diff --git a/server/evaluate/__init__.py b/server/evaluate/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/server/evaluate/evaluate_transcription.py b/server/evaluate/evaluate_transcription.py deleted file mode 100644 index a55c8ee4..00000000 --- a/server/evaluate/evaluate_transcription.py +++ /dev/null @@ -1,204 +0,0 @@ -import re -from pathlib import Path -from typing import Any, List - -from jiwer import wer -from Levenshtein import distance -from pydantic import BaseModel, Field, field_validator -from tqdm.auto import tqdm -from whisper.normalizers import EnglishTextNormalizer - - -class EvaluationResult(BaseModel): - """ - Result object of the model evaluation - """ - accuracy: float = Field(default=0.0) - total_test_samples: int = Field(default=0) - - -class EvaluationTestSample(BaseModel): - """ - Represents one test sample - """ - - reference_text: str - predicted_text: str - - def update(self, reference_text:str, predicted_text:str) -> None: - self.reference_text = reference_text - self.predicted_text = predicted_text - - -class TestDatasetLoader(BaseModel): - """ - Test samples loader - """ - - test_dir: Path = Field(default=Path(__file__).parent) - total_samples: int = Field(default=0) - - @field_validator("test_dir") - def validate_file_path(cls, path): - """ - Check the file path - """ - if not path.exists(): - raise ValueError("Path does not exist") - return path - - def _load_test_data(self) -> tuple[Path, Path]: - """ - Loader function to validate input files and generate samples - """ - PREDICTED_TEST_SAMPLES_DIR = self.test_dir / "predicted_texts" - REFERENCE_TEST_SAMPLES_DIR = self.test_dir / "reference_texts" - - for filename in PREDICTED_TEST_SAMPLES_DIR.iterdir(): - match = re.search(r"(\d+)\.txt$", filename.as_posix()) - if match: - sample_id = match.group(1) - pred_file_path = PREDICTED_TEST_SAMPLES_DIR / filename - ref_file_name = "ref_sample_" + str(sample_id) + ".txt" - ref_file_path = REFERENCE_TEST_SAMPLES_DIR / ref_file_name - if ref_file_path.exists(): - self.total_samples += 1 - yield ref_file_path, pred_file_path - - def __iter__(self) -> EvaluationTestSample: - """ - Iter method for the test loader - """ - for pred_file_path, ref_file_path in self._load_test_data(): - with open(pred_file_path, "r", encoding="utf-8") as file: - pred_text = file.read() - with open(ref_file_path, "r", encoding="utf-8") as file: - ref_text = file.read() - yield EvaluationTestSample(reference_text=ref_text, predicted_text=pred_text) - - -class EvaluationConfig(BaseModel): - """ - Model for evaluation parameters - """ - insertion_penalty: int = Field(default=1) - substitution_penalty: int = Field(default=1) - deletion_penalty: int = Field(default=1) - normalizer: Any = Field(default=EnglishTextNormalizer()) - test_directory: str = Field(default=str(Path(__file__).parent)) - - -class ModelEvaluator: - """ - Class that comprises all model evaluation related processes and methods - """ - - # The 2 popular methods of WER differ slightly. More dimensions of accuracy - # will be added. For now, the average of these 2 will serve as the metric. - WEIGHTED_WER_LEVENSHTEIN = 0.0 - WER_LEVENSHTEIN = [] - WEIGHTED_WER_JIWER = 0.0 - WER_JIWER = [] - - evaluation_result = EvaluationResult() - test_dataset_loader = None - evaluation_config = None - - def __init__(self, **kwargs): - self.evaluation_config = EvaluationConfig(**kwargs) - self.test_dataset_loader = TestDatasetLoader(test_dir=self.evaluation_config.test_directory) - - def __repr__(self): - return f"ModelEvaluator({self.evaluation_config})" - - def describe(self) -> dict: - """ - Returns the parameters defining the evaluator - """ - return self.evaluation_config.model_dump() - - def _normalize(self, sample: EvaluationTestSample) -> None: - """ - Normalize both reference and predicted text - """ - sample.update( - self.evaluation_config.normalizer(sample.reference_text), - self.evaluation_config.normalizer(sample.predicted_text), - ) - - def _calculate_wer(self, sample: EvaluationTestSample) -> float: - """ - Based on weights for (insert, delete, substitute), calculate - the Word Error Rate - """ - levenshtein_distance = distance( - s1=sample.reference_text, - s2=sample.predicted_text, - weights=( - self.evaluation_config.insertion_penalty, - self.evaluation_config.deletion_penalty, - self.evaluation_config.substitution_penalty, - ), - ) - wer = levenshtein_distance / len(sample.reference_text) - return wer - - def _calculate_wers(self) -> None: - """ - Compute WER - """ - for sample in tqdm(self.test_dataset_loader, desc="Evaluating"): - self._normalize(sample) - wer_item_l = { - "wer": self._calculate_wer(sample), - "no_of_words": len(sample.reference_text), - } - wer_item_j = { - "wer": wer(sample.reference_text, sample.predicted_text), - "no_of_words": len(sample.reference_text), - } - self.WER_LEVENSHTEIN.append(wer_item_l) - self.WER_JIWER.append(wer_item_j) - - def _calculate_weighted_wer(self, wers: List[float]) -> float: - """ - Calculate the weighted WER from WER - """ - total_wer = 0.0 - total_words = 0.0 - for item in wers: - total_wer += item["no_of_words"] * item["wer"] - total_words += item["no_of_words"] - return total_wer / total_words - - def _calculate_model_accuracy(self) -> None: - """ - Compute model accuracy - """ - self._calculate_wers() - weighted_wer_levenshtein = self._calculate_weighted_wer(self.WER_LEVENSHTEIN) - weighted_wer_jiwer = self._calculate_weighted_wer(self.WER_JIWER) - - final_weighted_wer = (weighted_wer_levenshtein + weighted_wer_jiwer) / 2 - self.evaluation_result.accuracy = (1 - final_weighted_wer) * 100 - - def evaluate(self, recalculate: bool = False) -> EvaluationResult: - """ - Triggers the model evaluation - """ - if not self.evaluation_result.accuracy or recalculate: - self._calculate_model_accuracy() - return EvaluationResult( - accuracy=self.evaluation_result.accuracy, - total_test_samples=self.test_dataset_loader.total_samples - ) - - -eval_config = {"insertion_penalty": 1, "deletion_penalty": 2, "substitution_penalty": 1} - -evaluator = ModelEvaluator(**eval_config) -evaluation = evaluator.evaluate() - -print(evaluator) -print(evaluation) -print("Model accuracy : {:.2f} %".format(evaluation.accuracy)) diff --git a/server/evaluate/predicted_texts/pred_sample_1.txt b/server/evaluate/predicted_texts/pred_sample_1.txt deleted file mode 100644 index d5fa2fa1..00000000 --- a/server/evaluate/predicted_texts/pred_sample_1.txt +++ /dev/null @@ -1 +0,0 @@ -We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas . Thank you for having me . You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . So , it 's a busy conference , but lots of exciting stuff going on . Yeah , it 's incredible . I mean , I want to zoom out for a second to start with , which is that this is obviously not your first time taking and packaging new technology breakthroughs for the enterprise . Both in your time at Oracle and now CEO of Google Cloud , this is something that you 've been doing for quite some time now . When you sort of zoom all the way out , what do you think are some of the things that have some of your principles , or some of your thoughts and enabling these technological breakthroughs and actually enabling the enterprise with them ? And what are the key insights that you have there ? Thank you . A lot of the work . So first of all , we 've really built out the organization the last three years . We 've seen a huge ramp up in our business , credit to all the people who joined us at one point over 70 % of organization that joined your in COVID . So they had n't met anybody . They could n't meet their managers , but they all did an amazing job together . The adoption of technology by companies , and I 'll give you just some elements , particularly in the application of AI in different domains that we 've seen . We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud . You know , fraud detection and banking , there 's a lot of false positives , which makes it hard to really , you know , to a very expensive people doing something called anti-money laundering . And our AI algorithms are really able to be super precise on detection . Explainability is a critical thing there , right ? So people ask , why did you , why did you approve , why did you flag this one and not that one ? Because regulators are involved . So explainability becomes a big deal . We help , we help renewal , for example , monitor all of the factories . The process roughly , a billion data sets every day . Obviously , humans can process that . But making it super simple to , and you guys have given all your expertise in labeling and other things , you would get a sense . Factory floor data is not clean data . And so you have to actually clean , imagine doing a billion data sets into an environment every single day . You have to give the data pipelines really good . And so a lot of technology work happens to make that possible for companies . Third is , if you shop at IKEA , for example , behind IKEA is systems , it 's our recommendation system . find IKEA is systems , it 's our recommendation system . And the way that people shop for furniture and products is not the same in all countries . And so how are you able to one deal with the benefits you get from a global model , but also to contextually the specific elements in each country because people have different buying habits . Those are all things that we 've learned applying our AI in different contexts in different parts of the world . Yeah . You 've sort of glossed over this , but you 've led since you took over at Google Cloud , just a meteoric growth of the platform . You know , I think the past few years , you 've tripled your sales force and ending last year , you obviously ca n't come in this , but end the last year at , I believe , 20 billion of annual revenue , which is incredible and this incredible growth journey . What do you attribute your success to ? And how do you think you 've been able to drive just to an incredible growth and success ? From our point of view , every industry , virtually in the world , is now becoming a software powered technology industry . If you talk to automobile companies , they 're increasingly vehicles are more about software than mechanical systems . If you talk to telecommunications companies , the networks are commodities unless they can make them platforms to deliver applications , so they need new ways to slice , manage the network . If you look at banks at the end of the day , they 're about all the products of a bank or data , and all of that becomes how do you differentiate in the value delivering clients through a digital medium ? Because increasingly , I 'm sure all of you look at yourselves and go when was the last time I went to a branch of a bank . So a lot of our work has been pushing the technology innovation really far , but bringing that technology super easily to people in different industries . And given the demand that people have for a hair , I really want , I need the technology to help me power my industry , the change I 'm seeing in my industry , the more accessible we can make it , the easier and the faster we get adoption , and our approach has been to be completely open . And when to be completely open . And when I say completely open , we offer every part of the stack that we have from the hardware and network to the software abstractions above to things that are more packaged because different organizations have different levels at which they have expertise and want to adopt technology . Yeah . I mean it 's been , mean it 's been obviously incredible . You know going back to AI for a second , Google , Google obviously is an early mover in AI and Google Cloud has also been through , you know , starting with TensorFlow and Vertex AI and AutoML and so many incredibly innovative technologies . And AI has been obviously kind of a buzzword for some time now within the industry . And I think we see this in use as well . The adoption has maybe been a bit slower than we would expected until now . What do you think have been the barriers to greater levels of AI adoption , greater levels of enterprise that 's in value from AI ? And what do you think the future holds ? So we 've worked with a huge number of companies doing work , having them adopt AI . A lot of the lessons we 've seen and observed from it are the barriers to adoption are rarely about the algorithm itself . It 's often the barriers to adoption about very algorithm itself . It 's often the various adoption about very different things . So when we work with customers in many , many industries , take retailers an example , and you think of a very mundane example , like recommendations , to make product discovery on the web much easier for their own products . The biggest challenges standardizing the meaning of the product and the catalog . Because unless you have a standardized definition of the products and the data behind the algorithm is clean , it 's super hard to actually get to recommendation . And so in the work we did with H & M , for example , or at Macy 's , or at IKEA , or Bloomingdale 's , a huge number of these brands , the big part of the program is actually how do you label and clean the data upfront and standardize it before you get into the algorithmic phase . So that 's one part of things we see . Second part is for large organizations to adopt AI , they have to need to integrate the results of the algorithm back into their core processes . So , you know , practical example , we work with OGE , OGE is a large , large electric producer , electricity and power producer in Europe . They are probably one of the largest renewable energy producer in the world . They use wind farms . One of the things they really struggled with was , how do you predict how much wind is going to be there three days from now ? Because the power grid requires that prediction in order to capacity plan how much power is going into the grid . So they work with us and they use our AI to do that . But that needs to be tied into how they 're telling the rest of the power sources that work on the grid . Hey , if this went to wind is coming in , here 's all the other sources in each generation . So tying it back in is not as simple as people think . And so a lot of time is that the third on the people side , there 's change management you go through to get people to trust the algorithm . So one of the things we 've done work with many banks , particularly during the pandemic , when the government issued small business loans . There was a giant bottleneck in being able to get loans out to individual consumers . And frankly , because the banks did n't want to bring a huge army of loan officers in , they had to use software and algorithms to process it . Now the challenge people had is they needed to trust the algorithm was being fair in saying yes to some and no to others and that it would mirror for example the recommendations that their best mortgage bankers would do , right ? Just as a loan office as we do . So it gave them the benefit of scale because we processed literally millions and millions of mortgages through our technology , but it required them to get comfortable that things like fairness and other things were working . So often when people look at AI , they think it 's a skills issue . There 's certainly a skill issue involved . There 's not enough talent in the ecosystem . But things are getting easier and easier as the models get more and more sophisticated . Often people forget about these other issues that are important in getting adoption . Yeah . I mean , you 're preaching the choir when you mention the data challenges that all these enterprises face and how critical that is to getting working in the early days . One of the things that I think is interesting about Google Cloud strategies that you really have products at different layers of the stack and different layers of closest to the bare metal all the way up to these package solutions . In what way do you think that the enterprise world and even the broader business world is going to adopt these AI technologies ? Do you think that the end state is that a lot of them are using your lower level , more infrastructure ? Products , or do you think that many of them are going to adopt solutions ? How do you think this plays out over the next few years ? So we offer four layers of technology for people . There 's a set of people who say , look , I just need your computational infrastructure , your large systems . We build something called tens of processing unit , which is our large scale systems . We 're also working with Crossing Unit , which is our large-scale systems . We 're also working within video to build a really high-scale GPU Bay system . But many people , some customers say , look , I just need access to that . And we make that available because the TPUs are what we use within Google . And we make that available along with the compilation software to optimize models on the TPUs . Take as an example , LG , the Korean company that makes appliances , their team is built a large , I mean , multi-hundred billion parameter model , because they wanted to make that a way that people can interact with appliances without having to press buttons on them . So they built a model . They said , I just need access to your infrastructures . That 's one way we offer a peak capability . A second level is people say look , I really do n't need access to the raw infrastructure itself . What I need is the ability to build models using your platform . And so we offer a platform called Vertex and people build models and push them using our machine learning platform . And there are many , many organizations in logistics and financial services in retail and others who build their own models on top of the platform . The third is to make things even easier , we 've taken some of the core pieces , translation , documents , image processing , video . And we 've said , we can offer an auto-email based solution , which further simplifies how you use our platforms . And so for example , translation , we have a capability to handle translation in 135 languages . One of the important things that people ask when they go to many languages is if you look at the data sets that I used to train models , they are primarily , there 's a large set in English , because you have the whole internet is primarily in a very small number of languages . But once you get to more narrow languages , for instance , Swahili or some of the African languages , or even in Asia , there are many languages , even from where I grew up in India . There are languages that are not as widely represented on the internet . Can your model in translation provide equivalent fidelity in sparse languages ? Because it 's always important to those people only understand that language that they get a high fidelity result . So we 've built something called translation hub and it 's being used in very mundane places but with extraordinary impact . For example , when people announce COVID guidelines or recently monkey parks , for example , which is another thing , they needed translate many , many languages . And normally the process would take a long time . We have movie studios , for example , in a different example , saying , hey , when we launch a movie , we have a high fidelity set of languages , we 're actually going to hold the movie up and show that people do it . But for the long tail , we just need captioning . We 're not necessarily going to do voice dubbing . We 're going to do captioning . And they use our translation solutions to go to that . Even within companies , every medicine , for example , uses it to translate all their instruction manuals into many languages for their technicians . And then lastly , in some places , there are companies like retailers who tell us , look , a handful of the largest retailers may build their own software teams . But some of us who are small merchants , we 're not software companies . And telling us , you 've got to be a software company to use AI is not fair . So for some industries , we actually build fully packet solutions . If you call many telephone companies , the context center , behind it , sits our voice agent . And the rationale behind that was super simple , when a new smartphone launches like an iPhone or a Pixel , typically in the morning of the launch , some of these contact centers get three , four million calls in an hour . And it 's hard to hire that many agents to handle the phones . So we said , why would n't software be able to handle it ? We then evolved it so that the natural language interface can become actually the workflow for these organizations . But that 's a much more of a package solution so that telephone companies do n't have to have armies of data scientists to do it . So our work spans all of these because people have different needs and we find that as you improve the maturation of this and you make it more easy for people to adopt it . You will get broader proliferation and adoption of AI as a whole . Yeah , you know , you walk through so many different use cases and so many applications to the technology . I imagine one , and there 's so desperately , you know , everywhere from , you know , fraud detection to translation to translation of manuals , you know , there 's such a wide translation of manuals . There 's such a wide array of use cases . How do you all like Google Cloud think about helping businesses understand what is AI good for ? What can they use AI for ? There 's obviously such a wide diversity of different use cases , but what at a framework level do you tell them , how can I use AI within my business ? It 's a really good question . I mean , a lot of our work actually comes from clients asking us now , and that 's actually an encouraging thing . Because you know , see from up on the view , some simple things , how many of you believe in a few years ' time there 's gon na be intelligence software and non-intelligence software , right ? I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software . I mean , nobody would say in three , four years ' time , we 're going to write software that has not powered in some form of fashion by AI . So you know , in most companies actually , it 's really encouraging to see that they look at domain problems they 're having and say , for instance , I used to do it using a rules engine , which is an older model for defining kind of workflow within organizations . Can you apply AI to do it in a new way ? I used to do this in a specific way . I heard about image recognition . One example really fun or interesting one , US Navy , when you have corrosion on the base of ships , the old way was to lift it into dry dark and take a look at it . If you 've ever seen one of these ships , you can imagine lifting to dry dark is not an easy thing . So they said , can we fly a drone with your camera image recognition around it and detect corrosion ? And so what we 've seen is that as you lift up the capability where image , audio , text , et cetera , all these forms of input can be processed extremely accurately , most customers start figuring it out . And so they call us with , most of our work has come from customers calling us , saying , hey , I have this need . Can I apply AI to it ? And so we talk to them about how and when it makes sense to use AI . But we also talk to them about the consequences if the models are not handling things like skew in the data . How do you ensure that , for example , you 're treating fairness properly ? How do you ensure that the model is safe , etc . Yeah , I think it 's , I mean , all the use cases , the variety is incredibly exciting . It 's cool that these customers are coming to you directly with many of them . What is , again , kind of thinking bigger picture , what is machine learning an AI mean for Google Cloud on the whole over the next call 510 years ? So we feel that the boundary of what machine learning and what AI can do will change over time . When it started , it was about doing what we would call assistive things . Assistive things are where a human being is able to do it , but the computer assists the human being in some ways to do it better . Right ? So common examples people talk about is , hey , your doctor or radiologist , you used to look at x-ray images . Now , a computer is going to look at it and detect tumors , but it 's assisting you to find something that you may have done another way . So that 's the first phase and a lot of the work we see is primarily in that phase today . The second phase is to do something where you could n't do it with a human because the quantity of data you need to process or the amount of people you need would be just far too significant . And so the machine is doing something that humans could n't do , but it 's still an incremental element on top of what humans could do themselves . The third phase , I think , is where we think generative AI , for example , goes , because it 's about enabling people to express themselves in a different way , and to assist them in expressiveness . So I 'll give you a practical example . A lot of you probably use tools , slides , and things like that in your day to day job . PowerPoint was invented a long time ago and was really just about drawing things . You know , I 've got a 14 year old . And so if you look at the younger generation , if you look at what slides were , they were really tools to help people draw . And then to take what was on the slide projector and presented . Then the younger generation says , hey , I do n't want to draw things that 's really old-fashioned . I 'm going to go to the internet and copy images , right ? Because when they do class projects , they 're copying images into the slides . And then , as people observe , you know , on the social media environment , people going from text , which may have been Facebook to short images , which is Instagram to short video TikTok , we would say , hey , why would n't we be able to record short video ? And be used that as a mechanism to share . But recording short video is still capturing the real world through the lens of the camera . What people want is a more expressive way of saying , I have an idea , can I translate it ? And it may not be something I can capture . Imagine a kid in California and a school saying saying I want to capture how the landscape and outside of Paris and France is right now . I think they need to be able to generate some of the ideas that they could capture by physically being there . And so we 're working on all of this and we 're bringing some of these into our products to change what people could possibly do through the application of AI so they improve expressiveness for people . And so every boundary as the technology gets more sophisticated we think it moves from just assistance to assistance on things that human beings may not have been able to just linearly do to now things like expressiveness , which is a very different capability than people could actually do themselves . Yeah , I mean , all of this is very obviously incredibly exciting and we 're all watching it happen in real time . There 's an artist who actually described the image generation models as , he sort of image generation models as he was , he sort of said like , you kind of think about like a camera . Like it 's a new tool that allows you to create fundamentally new forms of art . That 's right . Yeah . And not just one medium of art , right ? Because if you look in the past , people said , you were a painter , you were a sculpture , you were a musician , and now these technologies allow you to blend all of it as a form of expressiveness . Yeah . You know , the last question I have for you is , you know , you obviously sit down with many of the sort of leading CEOs and business leaders of of the sort of largest organizations in the world . And I 'm sure one thing that is on many of their minds is sort of as AI technology develops and it continues to progress is potential disruption that might come from art of film intelligence . What sort of , how do you approach that conversation ? What sort of your advice to these business leaders who are looking at this powerful new technology and thinking about what that might mean for the businesses and the business landscape . When we talk to CEOs , I mean the biggest things we talk to them about number one , productivity in the long term , productivity has always been the primary driver of improving both company productivity , meaning their own companies , as well as societal benefit , things like affluence of a society , etc . And the means and equality of distribution of income to people across all spectrum society . Eventually , the most important metric , and you can look at any economic textbook is productivity . Software and technology has probably been the biggest boomer productivity over the last 30 , 40 years . This is the next evolution of that . And so we always say , if you approach it the right way , for example , labor shortages are going on right now . The biggest potential benefit is the application of some of these platforms like AI to do in that . The second , with any technological generation revolution , like artificial intelligence , but if you went back in time and looked at the industrial revolution , etc . There are always during the period of transition , anxiety about the consequences of that technology . And it does n't mean the technology by itself is good or bad . It 's the application of the technology that 's good or bad . So it 's incumbent upon both the technology providers and the users of the technology to ensure that the negative consequences of it are managed properly . Right ? The obvious example is , for instance , if you look at a very simple thing , image recognition . Image recognition can help doctors find tumors way better than having the best radiographer . It 's a system in that context and it 's like helping people with a better quality microscope than they had before . Object recognition is helping people find , for example , people who are in the ocean much more accurately so the coastguard can rescue them . At the same time , being able to use a camera and say that 's Thomas Korean has , you know , a lot of potential negative consequences . And so as a provider of technology , we at Google have chosen not to do that third part . But we also tell companies , it 's important not just to say , this is what 's regularly allowed by the legal framework , because law in many countries is not yet keeping up with how fast AI technology is moving . But to take the responsibility as a company CEO to say , here 's what I believe comfortable with , and here 's what I wo n't be comfortable with . Yeah . Well , Thomas , thank you so much for such incredible conversations . I think I 'm very heartened to hear all the incredible work that Google Cloud is doing to make artificial intelligence accessible to the entire business world and all of every enterprise around the globe . And I 'm so excited that you 're able to join us . Thank you so much . Thank you so much for having me . Thank you . Thank you . \ No newline at end of file diff --git a/server/evaluate/predicted_texts/pred_sample_2.txt b/server/evaluate/predicted_texts/pred_sample_2.txt deleted file mode 100644 index ca15f6c5..00000000 --- a/server/evaluate/predicted_texts/pred_sample_2.txt +++ /dev/null @@ -1 +0,0 @@ - Well, health technologies ticker civil W-E-L-L. on the TSX recently reported it's 2023 Q1 results beating the streets consensus estimate for revenue and adjusted EBITDA. And in a report. you this week. Raymond James, and we'll say quote, we're impressed by Wells capacity to drive. powerful growth across its diverse business units in the absence of M&A. me today, you'll see your Honour Tribazi to, okay, what's next for well health. Good to see you Sir, how are you? Great to see you Richard. Thanks very much for having me. Great to have you. congratulations on your 17th consecutive quarter of record grab you. And you share some insights into what's driven these results historically and in the past Porter was found. Yeah, thank you. We were very excited about our Q1220. three results. And as you mentioned, we've had a long, you know, successful. stream of continued growth and record growth. also had accelerating organic growth. And I think a big part of the success of our franchise here is the incredibly sticky and predictable revenue that we have. Well over 90% of our business is either highly vio-curring as in. the, you know, highly predictable results of our two-sided network of patients. of providers or truly recurring as in schedule or subscribe. revenues. And this allows us to essentially make sure that you know we're on track it obviously you know like any other business things happen. And sometimes it's hard to meet those results, but what's really being unique about our platform is do you have exposure to all kinds of different aspects of health care? You know, we have primary care. care and specialized care on both sides of the border in the U.S. and Canada. So we have exposure to different types of care. It's a business model, we have exposure to the US payer network, which has higher per unit economics. and Canada and of course the stability and and sort of of higher fidelity kind of collections and revenue cycle process the candidate has over United States where you don't have to kind of deal with all of that payment noise. just a lot of, I'd extract built into the platform because of the diversity of different health care. businesses that we support. And Where do you see Wells future growth coming from which part of the business? excites you the most right now. Yeah, we'll look the centrifugal force of well is health care provider, and we exist to tech, enable, and amelor the business of that tech provider. And that's what we're laser focused on, and what we're seeing is providers not wanting to this is anymore. It's very simple. And so we have a digital platform and providers can either acquire what they want and be from our digital platform and implement themselves or they can decide that they don't want to run a business anymore and they don't want to configure and manage technology, which is becoming a bigger and bigger part of the world every single day. day and when we see what we've seen with that dynamic is that it's a lot of them are now just wanting to work in a place where all the technologies configured for them. It's wrapped around them and they have a competent operating partner. that is supporting the practice and take care. can care of the front office in the back office so that they can focus on providing care. This results in them seeing patients and being happier because, you know, they became doctors to see patients, not so they can manage workers and deal with HR issues and deal with lack of And all that kind of stuff. Excellent. And I know too that acquisition supply to keep roll it in well. Can you share it inside into our positions fit in? to Wells growth strategy. Sure, and look at 2020. 2020 and 2021 we did a lot of acquisitions and 2022 we took a bit of a breather and we really focused on it integration, and I think that's one of the reasons why you saw this accelerating organic growth. We really we're able to demonstrate that we could bring together or the different elements of our technology platform. We started to sell bundles. We started to really derive synergy. and activate, you know, more sales as a result of selling all their different products as services with one voice, with one vision. So we made it easier for providing to use their technology. And I think that was a big reason for our growth. now M&A as you know we're a capital allocation company we're never far from And so we did continue to have, you know, toughens here and there. And in fact, today, We announced that we've acquired the Alberta operations of MCI-1 health. another publicly traded company who is looking to raise funds to support their business. We're very pleased with this acquisition and it just demonstrates our continued discipline. plan. These are, you know, great primary care clinics in Canada. It right in the greater Calgary area and you know, just allows us to grow our footprint. in Alberta, which is an important province for us, and it's if you look at the price. If you look at what we're getting, you know, it's just a positive of our continued. discipline. And just, you know, a few days ago at our conference call I mentioned that we have you know a really strong line up of acquisitions and you know they're starting to I think come to fruition for us. I'm helping you on the road no question. you recently announced a new AI investment program last month. What's specific areas of health care technology your AI are you focusing on and what's the when it comes to AI. Yes, I look AI. as as as as I'm sure you're aware is it's become you know really an incredibly important topic taken in all aspects of life. of business and, you know, not just business socially as well. Everyone's talking about. this new breakthrough disruptive technology, the large language models, and gender to the AI. I mean, look, AI has been about a 80 year old overnight success. a lot of people have been working on this for a long time. Generative AI is just sort of the culmination of a lot of things coming together and working, but it is uncorrect, enormous. innovation and and we think that this there's a very good news story about this in healthcare, particularly where we were looking to look, we were looking to look. unlock the value of the data that we all produce every single day. as humans. And so we've established an AI investment program. because no one company can tackle all of these innovations themselves and what will done, too, is taken a very much an ecosystem approach by establishing its app stock. marketplace. And so we're very excited about not only allocating capital into promising young AI companies that are focused on digital health and so I'll health care problems, but also giving them access to, you know, safely purely to our provider network, to our, you know, to our outpatient clinic. which is the largest owned and operated network in Canada by far. So, And when these, and it's, it was remarkable. We, we announced this, program. We've had just in the in the first week to 10 days we've had over a 100 inbound prospects come in that wanted to, collaborate with us. And again, I don't think that's necessarily for the money, you know, we're saying we would invest in it. minimum of a quarter of a million dollars, you know, a lot of them will likely be higher than a quarter of a million dollars. So it's not life-changing money, but our structural advantages and and the benefits that we have in the well network. Those are extremely hard to come by. And I think you'll see us, you know, help some of these companies. and these succeed and they will help us drive, you know, more innovative. that helps the provider. It's speaking of this very interesting AI. I know you're coming. just launched well AI voice. This is super interesting. Tell me what it is and the impact it could have on health care provides. Yeah thanks for asking our providers are thrilled with this. We've had a number of of our own well providers testing this technology. and it really feels like magic to them. It's essentially an ambient AIS. powered scribe. So it's a service that with the consent of the party involved listens to the conversation between a patient and provider. And then essentially condenses that into a medically relevant note for the chart files. Typically that is a lengthy process, a doctor has to transcribe notes, then review those notes and make sure that a appropriate medically oriented instruction. should notice is prepared and put into the chart. And that could take, you know, sometimes more than more time than the actual consultation. time. And so we believe that on average if it's you. regularly and consistently, this can give providers back at least a third of their day. And it's just a game changer. And it's just a game changer. And it's just a game changer. And We have now gone into general release with this product. It's widely available and Canada, it has been integrated into our EMR, which makes it even more valuable tools like this are going to start popping up, but if they are not integrated into your practice management system, then you have to kind of have data in in more than one place and move that around a little bit, which makes it a little bit more defensive. difficult, especially with HIPAA requirements and regulations. So again, I think this is the first of many types of different products and services that allow doctors to place more emphasis and focus on the patient and experience instead of having their head in a laptop and looking at you once to the wild, they'll be looking at you and speaking to their practice management system. And I think this, you know, think about it as a lot. except for for doctors, you know, this disability to speak. and have, you know, voice driven AI assistant that does things like this, I think are going to be incredibly helpful and valuable for healthcare providers. super fascinating. I mean we're just hearing you know more about AI maybe AI the first time, but here you are with the product already on the market and in the health care field. That's got to be pretty attractive to the out there right ahead of many other people, right? Thank you Richard. Thanks for that recognition that that's being our intention. We we want to demonstrate that we're all in on. ensuring that technology, the benefits providers is, is, is, is, is, accelerated and de-risk and provided, you know, and in a timely way, you know, providers need this help, we have a health care crisis in the country that is generally characterized as a lack of doctors. And so imagine if we can get our doctors to be 20 or 30% more productive. through the use of these types of tools. Well, they're going to see more pickations. And that's going to help all of us and and look if you step back well as be This model is all about having exposure to the success of doctors and doing our best to help them be more successful because we're going to revenue share relationship with most of the doctors with. And so this is good for the ecosystem. It's great for the provider and it's great for well as well. Super fascinating, Hamed Shabazi CEO, well technologies, ticker, W-E-L-L, great to catch up again. Thank you, sir. Thank you Richard appreciate how you having this \ No newline at end of file diff --git a/server/evaluate/predicted_texts/pred_sample_3.txt b/server/evaluate/predicted_texts/pred_sample_3.txt deleted file mode 100644 index 57d1582d..00000000 --- a/server/evaluate/predicted_texts/pred_sample_3.txt +++ /dev/null @@ -1 +0,0 @@ - medicine is hard work, as most of them sit easy. It takes your lectures and notes to create a Personalized study plan with exclusive videos, practice questions, and flashcards. and so much more. Try it free today. In diabetes Melodies, your body has trouble moving glucose, which is a type of sugar from your blood into your cells. This leads to high levels of glucose in your blood and not enough of it in your cells. And remember that your cells need glucose. glucose as a source of energy. So not letting the glucose enter means that the cell's star for energy. this fight having glucose right on their doorstep. In general, The body controls how much glucose is in the blood relative to how much gets into the cells with two insulin and boogogon insulin is used to reduce bloke glucose levels and glucogannas used to increase bloke glucose levels. Both of these hormones are produced by clusters of cells in the pancreas called eyelets of longer Insulin is secreted by beta cells in the center of these islands, and Lukagan is secreted by Elfisels in the periphery of the islands. the amount of glucose in the blood by binding the insulin receptors embedded in the cell membrane is insulin-responseed tissues, like muscle cells in adipose tissue. When activated, the insulin receptors cause vesicles containing glucose transporter that are inside the cell to fuse with the cell membrane, allowing glucose to be transported into the cell. Who could go on does exactly the opposite? It raises the boy group of glucose levels by getting the liver to generate new molecules of glucose from other molecules. And also, break down glycogen into glucose so that it can all get dumped into the blood. Diabetes notice is diagnosed when blood glucose levels get too high and this is seen among 10% of the world population. There are two types of diabetes, type 1 in type 2. And the main difference between them is the underlying mechanism that causes the blood glucose level. to rise. About 10% of people with diabetes have type 1 and the remaining 9% 90% of people with diabetes have type 2. Let's start with type 1 diabetes. some times just called type 1 diabetes. In this situation the body doesn't make enough insulin. The reason this happens is that in type 1 diabetes there's type 4 hypersensitivity response or a cell mediated immune response, where a person own T-cells attack the pancreas. As a quick review, remember that the immune system has T cells that react to all sorts of antigens, which are usually small peptides. polysaccharides or lipids. And that's some of these antigens are part of our own body's cell. It doesn't make sense to allow T-cells that will attack our own cells to hang around. until there's this process to eliminate them called self-tolerance. In type 1 diabetes, there's a genetic abnormality that causes a loss of self-tolerance T cells that specifically target the beta cell antigens losing cell Lawrence means that these T-cells are allowed to recruit other immune cells, and coordinate on these beta cells, losing beta cells means less insulin. and less insulin means that glucose piles up in the blood, because it can't enter the body's cells. One really important group of genes involved in regulation of the immune response is the human glucoseide antigen system, or HLA system. Even though it's called a system, it's basically this group of genes on chromosome 6 that encode the major histocompatibility. complex or MHC, which is a protein that's extremely important in helping the immune system recognized foreign molecules, as well as maintaining self-tolerance. MHC is like the serving platter that antigenes are presented to the immune cells on. Interestingly, people with Type 1 diabetes often have specific HLA genes in common with each other. One called HLADR3 and another called HLADR4. But this is just a genetic clue, right? Because not every one of HLADR3 and HLADR3 and LATR4 develops diabetes. In diabetes mode, it's type 1. destruction of beta cells usually starts early in life. But sometimes up to 90% of the beta cells are destroyed before symptoms crop up. control diabetes that all sound similar are polyphasia, like osteoorosis. polyureka and polydipsia. Let's go through them one by one. Even though there's a lot of glucose in the blood, it cannot get into the cells, which leaves cell star for energy. So in response, Adobos tissue starts breaking down fat. called like policies and muscle tissue starts breaking down proteins. Both of which results in weight loss for someone with uncontrolled diabetes. This catabolic state leaves people viewing hungry. Also known as Polyphasia. Phasia means eating and Poly means Now with high glucose levels that means that when blood gets filter through the kidneys, some of it starts to spill into the urine, called glycosurio. first of glucose and urea the urine. Since glucose is asthma automatically active, water tends to follow it, resulting in an increase in urination or polyurethane Allie again refers to a lot and you're yeah again refers to urine Finally, because there's so much urination, people with uncontrolled diabetes. become dehydrated and thirsty, or polydipsia. Poly means a lot. dipcr means thirst. Even though people with diabetes are not able to to produce their own insulin, they can still respond to insulin. So treatment involves life long. insulin therapy to regulate their blood glucose levels and basically enable their cells to use glucose. One really serious complication with type one diabetes is called diabetic keto acidosis or D.K.A. Don't understand it, let's go back to the video. the process of light policies, or fat is broken down into free fatty acids. After that happens, the liver turns the fatty acids into ketone bodies. Like a silo, a silo. acid and beta hydroxypeteric acid. A cdoc acid is a keto acid. because it has a ketone group and a carboxylic acid group. Beta hydroxy beturus. acid on the other hand, even though it's still one of the ketone bodies, isn't technically a keto acid. so that's keytown group has been reduced to a hydroxyl group is keytown bodies are important because they can be used by cells for energy, but they also increase the acidity of the blood, which is why it's called keto acidosis, and the blood coming really acidic can have major effects throughout the body. Individuals can develop a small respirator. which is a deep and labored breathing as the body tries to move carbon dioxide out of the blood. In an effort to reduce its acidity, cells also have a transport order that exchanges hydrogen ions or protons for potassium. When the blood gets acidic, it's by definition loaded with protons, like it's said in the cells while potassium gets sent into the fluid outside cells. And keep in mind is that in addition to helping glucose enter cells, insulin stimulates async ATPases, which help potassium get into the cells, and so without insulin more potassium stays in the fluid outside cells. Both of these mechanisms lead to increased potassium in the fluid outside cells, which quickly makes it into the blood and causes hyperphysis. The potassium is then extruded, so over time, even though the blood potassium levels remain high. Overall stores of potassium in the body, which include potassium inside cells starts to run low. Individuals will also have a high anion gap, which reflects a large difference in the unmeasured negative and positive ions in serum largely due to the buildup of keto acids, diabetic keto Asadosis can happen even in people who've already been diagnosed with diabetes and currently have some sort of of insulin therapy. This is up in that frame, which in turn stimulates the release of Glucogon. Too much Glucogon. can tip the delicate hormonal balance of Boopagonan insulin in favor of elevating blood sugars and can lead to a cascade of events we just described. Increase glucose in the blood. loss of glucose in the urine, loss of water, dehydration, and in parallel and need for alternative energy, generation of ketone bodies, and keto acidosis. Interestingly, both ketone bodies break down into acetone, and escape as a gas by getting breathed. out the lungs, which gives us sweet, fruity smell to a person's breath. Although that's the only sweet thing about this illness, which also causes nausea, vomiting and if severe, mental status changes in acute cerebral edema. Treatment of a DKA episode involves giving plenty of fluids, which helps with dehydration. insulin which helps lower blood glucose levels and replacement of electrolytes like potassium. All of which help to reverse the acidosis. Now let's switch gears and talk about type 2 diabetes, which we're at the body makes instance. but the tissues don't respond as well to it. The exact reason why cells don't respond isn't fully understood. Essentially the body is providing the of insulin, but the cells don't move their glucose transporters to their membrane in response. Which remember is needed for the glucose to get into the cells? These cells therefore have insulin resistance. Some risk factors for insulin resistance are obesity. city, lack of exercise, and hypertension. The exact mechanisms are still being explored. For example, an excess of adipose tissue or fat. It's not to cause the release of free fatty acids and so-called adabokines, which are signaling fuel-second cause inflammation, which seems related to insulin resistance. However, many people that are obese are not diabetic, so genetic will play a major role as well. We see this when we look at twin studies as well. We're having a twin with Type II Diabetes increased at the risk of developing Type II Diabetes. completely independently of other environmental risk factors. And type 2 diabetes. Since tissues don't respond as well to normal levels of insulin, the body ends up producing more insulin in order to get the same effect and move glucose out of the blood. through beta cell hyperplasia and increased number of beta cells. and beta cell hypertrophy, where they actually grow in size. All in this attempt to pump out more insulin. This works for a while and by keeping insulin levels higher than normal, but glucose levels can be kept normal, called normal glycemia. Now, along with insulin, beta cells also secret eyelet and alloy polypeptide. So while Beta cells are cranking out insulin, they also secrete an increase. the amount of amuline. Over time, amuline builds up and aggregates in the islands. This beta cell compensation, though, is not sustainable. and over time those maxed out beta cells get exhausted, they become dysfunctional, and high-potrophy and get smaller, as well as high-boa-plasia and die-off. As beta cells are lost in insulin levels decrease, glucose levels in the blood start to increase in patients develop hyper glycemia, which leads to similar clinical signs that we mentioned before, like Polyphasia, Bikosuria, Polyuria. and polydipsia. But unlike type 1 diabetes, there's generally some circulating insulin and type 2 diabetes from the beta cells that are trying to compensate for the insulin resistance. This means that the insulin book gun balances such that diabetic ketoacidosis is not usually developed. Having said that, a complication called hyper-osmola hyper glycemic state, or HHS, is much more common in type 2 diabetes than type one diabetes and it causes increased plasma ultimately due to extreme dehydration. and concentration of the blood. To help understand this remember that glucose is a pull their molecule that cannot passively diffuse across cell membranes, which means that it is a solid. So when levels of glucose are super high in the blood, meaning it's a high for Osmolir State, water starts to leave the body cells and enter the blood vessels. Even the cells relatively dry and trival, rather than plump in juicy. blood vessels that are full of water lead to increased urination and total body dehydration. This is a very serious situation because the dehydration of the body cells and in particular the brain can cause a number of symptoms, including mental status changes. In HHS, you can sometimes see mild hedonemia and acidosis. but not to the extent that it's seen in DKA. And in DKA, you can see some hyper-oscalary. So there's definitely overlap between these two syndromes. type 1 and type 2 diabetes are also a couple other sub types of diabetes melodies. The stational diabetes is when pregnant women have increased blood glucose, which is particularly during the third trimester. Although ultimately unknown, because it's thought to be related it's a pregnancy hormones that interfere with insulin action on insulin receptors. Also, sometimes you build and develop drug-induced diabetes, which is where medications have side effects that tend to increase blood glucose levels. for both of these is thought to be related to insulin resistance, like type 2 diabetes. an autoimmune destruction process like in type 1 diabetes, diagnosing type one or type 2 diabetes is done by getting a sense for how much glucose is floating around in the blood. and has specific standards that the World Health Organization uses. and glucose tests is taken where the person doesn't eat or drink. Except the water, that's okay. for a total of eight hours and then has their blood tested for glucose levels levels of 100 milligrams per deciliter to 125 milligrams per deciliter indicates pre-dite. and 126 milligrams per dec leader or higher indicates diabetes. A non-fasting or random glucose test can be done at any time. with 200 milligrams per deciliter or higher being a red flag for diabetes. Another test is called an oral glucose tolerance test, where a person is giving glucose and then blood samples are taking at time intervals to figure out how well it's being cleared from the blood. And most importantly, interval being 2 hours later. Levels of 140 milligrams per desuiter to 109 99 milligrams per deswitter indicate pre-diabetes. 200 or above indicates diabetes. Another thing to know is that when blood the glucose levels get high, the glucose can also stick to proteins that are floating around in the blood or in cells. So that brings us to another type of test that can be done, which is the HBA1C. test, which tests for the proportion of hemoglobin and red blood cells that has glucose stuck to it. to it, of glycated hemoglobin, HPA1C levels of 5. 0.7% to 6.4% indicate pre-diabetes and 6.5% are higher indicates diabetes. This proportion of glycated hemoglobin doesn't change day to day So it gives a sense for whether the blunt glucose levels have been high over the past two to three months. Finally, we have the CPATH diet test, which tests for bi-products. of insulin production. If the level of CPF tied is low or absent, it means the pancreas the level of CPF tied is low or absent. no longer producing enough insulin, and the glucose cannot enter the cells. For type 1 diabetes, insulin is the only treatment option. For type 2. two diabetes on the other hand. Lifestyle changes like weight loss and exercise. along with the healthy diet and oral anti-diabetic medication, like met for women in several other classes. in some times to be enough to reverse some of that insulin resistance and keep blood sugar levels in However, if oral anti-diabetic medications fail, I have two different types of medications. diabetes can also be treated with insulin. Something to bear in mind is that insulin treatment comes with a risk of hypoglycemia, especially if insulin is taken without a meal. Symptoms of hypoglycemia can be mild, like weakness, hunger, and shaking, but they can progress to a loss of consciousness in seizures in severe cases. In mild cases, drinking juices or eating candy or sugar, might be enough to bring blood sugar up, but in severe cases intravenous glucose should be given as soon as possible. The FDA is also recently approved the treatment for severe hypobacemia. All right, now over time, high glucose levels can cause damage to tiny blood vessels, while the micro-abache of the turret. across this called high line arteriolosthlerosis is where the walls of the arteriolos develop deposits, which are deposits of proteins, and these make them hard and inflex In capillaries, the basement membrane can flick it, and make it difficult for oxygen. to easily move from the capillary to the tissues, causing hypoxia. One of the most significant effects is that diabetes increases the risk of medium and large arterial wall damage, and subsequent atherosclerosis, which can lead to heart attack. and strokes, which are major causes of morbidity and mortality for patients with diabetes. In the eyes diabetes can lead to retinopathy and evidence of that can be seen on to find a scopic example that shows cotton wool spots or flare hemorrhages and can eventually In the kidneys, the afferent and e-ferrent arterioles as well as the glimmerialist itself can get damaged, which can lead to a nephrodic syndrome that's slowly damaged. diminishes the kidney's ability to filter blood over time, and can ultimately lead to dialysis. Diabetes can also affect the function of nerves causing symptoms like a decrease in sensation in the toes and fingers. Sometimes called a stock and glove distribution. As well as causes the autonomic nervous system to malfunction. That system controls a number. of body functions, everything from sweating to passing gas. about the poor blood supply and nerve damage can lead to ulcers, typically on the feet. And don't heal quickly and can get pretty severe and need to be amputated. These are some of the complications of uncontrolled diabetes, which is why it's important to diagnose and control diabetes. IPDs through a healthy lifestyle, medications to reduce insulin resistance. and even insulin therapy if beta cells have been exhausted. Well, type 1 diabetes kidney. not be prevented. Type 2 diabetes can. In fact, many people diabetes can control their blood sugar levels really effectively and live a full and active life without any of the complications. patients. Thanks for watching. If you're interested in the deeper dive on this topic. Take a look at us most is.org where we have flashcards, questions, and tools to help you learn medicine. \ No newline at end of file diff --git a/server/evaluate/reference_texts/ref_sample_1.txt b/server/evaluate/reference_texts/ref_sample_1.txt deleted file mode 100644 index 341d622f..00000000 --- a/server/evaluate/reference_texts/ref_sample_1.txt +++ /dev/null @@ -1,1544 +0,0 @@ -CEO of Google cloud and Alexander Wang - -CEO and founder of scale AI Thomas - -joined Google in November 2018 as the - -CEO of Google Cloud prior to Google - -Thomas spent 22 years at Oracle where - -most recently he was president of - -product development before that Thomas - -worked at McKinsey as a business analyst - -and engagement manager his nearly 30 - -years of experience have given him a - -deep knowledge of engineering Enterprise - -relationships and Leadership of large - -organizations Thomas's degrees include - -an MBA in administration and management - -from Stanford University as an RJ Miller - -scholar and a bsee in electrical - -engineering and computer science from - -Princeton University where he graduated - -summa laude Thomas serves as a - -member of the Stanford Graduate School - -of Business advisory Council and - -Princeton University School of - -Engineering advisory Council please - -welcome to the stage Thomas kurian and - -Alexander Wang - -[Music] - -this is a super exciting conversation - -thanks for uh thanks so much for being - -here Thomas thank you for having me you - -all just came off of uh your incredible - -Google Cloud next conference Where You - -released a wide variety of functionality - -and features and sort of new products - -across artificial intelligence but also - -across the entire sort of cloud - -ecosystem do you want to just first by - -walking through uh first start by - -walking through uh all the innovations - -that that you sort of released and uh - -and what you're excited about when it - -comes to Google Cloud - -you know our vision is super simple if - -you look at - -what smartphones did for a consumer you - -know they took - -a computer - -an internet browser a communication - -device and a camera and made it so that - -it's in everybody's pocket so it really - -brought computation to every person - -we feel that you know our our what we're - -trying to do is take all the - -technological innovation that Google's - -doing - -but make it super simple so that - -everyone can consume it and so that - -includes our global data center - -footprint - -all the new types of hardware and - -large-scale systems we work on - -the software that we're making available - -for people to do high-scale computation - -tools for data processing tools for - -cyber security - -tools for machine learning but make it - -so simple that everyone can use it - -and every step that we do to simplify - -things for people we think adoption can - -grow and so that's a lot of what we've - -done these last three four years and we - -made a number of announcements that next - -in in machine learning and AI in - -particular you know we look at our work - -as four elements - -how we take our large-scale compute - -systems that we're building for AI and - -how we make that available to everybody - -second what we're doing with the - -software stacks on top of it things like - -Jacks and other things and how we're - -making those available to everybody - -third is advances because different - -people have different levels of - -expertise some people say I need the - -hardware to build my own large language - -model or algorithm other people say look - -I really need to use a building block - -you guys give me so third is we've done - -a lot with automl and we announced new - -capability for image video and - -translation to make it available to - -everybody and then lastly we're also - -building completely packaged solutions - -for some areas and we announced some new - -stuff so it was a busy conference but - -you know lots of exciting stuff going on - -yeah it's incredible I mean I want to - -zoom out for a second to start with - -which is that this is obviously not your - -first time taking and packaging new - -technology breakthroughs for for the - -Enterprise you know both in your time at - -Oracle and now CEO of Google Cloud this - -is something that you've been doing for - -quite some time now when you sort of - -Zoom all the way out what do you think - -are some of the things that have some of - -of your principles or some of your - -thoughts and enabling these - -technological breakthroughs and actually - -enabling the Enterprise with them and - -what are sort of the key insights that - -you have there thank you a lot of the - -work so first of all we've really built - -out the organization the last three - -years we've seen a huge ramp up in our - -business credit to all the people you - -know who joined us - -at one point over 70 percent of - -organizations that joined during covid - -so they hadn't met anybody they couldn't - -meet their managers but they all did an - -amazing job together - -the adoption of Technology by companies - -and I'll give you just some elements - -particularly in the application of AI in - -different domains that we've seen - -we work with a large financial - -institution in Hong Kong and Shanghai - -bank which uses our machine learning to - -detect fraud - -you know fraud detection and banking - -there's a lot of false positives which - -makes it hard to really you know to it's - -very expensive for people doing - -something called anti-money laundering - -and our AI algorithms are really able to - -be super precise on detection - -explainability is a critical thing there - -right so people ask why did you why did - -you approve why did you flag this one - -and not that one because Regulators are - -involved so explainability becomes a big - -deal - -um we helped we helped uh Renault for - -example monitor all of the factories - -they process roughly a billion data sets - -every day obviously humans can process - -that - -but making it super simple to and you - -guys had given all your expertise in - -labeling and other things you would get - -a sense Factory floor data is not clean - -data and so you have to actually clean - -imagine doing a billion data sets into - -an environment every single day you have - -to get the data pipelines really good - -and so a lot of Technology work happens - -to make that possible for companies - -um third is if you shop at Ikea for - -example behind Ikea is systems it's our - -recommendation system - -and the way that people shop for - -furniture - -and products is not the same in all - -countries and so how are you able to one - -deal with the benefits you get from a - -global model - -but also take contextually the specific - -elements in each country because people - -have different buying habits those are - -all things that we've learned applying - -our AI in different contexts in - -different parts of the world yeah you - -know you've you've you're uh you sort of - -uh glossed over this but you've LED - -since you took over at Google Cloud just - -a meteoric growth of the of the platform - -you know I think in the past few years - -you've tripled your sales force and - -ending last year you obviously can't - -comment on this but ended last year at I - -believe 20 billion uh of annual revenue - -which is which is incredible and and - -this incredible growth Journey what do - -you attribute your success to and how do - -you think you've been able to to drive - -to such an incredible incredible growth - -and success - -you know from our point of view every - -every industry virtually in the world is - -now becoming a software powered you know - -technology industry right if you talk to - -automobile companies they're - -increasingly their vehicles are more - -about software than mechanical systems - -if you talk to telecommunications - -companies their networks are Commodities - -unless they can make them platforms to - -deliver applications so they need new - -ways to slice manage the network - -if you look at banks at the end of the - -day they're about all the products of a - -bank are data and all of that becomes - -how do you differentiate in the value - -you're delivering clients through a - -digital medium because increasingly I'm - -sure all of you look at yourselves and - -go when was the last time I went to a - -branch of a bank so a lot of our work - -has been pushing the Technology - -Innovation really far but bringing that - -technology super easily to people in - -different Industries and given the - -demand that people have for a hey I - -really want I need the technology to - -help me power my industry that the - -change I'm seeing in my industry the - -more accessible we can make it the - -easier and the faster we get adoption - -and our approach has been to be - -completely open and when I say - -completely open we offer every part of - -the stack that we have from the hardware - -and network to the software abstractions - -above - -two things that are more packaged - -because different organizations have - -different levels at which they have - -expertise and want to adopt technology - -yeah yeah I mean it's been I mean it's - -been obviously incredible you know going - -back to AI for a second Google Google - -obviously is is an early mover in Ai and - -Google cloud has also been through you - -know or starting with tensorflow and - -vertex Ai and automl and so many - -incredibly Innovative Technologies and - -uh ai's been obviously kind of a a - -buzzword for some time now within the - -industry and and - -um you know I think we see this and you - -see as well the adoption has maybe been - -a bit slower than we would have expected - -until now what do you think have been - -the barriers to Greater levels of AI - -adoption greater levels of of - -Enterprises seeing value from Ai and and - -what do you think the future holds - -so we work with a huge number of - -companies doing work having them adopt - -AI - -a lot of the lessons we've seen and - -observed from it - -are the barriers to adoption are rarely - -about the algorithm itself right it's - -often the barriers to adoption about - -very different things so when we work - -with customers in many many Industries - -take retail as an example - -and you think of a very mundane example - -like recommendations to make product - -discovery on the web much easier for - -their own products the biggest challenge - -is standardizing the meaning of the - -product and the catalog Because unless - -you have a standardized definition of - -the products and the data behind the - -algorithm is clean it's super hard to - -actually get a recommendation and so in - -the work we did with h m for example or - -at Macy's or at Ikea or Bloomingdale's a - -huge number of these Brands the big part - -of the program is actually how do you - -label and clean the data up front and - -standardize it before you get into the - -algorithmic phase so that's one part of - -things we see - -second part is for large organizations - -to adopt AI they have to need to - -integrate the the the results of the - -algorithm back into their core processes - -so you know practical example we work - -with Angie Angie is a large large - -electric producer electricity and power - -producer in Europe - -they are probably the one of the largest - -renewable energy producer in the world - -they use wind farms - -one of the things they really struggled - -with was how do you predict how much - -wind is going to be there three days - -from now because the power grid requires - -that prediction in order to capacity - -plan how much power is going into the - -grid so they work with us and they use - -our AI to do that - -but that needs to be tied into how - -they're telling the rest of the power - -sources that work on the grade hey if - -this much wind is coming in here's all - -the other sources need to generate so - -tying it back in is not as simple as - -people think and so a lot of time is - -that - -the third on the people side there's - -change management you go through to get - -people to trust the algorithm so one of - -the things we've done work with many - -banks particularly during the pandemic - -when the government issued small - -business loans - -there was a giant bottleneck in being - -able to get loans out to individual - -consumers - -and frankly because the banks didn't - -want to bring a huge Army of loan - -officers in - -they had to use software and algorithms - -to process it now the challenge people - -had is they needed to trust the - -algorithm was being fair in saying yes - -to some and no to others - -and that it would mirror for example the - -recommendations that their best mortgage - -you know Bankers would do right just as - -a loan officers would do so it gave them - -the benefit of scale because we - -processed literally millions and - -millions of mortgages through our - -technology but it required them to get - -comfortable that things like fairness - -and other things were working so often - -when people look at AI they think it's a - -skills issue there's certainly a skill - -issue involved there's not enough talent - -in the ecosystem but things are getting - -easier and easier as the models get more - -and more sophisticated often people - -forget about these other issues that are - -important in getting adoption yeah I - -mean you're uh you're preaching in the - -choir when you mentioned the the data - -challenges that all these Enterprises uh - -face and uh and how critical that is to - -getting AI working in the early days - -um you know one of one of the things - -that I think is interesting about Google - -Cloud strategy is that you really have - -products that different layers of sort - -of the stack and different layers of of - -um you know closest to the bare metal - -all the way up to these package - -Solutions you know I'm with in what way - -do you think that the Enterprise world - -and even the the sort of broader - -business world is going to adopt these - -AI Technologies do you think that the - -end stated that a lot of them are using - -your lower level more infrastructure uh - -products or do you think that many of - -them are going to adopt Solutions how do - -you think this plays out over the the - -next few years - -so we offer four layers of technology - -for people - -there's a set of people who say look I - -just need your - -you know computational infrastructure - -your large systems we build something - -called tensor Processing Unit which is - -our large scale systems we're also - -working with Nvidia to build a really - -high scale gpu-based system - -but many people some some customers say - -look I just need access to that and we - -make that available because the tpus are - -what we use within Google and we make - -that available along with the - -compilation software to optimize models - -on the tpus - -take as an example of LG you know the - -the Korean company that makes appliances - -their team has built a a large I mean - -multi-hundred billion parameter model - -because they wanted to make that a way - -that people can interact with appliances - -without having to press buttons on them - -they so they built a model they said I - -just need access to your infrastructure - -so that's one way we offer capability - -a second level is people say look I - -really don't need access to the raw - -infrastructure itself what I need is the - -ability to build models using your - -platform and so we offer a platform - -called vertex and people build models - -and push them using our machine learning - -platform and there are many many - -organizations in logistics and financial - -services in retail and others who build - -their own models on top of the platform - -the third is to make things even easier - -we've taken some of the core pieces - -translation documents - -uh image processing video - -and we've said we can offer an automl - -based solution which further simplifies - -how you use our platforms - -and so for example translation we have a - -capability to handle translation in 135 - -languages - -one of the important things that people - -ask when they go to many languages is - -the if you look at the data sets that - -are used to - -to train models - -they are primarily there's a large set - -in English because you have the whole - -internet is primarily in a very small - -number of languages but once you get to - -more narrow languages for instance - -Swahili or some of the African languages - -or even in Asia there are many languages - -even from where I grew up in India there - -are languages that are not as widely - -represented on the internet can you - -model in Translation provide equivalent - -Fidelity in sparse languages because - -it's always important to those people - -who only understand that language that - -they get a high fidelity result - -so we built something called translation - -Hub and it's being used in very mundane - -places but with extraordinary impact for - -example when people announce covet - -guidelines or recently monkey pox for - -example which is another thing they need - -to translate in many many languages and - -normally the process would take a long - -time - -we have movie studios for example in a - -in a different example saying hey when - -we launch a movie - -uh we have a high fidelity set of - -languages we're actually going to hold - -the movie up and show that people do it - -but for the long tail we just need - -captioning uh we're not necessarily - -going to do voice dubbing we're going to - -do captioning and they use our - -translation solutions to go to that even - -within companies Avery Dennison for - -example uses it to translate all their - -instruction manuals into many languages - -for their technicians - -and then lastly in some places there are - -companies like retailers who tell us - -look a handful of the largest retailers - -May build their own software teams - -but some of us who are small Merchants - -we're not software companies and telling - -us you got to be a software company to - -use AI is not fair - -so for some Industries we actually build - -fully packaged Solutions if you if you - -call many telephone companies their - -contact center behind it sits a voice - -agent - -and the rationale behind that was super - -simple when a new smartphone launches - -like an iPhone or a pixel typically in - -the morning of the launch some of these - -contact centers get three four million - -calls in an hour - -and it's hard to hire that many agents - -to handle the phones so we said why - -wouldn't software be able to handle it - -we then evolved it so that the natural - -language interface can become actually - -the workflow for these organizations but - -that's a much more of a package solution - -so that telephone companies don't have - -to have armies of data scientists to do - -it so our work spans all of these - -because people have different needs and - -we find that you know as you improve the - -maturation of this and you make it more - -easy for people to adopt it you will get - -broader proliferation and Adoption of AI - -as a whole - -yeah you know you walk through so many - -different use cases and so many - -applications of the technology I imagine - -one um and they're so desperate you know - -everywhere from uh you know fraud - -detection to translation to sort of - -translation of manuals you know there's - -such a wide array of use cases how do - -you you all at Google Cloud think about - -helping businesses understand what what - -is AI good for you know what what can - -they use AI for you know there's there's - -obviously such a wide - -um uh diversity of different use cases - -but what at a framework level do you do - -you tell them like how can I use AI - -within my business - -it's a really good question I mean a lot - -of our work actually comes from clients - -asking us now and that's actually - -an encouraging thing because you know - -see from our point of view some simple - -things how many of you believe in a few - -years time there's going to be - -intelligent software and - -non-intelligence software - -right I mean nobody would say in three - -four years time we're going to write - -software that has not powered in some - -form of fashion by AI so you know and - -most companies actually it's really - -encouraging to see that they look at - -domain problems they're having and say - -for instance I used to do it using a - -rules engine which is an older model for - -defining kind of workflow within - -organizations can you apply AI to do it - -in a new way - -um I used to do this in a specific way I - -heard about image recognition but you - -know one example really fun or - -interesting one U.S Navy - -um when you have corrosion on the base - -of ships the old way was to lift it into - -Dry Dock and take a look at it if you've - -ever seen one of these ships you can - -imagine lifting into dry dock is not an - -easy thing so they said can we fly a - -drone with Geo camera image recognition - -around it and detect corrosion and it's - -so the what we've seen is that as you - -lift up the capability where image audio - -text Etc all these forms of input - -can be processed extremely accurately - -most customers start figuring it out and - -so they call us with most of our work - -has come from customers calling us - -saying hey I have this need can I apply - -AI to it and so we talk to them about - -how and when it makes sense to use AI - -but we also talk to them about the - -consequences if the models are not you - -know handling things like skew in the - -data how do you ensure that for example - -you're treating fairness properly how do - -you ensure that the model is safe etc - -etc - -yeah you know I think uh it's it's - -exciting I mean all the use cases the - -variety is is incredibly exciting it's - -cool that these customers are coming to - -you - -um directly with many of them what what - -is you know again kind of uh thinking - -bigger picture what is machine learning - -and AI mean for Google Cloud on the - -whole over the next call it five ten - -years - -so we feel that the boundary of what - -machine learning and what AI can do will - -change over time - -uh when it's started it was about doing - -what you know what we would call - -assistive things - -assist if things are where a human being - -is able to do it but the computer - -assists the human being in some ways to - -do it better right so common examples - -people talk about is hey you're a doctor - -or radiologist - -you used to look at x-ray images now a - -computer is going to look at it and - -detect tumors but it's assisting you to - -find something that you may have done - -another way - -so that's the first phase and a lot of - -the work we see is is primarily in that - -phase today - -the the second phase is to do something - -where you couldn't do it with a human - -because the quantity of data you need to - -process or the amount of people you need - -would be just far too significant and so - -the machine is doing something that - -humans couldn't do but it's still an - -incremental element on top of what - -humans could do themselves - -the third phase I think is where we - -think generative AI for example goes - -because it's about enabling people to - -express themselves in a different way - -right and to assist them in - -expressiveness so I'll give you a - -practical example a lot of you probably - -use tools uh like slides and things like - -that in your day-to-day job right - -PowerPoint was invented a long time ago - -and was really just about drawing things - -you know I've got a 14 year old and so - -if you look at the younger generation - -if you look at what slides were they - -were really tools to help people draw - -and then to take what was on the slide - -projector and present it - -then P you know the the younger - -generation says hey I don't want to draw - -things that's like really old-fashioned - -I'm going to go to the internet and copy - -images right because they when they do - -class projects They're copying images - -into the slides - -and then you know as as people observe - -you know on the social media environment - -people going from text which may have - -been Facebook to short to images which - -is Instagram to short video Tick Tock - -people say hey why wouldn't we able to - -record short video and we use that as a - -mechanism to share but recording short - -video is still capturing the real world - -through the lens of the camera - -what people want is a more expressive - -way of saying I have an idea can I - -translate it and it may not be something - -I can capture imagine a kid in - -California in a school saying I want to - -capture how - -the landscape and outside of Paris and - -Francis right now I think they need to - -be able to generate some of the ideas - -that they couldn't capture by physically - -being there and so we're working on all - -of this and we're bringing some of these - -into our products to change what people - -could possibly do through the - -application of AI so they improve - -expressiveness for people - -and so every boundary as the technology - -gets more sophisticated we think it - -moves from just assistance to assistance - -on things that human beings may not have - -been able to just linearly do - -to now things like expressiveness which - -is a very different capability than - -people could actually do themselves - -uh yeah it's an it's I mean all this is - -very is obviously incredibly exciting - -and we're all watching it happen in real - -time you know there's an artist uh who - -actually described the these sort of - -image generation models as he was he - -sort of said like you kind of have to - -think about like a like a camera like - -it's a new tool that allows you to - -create fundamentally new uh you know - -forms of art that's right yeah and not - -just one medium of art right because if - -you look in the past people said you - -were a painter you were a sculpture - -you're a musician and now these - -Technologies allow you to blend all of - -it as a form of expressiveness yeah you - -know the the last question I have for - -you is you you know you obviously sit - -down with many of the sort of leading - -CEOs and Business Leaders of many of the - -the sort of largest uh organizations in - -the world and I'm sure one thing that is - -on many of their minds is sort of um as - -AI technology develops and it continues - -to progress is potential disruption that - -might come from from artificial - -intelligence what sort of how do you - -approach that conversation what's sort - -of your advice to these these Business - -Leaders who are looking at this powerful - -new technology and thinking about what - -that might mean for for the businesses - -and and the business landscape - -when we talk to CEOs I mean the biggest - -things we talk to them about number one - -you know uh productivity in the long - -term - -productivity has always been the primary - -driver of improving you know both - -company productivity meaning their own - -companies as well as societal you know - -benefit things like affluence of a - -society Etc and the means and equality - -of distribution of income to people - -across all Spectrum Society eventually - -the most important metric and you can - -look at any economics textbook is - -productivity - -uh software and technology has probably - -been the biggest Boon of productivity - -over the last 30 40 years - -this is the next evolution of that and - -so we always say if you approach it the - -right way for example labor shortages - -are going on right now - -the biggest potential benefit is the - -application of some of these platforms - -like AI to doing that - -the second - -with any technological generation - -Revolution like artificial intelligence - -but if you went back in time and looked - -at the Industrial Revolution Etc they're - -always During the period of transition - -anxiety about the consequences of that - -technology - -and it doesn't mean that technology by - -itself is good or bad it's the - -application of the technology that's - -good or bad - -so it's incumbent upon both the - -technology providers and the users of - -the technology to ensure that the - -negative consequences of it are managed - -properly right - -the obvious example is for instance if - -you look at a very simple thing image - -recognition - -image recognition can help doctors find - -tumors way better than having the best - -radiographer - -it's assistive in that context and it's - -like helping people with a better - -quality microscope than they had before - -object recognition is helping people - -find for example people who are in the - -ocean much more accurately so the Coast - -Guard can rescue them - -at the same time being able to use a - -camera and say that's Thomas kurian - -has uh you know a lot of potential - -negative consequences and so as a - -provider of Technology we at Google have - -chosen not to do that third part but we - -also tell companies it's important not - -just to say this is what's regulatory - -Allowed by the legal framework because - -law in many countries is not yet keeping - -up with how fast AI Technologies is - -moving but to take the responsibility as - -a Company CEO to say here's what I'd be - -comfortable with and here's what I won't - -be comfortable with yeah well Thomas - -thank you so much for uh such an - -incredible conversations I think uh I - -think I'm I'm very heartened to hear all - -the incredible work that Google cloud is - -doing to make artificial intelligence - -accessible to you know the entire - -business world and all of every - -Enterprise around the globe and uh I'm - -so excited that you're able to join us - -thank you so much thank you so much for - -having me - -[Music] - diff --git a/server/evaluate/reference_texts/ref_sample_2.txt b/server/evaluate/reference_texts/ref_sample_2.txt deleted file mode 100644 index f407f205..00000000 --- a/server/evaluate/reference_texts/ref_sample_2.txt +++ /dev/null @@ -1,620 +0,0 @@ -Technologies ticker symbol w-e-l-l on - -the TSX recently reported its 2023 q1 - -results beating the streets consensus - -estimate for revenue and adjusted ebitda - -and in a report issued this week Raymond - -James analyst said quote we're impressed - -by Wells capacity to drive powerful - -growth across its diverse business units - -in the absence of M A joining me today - -is CEO Hamed chabazi to look at what's - -next for well health good to see you sir - -how are you great to see you Richard - -thanks very much for having me great to - -have you uh congratulations on your 17th - -consecutive quarter of record Revenue - -can you share some insights into what's - -Driven these results historically and in - -the past quarter as well - -yeah thank you we we're very excited - -about our uh q1 2023 results and as you - -mentioned uh we've had a long you know - -successful uh string of of uh you know - -continued growth and record growth - -um we also had accelerating organic - -growth and I think um a big part of the - -success of our franchise here is the - -incredibly sticky and predictable - -Revenue that we have you know well over - -90 of our business is either highly - -reoccurring as in uh the you know highly - -predictable uh results of our two-sided - -network of patients and providers or - -truly recurring as in scheduled or - -subscribed revenues and this allows us - -to essentially make sure that that uh - -you know we're on track it obviously you - -know like any other business things - -happen uh and sometimes it's hard to - -meet those results but what's really - -being unique about our platform is we do - -have exposure to all kinds of different - -aspects of healthcare you know we have - -Prime primary care and Specialized Care - -on both sides of the Border in the US - -and Canada so we have exposure to - -different types of business models we - -have exposure to the U.S payer Network - -which has higher per unit economics than - -Canada and of course the stability and - -uh and and sort of higher Fidelity uh - -kind of Collections and revenue cycle - -process that Canada has over the United - -States where you don't have to kind of - -deal with all of that uh at that payment - -noise so just a lot of I think strength - -built into the platform because of the - -diversity of different Healthcare - -businesses that we support - -and uh where do you see Well's future - -growth coming from which part of the - -business uh excites you the most right - -now yeah well look the centrifugal force - -of well is the healthcare provider and - -we exist to uh Tech enable and - -ameliorate the business of that of that - -Tech of that healthcare provider uh and - -and and that's what we're laser focused - -on and and what we're seeing is - -providers not wanting to run businesses - -anymore it's very simple and so we have - -a digital platform and providers can - -either acquire what they want and need - -from our digital platform and implement - -it themselves - -or they can decide that they don't want - -to run a business anymore they don't - -want to configure and manage technology - -which is becoming a bigger and bigger - -part of their world every single day and - -when we see what we've seen with that - -Dynamic is that uh is that a lot of them - -are now just wanting to work in a place - -where where all the technology is - -configured for them it's wrapped around - -them and they have a competent operating - -partner that is supporting the organ the - -the practice uh and and taking care of - -the front office in the back office so - -that they can focus on providing care - -this results in them seeing more - -patients uh and and being happier - -because you know they became doctors to - -see patients not so they can manage uh - -workers and and deal with HR issues and - -deal with labs and all that kind of - -stuff excellent and I know too that - -Acquisitions have played a key role in - -well can you share any insights into how - -the Acquisitions fit into Wells growth - -strategy - -sure in in look in 2020 and 2021 we did - -a lot of Acquisitions in 2022 we took a - -bit of a breather and we've really - -focused on integration and I think - -that's one of the reasons why you saw - -this accelerating organic growth we - -really were able to demonstrate that we - -could bring together the different - -elements of our technology platform we - -started to sell bundles we started to - -really derive Synergy uh and activate uh - -you know more sales as a result of - -selling uh all the different products - -and services with one voice with One - -Vision uh so we made it easier for - -providers to use their technology and I - -think that was a big reason uh for our - -growth now M A as you know where Capital - -allocation company we're never far from - -it and so we did continue to have you - -know tuck-ins here and there and in fact - -today uh we announced that we've - -acquired uh the Alberta operations of uh - -MCI one Health and other publicly traded - -company uh who was looking to raise - -funds to support their business we're - -very pleased with with this acquisition - -it just demonstrates our continued - -discipline these are you know great - -primary care clinics in in Canada right - -in the greater Calgary area and uh uh - -you know just allows us to grow our - -footprint in Alberta which is an - -important Province for us and it it's - -it's if you look at the price if you - -look at what we're getting uh you know - -it's just demonstrative of our continued - -uh discipline and just you know a few - -days ago at our conference call I - -mentioned uh that we had you know a - -really strong lineup of Acquisitions uh - -and you know they're starting to uh uh I - -think uh come to fruition for us - -a company on the grown-up question I you - -recently announced a new AI investment - -program last month what specific areas - -of healthcare technology or AI are you - -focusing on and what's the strategy when - -it comes to AI - -yes uh look AI as as I'm sure you're - -aware is it's become you know really uh - -an incredibly important topic in in all - -aspects of of business and and you know - -not just business socially as well - -everyone's talking about uh this this - -new breakthrough disruptive technology - -the large language models and generative - -AI - -um I mean look AI uh has been about a 80 - -year old overnight success a lot of - -people have been working on this for a - -long time generative AI is just sort of - -you know the culmination of a lot of - -things coming together and working uh - -but it is uncorked enormous uh - -Innovation and and we think that um this - -there's a very good news story about - -this in healthcare particularly where we - -were looking to look we were looking to - -unlock uh the value of of the data that - -that we all produce every single day - -um as as humans and and so we've - -established an AI investment program - -because no one company can can tackle - -all of these Innovations themselves and - -what well has done too is it's taken a - -very much an ecosystem approach by - -establishing its apps.health Marketplace - -and so we're very excited about not only - -uh allocating Capital into promising - -young AI companies that are focused on - -digital health and solving Healthcare - -problems but also giving them access to - -um you know safely and securely to our - -provider Network to our uh you know to - -to our Outpatient Clinic Network which - -is the largest owned and operated - -Network in Canada by far uh so - -um and and when these and it's it was - -remarkable when we announced this - -program we've had just in the in the - -first uh week to 10 days we've had over - -a hundred uh inbound prospects come in - -uh that that wanted to you know - -collaborate with us and again I don't - -think that's necessarily for the money - -you know we're saying we would invest a - -minimum of a quarter of a million - -dollars you know a lot of them will - -likely be higher than a quarter of a - -million dollars - -so it's not life-changing money but but - -our structural advantages and and and - -the benefits that we have in the Well - -Network those are extremely hard to come - -by uh and I think and I think uh uh - -you'll see us uh you know help some of - -these companies uh succeed and they will - -help us drive uh you know more - -Innovation to that helps the provider - -but speaking of this very interesting AI - -I know your company just launched well - -AI voice this is super interesting tell - -me what it is and the impact it could - -have on health care providers - -yeah thanks for uh asking Richard our - -providers uh are thrilled with this you - -know we've we've had a number of of of - -our own well providers testing this - -technology and it it it really feels - -like magic to them it's essentially an - -ambient AI powered scribe so it's a it's - -a service that with the consent of the - -parties involved listens to the - -conversation between a patient and - -provider and then uh essentially - -condenses that into a medically relevant - -note for the chart files uh typically - -that is a lengthy process a doctor has - -to transcribe notes then review those - -notes and make sure that uh a a a a - -appropriate medically oriented and - -structured node is is is uh prepared and - -put into the chart and that could take - -you know sometimes more than more time - -than the actual consultation uh time and - -so we believe that on average if it's - -used regularly and consistently this can - -give providers back at least a third of - -their day - -um and and it's it's just a game changer - -uh and and uh we have now gone into - -General release with this product it's - -widely available in Canada uh it has - -been integrated into our EMR which makes - -it even more valuable tools like this - -are going to start popping up but if - -they're not integrated into your - -practice management system then you have - -to kind of have data in in more than one - -place and and move that around a little - -bit which which makes it a little bit - -more difficult especially with HIPAA - -requirements and and regulations so - -again I think this is the first of many - -types of different products and services - -that allow doctors to place more - -emphasis and focus on the patient - -experience instead of having their head - -in a laptop and looking at you once in a - -while they'll be looking at you and - -speaking to their practice management - -system and I think this you know think - -about it as Alexa for for our doctors uh - -you know this this ability to speak uh - -and and have you know uh you know Voice - -driven AI assistant that does things - -like this I think are going to be you - -know incredibly helpful and valuable uh - -for for healthcare providers - -super fascinating I mean we're just - -hearing you know more about AI maybe AI - -for the first time but here you are with - -a product already on the market in the - -in the healthcare field that's going to - -be pretty attractive to be out there uh - -right ahead of many other people right - -thank you Richard thanks for that - -recognition that's been Our intention we - -we want to demonstrate that we uh you - -know that we're all in on ensuring that - -technology that benefits providers uh is - -is is accelerated and uh de-risked and - -provided uh you know um in in a timely - -way you know providers need this help we - -we have a healthcare crisis in the - -country that is generally characterized - -as a as a lack of doctors and so imagine - -if we can get our doctors to be 20 or 30 - -percent more productive through the use - -of these types of tools well they're - -going to just see more patience and and - -that's going to help all of us and uh - -and look if you step back Wells business - -model is all about having exposure to - -the success of doctors and doing our - -best to help them be more successful - -because we're in a revenue share - -relationship with most of the doctors - -that we work with and so this uh this is - -good for the ecosystem it's great for - -the provider and it's great for well as - -well super fascinating I'm Ed shabazzi - -CEO well Health Technologies ticker - -w-e-l-l great to catch up again thank - -you sir - -thank you Richard appreciate you having - -me - -[Music] - -thank you - diff --git a/server/evaluate/reference_texts/ref_sample_3.txt b/server/evaluate/reference_texts/ref_sample_3.txt deleted file mode 100644 index a589de5f..00000000 --- a/server/evaluate/reference_texts/ref_sample_3.txt +++ /dev/null @@ -1,970 +0,0 @@ -learning medicine is hard work osmosis - -makes it easy it takes our lectures and - -notes to create a personalized study - -plan with exclusive videos practice - -questions and flashcards and so much - -more try it free today - -in diabetes mellitus your body has - -trouble moving glucose which is the type - -of sugar from your blood into your cells - -this leads to high levels of glucose in - -your blood and not enough of it in your - -cells and remember that your cells need - -glucose as a source of energy so not - -letting the glucose enter means that the - -cells star for energy despite having - -glucose right on their doorstep in - -general the body controls how much - -glucose is in the blood relative to how - -much gets into the cells with two - -hormones insulin and glucagon insulin is - -used to reduce blood glucose levels and - -glucagon is used to increase blood - -glucose levels both of these hormones - -are produced by clusters of cells in the - -pancreas called islets of langerhans - -insulin is secreted by beta cells in the - -center of these islets and glucagon is - -secreted by alpha cells in the periphery - -of the islets insulin reduces the amount - -of glucose in the blood by binding to - -insulin receptors embedded in the cell - -membrane of various insulin responsive - -tissues like muscle cells in adipose - -tissue when activated the insulin - -receptors cause vesicles containing - -glucose transporter that are inside the - -cell to fuse with the cell membrane - -allowing glucose to be transported into - -the cell glucagon does exactly the - -opposite it raises the blood glucose - -levels by getting the liver to generate - -new molecules of glucose from other - -molecules and also break down glycogen - -into glucose so that I can all get - -dumped into the blood diabetes mellitus - -is diagnosed when blood glucose levels - -get too high and this is seen among 10 - -percent of the world population there - -are two types of diabetes type 1 and - -type 2 and the main difference between - -them is the underlying mechanism that - -causes the blood glucose levels to rise - -about 10% of people with diabetes have - -type 1 and the remaining 90% of people - -with diabetes have type 2 let's start - -with type 1 diabetes mellitus sometimes - -just called type 1 diabetes in this - -situation the body doesn't make enough - -insulin the reason this happens is that - -in type 1 diabetes there's a type 4 - -hypersensitivity response or a cell - -mediated immune response where a - -person's own T cells at - -the pancreas as a quick review remember - -that the immune system has T cells that - -react to all sorts of antigens which are - -usually small peptides polysaccharides - -or lipids and that some of these - -antigens are part of our own body cells - -it doesn't make sense to allow T cells - -that will attack our own cells to hang - -around until there's this process to - -eliminate them called self tolerance in - -type 1 diabetes there's a genetic - -abnormality that causes a loss of self - -tolerance among T cells that - -specifically target the beta cell - -antigens losing self tolerance means - -that these T cells are allowed to - -recruit other immune cells and - -coordinate an attack on these beta cells - -losing beta cells means less insulin and - -less insulin means that glucose piles up - -in the blood because it can't enter the - -body's cells one really important group - -of genes involved in regulation of the - -immune response is the human leukocyte - -antigen system or HLA system even though - -it's called a system it's basically this - -group of genes on chromosome 6 that - -encode the major histocompatibility - -complex or MHC which is a protein that's - -extremely important in helping the - -immune system recognize foreign - -molecules as well as maintaining self - -tolerance MHC is like the serving - -platter that antigens are presented to - -the immune cells on interestingly people - -with type 1 diabetes often have specific - -HLA genes in common with each other one - -called - -HLA dr3 and another called HLA dr4 but - -this is just a genetic clue right - -because not everyone with HLA dr3 and - -HLA dr4 develops diabetes in diabetes - -mellitus type 1 destruction of beta - -cells usually starts early in life but - -sometimes up to 90% of the beta cells - -are destroyed before symptoms crop up - -for clinical symptoms of uncontrolled - -diabetes that all sound similar our - -polyphagia glycosuria polyuria and - -polydipsia let's go through them one by - -one even though there's a lot of glucose - -in the blood it cannot get into the - -cells which leaves cells starved for - -energy so in response adipose tissue - -starts breaking down fat called - -lipolysis - -and muscle tissue starts breaking down - -proteins both of which results in weight - -loss for someone with uncontrolled - -diabetes this catabolic state leaves - -people feeling hungry - -also known as poly fascia Faiza means - -eating and poly means a lot now with - -high glucose levels that means that when - -blood gets filtered through the kidneys - -some of it starts to spill into the - -urine called glycosuria glyco surfers to - -glucose and urea the urine since glucose - -is osmotically active water tends to - -follow it resulting in an increase in - -urination or polyuria poly again refers - -to a lot and urea again refers to urine - -finally because there's so much - -urination people with uncontrolled - -diabetes become dehydrated and thirsty - -or polydipsia poly means a lot and dip - -SIA means thirst even though people with - -diabetes are not able to produce their - -own insulin they can still respond to - -insulin so treatment involves lifelong - -insulin therapy to regulate their blood - -glucose levels and basically enable - -their cells to use glucose - -one really serious complication with - -type 1 diabetes is called diabetic - -ketoacidosis or DKA to understand it - -let's go back to the process of - -lipolysis where fat is broken down into - -free fatty acids after that happens the - -liver turns the fatty acids into ketone - -bodies like Osito acetic acid in beta - -hydroxy butyrate acid a seed of acetic - -acid is a keto acid because it has a - -ketone group in a carboxylic acid group - -beta hydroxy rhetoric acid on the other - -hand even though it's still one of the - -ketone bodies isn't technically a keto - -acid since its ketone group has been - -reduced to a hydroxyl group these ketone - -bodies are important because they can be - -used by cells for energy but they also - -increase the acidity of the blood which - -is why it's called ketoacidosis and the - -blood becoming really acidic can have - -major effects throughout the body - -individuals can develop custom all - -respiration which is a deep and labored - -breathing as the body tries to move - -carbon dioxide out of the blood in an - -effort to reduce its acidity cells also - -have a transporter that exchanges - -hydrogen ions or protons for potassium - -when the blood gets acidic it's by - -definition loaded with protons that get - -sent into cells while potassium gets - -sent into the fluid outside cells - -another thing to keep in mind is that in - -addition to helping glucose enter cells - -insulin stimulates the sodium potassium - -ATPase --is which help potassium get - -into the cells and so without insulin - -more potassium stays in the fluid - -outside cells both of these mechanisms - -lead to increased potassium in the fluid - -outside cells which quickly makes it - -into the blood and causes hyperkalemia - -the potassium is then excreted so over - -time even though the blood potassium - -levels remain high over all stores of - -potassium in the body which include - -potassium inside cells starts to run low - -individuals will also have a high anion - -gap which reflects a large difference in - -the unmeasured negative and positive - -ions in the serum largely due to the - -build-up of ketoacids - -diabetic ketoacidosis can happen even in - -people who have already been diagnosed - -with diabetes and currently have some - -sort of insulin therapy - -in states of stress like an infection - -the body releases epinephrine which in - -turn stimulates the release of glucagon - -too much glucagon can tip the delicate - -hormonal balance of glucagon and insulin - -in favor of elevating blood sugars and - -can lead to a cascade of events we just - -described increased glucose in the blood - -loss of glucose in the urine loss of - -water dehydration and in parallel and - -need for alternative energy generation - -of ketone bodies and ketoacidosis - -interestingly both ketone bodies break - -down into acetone and escape as a gas by - -getting breathed out the lungs which - -gives us sweet fruity smell to a - -person's breath in general though that's - -the only sweet thing about this illness - -which also causes nausea vomiting and if - -severe mental status changes and acute - -cerebral edema - -treatment of a DKA episode involves - -giving plenty of fluids which helps with - -dehydration insulin which helps lower - -blood glucose levels and replacement of - -electrolytes like potassium all of which - -help to reverse the acidosis now let's - -switch gears and talk about type 2 - -diabetes which is where the body makes - -insulin but the tissues don't respond as - -well to it the exact reason why cells - -don't respond isn't fully understood - -essentially the body's providing the - -normal amount of insulin but the cells - -don't move their glucose transporters to - -their membrane in response which - -remember is needed for the glucose to - -get into the cells these cells therefore - -have insulin resistance some risk - -factors for insulin resistance are - -obesity lack of exercise and - -hypertension the exact mechanisms are - -still being explored for example in - -excess of adipose tissue or fat is - -thought to cause the release of free - -fatty acids in so-called edible kinds - -which are signaling molecules that can - -cause inflammation which seems related - -to insulin resistance - -however many people that are obese are - -not diabetic so genetic factors probably - -play a major role as well we see this - -when we look at twin studies as well - -we're having a twin with type-2 diabetes - -increases the risk of developing type 2 - -diabetes completely independently of - -other environmental risk factors in type - -2 diabetes since tissues don't respond - -as well to normal levels of insulin the - -body ends up producing more insulin in - -order to get the same effect and move - -glucose out of the blood - -they do this through beta cell - -hyperplasia an increased number of beta - -cells and beta cell hypertrophy where - -they actually grow in size all in this - -attempt to pump out more insulin this - -works for a while and by keeping insulin - -levels higher than normal blood glucose - -levels can be kept normal called normal - -glycemia now along with insulin beta - -cells also secrete islet amyloid - -polypeptide or amylin so while beta - -cells are cranking out insulin they also - -secrete an increased amount of amylin - -over time Emlyn builds up and aggregates - -in the islets this beta cell - -compensation though is not sustainable - -and over time those maxed out beta cells - -get exhausted and they become - -dysfunctional and undergo hypo trophy - -and get smaller as well as hypoplasia - -and die off as beta cells are lost in - -insulin levels decrease glucose levels - -in the blood start to increase in - -patients develop hyperglycemia which - -leads to similar clinical signs that we - -mentioned before like Paul aphasia - -glycosuria polyuria polydipsia but - -unlike type 1 diabetes there's generally - -some circulating insulin in type 2 - -diabetes from the beta cells that are - -trying to compensate for the insulin - -resistance this means that the insulin - -glucagon balances such that diabetic - -ketoacidosis does not usually develop - -having said that a complication called - -hyperosmolar hyperglycemic state or HHS - -is much more common in type 2 diabetes - -than type 1 diabetes and it causes - -increased plasma osmolarity due to - -extreme dehydration and concentration of - -the blood to help understand this - -remember that glucose is a polar - -molecule that cannot passively diffuse - -across cell membranes which means that - -it acts as a solute so when levels of - -glucose are super high in the blood - -meaning it's a hyperosmolar State water - -starts to leave the body cells and enter - -the blood vessels leaving the cells were - -relatively dry in travailed rather than - -plump and juicy blood vessels that are - -full of water lead to increased - -urination and total body dehydration and - -this is a very serious situation because - -the dehydration of the body's cells and - -in particular the brain can cause a - -number of symptoms including mental - -status changes in HHS you can sometimes - -see mild ketone emia and acidosis but - -not to the extent that it's seen in DKA - -and in DKA you can see some hyper - -osmolarity so there's definitely overlap - -between these two syndromes - -besides type 1 and type 2 diabetes there - -are also a couple other subtypes of - -diabetes mellitus gestational diabetes - -is when pregnant women have increased - -blood glucose which is particularly - -during the third trimester although - -ultimately unknown the cause is thought - -to be related to pregnancy hormones that - -interfere with insulins action on - -insulin receptors also sometimes people - -can develop drug-induced diabetes which - -is where medications have side effects - -that tend to increase blood glucose - -levels the mechanism for both of these - -is thought to be related to insulin - -resistance like type 2 diabetes rather - -than an autoimmune destruction process - -like in type 1 diabetes diagnosing type - -1 or type 2 diabetes is done by getting - -a sense for how much glucose is floating - -around in the blood and has specific - -standards that the World Health - -Organization uses very commonly a - -fasting glucose test is taken where the - -person doesn't eat or drink except the - -water that's okay for a total of eight - -hours and then has their blood tested - -for glucose levels levels of 100 - -milligrams per deciliter to 120 - -five milligrams per deciliter indicates - -pre-diabetes and 126 milligrams per - -deciliter or higher indicates diabetes a - -non fasting a random glucose test can be - -done at any time with 200 milligrams per - -deciliter or higher being a red flag for - -diabetes another test is called an oral - -glucose tolerance test where person is - -given glucose and then blood samples are - -taken at time intervals to figure out - -how well it's being cleared from the - -blood the most important interval being - -two hours later levels of 140 milligrams - -per deciliter to 199 milligrams per - -deciliter indicate pre-diabetes - -and 200 or above indicates diabetes - -another thing to know is that when blood - -glucose levels get high the glucose can - -also stick to proteins that are floating - -around in the blood or in cells so that - -brings us to another type of test that - -can be done which is the hba1c test - -which tests for the proportion of - -hemoglobin in red blood cells that has - -glucose stuck to it called glycated - -hemoglobin hba1c levels of 5.7% 26.4% - -indicate pre-diabetes - -and 6.5 percent or higher indicates - -diabetes this proportion of glycated - -hemoglobin doesn't change day to day so - -it gives a sense for whether the blood - -glucose levels have been high over the - -past two to three months finally we have - -the c-peptide test which tests for - -byproducts of insulin production if the - -level of c-peptide is low or absent it - -means the pancreas is no longer - -producing enough insulin and the glucose - -cannot enter the cells - -for type one diabetes insulin is the - -only treatment option for type 2 - -diabetes on the other hand lifestyle - -changes like weight loss and exercise - -along with a healthy diet and an oral - -anti-diabetic medication like metformin - -in several other classes can sometimes - -be enough to reverse some of that - -insulin resistance and keep blood sugar - -levels in check however if oral - -anti-diabetic medications fail type 2 - -diabetes can also be treated with - -insulin something to bear in mind is - -that insulin treatment comes with a risk - -of hypoglycemia especially if insulin is - -taken without a meal symptoms of - -hypoglycemia can be mild like weakness - -hunger and shaking but they can progress - -to a loss of consciousness in seizures - -in severe cases in mild cases drinking - -juices or eating candy or sugar might be - -enough to bring blood sugar up but in - -severe cases intravenous glucose should - -be given as soon as possible - -the FDA has also recently approved - -intranasal glucagon as a treatment for - -severe hypoglycemia all right now over - -time high glucose levels can cause - -damage to tiny blood vessels while the - -micro vasculature in arterioles a - -process called hyaline - -arteriolosclerosis is where the walls of - -the arterioles develop hyaline deposits - -which are deposits of proteins and these - -make them hard and inflexible in - -capillaries the basement membrane can - -thicken and make it difficult for oxygen - -to easily move from the capillary to the - -tissues causing hypoxia - -one of the most significant effects is - -that diabetes increases the risk of - -medium and large arterial wall damage - -and subsequent atherosclerosis which can - -lead to heart attacks and strokes which - -are major causes of morbidity and - -mortality for patients with diabetes in - -the eyes diabetes can lead to - -retinopathy and evidence of that can be - -seen on a fundus copic exam that shows - -cotton-wool spots or flare hemorrhages - -and can eventually cause blindness in - -the kidneys the a ferrant and efferent - -arterioles as well as the glomerulus - -itself can get damaged which can lead to - -an F Radek syndrome that slowly - -diminishes the kidneys ability to filter - -blood over time and can ultimately lead - -to dialysis diabetes can also affect the - -function of nerves causing symptoms like - -a decrease in sensation in the toes and - -fingers sometimes called a stocking - -glove distribution as well as causes the - -autonomic nervous system to malfunction - -and that system controls a number of - -body functions - -everything from sweating to passing gas - -finally both the poor blood supply and - -nerve damage can lead to ulcers - -typically on the feet that don't heal - -quickly and can get pretty severe and - -need to be amputated these are some of - -the complications of uncontrolled - -diabetes which is why it's important to - -diagnose and control diabetes through a - -healthy lifestyle medications to reduce - -insulin resistance and even insulin - -therapy if beta cells have been - -exhausted while type 1 diabetes cannot - -be prevented type 2 diabetes can in fact - -many people with diabetes can control - -their blood sugar levels really - -effectively and live a full and active - -life without any of the complications - -thanks for watching if you're interested - -in a deeper dive on this topic take a - -look at as Moses org where we have - -flashcards questions and other awesome - -tools to help you learn medicine - -you - diff --git a/server/notebooks/Viz-experiments.ipynb b/server/notebooks/Viz-experiments.ipynb deleted file mode 100644 index 331c2061..00000000 --- a/server/notebooks/Viz-experiments.ipynb +++ /dev/null @@ -1,860 +0,0 @@ -{ - "cells": [ - { - "cell_type": "markdown", - "id": "a5ace857", - "metadata": {}, - "source": [ - "# Visualization Experiments" - ] - }, - { - "cell_type": "markdown", - "id": "9bfc569d", - "metadata": {}, - "source": [ - "Lets load the data artefacts to local memory. These files are to be downloaded from S3 as the pipeline automatically uploads them to the pre-configured S3 bucket." - ] - }, - { - "cell_type": "code", - "execution_count": 18, - "id": "edc584b2", - "metadata": {}, - "outputs": [ - { - "name": "stderr", - "output_type": "stream", - "text": [ - "\u001b[32m2023-06-23 15:01:36.558\u001b[0m | \u001b[1mINFO \u001b[0m | \u001b[36mfile_utilities\u001b[0m:\u001b[36mdownload_files\u001b[0m:\u001b[36m36\u001b[0m - \u001b[1mDownloading file df_06-23-2023_06:10:03.pkl\u001b[0m\n", - "\u001b[32m2023-06-23 15:01:38.450\u001b[0m | \u001b[1mINFO \u001b[0m | \u001b[36mfile_utilities\u001b[0m:\u001b[36mdownload_files\u001b[0m:\u001b[36m36\u001b[0m - \u001b[1mDownloading file mappings_06-23-2023_06:10:03.pkl\u001b[0m\n", - "\u001b[32m2023-06-23 15:01:39.179\u001b[0m | \u001b[1mINFO \u001b[0m | \u001b[36mfile_utilities\u001b[0m:\u001b[36mdownload_files\u001b[0m:\u001b[36m36\u001b[0m - \u001b[1mDownloading file transcript_with_timestamp_06-23-2023_06:10:03.txt\u001b[0m\n" - ] - } - ], - "source": [ - "from file_utilities import download_files\n", - "import pickle\n", - "\n", - "# Download files from S3 bucket. You can download multiple files at a time by passing a list of names\n", - "# files_to_download = [\"df.pkl\",\n", - "# \"mappings.pkl\",\n", - "# \"transcript_timestamps.txt\"]\n", - "\n", - "# set the timestamp \n", - "timestamp = \"06-23-2023_06:10:03\"\n", - "\n", - "# df,mappings,transcript_timestamps file names\n", - "df_file_name = \"df_\" + timestamp + \".pkl\"\n", - "mappings_file_name = \"mappings_\" + timestamp + \".pkl\"\n", - "transcript_file_name = \"transcript_with_timestamp_\" + timestamp + \".txt\"\n", - "\n", - "\n", - "files_to_download = [df_file_name,\n", - " mappings_file_name,\n", - " transcript_file_name] \n", - "download_files(files_to_download)" - ] - }, - { - "cell_type": "code", - "execution_count": 19, - "id": "5027fe25", - "metadata": {}, - "outputs": [], - "source": [ - "# Download spacy model for the first time\n", - "import nltk\n", - "import spacy\n", - "from nltk.corpus import stopwords\n", - "\n", - "nltk.download('punkt', quiet=True)\n", - "nltk.download('stopwords', quiet=True)\n", - "spaCy_model = \"en_core_web_md\"\n", - "nlp = spacy.load(spaCy_model)\n", - "spacy_stopwords = nlp.Defaults.stop_words\n", - "STOPWORDS = set(spacy_stopwords).union(set(stopwords.words('english')))" - ] - }, - { - "cell_type": "markdown", - "id": "8abc435d", - "metadata": {}, - "source": [ - "## Example template 1" - ] - }, - { - "cell_type": "markdown", - "id": "2b1a4834", - "metadata": {}, - "source": [ - "## Scatter plot of transcription with Topic modelling" - ] - }, - { - "cell_type": "code", - "execution_count": 23, - "id": "55a75dcf", - "metadata": {}, - "outputs": [ - { - "data": { - "text/html": [ - "
\n", - "\n", - "\n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - " \n", - "
timestamptextts_to_topic_mapping_top_1ts_to_topic_mapping_top_2
0(0.0, 12.36)this . Okay , yeah , so it looks like I am re...TAMFounders
1(12.36, 25.76)because Goku needs that for the audio plus the...FoundersTAM
2(25.76, 30.32)the rest of the team did . So I want to just ...FoundersAGENDA
3(30.32, 35.52)then we can ask questions or how do you want t...TAMFounders
4(35.52, 49.56)introduction . So what I , it all started wit...FoundersTAM
...............
554(3323.0, 3326.56)It 's crazy . But definitely with theFoundersTAM
555(3326.56, 3332.24)local models , we have n't found a way to work...FoundersTAM
556(3332.24, 3337.2)if you 'd have 90 minutes of audio to transfer...TAMFounders
557(3338.32, 3344.4)We actually have a preprocessor to resolve wha...FoundersTAM
558(3344.4, None)there 's still some struggles on the local mod...FoundersTAM
\n", - "

559 rows × 4 columns

\n", - "
" - ], - "text/plain": [ - " timestamp text \\\n", - "0 (0.0, 12.36) this . Okay , yeah , so it looks like I am re... \n", - "1 (12.36, 25.76) because Goku needs that for the audio plus the... \n", - "2 (25.76, 30.32) the rest of the team did . So I want to just ... \n", - "3 (30.32, 35.52) then we can ask questions or how do you want t... \n", - "4 (35.52, 49.56) introduction . So what I , it all started wit... \n", - ".. ... ... \n", - "554 (3323.0, 3326.56) It 's crazy . But definitely with the \n", - "555 (3326.56, 3332.24) local models , we have n't found a way to work... \n", - "556 (3332.24, 3337.2) if you 'd have 90 minutes of audio to transfer... \n", - "557 (3338.32, 3344.4) We actually have a preprocessor to resolve wha... \n", - "558 (3344.4, None) there 's still some struggles on the local mod... \n", - "\n", - " ts_to_topic_mapping_top_1 ts_to_topic_mapping_top_2 \n", - "0 TAM Founders \n", - "1 Founders TAM \n", - "2 Founders AGENDA \n", - "3 TAM Founders \n", - "4 Founders TAM \n", - ".. ... ... \n", - "554 Founders TAM \n", - "555 Founders TAM \n", - "556 TAM Founders \n", - "557 Founders TAM \n", - "558 Founders TAM \n", - "\n", - "[559 rows x 4 columns]" - ] - }, - "execution_count": 23, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "\n", - "df = pd.read_pickle(df_file_name)\n", - "df" - ] - }, - { - "cell_type": "markdown", - "id": "a795137e", - "metadata": {}, - "source": [ - "Change the values of \"category\", \"category_name\" to one agenda topic and change the value of \"not_category_name\" and see different plots." - ] - }, - { - "cell_type": "code", - "execution_count": 25, - "id": "43e01074", - "metadata": {}, - "outputs": [], - "source": [ - "import pandas as pd\n", - "import scattertext as st\n", - "\n", - "df = pd.read_pickle(df_file_name)\n", - "\n", - "def plot_topic_modelling_and_word_to_sentence_search(df, cat_1, cat_1_name, cat_2_name):\n", - " df = df.assign(parse=lambda df: df.text.apply(st.whitespace_nlp_with_sentences))\n", - "\n", - " corpus = st.CorpusFromParsedDocuments(\n", - " df, category_col='ts_to_topic_mapping_top_1', parsed_col='parse'\n", - " ).build().get_unigram_corpus().remove_terms(STOPWORDS, ignore_absences=True).compact(st.AssociationCompactor(2000))\n", - " \n", - " html = st.produce_scattertext_explorer(\n", - " corpus,\n", - " category=cat_1, category_name=cat_1_name, not_category_name=cat_2_name,\n", - " minimum_term_frequency=0, pmi_threshold_coefficient=0,\n", - " width_in_pixels=1000,\n", - " transform=st.Scalers.dense_rank\n", - " )\n", - " open('./new_viz_' + timestamp + '.html', 'w').write(html)\n", - "\n", - "plot_topic_modelling_and_word_to_sentence_search(df,\n", - " cat_1=\"Founders\",\n", - " cat_1_name=\"Founders\",\n", - " cat_2_name=\"TAM\")\n", - "\n", - "# once you are done, check the generated HTML file\n" - ] - }, - { - "cell_type": "markdown", - "id": "e9994c87", - "metadata": {}, - "source": [ - "## Example template 2" - ] - }, - { - "cell_type": "markdown", - "id": "35c4f7fd", - "metadata": {}, - "source": [ - "## Time driven Insights" - ] - }, - { - "cell_type": "code", - "execution_count": 26, - "id": "7cdcd66f", - "metadata": {}, - "outputs": [], - "source": [ - "mappings = pickle.load(open(mappings_file_name, \"rb\"))\n", - "timestamp_to_topic_first_match = mappings[0]\n", - "timestamp_to_topic_second_match = mappings[1]\n", - "topic_to_timestamp_first_match = mappings[2]\n", - "topic_to_timestamp_second_match = mappings[3]" - ] - }, - { - "cell_type": "code", - "execution_count": 27, - "id": "11221022", - "metadata": {}, - "outputs": [ - { - "data": { - "image/png": "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\n", - "text/plain": [ - "
" - ] - }, - "metadata": { - "needs_background": "light" - }, - "output_type": "display_data" - }, - { - "data": { - "image/png": "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\n", - "text/plain": [ - "
" - ] - }, - "metadata": { - "needs_background": "light" - }, - "output_type": "display_data" - } - ], - "source": [ - "import collections \n", - "import seaborn as sns\n", - "import matplotlib.pyplot as plt\n", - "\n", - "def plot_time_spent_for_topic(mapping, order):\n", - " topic_times = collections.defaultdict(int)\n", - " for key in mapping.keys():\n", - " if key[1] is None or key[0] is None:\n", - " continue\n", - " duration = key[1] - key[0]\n", - " topic_times[mapping[key]] += duration\n", - " \n", - " keys = list(topic_times.keys())\n", - " vals = [int(topic_times[k]) for k in keys] \n", - " plt.figure(figsize=(10,8))\n", - " sns.barplot(x=vals, y=keys).set(title='Time spent on ' + order + ' matched topic')\n", - "\n", - " \n", - "\n", - "plot_time_spent_for_topic(timestamp_to_topic_first_match, \"first\")\n", - "plot_time_spent_for_topic(timestamp_to_topic_second_match, \"second\")" - ] - }, - { - "cell_type": "markdown", - "id": "e9ae6e25", - "metadata": {}, - "source": [ - "## Example template 3" - ] - }, - { - "cell_type": "markdown", - "id": "69be38ce", - "metadata": {}, - "source": [ - "## Enhanced search for timelines" - ] - }, - { - "cell_type": "markdown", - "id": "f8a47348", - "metadata": {}, - "source": [ - "We can already search for a particular word in the interactive HTML document from example 1 to see a list of all transcribed sentences having an occurence of the word (in the context of the chosen topic). \n", - "\n", - "We can also retrieve all the segments(timestamps)in the transcription, related to a particular topic, to\n", - "\n", - "i) Segregrate all content on a particular topic of importance.\n", - "\n", - "ii) Perform selective summarization of the segregated content to make productive follow-ups. (Maybe use a model to extract action items and announcements from the transcription or selective summary ? )\n", - "\n", - "iii) Use the timestamps to highlight video / audio / transcription segments.\n", - "\n", - "iv) Jump to a desired segment of video / audio / transcription." - ] - }, - { - "cell_type": "code", - "execution_count": 28, - "id": "69d814c9", - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Timelines where Founders was covered : \n" - ] - }, - { - "data": { - "text/plain": [ - "[(12.36, 25.76),\n", - " (25.76, 30.32),\n", - " (35.52, 49.56),\n", - " (76.64, 87.76),\n", - " (87.76, 96.08),\n", - " (104.32, 105.62),\n", - " (107.02, 111.0),\n", - " (119.24, 123.14),\n", - " (125.64, 130.64),\n", - " (130.8, 132.6),\n", - " (142.72, 145.24),\n", - " (152.88, 155.14),\n", - " (155.14, 157.8),\n", - " (157.8, 160.96),\n", - " (171.0, 188.68),\n", - " (188.68, 197.28),\n", - " (202.84, 214.48),\n", - " (214.48, 219.28),\n", - " (219.84, 228.44),\n", - " (228.44, 238.76),\n", - " (246.46, 248.84),\n", - " (248.84, 251.38),\n", - " (256.78, 259.5),\n", - " (259.5, 261.94),\n", - " (270.96, 278.4),\n", - " (278.4, 288.32),\n", - " (291.44, 293.84),\n", - " (336.0, 339.12),\n", - " (351.48, 352.32),\n", - " (362.64, 393.44),\n", - " (402.72, 403.04),\n", - " (427.12, 432.12),\n", - " (438.96, 441.34),\n", - " (448.72, 453.84),\n", - " (453.84, 463.36),\n", - " (463.36, 466.5),\n", - " (466.5, 468.26),\n", - " (471.66, 473.06),\n", - " (482.28, 487.6),\n", - " (515.84, 522.64),\n", - " (529.66, 533.24),\n", - " (533.24, 535.52),\n", - " (540.42, 543.16),\n", - " (544.0, 550.0),\n", - " (556.0, 564.0),\n", - " (564.0, 568.76),\n", - " (568.76, 569.6),\n", - " (573.36, 574.56),\n", - " (577.52, 580.96),\n", - " (580.96, 584.6),\n", - " (584.6, 605.62),\n", - " (607.86, 609.18),\n", - " (619.42, 621.54),\n", - " (621.54, 623.98),\n", - " (623.98, 626.88),\n", - " (630.24, 634.96),\n", - " (634.96, 639.96),\n", - " (641.68, 646.66),\n", - " (659.28, 661.96),\n", - " (666.3, 667.14),\n", - " (670.48, 673.92),\n", - " (673.92, 677.02),\n", - " (677.02, 678.52),\n", - " (678.52, 680.72),\n", - " (680.72, 683.48),\n", - " (683.48, 746.8),\n", - " (758.32, 773.0),\n", - " (797.28, 813.36),\n", - " (813.36, 819.6),\n", - " (820.32, 824.08),\n", - " (821.7, 824.86),\n", - " (828.56, 830.94),\n", - " (831.78, 832.88),\n", - " (832.88, 837.04),\n", - " (837.04, 840.68),\n", - " (840.68, 844.64),\n", - " (854.78, 860.2),\n", - " (860.2, 865.4),\n", - " (865.4, 870.0),\n", - " (870.0, 872.2),\n", - " (877.24, 880.94),\n", - " (880.94, 882.94),\n", - " (882.94, 885.96),\n", - " (889.52, 892.96),\n", - " (896.4, 900.88),\n", - " (900.88, 903.24),\n", - " (908.36, 911.42),\n", - " (911.42, 912.74),\n", - " (915.26, 917.68),\n", - " (921.32, 924.0),\n", - " (924.0, 953.04),\n", - " (982.0, 988.0),\n", - " (988.0, 995.0),\n", - " (995.0, 1009.52),\n", - " (1010.24, 1024.8),\n", - " (1022.36, 1025.5),\n", - " (1033.8, 1036.58),\n", - " (1040.9, 1044.56),\n", - " (1047.44, 1070.4),\n", - " (1070.48, 1076.6),\n", - " (1080.56, 1088.32),\n", - " (1088.32, 1091.08),\n", - " (1108.12, 1109.4),\n", - " (1115.16, 1119.1),\n", - " (1131.44, 1134.4),\n", - " (1134.4, 1135.64),\n", - " (1141.8, 1144.46),\n", - " (1151.52, 1152.88),\n", - " (1155.16, 1156.44),\n", - " (1161.36, 1166.24),\n", - " (1166.24, 1172.32),\n", - " (1198.36, 1203.16),\n", - " (1205.28, 1206.12),\n", - " (1211.94, 1213.8),\n", - " (1216.16, 1219.56),\n", - " (1221.84, 1224.2),\n", - " (1224.2, 1245.0),\n", - " (1254.34, 1256.64),\n", - " (1257.96, 1259.04),\n", - " (1259.04, 1260.76),\n", - " (1260.76, 1269.0),\n", - " (1269.0, 1275.0),\n", - " (1275.0, 1280.0),\n", - " (1290.64, 1293.92),\n", - " (1293.92, 1296.6),\n", - " (1296.6, 1300.16),\n", - " (1300.16, 1303.16),\n", - " (1303.16, 1307.2),\n", - " (1330.72, 1337.44),\n", - " (1394.92, 1401.62),\n", - " (1401.62, 1408.2),\n", - " (1408.2, 1409.56),\n", - " (1413.54, 1417.68),\n", - " (1417.68, 1420.92),\n", - " (1420.92, 1422.56),\n", - " (1422.56, 1424.36),\n", - " (1424.36, 1426.68),\n", - " (1427.52, 1428.86),\n", - " (1428.86, 1429.7),\n", - " (1429.7, 1430.54),\n", - " (1436.84, 1439.84),\n", - " (1455.56, 1456.64),\n", - " (1465.72, 1468.38),\n", - " (1468.38, 1469.54),\n", - " (1476.24, 1481.24),\n", - " (1482.64, 1490.0),\n", - " (1490.0, 1539.0),\n", - " (1541.0, 1547.68),\n", - " (1547.68, 1553.2),\n", - " (1553.2, 1558.96),\n", - " (1563.04, 1566.6),\n", - " (1566.6, 1568.4),\n", - " (1569.48, 1571.1),\n", - " (1571.1, 1574.0),\n", - " (1574.0, 1575.62),\n", - " (1577.2, 1578.98),\n", - " (1581.7, 1583.04),\n", - " (1583.04, 1584.32),\n", - " (1584.32, 1591.0),\n", - " (1596.0, 1606.92),\n", - " (1606.92, 1611.72),\n", - " (1614.04, 1615.88),\n", - " (1618.88, 1619.8),\n", - " (1625.72, 1630.44),\n", - " (1634.04, 1635.04),\n", - " (1646.4, 1648.52),\n", - " (1648.52, 1649.72),\n", - " (1649.72, 1652.04),\n", - " (1652.04, 1655.16),\n", - " (1658.96, 1660.88),\n", - " (1666.12, 1668.68),\n", - " (1672.84, 1674.56),\n", - " (1674.56, 1676.88),\n", - " (1676.88, 1680.44),\n", - " (1682.68, 1688.0),\n", - " (1694.0, 1702.0),\n", - " (1702.0, 1749.0),\n", - " (1769.36, 1772.84),\n", - " (1784.32, 1789.2),\n", - " (1821.24, 1826.8),\n", - " (1826.8, 1833.44),\n", - " (1839.68, 1854.0),\n", - " (1871.0, 1877.0),\n", - " (1877.0, 1883.0),\n", - " (1885.32, 1888.4),\n", - " (1888.4, 1891.36),\n", - " (1893.72, 1896.36),\n", - " (1901.52, 1903.04),\n", - " (1904.96, 1909.94),\n", - " (1915.06, 1918.38),\n", - " (1922.78, 1927.08),\n", - " (1927.08, 1933.0),\n", - " (1933.0, 1939.68),\n", - " (1939.68, 1946.62),\n", - " (1952.86, 1958.66),\n", - " (1958.66, 1961.74),\n", - " (1961.74, 1964.54),\n", - " (1964.54, 1987.12),\n", - " (1995.6, 2000.24),\n", - " (2006.5, 2009.9),\n", - " (2009.9, 2011.14),\n", - " (2016.94, 2017.86),\n", - " (2030.48, 2032.12),\n", - " (2034.52, 2036.64),\n", - " (2036.64, 2039.12),\n", - " (2039.12, 2041.36),\n", - " (2042.24, 2050.0),\n", - " (2050.0, 2070.08),\n", - " (2070.08, 2079.28),\n", - " (2079.28, 2087.44),\n", - " (2087.44, 2092.24),\n", - " (2100.16, 2104.64),\n", - " (2105.0, 2108.24),\n", - " (2112.8, 2115.12),\n", - " (2125.0, 2129.0),\n", - " (2136.0, 2139.0),\n", - " (2139.0, 2141.0),\n", - " (2141.0, 2143.0),\n", - " (2143.0, 2145.1),\n", - " (2145.1, 2148.98),\n", - " (2148.98, 2153.98),\n", - " (2162.38, 2165.68),\n", - " (2165.68, 2169.92),\n", - " (2172.02, 2173.66),\n", - " (2181.18, 2183.28),\n", - " (2183.28, 2209.52),\n", - " (2209.52, 2213.0),\n", - " (2213.0, 2214.44),\n", - " (2214.44, 2216.32),\n", - " (2216.32, 2217.72),\n", - " (2246.8, 2251.44),\n", - " (2292.0, 2307.76),\n", - " (2307.76, 2321.52),\n", - " (2338.0, 2341.0),\n", - " (2343.0, 2365.92),\n", - " (2365.92, 2370.76),\n", - " (2370.76, 2375.12),\n", - " (2378.96, 2382.32),\n", - " (2386.54, 2392.52),\n", - " (2399.18, 2409.92),\n", - " (2415.6, 2428.0),\n", - " (2446.0, 2448.0),\n", - " (2461.0, 2467.0),\n", - " (2471.3, 2475.8),\n", - " (2495.04, 2498.72),\n", - " (2526.72, 2535.76),\n", - " (2536.96, 2542.4),\n", - " (2545.48, 2547.16),\n", - " (2547.16, 2550.28),\n", - " (2550.28, 2554.44),\n", - " (2554.44, 2561.88),\n", - " (2561.88, 2563.64),\n", - " (2563.64, 2590.0),\n", - " (2590.8, 2595.36),\n", - " (2595.92, 2609.0),\n", - " (2609.0, 2612.0),\n", - " (2620.0, 2646.12),\n", - " (2650.58, 2651.92),\n", - " (2706.94, 2708.52),\n", - " (2708.52, 2710.6),\n", - " (2710.6, 2712.64),\n", - " (2715.68, 2719.76),\n", - " (2719.76, 2722.08),\n", - " (2722.08, 2723.6),\n", - " (2723.6, 2725.64),\n", - " (2726.88, 2728.32),\n", - " (2732.6, 2733.44),\n", - " (2739.76, 2741.72),\n", - " (2745.68, 2748.7),\n", - " (2751.14, 2754.6),\n", - " (2754.6, 2755.6),\n", - " (2764.32, 2766.4),\n", - " (2766.4, 2769.08),\n", - " (2774.04, 2776.2),\n", - " (2776.2, 2778.68),\n", - " (2778.68, 2782.06),\n", - " (2782.06, 2784.96),\n", - " (2784.96, 2804.24),\n", - " (2826.0, 2827.0),\n", - " (2838.0, 2853.0),\n", - " (2872.6, 2881.64),\n", - " (2881.64, 2889.2),\n", - " (2889.2, 2913.12),\n", - " (2913.12, 2926.0),\n", - " (2926.0, 2928.0),\n", - " (2928.0, 2933.0),\n", - " (2933.0, 2938.0),\n", - " (2938.0, 2940.0),\n", - " (2944.0, 2945.2),\n", - " (2953.3, 2955.5),\n", - " (2960.18, 2961.84),\n", - " (2968.24, 2975.2),\n", - " (2975.2, 2981.76),\n", - " (2986.0, 2992.78),\n", - " (3005.14, 3013.4),\n", - " (3013.4, 3049.0),\n", - " (3049.0, 3051.0),\n", - " (3051.0, 3053.0),\n", - " (3063.0, 3085.28),\n", - " (3091.2, 3097.04),\n", - " (3123.84, 3172.0),\n", - " (3209.44, 3216.72),\n", - " (3227.0, 3231.0),\n", - " (3231.0, 3234.0),\n", - " (3240.0, 3241.0),\n", - " (3244.56, 3249.92),\n", - " (3255.88, 3257.44),\n", - " (3257.44, 3261.24),\n", - " (3261.24, 3268.4),\n", - " (3268.4, 3275.28),\n", - " (3323.0, 3326.56),\n", - " (3326.56, 3332.24),\n", - " (3338.32, 3344.4),\n", - " (3344.4, None)]" - ] - }, - "execution_count": 28, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "def retrieve_time_segments(topic):\n", - " return topic_to_timestamp_first_match[topic]\n", - "\n", - "search_topic = \"Founders\"\n", - "print(\"Timelines where \" + search_topic + \" was covered : \")\n", - "time_segments_of_interest = retrieve_time_segments(topic=search_topic)\n", - "time_segments_of_interest" - ] - }, - { - "cell_type": "markdown", - "id": "b587da79", - "metadata": {}, - "source": [ - "## Selective segregation of content" - ] - }, - { - "cell_type": "code", - "execution_count": 31, - "id": "5dc2014f", - "metadata": {}, - "outputs": [], - "source": [ - "import json\n", - "import ast\n", - "\n", - "time_segments_of_interest = retrieve_time_segments(\"Founders\")\n", - "\n", - "ts_transcript = {}\n", - "with open(transcript_file_name, \"r\") as f:\n", - " ts_transcript = f.read()\n", - "ts_transcript = ast.literal_eval(ts_transcript)\n", - "\n", - "selective_transcribed_content = \"\"\n", - "for chunk in ts_transcript[\"chunks\"]:\n", - " if chunk[\"timestamp\"] in time_segments_of_interest:\n", - " selective_transcribed_content += chunk[\"text\"]" - ] - }, - { - "cell_type": "code", - "execution_count": 32, - "id": "caeff7f1", - "metadata": {}, - "outputs": [ - { - "data": { - "text/plain": [ - "\"because Goku needs that for the audio plus the transcript plus the timestamps . So cool . Okay , so we can have our discussion as planned about ontology prompt . So JDC , I just wanted to learn about it and I think the rest of the team did . So I want to just start off , maybe you want to give us some context and introduction . So what I , it all started with the demo from Palantir. , as you know , they were able to make this AIP thing where they were such a powerful thing perhaps because they use an ontology . So it means that they have a definition of entities and actions that they can take . So then they control this to an LLM . So I started exploring with that with Jamo . in our hierarchical way using just like texts . And it kind of works . So in the end , I decided that one good pattern operations , mutations and queries over certain entities . So I assume that many of the LLMs have seen a lot of this data and know how to operate . So people from OpenAI and many companies So that 's how I started . So one of the problem domains that we could map to this , it 's CP because it 's some actions that you need to take . So that 's basically the system I end up developing . But that has some challenges . So the first thing is that , let me share the screen . And I can take you over . It 's amazing that Corey , Michal , and John are attending , because I 'm reaching to the point that I would love to have more ideas . Can you see the screen ? to you independently , just very late for him today . So I started by like mocking these kind of entities that could work with Zibi . So we have like employees , candidates . I just took a look at out of the comments that we support right now , reminders , and then put together this GraphQL thing . Yeah . Right now I see the whereby window . Is that what you 're showing ? Oh , sorry . No . Let me see . Maybe I 'm just sharing one , not the whole screen . Okay . I think this will solve it . that we have for many purposes , like requesting vacations and some other things . So there are some entities that are employees . I just like mocked some of them . just mocked up . Get reminder . This was the first thing . There is the vacations input , create reminder , and some mutation . So as you can see , this is part of the first test that I did . So let me show you here how this can work . So let 's say in the first day of July , I 'm trying to , sorry , I was in the middle of some page here . Oh , my God . It 's always show up . Yeah , yeah . I 'm refactoring like heavily this , but you see , it does work very well . So wait , let me change the strategy here . Let me see another example . Okay . So let 's say vacations . Because I 've been experimenting a lot . Just want to show you one related with vacation . Well . Let me see if it can work this way . I have many modes of operation . It behaves very good for this kind of . So it creates vacations , it fills all the things here . So this can basically be adapted to any CLI . GraphQL pretty well . command line interface for many things as long as you can describe the things like this . And this is a very flexible pattern . It was really taught by the people that invented this on Facebook . It divides the things by queries and mutations , and I 've tested heavily . What 's the challenge ? So the initial LLMs that we had access to , especially from OpenAI , I 'm giving it . And on the basis of this , I create like a subtree because this thing generates So there are two things reference here to types . There is vacations input and employee . This has not like further reference to other types . And then I come to employee here , and there are no further references . So I 've been implementing strategies for this . So one strategy kind of infers what do I need to use on the basis of the description . So let me show you , to Michal and Shonda , are very familiar . with these descriptions . and then identify or just use that strategy of the top K to know which ones should I include . Then I create this like sub tree that fits in most of the time in the GraphQL . Then I can use this . There is another strategy where I directly ask the LLM . So instead of using embeddings , I ask the LLM as if they were tools . That is the thing that is now like very popular . So I asked the LLM , this is the query I have . with their own descriptions . And then on the basis of this , I create again , these like sub tree of the GraphQL , and then I can obtain the query . with using the LLM in the case of , says me concretely , that it 's the use case that Adam wanted to test . This thing has so , the number of queries is so big , So let 's say , let 's say with GitHub . Let me see . So this is the query I 'm giving for the Rails organization , get the Rails repo on the last full request with the status open . This is the GraphQL that I 'm passing . The schema strategy , I 'm using embeddings . So I 'm just creating the embeddings and then getting the top K , rebuilding that tree so I can fit it to the LLM . Sorry . This is in an unstable state right now . Let me try to uncomment some things that I 'm thinking right now . Let me test another one , stars . Yeah , it 's better for me to use the LLM strategy here . So this usually can take a while . So there are even some cases where the GraphQL is so big that even after skimming it and just getting the parts , it 's too big . But with the Anthropic LLM , I can make it , idea is that you have a human query , then you have a description of the mutations and queries that you can take . But to be able to feed that to the LLM , you need a strategy to tell the LLM which parts it needs to use . how big it is , not even in the bigger ones , in the bigger Anthropic . So it 's 58,000 lines . And maybe make an average of two tokens per each . So that crosses even the 100K window of this LLM . So I just need to pick the entities that are relevant with whatever strategy . Right now there are two , like LLMs and embeddings . with whatever strategy . Right now there are two like LLMs and embeddings . and let me show it for you . Quick question . Do they all have like the really verbose comments there ? All the GraphQL queries ? Are they all , you know , do they have rich comments like that ? shrink that by eliminating those . Yeah . Right now that 's one of the requirements for the strategies . Because if you got rid of the comments , then the LLM would not maybe know as well . Yeah , as well . Yeah . because of the lower tokens , but then , you know , depending on whether the variable names are self-describing enough . Exactly , exactly . So I 've implemented some strategies . the tools that I have available . And then when I include the GraphQL , It 's all about like trying to save like characters always , not just for like cause reasons , But yeah , all of those combinations are available . and it might work if these variable names are self-descriptive . As you have seen , this one , it 's really without any comments , and it works pretty well . Let me show you how these things work . There is this Explorer from ... Let 's see . I asked this , get the ROS projects with most stars . so it 's , like , ordered . And then you 've got the stars and then you can . So that 's , let me try another one . Yeah , and maybe I could share a use case that what we were talking about , KTC , and maybe the others on the call have already seen it in the chat , but with that kind of US federal initiative says me the smart manufacturing Institute and then the think IQ platform they 're building . They 're really pushing tons of small medium sized manufacturers to leverage these open data standards to collect data for like incompatible appliances , like let 's say mixers on a manufacturing line from different manufacturers that do n't interoperate . let 's say mixers on a manufacturing line from different manufacturers that do n't interoperate . to do that with just their knowledge . then improving the GUI interface that they have , And I think where we can step in potentially what we 've been exploring is to go beyond the GUI interface where it 's like a lot of , they use GraphiQL actually , a lot of like pointing and clicking to select the specific appliances and data elements . How do we take that a step further and basically use language like JDC is describing here to say , Hey , I want to get a historian graph for all of the mixers that we had on the line for the last four Saturdays because you know , So that 's the idea to spit out a graph QL endpoint that would put that specific data for them , right ? That 's kind of the idea we 've been exploring at least . but by just describing . for a given instance after we 've specified time range . 1690 , max 10 samples within the start time , blah , blah , blah , end time . of more things , just the first , Yeah , I see that that 's an example . like English , right ? would be able to think about the information they need and have an awareness of that . And honestly , most engineers in manufacturing spaces are familiar with that type of notation , through this interface , then that 's incredibly robust . Exactly right . Yeah , because I 'm using Anthropic , That 's another story of like tweaking the prompt . then use the cheaper models that are faster . So right now it 's in a state that it 's like a Frankenstein , because I 'm using some parts from OpenAI and other from Anthropic . But right now it 's like the embeddings . It 's using like embedding from OpenAI and getting the tools that I need to use . And then when I get the GraphQL , I use it in Anthropic , Sesame 's GraphQL , it 's like , whoa . 15,000 , yeah . Yeah , 15,000 . Just let 's say like 30K , 30K tokens , making an average . So that goes into every prompt . It has to have all 30,000 tokens , so you 're left with 70,000 for our response , essentially . Or no , 70,000 for context , I guess . as context , that it 's the prompt plus all the data that I include , that it 's in context data , and that includes the GraphQL description . So that 's why I need to like to save as much as possible , trying to squeeze and getting just the parts that are relevant to the query . request . No , I 'm not . I 'm not . I 'm not like right now . I 'm not sampling that , but it 's less than that because the whole idea . I mean , once let me , let me explain you better . So the strategy that my point , but I think that the Miha and Sean , do you get the point ? Like it 's these things like reference other types , so you need to include them like all . Yeah , yeah , you do like an LLM guided tree shaking to minimize the input code . Yeah , yeah . And on the basis of this , I kind of like , how do you say , cut this huge tree of GraphQL . Like tree shaking , yeah . That 's basically . Yeah . To include just the things that I will know . And all the dependencies . Yeah , exactly . Yeah . Is it only the human language descriptions that I use for Anthropic . You must use the following criteria , blah blah , blah , context , date , time . Like these rules , the context , date , time , and that 's it . Yeah , but on the OpenAI side , when you generate the embeddings to find what you want to use from this . Let me show you . So I came up with this like a strategy . So there is this general thing , schema 'm passing , I return the whole schema . LLM strategy . So , it works like this . Get tool descriptions by operation . So , what it basically does , it gets the ... I mean , it 's easier to show it here . So , let 's say I get this , just this part that says code of conduct , And then I put two dots and say like this , look up the code of conduct by scheme . So I pass all of these in this format . You see here that key value . This is the descriptions of the tools . So it 's like , it 's the symbol name symbol name colon and then the comment . Yeah , yeah exactly . That 's the LLM strategy . So does it incorporate the comment on the type name as well ? Because it 's like the whole tree right or what ? Yeah and this list of tools separated by commas , answer the following question as best as you can , blah , blah , blah . And here you have the tool description . So that 's the format of code of conduct , comma . So that 's the LLM strategy . The other strategy is simpler . And it 's embedding strategy . So for the embedding strategy , I basically construct these documents that are the same thing , you see ? So for the LLM strategy , do you run the LLM once for every single field of every type in the source schema ? No , no , no , it 's not . I chunk this , like code of conduct to dot , look up the code of conduct by its key . Yeah , yeah . but I 'm not talking that case , but that 's not very hard to do . Yeah . and all the mutations and it still it works . It does n't work for Sesame . So that 's why I wrote embeddings . Because embeddings , you know , I just create embedding , Let 's say the tools are the queries and mutations shrink and tree , shrink and graph QL tree . and mutation descriptions . And then I create this document using the Java index capabilities . It 's just that , let 's say it will say like Enterprise and Lookup Enterprise in this . So you see here , for all in blah , blah , blah , get keys . And then I return the proper documents . So this is what it identifies the operations , order by a score , and then build the schema from operations . That 's what I call the thing . Once I know which queries or mutations I could be needing , this is the function works . But remarkably , it 's able to tackle these very huge schemas really well . You see ? You 've got it . JVC , what are your ? So two questions . One , I have n't seen if you put out a repo for this yet . I understand you 're refactoring . But if I want to run this on my own machine , are you going to be done refactoring soon ? Or what 's your thoughts on that ? open AI and for the GraphQL , So it 's here . What 's the other advantage of this ? awesome examples and maybe like two or three examples that like are totally out of scope and not feasible , and then take some time to demo that to them and maybe start the conversation us eager ? How could this make maybe some additional companies adopt this that are like , you know , they 're afraid of using a graph to a GUI to structure endpoints and then dump them into some sort of data explorer , right ? And then demonstrate the query , but then also maybe to be able to see the actual data being run itself . So if we put some sort of like a graphana type thing , whatever , on the third end there , I 'm wondering how much of this that , how elegant that could be really , right ? but at least maybe certain things it works with . Yeah , I 've tried with many of these queries , and give them extra details when those are missing , of all attributes . So query equipment , you see this one , equipment . My impression is that this kind of use case , that then they put into some like dashboard or anything . Because in the end , you will need , I think you will need some expertise because sometimes , unless you describe all the fields , it will miss some fields or unless you , let 's say , sometimes it just selects the first 10 . So unless you understand how to get rid of that . and specific interfaces for , again , key questions . Give me this particular data for these types of machines . Or give me a list of these machines . Or give me a list of these locations and all the machines , you know , and all the machines that are like the combination of all these queries . I think that that 's really how our customers are utilizing that to create these kind of custom reports . But like , what if that did n't have to happen ? And instead , you could have that natural language query that maybe works for like the simpler or majority of requests . And then you maybe it bypasses , I guess for instance , this one . A Grafana visualization of that , right ? but for any reason uncontrollable , it just got the first 10 . That 's another thing , in the GraphQL , but that 's another story . So there might be some inconsistency . Cause I just downloaded these from someplace here . Like , yeah . And in that case , we would need some sort of like a post-processor for these very specific errors , right ? To basically resolve them . Yeah . That 's ... I kind of see like more potential on some other use cases like general interface for general command line interface . There are some efforts on it . So that 's also worth mentioning maybe to the guys here . So these people from Stanford tried to make basically this kind of the same thing . So , scaling tools . So , these tool patterns , that is , you have some query and then you have some tools available to solve it , but you always have the that is some YAMA , and fine-tuned it on the instructions . That is some YAMA , and fine-tune it on the instructions . This is using tools such as Hugging Face Models and things like that . But it 's way more complex , in my opinion , I think , because they tried to format everything as function calls . And I think this thing has seen more ... rather than SQL or some function code . You know , there is this comas and things . GraphQL , it 's kind of simpler . You just know the fields . You might miss some fields and still the thing goes on . So it 's kind of in between in the spectrum of this contract . And on the other end , you have like things like REST , but in the , sorry , in the other end , you have just JSON , you have like REST and GraphQL . But there are many efforts doing this . And you know , right now also OpenAI has included natively a functionality to call , to do this kind of thing . I think behind the scenes , they are doing the same . So it does n't , one can just use the same thing and it will take advantage of the training that they put already . But underneath , it 's doing a very similar thing , I guess . It 's called OpenAI functions . Yeah , OK . For the graphic , sorry . Sorry , go ahead . them and get their perspective . And then , you know , I can summarize those learnings for us to think There is some ... It says that it has an error , but it 's not erroneous . That 's something I do n't understand sometimes with GraphQL . I instructed for the Rails organization , get the repo Rails and the last 10 pull request titles with open status . Let me show you . Sorry , Sharon , let me cut you off . I just want to close out that item . Yeah . So you can see here . Also , response request not issues . So , qualify association . It 's here , it 's in there . Get rid of unnecessary gem . You see it 's , and it 's kind of a complex query . I just said like for the Rails organization , get the repo Rails and the last 10 pull requests with titles with open status . And it has enums and things like that that are not trivial . This is an enum . This is this kind of nesting pattern that it needs to use . So I feel very confident with the thing . And it also did some good work with Sesme . One idea is that this can form the basis of some assistant , local assistant , so you could thing that I 've seen that people did with this Gorilla . So if you ... This is the ... That 's related with the other thing URLs , internal URLs ? These people are moving in that direction . Yeah , okay , yeah . like this text interface for many things . This is the one I was referencing , the AIP thing . Let 's say it receives an alert , and then it says , show me more details . You can just stuff the prompt , show me more details , but they are referencing these entities . So you know , immediately what 's the query that it 's involved there and even the IDs that you can like stuff into the prompt . So this is the one dimension that was the motivation to start with this . And it 's remarkable that these guys , there is some place where it 's shown that they are using open source model here . You see , they disclose this . So they 're using Flan , GPT-NEWX . Thank you , God . I got all of these screenshotted . You can go to AIP , or maybe at the start of ontology prompt . And I was analyzing all of these screens that they disclose . Yeah . Remember , they also disclose some very interesting thing here of how they work . So they just use the LLM as some other user . They do n't have this problem of what it can access or not . It 's just like another user . So whatever that user has access to , the LLM has access to . And this is the actions that it can take . They include one thing called suggestions . Yeah , that 's a main thing that they advertise , just like , figure out LLM security . Look , this other thing , this other thing has also a bunch of clues of how they , so instead of getting this PII problem embedded into the model , they just put an input filter to whatever PII is involved . They just cut it , then it 's the model , and then there is the validation . Very cool , yeah . Sorry , Tom , you were about to ask a question earlier . Yeah , so for the output of the GraphQL , are you using the constraint generation or whatever ? Or is it just opening the syntactically correct GraphQL on its own ? it 's open and it 's working well . I 'm not like constraining it to some like concrete schema . with knowledge of the schema . Yeah , definitely . But right now , it 's not constrained at all . Yeah , the only protection that I have right now for this . You validate afterwards , right ? Oh , I do two things . Like , let do two things . Let me show you . I think here it 's in main . Oh , okay . because there could be a bunch of errors . GraphQL at all , but I 've never encountered that . but the prompts that I 'm using are already like covering for that . like , hey , take into account that this data , it 's in this format , something like that . for GraphQL queries or mutations . Oh , like to extract the part of the result . Yeah , and then I do the validation . But I 'm really not very concerned . I think right now this is not a problem . I mean , it behaves pretty well . Okay , in that in that front . But yeah , you could like constrain I think there are some efforts and I there I read a debate of what was I read a debate of what was OpenAI really doing , because you could do two things . One , it 's this open-ended thing , and then constrain it with regex . And in the case that it misses , you can feed it back and tell it , hey , this is bad . the valid ones . That 's another solution that I 've seen . And finally , it might be the one that you have in mind that it 's like constrained , that you have to hack into the LLM latest layer , and then you can squash the probability because in that layer , you will have one output per token , so there will be some invalid tokens . There 's some guy that did that . Yeah . That 's kind of the approach that Gorilla ... They do n't need this . They are able to chunk thousands of tools without having to select the tool . It 's just the training that speeds that . Yeah , they did say . But it has some small errors . I think I prefer to have a way more powerful LLM rather than have that ... From what I 've seen , it 's very easy just to use the regex and then validate . Yeah , fair . Especially since so much of this is plain language already . The way the function calls are , they 're kind of intuitive when you read them in a way , right ? Yeah . Yeah , I 've seen that this model , because it 's this fine-tuning , foundational model really helps in that situation . from English to Chinese if you come here . And it selected the French model . So there should be , I do n't know why they include this example in their showcase , but they are being honest with that . So it 's like , how do you call that ? It 's like overfitting because you 're not even able to generalize these codes . You just use the model . completely biased , it 's like overfitted . So I kind of prefer the more general model than Google did this but for SQL . And if you read the paper like very , with a lot of attention , they did n't gain much by fine tuning . You are comparing it with others . Look at the figures . The difference between fine-tune and the other one was just like 1 % . Put the analysis here . Yeah . 77.3 to ... Look3 . 77.3 to 78.3 . Look , so the only difference ... Okay , this is the whole dataset . Sorry . So , the difference between a few-shot stuff model to a fine-tuned model , it was just 1 % . So , I really prefer to have the few shots . That 's something that is also emerging . samples to stuff in . And that might work better than just fine-tuning and that 's like costly . They changed their docs , but let me search for samples . Wait . Or read Or retreaters . Well , they have this concept , you know , that on the basis of the prompt , they fetch the most relevant view shot samples to stuff in . And according to what I 've seen , it might work better than just like fine-tuning . Yeah , but at some point it might make sense to fine-tune . Yeah , this is very exciting . I 'm looking forward to running it on my own machine and kind of understanding a little bit deeper . And yeah , exploring more use cases , that sounds really exciting as well . Yeah , so I play with ... Let me ... Can be stuffy . But it did work also for these huge GraphQLs . I 'm not using toy things . GitHub . Sorry ? Yeah . Well , at this point , just to comply with monadicals thing , sending the . Yeah , and that 's where Sean 's ideas about how to overcome the context window size and local processing would maybe help . If it was feasible . Not necessary . No , it 's just using a privately hosted LLMs . I 'm using Anthropix and OpenAI . It 's crazy . But definitely with the local models , we have n't found a way to work with that . Plus , you know , if some of them , it 's like , We actually have a preprocessor to resolve what we 're seeing on the screen right now . But yeah , there 's still some struggles on the local model that we have to figure out and get creative with . Okay , I 'm going to stop the recording unless anybody else has any last questions model could be useful for the token . Contact size problem as well . Like if something that could be applied to the scheme . Get rid of this . Anyway , that 's all I just wanted to mention . Yeah , yeah . Awesome . Hey , thanks , everyone . \"" - ] - }, - "execution_count": 32, - "metadata": {}, - "output_type": "execute_result" - } - ], - "source": [ - "selective_transcribed_content" - ] - }, - { - "cell_type": "markdown", - "id": "a20896b4", - "metadata": {}, - "source": [ - "## Selective topic summarization" - ] - }, - { - "cell_type": "markdown", - "id": "6f8ab415", - "metadata": {}, - "source": [ - "We can use this selective content to now summarize using the already available pipeline !" - ] - }, - { - "cell_type": "markdown", - "id": "06f009d5", - "metadata": {}, - "source": [ - "# And Much More !!" - ] - } - ], - "metadata": { - "kernelspec": { - "display_name": "Python 3", - "language": "python", - "name": "python3" - }, - "language_info": { - "codemirror_mode": { - "name": "ipython", - "version": 3 - }, - "file_extension": ".py", - "mimetype": "text/x-python", - "name": "python", - "nbconvert_exporter": "python", - "pygments_lexer": "ipython3", - "version": "3.8.8" - } - }, - "nbformat": 4, - "nbformat_minor": 5 -} diff --git a/server/notebooks/incsum.ipynb b/server/notebooks/incsum.ipynb deleted file mode 100644 index 471ae08d..00000000 --- a/server/notebooks/incsum.ipynb +++ /dev/null @@ -1,2534 +0,0 @@ -{ - "cells": [ - { - "cell_type": "code", - "execution_count": 3, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "Performing chunk summary : mpt-7B\n" - ] - }, - { - "name": "stderr", - "output_type": "stream", - "text": [ - "/Users/user/.pyenv/versions/3.11.2/lib/python3.11/site-packages/spacy/pipeline/lemmatizer.py:211: UserWarning: [W108] The rule-based lemmatizer did not find POS annotation for one or more tokens. Check that your pipeline includes components that assign token.pos, typically 'tagger'+'attribute_ruler' or 'morphologizer'.\n", - " warnings.warn(Warnings.W108)\n" - ] - } - ], - "source": [ - "print(\"Performing chunk summary : \" + \"mpt-7B\")\n", - "\n", - "from langchain import PromptTemplate\n", - "import torch\n", - "import transformers\n", - "from transformers import AutoTokenizer\n", - "from langchain.llms import TextGen\n", - "from langchain.prompts import Prompt\n", - "from langchain.chains.summarize import load_summarize_chain\n", - "from langchain.text_splitter import SpacyTextSplitter\n", - "\n", - "text_splitter = SpacyTextSplitter(\n", - " chunk_size = 5000,\n", - " chunk_overlap = 200,\n", - " length_function = len\n", - ")\n", - "\n", - "with open(\"transcript.txt\") as f:\n", - " txt = f.read()\n", - "\n", - "docs = text_splitter.create_documents([txt])\n" - ] - }, - { - "cell_type": "code", - "execution_count": 12, - "metadata": {}, - "outputs": [], - "source": [ - "from langchain import LLMChain\n", - "llm_summary = TextGen(model_url=\"http://216.153.52.83:5000\", max_new_tokens=250)\n", - "summary_prompt_template = \"\"\"Write a concise two line summary of the following:\n", - "\n", - "\n", - "{text}\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\"\"\"\n", - "summary_prompt = PromptTemplate(template=summary_prompt_template, input_variables=[\"text\"])\n", - "summary_chain = LLMChain(llm=llm_summary, prompt=summary_prompt, verbose=True)" - ] - }, - { - "cell_type": "code", - "execution_count": 13, - "metadata": {}, - "outputs": [], - "source": [ - "llm_subject = TextGen(model_url=\"http://216.153.52.83:5000\", max_new_tokens=100)\n", - "subject_prompt_template = \"\"\"Summarize the text below in a subject line:\n", - "\n", - "\n", - "{text}\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\"\"\"\n", - "subject_prompt = PromptTemplate(template=subject_prompt_template, input_variables=[\"text\"])\n", - "subject_chain = LLMChain(llm=llm_subject, prompt=subject_prompt, verbose=True)" - ] - }, - { - "cell_type": "code", - "execution_count": 15, - "metadata": {}, - "outputs": [ - { - "name": "stdout", - "output_type": "stream", - "text": [ - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mWrite a concise two line summary of the following:\n", - "\n", - "\n", - "We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI .\n", - "\n", - "Thomas joined Google in November 2018 as the CEO of Google Cloud .\n", - "\n", - "Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development .\n", - "\n", - "Before that , Thomas worked at McKinsey as a business analyst and engagement manager .\n", - "\n", - "His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations .\n", - "\n", - "Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . \n", - "\n", - "Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council .\n", - "\n", - "Please welcome to the stage , Thomas Curian and Alexander Wang .\n", - "\n", - "This is a super exciting conversation .\n", - "\n", - "Thanks for being here , Thomas .\n", - "\n", - "Thank you for having me .\n", - "\n", - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem .\n", - "\n", - "You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ?\n", - "\n", - "Now our vision is super simple . \n", - "\n", - "If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person .\n", - "\n", - "We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it .\n", - "\n", - "And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it .\n", - "\n", - "And every step that we do to simplify things for people , we think adoption can grow .\n", - "\n", - "And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . \n", - "\n", - "Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody .\n", - "\n", - "Third is advances because different people have different levels of expertise .\n", - "\n", - "Some people say I need the hardware to build my own large language model or algorithm .\n", - "\n", - "Other people say , look , I really need to use a building block .\n", - "\n", - "You guys give me . \n", - "\n", - "So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . \n", - "\n", - "And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . \n", - "\n", - "So , it 's a busy conference , but lots of exciting stuff going on .\n", - "\n", - "Yeah , it 's incredible .\n", - "\n", - "I mean , I want to zoom out for a second to start with , which is that this is obviously not your first time taking and packaging new technology breakthroughs for the enterprise .\n", - "\n", - "Both in your time at Oracle and now CEO of Google Cloud , this is something that you 've been doing for quite some time now .\n", - "\n", - "When you sort of zoom all the way out , what do you think are some of the things that have some of your principles , or some of your thoughts and enabling these technological breakthroughs and actually enabling the enterprise with them ?\n", - "\n", - "And what are the key insights that you have there ?\n", - "\n", - "Thank you .\n", - "\n", - "A lot of the work .\n", - "\n", - "So first of all , we 've really built out the organization the last three years . \n", - "\n", - "We 've seen a huge ramp up in our business , credit to all the people who joined us at one point over 70 % of organization that joined your in COVID . \n", - "\n", - "So they had n't met anybody . \n", - "\n", - "They could n't meet their managers , but they all did an amazing job together .\n", - "\n", - "The adoption of technology by companies , and I 'll give you just some elements , particularly in the application of AI in different domains that we 've seen .\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\u001b[0m\n", - "Write a concise two line summary of the following:\n", - "\n", - "\n", - "We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI .\n", - "\n", - "Thomas joined Google in November 2018 as the CEO of Google Cloud .\n", - "\n", - "Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development .\n", - "\n", - "Before that , Thomas worked at McKinsey as a business analyst and engagement manager .\n", - "\n", - "His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations .\n", - "\n", - "Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . \n", - "\n", - "Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council .\n", - "\n", - "Please welcome to the stage , Thomas Curian and Alexander Wang .\n", - "\n", - "This is a super exciting conversation .\n", - "\n", - "Thanks for being here , Thomas .\n", - "\n", - "Thank you for having me .\n", - "\n", - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem .\n", - "\n", - "You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ?\n", - "\n", - "Now our vision is super simple . \n", - "\n", - "If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person .\n", - "\n", - "We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it .\n", - "\n", - "And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it .\n", - "\n", - "And every step that we do to simplify things for people , we think adoption can grow .\n", - "\n", - "And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . \n", - "\n", - "Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody .\n", - "\n", - "Third is advances because different people have different levels of expertise .\n", - "\n", - "Some people say I need the hardware to build my own large language model or algorithm .\n", - "\n", - "Other people say , look , I really need to use a building block .\n", - "\n", - "You guys give me . \n", - "\n", - "So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . \n", - "\n", - "And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . \n", - "\n", - "So , it 's a busy conference , but lots of exciting stuff going on .\n", - "\n", - "Yeah , it 's incredible .\n", - "\n", - "I mean , I want to zoom out for a second to start with , which is that this is obviously not your first time taking and packaging new technology breakthroughs for the enterprise .\n", - "\n", - "Both in your time at Oracle and now CEO of Google Cloud , this is something that you 've been doing for quite some time now .\n", - "\n", - "When you sort of zoom all the way out , what do you think are some of the things that have some of your principles , or some of your thoughts and enabling these technological breakthroughs and actually enabling the enterprise with them ?\n", - "\n", - "And what are the key insights that you have there ?\n", - "\n", - "Thank you .\n", - "\n", - "A lot of the work .\n", - "\n", - "So first of all , we 've really built out the organization the last three years . \n", - "\n", - "We 've seen a huge ramp up in our business , credit to all the people who joined us at one point over 70 % of organization that joined your in COVID . \n", - "\n", - "So they had n't met anybody . \n", - "\n", - "They could n't meet their managers , but they all did an amazing job together .\n", - "\n", - "The adoption of technology by companies , and I 'll give you just some elements , particularly in the application of AI in different domains that we 've seen .\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - " We're joining today via Zoom, along with Tom Curian (CEO) & Alex WANG(Founder), both co-founders/executive officers @ScaleAI. \n", - " Priorities discussed included 1.) The company's mission 2.) How scaleai has grown 3). Their approach towards hiring 4.). What makes scaling ML models difficult 5 ). And more... \n", - "\n", - "\n", - "\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mSummarize the text below in a subject line:\n", - "\n", - "\n", - "We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI .\n", - "\n", - "Thomas joined Google in November 2018 as the CEO of Google Cloud .\n", - "\n", - "Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development .\n", - "\n", - "Before that , Thomas worked at McKinsey as a business analyst and engagement manager .\n", - "\n", - "His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations .\n", - "\n", - "Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . \n", - "\n", - "Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council .\n", - "\n", - "Please welcome to the stage , Thomas Curian and Alexander Wang .\n", - "\n", - "This is a super exciting conversation .\n", - "\n", - "Thanks for being here , Thomas .\n", - "\n", - "Thank you for having me .\n", - "\n", - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem .\n", - "\n", - "You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ?\n", - "\n", - "Now our vision is super simple . \n", - "\n", - "If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person .\n", - "\n", - "We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it .\n", - "\n", - "And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it .\n", - "\n", - "And every step that we do to simplify things for people , we think adoption can grow .\n", - "\n", - "And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . \n", - "\n", - "Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody .\n", - "\n", - "Third is advances because different people have different levels of expertise .\n", - "\n", - "Some people say I need the hardware to build my own large language model or algorithm .\n", - "\n", - "Other people say , look , I really need to use a building block .\n", - "\n", - "You guys give me . \n", - "\n", - "So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . \n", - "\n", - "And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . \n", - "\n", - "So , it 's a busy conference , but lots of exciting stuff going on .\n", - "\n", - "Yeah , it 's incredible .\n", - "\n", - "I mean , I want to zoom out for a second to start with , which is that this is obviously not your first time taking and packaging new technology breakthroughs for the enterprise .\n", - "\n", - "Both in your time at Oracle and now CEO of Google Cloud , this is something that you 've been doing for quite some time now .\n", - "\n", - "When you sort of zoom all the way out , what do you think are some of the things that have some of your principles , or some of your thoughts and enabling these technological breakthroughs and actually enabling the enterprise with them ?\n", - "\n", - "And what are the key insights that you have there ?\n", - "\n", - "Thank you .\n", - "\n", - "A lot of the work .\n", - "\n", - "So first of all , we 've really built out the organization the last three years . \n", - "\n", - "We 've seen a huge ramp up in our business , credit to all the people who joined us at one point over 70 % of organization that joined your in COVID . \n", - "\n", - "So they had n't met anybody . \n", - "\n", - "They could n't meet their managers , but they all did an amazing job together .\n", - "\n", - "The adoption of technology by companies , and I 'll give you just some elements , particularly in the application of AI in different domains that we 've seen .\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\u001b[0m\n", - "Summarize the text below in a subject line:\n", - "\n", - "\n", - "We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI .\n", - "\n", - "Thomas joined Google in November 2018 as the CEO of Google Cloud .\n", - "\n", - "Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development .\n", - "\n", - "Before that , Thomas worked at McKinsey as a business analyst and engagement manager .\n", - "\n", - "His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations .\n", - "\n", - "Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . \n", - "\n", - "Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council .\n", - "\n", - "Please welcome to the stage , Thomas Curian and Alexander Wang .\n", - "\n", - "This is a super exciting conversation .\n", - "\n", - "Thanks for being here , Thomas .\n", - "\n", - "Thank you for having me .\n", - "\n", - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem .\n", - "\n", - "You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ?\n", - "\n", - "Now our vision is super simple . \n", - "\n", - "If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person .\n", - "\n", - "We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it .\n", - "\n", - "And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it .\n", - "\n", - "And every step that we do to simplify things for people , we think adoption can grow .\n", - "\n", - "And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . \n", - "\n", - "Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody .\n", - "\n", - "Third is advances because different people have different levels of expertise .\n", - "\n", - "Some people say I need the hardware to build my own large language model or algorithm .\n", - "\n", - "Other people say , look , I really need to use a building block .\n", - "\n", - "You guys give me . \n", - "\n", - "So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . \n", - "\n", - "And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . \n", - "\n", - "So , it 's a busy conference , but lots of exciting stuff going on .\n", - "\n", - "Yeah , it 's incredible .\n", - "\n", - "I mean , I want to zoom out for a second to start with , which is that this is obviously not your first time taking and packaging new technology breakthroughs for the enterprise .\n", - "\n", - "Both in your time at Oracle and now CEO of Google Cloud , this is something that you 've been doing for quite some time now .\n", - "\n", - "When you sort of zoom all the way out , what do you think are some of the things that have some of your principles , or some of your thoughts and enabling these technological breakthroughs and actually enabling the enterprise with them ?\n", - "\n", - "And what are the key insights that you have there ?\n", - "\n", - "Thank you .\n", - "\n", - "A lot of the work .\n", - "\n", - "So first of all , we 've really built out the organization the last three years . \n", - "\n", - "We 've seen a huge ramp up in our business , credit to all the people who joined us at one point over 70 % of organization that joined your in COVID . \n", - "\n", - "So they had n't met anybody . \n", - "\n", - "They could n't meet their managers , but they all did an amazing job together .\n", - "\n", - "The adoption of technology by companies , and I 'll give you just some elements , particularly in the application of AI in different domains that we 've seen .\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - " The bank has used its ML models, trained using GCP’s auto ml platform (Auto Pipeline), since 2019. It says “the accuracy rate [of detecting fraudulent transactions] increased significantly” after adopting google ‘ s solution.” \n", - " In addition,”google ’ s ability to scale quickly helped shb become more agile”, according to chief information officer chris wong yan fook ”we needed someone else's help”. Google claims 90%+ reduction rates against traditional methods such as\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mWrite a concise two line summary of the following:\n", - "\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "You know , fraud detection and banking , there 's a lot of false positives , which makes it hard to really , you know , to a very expensive people doing something called anti-money laundering .\n", - "\n", - "And our AI algorithms are really able to be super precise on detection .\n", - "\n", - "Explainability is a critical thing there , right ?\n", - "\n", - "So people ask , why did you , why did you approve , why did you flag this one and not that one ?\n", - "\n", - "Because regulators are involved .\n", - "\n", - "So explainability becomes a big deal .\n", - "\n", - "We help , we help renewal , for example , monitor all of the factories .\n", - "\n", - "The process roughly , a billion data sets every day .\n", - "\n", - "Obviously , humans can process that .\n", - "\n", - "But making it super simple to , and you guys have given all your expertise in labeling and other things , you would get a sense .\n", - "\n", - "Factory floor data is not clean data .\n", - "\n", - "And so you have to actually clean , imagine doing a billion data sets into an environment every single day .\n", - "\n", - "You have to give the data pipelines really good .\n", - "\n", - "And so a lot of technology work happens to make that possible for companies .\n", - "\n", - "Third is , if you shop at IKEA , for example , behind IKEA is systems , it 's our recommendation system .\n", - "\n", - "find IKEA is systems , it 's our recommendation system . \n", - "\n", - "And the way that people shop for furniture and products is not the same in all countries .\n", - "\n", - "And so how are you able to one deal with the benefits you get from a global model , but also to contextually the specific elements in each country because people have different buying habits . \n", - "\n", - "Those are all things that we 've learned applying our AI in different contexts in different parts of the world .\n", - "\n", - "Yeah . \n", - "\n", - "You 've sort of glossed over this , but you 've led since you took over at Google Cloud , just a meteoric growth of the platform .\n", - "\n", - "You know , I think the past few years , you 've tripled your sales force and ending last year , you obviously ca n't come in this , but end the last year at , I believe , 20 billion of annual revenue , which is incredible and this incredible growth journey .\n", - "\n", - "What do you attribute your success to ?\n", - "\n", - "And how do you think you 've been able to drive just to an incredible growth and success ?\n", - "\n", - "From our point of view , every industry , virtually in the world , is now becoming a software powered technology industry .\n", - "\n", - "If you talk to automobile companies , they 're increasingly vehicles are more about software than mechanical systems .\n", - "\n", - "If you talk to telecommunications companies , the networks are commodities unless they can make them platforms to deliver applications , so they need new ways to slice , manage the network .\n", - "\n", - "If you look at banks at the end of the day , they 're about all the products of a bank or data , and all of that becomes how do you differentiate in the value delivering clients through a digital medium ?\n", - "\n", - "Because increasingly , I 'm sure all of you look at yourselves and go when was the last time I went to a branch of a bank .\n", - "\n", - "So a lot of our work has been pushing the technology innovation really far , but bringing that technology super easily to people in different industries .\n", - "\n", - "And given the demand that people have for a hair , I really want , I need the technology to help me power my industry , the change I 'm seeing in my industry , the more accessible we can make it , the easier and the faster we get adoption , and our approach has been to be completely open . \n", - "\n", - "And when to be completely open .\n", - "\n", - "And when I say completely open , we offer every part of the stack that we have from the hardware and network to the software abstractions above to things that are more packaged because different organizations have different levels at which they have expertise and want to adopt technology .\n", - "\n", - "Yeah . \n", - "\n", - "I mean it 's been , mean it 's been obviously incredible . \n", - "\n", - "You know going back to AI for a second , Google , Google obviously is an early mover in AI and Google Cloud has also been through , you know , starting with TensorFlow and Vertex AI and AutoML and so many incredibly innovative technologies .\n", - "\n", - "And AI has been obviously kind of a buzzword for some time now within the industry .\n", - "\n", - "And I think we see this in use as well .\n", - "\n", - "The adoption has maybe been a bit slower than we would expected until now .\n", - "\n", - "What do you think have been the barriers to greater levels of AI adoption , greater levels of enterprise that 's in value from AI ?\n", - "\n", - "And what do you think the future holds ?\n", - "\n", - "So we 've worked with a huge number of companies doing work , having them adopt AI .\n", - "\n", - "A lot of the lessons we 've seen and observed from it are the barriers to adoption are rarely about the algorithm itself .\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\u001b[0m\n", - "Write a concise two line summary of the following:\n", - "\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "You know , fraud detection and banking , there 's a lot of false positives , which makes it hard to really , you know , to a very expensive people doing something called anti-money laundering .\n", - "\n", - "And our AI algorithms are really able to be super precise on detection .\n", - "\n", - "Explainability is a critical thing there , right ?\n", - "\n", - "So people ask , why did you , why did you approve , why did you flag this one and not that one ?\n", - "\n", - "Because regulators are involved .\n", - "\n", - "So explainability becomes a big deal .\n", - "\n", - "We help , we help renewal , for example , monitor all of the factories .\n", - "\n", - "The process roughly , a billion data sets every day .\n", - "\n", - "Obviously , humans can process that .\n", - "\n", - "But making it super simple to , and you guys have given all your expertise in labeling and other things , you would get a sense .\n", - "\n", - "Factory floor data is not clean data .\n", - "\n", - "And so you have to actually clean , imagine doing a billion data sets into an environment every single day .\n", - "\n", - "You have to give the data pipelines really good .\n", - "\n", - "And so a lot of technology work happens to make that possible for companies .\n", - "\n", - "Third is , if you shop at IKEA , for example , behind IKEA is systems , it 's our recommendation system .\n", - "\n", - "find IKEA is systems , it 's our recommendation system . \n", - "\n", - "And the way that people shop for furniture and products is not the same in all countries .\n", - "\n", - "And so how are you able to one deal with the benefits you get from a global model , but also to contextually the specific elements in each country because people have different buying habits . \n", - "\n", - "Those are all things that we 've learned applying our AI in different contexts in different parts of the world .\n", - "\n", - "Yeah . \n", - "\n", - "You 've sort of glossed over this , but you 've led since you took over at Google Cloud , just a meteoric growth of the platform .\n", - "\n", - "You know , I think the past few years , you 've tripled your sales force and ending last year , you obviously ca n't come in this , but end the last year at , I believe , 20 billion of annual revenue , which is incredible and this incredible growth journey .\n", - "\n", - "What do you attribute your success to ?\n", - "\n", - "And how do you think you 've been able to drive just to an incredible growth and success ?\n", - "\n", - "From our point of view , every industry , virtually in the world , is now becoming a software powered technology industry .\n", - "\n", - "If you talk to automobile companies , they 're increasingly vehicles are more about software than mechanical systems .\n", - "\n", - "If you talk to telecommunications companies , the networks are commodities unless they can make them platforms to deliver applications , so they need new ways to slice , manage the network .\n", - "\n", - "If you look at banks at the end of the day , they 're about all the products of a bank or data , and all of that becomes how do you differentiate in the value delivering clients through a digital medium ?\n", - "\n", - "Because increasingly , I 'm sure all of you look at yourselves and go when was the last time I went to a branch of a bank .\n", - "\n", - "So a lot of our work has been pushing the technology innovation really far , but bringing that technology super easily to people in different industries .\n", - "\n", - "And given the demand that people have for a hair , I really want , I need the technology to help me power my industry , the change I 'm seeing in my industry , the more accessible we can make it , the easier and the faster we get adoption , and our approach has been to be completely open . \n", - "\n", - "And when to be completely open .\n", - "\n", - "And when I say completely open , we offer every part of the stack that we have from the hardware and network to the software abstractions above to things that are more packaged because different organizations have different levels at which they have expertise and want to adopt technology .\n", - "\n", - "Yeah . \n", - "\n", - "I mean it 's been , mean it 's been obviously incredible . \n", - "\n", - "You know going back to AI for a second , Google , Google obviously is an early mover in AI and Google Cloud has also been through , you know , starting with TensorFlow and Vertex AI and AutoML and so many incredibly innovative technologies .\n", - "\n", - "And AI has been obviously kind of a buzzword for some time now within the industry .\n", - "\n", - "And I think we see this in use as well .\n", - "\n", - "The adoption has maybe been a bit slower than we would expected until now .\n", - "\n", - "What do you think have been the barriers to greater levels of AI adoption , greater levels of enterprise that 's in value from AI ?\n", - "\n", - "And what do you think the future holds ?\n", - "\n", - "So we 've worked with a huge number of companies doing work , having them adopt AI .\n", - "\n", - "A lot of the lessons we 've seen and observed from it are the barriers to adoption are rarely about the algorithm itself .\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - " The author works closely with customers using artificial intelligence (AI) solutions provided by their company. They describe three areas where these customer projects apply advanced computer vision, natural language processing/understanding models trained via deep neural nets; 1.) Fraud Detection & Anti Money Laundering 2.), Factory Floor Monitoring 3). Recommendation Systems used across multiple vertical markets including retail shopping experiences offered online /in store environments globally - serving up personalized product recommendations based upon user behavior patterns detected throughout massive datasets collected daily generated during millions transactions processed per minute! All while maintaining strict privacy standards required under GDPR regulations enforced today worldwide due to recent EU legislation passed earlier 2018 requiring stringent consent requirements before collecting personal identifiable information pertaining directly towards individual users browsing activity conducted whilst visiting websites operated publicly available internet spaces owned privately maintained cloud computing services hosted internally private corporate intranets dedicated exclusively internal employees only!!!!!! \n", - " In addition providing highly accurate realtime predictions regarding potential fraudulent activities perpetrated against major international finance institutions operating out hong kongshanghai commercial hub located downtown core central district surrounded heavily dense population density densely populated urban area situated southern hemisphere equatorial region tropical climate zone characterized warm humid subtropical monsoon influenced weather pattern typical rainy season lasts nine months wettest month june average\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mSummarize the text below in a subject line:\n", - "\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "You know , fraud detection and banking , there 's a lot of false positives , which makes it hard to really , you know , to a very expensive people doing something called anti-money laundering .\n", - "\n", - "And our AI algorithms are really able to be super precise on detection .\n", - "\n", - "Explainability is a critical thing there , right ?\n", - "\n", - "So people ask , why did you , why did you approve , why did you flag this one and not that one ?\n", - "\n", - "Because regulators are involved .\n", - "\n", - "So explainability becomes a big deal .\n", - "\n", - "We help , we help renewal , for example , monitor all of the factories .\n", - "\n", - "The process roughly , a billion data sets every day .\n", - "\n", - "Obviously , humans can process that .\n", - "\n", - "But making it super simple to , and you guys have given all your expertise in labeling and other things , you would get a sense .\n", - "\n", - "Factory floor data is not clean data .\n", - "\n", - "And so you have to actually clean , imagine doing a billion data sets into an environment every single day .\n", - "\n", - "You have to give the data pipelines really good .\n", - "\n", - "And so a lot of technology work happens to make that possible for companies .\n", - "\n", - "Third is , if you shop at IKEA , for example , behind IKEA is systems , it 's our recommendation system .\n", - "\n", - "find IKEA is systems , it 's our recommendation system . \n", - "\n", - "And the way that people shop for furniture and products is not the same in all countries .\n", - "\n", - "And so how are you able to one deal with the benefits you get from a global model , but also to contextually the specific elements in each country because people have different buying habits . \n", - "\n", - "Those are all things that we 've learned applying our AI in different contexts in different parts of the world .\n", - "\n", - "Yeah . \n", - "\n", - "You 've sort of glossed over this , but you 've led since you took over at Google Cloud , just a meteoric growth of the platform .\n", - "\n", - "You know , I think the past few years , you 've tripled your sales force and ending last year , you obviously ca n't come in this , but end the last year at , I believe , 20 billion of annual revenue , which is incredible and this incredible growth journey .\n", - "\n", - "What do you attribute your success to ?\n", - "\n", - "And how do you think you 've been able to drive just to an incredible growth and success ?\n", - "\n", - "From our point of view , every industry , virtually in the world , is now becoming a software powered technology industry .\n", - "\n", - "If you talk to automobile companies , they 're increasingly vehicles are more about software than mechanical systems .\n", - "\n", - "If you talk to telecommunications companies , the networks are commodities unless they can make them platforms to deliver applications , so they need new ways to slice , manage the network .\n", - "\n", - "If you look at banks at the end of the day , they 're about all the products of a bank or data , and all of that becomes how do you differentiate in the value delivering clients through a digital medium ?\n", - "\n", - "Because increasingly , I 'm sure all of you look at yourselves and go when was the last time I went to a branch of a bank .\n", - "\n", - "So a lot of our work has been pushing the technology innovation really far , but bringing that technology super easily to people in different industries .\n", - "\n", - "And given the demand that people have for a hair , I really want , I need the technology to help me power my industry , the change I 'm seeing in my industry , the more accessible we can make it , the easier and the faster we get adoption , and our approach has been to be completely open . \n", - "\n", - "And when to be completely open .\n", - "\n", - "And when I say completely open , we offer every part of the stack that we have from the hardware and network to the software abstractions above to things that are more packaged because different organizations have different levels at which they have expertise and want to adopt technology .\n", - "\n", - "Yeah . \n", - "\n", - "I mean it 's been , mean it 's been obviously incredible . \n", - "\n", - "You know going back to AI for a second , Google , Google obviously is an early mover in AI and Google Cloud has also been through , you know , starting with TensorFlow and Vertex AI and AutoML and so many incredibly innovative technologies .\n", - "\n", - "And AI has been obviously kind of a buzzword for some time now within the industry .\n", - "\n", - "And I think we see this in use as well .\n", - "\n", - "The adoption has maybe been a bit slower than we would expected until now .\n", - "\n", - "What do you think have been the barriers to greater levels of AI adoption , greater levels of enterprise that 's in value from AI ?\n", - "\n", - "And what do you think the future holds ?\n", - "\n", - "So we 've worked with a huge number of companies doing work , having them adopt AI .\n", - "\n", - "A lot of the lessons we 've seen and observed from it are the barriers to adoption are rarely about the algorithm itself .\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\u001b[0m\n", - "Summarize the text below in a subject line:\n", - "\n", - "\n", - "We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud .\n", - "\n", - "You know , fraud detection and banking , there 's a lot of false positives , which makes it hard to really , you know , to a very expensive people doing something called anti-money laundering .\n", - "\n", - "And our AI algorithms are really able to be super precise on detection .\n", - "\n", - "Explainability is a critical thing there , right ?\n", - "\n", - "So people ask , why did you , why did you approve , why did you flag this one and not that one ?\n", - "\n", - "Because regulators are involved .\n", - "\n", - "So explainability becomes a big deal .\n", - "\n", - "We help , we help renewal , for example , monitor all of the factories .\n", - "\n", - "The process roughly , a billion data sets every day .\n", - "\n", - "Obviously , humans can process that .\n", - "\n", - "But making it super simple to , and you guys have given all your expertise in labeling and other things , you would get a sense .\n", - "\n", - "Factory floor data is not clean data .\n", - "\n", - "And so you have to actually clean , imagine doing a billion data sets into an environment every single day .\n", - "\n", - "You have to give the data pipelines really good .\n", - "\n", - "And so a lot of technology work happens to make that possible for companies .\n", - "\n", - "Third is , if you shop at IKEA , for example , behind IKEA is systems , it 's our recommendation system .\n", - "\n", - "find IKEA is systems , it 's our recommendation system . \n", - "\n", - "And the way that people shop for furniture and products is not the same in all countries .\n", - "\n", - "And so how are you able to one deal with the benefits you get from a global model , but also to contextually the specific elements in each country because people have different buying habits . \n", - "\n", - "Those are all things that we 've learned applying our AI in different contexts in different parts of the world .\n", - "\n", - "Yeah . \n", - "\n", - "You 've sort of glossed over this , but you 've led since you took over at Google Cloud , just a meteoric growth of the platform .\n", - "\n", - "You know , I think the past few years , you 've tripled your sales force and ending last year , you obviously ca n't come in this , but end the last year at , I believe , 20 billion of annual revenue , which is incredible and this incredible growth journey .\n", - "\n", - "What do you attribute your success to ?\n", - "\n", - "And how do you think you 've been able to drive just to an incredible growth and success ?\n", - "\n", - "From our point of view , every industry , virtually in the world , is now becoming a software powered technology industry .\n", - "\n", - "If you talk to automobile companies , they 're increasingly vehicles are more about software than mechanical systems .\n", - "\n", - "If you talk to telecommunications companies , the networks are commodities unless they can make them platforms to deliver applications , so they need new ways to slice , manage the network .\n", - "\n", - "If you look at banks at the end of the day , they 're about all the products of a bank or data , and all of that becomes how do you differentiate in the value delivering clients through a digital medium ?\n", - "\n", - "Because increasingly , I 'm sure all of you look at yourselves and go when was the last time I went to a branch of a bank .\n", - "\n", - "So a lot of our work has been pushing the technology innovation really far , but bringing that technology super easily to people in different industries .\n", - "\n", - "And given the demand that people have for a hair , I really want , I need the technology to help me power my industry , the change I 'm seeing in my industry , the more accessible we can make it , the easier and the faster we get adoption , and our approach has been to be completely open . \n", - "\n", - "And when to be completely open .\n", - "\n", - "And when I say completely open , we offer every part of the stack that we have from the hardware and network to the software abstractions above to things that are more packaged because different organizations have different levels at which they have expertise and want to adopt technology .\n", - "\n", - "Yeah . \n", - "\n", - "I mean it 's been , mean it 's been obviously incredible . \n", - "\n", - "You know going back to AI for a second , Google , Google obviously is an early mover in AI and Google Cloud has also been through , you know , starting with TensorFlow and Vertex AI and AutoML and so many incredibly innovative technologies .\n", - "\n", - "And AI has been obviously kind of a buzzword for some time now within the industry .\n", - "\n", - "And I think we see this in use as well .\n", - "\n", - "The adoption has maybe been a bit slower than we would expected until now .\n", - "\n", - "What do you think have been the barriers to greater levels of AI adoption , greater levels of enterprise that 's in value from AI ?\n", - "\n", - "And what do you think the future holds ?\n", - "\n", - "So we 've worked with a huge number of companies doing work , having them adopt AI .\n", - "\n", - "A lot of the lessons we 've seen and observed from it are the barriers to adoption are rarely about the algorithm itself .\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - " The Future Of Artificial Intelligence In Business And Society \n", - " Summary : We spoke today, April 24th 2019. You were interested hearing updates around artificial intelligence (AI) trends across business & society - specifically regarding recent advancements made by leading tech giants like Amazon Web Services Inc., Microsoft Corp.’s Azure division etc.. Also discussed key challenges faced while implementing these solutions internally including lack of skilled talent pool available globally! Finally shared exciting news related towards upcoming developments planned under GCP's Machine Learning Platform space...\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mWrite a concise two line summary of the following:\n", - "\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "So when we work with customers in many , many industries , take retailers an example , and you think of a very mundane example , like recommendations , to make product discovery on the web much easier for their own products .\n", - "\n", - "The biggest challenges standardizing the meaning of the product and the catalog .\n", - "\n", - "Because unless you have a standardized definition of the products and the data behind the algorithm is clean , it 's super hard to actually get to recommendation .\n", - "\n", - "And so in the work we did with H & M , for example , or at Macy 's , or at IKEA , or Bloomingdale 's , a huge number of these brands , the big part of the program is actually how do you label and clean the data upfront and standardize it before you get into the algorithmic phase . \n", - "\n", - "So that 's one part of things we see . \n", - "\n", - "Second part is for large organizations to adopt AI , they have to need to integrate the results of the algorithm back into their core processes .\n", - "\n", - "So , you know , practical example , we work with OGE , OGE is a large , large electric producer , electricity and power producer in Europe .\n", - "\n", - "They are probably one of the largest renewable energy producer in the world .\n", - "\n", - "They use wind farms . \n", - "\n", - "One of the things they really struggled with was , how do you predict how much wind is going to be there three days from now ?\n", - "\n", - "Because the power grid requires that prediction in order to capacity plan how much power is going into the grid . \n", - "\n", - "So they work with us and they use our AI to do that . \n", - "\n", - "But that needs to be tied into how they 're telling the rest of the power sources that work on the grid . \n", - "\n", - "Hey , if this went to wind is coming in , here 's all the other sources in each generation . \n", - "\n", - "So tying it back in is not as simple as people think .\n", - "\n", - "And so a lot of time is that the third on the people side , there 's change management you go through to get people to trust the algorithm .\n", - "\n", - "So one of the things we 've done work with many banks , particularly during the pandemic , when the government issued small business loans . \n", - "\n", - "There was a giant bottleneck in being able to get loans out to individual consumers .\n", - "\n", - "And frankly , because the banks did n't want to bring a huge army of loan officers in , they had to use software and algorithms to process it . \n", - "\n", - "Now the challenge people had is they needed to trust the algorithm was being fair in saying yes to some and no to others and that it would mirror for example the recommendations that their best mortgage bankers would do , right ?\n", - "\n", - "Just as a loan office as we do .\n", - "\n", - "So it gave them the benefit of scale because we processed literally millions and millions of mortgages through our technology , but it required them to get comfortable that things like fairness and other things were working .\n", - "\n", - "So often when people look at AI , they think it 's a skills issue .\n", - "\n", - "There 's certainly a skill issue involved .\n", - "\n", - "There 's not enough talent in the ecosystem .\n", - "\n", - "But things are getting easier and easier as the models get more and more sophisticated .\n", - "\n", - "Often people forget about these other issues that are important in getting adoption .\n", - "\n", - "Yeah .\n", - "\n", - "I mean , you 're preaching the choir when you mention the data challenges that all these enterprises face and how critical that is to getting working in the early days .\n", - "\n", - "One of the things that I think is interesting about Google Cloud strategies that you really have products at different layers of the stack and different layers of closest to the bare metal all the way up to these package solutions .\n", - "\n", - "In what way do you think that the enterprise world and even the broader business world is going to adopt these AI technologies ?\n", - "\n", - "Do you think that the end state is that a lot of them are using your lower level , more infrastructure ?\n", - "\n", - "Products , or do you think that many of them are going to adopt solutions ?\n", - "\n", - "How do you think this plays out over the next few years ?\n", - "\n", - "So we offer four layers of technology for people .\n", - "\n", - "There 's a set of people who say , look , I just need your computational infrastructure , your large systems .\n", - "\n", - "We build something called tens of processing unit , which is our large scale systems . \n", - "\n", - "We 're also working with Crossing Unit , which is our large-scale systems .\n", - "\n", - "We 're also working within video to build a really high-scale GPU Bay system .\n", - "\n", - "But many people , some customers say , look , I just need access to that .\n", - "\n", - "And we make that available because the TPUs are what we use within Google .\n", - "\n", - "And we make that available along with the compilation software to optimize models on the TPUs .\n", - "\n", - "Take as an example , LG , the Korean company that makes appliances , their team is built a large , I mean , multi-hundred billion parameter model , because they wanted to make that a way that people can interact with appliances without having to press buttons on them .\n", - "\n", - "So they built a model .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\u001b[0m\n", - "Write a concise two line summary of the following:\n", - "\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "So when we work with customers in many , many industries , take retailers an example , and you think of a very mundane example , like recommendations , to make product discovery on the web much easier for their own products .\n", - "\n", - "The biggest challenges standardizing the meaning of the product and the catalog .\n", - "\n", - "Because unless you have a standardized definition of the products and the data behind the algorithm is clean , it 's super hard to actually get to recommendation .\n", - "\n", - "And so in the work we did with H & M , for example , or at Macy 's , or at IKEA , or Bloomingdale 's , a huge number of these brands , the big part of the program is actually how do you label and clean the data upfront and standardize it before you get into the algorithmic phase . \n", - "\n", - "So that 's one part of things we see . \n", - "\n", - "Second part is for large organizations to adopt AI , they have to need to integrate the results of the algorithm back into their core processes .\n", - "\n", - "So , you know , practical example , we work with OGE , OGE is a large , large electric producer , electricity and power producer in Europe .\n", - "\n", - "They are probably one of the largest renewable energy producer in the world .\n", - "\n", - "They use wind farms . \n", - "\n", - "One of the things they really struggled with was , how do you predict how much wind is going to be there three days from now ?\n", - "\n", - "Because the power grid requires that prediction in order to capacity plan how much power is going into the grid . \n", - "\n", - "So they work with us and they use our AI to do that . \n", - "\n", - "But that needs to be tied into how they 're telling the rest of the power sources that work on the grid . \n", - "\n", - "Hey , if this went to wind is coming in , here 's all the other sources in each generation . \n", - "\n", - "So tying it back in is not as simple as people think .\n", - "\n", - "And so a lot of time is that the third on the people side , there 's change management you go through to get people to trust the algorithm .\n", - "\n", - "So one of the things we 've done work with many banks , particularly during the pandemic , when the government issued small business loans . \n", - "\n", - "There was a giant bottleneck in being able to get loans out to individual consumers .\n", - "\n", - "And frankly , because the banks did n't want to bring a huge army of loan officers in , they had to use software and algorithms to process it . \n", - "\n", - "Now the challenge people had is they needed to trust the algorithm was being fair in saying yes to some and no to others and that it would mirror for example the recommendations that their best mortgage bankers would do , right ?\n", - "\n", - "Just as a loan office as we do .\n", - "\n", - "So it gave them the benefit of scale because we processed literally millions and millions of mortgages through our technology , but it required them to get comfortable that things like fairness and other things were working .\n", - "\n", - "So often when people look at AI , they think it 's a skills issue .\n", - "\n", - "There 's certainly a skill issue involved .\n", - "\n", - "There 's not enough talent in the ecosystem .\n", - "\n", - "But things are getting easier and easier as the models get more and more sophisticated .\n", - "\n", - "Often people forget about these other issues that are important in getting adoption .\n", - "\n", - "Yeah .\n", - "\n", - "I mean , you 're preaching the choir when you mention the data challenges that all these enterprises face and how critical that is to getting working in the early days .\n", - "\n", - "One of the things that I think is interesting about Google Cloud strategies that you really have products at different layers of the stack and different layers of closest to the bare metal all the way up to these package solutions .\n", - "\n", - "In what way do you think that the enterprise world and even the broader business world is going to adopt these AI technologies ?\n", - "\n", - "Do you think that the end state is that a lot of them are using your lower level , more infrastructure ?\n", - "\n", - "Products , or do you think that many of them are going to adopt solutions ?\n", - "\n", - "How do you think this plays out over the next few years ?\n", - "\n", - "So we offer four layers of technology for people .\n", - "\n", - "There 's a set of people who say , look , I just need your computational infrastructure , your large systems .\n", - "\n", - "We build something called tens of processing unit , which is our large scale systems . \n", - "\n", - "We 're also working with Crossing Unit , which is our large-scale systems .\n", - "\n", - "We 're also working within video to build a really high-scale GPU Bay system .\n", - "\n", - "But many people , some customers say , look , I just need access to that .\n", - "\n", - "And we make that available because the TPUs are what we use within Google .\n", - "\n", - "And we make that available along with the compilation software to optimize models on the TPUs .\n", - "\n", - "Take as an example , LG , the Korean company that makes appliances , their team is built a large , I mean , multi-hundred billion parameter model , because they wanted to make that a way that people can interact with appliances without having to press buttons on them .\n", - "\n", - "So they built a model .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - " It's common for companies adopting artificial intelligence (AI) tools, such as machine learning frameworks used by computer scientists/data engineers. The most difficult aspect isn’t necessarily understanding complex mathematical concepts underlying those techniques; rather, implementing new methods successfully depends heavily upon cleaning existing datasets accurately prior to training any given ML framework -- otherwise known simply as “labeling” said dataset correctly beforehand! This step alone may prove challenging due to differences between internal definitions across multiple departments responsible managing inventory lists containing thousands items sold online via ecommerce platforms operated independently throughout separate divisions under corporate umbrella...or perhaps lack thereof altogether? Regardless whether dealing directly wth IT teams tasked maintaining servers hosting proprietary databases storing sensitive customer information collected digitally while browsing website(s), marketing department attempting convince upper levels executive board members invest budget dollars allocated towards advertising campaigns promoting latest fashion trends advertised exclusively mobile applications developed entirely housekeeping staff hired specifically maintain physical appearance home environment maintained impeccable condition despite children running wild upstairs bedrooms downstairs kitchen dining room living area basement garage attic storage closet laundry utility bathroom toilet guest bedroom nursery playroom study library hallway entryway stairwell hallways closets pantry mudrooms attics basements crawl spaces garages sheds barn stables corrals stalls kennels paddocks aren\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mSummarize the text below in a subject line:\n", - "\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "So when we work with customers in many , many industries , take retailers an example , and you think of a very mundane example , like recommendations , to make product discovery on the web much easier for their own products .\n", - "\n", - "The biggest challenges standardizing the meaning of the product and the catalog .\n", - "\n", - "Because unless you have a standardized definition of the products and the data behind the algorithm is clean , it 's super hard to actually get to recommendation .\n", - "\n", - "And so in the work we did with H & M , for example , or at Macy 's , or at IKEA , or Bloomingdale 's , a huge number of these brands , the big part of the program is actually how do you label and clean the data upfront and standardize it before you get into the algorithmic phase . \n", - "\n", - "So that 's one part of things we see . \n", - "\n", - "Second part is for large organizations to adopt AI , they have to need to integrate the results of the algorithm back into their core processes .\n", - "\n", - "So , you know , practical example , we work with OGE , OGE is a large , large electric producer , electricity and power producer in Europe .\n", - "\n", - "They are probably one of the largest renewable energy producer in the world .\n", - "\n", - "They use wind farms . \n", - "\n", - "One of the things they really struggled with was , how do you predict how much wind is going to be there three days from now ?\n", - "\n", - "Because the power grid requires that prediction in order to capacity plan how much power is going into the grid . \n", - "\n", - "So they work with us and they use our AI to do that . \n", - "\n", - "But that needs to be tied into how they 're telling the rest of the power sources that work on the grid . \n", - "\n", - "Hey , if this went to wind is coming in , here 's all the other sources in each generation . \n", - "\n", - "So tying it back in is not as simple as people think .\n", - "\n", - "And so a lot of time is that the third on the people side , there 's change management you go through to get people to trust the algorithm .\n", - "\n", - "So one of the things we 've done work with many banks , particularly during the pandemic , when the government issued small business loans . \n", - "\n", - "There was a giant bottleneck in being able to get loans out to individual consumers .\n", - "\n", - "And frankly , because the banks did n't want to bring a huge army of loan officers in , they had to use software and algorithms to process it . \n", - "\n", - "Now the challenge people had is they needed to trust the algorithm was being fair in saying yes to some and no to others and that it would mirror for example the recommendations that their best mortgage bankers would do , right ?\n", - "\n", - "Just as a loan office as we do .\n", - "\n", - "So it gave them the benefit of scale because we processed literally millions and millions of mortgages through our technology , but it required them to get comfortable that things like fairness and other things were working .\n", - "\n", - "So often when people look at AI , they think it 's a skills issue .\n", - "\n", - "There 's certainly a skill issue involved .\n", - "\n", - "There 's not enough talent in the ecosystem .\n", - "\n", - "But things are getting easier and easier as the models get more and more sophisticated .\n", - "\n", - "Often people forget about these other issues that are important in getting adoption .\n", - "\n", - "Yeah .\n", - "\n", - "I mean , you 're preaching the choir when you mention the data challenges that all these enterprises face and how critical that is to getting working in the early days .\n", - "\n", - "One of the things that I think is interesting about Google Cloud strategies that you really have products at different layers of the stack and different layers of closest to the bare metal all the way up to these package solutions .\n", - "\n", - "In what way do you think that the enterprise world and even the broader business world is going to adopt these AI technologies ?\n", - "\n", - "Do you think that the end state is that a lot of them are using your lower level , more infrastructure ?\n", - "\n", - "Products , or do you think that many of them are going to adopt solutions ?\n", - "\n", - "How do you think this plays out over the next few years ?\n", - "\n", - "So we offer four layers of technology for people .\n", - "\n", - "There 's a set of people who say , look , I just need your computational infrastructure , your large systems .\n", - "\n", - "We build something called tens of processing unit , which is our large scale systems . \n", - "\n", - "We 're also working with Crossing Unit , which is our large-scale systems .\n", - "\n", - "We 're also working within video to build a really high-scale GPU Bay system .\n", - "\n", - "But many people , some customers say , look , I just need access to that .\n", - "\n", - "And we make that available because the TPUs are what we use within Google .\n", - "\n", - "And we make that available along with the compilation software to optimize models on the TPUs .\n", - "\n", - "Take as an example , LG , the Korean company that makes appliances , their team is built a large , I mean , multi-hundred billion parameter model , because they wanted to make that a way that people can interact with appliances without having to press buttons on them .\n", - "\n", - "So they built a model .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\u001b[0m\n", - "Summarize the text below in a subject line:\n", - "\n", - "\n", - "It 's often the barriers to adoption about very algorithm itself . \n", - "\n", - "It 's often the various adoption about very different things .\n", - "\n", - "So when we work with customers in many , many industries , take retailers an example , and you think of a very mundane example , like recommendations , to make product discovery on the web much easier for their own products .\n", - "\n", - "The biggest challenges standardizing the meaning of the product and the catalog .\n", - "\n", - "Because unless you have a standardized definition of the products and the data behind the algorithm is clean , it 's super hard to actually get to recommendation .\n", - "\n", - "And so in the work we did with H & M , for example , or at Macy 's , or at IKEA , or Bloomingdale 's , a huge number of these brands , the big part of the program is actually how do you label and clean the data upfront and standardize it before you get into the algorithmic phase . \n", - "\n", - "So that 's one part of things we see . \n", - "\n", - "Second part is for large organizations to adopt AI , they have to need to integrate the results of the algorithm back into their core processes .\n", - "\n", - "So , you know , practical example , we work with OGE , OGE is a large , large electric producer , electricity and power producer in Europe .\n", - "\n", - "They are probably one of the largest renewable energy producer in the world .\n", - "\n", - "They use wind farms . \n", - "\n", - "One of the things they really struggled with was , how do you predict how much wind is going to be there three days from now ?\n", - "\n", - "Because the power grid requires that prediction in order to capacity plan how much power is going into the grid . \n", - "\n", - "So they work with us and they use our AI to do that . \n", - "\n", - "But that needs to be tied into how they 're telling the rest of the power sources that work on the grid . \n", - "\n", - "Hey , if this went to wind is coming in , here 's all the other sources in each generation . \n", - "\n", - "So tying it back in is not as simple as people think .\n", - "\n", - "And so a lot of time is that the third on the people side , there 's change management you go through to get people to trust the algorithm .\n", - "\n", - "So one of the things we 've done work with many banks , particularly during the pandemic , when the government issued small business loans . \n", - "\n", - "There was a giant bottleneck in being able to get loans out to individual consumers .\n", - "\n", - "And frankly , because the banks did n't want to bring a huge army of loan officers in , they had to use software and algorithms to process it . \n", - "\n", - "Now the challenge people had is they needed to trust the algorithm was being fair in saying yes to some and no to others and that it would mirror for example the recommendations that their best mortgage bankers would do , right ?\n", - "\n", - "Just as a loan office as we do .\n", - "\n", - "So it gave them the benefit of scale because we processed literally millions and millions of mortgages through our technology , but it required them to get comfortable that things like fairness and other things were working .\n", - "\n", - "So often when people look at AI , they think it 's a skills issue .\n", - "\n", - "There 's certainly a skill issue involved .\n", - "\n", - "There 's not enough talent in the ecosystem .\n", - "\n", - "But things are getting easier and easier as the models get more and more sophisticated .\n", - "\n", - "Often people forget about these other issues that are important in getting adoption .\n", - "\n", - "Yeah .\n", - "\n", - "I mean , you 're preaching the choir when you mention the data challenges that all these enterprises face and how critical that is to getting working in the early days .\n", - "\n", - "One of the things that I think is interesting about Google Cloud strategies that you really have products at different layers of the stack and different layers of closest to the bare metal all the way up to these package solutions .\n", - "\n", - "In what way do you think that the enterprise world and even the broader business world is going to adopt these AI technologies ?\n", - "\n", - "Do you think that the end state is that a lot of them are using your lower level , more infrastructure ?\n", - "\n", - "Products , or do you think that many of them are going to adopt solutions ?\n", - "\n", - "How do you think this plays out over the next few years ?\n", - "\n", - "So we offer four layers of technology for people .\n", - "\n", - "There 's a set of people who say , look , I just need your computational infrastructure , your large systems .\n", - "\n", - "We build something called tens of processing unit , which is our large scale systems . \n", - "\n", - "We 're also working with Crossing Unit , which is our large-scale systems .\n", - "\n", - "We 're also working within video to build a really high-scale GPU Bay system .\n", - "\n", - "But many people , some customers say , look , I just need access to that .\n", - "\n", - "And we make that available because the TPUs are what we use within Google .\n", - "\n", - "And we make that available along with the compilation software to optimize models on the TPUs .\n", - "\n", - "Take as an example , LG , the Korean company that makes appliances , their team is built a large , I mean , multi-hundred billion parameter model , because they wanted to make that a way that people can interact with appliances without having to press buttons on them .\n", - "\n", - "So they built a model .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - " The future will belong to those companies whose employees understand artificial intelligence. - Peter Thiel, co founder PayPal \n", - "\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mWrite a concise two line summary of the following:\n", - "\n", - "\n", - "So they built a model . \n", - "\n", - "They said , I just need access to your infrastructures .\n", - "\n", - "That 's one way we offer a peak capability .\n", - "\n", - "A second level is people say look , I really do n't need access to the raw infrastructure itself .\n", - "\n", - "What I need is the ability to build models using your platform . \n", - "\n", - "And so we offer a platform called Vertex and people build models and push them using our machine learning platform .\n", - "\n", - "And there are many , many organizations in logistics and financial services in retail and others who build their own models on top of the platform .\n", - "\n", - "The third is to make things even easier , we 've taken some of the core pieces , translation , documents , image processing , video .\n", - "\n", - "And we 've said , we can offer an auto-email based solution , which further simplifies how you use our platforms .\n", - "\n", - "And so for example , translation , we have a capability to handle translation in 135 languages .\n", - "\n", - "One of the important things that people ask when they go to many languages is if you look at the data sets that I used to train models , they are primarily , there 's a large set in English , because you have the whole internet is primarily in a very small number of languages .\n", - "\n", - "But once you get to more narrow languages , for instance , Swahili or some of the African languages , or even in Asia , there are many languages , even from where I grew up in India .\n", - "\n", - "There are languages that are not as widely represented on the internet .\n", - "\n", - "Can your model in translation provide equivalent fidelity in sparse languages ?\n", - "\n", - "Because it 's always important to those people only understand that language that they get a high fidelity result .\n", - "\n", - "So we 've built something called translation hub and it 's being used in very mundane places but with extraordinary impact .\n", - "\n", - "For example , when people announce COVID guidelines or recently monkey parks , for example , which is another thing , they needed translate many , many languages .\n", - "\n", - "And normally the process would take a long time .\n", - "\n", - "We have movie studios , for example , in a different example , saying , hey , when we launch a movie , we have a high fidelity set of languages , we 're actually going to hold the movie up and show that people do it .\n", - "\n", - "But for the long tail , we just need captioning .\n", - "\n", - "We 're not necessarily going to do voice dubbing .\n", - "\n", - "We 're going to do captioning .\n", - "\n", - "And they use our translation solutions to go to that .\n", - "\n", - "Even within companies , every medicine , for example , uses it to translate all their instruction manuals into many languages for their technicians .\n", - "\n", - "And then lastly , in some places , there are companies like retailers who tell us , look , a handful of the largest retailers may build their own software teams .\n", - "\n", - "But some of us who are small merchants , we 're not software companies .\n", - "\n", - "And telling us , you 've got to be a software company to use AI is not fair .\n", - "\n", - "So for some industries , we actually build fully packet solutions .\n", - "\n", - "If you call many telephone companies , the context center , behind it , sits our voice agent .\n", - "\n", - "And the rationale behind that was super simple , when a new smartphone launches like an iPhone or a Pixel , typically in the morning of the launch , some of these contact centers get three , four million calls in an hour .\n", - "\n", - "And it 's hard to hire that many agents to handle the phones .\n", - "\n", - "So we said , why would n't software be able to handle it ?\n", - "\n", - "We then evolved it so that the natural language interface can become actually the workflow for these organizations .\n", - "\n", - "But that 's a much more of a package solution so that telephone companies do n't have to have armies of data scientists to do it .\n", - "\n", - "So our work spans all of these because people have different needs and we find that as you improve the maturation of this and you make it more easy for people to adopt it .\n", - "\n", - "You will get broader proliferation and adoption of AI as a whole .\n", - "\n", - "Yeah , you know , you walk through so many different use cases and so many applications to the technology .\n", - "\n", - "I imagine one , and there 's so desperately , you know , everywhere from , you know , fraud detection to translation to translation of manuals , you know , there 's such a wide translation of manuals .\n", - "\n", - "There 's such a wide array of use cases .\n", - "\n", - "How do you all like Google Cloud think about helping businesses understand what is AI good for ?\n", - "\n", - "What can they use AI for ?\n", - "\n", - "There 's obviously such a wide diversity of different use cases , but what at a framework level do you tell them , how can I use AI within my business ?\n", - "\n", - "It 's a really good question .\n", - "\n", - "I mean , a lot of our work actually comes from clients asking us now , and that 's actually an encouraging thing .\n", - "\n", - "Because you know , see from up on the view , some simple things , how many of you believe in a few years ' time there 's gon na be intelligence software and non-intelligence software , right ? \n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\u001b[0m\n", - "Write a concise two line summary of the following:\n", - "\n", - "\n", - "So they built a model . \n", - "\n", - "They said , I just need access to your infrastructures .\n", - "\n", - "That 's one way we offer a peak capability .\n", - "\n", - "A second level is people say look , I really do n't need access to the raw infrastructure itself .\n", - "\n", - "What I need is the ability to build models using your platform . \n", - "\n", - "And so we offer a platform called Vertex and people build models and push them using our machine learning platform .\n", - "\n", - "And there are many , many organizations in logistics and financial services in retail and others who build their own models on top of the platform .\n", - "\n", - "The third is to make things even easier , we 've taken some of the core pieces , translation , documents , image processing , video .\n", - "\n", - "And we 've said , we can offer an auto-email based solution , which further simplifies how you use our platforms .\n", - "\n", - "And so for example , translation , we have a capability to handle translation in 135 languages .\n", - "\n", - "One of the important things that people ask when they go to many languages is if you look at the data sets that I used to train models , they are primarily , there 's a large set in English , because you have the whole internet is primarily in a very small number of languages .\n", - "\n", - "But once you get to more narrow languages , for instance , Swahili or some of the African languages , or even in Asia , there are many languages , even from where I grew up in India .\n", - "\n", - "There are languages that are not as widely represented on the internet .\n", - "\n", - "Can your model in translation provide equivalent fidelity in sparse languages ?\n", - "\n", - "Because it 's always important to those people only understand that language that they get a high fidelity result .\n", - "\n", - "So we 've built something called translation hub and it 's being used in very mundane places but with extraordinary impact .\n", - "\n", - "For example , when people announce COVID guidelines or recently monkey parks , for example , which is another thing , they needed translate many , many languages .\n", - "\n", - "And normally the process would take a long time .\n", - "\n", - "We have movie studios , for example , in a different example , saying , hey , when we launch a movie , we have a high fidelity set of languages , we 're actually going to hold the movie up and show that people do it .\n", - "\n", - "But for the long tail , we just need captioning .\n", - "\n", - "We 're not necessarily going to do voice dubbing .\n", - "\n", - "We 're going to do captioning .\n", - "\n", - "And they use our translation solutions to go to that .\n", - "\n", - "Even within companies , every medicine , for example , uses it to translate all their instruction manuals into many languages for their technicians .\n", - "\n", - "And then lastly , in some places , there are companies like retailers who tell us , look , a handful of the largest retailers may build their own software teams .\n", - "\n", - "But some of us who are small merchants , we 're not software companies .\n", - "\n", - "And telling us , you 've got to be a software company to use AI is not fair .\n", - "\n", - "So for some industries , we actually build fully packet solutions .\n", - "\n", - "If you call many telephone companies , the context center , behind it , sits our voice agent .\n", - "\n", - "And the rationale behind that was super simple , when a new smartphone launches like an iPhone or a Pixel , typically in the morning of the launch , some of these contact centers get three , four million calls in an hour .\n", - "\n", - "And it 's hard to hire that many agents to handle the phones .\n", - "\n", - "So we said , why would n't software be able to handle it ?\n", - "\n", - "We then evolved it so that the natural language interface can become actually the workflow for these organizations .\n", - "\n", - "But that 's a much more of a package solution so that telephone companies do n't have to have armies of data scientists to do it .\n", - "\n", - "So our work spans all of these because people have different needs and we find that as you improve the maturation of this and you make it more easy for people to adopt it .\n", - "\n", - "You will get broader proliferation and adoption of AI as a whole .\n", - "\n", - "Yeah , you know , you walk through so many different use cases and so many applications to the technology .\n", - "\n", - "I imagine one , and there 's so desperately , you know , everywhere from , you know , fraud detection to translation to translation of manuals , you know , there 's such a wide translation of manuals .\n", - "\n", - "There 's such a wide array of use cases .\n", - "\n", - "How do you all like Google Cloud think about helping businesses understand what is AI good for ?\n", - "\n", - "What can they use AI for ?\n", - "\n", - "There 's obviously such a wide diversity of different use cases , but what at a framework level do you tell them , how can I use AI within my business ?\n", - "\n", - "It 's a really good question .\n", - "\n", - "I mean , a lot of our work actually comes from clients asking us now , and that 's actually an encouraging thing .\n", - "\n", - "Because you know , see from up on the view , some simple things , how many of you believe in a few years ' time there 's gon na be intelligence software and non-intelligence software , right ? \n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - " So, according to DeepMind's CEO, Demis Hassabis (as quoted), \"AI has been around since 1956. It took until 2016 before anyone realized its potential.\" The quote refers specifically to artificial general intelligence -- AGI - i.e., machines capable of intelligent behavior similar to humans'. In fact, he goes onto state explicitly later in his talk :\"AGI could change everything\". This statement reflects current thinking among leading experts working in the field; however, most agree that achieving full human parity remains extremely challenging given the vast complexity involved.[1] Despite this challenge, significant progress towards developing systems demonstrating strong forms of intelligence continues apace across multiple disciplines including computer vision,[2][3] speech recognition[4], robotics [5],[6]and Natural Language Processing(NLP) / Machine Learning ([7]). As noted by Dr Ian Goodfellow during his recent TED Talk entitled “Deepfakes & Fake News”, NLP represents perhaps the single greatest area of active research today due to both its broad applicability throughout society coupled wth rapid advances made possible via deep neural networks trained over massive datasets provided freely online courtesy of giants like google/youtube etc... \n", - "\n", - "In addition to providing powerful tools enabling researchers worldwide develop increasingly\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mSummarize the text below in a subject line:\n", - "\n", - "\n", - "So they built a model . \n", - "\n", - "They said , I just need access to your infrastructures .\n", - "\n", - "That 's one way we offer a peak capability .\n", - "\n", - "A second level is people say look , I really do n't need access to the raw infrastructure itself .\n", - "\n", - "What I need is the ability to build models using your platform . \n", - "\n", - "And so we offer a platform called Vertex and people build models and push them using our machine learning platform .\n", - "\n", - "And there are many , many organizations in logistics and financial services in retail and others who build their own models on top of the platform .\n", - "\n", - "The third is to make things even easier , we 've taken some of the core pieces , translation , documents , image processing , video .\n", - "\n", - "And we 've said , we can offer an auto-email based solution , which further simplifies how you use our platforms .\n", - "\n", - "And so for example , translation , we have a capability to handle translation in 135 languages .\n", - "\n", - "One of the important things that people ask when they go to many languages is if you look at the data sets that I used to train models , they are primarily , there 's a large set in English , because you have the whole internet is primarily in a very small number of languages .\n", - "\n", - "But once you get to more narrow languages , for instance , Swahili or some of the African languages , or even in Asia , there are many languages , even from where I grew up in India .\n", - "\n", - "There are languages that are not as widely represented on the internet .\n", - "\n", - "Can your model in translation provide equivalent fidelity in sparse languages ?\n", - "\n", - "Because it 's always important to those people only understand that language that they get a high fidelity result .\n", - "\n", - "So we 've built something called translation hub and it 's being used in very mundane places but with extraordinary impact .\n", - "\n", - "For example , when people announce COVID guidelines or recently monkey parks , for example , which is another thing , they needed translate many , many languages .\n", - "\n", - "And normally the process would take a long time .\n", - "\n", - "We have movie studios , for example , in a different example , saying , hey , when we launch a movie , we have a high fidelity set of languages , we 're actually going to hold the movie up and show that people do it .\n", - "\n", - "But for the long tail , we just need captioning .\n", - "\n", - "We 're not necessarily going to do voice dubbing .\n", - "\n", - "We 're going to do captioning .\n", - "\n", - "And they use our translation solutions to go to that .\n", - "\n", - "Even within companies , every medicine , for example , uses it to translate all their instruction manuals into many languages for their technicians .\n", - "\n", - "And then lastly , in some places , there are companies like retailers who tell us , look , a handful of the largest retailers may build their own software teams .\n", - "\n", - "But some of us who are small merchants , we 're not software companies .\n", - "\n", - "And telling us , you 've got to be a software company to use AI is not fair .\n", - "\n", - "So for some industries , we actually build fully packet solutions .\n", - "\n", - "If you call many telephone companies , the context center , behind it , sits our voice agent .\n", - "\n", - "And the rationale behind that was super simple , when a new smartphone launches like an iPhone or a Pixel , typically in the morning of the launch , some of these contact centers get three , four million calls in an hour .\n", - "\n", - "And it 's hard to hire that many agents to handle the phones .\n", - "\n", - "So we said , why would n't software be able to handle it ?\n", - "\n", - "We then evolved it so that the natural language interface can become actually the workflow for these organizations .\n", - "\n", - "But that 's a much more of a package solution so that telephone companies do n't have to have armies of data scientists to do it .\n", - "\n", - "So our work spans all of these because people have different needs and we find that as you improve the maturation of this and you make it more easy for people to adopt it .\n", - "\n", - "You will get broader proliferation and adoption of AI as a whole .\n", - "\n", - "Yeah , you know , you walk through so many different use cases and so many applications to the technology .\n", - "\n", - "I imagine one , and there 's so desperately , you know , everywhere from , you know , fraud detection to translation to translation of manuals , you know , there 's such a wide translation of manuals .\n", - "\n", - "There 's such a wide array of use cases .\n", - "\n", - "How do you all like Google Cloud think about helping businesses understand what is AI good for ?\n", - "\n", - "What can they use AI for ?\n", - "\n", - "There 's obviously such a wide diversity of different use cases , but what at a framework level do you tell them , how can I use AI within my business ?\n", - "\n", - "It 's a really good question .\n", - "\n", - "I mean , a lot of our work actually comes from clients asking us now , and that 's actually an encouraging thing .\n", - "\n", - "Because you know , see from up on the view , some simple things , how many of you believe in a few years ' time there 's gon na be intelligence software and non-intelligence software , right ? \n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\u001b[0m\n", - "Summarize the text below in a subject line:\n", - "\n", - "\n", - "So they built a model . \n", - "\n", - "They said , I just need access to your infrastructures .\n", - "\n", - "That 's one way we offer a peak capability .\n", - "\n", - "A second level is people say look , I really do n't need access to the raw infrastructure itself .\n", - "\n", - "What I need is the ability to build models using your platform . \n", - "\n", - "And so we offer a platform called Vertex and people build models and push them using our machine learning platform .\n", - "\n", - "And there are many , many organizations in logistics and financial services in retail and others who build their own models on top of the platform .\n", - "\n", - "The third is to make things even easier , we 've taken some of the core pieces , translation , documents , image processing , video .\n", - "\n", - "And we 've said , we can offer an auto-email based solution , which further simplifies how you use our platforms .\n", - "\n", - "And so for example , translation , we have a capability to handle translation in 135 languages .\n", - "\n", - "One of the important things that people ask when they go to many languages is if you look at the data sets that I used to train models , they are primarily , there 's a large set in English , because you have the whole internet is primarily in a very small number of languages .\n", - "\n", - "But once you get to more narrow languages , for instance , Swahili or some of the African languages , or even in Asia , there are many languages , even from where I grew up in India .\n", - "\n", - "There are languages that are not as widely represented on the internet .\n", - "\n", - "Can your model in translation provide equivalent fidelity in sparse languages ?\n", - "\n", - "Because it 's always important to those people only understand that language that they get a high fidelity result .\n", - "\n", - "So we 've built something called translation hub and it 's being used in very mundane places but with extraordinary impact .\n", - "\n", - "For example , when people announce COVID guidelines or recently monkey parks , for example , which is another thing , they needed translate many , many languages .\n", - "\n", - "And normally the process would take a long time .\n", - "\n", - "We have movie studios , for example , in a different example , saying , hey , when we launch a movie , we have a high fidelity set of languages , we 're actually going to hold the movie up and show that people do it .\n", - "\n", - "But for the long tail , we just need captioning .\n", - "\n", - "We 're not necessarily going to do voice dubbing .\n", - "\n", - "We 're going to do captioning .\n", - "\n", - "And they use our translation solutions to go to that .\n", - "\n", - "Even within companies , every medicine , for example , uses it to translate all their instruction manuals into many languages for their technicians .\n", - "\n", - "And then lastly , in some places , there are companies like retailers who tell us , look , a handful of the largest retailers may build their own software teams .\n", - "\n", - "But some of us who are small merchants , we 're not software companies .\n", - "\n", - "And telling us , you 've got to be a software company to use AI is not fair .\n", - "\n", - "So for some industries , we actually build fully packet solutions .\n", - "\n", - "If you call many telephone companies , the context center , behind it , sits our voice agent .\n", - "\n", - "And the rationale behind that was super simple , when a new smartphone launches like an iPhone or a Pixel , typically in the morning of the launch , some of these contact centers get three , four million calls in an hour .\n", - "\n", - "And it 's hard to hire that many agents to handle the phones .\n", - "\n", - "So we said , why would n't software be able to handle it ?\n", - "\n", - "We then evolved it so that the natural language interface can become actually the workflow for these organizations .\n", - "\n", - "But that 's a much more of a package solution so that telephone companies do n't have to have armies of data scientists to do it .\n", - "\n", - "So our work spans all of these because people have different needs and we find that as you improve the maturation of this and you make it more easy for people to adopt it .\n", - "\n", - "You will get broader proliferation and adoption of AI as a whole .\n", - "\n", - "Yeah , you know , you walk through so many different use cases and so many applications to the technology .\n", - "\n", - "I imagine one , and there 's so desperately , you know , everywhere from , you know , fraud detection to translation to translation of manuals , you know , there 's such a wide translation of manuals .\n", - "\n", - "There 's such a wide array of use cases .\n", - "\n", - "How do you all like Google Cloud think about helping businesses understand what is AI good for ?\n", - "\n", - "What can they use AI for ?\n", - "\n", - "There 's obviously such a wide diversity of different use cases , but what at a framework level do you tell them , how can I use AI within my business ?\n", - "\n", - "It 's a really good question .\n", - "\n", - "I mean , a lot of our work actually comes from clients asking us now , and that 's actually an encouraging thing .\n", - "\n", - "Because you know , see from up on the view , some simple things , how many of you believe in a few years ' time there 's gon na be intelligence software and non-intelligence software , right ? \n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mWrite a concise two line summary of the following:\n", - "\n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "I mean , nobody would say in three , four years ' time , we 're going to write software that has not powered in some form of fashion by AI .\n", - "\n", - "So you know , in most companies actually , it 's really encouraging to see that they look at domain problems they 're having and say , for instance , I used to do it using a rules engine , which is an older model for defining kind of workflow within organizations . \n", - "\n", - "Can you apply AI to do it in a new way ?\n", - "\n", - "I used to do this in a specific way .\n", - "\n", - "I heard about image recognition .\n", - "\n", - "One example really fun or interesting one , US Navy , when you have corrosion on the base of ships , the old way was to lift it into dry dark and take a look at it .\n", - "\n", - "If you 've ever seen one of these ships , you can imagine lifting to dry dark is not an easy thing .\n", - "\n", - "So they said , can we fly a drone with your camera image recognition around it and detect corrosion ?\n", - "\n", - "And so what we 've seen is that as you lift up the capability where image , audio , text , et cetera , all these forms of input can be processed extremely accurately , most customers start figuring it out .\n", - "\n", - "And so they call us with , most of our work has come from customers calling us , saying , hey , I have this need .\n", - "\n", - "Can I apply AI to it ?\n", - "\n", - "And so we talk to them about how and when it makes sense to use AI .\n", - "\n", - "But we also talk to them about the consequences if the models are not handling things like skew in the data .\n", - "\n", - "How do you ensure that , for example , you 're treating fairness properly ?\n", - "\n", - "How do you ensure that the model is safe , etc .\n", - "\n", - "Yeah , I think it 's , I mean , all the use cases , the variety is incredibly exciting .\n", - "\n", - "It 's cool that these customers are coming to you directly with many of them .\n", - "\n", - "What is , again , kind of thinking bigger picture , what is machine learning an AI mean for Google Cloud on the whole over the next call 510 years ?\n", - "\n", - "So we feel that the boundary of what machine learning and what AI can do will change over time .\n", - "\n", - "When it started , it was about doing what we would call assistive things .\n", - "\n", - "Assistive things are where a human being is able to do it , but the computer assists the human being in some ways to do it better .\n", - "\n", - "Right ?\n", - "\n", - "So common examples people talk about is , hey , your doctor or radiologist , you used to look at x-ray images .\n", - "\n", - "Now , a computer is going to look at it and detect tumors , but it 's assisting you to find something that you may have done another way . \n", - "\n", - "So that 's the first phase and a lot of the work we see is primarily in that phase today . \n", - "\n", - "The second phase is to do something where you could n't do it with a human because the quantity of data you need to process or the amount of people you need would be just far too significant . \n", - "\n", - "And so the machine is doing something that humans could n't do , but it 's still an incremental element on top of what humans could do themselves .\n", - "\n", - "The third phase , I think , is where we think generative AI , for example , goes , because it 's about enabling people to express themselves in a different way , and to assist them in expressiveness .\n", - "\n", - "So I 'll give you a practical example .\n", - "\n", - "A lot of you probably use tools , slides , and things like that in your day to day job .\n", - "\n", - "PowerPoint was invented a long time ago and was really just about drawing things .\n", - "\n", - "You know , I 've got a 14 year old .\n", - "\n", - "And so if you look at the younger generation , if you look at what slides were , they were really tools to help people draw .\n", - "\n", - "And then to take what was on the slide projector and presented .\n", - "\n", - "Then the younger generation says , hey , I do n't want to draw things that 's really old-fashioned .\n", - "\n", - "I 'm going to go to the internet and copy images , right ? \n", - "\n", - "Because when they do class projects , they 're copying images into the slides .\n", - "\n", - "And then , as people observe , you know , on the social media environment , people going from text , which may have been Facebook to short images , which is Instagram to short video TikTok , we would say , hey , why would n't we be able to record short video ?\n", - "\n", - "And be used that as a mechanism to share .\n", - "\n", - "But recording short video is still capturing the real world through the lens of the camera .\n", - "\n", - "What people want is a more expressive way of saying , I have an idea , can I translate it ? \n", - "\n", - "And it may not be something I can capture . \n", - "\n", - "Imagine a kid in California and a school saying saying I want to capture how the landscape and outside of Paris and France is right now .\n", - "\n", - "I think they need to be able to generate some of the ideas that they could capture by physically being there .\n", - "\n", - "And so we 're working on all of this and we 're bringing some of these into our products to change what people could possibly do through the application of AI so they improve expressiveness for people .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\u001b[0m\n", - "Write a concise two line summary of the following:\n", - "\n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "I mean , nobody would say in three , four years ' time , we 're going to write software that has not powered in some form of fashion by AI .\n", - "\n", - "So you know , in most companies actually , it 's really encouraging to see that they look at domain problems they 're having and say , for instance , I used to do it using a rules engine , which is an older model for defining kind of workflow within organizations . \n", - "\n", - "Can you apply AI to do it in a new way ?\n", - "\n", - "I used to do this in a specific way .\n", - "\n", - "I heard about image recognition .\n", - "\n", - "One example really fun or interesting one , US Navy , when you have corrosion on the base of ships , the old way was to lift it into dry dark and take a look at it .\n", - "\n", - "If you 've ever seen one of these ships , you can imagine lifting to dry dark is not an easy thing .\n", - "\n", - "So they said , can we fly a drone with your camera image recognition around it and detect corrosion ?\n", - "\n", - "And so what we 've seen is that as you lift up the capability where image , audio , text , et cetera , all these forms of input can be processed extremely accurately , most customers start figuring it out .\n", - "\n", - "And so they call us with , most of our work has come from customers calling us , saying , hey , I have this need .\n", - "\n", - "Can I apply AI to it ?\n", - "\n", - "And so we talk to them about how and when it makes sense to use AI .\n", - "\n", - "But we also talk to them about the consequences if the models are not handling things like skew in the data .\n", - "\n", - "How do you ensure that , for example , you 're treating fairness properly ?\n", - "\n", - "How do you ensure that the model is safe , etc .\n", - "\n", - "Yeah , I think it 's , I mean , all the use cases , the variety is incredibly exciting .\n", - "\n", - "It 's cool that these customers are coming to you directly with many of them .\n", - "\n", - "What is , again , kind of thinking bigger picture , what is machine learning an AI mean for Google Cloud on the whole over the next call 510 years ?\n", - "\n", - "So we feel that the boundary of what machine learning and what AI can do will change over time .\n", - "\n", - "When it started , it was about doing what we would call assistive things .\n", - "\n", - "Assistive things are where a human being is able to do it , but the computer assists the human being in some ways to do it better .\n", - "\n", - "Right ?\n", - "\n", - "So common examples people talk about is , hey , your doctor or radiologist , you used to look at x-ray images .\n", - "\n", - "Now , a computer is going to look at it and detect tumors , but it 's assisting you to find something that you may have done another way . \n", - "\n", - "So that 's the first phase and a lot of the work we see is primarily in that phase today . \n", - "\n", - "The second phase is to do something where you could n't do it with a human because the quantity of data you need to process or the amount of people you need would be just far too significant . \n", - "\n", - "And so the machine is doing something that humans could n't do , but it 's still an incremental element on top of what humans could do themselves .\n", - "\n", - "The third phase , I think , is where we think generative AI , for example , goes , because it 's about enabling people to express themselves in a different way , and to assist them in expressiveness .\n", - "\n", - "So I 'll give you a practical example .\n", - "\n", - "A lot of you probably use tools , slides , and things like that in your day to day job .\n", - "\n", - "PowerPoint was invented a long time ago and was really just about drawing things .\n", - "\n", - "You know , I 've got a 14 year old .\n", - "\n", - "And so if you look at the younger generation , if you look at what slides were , they were really tools to help people draw .\n", - "\n", - "And then to take what was on the slide projector and presented .\n", - "\n", - "Then the younger generation says , hey , I do n't want to draw things that 's really old-fashioned .\n", - "\n", - "I 'm going to go to the internet and copy images , right ? \n", - "\n", - "Because when they do class projects , they 're copying images into the slides .\n", - "\n", - "And then , as people observe , you know , on the social media environment , people going from text , which may have been Facebook to short images , which is Instagram to short video TikTok , we would say , hey , why would n't we be able to record short video ?\n", - "\n", - "And be used that as a mechanism to share .\n", - "\n", - "But recording short video is still capturing the real world through the lens of the camera .\n", - "\n", - "What people want is a more expressive way of saying , I have an idea , can I translate it ? \n", - "\n", - "And it may not be something I can capture . \n", - "\n", - "Imagine a kid in California and a school saying saying I want to capture how the landscape and outside of Paris and France is right now .\n", - "\n", - "I think they need to be able to generate some of the ideas that they could capture by physically being there .\n", - "\n", - "And so we 're working on all of this and we 're bringing some of these into our products to change what people could possibly do through the application of AI so they improve expressiveness for people .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - " Machine Learning (ML) & Artificial Intelligence (AI), including Generative AI, enable computers to learn complex patterns based off large amounts of historical training datasets. ML/AI technologies power applications across industries such as healthcare, finance, retail, manufacturing - helping businesses automate processes while improving efficiency / accuracy; meanwhile empowering individuals via consumer apps e.g., photo search capabilities enabled by Image Recognition technology. Over the past decade cloud computing providers Microsoft Azure, Amazon Web Services AWS], Alphabet's GCP [Google Cloud Platform] IBM Bluemix ]have invested heavily building their own proprietary platforms leveraging open source frameworks i.e Apache Spark + TensorFlow ; offering services built upon those platform APIs -- allowing developers access without needing deep expertise required previously needed w/ legacy systems requiring custom code development.] In 2023 expect continued rapid innovation driven largely b\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mSummarize the text below in a subject line:\n", - "\n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "I mean , nobody would say in three , four years ' time , we 're going to write software that has not powered in some form of fashion by AI .\n", - "\n", - "So you know , in most companies actually , it 's really encouraging to see that they look at domain problems they 're having and say , for instance , I used to do it using a rules engine , which is an older model for defining kind of workflow within organizations . \n", - "\n", - "Can you apply AI to do it in a new way ?\n", - "\n", - "I used to do this in a specific way .\n", - "\n", - "I heard about image recognition .\n", - "\n", - "One example really fun or interesting one , US Navy , when you have corrosion on the base of ships , the old way was to lift it into dry dark and take a look at it .\n", - "\n", - "If you 've ever seen one of these ships , you can imagine lifting to dry dark is not an easy thing .\n", - "\n", - "So they said , can we fly a drone with your camera image recognition around it and detect corrosion ?\n", - "\n", - "And so what we 've seen is that as you lift up the capability where image , audio , text , et cetera , all these forms of input can be processed extremely accurately , most customers start figuring it out .\n", - "\n", - "And so they call us with , most of our work has come from customers calling us , saying , hey , I have this need .\n", - "\n", - "Can I apply AI to it ?\n", - "\n", - "And so we talk to them about how and when it makes sense to use AI .\n", - "\n", - "But we also talk to them about the consequences if the models are not handling things like skew in the data .\n", - "\n", - "How do you ensure that , for example , you 're treating fairness properly ?\n", - "\n", - "How do you ensure that the model is safe , etc .\n", - "\n", - "Yeah , I think it 's , I mean , all the use cases , the variety is incredibly exciting .\n", - "\n", - "It 's cool that these customers are coming to you directly with many of them .\n", - "\n", - "What is , again , kind of thinking bigger picture , what is machine learning an AI mean for Google Cloud on the whole over the next call 510 years ?\n", - "\n", - "So we feel that the boundary of what machine learning and what AI can do will change over time .\n", - "\n", - "When it started , it was about doing what we would call assistive things .\n", - "\n", - "Assistive things are where a human being is able to do it , but the computer assists the human being in some ways to do it better .\n", - "\n", - "Right ?\n", - "\n", - "So common examples people talk about is , hey , your doctor or radiologist , you used to look at x-ray images .\n", - "\n", - "Now , a computer is going to look at it and detect tumors , but it 's assisting you to find something that you may have done another way . \n", - "\n", - "So that 's the first phase and a lot of the work we see is primarily in that phase today . \n", - "\n", - "The second phase is to do something where you could n't do it with a human because the quantity of data you need to process or the amount of people you need would be just far too significant . \n", - "\n", - "And so the machine is doing something that humans could n't do , but it 's still an incremental element on top of what humans could do themselves .\n", - "\n", - "The third phase , I think , is where we think generative AI , for example , goes , because it 's about enabling people to express themselves in a different way , and to assist them in expressiveness .\n", - "\n", - "So I 'll give you a practical example .\n", - "\n", - "A lot of you probably use tools , slides , and things like that in your day to day job .\n", - "\n", - "PowerPoint was invented a long time ago and was really just about drawing things .\n", - "\n", - "You know , I 've got a 14 year old .\n", - "\n", - "And so if you look at the younger generation , if you look at what slides were , they were really tools to help people draw .\n", - "\n", - "And then to take what was on the slide projector and presented .\n", - "\n", - "Then the younger generation says , hey , I do n't want to draw things that 's really old-fashioned .\n", - "\n", - "I 'm going to go to the internet and copy images , right ? \n", - "\n", - "Because when they do class projects , they 're copying images into the slides .\n", - "\n", - "And then , as people observe , you know , on the social media environment , people going from text , which may have been Facebook to short images , which is Instagram to short video TikTok , we would say , hey , why would n't we be able to record short video ?\n", - "\n", - "And be used that as a mechanism to share .\n", - "\n", - "But recording short video is still capturing the real world through the lens of the camera .\n", - "\n", - "What people want is a more expressive way of saying , I have an idea , can I translate it ? \n", - "\n", - "And it may not be something I can capture . \n", - "\n", - "Imagine a kid in California and a school saying saying I want to capture how the landscape and outside of Paris and France is right now .\n", - "\n", - "I think they need to be able to generate some of the ideas that they could capture by physically being there .\n", - "\n", - "And so we 're working on all of this and we 're bringing some of these into our products to change what people could possibly do through the application of AI so they improve expressiveness for people .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\u001b[0m\n", - "Summarize the text below in a subject line:\n", - "\n", - "\n", - "I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software .\n", - "\n", - "I mean , nobody would say in three , four years ' time , we 're going to write software that has not powered in some form of fashion by AI .\n", - "\n", - "So you know , in most companies actually , it 's really encouraging to see that they look at domain problems they 're having and say , for instance , I used to do it using a rules engine , which is an older model for defining kind of workflow within organizations . \n", - "\n", - "Can you apply AI to do it in a new way ?\n", - "\n", - "I used to do this in a specific way .\n", - "\n", - "I heard about image recognition .\n", - "\n", - "One example really fun or interesting one , US Navy , when you have corrosion on the base of ships , the old way was to lift it into dry dark and take a look at it .\n", - "\n", - "If you 've ever seen one of these ships , you can imagine lifting to dry dark is not an easy thing .\n", - "\n", - "So they said , can we fly a drone with your camera image recognition around it and detect corrosion ?\n", - "\n", - "And so what we 've seen is that as you lift up the capability where image , audio , text , et cetera , all these forms of input can be processed extremely accurately , most customers start figuring it out .\n", - "\n", - "And so they call us with , most of our work has come from customers calling us , saying , hey , I have this need .\n", - "\n", - "Can I apply AI to it ?\n", - "\n", - "And so we talk to them about how and when it makes sense to use AI .\n", - "\n", - "But we also talk to them about the consequences if the models are not handling things like skew in the data .\n", - "\n", - "How do you ensure that , for example , you 're treating fairness properly ?\n", - "\n", - "How do you ensure that the model is safe , etc .\n", - "\n", - "Yeah , I think it 's , I mean , all the use cases , the variety is incredibly exciting .\n", - "\n", - "It 's cool that these customers are coming to you directly with many of them .\n", - "\n", - "What is , again , kind of thinking bigger picture , what is machine learning an AI mean for Google Cloud on the whole over the next call 510 years ?\n", - "\n", - "So we feel that the boundary of what machine learning and what AI can do will change over time .\n", - "\n", - "When it started , it was about doing what we would call assistive things .\n", - "\n", - "Assistive things are where a human being is able to do it , but the computer assists the human being in some ways to do it better .\n", - "\n", - "Right ?\n", - "\n", - "So common examples people talk about is , hey , your doctor or radiologist , you used to look at x-ray images .\n", - "\n", - "Now , a computer is going to look at it and detect tumors , but it 's assisting you to find something that you may have done another way . \n", - "\n", - "So that 's the first phase and a lot of the work we see is primarily in that phase today . \n", - "\n", - "The second phase is to do something where you could n't do it with a human because the quantity of data you need to process or the amount of people you need would be just far too significant . \n", - "\n", - "And so the machine is doing something that humans could n't do , but it 's still an incremental element on top of what humans could do themselves .\n", - "\n", - "The third phase , I think , is where we think generative AI , for example , goes , because it 's about enabling people to express themselves in a different way , and to assist them in expressiveness .\n", - "\n", - "So I 'll give you a practical example .\n", - "\n", - "A lot of you probably use tools , slides , and things like that in your day to day job .\n", - "\n", - "PowerPoint was invented a long time ago and was really just about drawing things .\n", - "\n", - "You know , I 've got a 14 year old .\n", - "\n", - "And so if you look at the younger generation , if you look at what slides were , they were really tools to help people draw .\n", - "\n", - "And then to take what was on the slide projector and presented .\n", - "\n", - "Then the younger generation says , hey , I do n't want to draw things that 's really old-fashioned .\n", - "\n", - "I 'm going to go to the internet and copy images , right ? \n", - "\n", - "Because when they do class projects , they 're copying images into the slides .\n", - "\n", - "And then , as people observe , you know , on the social media environment , people going from text , which may have been Facebook to short images , which is Instagram to short video TikTok , we would say , hey , why would n't we be able to record short video ?\n", - "\n", - "And be used that as a mechanism to share .\n", - "\n", - "But recording short video is still capturing the real world through the lens of the camera .\n", - "\n", - "What people want is a more expressive way of saying , I have an idea , can I translate it ? \n", - "\n", - "And it may not be something I can capture . \n", - "\n", - "Imagine a kid in California and a school saying saying I want to capture how the landscape and outside of Paris and France is right now .\n", - "\n", - "I think they need to be able to generate some of the ideas that they could capture by physically being there .\n", - "\n", - "And so we 're working on all of this and we 're bringing some of these into our products to change what people could possibly do through the application of AI so they improve expressiveness for people .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mWrite a concise two line summary of the following:\n", - "\n", - "\n", - "And so every boundary as the technology gets more sophisticated we think it moves from just assistance to assistance on things that human beings may not have been able to just linearly do to now things like expressiveness , which is a very different capability than people could actually do themselves .\n", - "\n", - "Yeah , I mean , all of this is very obviously incredibly exciting and we 're all watching it happen in real time .\n", - "\n", - "There 's an artist who actually described the image generation models as , he sort of image generation models as he was , he sort of said like , you kind of think about like a camera .\n", - "\n", - "Like it 's a new tool that allows you to create fundamentally new forms of art .\n", - "\n", - "That 's right .\n", - "\n", - "Yeah .\n", - "\n", - "And not just one medium of art , right ?\n", - "\n", - "Because if you look in the past , people said , you were a painter , you were a sculpture , you were a musician , and now these technologies allow you to blend all of it as a form of expressiveness .\n", - "\n", - "Yeah .\n", - "\n", - "You know , the last question I have for you is , you know , you obviously sit down with many of the sort of leading CEOs and business leaders of of the sort of largest organizations in the world . \n", - "\n", - "And I 'm sure one thing that is on many of their minds is sort of as AI technology develops and it continues to progress is potential disruption that might come from art of film intelligence .\n", - "\n", - "What sort of , how do you approach that conversation ?\n", - "\n", - "What sort of your advice to these business leaders who are looking at this powerful new technology and thinking about what that might mean for the businesses and the business landscape .\n", - "\n", - "When we talk to CEOs , I mean the biggest things we talk to them about number one , productivity in the long term , productivity has always been the primary driver of improving both company productivity , meaning their own companies , as well as societal benefit , things like affluence of a society , etc . \n", - "\n", - "And the means and equality of distribution of income to people across all spectrum society .\n", - "\n", - "Eventually , the most important metric , and you can look at any economic textbook is productivity .\n", - "\n", - "Software and technology has probably been the biggest boomer productivity over the last 30 , 40 years .\n", - "\n", - "This is the next evolution of that .\n", - "\n", - "And so we always say , if you approach it the right way , for example , labor shortages are going on right now .\n", - "\n", - "The biggest potential benefit is the application of some of these platforms like AI to do in that .\n", - "\n", - "The second , with any technological generation revolution , like artificial intelligence ,\n", - "\n", - "but if you went back in time and looked at the industrial revolution , etc . \n", - "\n", - "There are always during the period of transition , anxiety about the consequences of that technology . \n", - "\n", - "And it does n't mean the technology by itself is good or bad .\n", - "\n", - "It 's the application of the technology that 's good or bad . \n", - "\n", - "So it 's incumbent upon both the technology providers and the users of the technology to ensure that the negative consequences of it are managed properly .\n", - "\n", - "Right ?\n", - "\n", - "The obvious example is , for instance , if you look at a very simple thing , image recognition .\n", - "\n", - "Image recognition can help doctors find tumors way better than having the best radiographer .\n", - "\n", - "It 's a system in that context and it 's like helping people with a better quality microscope than they had before .\n", - "\n", - "Object recognition is helping people find , for example , people who are in the ocean much more accurately so the coastguard can rescue them .\n", - "\n", - "At the same time , being able to use a camera and say that 's Thomas Korean has , you know , a lot of potential negative consequences . \n", - "\n", - "And so as a provider of technology , we at Google have chosen not to do that third part . \n", - "\n", - "But we also tell companies , it 's important not just to say , this is what 's regularly allowed by the legal framework , because law in many countries is not yet keeping up with how fast AI technology is moving .\n", - "\n", - "But to take the responsibility as a company CEO to say , here 's what I believe comfortable with , and here 's what I wo n't be comfortable with .\n", - "\n", - "Yeah .\n", - "\n", - "Well , Thomas , thank you so much for such incredible conversations .\n", - "\n", - "I think I 'm very heartened to hear all the incredible work that Google Cloud is doing to make artificial intelligence accessible to the entire business world and all of every enterprise around the globe .\n", - "\n", - "And I 'm so excited that you 're able to join us .\n", - "\n", - "Thank you so much .\n", - "\n", - "Thank you so much for having me . \n", - "\n", - "Thank you .\n", - "\n", - "Thank you .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - "\u001b[0m\n", - "Write a concise two line summary of the following:\n", - "\n", - "\n", - "And so every boundary as the technology gets more sophisticated we think it moves from just assistance to assistance on things that human beings may not have been able to just linearly do to now things like expressiveness , which is a very different capability than people could actually do themselves .\n", - "\n", - "Yeah , I mean , all of this is very obviously incredibly exciting and we 're all watching it happen in real time .\n", - "\n", - "There 's an artist who actually described the image generation models as , he sort of image generation models as he was , he sort of said like , you kind of think about like a camera .\n", - "\n", - "Like it 's a new tool that allows you to create fundamentally new forms of art .\n", - "\n", - "That 's right .\n", - "\n", - "Yeah .\n", - "\n", - "And not just one medium of art , right ?\n", - "\n", - "Because if you look in the past , people said , you were a painter , you were a sculpture , you were a musician , and now these technologies allow you to blend all of it as a form of expressiveness .\n", - "\n", - "Yeah .\n", - "\n", - "You know , the last question I have for you is , you know , you obviously sit down with many of the sort of leading CEOs and business leaders of of the sort of largest organizations in the world . \n", - "\n", - "And I 'm sure one thing that is on many of their minds is sort of as AI technology develops and it continues to progress is potential disruption that might come from art of film intelligence .\n", - "\n", - "What sort of , how do you approach that conversation ?\n", - "\n", - "What sort of your advice to these business leaders who are looking at this powerful new technology and thinking about what that might mean for the businesses and the business landscape .\n", - "\n", - "When we talk to CEOs , I mean the biggest things we talk to them about number one , productivity in the long term , productivity has always been the primary driver of improving both company productivity , meaning their own companies , as well as societal benefit , things like affluence of a society , etc . \n", - "\n", - "And the means and equality of distribution of income to people across all spectrum society .\n", - "\n", - "Eventually , the most important metric , and you can look at any economic textbook is productivity .\n", - "\n", - "Software and technology has probably been the biggest boomer productivity over the last 30 , 40 years .\n", - "\n", - "This is the next evolution of that .\n", - "\n", - "And so we always say , if you approach it the right way , for example , labor shortages are going on right now .\n", - "\n", - "The biggest potential benefit is the application of some of these platforms like AI to do in that .\n", - "\n", - "The second , with any technological generation revolution , like artificial intelligence ,\n", - "\n", - "but if you went back in time and looked at the industrial revolution , etc . \n", - "\n", - "There are always during the period of transition , anxiety about the consequences of that technology . \n", - "\n", - "And it does n't mean the technology by itself is good or bad .\n", - "\n", - "It 's the application of the technology that 's good or bad . \n", - "\n", - "So it 's incumbent upon both the technology providers and the users of the technology to ensure that the negative consequences of it are managed properly .\n", - "\n", - "Right ?\n", - "\n", - "The obvious example is , for instance , if you look at a very simple thing , image recognition .\n", - "\n", - "Image recognition can help doctors find tumors way better than having the best radiographer .\n", - "\n", - "It 's a system in that context and it 's like helping people with a better quality microscope than they had before .\n", - "\n", - "Object recognition is helping people find , for example , people who are in the ocean much more accurately so the coastguard can rescue them .\n", - "\n", - "At the same time , being able to use a camera and say that 's Thomas Korean has , you know , a lot of potential negative consequences . \n", - "\n", - "And so as a provider of technology , we at Google have chosen not to do that third part . \n", - "\n", - "But we also tell companies , it 's important not just to say , this is what 's regularly allowed by the legal framework , because law in many countries is not yet keeping up with how fast AI technology is moving .\n", - "\n", - "But to take the responsibility as a company CEO to say , here 's what I believe comfortable with , and here 's what I wo n't be comfortable with .\n", - "\n", - "Yeah .\n", - "\n", - "Well , Thomas , thank you so much for such incredible conversations .\n", - "\n", - "I think I 'm very heartened to hear all the incredible work that Google Cloud is doing to make artificial intelligence accessible to the entire business world and all of every enterprise around the globe .\n", - "\n", - "And I 'm so excited that you 're able to join us .\n", - "\n", - "Thank you so much .\n", - "\n", - "Thank you so much for having me . \n", - "\n", - "Thank you .\n", - "\n", - "Thank you .\n", - "\n", - "\n", - "CONCISE SUMMARY:\n", - "\n", - " And so, when talking to executives (CEOs) regarding Artificial Intelligence development & applications within large corporations/businesses - our main focus should primarily revolve towards increasing overall efficiency /productivity. We must remember however; there will inevitably still likely exist certain ethical concerns surrounding various aspects related to A.I., therefore proper management strategies need to remain top priority throughout each stage of implementation process! In addition- ensuring appropriate legislation exists prior adopting cutting edge tech solutions into daily operations...all while maintaining full transparency along whole journey! Lastly -- never forget importance prioritizing PEOPLE first above everything else!! Always keep safety foremost mind!! :) Cheers everyone!!! Have great day ahead :-) Stay safe out there <3 Peace Love&Light :) xoxoxo Team Ai4All #AIforall @googlecloudteam\n", - "\n", - "\n", - "\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n", - "\n", - "\n", - "\u001b[1m> Entering new LLMChain chain...\u001b[0m\n", - "Prompt after formatting:\n", - "\u001b[32;1m\u001b[1;3mSummarize the text below in a subject line:\n", - "\n", - "\n", - "And so every boundary as the technology gets more sophisticated we think it moves from just assistance to assistance on things that human beings may not have been able to just linearly do to now things like expressiveness , which is a very different capability than people could actually do themselves .\n", - "\n", - "Yeah , I mean , all of this is very obviously incredibly exciting and we 're all watching it happen in real time .\n", - "\n", - "There 's an artist who actually described the image generation models as , he sort of image generation models as he was , he sort of said like , you kind of think about like a camera .\n", - "\n", - "Like it 's a new tool that allows you to create fundamentally new forms of art .\n", - "\n", - "That 's right .\n", - "\n", - "Yeah .\n", - "\n", - "And not just one medium of art , right ?\n", - "\n", - "Because if you look in the past , people said , you were a painter , you were a sculpture , you were a musician , and now these technologies allow you to blend all of it as a form of expressiveness .\n", - "\n", - "Yeah .\n", - "\n", - "You know , the last question I have for you is , you know , you obviously sit down with many of the sort of leading CEOs and business leaders of of the sort of largest organizations in the world . \n", - "\n", - "And I 'm sure one thing that is on many of their minds is sort of as AI technology develops and it continues to progress is potential disruption that might come from art of film intelligence .\n", - "\n", - "What sort of , how do you approach that conversation ?\n", - "\n", - "What sort of your advice to these business leaders who are looking at this powerful new technology and thinking about what that might mean for the businesses and the business landscape .\n", - "\n", - "When we talk to CEOs , I mean the biggest things we talk to them about number one , productivity in the long term , productivity has always been the primary driver of improving both company productivity , meaning their own companies , as well as societal benefit , things like affluence of a society , etc . \n", - "\n", - "And the means and equality of distribution of income to people across all spectrum society .\n", - "\n", - "Eventually , the most important metric , and you can look at any economic textbook is productivity .\n", - "\n", - "Software and technology has probably been the biggest boomer productivity over the last 30 , 40 years .\n", - "\n", - "This is the next evolution of that .\n", - "\n", - "And so we always say , if you approach it the right way , for example , labor shortages are going on right now .\n", - "\n", - "The biggest potential benefit is the application of some of these platforms like AI to do in that .\n", - "\n", - "The second , with any technological generation revolution , like artificial intelligence ,\n", - "\n", - "but if you went back in time and looked at the industrial revolution , etc . \n", - "\n", - "There are always during the period of transition , anxiety about the consequences of that technology . \n", - "\n", - "And it does n't mean the technology by itself is good or bad .\n", - "\n", - "It 's the application of the technology that 's good or bad . \n", - "\n", - "So it 's incumbent upon both the technology providers and the users of the technology to ensure that the negative consequences of it are managed properly .\n", - "\n", - "Right ?\n", - "\n", - "The obvious example is , for instance , if you look at a very simple thing , image recognition .\n", - "\n", - "Image recognition can help doctors find tumors way better than having the best radiographer .\n", - "\n", - "It 's a system in that context and it 's like helping people with a better quality microscope than they had before .\n", - "\n", - "Object recognition is helping people find , for example , people who are in the ocean much more accurately so the coastguard can rescue them .\n", - "\n", - "At the same time , being able to use a camera and say that 's Thomas Korean has , you know , a lot of potential negative consequences . \n", - "\n", - "And so as a provider of technology , we at Google have chosen not to do that third part . \n", - "\n", - "But we also tell companies , it 's important not just to say , this is what 's regularly allowed by the legal framework , because law in many countries is not yet keeping up with how fast AI technology is moving .\n", - "\n", - "But to take the responsibility as a company CEO to say , here 's what I believe comfortable with , and here 's what I wo n't be comfortable with .\n", - "\n", - "Yeah .\n", - "\n", - "Well , Thomas , thank you so much for such incredible conversations .\n", - "\n", - "I think I 'm very heartened to hear all the incredible work that Google Cloud is doing to make artificial intelligence accessible to the entire business world and all of every enterprise around the globe .\n", - "\n", - "And I 'm so excited that you 're able to join us .\n", - "\n", - "Thank you so much .\n", - "\n", - "Thank you so much for having me . \n", - "\n", - "Thank you .\n", - "\n", - "Thank you .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - "\u001b[0m\n", - "Summarize the text below in a subject line:\n", - "\n", - "\n", - "And so every boundary as the technology gets more sophisticated we think it moves from just assistance to assistance on things that human beings may not have been able to just linearly do to now things like expressiveness , which is a very different capability than people could actually do themselves .\n", - "\n", - "Yeah , I mean , all of this is very obviously incredibly exciting and we 're all watching it happen in real time .\n", - "\n", - "There 's an artist who actually described the image generation models as , he sort of image generation models as he was , he sort of said like , you kind of think about like a camera .\n", - "\n", - "Like it 's a new tool that allows you to create fundamentally new forms of art .\n", - "\n", - "That 's right .\n", - "\n", - "Yeah .\n", - "\n", - "And not just one medium of art , right ?\n", - "\n", - "Because if you look in the past , people said , you were a painter , you were a sculpture , you were a musician , and now these technologies allow you to blend all of it as a form of expressiveness .\n", - "\n", - "Yeah .\n", - "\n", - "You know , the last question I have for you is , you know , you obviously sit down with many of the sort of leading CEOs and business leaders of of the sort of largest organizations in the world . \n", - "\n", - "And I 'm sure one thing that is on many of their minds is sort of as AI technology develops and it continues to progress is potential disruption that might come from art of film intelligence .\n", - "\n", - "What sort of , how do you approach that conversation ?\n", - "\n", - "What sort of your advice to these business leaders who are looking at this powerful new technology and thinking about what that might mean for the businesses and the business landscape .\n", - "\n", - "When we talk to CEOs , I mean the biggest things we talk to them about number one , productivity in the long term , productivity has always been the primary driver of improving both company productivity , meaning their own companies , as well as societal benefit , things like affluence of a society , etc . \n", - "\n", - "And the means and equality of distribution of income to people across all spectrum society .\n", - "\n", - "Eventually , the most important metric , and you can look at any economic textbook is productivity .\n", - "\n", - "Software and technology has probably been the biggest boomer productivity over the last 30 , 40 years .\n", - "\n", - "This is the next evolution of that .\n", - "\n", - "And so we always say , if you approach it the right way , for example , labor shortages are going on right now .\n", - "\n", - "The biggest potential benefit is the application of some of these platforms like AI to do in that .\n", - "\n", - "The second , with any technological generation revolution , like artificial intelligence ,\n", - "\n", - "but if you went back in time and looked at the industrial revolution , etc . \n", - "\n", - "There are always during the period of transition , anxiety about the consequences of that technology . \n", - "\n", - "And it does n't mean the technology by itself is good or bad .\n", - "\n", - "It 's the application of the technology that 's good or bad . \n", - "\n", - "So it 's incumbent upon both the technology providers and the users of the technology to ensure that the negative consequences of it are managed properly .\n", - "\n", - "Right ?\n", - "\n", - "The obvious example is , for instance , if you look at a very simple thing , image recognition .\n", - "\n", - "Image recognition can help doctors find tumors way better than having the best radiographer .\n", - "\n", - "It 's a system in that context and it 's like helping people with a better quality microscope than they had before .\n", - "\n", - "Object recognition is helping people find , for example , people who are in the ocean much more accurately so the coastguard can rescue them .\n", - "\n", - "At the same time , being able to use a camera and say that 's Thomas Korean has , you know , a lot of potential negative consequences . \n", - "\n", - "And so as a provider of technology , we at Google have chosen not to do that third part . \n", - "\n", - "But we also tell companies , it 's important not just to say , this is what 's regularly allowed by the legal framework , because law in many countries is not yet keeping up with how fast AI technology is moving .\n", - "\n", - "But to take the responsibility as a company CEO to say , here 's what I believe comfortable with , and here 's what I wo n't be comfortable with .\n", - "\n", - "Yeah .\n", - "\n", - "Well , Thomas , thank you so much for such incredible conversations .\n", - "\n", - "I think I 'm very heartened to hear all the incredible work that Google Cloud is doing to make artificial intelligence accessible to the entire business world and all of every enterprise around the globe .\n", - "\n", - "And I 'm so excited that you 're able to join us .\n", - "\n", - "Thank you so much .\n", - "\n", - "Thank you so much for having me . \n", - "\n", - "Thank you .\n", - "\n", - "Thank you .\n", - "\n", - "\n", - "SUBJECT LINE:\n", - "\n", - " And So Every Boundary As The Technology Gets More Sophisticated We Think It Moves From Just Assistance To Assisting On Things That Human Beings May Not Have Been Able To Linear Do, Now Things Like Expressiveness Which Is Very Different Capability Than People Could Actually Do. \n", - "\n", - "\n", - "\u001b[1m> Finished chain.\u001b[0m\n" - ] - } - ], - "source": [ - "summaries = []\n", - "subjects = []\n", - "for doc in docs:\n", - " summary = summary_chain.run({\"text\": doc.page_content})\n", - " summaries.append(summary)\n", - " subject = subject_chain.run({\"text\": doc.page_content})\n", - " subjects.append(subject)\n", - "\n", - "\n", - "# for c in chunks:\n", - "# summary = llm.generate(c, max_length=100, num_return_sequences=1)[0]\n", - "# summaries.append(summary)\n", - "\n", - "with open(\"mpt-7b-summaries.txt\", \"a\") as f:\n", - " for summary, subject in zip(summaries, subjects):\n", - " f.write(\"SUBJECT: \" + subject + \"\\n\")\n", - " f.write(\"SUMMARY: \" + summary + \"\\n\\n\")\n" - ] - } - ], - "metadata": { - "kernelspec": { - "display_name": "myenv", - "language": "python", - "name": "python3" - }, - "language_info": { - "codemirror_mode": { - "name": "ipython", - "version": 3 - }, - "file_extension": ".py", - "mimetype": "text/x-python", - "name": "python", - "nbconvert_exporter": "python", - "pygments_lexer": "ipython3", - "version": "3.11.2" - }, - "orig_nbformat": 4 - }, - "nbformat": 4, - "nbformat_minor": 2 -} diff --git a/server/reflector-local/0-reflector-local.py b/server/reflector-local/0-reflector-local.py deleted file mode 100644 index 4d5cebda..00000000 --- a/server/reflector-local/0-reflector-local.py +++ /dev/null @@ -1,34 +0,0 @@ -import os -import subprocess -import sys - -from loguru import logger - -# Get the input file name from the command line argument -input_file = sys.argv[1] -# example use: python 0-reflector-local.py input.m4a agenda.txt - -# Get the agenda file name from the command line argument if provided -if len(sys.argv) > 2: - agenda_file = sys.argv[2] -else: - agenda_file = "agenda.txt" -# example use: python 0-reflector-local.py input.m4a my_agenda.txt - -# Check if the agenda file exists -if not os.path.exists(agenda_file): - logger.error("agenda_file is missing") - -# Check if the input file is .m4a, if so convert to .mp4 -if input_file.endswith(".m4a"): - subprocess.run(["ffmpeg", "-i", input_file, f"{input_file}.mp4"]) - input_file = f"{input_file}.mp4" - -# Run the first script to generate the transcript -subprocess.run(["python3", "1-transcript-generator.py", input_file, f"{input_file}_transcript.txt"]) - -# Run the second script to compare the transcript to the agenda -subprocess.run(["python3", "2-agenda-transcript-diff.py", agenda_file, f"{input_file}_transcript.txt"]) - -# Run the third script to summarize the transcript -subprocess.run(["python3", "3-transcript-summarizer.py", f"{input_file}_transcript.txt", f"{input_file}_summary.txt"]) diff --git a/server/reflector-local/1-transcript-generator.py b/server/reflector-local/1-transcript-generator.py deleted file mode 100755 index e19c41da..00000000 --- a/server/reflector-local/1-transcript-generator.py +++ /dev/null @@ -1,62 +0,0 @@ -import argparse -import os - -import moviepy.editor -import whisper -from loguru import logger - -WHISPER_MODEL_SIZE = "base" - - -def init_argparse() -> argparse.ArgumentParser: - parser = argparse.ArgumentParser( - usage="%(prog)s ", - description="Creates a transcript of a video or audio file using the OpenAI Whisper model" - ) - parser.add_argument("location", help="Location of the media file") - parser.add_argument("output", help="Output file path") - return parser - - -def main(): - import sys - sys.setrecursionlimit(10000) - - parser = init_argparse() - args = parser.parse_args() - - media_file = args.location - logger.info(f"Processing file: {media_file}") - - # Check if the media file is a valid audio or video file - if os.path.isfile(media_file) and not media_file.endswith( - ('.mp3', '.wav', '.ogg', '.flac', '.mp4', '.avi', '.flv')): - logger.error(f"Invalid file format: {media_file}") - return - - # If the media file we just retrieved is an audio file then skip extraction step - audio_filename = media_file - logger.info(f"Found audio-only file, skipping audio extraction") - - audio = moviepy.editor.AudioFileClip(audio_filename) - - logger.info("Selected extracted audio") - - # Transcribe the audio file using the OpenAI Whisper model - logger.info("Loading Whisper speech-to-text model") - whisper_model = whisper.load_model(WHISPER_MODEL_SIZE) - - logger.info(f"Transcribing file: {media_file}") - whisper_result = whisper_model.transcribe(media_file) - - logger.info("Finished transcribing file") - - # Save the transcript to the specified file. - logger.info(f"Saving transcript to: {args.output}") - transcript_file = open(args.output, "w") - transcript_file.write(whisper_result["text"]) - transcript_file.close() - - -if __name__ == "__main__": - main() diff --git a/server/reflector-local/2-agenda-transcript-diff.py b/server/reflector-local/2-agenda-transcript-diff.py deleted file mode 100644 index 30886dc0..00000000 --- a/server/reflector-local/2-agenda-transcript-diff.py +++ /dev/null @@ -1,68 +0,0 @@ -import argparse - -import spacy -from loguru import logger - - -# Define the paths for agenda and transcription files -def init_argparse() -> argparse.ArgumentParser: - parser = argparse.ArgumentParser( - usage="%(prog)s ", - description="Compares the transcript of a video or audio file to an agenda using the SpaCy model" - ) - parser.add_argument("agenda", help="Location of the agenda file") - parser.add_argument("transcription", help="Location of the transcription file") - return parser - - -args = init_argparse().parse_args() -agenda_path = args.agenda -transcription_path = args.transcription - -# Load the spaCy model and add the sentencizer -spaCy_model = "en_core_web_md" -nlp = spacy.load(spaCy_model) -nlp.add_pipe('sentencizer') -logger.info("Loaded spaCy model " + spaCy_model) - -# Load the agenda -with open(agenda_path, "r") as f: - agenda = [line.strip() for line in f.readlines() if line.strip()] -logger.info("Loaded agenda items") - -# Load the transcription -with open(transcription_path, "r") as f: - transcription = f.read() -logger.info("Loaded transcription") - -# Tokenize the transcription using spaCy -doc_transcription = nlp(transcription) -logger.info("Tokenized transcription") - -# Find the items covered in the transcription -covered_items = {} -for item in agenda: - item_doc = nlp(item) - for sent in doc_transcription.sents: - if not sent or not all(token.has_vector for token in sent): - # Skip an empty span or one without any word vectors - continue - similarity = sent.similarity(item_doc) - similarity_threshold = 0.7 - if similarity > similarity_threshold: # Set the threshold to determine what is considered a match - covered_items[item] = True - break - -# Count the number of items covered and calculatre the percentage -num_covered_items = sum(covered_items.values()) -percentage_covered = num_covered_items / len(agenda) * 100 - -# Print the results -print("💬 Agenda items covered in the transcription:") -for item in agenda: - if item in covered_items and covered_items[item]: - print("✅ ", item) - else: - print("❌ ", item) -print("📊 Coverage: {:.2f}%".format(percentage_covered)) -logger.info("Finished comparing agenda to transcription with similarity threshold of " + str(similarity_threshold)) diff --git a/server/reflector-local/3-transcript-summarizer.py b/server/reflector-local/3-transcript-summarizer.py deleted file mode 100644 index 58a75451..00000000 --- a/server/reflector-local/3-transcript-summarizer.py +++ /dev/null @@ -1,94 +0,0 @@ -import argparse - -import nltk - -nltk.download('stopwords') -from nltk.corpus import stopwords -from nltk.tokenize import word_tokenize, sent_tokenize -from heapq import nlargest -from loguru import logger - - -# Function to initialize the argument parser -def init_argparse(): - parser = argparse.ArgumentParser( - usage="%(prog)s ", - description="Summarization" - ) - parser.add_argument("transcript", type=str, default="transcript.txt", help="Path to the input transcript file") - parser.add_argument("summary", type=str, default="summary.txt", help="Path to the output summary file") - parser.add_argument("--num_sentences", type=int, default=5, help="Number of sentences to include in the summary") - return parser - - -# Function to read the input transcript file -def read_transcript(file_path): - with open(file_path, "r") as file: - transcript = file.read() - return transcript - - -# Function to preprocess the text by removing stop words and special characters -def preprocess_text(text): - stop_words = set(stopwords.words('english')) - words = word_tokenize(text) - words = [w.lower() for w in words if w.isalpha() and w.lower() not in stop_words] - return words - - -# Function to score each sentence based on the frequency of its words and return the top sentences -def summarize_text(text, num_sentences): - # Tokenize the text into sentences - sentences = sent_tokenize(text) - - # Preprocess the text by removing stop words and special characters - words = preprocess_text(text) - - # Calculate the frequency of each word in the text - word_freq = nltk.FreqDist(words) - - # Calculate the score for each sentence based on the frequency of its words - sentence_scores = {} - for i, sentence in enumerate(sentences): - sentence_words = preprocess_text(sentence) - for word in sentence_words: - if word in word_freq: - if i not in sentence_scores: - sentence_scores[i] = word_freq[word] - else: - sentence_scores[i] += word_freq[word] - - # Select the top sentences based on their scores - top_sentences = nlargest(num_sentences, sentence_scores, key=sentence_scores.get) - - # Sort the top sentences in the order they appeared in the original text - summary_sent = sorted(top_sentences) - summary = [sentences[i] for i in summary_sent] - - return " ".join(summary) - - -def main(): - # Initialize the argument parser and parse the arguments - parser = init_argparse() - args = parser.parse_args() - - # Read the input transcript file - logger.info(f"Reading transcript from: {args.transcript}") - transcript = read_transcript(args.transcript) - - # Summarize the transcript using the nltk library - logger.info("Summarizing transcript") - summary = summarize_text(transcript, args.num_sentences) - - # Write the summary to the output file - logger.info(f"Writing summary to: {args.summary}") - with open(args.summary, "w") as f: - f.write("Summary of: " + args.transcript + "\n\n") - f.write(summary) - - logger.info("Summarization completed") - - -if __name__ == "__main__": - main() diff --git a/server/reflector-local/30min-CyberHR/30min-CyberHR-agenda.txt b/server/reflector-local/30min-CyberHR/30min-CyberHR-agenda.txt deleted file mode 100644 index 34717a1c..00000000 --- a/server/reflector-local/30min-CyberHR/30min-CyberHR-agenda.txt +++ /dev/null @@ -1,4 +0,0 @@ -# Deloitte HR @ NYS Cybersecurity Conference -- ways to retain and grow your workforce -- how to enable cybersecurity professionals to do their best work -- low-budget activities that can be implemented starting tomorrow \ No newline at end of file diff --git a/server/reflector-local/30min-CyberHR/30min-CyberHR-transcript.txt b/server/reflector-local/30min-CyberHR/30min-CyberHR-transcript.txt deleted file mode 100644 index 6501ff02..00000000 --- a/server/reflector-local/30min-CyberHR/30min-CyberHR-transcript.txt +++ /dev/null @@ -1 +0,0 @@ - I don't know if everything's going well so far. Hold on. OK, great. One in. Join us. This is for you to keep us low-keyed. Very casual. Absolutely. All right. Well, thank you all for being here. We're very honored that you came to join us for this conversation. We are going to spend a little bit of time, I know it was written up, right? But thinking about how you actually grow, retain your employees, we know that there's a moron talent, especially in the cyber space right now, right? Everybody's trying to get everybody in the door. Pipelines are a little bit dry, a little bit hard to find. It's a tricky scenario. What we want to do, if you allow us for this session, is to almost park that call to the site. We're just going to suspend this belief we have for a moment. We're just going to put that aside. And instead what we're going to focus on are the employees that you actually have on board. All right. We know that the hiring piece is complex, it requires dollars, an HR and a whole bunch of stuff. That's that. We're going to focus on the rest, which is your workforce today, how do you grow them, how do you retain them? And in doing so, you actually find that they'll become more effective and efficient and committed to your organization, so that then they'll go ahead and become random ambassadors for you, which helps that talent recruit group pipeline in the end anyway. Okay. Great. You're going to jump in with a couple of introductions and then look what we're going to be. So my name is Susie Candace. I am a senior manager in the Lloyds Cuban Cabo practice. I focus on organizational transformation. What does that mean? That's where we put all of our organizational design work, so how offices are aligned up the org chart, and how the information and decisions flow. In addition, it's where we hold our culture and our communications work. I've spent my career at Deloitte and Beyond, previously, focused on culture and communications and employee and workforce programs and how you encourage that engagement and your job experience is the best taking day. I've done that in a couple of client spaces, including the federal state, state, state, local, and space, including with some cyber agencies as well. But that'll be excellent. Okay. Thanks, Susie. Good afternoon, everyone. I'm Mara Patashnik. I'm also a senior manager at Deloitte, and I lead workforce experience for the government and public services practice at our firm. I, along the forefront, have worked for our success experience issues for commercial cities, state and federal clients, focusing on training, attracting talent, recruiting, retention, and really today, what we want to do is share some best practices, tools, what we've seen from other clients, and help you give you some ideas about how to become an employer and choice, and how to retain and attract high-performing cyber talent. So we'll dive right in. Before we do, I just want to give one copy out of you. Because it's day two of the conference, now I've been thinking outside for a while. I would say Mara and I both want to copy up the neither of us are going to be cyber experts, especially in the room of this caliber. We do this with cyber clients, but we also do this with other clients. Our invitation to you is to take what we're going to share here and think about how that applies to your day-to-day in your organization. So we kind of did this a little bit to stretch that thinking outside of just this formal cyber realm, and bring you some ideas of what we're seeing from outside. That works. Great. So, absolutely, we define workforce experience holistically as the sum of the workers lived experience at work and how they feel about their organization. And so, really, this is shaped by eight key dimensions that impact worker experience overall. So we'll start by walking through each of these. So really, this framework is backed by research, as well as testing solutions with both our commercial, as well as our government clients, and we start with the people I work with. And this is really support and recognition from managers and who interact with day-to-day. That's commonly cited as one of the top five factors for talent retention, for the technology that I use. So that's very much about frustration-free technology. So are you having to do work-or-arms? Are you having to call IT or spend a lot of your day trying to figure out how to actually use the technology to do your job? And are there ways to use technology to improve collaboration, coordination, and communication in a way that you're able to do your work more efficiently? The maximum focus on a places I do work, which is really focusing on the flexibility of the physical workspace and how that improves employee productivity, increases jobs as faction, and lowers overall workplace stress. Another one that we focus on is a sense of belonging and worthiness that an organization creates. So that's very much about increasing your job performance. I think we've cited that it can lead to a 56% increase in job performance and a 50% reduction in turnover risk. And the work I do, I think this is particularly important to cyber employees where a lot of government employees are driven by the mission of the work that they're doing. And so connecting the work to meaningful experiences and re-enerating that fulfilling and contributes to something bigger themselves is a great way to motivate and keep employees. And that's something that I think government organizations have an advantage over commercial organizations. Another one is well-being. So this is really focusing on personal life. So, organizations can create greater flexibility to attend to well-being means. That increases worker satisfaction. And then another one is really focusing on the mission. So, identifying an organization's purpose and connecting it to their own personal values as well as the mission of the work organization. And finally, education. So this is really about high performing organizations and the fact that 30 times more likely workers are going to be able to achieve their long-term goals and stay within organization if they have the proper training and support and experiences to help them get there. So really when thinking about workforce experience, organizations know often are sure where to start or what elements impact their employees. So this framework is really what we use as a starting point to help orient our clients and to also help them prioritize certain employee activities and relationships. So these are the key contributors that we've seen in our experience to employees really engage and motivated, which has directly contributed to retaining and attracting talent. So with that, we are going to move on to the next slide. So just start off cybersecurity employees we know are experiencing disruption unlike ever before. And so just to a frame today's discussion, we first want to start by walking through how one agency has approached the issue and use workforce experience to address it. So a large federal agency of the Angus Cybersecurity Priorities had long-suing workforce issues. You know, lots of attrition, tough time attracting the right talent for the roles that they needed. And so they really double down and focusing on career growth and development, culture, recognition, morale, and investing in trust in leadership. So as a result of reimagining the workforce experience, they were able to retain critical employees or a major transition. And a few things with these thoughts were as a result of investment in workforce experience programs include the increase in frequency of career-focused conversations, increase in informal recognition across the entire organization, and employees really felt that they had the tools to help the quantum to do their job center and also to prevent burnout amongst their teens, which was an issue that this organization was facing. So next, I'll pass it over to Susie and to speak about the cyber workforce today, probably a little bit about what you all are seeing. Thanks, Mara. So there are a couple of stats up on this slide. I don't think they're going to surprise anybody, but I just want to call them out, because I think they're an important framing for what we are seeing in this hybrid industry. This is one of the first groups you are all experts in. So in 2020, you last hear Des Moines and the National Association of State CIOs did a cybersecurity study. Is anyone familiar with the study? No, no. No, I think that. Okay, a little later. Afternoon reading board. Yeah. And so, it was a survey of all 50 state CSOs and three territories, CSOs, so very comprehensive across the United States. And they were asking questions about the general cybersecurity environment, and the conditions, and things came out around the workforce. So when those CSOs were asked about how they can support emerging threats, and what their biggest challenge is, number one, they said legacy infrastructure and solutions. Number two, they said workforce. The number two thing holding these state CSOs back from responding to the threats that are coming is the workforce. That's a pretty big issue. In addition, 62% out two-thirds reported that their staff have a gap in competencies. Don't have the skills to do the job. Another pretty big issue. You can step back and look at it. When we talk about the belonging piece that Mara just talked about and shared, 23% with the respondents didn't know if their organization has established DEI leadership positions. So is there someone in the organization really focused on that inclusion piece and bringing all of your workforce along? And only 25% reported offering remote work options, which I know is tricky in the cyberspace. When you think about the SOC and everything else, but there are ways to do it with at least parts of the workforce. And we know that that's super important to the millennials and slunder representing groups. Remote work has certainly been a topic since the pandemic, right? I see people nodding. It's just on everybody's tongue. It's front of mind. So how do you make that work? In addition to all of this, I think what's really important to note, right, is that cyber threats have not reduced in any way. So there's one stat that says that as a result of COVID-19, it's been linked to a 238% increase in cybercrack. So the numbers are going up. The need continues and the resourcing of the workforce maybe is a little bit lagging behind. In addition to this in the cyberspace, we're also seeing some pretty big macro trends in workforce space in general that are contributing to some of the challenges going on today. And these also probably won't surprise you. Number one, exhaustion. So this is where that burnout topic that someone you people are talking about is showing up. And in cyber agencies, we have frontline workers, right? There's a 24 set of an operation. And so you're going to have this burnout challenge, unfortunately. This is also contributed by working from home in some ways and that blurring of that work life, home life, when the end of the day happens, if it happens, and who stays on their laptop all night working. In addition, between 2021 and 2022, the number of meetings we all experience increased 288%. So you're getting your Zoom fatigue, your Teams fatigue, right? Does that meeting fatigue along the way? Resignation. Who's heard of the great resignation coming out of the pandemic area? Yeah. So that's where the stomach fits in. We know that a lot of people left the workforce voluntarily and in fact in 2021, more people than ever before left the workforce according to the Bureau of Labor statistics. Some people have come back and some people haven't. So now are options for who's out there working and interested in being in the workforce. And the talent and diversity that they bring to your organizations could be a little more limited. And then reshuffling. 34% of US workers shifted their cities since the start of COVID. 34, a third of US workers have moved cities. And so this is where that remote work conversation becomes so important about can you get the right people? And so people again that you have on board, how are you engaging them, keeping them, you know, working at their most effective to support themselves in your organization? Why does all this matter? Because we know that losing employees is costly. Giving them a positive workforce experience means that they stay with your organization. They are committed. They are loyal. And they continue to grow and build within your organization. One study shows that losing an employee and having to recruit a new one costs three to four times their salary. The opportunity costs, the knowledge drain, the process of just onboarding, putting the job up on the boards, bringing them on three to four times their salary. So if you can get it right on the front end and really have that workforce experience be positive and keep your people, it's going to pay off dividends in the end. I'm going to pass it back tomorrow. Who's going to tie this back to that model for us? Great. Building on what Susie said, workforce experience has never been more important given this crisis and the urgency around keeping talent and finding the right talent. Since the workforce is an organization's most valuable asset, investing in workforce experience activities we've found has lead to more productive work, more efficient work, more innovative approaches to the work, and more engaged teams which ultimately results in better mission outcomes for your organization. And we found a direct correlation with the ability to retain employees, improve efficiency overall in your organization, and manage organizational changes and transitions more effectively. From our work with other agencies as well as commercial organizations, we found that investing in workforce experience is the most sustainable way to create a competitive advantage that improves building business and talent outcomes. And so, if investing in workforce experience is a single best way that organizations can minimize workforce disruption in the short term and get ahead of future work, how do we do that? And so, in this next section, we're going to focus on how cyber employers can win at workforce experience. So, based on a 2022 state CIO, NACIO study that included 51 participants, it was identified that the following areas were some of the top priorities around how to improve workforce experience. We found that 18% of respondents decided increasing remote work options as particularly important. 8% focused on flexible work schedules and how work affects their life. 35% focused on how I grow as a human, so focusing on sweet-skilling opportunities for rotation programs and the right training to make sure that the workforce is able to meet the needs of the modern IT demands. And so, with this, Susie is going to walk us through a deeper dive of the different options on how to address these results and what this actually might look like in practice. Great. Thank you, Mara. So, again, just to take it from the top, right? We know that the talent pipeline is tough. We know that the recruiting process is challenging. Diving a little deeper into some of these things, we're going to walk through a couple of activities, programs, initiatives, ideas, just to hopefully give you something to walk away with today that you can take back to your organization and implement. And the idea behind these is you can implement them tomorrow. You don't need to get into an IT backlog of some system bill. You don't need to go to HR and say I need millions of dollars to hire people and fill the gaps. These are things that you can do with your current workforce in place today to improve that workforce experience. So, starting with how I grow with a human, and again, 35% of the CIO said that this was one of the most important things. A couple of ideas here for things that can be done and things that we've seen be successful elsewhere. Number one, integrating training into the day to day. This is not about sending an employee off to go take a training, get a certification, step out of the workplace and come back. We know that most adult learners prefer experiential learning. And so the way to do that is apprenticeship, mentorship, shadowing, rotational programs, lunch and learns, informal training opportunities, action, after action reports where you can also around and talk about what did we do well, what didn't we do, what are we going to do differently. And some of this also is just documenting what it is you do, so setting up some of those SOPs so that you can pass those over to somebody and say, well this is the best way to do it. Identifying opportunities for employees to use their strengths daily. When we talk about strengths, we don't just mean one of my good at it. What we mean by strengths is what gives me individually energy, what energizes me, what excites me. And so I may be very good at data analytics, but that doesn't mean that that's my passion. I may actually prefer competing teams or communications or something like that. And so checking in with your teams about what is it that each person enjoys most? And trying to align them to that work, studies have shown that if you align people to their strengths, again the things that energize them, they will actually outperform and do way better than if you give them a challenge task or activity that then they have to sort of overcome and be able to accomplish. We are good when we're playing to our strengths. So it doesn't mean that an employee can only do their strength all day and be happy, but trying to find ways for them to incorporate their strengths into the day and day as much as possible. And then another thing that tends to be useful is stepping back and looking at career pathways. So how are you going to grow people from level to level, from ability to ability? Is anybody here familiar with the nice framework? Come here, Nick. Yeah, couple of hands. So this is a great resource if you haven't seen and I highly recommend going onto the website. They have work titles, they have job descriptions, they have competencies, they have skills, knowledge, experiences, and you can stack roles so you can say, okay, if I'm an IT leadership management position today and I want to become a data analyst and they will help you kind of they map that path a little bit for you with the overlap between those two roles, for example, and then some of the knowledge skills and abilities to go get for one to the other. So a very, very useful tool used by a lot of cyber agencies and organizations highly, highly, highly recommend using it or at least checking it out and see if there's anything useful for you there. But really looking at where are my people going and how do I grow them? Because we know also that one of the most important things for employees is them feeling like they have a growth path within an organization that somebody is invested in their development. When we move on to places I do work, I know this becomes a sticky conversation in a cyber environment. I know the sock, you know, is what it is, we know that piece. But there are conversations still to be having around work location. What we have found is most successful is when organizations really focus in on work location, in alignment to your desires, your goals, your strategy. There are some things that working remote are great for. Heads down time, idea generation, focus, and productivity. There are some things that working in person are great for. Teamwork, apprenticeship, and some of those sort of incidental connection points, right? Running into somebody, you know, by the water cooler and finding out that we're doing similar things and we can trade notes. Actually, Mara and I just did that and I said, oh, send me that if you could because I want to see that. So designing intentionally for what work can be done in person, what work can be done remotely, maybe it's part and part. Maybe there's a single initiative or project where you can send people to go home and work remotely and then come back. And they're still feeling like they're getting a taste of that opportunity if they're normally needing to be on site. So really being smart and intentional and purposeful about how you design that. Looking at the digital communications and information sharing, so a lot of us have moved on to teens and Slack and Zoom and these digital tools. Are you really using them for their maximum capability to support your organization? Is there somebody in your organization who really is tasked with understanding all of the features that you have licensed to and bringing that goodness to your organization to help your people work even better? And then the third one there is forcing connectivity and networking, especially in a high rate or remote environment. We know that happiness at work and elsewhere is tied to social connection. So how are you forcing some of that social connection? Maybe it is in those brown bag trainings or after action reports. Maybe it is taking a moment before diving into a meeting and saying, hey, how is everybody's weekend or what are you doing this weekend or tell me something funny or if you were going to be an office supply, what would you be, whatever it may be, right? But finding a way to sort of force a little door opening and having people build that connection. And then the well-being piece. And so I shared with you about the exhaustion and the burnout rates and all that earlier. A lot of organizations, especially since during instance the pandemic, have stood up well-being activities. They're not necessarily aligned to a strategy. What is your mission with those well-being activities? How does that align to your broader organizational strategy? What are you trying to achieve? Then what are the initiatives that you lay in with that? So it's not just sort of a scatter plot of activities, but they're sort of intentional and organized, aligning back to those goals that you have. And then matter of the impact of that and adjust as needed. And if you're engaging your workforce correctly, then you will have lots of data points about what's working and what's not in order to make those adjustments. Start your wellness a little bit, but again, thinking creatively, not just about remote work, but how shift work is scheduled. So a lot of the conversation around workforce and economic development is around people who may be to meet their child care duties at home while still trying to get into the workforce. Or they're taking classes at night while trying to get into the workforce. What is it that you can do to maybe look at four hour shifts or six hour shifts different than your typical shakes to bring in more people and diversity. And also alleviate some of the burnout and pressure that's happening for the workforce in place today. There are a bunch of apps that focus on this. A lot of them are focusing in the healthcare space of the transportation frontline worker space, but they could easily be leveraged over into the cyber space as well. And then the third one here conducting a culture assessment. A lot of organizations assume a culture sort of just happens and it doesn't. It's important to be intentional about culture. And what do we mean when we talk about culture, right? Because we all sort of live in it all the time. But what is that? When we talk about culture, what we talk about are the intangibles, your beliefs and your values as an organization. Are we more individually individualistic, the focused or are we more collaborative? Are we more risk averse or are we more innovative and risk takers? Are we more internally focused or do we have external customers that we've really organized around? And so when you do a cultural assessment, you look at where you are today versus where you want your organization to be and what are the steps to get from A to Z. And how do you do that in a way that then filters and flows through your entire organization so that by the time you get to performance reviews or your next strategic planning cycle or standing up on a new program or redesigning your organization, there isn't friction with what it is you really good. And what it is you really want from your employees and what you're trying to achieve as an organization and how you're getting there. So cultural assessment, well, it seems like a sort of unnecessary. It's actually very effective and super supportive of your broader goals. Okay, I've talked a lot. Lots of ideas. Hopefully a good menu for you to pick at least something from it if it's useful. I'm going to pass it back tomorrow to kind of bring it all together for us. Thank you, Cici. So how can employers win at work for our experience? So practically speaking, the key to implementing a successful workforce experience really starts with understanding what your workforce wants. That can be through a survey or focus groups doing user research, but really putting their voice at the center of these initiatives. And then working to prioritize which ones within your organization can help you achieve your overall business objectives. So we're going to walk through our perspective on what it means to really bring workforce experience to an organization. So starting with employers need to think about beyond employee engagement. And so what we mean by this is it's not just one data point. Are my employees engaged? How have I retained my staff? Or did I recruit an attractor? You know this number of individuals? It's beyond that an accommodation of those eight characteristics that really helps to retain employees and attract the right talent and enhance the overall workforce experience. So an example of that is looking at physical workspace. So do they are they able to do their work remotely? In some cases, as Susie mentioned, do they have everything they need in the office in order to be productive, to get things done in terms of head-down work versus team collaboration? Culture, do they, you know, are they recognized for their work? It's great as a hundred dollar gift card is for a great job and something. In a lot of cases what we found is it's equally just as advantageous to have a manager reach out and say, thank you so much. You did a great job and acknowledged that. And that's often what can motivate and keep employees engaged. And another area is technology. I think we mentioned, you know, it's a great way to help collaborate, work remotely, but it can also be an impediment to when technology isn't easily accessible within an organization. There are challenges, IT is available to help with certain things and can often be like a source of frustration for employees. And so really the way we look at it is not just one area at an employee engagement, but how do you approach across those eight dimensions, multiple areas of prioritizing different initiatives? The second one that we really focus on and this goes back to what Susie said about the assessment and also what I mentioned around the voice of the workforce. And so incorporating that throughout the design is incredibly important. One of the first things that we often do is a poll survey of a particular group to understand what they really hear about. What are their preferences? If they had, it's not just, I want more money, but rather if you had a choice between a little bit more flexibility in your work schedule versus, you know, increased recognition, where do they really stand in terms of what should actually get prioritized? And when you drill down into the heart of it, often is it what you think? Is it the obvious answers around like, I want flexibility? It's like, but what does that really need and how can that work for your organization in a way that maybe hybrid works for some workforces, but not necessarily for a cyber workforce? It's really, you know, is there a way that they can get different experiences for rotation program? As Susie mentioned. And so worker input, I would say, is the number one thing to invest in and even, you know, communicating that you're doing this and engaging the workforce early on and incorporating them into this because this is important to the organization and to you as leaders is an important way, I think also to retain talent. The third one is a continuous listening approach. And this one really focuses on not just pulsing a workforce once a year through an annual HR survey of, how do you really feel like, you know, what leadership considerations should we implement or, you know, how can we enhance the performance management process? It's really on an ongoing basis and even in an informal way, you as leaders and managers of teams asking your teams, you know, what they care about, what they're frustrated about, what their preferences are, and taking small steps where you can to try to incorporate that into how they work. And, you know, sometimes it's challenging to do organization-wide changes that take a lot of funding, investment, and capabilities. So oftentimes a way to do that is through, you know, more informal touch points on individual teams that you're leading. We found that that's a cost-effective, quick way to be able to get a lot of employee engagement and retain and keep your talent happy. The final one is workplace technology, and I know we've talked about this a little bit. And I think really with this, it's, how do you boost productivity? Workers want to do a great job. They're, they're the performed well. They, you know, are often very mission driven. They want to grow their careers and they, you know, want to try to be as efficient effective as they can in their job. But oftentimes, you know, what the organization decides in terms of technology versus like what the worker actually needs. There's a bit of a disconnect. And so really looking at, you know, how this can be a major accelerator based on how you work. Do you want an ants collaboration? Do you want to try to create more opportunities for high-grader of work? And, you know, through technology, that can really enhance and accelerate a lot of those activities. And so with all of these four areas combined, you know, this is what we've seen across, you know, Fortune 500 companies, federal agencies, the city, state agencies as well as kind of the key characteristics and commonalities amongst the most effective employee engagement and workforce experience programs. And just kind of round things out with this quote is, you know, if workforce experience I recognize, you know, can seem kind of fluffy and like, oh, that's a nice to have. But our premise is that your workforce is the most important asset in your organization. And if they are the most important asset, you really want to invest in them to, you know, day to day be the driver of change in, you know, be the builder productivity. We've just found that, you know, by investing in this and putting the workforce as, you know, the center part of what you invest in as an organization and leaders, it's not only about retention, talent, you know, the cyber workforce crisis, but people want to do work well and they're able to get more done and achieve more without you, you know, directly supervising and micromanaging or looking at everything because, you know, you know, you know, you're not going to be able to do anything. And you know, you know, you're not going to be able to do anything because you're not going to be able to do anything because you're not going to do anything because you're not going to do anything. So with that, that is our presentation for today. I hope there was a little bit of, you know, the landscape of the cyber workforce with some practical tips that you can take away for how to just think about, you know, improving the overall workforce experience and investing in your employees. So with this, you know, we know that all of you are in the trenches every day, you're facing this, you're living this, and we are just interested to hear from all of you, you know, just to start, like, what's one thing that has worked well in your organization in terms of enhancing or investing in the workforce experience? I'd love to get someone to start. Don't be shy. Yeah, sometimes it's important to have something or what hasn't worked well, you know, that's something that's easier to start with, you know. But we'd love to just hear from all of you, because I think, you know, how we've aggregated a lot of these best practices and what we've come with is by, you know, hearing other experiences from other organizations. And so one of the best and most effective things that I feel like I take away from conferences often is hearing from my peers and what they're facing and, you know, let it, you know, similarly I could bring back to what I'm leading. What about you? Oh, I see a hand back there. \ No newline at end of file diff --git a/server/reflector-local/30min-CyberHR/30min-CyberHR.m4a.mp4_summary.txt b/server/reflector-local/30min-CyberHR/30min-CyberHR.m4a.mp4_summary.txt deleted file mode 100644 index 0c0fbd16..00000000 --- a/server/reflector-local/30min-CyberHR/30min-CyberHR.m4a.mp4_summary.txt +++ /dev/null @@ -1,3 +0,0 @@ -Summary of: 30min-CyberHR/30min-CyberHR.m4a.mp4_transcript.txt - -Since the workforce is an organization's most valuable asset, investing in workforce experience activities, we've found has lead to more productive work, more efficient work, more innovative approaches to the work, and more engaged teams which ultimately results in better mission outcomes for your organization. And this one really focuses on not just pulsing a workforce once a year through an annual HR survey of, how do you really feel like, you know, what leadership considerations should we implement or, you know, how can we enhance the performance management process. We've just found that, you know, by investing in this and putting the workforce as, you know, the center part of what you invest in as an organization and leaders, it's not only about retention, talent, you know, the cyber workforce crisis, but people want to do work well and they're able to get more done and achieve more without you, you know, directly supervising and micromanaging or looking at everything because, you know, you know, you know, you're not going to be able to do anything. I hope there was a little bit of, you know, the landscape of the cyber workforce with some practical tips that you can take away for how to just think about, you know, improving the overall workforce experience and investing in your employees. So with this, you know, we know that all of you are in the trenches every day, you're facing this, you're living this, and we are just interested to hear from all of you, you know, just to start, like, what's one thing that has worked well in your organization in terms of enhancing or investing in the workforce experience? \ No newline at end of file diff --git a/server/reflector-local/30min-CyberHR/30min-CyberHR.m4a.mp4_transcript.txt b/server/reflector-local/30min-CyberHR/30min-CyberHR.m4a.mp4_transcript.txt deleted file mode 100644 index 809308ed..00000000 --- a/server/reflector-local/30min-CyberHR/30min-CyberHR.m4a.mp4_transcript.txt +++ /dev/null @@ -1 +0,0 @@ - I don't know if everything's going well so far. Hold on. OK, great. One in. Join us. This is for you to keep us low-keyed. Very casual. Absolutely. All right. Well, thank you all for being here. We're very honored that you came to join us for this conversation. We are going to spend a little bit of time, I know it was written up, right? But thinking about how you actually grow, retain your employees, we know that there's a moron talent, especially in the cyber space right now, right? Everybody's trying to get everybody in the door. Pipelines are a little bit dry, a little bit hard to find. It's a tricky scenario. What we want to do, if you allow us for this session, is to almost park that call to the site. We're just going to suspend this belief we have for a moment. We're just going to put that aside. And instead what we're going to focus on are the employees that you actually have on board. All right. We know that the hiring piece is complex, it requires dollars, an HR and a whole bunch of stuff. That's that. We're going to focus on the rest, which is your workforce today, how do you grow them, how do you retain them? And in doing so, you actually find that they'll become more effective and efficient and committed to your organization, so that then they'll go ahead and become random ambassadors for you, which helps that talent recruit group pipeline in the end anyway. Okay. Great. You're going to jump in with a couple of introductions and then look what we're going to be. So my name is Susie Candace. I am a senior manager in the Lloyds Cuban Cabo practice. I focus on organizational transformation. What does that mean? That's where we put all of our organizational design work, so how offices are aligned up the org chart, and how the information and decisions flow. In addition, it's where we hold our culture and our communications work. I've spent my career at Deloitte and Beyond, previously, focused on culture and communications and employee and workforce programs and how you encourage that engagement and your job experience is the best taking day. I've done that in a couple of client spaces, including the federal state, state, state, local, and space, including with some cyber agencies as well. But that'll be excellent. Okay. Thanks, Susie. Good afternoon, everyone. I'm Mara Patashnik. I'm also a senior manager at Deloitte, and I lead workforce experience for the government and public services practice at our firm. I, along the forefront, have worked for our success experience issues for commercial cities, state and federal clients, focusing on training, attracting talent, recruiting, retention, and really today, what we want to do is share some best practices, tools, what we've seen from other clients, and help you give you some ideas about how to become an employer and choice, and how to retain and attract high-performing cyber talent. So we'll dive right in. Before we do, I just want to give one copy out of you. Because it's day two of the conference, now I've been thinking outside for a while. I would say Mara and I both want to copy up the neither of us are going to be cyber experts, especially in the room of this caliber. We do this with cyber clients, but we also do this with other clients. Our invitation to you is to take what we're going to share here and think about how that applies to your day-to-day in your organization. So we kind of did this a little bit to stretch that thinking outside of just this formal cyber realm, and bring you some ideas of what we're seeing from outside. That works. Great. So, absolutely, we define workforce experience holistically as the sum of the workers lived experience at work and how they feel about their organization. And so, really, this is shaped by eight key dimensions that impact worker experience overall. So we'll start by walking through each of these. So really, this framework is backed by research, as well as testing solutions with both our commercial, as well as our government clients, and we start with the people I work with. And this is really support and recognition from managers and who interact with day-to-day. That's commonly cited as one of the top five factors for talent retention, for the technology that I use. So that's very much about frustration-free technology. So are you having to do work-or-arms? Are you having to call IT or spend a lot of your day trying to figure out how to actually use the technology to do your job? And are there ways to use technology to improve collaboration, coordination, and communication in a way that you're able to do your work more efficiently? The maximum focus on a places I do work, which is really focusing on the flexibility of the physical workspace and how that improves employee productivity, increases jobs as faction, and lowers overall workplace stress. Another one that we focus on is a sense of belonging and worthiness that an organization creates. So that's very much about increasing your job performance. I think we've cited that it can lead to a 56% increase in job performance and a 50% reduction in turnover risk. And the work I do, I think this is particularly important to cyber employees where a lot of government employees are driven by the mission of the work that they're doing. And so connecting the work to meaningful experiences and re-enerating that fulfilling and contributes to something bigger themselves is a great way to motivate and keep employees. And that's something that I think government organizations have an advantage over commercial organizations. Another one is well-being. So this is really focusing on personal life. So, organizations can create greater flexibility to attend to well-being means. That increases worker satisfaction. And then another one is really focusing on the mission. So, identifying an organization's purpose and connecting it to their own personal values as well as the mission of the work organization. And finally, education. So this is really about high performing organizations and the fact that 30 times more likely workers are going to be able to achieve their long-term goals and stay within organization if they have the proper training and support and experiences to help them get there. So really when thinking about workforce experience, organizations know often are sure where to start or what elements impact their employees. So this framework is really what we use as a starting point to help orient our clients and to also help them prioritize certain employee activities and relationships. So these are the key contributors that we've seen in our experience to employees really engage and motivated, which has directly contributed to retaining and attracting talent. So with that, we are going to move on to the next slide. So just start off cybersecurity employees we know are experiencing disruption unlike ever before. And so just to a frame today's discussion, we first want to start by walking through how one agency has approached the issue and use workforce experience to address it. So a large federal agency of the Angus Cybersecurity Priorities had long-suing workforce issues. You know, lots of attrition, tough time attracting the right talent for the roles that they needed. And so they really double down and focusing on career growth and development, culture, recognition, morale, and investing in trust in leadership. So as a result of reimagining the workforce experience, they were able to retain critical employees or a major transition. And a few things with these thoughts were as a result of investment in workforce experience programs include the increase in frequency of career-focused conversations, increase in informal recognition across the entire organization, and employees really felt that they had the tools to help the quantum to do their job center and also to prevent burnout amongst their teens, which was an issue that this organization was facing. So next, I'll pass it over to Susie and to speak about the cyber workforce today, probably a little bit about what you all are seeing. Thanks, Mara. So there are a couple of stats up on this slide. I don't think they're going to surprise anybody, but I just want to call them out, because I think they're an important framing for what we are seeing in this hybrid industry. So in 2020, you last hear DeLoi and the National Association of State CIOs to do cyber security study. Is anyone familiar with the study? Can you see it? No? Not, I think that. Okay. A little off, a little. Afternoon reading board. Yeah. And so it was a survey of all 50 state CSOs and three territories, CSOs, so very comprehensive across the United States. And they were asking questions about the general cyber security environment, but in addition, some things came out around the workforce. So when those CSOs were asked about how they can support emerging threats and what their biggest challenge is, number one, they said legacy infrastructure and solutions. Number two, they said workforce. The number two thing holding these state CSOs back from responding to the threats that are coming is the workforce. That's a pretty big issue. In addition, 62% out two-thirds reported that their staff have a gap in competencies. Don't have the skills to do the job. Another pretty big issue. Can you step back and look at it? When we talk about the belonging piece that Mara just talked about and shared, 23% with the respondents didn't know if their organization has established DEI leadership positions. So is there someone in the organization really focused on that inclusion piece and bringing all of your workforce along? And only 25% reported offering remote work options, which I know is tricky in the cyberspace. When you think about the SOC and everything else, but there are ways to do it with at least parts of the workforce. And we know that that's super important to the millennials and slunder representing groups. Remote work has certainly been a topic since the pandemic, right? I see people nodding. It's just on everybody's tongue. It's front of mind. So how do you make that work? In addition to all of this, I think what's really important to note, right, is that cyber threats have not reduced in any way. So there's one stat that says that as a result of COVID-19, it's been linked to a 238% increase in cybercrack. So the numbers are going up. The need continues and the resourcing of the workforce maybe is a little bit lagging behind. In addition to this in the cyberspace, we're also seeing some pretty big macro trends in workforce space in general that are contributing to some of the challenges going on today. And these also probably won't surprise you. Number one, exhaustion. So this is where that burnout topic that someone you people are talking about is showing up. And in cyber agencies, we have frontline workers, right? There's a 24 set of an operation. And so you're going to have this burnout challenge, unfortunately. This is also contributed by working from home in some ways and that blurring of that work life, home life, when the end of the day happens, if it happens, and who stays on their laptop all night working. In addition, between 2021 and 2022, the number of meetings we all experience increased 288%. So you're getting your Zoom fatigue, your Teams fatigue, right? Does that meeting fatigue along the way? Resignation. Who's heard of the great resignation coming out of the pandemic area? Yeah. So that's where this topic fits in. We know that a lot of people left the workforce voluntarily, and in fact, in 2021, more people than ever before left the workforce supporting to the Bureau of Labor Statistics. Some people have come back and some people haven't. So now are options for who's out there working, and interested in being in the workforce, and the talent and diversity that they bring to your organizations could be a little more limited. And then reshuffling. 34% of U.S. workers shifted their cities since the start of COVID. 34, a third of U.S. workers have moved cities. And so this is where that remote work conversation becomes so important about the right people. And so people, again, that you have on board, how are you engaging them, keeping them, you know, working at their most effective to support themselves in your organization? Why does all this matter? Because we know that losing employees is costly. Giving them a positive workforce experience means that they stay with your organization. They are committed. They are loyal. And they continue to grow and build within your organization. One study shows that losing an employee and having to recruit a new one costs three to four times their salary. The opportunity costs, the knowledge drain, the process of just onboarding, completing the job up on the boards, bringing them on three to four times their salary. So if you can get it right on the front end and really have that workforce experience be positive and keep your people, it's going to pay off dividends in the end. I'm going to pass it back tomorrow. Who's going to tie this back to that model for us? Great. Building on what Susie said, workforce experience has never been more important given this crisis and the urgency around keeping talent and finding the right talent. Since the workforce is an organization's most valuable asset, investing in workforce experience activities, we've found has lead to more productive work, more efficient work, more innovative approaches to the work, and more engaged teams which ultimately results in better mission outcomes for your organization. And we found a direct correlation with the ability to retain employees, improve efficiency overall in your organization, and manage organizational changes and transitions more effectively from our work with other agencies as well as commercial organizations. We found that investing in workforce experience is the most sustainable way to create a competitive advantage that improves the business and talent outcomes. And so if investing in workforce experience is a single best way that organizations can minimize workforce disruption in the short term and get ahead of future work, how do we do that? And so in this next section, we're going to focus on how cyber employers can win at workforce experience. So based on a 2022 state CIO, Nassio Ceti that included 51 participants, it was identified that the following areas were some of the top priorities around how to improve workforce experience. We found that 18% of respondents decided increasing remote work options as particularly important. 8% focused on flexible work schedules and how work affects their life. 35% focused on how I grow as a human, so focusing on three-skilling opportunities for rotation programs and the right training to make sure that the workforce is able to meet the needs of the modern IT demands. And so with this, Susie is going to walk us through a deeper dive of the different options on how to address these results if what this actually might look like in practice. Great, thank you, Mara. So again, just to take it from the top, right, we know that the talent pipeline is tough. We know that the recruiting process is challenging. Diving a little deeper into some of these things, we're going to walk through a couple of activities, programs, initiatives, ideas, just to hopefully give you something to walk away with today that you can take back to your organization and implement. And the idea behind these is you can implement them tomorrow. You don't need to get into an IT backlog of some system bill. You don't need to go to HR and say I need millions of dollars to hire people and fill the gaps. These are things that you can do with your current workforce in place today to improve that workforce experience. So starting with how I grow with a human, and again, 35% of the CIO said that this was one of the most important things. A couple of ideas here for things that can be done and things that we've seen be successful elsewhere. Number one, integrating training into the day to day. This is not about sending an employee off to go take a training, get a certification, step out of the workplace and come back. We know that most adult learners prefer experiential learning. And so the way to do that is apprenticeship, mentorship, shadowing, rotational programs, lunch and learns, informal training opportunities, action, after action reports where you can also sit around and talk about what did we do well, what didn't we do, what did we do differently. And some of this also is just documenting what it is you do. So setting up some of those SOPs so that you can pass those over to somebody and say, well this is the best way to do it. And here's lessons learned. Identifying opportunities for employees to use their strengths daily. When we talk about strengths, we don't just mean one of my good at it. What we mean by strengths is what gives me individually energy. What energizes me, what excites me. And so I may be very good at data analytics, but that doesn't mean that that's my passion. I may actually prefer competing teams or communications or something like that. And so checking in with your teams about what is it that each person enjoys most and trying to align them to that work. Studies have shown that if you align people to their strengths, again the things that energize them, they will actually outperform. And do way better than if you give them a challenge task or activity that then they have to sort of overcome and be able to accomplish. We are good when we're playing to our strengths. So it doesn't mean that an employee can only do their strength all day to be happy, but trying to find ways for them to incorporate their strengths into the day to day as much as possible. And then another thing that tends to be useful is stepping back and really looking at career pathways. So how are you going to grow people from level to level, from ability to ability? Is anybody here familiar with the nice framework? Come here, Nick. Yeah, a couple of hands. So this is a great resource if you haven't seen it, and I highly recommend going onto the website. They have work titles, they have job descriptions, they have competencies, they have skills, knowledge, experiences. And you can stack roles so you can say, okay, if I'm an IT leadership management position today and I want to become a data analyst, and they will help you kind of, they map that path a little bit for you with the overlap between those two roles, for example, and then some of the knowledge skills and abilities to go get for one to the other. So a very, very useful tool used by a lot of cyber agencies and organizations, highly, highly, highly recommend using it, or at least checking it out and see if there's anything useful for you there. But really looking at where are my people going and how do I grow them? Because we know also that one of the most important things for employees is them feeling like they have a growth path within an organization that somebody is invested in their development. When we move on to places I do work, I know this becomes a sticky conversation in a cyber environment. I know the sock is what it is, we know that piece. But there are conversations still to be having around work location. What we have found is most successful is when organizations really focus in on work location, in alignment to your desires, your goals, your strategy. There are some things that working remote are great for. Heads down time, idea generation, focus, and productivity. There are some things that working in person are great for. Teamwork, apprenticeship, and some of those sort of incidental connection points, running into somebody by the water cooler and finding out that we're doing similar things, and we can trade notes. Actually, Mara and I just did that, and I said, oh, send me that if you could, because I really want to see that. So designing intentionally for what work can be done in person, what work can be done remotely, maybe it's part and part, maybe there's a single initiative or project where you can send people to go home and work remotely and then come back. And they're still feeling like they're getting a taste of that opportunity if they're normally needing to be on site. So really being smart and intentional and purposeful about how you design that. Looking at the digital communications and information sharing, so a lot of us have moved on to teens and Slack and Zoom and these digital tools. Are you really using them for their maximum capability to support your organization? Is there somebody in your organization who really is tasked with understanding all of the features that you have licensed to and bringing that goodness to your organization to help your people work even better? And then the third one there is forcing connectivity and networking, especially in a high rate or remote environment. We know that happiness, at work, and elsewhere is tied to social connection. So how are you forcing some of that social connection? Maybe it is in those brown bag trainings or after action reports. Maybe it is taking a moment before diving into a meeting and saying, hey, how is everybody's weekend or what are you doing this weekend or tell me something funny or if you were going to be a office supply, what would you be? Whatever it may be, right? But finding a way to sort of force a little door opening and having people build that connection. And then the well-being piece. And so I shared with you about the exhaustion and the burnout rates and all that earlier. A lot of organizations, especially since during instance the pandemic, have stood up well-being activities. They are not necessarily aligned to a strategy. What is your mission with those well-being activities? How does that align to your broader organizational strategy? What are you trying to achieve? Then what are the initiatives that you lay in with that? So it is not just a scatter plot of activities, but they are sort of intentional and organized, aligning back to those goals that you have. And then the matter of the impact of that and adjust as needed. And if you are engaging your workforce correctly, then you will have lots of data points about what is working and what is not in order to make those adjustments. Starting on this a little bit. But again, thinking creatively, not just about remote work, but how shift work is scheduled. So a lot of the conversation around workforce and economic development is around people who may be to meet their child care duties at home while still trying to get into the workforce. Or they are taking classes at night while trying to get into the workforce. What is it that you can do to maybe look at four hour shifts or six hour shifts different than your typical shifts to bring in more people and diversity and also alleviate some of the burnout and pressure that is happening for the workforce in place today? There are a bunch of apps that focus on this. A lot of them are focusing in the healthcare space of the transportation frontline worker space, but they could easily be leveraged over into the cyber space as well. And then the third one here, conducting a culture assessment. A lot of organizations assume a culture sort of just happens and it doesn't. It is important to be intentional about culture. And what do we mean when we talk about culture? We all live in it all the time, but what is that? When we talk about culture, what we talk about are the intangibles, your beliefs and your values as an organization. Are we more individually, individualistic, the focused, or are we more collaborative? Are we more risk of first, or are we more innovative and risk takers? Are we more internally focused, or do we have external customers that we've really organized around? So when you do a cultural assessment, you look at where you are today versus where you want your organization to be and what are the steps to get from A to Z. And how do you do that in a way that then filters and flows through your entire organization so that by the time you get to performance reviews or your next strategic planning cycle or standing up on your program or redesigning your organization, there isn't friction with what it is you really want from your employees and what you're trying to achieve as an organization and how you're getting there. So cultural assessment, well it seems like a sort of unnecessary, it's actually very effective and super supportive of your protocols. Okay, I've talked a lot, lots of ideas, hopefully a good menu for you to pick at least something from if it's useful. I'm going to pass it back to Mara to kind of bring it all together for us. Thank you, CZ. So how can employers win at workforce experience? So when you're practically speaking, the key to implementing a successful workforce experience really starts with understanding what your workforce wants. That can be through a survey or focus groups doing user research, but really putting their voice at the center of these initiatives and then working to prioritize which ones within your organization can help you achieve your overall business objectives. So we're going to walk through our perspective on what it means to really bring workforce experience to an organization. So starting with employers need to think about beyond employee engagement. And so what we mean by this is it's not just one data point, are my employees engaged, how have I retained my staffer, did I recruit an attract, you know, this number of individuals. It's beyond that an accommodation of those eight characteristics that really helps to retain employees and attract the right talent and enhance the overall workforce experience. So an example of that is looking at physical workspace. So do they are they able to do their work remotely in some cases, as Susie mentioned, do they have everything they need in the office in order to be productive, to get things done in terms of head sound work versus team collaboration. Culture, do they, you know, are they recognized for the work? It's great as a hundred dollar gift card is for a great job and something. In a lot of cases what we found is it's equally just as advantageous to have a manager reach out and say, thank you so much. You did a great job and acknowledge that and that's often what can motivate and keep employees engaged. And another area is technology. I think we mentioned, you know, it's a great way to help collaborate, work remotely, but it can also be an impediment to when technology isn't easily accessible within an organization. There are challenges, IT is available to help with certain things and can often be like a source of frustration for employees. And so really the way we look at it is not just one area at an employee engagement, but how do you approach across those eight dimensions, multiple areas of prioritizing different initiatives. The second one that we really focus on and this goes back to what Susie said about the assessment and also what I mentioned around the voice of the workforce. And so incorporating that throughout the design is incredibly important. One of the first things that we often do is a poll survey of a particular group to understand what they really hear about. What are their preferences? If they had, it's not just, I want more money, but rather if you had a choice between a little bit more flexibility in your work schedule versus, you know, increased recognition, where do they really stand in terms of what should actually get prioritized. And when you drill down into the heart of it, often is it what you think? Is it the obvious answers around like, I want flexibility? It's like, but what does that really need and how can that work for your organization in a way that maybe hybrid works for some workforces, but not necessarily for a cyber workforce. It's really, you know, is there a way that they can get different experiences for rotation program as Susie mentioned. And so worker input, I would say, is the number one thing to invest in and even, you know, communicating that you're doing this and engaging the workforce early on and incorporating them into this because this is important to the organization and to you as leaders is an important way, I think also to retain talent. The third one is a continuous listening approach. And this one really focuses on not just pulsing a workforce once a year through an annual HR survey of, how do you really feel like, you know, what leadership considerations should we implement or, you know, how can we enhance the performance management process. It's really on an ongoing basis and even in an informal way, you as leaders and managers of teams asking your teams, you know, what they care about, what they're frustrated about, what their preferences are, and taking small steps where you can to try to incorporate that into how they work. And, you know, sometimes it's challenging to do organization-wide changes that take a lot of funding, investment, and capabilities. So oftentimes a way to do that is through, you know, more informal touch points on individual teams that you're leading. We found that that's a cost-effective, quick way to be able to get a lot of employee engagement and routine and keep your talent happy. The final one is workplace technology, and I know we've talked about this a little bit. And I think really with this, it's, how do you boost productivity? Workers want to do a great job. They're, they're the performed well. They, you know, are often very mission driven. They want to grow their careers and they, you know, want to try to be as efficient effective as they can in their job. But oftentimes, you know, what the organization decides in terms of technology versus like what the worker actually needs. There's a bit of a disconnect. And so really looking at, you know, how this can be a major accelerator based on how you work. Do you want an ants collaboration? Do you want to try to create more opportunities for high-grader of work? And, you know, through technology, that can really enhance and accelerate a lot of those activities. And so with all of these four areas combined, you know, this is what we've seen across, you know, Fortune 500 companies, federal agencies, the city, state agencies as well as kind of the key characteristics and commonalities amongst the most effective employee engagement and workforce experience programs. And just kind of round things out with this quote is, you know, if workforce experience I recognize, you know, can seem kind of fluffy and like, oh, that's a nice to have. But our premise is that your workforce is the most important asset in your organization. And if they are the most important asset, you really want to invest in them to, you know, day to day be the driver of change in, you know, be the builder productivity. We've just found that, you know, by investing in this and putting the workforce as, you know, the center part of what you invest in as an organization and leaders, it's not only about retention, talent, you know, the cyber workforce crisis, but people want to do work well and they're able to get more done and achieve more without you, you know, directly supervising and micromanaging or looking at everything because, you know, you know, you know, you're not going to be able to do anything. And you know, you know, you're not going to be able to do anything because you're not going to be able to do anything because you're not going to do anything because you're not going to do anything. So with that, that is our presentation for today. I hope there was a little bit of, you know, the landscape of the cyber workforce with some practical tips that you can take away for how to just think about, you know, improving the overall workforce experience and investing in your employees. So with this, you know, we know that all of you are in the trenches every day, you're facing this, you're living this, and we are just interested to hear from all of you, you know, just to start, like, what's one thing that has worked well in your organization in terms of enhancing or investing in the workforce experience? I'd love to get someone to start. Don't be shy. Yeah, sometimes it's important to have something or what hasn't worked well, you know, that's something that's easier to start with, you know. But we'd love to just hear from all of you, because I think, you know, how we've aggregated a lot of these best practices and what we've come with is by, you know, hearing other experiences from other organizations. And so one of the best and most effective things that I feel like I take away from conferences often is hearing from my peers and what they're facing and, you know, let it, you know, similarly I could bring back to what I'm leading. What about you? Oh, I see a hand back there. \ No newline at end of file diff --git a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-AGENDA-FULL.txt b/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-AGENDA-FULL.txt deleted file mode 100644 index 8ad3ff1c..00000000 --- a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-AGENDA-FULL.txt +++ /dev/null @@ -1,47 +0,0 @@ -AGENDA: Most important things to look for in a start up - -TAM: Make sure the market is sufficiently large than once they win they can get rewarded -- Medium sized markets that should be winner take all can work -- TAM needs to be realistic of direct market size - -Product market fit: Being in a good market with a product than can satisfy that market -- Solves a problem -- Builds a solution a customer wants to buy -- Either saves the customer something (time/money/pain) or gives them something (revenue/enjoyment) - -Unit economics: Profit for delivering all-in cost must be attractive (% or $ amount) -- Revenue minus direct costs -- Raw input costs (materials, variable labour), direct cost of delivering and servicing the sale -- Attractive as a % of sales so it can contribute to fixed overhead -- Look for high incremental contribution margin - -LTV CAC: Life-time value (revenue contribution) vs cost to acquire customer must be healthy -- LTV = Purchase value x number of purchases x customer lifespan -- CAC = All-in costs of sales + marketing over number of new customer additions -- Strong reputation leads to referrals leads to lower CAC. Want customers evangelizing product/service -- Rule of thumb higher than 3 - -Churn: Fits into LTV, low churn leads to higher LTV and helps keep future CAC down -- Selling to replenish revenue every year is hard -- Can run through entire customer base over time -- Low churn builds strong net dollar retention - -Business: Must have sufficient barriers to entry to ward off copy-cats once established -- High switching costs (lock-in) -- Addictive -- Steep learning curve once adopted (form of switching cost) -- Two sided liquidity -- Patents, IP, Branding -- No hyper-scaler who can roll over you quickly -- Scale could be a barrier to entry but works against most start-ups, not for them -- Once developed, answer question: Could a well funded competitor starting up today easily duplicate this business or is it cheaper to buy the start up? - -Founders: Must be religious about their product. Believe they will change the world against all odds. -- Just money in the bank is not enough to build a successful company. Just good tech not enough -to build a successful company -- Founders must be motivated to build something, not (all) about money. They would be doing -this for free because they believe in it. Not looking for quick score -- Founders must be persuasive. They will be asking others to sacrifice to make their dream come -to life. They will need to convince investors this company can work and deserves funding. -- Must understand who the customer is and what problem they are helping to solve. -- Founders aren’t expected to know all the preceding points in this document but have an understanding of most of this, and be able to offer a vision. \ No newline at end of file diff --git a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-AGENDA-HEADERS.txt b/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-AGENDA-HEADERS.txt deleted file mode 100644 index fd8034a2..00000000 --- a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-AGENDA-HEADERS.txt +++ /dev/null @@ -1,8 +0,0 @@ -AGENDA: Most important things to look for in a start up -TAM: Make sure the market is sufficiently large than once they win they can get rewarded -Product market fit: Being in a good market with a product than can satisfy that market -Unit economics: Profit for delivering all-in cost must be attractive (% or $ amount) -LTV CAC: Life-time value (revenue contribution) vs cost to acquire customer must be healthy -Churn: Fits into LTV, low churn leads to higher LTV and helps keep future CAC down -Business: Must have sufficient barriers to entry to ward off copy-cats once established -Founders: Must be religious about their product. Believe they will change the world against all odds. \ No newline at end of file diff --git a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-Summary.txt b/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-Summary.txt deleted file mode 100644 index eb0762af..00000000 --- a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-Summary.txt +++ /dev/null @@ -1,10 +0,0 @@ -Summary of: recordings/42min-StartupsTechTalk.mp4 - -The speaker discusses their plan to launch an investment company, which will sit on a pool of cash raised from various partners and investors. They will take equity stakes in startups that they believe have the potential to scale and become successful. The speaker emphasizes the importance of investing in companies that have a large total addressable market (TAM) and good product-market fit. They also discuss the concept of unit economics and how it is important to ensure that the profit from selling a product or service outweighs the cost of producing it. The speaker encourages their team to keep an eye out for interesting startups and to send them their way if they come across any. - -The conversation is about the importance of unit economics, incremental margin, lifetime value, customer acquisition costs, churn, and barriers to entry in evaluating businesses for investment. The speaker explains that companies with good unit economics and high incremental contribution margins are ideal for investment. Lifetime value measures how much a customer will spend on a business over their entire existence, while customer acquisition costs measure the cost of acquiring a new customer. Churn refers to the rate at which customers leave a business, and businesses with low churn tend to have high lifetime values. High barriers to entry, such as high switching costs, can make it difficult for competitors to enter the market and kill established businesses. - -The speaker discusses various factors that can contribute to a company's success and create a competitive advantage. These include making the product addictive, having steep learning curves, creating two-sided liquidity for marketplaces, having patents or intellectual property, strong branding, and scale as a barrier to entry. The speaker also emphasizes the importance of founders having a plan to differentiate themselves from competitors and avoid being rolled over by larger companies. Additionally, the speaker mentions MasterCard and Visa as examples of companies that invented their markets, while Apple was able to build a strong brand despite starting with no developers or users. - -The speaker discusses the importance of founders in building successful companies, emphasizing that they must be passionate and believe in their product. They should also be charismatic and able to persuade others to work towards their vision. The speaker cites examples of successful CEOs such as Zuckerberg, Steve Jobs, Elon Musk, Bill Gates, Jeff Bezos, Travis Kalanick, and emphasizes that luck is also a factor in success. The speaker encourages listeners to have a critical eye when evaluating startups and to look for those with a clear understanding of their customers and the problem they are solving. - diff --git a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-Transcript.txt b/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-Transcript.txt deleted file mode 100644 index 8269c2cd..00000000 --- a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk-Transcript.txt +++ /dev/null @@ -1 +0,0 @@ - because Max is really aware of that. So basically we're pretty close and pretty finalized with that partners to launch funds. I say funds, but technically this structure will be a corporation. And the difference is if you do a funds, there's very strict rules and regulations and a lot of compliance work with financial authorities. And I've done that in my past job and I really don't wanna do that because I know how intensive it could be and how much of a time drag. So what we're gonna do is incorporate a holding company and call it an investment company. And it's gonna sit on a pool of cash that we raise for people and we'll just keep it in the bank until we find investments that we think are good or suitable startup investments. And we'll take equity stakes in those companies and help them grow. And the funds that we raise will be a mix of my connections, our connections. And end partners is gonna put us in front of a lot of different investors that they know, which is why I was working on the pitch deck and there's more to come. So I was planning on doing this talk a bit later, but we spoke to the head of that partners this week. He would prefer us to have more of a pipeline when we go speak to investors, meaning companies that were close to pulling the trigger on an investing. And your reality is a pretty long process to get to know a company and go through all the details and do all the research. But I'd like to get started at least meeting founders of these super early stage companies because that's what we're focused on. And the thing is, given we have 36, 37 employees all around the world, it makes way more sense. It can be helpful. If I give everyone on the team a bit of a grounding and just what you should be looking for, what some of the key characteristics are of a company that could scale well and become huge one day. Because you know, we're not looking to invest in the dry plingers down the street. Maybe a fine business, but it's not a business that you really take to become a large enterprise and make a ton of money, which is what we're focused on. So if I kind of train everyone or at least explain these concepts and you guys all have your own networks, you're all different parts of the world. You have friends and your friends, you're gonna hear stuff. And I'd like you to keep your ears open and your eyes open and when you come across interesting, even interesting companies if you don't know, if you just find the cool local tech company, you can send it my way. I can always reach out to the founder. Most people are always very, very happy to speak to investors because pretty much everybody in the startup world needs my. So that's the point today. Just before I jump into this document here, does anyone have any questions on that in the funds? I think we had one before, right? So Jan? I didn't understand. Okay, anyway, which part? I missed it. Okay, anyway, continue. Okay, so just in terms of what we're looking for, Sujan, I think you're asking about a return. I mean, realistically with startups, so many of them go bankrupt. Like you invest in the intentions. It's just what it's, you know, it's the nature of the game. So let's say we make 10 investments, I would expect maybe two of them or three of them could be aquahires where you get your money back. Maybe you break even maybe five or six are complete zeros which happens for a very small return, or you know, some sort of recovery, but a loss. And then maybe one or two to be home runs like 10 to 10 next. Right, and those pay for all of the mistakes. And that's really the purpose of do and venture investing or angel, which is even you, right? What we're doing. All right, so all you're in through, so first is Tom. I'm sure you guys have heard me talk about this before or what it stands for is total addressable market. So what that means is if you were a business and you captured 100% of the market, well, that looked like in sales. Let's say we were, I don't know, doing cloud computing and you know, obviously the biggest companies now are KWS and Azure and they have big market share. But let's say one company captured all of it and it was a trillion dollars in revenue a year. It's less than that of just using that example. So that would be the time. So when you were investing and I said, you don't really wanna look at like the drag cleaners down the street, it's cause we want a company that's starting small, but it's really going after a big market. There's a lot of really cool companies out there that solve a really niche problem and that's great. They could be more of a pet project or like a mom and pop store, but as an investor we don't really wanna touch that because the ability to get that 10, 20, 100 X is pretty diminished. If they're not solving a problem in a big addressable market, there's not a lot of potential upside. There's some exceptions like medium sized markets in work. So for an example of a medium sized total addressable market that worked pretty well as Etsy. I don't know, does anyone know Etsy? They do like local kind of arts and crafts are very customized and you deal with people online. So Etsy is not a huge market and one of the reasons a stock did do great at the start is people thought it was so nichey and so specialized that maybe Amazon would just either crush them which they didn't or the market wasn't big enough. Etsy proved that wrong and it wasn't huge but it was medium sized and Etsy stock has been extremely, extremely well. The market was bigger than people thought. Etsy captured not 100% but a very large percent of the total market and that's partly because bigger competitors like Amazon mostly ignored it because they didn't really see the potential. So there's other exceptions to the role where medium sized ones work like Etsy but in general we're gonna wanna go for big ones. Yeah, my wife buys it, kinda stuff off Etsy. Anyone have any questions on TAM or companies that you can think of? And they're curious if it's a big enough TAM or not. One question, Jordan, when you're talking about market, how do you define a market? Meaning that as we're located in different countries, how we can tell, this is going to be big here in my country, in my neighborhood, well, not in my neighborhood, but you know what I mean? Yeah, so depends on the company's plan but generally TAM would be the total market that you can reasonably address. So for Amazon, it's global online commerce, right? Like they touch everything. So any sort of retail online, that's Amazon's TAM. Cloud competing in any country, Amazon operates in all of them. So if you're a local company and you have zero plans, let's say you're in Canada and you have zero plans to go to the US, you can't really count the US but if it's been part of your long-term plan, is to go to the US and there's no roadblocks. Like let's say you're selling food, it's very hard to import food over borders, people do it, but it's harder. The US would not be your TAM. If you're a tech company, there's zero limitations, like Shopify, clearly Shopify is a Canadian company but their TAM is global because they sell around the world. So it depends on the type of company, but yeah, most companies we're going to be looking at are going to have global TAMs. So it's going to be worldwide or at least most of the developed world, which is the big chunk of the worldwide economy. Okay, so the next concept is product market fit. I wrote, being in a good market with a product that can satisfy the market. There's a few things here, so there's a lot of cool companies but maybe no one's actually going to pay for it. So you don't really find out until you get into the market and you start selling. Unfortunately, most of what we're going to be investing in is before that, they haven't actually launched the product. And of course, internally, we deal with a whole lot of companies like Virtupoker where we have a good idea, you know, what it's going to be like and we'll fix it. But we don't really know until we go into the market. One of the things we guess right, one of the things we guess wrong and how can we adjust it. So there's a few things to think about and we're going to be investing before companies are really selling. The good thing is you get them cheaper. Like if you invested in a good startup at five or six million bucks, it's probably because they haven't gone to market yet then all the sales. Like once they start selling and it's clear click, like most people consider that already ready for Series A, which is a further venture round. And a lot of the valuations could be like 20, 30 million. Coorsnet online, let's talk about like that BRD project he's doing. So that solves a real problem, right? It's both like breast rotors disease. It's the number one health impact that affects farmers per cows, above meat and milk. And I calculated the damage. The damage of the like the cows die. So total loss or after they recover from the disease, they don't gain as much weight. So you can't sell them for as much because it's meat times price. So it's pretty easy to calculate the total economic damage or harm. So if we stick in with BRD, if we solve that problem, meaning we recognize it earlier and prevent the farmer from having those losses, we can charge for part of the amount we're saving. The conversation I was having with them is why don't we charge about 20 by a percent? If the farm is going to have about $10,000 a year and economic damage, we can charge a quarter of that, $2,500, that's a real business. And it's a real solution, something wants to buy. We won't know how it works till we get in there and that's a product market fittest, but we can try. So the thing to focus on when looking at a company is it's saving the end buyer about time, money or pain. So for that example, let's use Uber. Uber solved the problem of getting a taxi was extremely painful. It was an old system. You had to call them on your phone and say, please come at 630. You couldn't see how close they were. So that saves a lot of, that saves people time and pain. Or it gives people, it gives people something like revenue. So if you're selling a business, some new product that gives them revenue, they're going to buy that. So for that example, think about booking.com. See if everyone knows booking. It's like the biggest travel website in the world. And the customer there on that side is the hotels. The hotels are all tied in. And the hotels know, oh, booking is aggregated. So many customers like you and Ray, you want to travel and gunpowder tells, if I'm a hotel in Milan, I better tie into booking.com. Because everyone's going to be searching for hotels in Milan. They're going to go to bookings. So if I tie an even a pack that pay booking 20% of the cost of the hotel might, or 15 to 20, which is what they charge, they're going to do it. Because they're going to bring a ton of revenue. And that's why bookings, a huge, huge business. And finally, could bring the customer something like enjoyment. So think about Netflix, right? It brings them joy. It brings them entertainment and they value that. Same with Nintendo, right? It doesn't save time or money, obviously. We're generating revenue, but it brings people entertainment that they're willing to pay for. So when we're evaluating companies, think of those three buckets, and really focus on is this company providing one of those in a way that people are going to want to pay for, does that make sense? Anyway, any questions there or thoughts about other companies you come across that do that well? OK, I'll keep moving along. I'll stop asking for questions and just jump in or raise your hand if you have them. OK, the next one is unit economics. So what does that mean? It's looking at the total profit for selling the product or service or whatever it is, minus the all in cost ability to, and you want it to be attractive. You don't want to like sell something and it costs 99 cents to deliver it. You sell per bucket cost 99 cents in your profits or 1% that's awful. So the way to measure this is revenue minus direct costs, and that's the unit economic itself, because unit economics mean building that one thing without all the overhead. So let's ignore, let's look at an article, for example. What are the unit economics? It's what we build out our developers at minus the developer's salary. That's the unit economics. Every company has certain amounts of overhead that aren't direct, but you still need to build off of. So let's say anyone on the upside, anyone in my role who's not writing code, not billed by the week or month, that would be more on the fixed cost side. Anyone on building code is kind of like in business we call it like a right book cost, and this is labor costs. So what we're looking for is companies that have really good unit economics, because that really allows them to scale and make a ton of money. And the next step of that is incremental margin. Let's say Facebook. Facebook has great unit economics, right? Like they serve ads, they sell ads. What's the cost of delivering that ad? If they get a dollar an ad revenue, it's like, I don't know, some basic server costs. Maybe it's like three cents per dollar of ad revenue. So it's huge. And what's the incremental cost? They're ready fully staffed, in fact, if I'm a people. So for every dollar of revenue they bring in, they'll get like 97 cents of gross profit, but they don't really need to add that money more operations people or that many more tech people and R&D. So their incremental contribution margin is huge. Like at the start it might be zero, because even if they're making 97 cents per dollar of incremental revenue, they still have to add operations people, they have to add tech people, they have to add sales people, and all that cost would eat up that 97 cents. But once you get to a certain level, it's completely incremental, and it works really well. Some other businesses with really high incremental contribution margin, MasterCard and Visa. Like they have some of the highest profit margins in the world, why? Because they're ready to set up, right? Like there's really just swipe the credit card, cost almost nothing. They get a fee every single time to do it. They're ready fully staffed. So you know, a hundred more businesses turn on tomorrow, say, hey, I'm gonna take MasterCard, it's free revenue for MasterCard. There's almost no cost associated with that. So really those are really good businesses. And I want you considering what the unit economics look like. Because you don't want to invest in something like, I don't know, some product that's just very low margin, it has no chance to get the high margin ever. So okay, I have low margins today, but you have to have high incremental unit economics to get to high margins eventually. The next one ties into that. So that's lifetime value and customer acquisition costs. So it's kind of measuring how attractive is it? How worth, how much is it worth to be spending money on marketing and sales to bring in a customer? So there's the big thing that's kind of math. The first is lifetime value. Lifetime value means how much is a customer gonna spend on your business over his whole customer existence? So obviously for customers gonna stick around for one year, it's not as valuable as a company that's gonna stick around or clients gonna stick around for 10 years. So it's how much they spend on average for transaction, times, how many purchases they'll make over their lifetime. So for a company with like a huge lifetime value, for customers think a Costco. You come in, you always spend like 500 bucks, it's more than you expected. Maybe you go like quite some month, you have a family. Nobody leaves Costco, they sign up for their membership like a hundred bucks a year and they're spending like a thousand bucks a month at Costco for like a decade. So the spending 12,000 a year times 10 years that's 120,000 that's huge. So even though Costco has like thin margins because people spend so much in the relief. Yeah, Hannah, you got kids, you need Costco, you teenagers. That's clear. We're just starting our Costco journey. These my kids are young, but yeah, my wife's there all the time now. But you're gonna keep going for a long time, maybe till your kids move out of the house and then it doesn't make sense together anymore. And then the other measurement here that's important is customer acquisition costs. I'm sorry, just the key state digital, like them value same thing for Amazon, right? Like once they acquire customer or the customer it tends to order, order, order. The lifetime value is huge and it's still growing because Amazon sees very little customers actually just outright quit. And since they, like there's been customers spying on Amazon since 1998 and they're still buy it, so that value just keeps going and going. Customer acquisition costs measures the cost of society and the first time so the way to calculate that is look at the cost of sales so that can be like advertising and marketing people plus the cost, yeah, so the cost of advertising, marketing people building up your brands, all of that. For Stock2Ware, it's generally those that go costs and what you measure is okay. We added 1,000 new customers this quarter. We spent a million dollars on marketing. So clearly the cost for new addition was 1,000. Let's say the average customer is gonna spend $4,000 over the lifetime, just spending $1,000 to gain a customer. They're gonna spend $4,000 over their lifetime, which is a pretty good ratio. That's 401. Anything over three is good. And for Stock2Ware, if we stick with the like, let's call it 80% Chris margins, because it's, you know, Stock2Ware, there's not a lot of cost to roll out one more customer. The $4,000 might translate to 3,000, the unit contribution, profit contribution. So you spend the thousands, get 4,000, sales and 3,000 profit, that's pretty good. That's why it's such an important measure, right? Like you created so much value. In that example, every customer you add just created $2,000 of value. So if you're an investor on Wall Street or an investor in early stage companies like us, you really wanna see that. And a lot of our companies won't have that yet, but it's still an important concept to understand, because as you scale and do other rounds, I promise that you're capital, that you're capitalists look at that very, very well. Something else to get your cash low, your customer acquisition costs lower, is you want your customers to vagalize it. Like the cheapest way to grow is a word of mouth, right? Like Amazon didn't spend any money on paid advertising for years and years and years. Like Jeff Bezos would just say, grand theost things, go to conferences, and be featured in barons. And like all these people who didn't know about Amazon in 97, like all of a sudden, they're getting all over the front page of Wall Street Journal, not paying for it. People are like, oh, I've gotta check out this new thing called Amazon. Actually, I was listening recently, Bezos went public early with Amazon when they were pretty small, because he thought it'd be great publicity and free press. So if you get that free press, that's the last money in advertising, that ratio goes up because your customer acquisition cost goes down. Think about GitHub, right? Like a lot of new engineers, they've got to really spend money to get them to start using GitHub. No, they just come to GitHub, because everyone's like, oh yeah, you gotta get on GitHub. I stored my code here, you gotta come check it out and start coding here too, and posting it all. So their customer acquisition cost is very low, which is why it's such a good business. So again, the rule of thumb, a thumb is three or higher, and under that, you don't really want it. I'll just wait a sec, any questions? Yeah, that's a fair point on Dress. Tesla had like way more demands than supply, so that any depend anything on ads. Now they supply cut up and the man's flattening a bit, so they actually have to start spending some money. Now you're saying, they're still spending very little. Yeah, you're right. Eventually they'll have to spend more though, but you're right. Okay, so actually have a question now. How much, like a normal company like Ezra, how much would you spend on advertising? Like how do you calculate how much do you spend on advertising? Well, it's an equation, right? Like a CAQ to LTV, and it kind of salt math at the end of the day. If you had perfect knowledge, and you need like one more piece of advertising, drove like 0.2 customers in each customer generates, like let's say you wanted to completely maximize, you'd make it say your contribution margin, on incremental sales, is just over what you're spending on ad revenue. Because that's just the math equation. Does that make sense? Like if you spend the dollar on ads, and it contributed $2,000 in gross profit, cool. You know, that's working. And without having to invest anymore in infrastructure, reality is a lot more complicated, right? Like if you're, I don't know, well, let's see, I got like you don't really want to advertise a ton in the huge and everywhere, and then getting to ubiquitous, because you grab it, damage your brands, but just like an economic textbook theory, and be like, it'd be that basic math. Okay, churn. So we all know what churn is, the churn in, churn out, canceling Netflix, whatever. So churn fits into lifetime value, right? Because lifetime value measures how long customers last. If a company has a lot of churn, and the customers don't last very long, and their lifetime value is low. On the inverse, that super sticky product, and everyone loves it, nobody leaves, or they just can't leave, because they're trapped, which is great from the business side. You're gonna have very high lifetime value. So I'll just call it a few businesses at a high churn and low churn, and it'll be kind of intuitive to you. So one is like selling through a publisher revenue every year is hard. If you have customers that are on repeat and growing with you, like some of the best tech companies, life is easy, right? Like you do nothing, 98% of your customers stick with you for next year, 2% leave, and then 98% of customers who stick with you probably group, because they're consuming more. So let's think about data dog snowflake. A lot of those companies, they're words still are some extent, like Wall Street darlings. Why? Because every year, their customers take more and more services from them. So data dog doesn't have to do anything. They literally could fire their, pretty much their whole sales force, and if their revenues are 100 bucks this year, they would drop to 98% because two percent of clients leave, and then go to 113, because their remainder consumed 15% more. So that's phenomenal, because it just such easy growth. Let's talk about something that are bad. So like meal kits, they lose like half their clients every single year, they turn out, let's look at Palatine. Now that COVID's done, a lot of people come into Palatine, they love it, they use it for a couple of years, and then they cancel. Maybe they're spending money, and they quit working out. I mean, that's like a super standard gym model too, in the real world, right? You sign up for the gym, you use it for a year or two, maybe you forget about it, and you're still paying and not using it. Eventually wake up, see your credit card bill, you're like, this is stupid, and you cancel it. So it's a very high-turn business. The good thing for what we're looking at is we're mostly going to be looking at tech companies and not retail tech, because retail tech does that by turn. But enterprise software tech tends to have some of the lowest turn around, and that's why it's some of the best businesses around. They get super sticky, and they're very hard to leave. Those are phenomenal businesses, and when you're looking at them, try to find businesses that have those characteristics. When you know people are going to be on it, they're going to stick with it, and not leave. And that kind of brings us to the next point, which is we're looking for businesses that have really high barriers to entry, to make sure copycats can't just come in, and mimic you and kill your business once you're established. So, you know, we're just talking about low-turn, so one of the things that makes it really good mo, meaning someone else can't just come in and duplicate you overnight, and they can't kill you overnight, is high-spotting costs, like you get locked in. There's a client that's really annoying to be locked in, but as a business, it's phenomenal. So, think about like Google Cloud, right? Like you move everything on cloud, we were speaking to Dochibo this week at the conference, and they're probably going to use Google for AI. Google's basically going to be giving in, they told us a bunch of fine-cuned foundational models off the Palm 2 for free, and see Janet and Shree have questions on that. And we're like, oh, you know, I asked Max and he were right, and like, why is Google just going to give you this stuff that I'm going to charge you? And it's like, Max, they want to lock everybody in. And the table's like exactly, we're going to be really cautious to like be able to move in a year if we need to, but Google's goal is going to be giving away foundational models, lock everyone in, make them use Google Cloud, make them use Google Tools, and it's going to be very hard to switch off. Any questions on that before I move on? Okay, the other things you want to look for is like make the product addictive, especially if it's in like the entertainment space, your video games, you wanted to addictive as help for the client, the customer, so they never leave in the key plane. I think that's evident. You want really steep learning curves sometimes for a product that's taken off, and that's a form of switching costs, right? Like if you think of how long and hard it is to really get your employees up to speed on something, they're not going to use something else. Like people, like designers, maybe just like, produce owners off Figma. They learn Figma, they're like Figma experts, they're not going to leave. Like that's why Adobe had to go and buy Figma. Figma, like that's Adobe's game plan, right? And Adobe was losing the market chair because Figma was so good, and all these people are like being trained on Figma. The best is when you see a company, and universities are offering classes and how to learn this, just like that's phenomenal. The next generation is just getting indoctrinated and trained how to do this. The universities are building up your software companies' value for free. So switching costs, yeah, Adobe for sure. So switching costs are high if you spend a lot of time trading someone internally, and it's hard to get people to use your product, but once they do, it makes it really sticky. So you kind of want them to become local experts on your thing, and it's just like a way you can make your business extra sticky. Okay, another one is two-sided liquidity. This is big. Basically, this is more for marketplaces, but like think about any business where there's two sides. We talked about booking before, right? You need to get all the consumers, and you're gonna use the hotels, and then you need, which is demand, and you need all the hotels themselves, which is supply. And if you have all this supply, but no consumer demands, the hotels are gonna live. If you have all the consumer demands, but they can't book anything on the site, they're gonna leave. So it's a bit of a chicken and egg, which is why it's very hard to replicate, and knock off the companies that are doing this. But if you do it well, like booking did, you scale both to a nice level, or if you do it well, like Uber did, you're always balancing to make sure you've been of drivers and enough passenger demands. Your drivers don't leave, and your customers keep staying. And you could grow, and it's really, really hard. It's the same with credit cards, right? Like they have two-sided liquidity. If I had, and this brings up my next point, if I had, or my last point, a billion dollars to go start a new credit card company, and I could just blow the billion to try to build it. Could I do it? Could I go to Merchant and be like, hey, you want to pay anything to take my credit card? Just gonna quick integration, and you're done. Okay, really, versus 2% for Visa. Are they really gonna offer it? If I'm like Jordan Card? No. If you have a huge brand in your global, you can start with something like Apple's doing, but Apple's the biggest company in the world. But generally, liquidity, two-sided, liquidity marketplaces, like credit card systems, are extremely hard to knock off. And that's why there's some of the best businesses where Visa and MasterCard have 60% margins. Other things that offer a nice mode that can really protect you is patents, obviously, like anything that's really patentable, hard to knock off, but not impossible. Patents aren't the best mode. Like I prefer two-sided liquidity systems to patents any day. Same with any type of intellectual property. It's also not as good as two-sided liquidity and some of the other things. And then branding, branding's huge too, right? We talked about Apple. Like there's nothing that overly unique. But as Andreas and I were talking about the start of the meeting, Apple did a great job building up two-sided liquidity on the iPhone when it was released, right? Like all the apps, you have to have the developers come and build on your platform, so that's supply. And you have to have tons of consumers have an iPhone, which is demand. And then if you're launching a new one, the developers are like, I'm not gonna build that new OS because there's no one using it. Wow, and I waste my time. I'm gonna stick with iPhone. I'm gonna stick with iOS. It's like way better. And that's the benefit of two-sided liquidity. And that's the benefit of Apple branding. And why maybe they'll have an impact in finance and ultimately have more Apple cards, not to take on master or visa card, but maybe they'll have a shot. Something else to look for. We do not want any hyperscalers who can roll you over quickly. So by that, I mean, I don't want to invest in a company. No one should invest in a company that like Microsoft or Amazon or Google can just copy you overnight and say, hey, we're gonna get into this. We're gonna give it away as part of our suite for free and fresh you. So like Zoom, stock took off during COVID. It's doing shit now. Why? Because that Microsoft gives away teams. Most enterprises, or a lot of them, they buy the office suite. So why are they gonna pay for Zoom? When teams calls are pretty good, like in my old company, we just, we had Zoom, we had Slack. We just dropped all of it. And we just went all in on Microsoft because it was way cheaper and pretty good. So we use teams for the Zoom equivalent. We use teams for the Slack equivalent. And we use teams for the Dropbox equivalent. And all those stocks came under tremendous pressure. And they kept growing the revenues, but the stocks went nowhere or down because nobody wanted to touch it. Because people know Microsoft is lurking there or has already started to chip in on their thing, their business and it's gonna have a big impact. So we wanna stay away from something that's super easy for those guys to roll and do. Now you can always like make the argument that they can get into anything, which is true. There was a port last week that Amazon wants to look at launching a phone plan in the US. But like we'll just deal in the realm of reality and what was logical for them to get into. It was logical because Microsoft had teams for years and they would eventually be doing this with teams. It was already around, it was already installed and everyone's a Windows computer. Next, scale could be a barrier to entry. It generally works against startups. Hyperscalers do work out, seems like some of these companies get bought. Yeah, so that'll be my last thing. So I get to that one's there, him. So scale could be a barrier to entry, but works against most startups not for them. So some companies like Amazon and think when they started, they didn't really have any barriers, right? Like they're just selling books on the internet that's what they started off with. Same with Walmart when they started. They were just telling a lot of different items. They knew that to actually create a mo and prevent me from opening up a bookstore online and competing against them was to get big fast. So they focused on getting as big as possible. They have as much scale as possible and then they actually created a mo purely out of that scale. Like if I wanted to compete against Amazon, how can I? They have like 90 planes, they own, they have like hundreds of DCs around the world. It just became extremely expensive. There's a reason they beat Barnes and Nobles starting from nothing when Barnes and Nobles was the biggest bookseller in the US is that they focused on getting big online quickly instead of just everywhere. They just focused online and they built DCs specifically to be able to ship direct to consumer, their fulfillment centers. Barnes and Nobles had a lot of DCs, but it was to ship huge pallets of books to book stores and then sell them to individuals from that process. They were really bad because it was a different business at actually shipping to individuals. And by the time they realized that, Amazon had already built a big mo by these barriers to entry. And then finally to kind of sum up that whole segment, one of the core questions that like good investors ask themselves all the time with thinking, is this a good or bad business? Could a well-funded competitor starting up today easily duplicate this business or is it cheaper to buy the startup? So that ties into what David was saying. Like David, you were saying like, I was saying some of the hyper scalars you don't want to compete against them. If you create a really different product then you get a big quick enough, sometimes it's cheaper for the hyper scalar to buy you. So let's look at like Figma and Adobe, right? Like Adobe bought Figma for like a crazy number and it's because of what I was saying earlier. Like Figma did such a good job and got there quickly. And people were being trained on it that it was at the risk of starting to cut into Adobe's market. So you know what, I'm not even sure if that deal will be approved by regulators. It might do it should not because it's clearly anti-competitive. It's Adobe which is like a monopoly business and has a lock on their customers. So a real threat for the first time in years and they're like, we're just gonna overpaying by this to kill this threat and own it internally. Because it would be too difficult for them and too hard to get everyone to switch over. Everyone had to inspect someone's time on the Figma ecosystem and learning how to they do it. And there's a lot of cases where it's just gonna be cheaper to buy the business outright than it is to compete against them and hire people. Like I said, this is a good or bad business. I gave the example earlier like MasterCard could a well-funded competitor start up today easily duplicate this business, like hell no, you can spend $10 billion and not have anywhere near the distribution of MasterCard and it's accepted everywhere. So that's a phenomenal business, right? Cause it's almost impossible to replicate today. Any questions on that? Yeah, I have a question. So how would you go about like, think Figma as like the example, right? Like if you were starting to develop Figma, you might say, okay, well Adobe is just gonna eat my lunch, right, like right away. But then obviously they didn't say that and they were able to get it to the point where they were repetitive to the point where Adobe wanted to buy them. So it seems like there needs to be some kind of like distinction in the analysis to figure out whether or not you can actually reach that threshold from wondering completely. It comes down to asking the founders what their plan is. They should be aware of it. And I say the founders for last, it's probably the most important point. Investing startups versus public companies. But yeah, the first thing that would have asked Figma is how are you gonna differentiate yourself for Adobe and how are you gonna make sure they don't just roll you over and they better have a good answer or I don't want to invest. If they haven't thought about that, like forget it. So there's. So sorry, like what would their answer possibly have been that like what do you think in their view, their answer would be that that question. I think you said we're gonna focus on a niche that it doesn't make sense for Adobe to get, go for and then roll it from there. So we're gonna get people on the ecosystem and learning how to use it because we're gonna focus on something really nichey that Adobe doesn't have anywhere on their roadmap. And they're gonna use that as a beach head to roll out to other products. Like anytime you're going up a well-established data company that's kind of the only way, right? Then you start super specialized and branch out after you have existing clients. You start adding more modules, more features and you start encroaching on Adobe's core business. And then Adobe sees it as a threat and they're like oh shit, we're just gonna buy these guys. But yeah, Adobe probably could have done that, but they probably would focus on like 100 different things. And any decent founder, if you're gonna invest in them, they better have an answer for that and it better be something smart. Cool. So MasterCard, they kind of invented the market, Sean, right? Like is that my visa? So like they kind of invented the card networks, right? It's like a MasterCard visa and American Express really, but American Express isn't as widely covered. The MasterCard is huge around the world. Like you can invent a space, which is what MasterCard and Visa basically did. They also had tons of travelers checks so they're already global to start with. They had a big travelers checks business. They had relationships with banks around the world from that. And from that they transitioned it to this great part system which is annoying her merchants useful for consumers, but insanely profitable monopolies. Apple, they started with no developers users. Branding is huge, right? Like Apple was on the verge of death until Steve Jobs came back. I think it was like 97. And Apple came out with those colorful Macs, which kind of like got them off the death bed and making money. And he kind of used that and crazy, crazy branding to bring her back from death. And that actually brings me to my last point, which is like founders or with Apple and CEO with Steve Jobs as the founders. So it's important for large companies. It's even more important for startups, right? Like it's number one, two, three. But the founders must be religious about their products and believe they're gonna change the world against all odds. Because you have to be a little bit crazy to be a founder, you have to be a little bit arrogant. Most of them are gonna be crazy arrogant and wrong but the ones who are crazy arrogant and right, like their good founders. So for founders, if you just have money in the bank, it's not enough to build a successful company. If you just have good tech, it's not enough to build a successful company. Like how many times have we talked internally, there's something like really smart academic who created something really often, had no idea how to sell it, had no idea how to raise the funds, had no idea how to motivate people to come work for him. And he just died on the vine and then like a year later, someone who has those skills basically just took that idea and ran with it. They did a great job. And it was massively successful. So founders, what I'm looking for is founders who are motivated to build something. They care about money, but it's not all about money. They would be doing this for free because they believe in it. They're not looking for a quick score, they're looking to build something. Like it's a bit of a cliche, but they're looking to change the world in some way or at least have their product to have a huge impact on the world and they truly believe that. Founders need to be persuasive. They're gonna be asking other people to make like sacrifices, to make the founders dream come true, which is bringing this company to life. They're gonna ask for long hours. They're gonna ask people to give in and just work on their vision, not the employees vision, so they need to be persuasive. They're also gonna need to be able to convince investors that this company is gonna work in the future and there's their funding. So that's what I mean. They need to be charismatic and they need to be able to tell the story well or it won't grow and it'll die on the vine. They also need to understand who their customer is and what problem they're helping to solve. So that ties into the answer, the question and the conversation before, right? A founder has to be able to articulate that. What exactly is the problem I'm solving and who am I gonna help and what's the approach to do that? I don't expect founders to know every single point that I just mentioned or in this presentation today, but they should have a general understanding of all these things. Like if we talked about, oh, you are gonna compete against Adobe. What's your plan on that? Like that's question one, if I'm sitting down with the big founders, they'd better have a good answer. And let's just talk about some of the super famous CEOs and the huge companies. Just so happens, these guys are all assholes. You do not need to be an asshole, do well, but you do need to character as excited as it explains. But like Zuckerberg, he's like all these guys are zealous. Like Zuckerberg, Steve Jobs, Elon Musk, Bill Gates, Jeff Bezos, Travis Klamik, AppLike Uber. All these guys have all those characteristics and they're not bigger than their companies, but I promised you at the same time, they would build in their companies. There was like 10 of a guys in the world who probably had the exact same idea as all of them. I mean, a lot of them weren't even first. Like Facebook obviously was like, I don't know, maybe the 20th social network out there. And yet it dominated because these guys were killers and they knew all this stuff, they could tell the story, they understood what their consumer was. They had the tech and the money, but they needed all those other characteristics to actually make it work versus everyone else who failed. So that's it for my thing here. So I'm happy to have a more general discussion if anyone has questions or wants to talk like specific companies or what they're thinking. And they were lucky, you need luck too. Yeah, that's basically life. Better be lucky than good. Yeah, you need all those things, but like you control what you can't control. So look, so when you go out there after today, Max and I are gonna be trying to find companies. We're gonna take a lot of meetings. A lot of them are gonna be bad. We're gonna pass on them, like that's just a numbers game. We're pretty much happy to sit down with anyone with like a happy, it's an idea of. Hopefully after this you have like a bit of a sense of what we're looking for and what could work. And now you have more of a critical eye. So when you see a startup or talk to a founder and he's saying these things in your head like, man, this isn't gonna work because of, you know, there's no tab or there's, you know, like Amazon's gonna roll these cuts over in like two days or whatever, you know, or the man, this is really interesting because not only they're not doing it and no one else is doing this, but like they're going after a big market. It makes perfect sense. They're solving a huge pain point and like this guy can tell a story for this girl, tell a story like that's what we want. So yeah, that's it guys. Hopefully that was like kind of different and interesting. Thanks Jordan. Cool, all right, I'll see you guys later. \ No newline at end of file diff --git a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_summary.txt b/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_summary.txt deleted file mode 100644 index d51af41b..00000000 --- a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_summary.txt +++ /dev/null @@ -1,3 +0,0 @@ -Summary of: 42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_transcript.txt - -If you had perfect knowledge, and you need like one more piece of advertising, drove like 0.2 customers in each customer generates, like let's say you wanted to completely maximize, you'd make it say your contribution margin, on incremental sales, is just over what you're spending on ad revenue. Like if you're, I don't know, well, let's see, I got like you don't really want to advertise a ton in the huge and everywhere, and then getting to ubiquitous, because you grab it, damage your brands, but just like an economic textbook theory, and be like, it'd be that basic math. And the table's like exactly, we're going to be really cautious to like be able to move in a year if we need to, but Google's goal is going to be giving away foundational models, lock everyone in, make them use Google Cloud, make them use Google Tools, and it's going to be very hard to switch off. Like if you were starting to develop Figma, you might say, okay, well Adobe is just gonna eat my lunch, right, like right away. So when you see a startup or talk to a founder and he's saying these things in your head like, man, this isn't gonna work because of, you know, there's no tab or there's, you know, like Amazon's gonna roll these cuts over in like two days or whatever, you know, or the man, this is really interesting because not only they're not doing it and no one else is doing this, but like they're going after a big market. \ No newline at end of file diff --git a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_transcript.txt b/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_transcript.txt deleted file mode 100644 index 8269c2cd..00000000 --- a/server/reflector-local/42min-StartupsTechTalk/42min-StartupsTechTalk.mp4_transcript.txt +++ /dev/null @@ -1 +0,0 @@ - because Max is really aware of that. So basically we're pretty close and pretty finalized with that partners to launch funds. I say funds, but technically this structure will be a corporation. And the difference is if you do a funds, there's very strict rules and regulations and a lot of compliance work with financial authorities. And I've done that in my past job and I really don't wanna do that because I know how intensive it could be and how much of a time drag. So what we're gonna do is incorporate a holding company and call it an investment company. And it's gonna sit on a pool of cash that we raise for people and we'll just keep it in the bank until we find investments that we think are good or suitable startup investments. And we'll take equity stakes in those companies and help them grow. And the funds that we raise will be a mix of my connections, our connections. And end partners is gonna put us in front of a lot of different investors that they know, which is why I was working on the pitch deck and there's more to come. So I was planning on doing this talk a bit later, but we spoke to the head of that partners this week. He would prefer us to have more of a pipeline when we go speak to investors, meaning companies that were close to pulling the trigger on an investing. And your reality is a pretty long process to get to know a company and go through all the details and do all the research. But I'd like to get started at least meeting founders of these super early stage companies because that's what we're focused on. And the thing is, given we have 36, 37 employees all around the world, it makes way more sense. It can be helpful. If I give everyone on the team a bit of a grounding and just what you should be looking for, what some of the key characteristics are of a company that could scale well and become huge one day. Because you know, we're not looking to invest in the dry plingers down the street. Maybe a fine business, but it's not a business that you really take to become a large enterprise and make a ton of money, which is what we're focused on. So if I kind of train everyone or at least explain these concepts and you guys all have your own networks, you're all different parts of the world. You have friends and your friends, you're gonna hear stuff. And I'd like you to keep your ears open and your eyes open and when you come across interesting, even interesting companies if you don't know, if you just find the cool local tech company, you can send it my way. I can always reach out to the founder. Most people are always very, very happy to speak to investors because pretty much everybody in the startup world needs my. So that's the point today. Just before I jump into this document here, does anyone have any questions on that in the funds? I think we had one before, right? So Jan? I didn't understand. Okay, anyway, which part? I missed it. Okay, anyway, continue. Okay, so just in terms of what we're looking for, Sujan, I think you're asking about a return. I mean, realistically with startups, so many of them go bankrupt. Like you invest in the intentions. It's just what it's, you know, it's the nature of the game. So let's say we make 10 investments, I would expect maybe two of them or three of them could be aquahires where you get your money back. Maybe you break even maybe five or six are complete zeros which happens for a very small return, or you know, some sort of recovery, but a loss. And then maybe one or two to be home runs like 10 to 10 next. Right, and those pay for all of the mistakes. And that's really the purpose of do and venture investing or angel, which is even you, right? What we're doing. All right, so all you're in through, so first is Tom. I'm sure you guys have heard me talk about this before or what it stands for is total addressable market. So what that means is if you were a business and you captured 100% of the market, well, that looked like in sales. Let's say we were, I don't know, doing cloud computing and you know, obviously the biggest companies now are KWS and Azure and they have big market share. But let's say one company captured all of it and it was a trillion dollars in revenue a year. It's less than that of just using that example. So that would be the time. So when you were investing and I said, you don't really wanna look at like the drag cleaners down the street, it's cause we want a company that's starting small, but it's really going after a big market. There's a lot of really cool companies out there that solve a really niche problem and that's great. They could be more of a pet project or like a mom and pop store, but as an investor we don't really wanna touch that because the ability to get that 10, 20, 100 X is pretty diminished. If they're not solving a problem in a big addressable market, there's not a lot of potential upside. There's some exceptions like medium sized markets in work. So for an example of a medium sized total addressable market that worked pretty well as Etsy. I don't know, does anyone know Etsy? They do like local kind of arts and crafts are very customized and you deal with people online. So Etsy is not a huge market and one of the reasons a stock did do great at the start is people thought it was so nichey and so specialized that maybe Amazon would just either crush them which they didn't or the market wasn't big enough. Etsy proved that wrong and it wasn't huge but it was medium sized and Etsy stock has been extremely, extremely well. The market was bigger than people thought. Etsy captured not 100% but a very large percent of the total market and that's partly because bigger competitors like Amazon mostly ignored it because they didn't really see the potential. So there's other exceptions to the role where medium sized ones work like Etsy but in general we're gonna wanna go for big ones. Yeah, my wife buys it, kinda stuff off Etsy. Anyone have any questions on TAM or companies that you can think of? And they're curious if it's a big enough TAM or not. One question, Jordan, when you're talking about market, how do you define a market? Meaning that as we're located in different countries, how we can tell, this is going to be big here in my country, in my neighborhood, well, not in my neighborhood, but you know what I mean? Yeah, so depends on the company's plan but generally TAM would be the total market that you can reasonably address. So for Amazon, it's global online commerce, right? Like they touch everything. So any sort of retail online, that's Amazon's TAM. Cloud competing in any country, Amazon operates in all of them. So if you're a local company and you have zero plans, let's say you're in Canada and you have zero plans to go to the US, you can't really count the US but if it's been part of your long-term plan, is to go to the US and there's no roadblocks. Like let's say you're selling food, it's very hard to import food over borders, people do it, but it's harder. The US would not be your TAM. If you're a tech company, there's zero limitations, like Shopify, clearly Shopify is a Canadian company but their TAM is global because they sell around the world. So it depends on the type of company, but yeah, most companies we're going to be looking at are going to have global TAMs. So it's going to be worldwide or at least most of the developed world, which is the big chunk of the worldwide economy. Okay, so the next concept is product market fit. I wrote, being in a good market with a product that can satisfy the market. There's a few things here, so there's a lot of cool companies but maybe no one's actually going to pay for it. So you don't really find out until you get into the market and you start selling. Unfortunately, most of what we're going to be investing in is before that, they haven't actually launched the product. And of course, internally, we deal with a whole lot of companies like Virtupoker where we have a good idea, you know, what it's going to be like and we'll fix it. But we don't really know until we go into the market. One of the things we guess right, one of the things we guess wrong and how can we adjust it. So there's a few things to think about and we're going to be investing before companies are really selling. The good thing is you get them cheaper. Like if you invested in a good startup at five or six million bucks, it's probably because they haven't gone to market yet then all the sales. Like once they start selling and it's clear click, like most people consider that already ready for Series A, which is a further venture round. And a lot of the valuations could be like 20, 30 million. Coorsnet online, let's talk about like that BRD project he's doing. So that solves a real problem, right? It's both like breast rotors disease. It's the number one health impact that affects farmers per cows, above meat and milk. And I calculated the damage. The damage of the like the cows die. So total loss or after they recover from the disease, they don't gain as much weight. So you can't sell them for as much because it's meat times price. So it's pretty easy to calculate the total economic damage or harm. So if we stick in with BRD, if we solve that problem, meaning we recognize it earlier and prevent the farmer from having those losses, we can charge for part of the amount we're saving. The conversation I was having with them is why don't we charge about 20 by a percent? If the farm is going to have about $10,000 a year and economic damage, we can charge a quarter of that, $2,500, that's a real business. And it's a real solution, something wants to buy. We won't know how it works till we get in there and that's a product market fittest, but we can try. So the thing to focus on when looking at a company is it's saving the end buyer about time, money or pain. So for that example, let's use Uber. Uber solved the problem of getting a taxi was extremely painful. It was an old system. You had to call them on your phone and say, please come at 630. You couldn't see how close they were. So that saves a lot of, that saves people time and pain. Or it gives people, it gives people something like revenue. So if you're selling a business, some new product that gives them revenue, they're going to buy that. So for that example, think about booking.com. See if everyone knows booking. It's like the biggest travel website in the world. And the customer there on that side is the hotels. The hotels are all tied in. And the hotels know, oh, booking is aggregated. So many customers like you and Ray, you want to travel and gunpowder tells, if I'm a hotel in Milan, I better tie into booking.com. Because everyone's going to be searching for hotels in Milan. They're going to go to bookings. So if I tie an even a pack that pay booking 20% of the cost of the hotel might, or 15 to 20, which is what they charge, they're going to do it. Because they're going to bring a ton of revenue. And that's why bookings, a huge, huge business. And finally, could bring the customer something like enjoyment. So think about Netflix, right? It brings them joy. It brings them entertainment and they value that. Same with Nintendo, right? It doesn't save time or money, obviously. We're generating revenue, but it brings people entertainment that they're willing to pay for. So when we're evaluating companies, think of those three buckets, and really focus on is this company providing one of those in a way that people are going to want to pay for, does that make sense? Anyway, any questions there or thoughts about other companies you come across that do that well? OK, I'll keep moving along. I'll stop asking for questions and just jump in or raise your hand if you have them. OK, the next one is unit economics. So what does that mean? It's looking at the total profit for selling the product or service or whatever it is, minus the all in cost ability to, and you want it to be attractive. You don't want to like sell something and it costs 99 cents to deliver it. You sell per bucket cost 99 cents in your profits or 1% that's awful. So the way to measure this is revenue minus direct costs, and that's the unit economic itself, because unit economics mean building that one thing without all the overhead. So let's ignore, let's look at an article, for example. What are the unit economics? It's what we build out our developers at minus the developer's salary. That's the unit economics. Every company has certain amounts of overhead that aren't direct, but you still need to build off of. So let's say anyone on the upside, anyone in my role who's not writing code, not billed by the week or month, that would be more on the fixed cost side. Anyone on building code is kind of like in business we call it like a right book cost, and this is labor costs. So what we're looking for is companies that have really good unit economics, because that really allows them to scale and make a ton of money. And the next step of that is incremental margin. Let's say Facebook. Facebook has great unit economics, right? Like they serve ads, they sell ads. What's the cost of delivering that ad? If they get a dollar an ad revenue, it's like, I don't know, some basic server costs. Maybe it's like three cents per dollar of ad revenue. So it's huge. And what's the incremental cost? They're ready fully staffed, in fact, if I'm a people. So for every dollar of revenue they bring in, they'll get like 97 cents of gross profit, but they don't really need to add that money more operations people or that many more tech people and R&D. So their incremental contribution margin is huge. Like at the start it might be zero, because even if they're making 97 cents per dollar of incremental revenue, they still have to add operations people, they have to add tech people, they have to add sales people, and all that cost would eat up that 97 cents. But once you get to a certain level, it's completely incremental, and it works really well. Some other businesses with really high incremental contribution margin, MasterCard and Visa. Like they have some of the highest profit margins in the world, why? Because they're ready to set up, right? Like there's really just swipe the credit card, cost almost nothing. They get a fee every single time to do it. They're ready fully staffed. So you know, a hundred more businesses turn on tomorrow, say, hey, I'm gonna take MasterCard, it's free revenue for MasterCard. There's almost no cost associated with that. So really those are really good businesses. And I want you considering what the unit economics look like. Because you don't want to invest in something like, I don't know, some product that's just very low margin, it has no chance to get the high margin ever. So okay, I have low margins today, but you have to have high incremental unit economics to get to high margins eventually. The next one ties into that. So that's lifetime value and customer acquisition costs. So it's kind of measuring how attractive is it? How worth, how much is it worth to be spending money on marketing and sales to bring in a customer? So there's the big thing that's kind of math. The first is lifetime value. Lifetime value means how much is a customer gonna spend on your business over his whole customer existence? So obviously for customers gonna stick around for one year, it's not as valuable as a company that's gonna stick around or clients gonna stick around for 10 years. So it's how much they spend on average for transaction, times, how many purchases they'll make over their lifetime. So for a company with like a huge lifetime value, for customers think a Costco. You come in, you always spend like 500 bucks, it's more than you expected. Maybe you go like quite some month, you have a family. Nobody leaves Costco, they sign up for their membership like a hundred bucks a year and they're spending like a thousand bucks a month at Costco for like a decade. So the spending 12,000 a year times 10 years that's 120,000 that's huge. So even though Costco has like thin margins because people spend so much in the relief. Yeah, Hannah, you got kids, you need Costco, you teenagers. That's clear. We're just starting our Costco journey. These my kids are young, but yeah, my wife's there all the time now. But you're gonna keep going for a long time, maybe till your kids move out of the house and then it doesn't make sense together anymore. And then the other measurement here that's important is customer acquisition costs. I'm sorry, just the key state digital, like them value same thing for Amazon, right? Like once they acquire customer or the customer it tends to order, order, order. The lifetime value is huge and it's still growing because Amazon sees very little customers actually just outright quit. And since they, like there's been customers spying on Amazon since 1998 and they're still buy it, so that value just keeps going and going. Customer acquisition costs measures the cost of society and the first time so the way to calculate that is look at the cost of sales so that can be like advertising and marketing people plus the cost, yeah, so the cost of advertising, marketing people building up your brands, all of that. For Stock2Ware, it's generally those that go costs and what you measure is okay. We added 1,000 new customers this quarter. We spent a million dollars on marketing. So clearly the cost for new addition was 1,000. Let's say the average customer is gonna spend $4,000 over the lifetime, just spending $1,000 to gain a customer. They're gonna spend $4,000 over their lifetime, which is a pretty good ratio. That's 401. Anything over three is good. And for Stock2Ware, if we stick with the like, let's call it 80% Chris margins, because it's, you know, Stock2Ware, there's not a lot of cost to roll out one more customer. The $4,000 might translate to 3,000, the unit contribution, profit contribution. So you spend the thousands, get 4,000, sales and 3,000 profit, that's pretty good. That's why it's such an important measure, right? Like you created so much value. In that example, every customer you add just created $2,000 of value. So if you're an investor on Wall Street or an investor in early stage companies like us, you really wanna see that. And a lot of our companies won't have that yet, but it's still an important concept to understand, because as you scale and do other rounds, I promise that you're capital, that you're capitalists look at that very, very well. Something else to get your cash low, your customer acquisition costs lower, is you want your customers to vagalize it. Like the cheapest way to grow is a word of mouth, right? Like Amazon didn't spend any money on paid advertising for years and years and years. Like Jeff Bezos would just say, grand theost things, go to conferences, and be featured in barons. And like all these people who didn't know about Amazon in 97, like all of a sudden, they're getting all over the front page of Wall Street Journal, not paying for it. People are like, oh, I've gotta check out this new thing called Amazon. Actually, I was listening recently, Bezos went public early with Amazon when they were pretty small, because he thought it'd be great publicity and free press. So if you get that free press, that's the last money in advertising, that ratio goes up because your customer acquisition cost goes down. Think about GitHub, right? Like a lot of new engineers, they've got to really spend money to get them to start using GitHub. No, they just come to GitHub, because everyone's like, oh yeah, you gotta get on GitHub. I stored my code here, you gotta come check it out and start coding here too, and posting it all. So their customer acquisition cost is very low, which is why it's such a good business. So again, the rule of thumb, a thumb is three or higher, and under that, you don't really want it. I'll just wait a sec, any questions? Yeah, that's a fair point on Dress. Tesla had like way more demands than supply, so that any depend anything on ads. Now they supply cut up and the man's flattening a bit, so they actually have to start spending some money. Now you're saying, they're still spending very little. Yeah, you're right. Eventually they'll have to spend more though, but you're right. Okay, so actually have a question now. How much, like a normal company like Ezra, how much would you spend on advertising? Like how do you calculate how much do you spend on advertising? Well, it's an equation, right? Like a CAQ to LTV, and it kind of salt math at the end of the day. If you had perfect knowledge, and you need like one more piece of advertising, drove like 0.2 customers in each customer generates, like let's say you wanted to completely maximize, you'd make it say your contribution margin, on incremental sales, is just over what you're spending on ad revenue. Because that's just the math equation. Does that make sense? Like if you spend the dollar on ads, and it contributed $2,000 in gross profit, cool. You know, that's working. And without having to invest anymore in infrastructure, reality is a lot more complicated, right? Like if you're, I don't know, well, let's see, I got like you don't really want to advertise a ton in the huge and everywhere, and then getting to ubiquitous, because you grab it, damage your brands, but just like an economic textbook theory, and be like, it'd be that basic math. Okay, churn. So we all know what churn is, the churn in, churn out, canceling Netflix, whatever. So churn fits into lifetime value, right? Because lifetime value measures how long customers last. If a company has a lot of churn, and the customers don't last very long, and their lifetime value is low. On the inverse, that super sticky product, and everyone loves it, nobody leaves, or they just can't leave, because they're trapped, which is great from the business side. You're gonna have very high lifetime value. So I'll just call it a few businesses at a high churn and low churn, and it'll be kind of intuitive to you. So one is like selling through a publisher revenue every year is hard. If you have customers that are on repeat and growing with you, like some of the best tech companies, life is easy, right? Like you do nothing, 98% of your customers stick with you for next year, 2% leave, and then 98% of customers who stick with you probably group, because they're consuming more. So let's think about data dog snowflake. A lot of those companies, they're words still are some extent, like Wall Street darlings. Why? Because every year, their customers take more and more services from them. So data dog doesn't have to do anything. They literally could fire their, pretty much their whole sales force, and if their revenues are 100 bucks this year, they would drop to 98% because two percent of clients leave, and then go to 113, because their remainder consumed 15% more. So that's phenomenal, because it just such easy growth. Let's talk about something that are bad. So like meal kits, they lose like half their clients every single year, they turn out, let's look at Palatine. Now that COVID's done, a lot of people come into Palatine, they love it, they use it for a couple of years, and then they cancel. Maybe they're spending money, and they quit working out. I mean, that's like a super standard gym model too, in the real world, right? You sign up for the gym, you use it for a year or two, maybe you forget about it, and you're still paying and not using it. Eventually wake up, see your credit card bill, you're like, this is stupid, and you cancel it. So it's a very high-turn business. The good thing for what we're looking at is we're mostly going to be looking at tech companies and not retail tech, because retail tech does that by turn. But enterprise software tech tends to have some of the lowest turn around, and that's why it's some of the best businesses around. They get super sticky, and they're very hard to leave. Those are phenomenal businesses, and when you're looking at them, try to find businesses that have those characteristics. When you know people are going to be on it, they're going to stick with it, and not leave. And that kind of brings us to the next point, which is we're looking for businesses that have really high barriers to entry, to make sure copycats can't just come in, and mimic you and kill your business once you're established. So, you know, we're just talking about low-turn, so one of the things that makes it really good mo, meaning someone else can't just come in and duplicate you overnight, and they can't kill you overnight, is high-spotting costs, like you get locked in. There's a client that's really annoying to be locked in, but as a business, it's phenomenal. So, think about like Google Cloud, right? Like you move everything on cloud, we were speaking to Dochibo this week at the conference, and they're probably going to use Google for AI. Google's basically going to be giving in, they told us a bunch of fine-cuned foundational models off the Palm 2 for free, and see Janet and Shree have questions on that. And we're like, oh, you know, I asked Max and he were right, and like, why is Google just going to give you this stuff that I'm going to charge you? And it's like, Max, they want to lock everybody in. And the table's like exactly, we're going to be really cautious to like be able to move in a year if we need to, but Google's goal is going to be giving away foundational models, lock everyone in, make them use Google Cloud, make them use Google Tools, and it's going to be very hard to switch off. Any questions on that before I move on? Okay, the other things you want to look for is like make the product addictive, especially if it's in like the entertainment space, your video games, you wanted to addictive as help for the client, the customer, so they never leave in the key plane. I think that's evident. You want really steep learning curves sometimes for a product that's taken off, and that's a form of switching costs, right? Like if you think of how long and hard it is to really get your employees up to speed on something, they're not going to use something else. Like people, like designers, maybe just like, produce owners off Figma. They learn Figma, they're like Figma experts, they're not going to leave. Like that's why Adobe had to go and buy Figma. Figma, like that's Adobe's game plan, right? And Adobe was losing the market chair because Figma was so good, and all these people are like being trained on Figma. The best is when you see a company, and universities are offering classes and how to learn this, just like that's phenomenal. The next generation is just getting indoctrinated and trained how to do this. The universities are building up your software companies' value for free. So switching costs, yeah, Adobe for sure. So switching costs are high if you spend a lot of time trading someone internally, and it's hard to get people to use your product, but once they do, it makes it really sticky. So you kind of want them to become local experts on your thing, and it's just like a way you can make your business extra sticky. Okay, another one is two-sided liquidity. This is big. Basically, this is more for marketplaces, but like think about any business where there's two sides. We talked about booking before, right? You need to get all the consumers, and you're gonna use the hotels, and then you need, which is demand, and you need all the hotels themselves, which is supply. And if you have all this supply, but no consumer demands, the hotels are gonna live. If you have all the consumer demands, but they can't book anything on the site, they're gonna leave. So it's a bit of a chicken and egg, which is why it's very hard to replicate, and knock off the companies that are doing this. But if you do it well, like booking did, you scale both to a nice level, or if you do it well, like Uber did, you're always balancing to make sure you've been of drivers and enough passenger demands. Your drivers don't leave, and your customers keep staying. And you could grow, and it's really, really hard. It's the same with credit cards, right? Like they have two-sided liquidity. If I had, and this brings up my next point, if I had, or my last point, a billion dollars to go start a new credit card company, and I could just blow the billion to try to build it. Could I do it? Could I go to Merchant and be like, hey, you want to pay anything to take my credit card? Just gonna quick integration, and you're done. Okay, really, versus 2% for Visa. Are they really gonna offer it? If I'm like Jordan Card? No. If you have a huge brand in your global, you can start with something like Apple's doing, but Apple's the biggest company in the world. But generally, liquidity, two-sided, liquidity marketplaces, like credit card systems, are extremely hard to knock off. And that's why there's some of the best businesses where Visa and MasterCard have 60% margins. Other things that offer a nice mode that can really protect you is patents, obviously, like anything that's really patentable, hard to knock off, but not impossible. Patents aren't the best mode. Like I prefer two-sided liquidity systems to patents any day. Same with any type of intellectual property. It's also not as good as two-sided liquidity and some of the other things. And then branding, branding's huge too, right? We talked about Apple. Like there's nothing that overly unique. But as Andreas and I were talking about the start of the meeting, Apple did a great job building up two-sided liquidity on the iPhone when it was released, right? Like all the apps, you have to have the developers come and build on your platform, so that's supply. And you have to have tons of consumers have an iPhone, which is demand. And then if you're launching a new one, the developers are like, I'm not gonna build that new OS because there's no one using it. Wow, and I waste my time. I'm gonna stick with iPhone. I'm gonna stick with iOS. It's like way better. And that's the benefit of two-sided liquidity. And that's the benefit of Apple branding. And why maybe they'll have an impact in finance and ultimately have more Apple cards, not to take on master or visa card, but maybe they'll have a shot. Something else to look for. We do not want any hyperscalers who can roll you over quickly. So by that, I mean, I don't want to invest in a company. No one should invest in a company that like Microsoft or Amazon or Google can just copy you overnight and say, hey, we're gonna get into this. We're gonna give it away as part of our suite for free and fresh you. So like Zoom, stock took off during COVID. It's doing shit now. Why? Because that Microsoft gives away teams. Most enterprises, or a lot of them, they buy the office suite. So why are they gonna pay for Zoom? When teams calls are pretty good, like in my old company, we just, we had Zoom, we had Slack. We just dropped all of it. And we just went all in on Microsoft because it was way cheaper and pretty good. So we use teams for the Zoom equivalent. We use teams for the Slack equivalent. And we use teams for the Dropbox equivalent. And all those stocks came under tremendous pressure. And they kept growing the revenues, but the stocks went nowhere or down because nobody wanted to touch it. Because people know Microsoft is lurking there or has already started to chip in on their thing, their business and it's gonna have a big impact. So we wanna stay away from something that's super easy for those guys to roll and do. Now you can always like make the argument that they can get into anything, which is true. There was a port last week that Amazon wants to look at launching a phone plan in the US. But like we'll just deal in the realm of reality and what was logical for them to get into. It was logical because Microsoft had teams for years and they would eventually be doing this with teams. It was already around, it was already installed and everyone's a Windows computer. Next, scale could be a barrier to entry. It generally works against startups. Hyperscalers do work out, seems like some of these companies get bought. Yeah, so that'll be my last thing. So I get to that one's there, him. So scale could be a barrier to entry, but works against most startups not for them. So some companies like Amazon and think when they started, they didn't really have any barriers, right? Like they're just selling books on the internet that's what they started off with. Same with Walmart when they started. They were just telling a lot of different items. They knew that to actually create a mo and prevent me from opening up a bookstore online and competing against them was to get big fast. So they focused on getting as big as possible. They have as much scale as possible and then they actually created a mo purely out of that scale. Like if I wanted to compete against Amazon, how can I? They have like 90 planes, they own, they have like hundreds of DCs around the world. It just became extremely expensive. There's a reason they beat Barnes and Nobles starting from nothing when Barnes and Nobles was the biggest bookseller in the US is that they focused on getting big online quickly instead of just everywhere. They just focused online and they built DCs specifically to be able to ship direct to consumer, their fulfillment centers. Barnes and Nobles had a lot of DCs, but it was to ship huge pallets of books to book stores and then sell them to individuals from that process. They were really bad because it was a different business at actually shipping to individuals. And by the time they realized that, Amazon had already built a big mo by these barriers to entry. And then finally to kind of sum up that whole segment, one of the core questions that like good investors ask themselves all the time with thinking, is this a good or bad business? Could a well-funded competitor starting up today easily duplicate this business or is it cheaper to buy the startup? So that ties into what David was saying. Like David, you were saying like, I was saying some of the hyper scalars you don't want to compete against them. If you create a really different product then you get a big quick enough, sometimes it's cheaper for the hyper scalar to buy you. So let's look at like Figma and Adobe, right? Like Adobe bought Figma for like a crazy number and it's because of what I was saying earlier. Like Figma did such a good job and got there quickly. And people were being trained on it that it was at the risk of starting to cut into Adobe's market. So you know what, I'm not even sure if that deal will be approved by regulators. It might do it should not because it's clearly anti-competitive. It's Adobe which is like a monopoly business and has a lock on their customers. So a real threat for the first time in years and they're like, we're just gonna overpaying by this to kill this threat and own it internally. Because it would be too difficult for them and too hard to get everyone to switch over. Everyone had to inspect someone's time on the Figma ecosystem and learning how to they do it. And there's a lot of cases where it's just gonna be cheaper to buy the business outright than it is to compete against them and hire people. Like I said, this is a good or bad business. I gave the example earlier like MasterCard could a well-funded competitor start up today easily duplicate this business, like hell no, you can spend $10 billion and not have anywhere near the distribution of MasterCard and it's accepted everywhere. So that's a phenomenal business, right? Cause it's almost impossible to replicate today. Any questions on that? Yeah, I have a question. So how would you go about like, think Figma as like the example, right? Like if you were starting to develop Figma, you might say, okay, well Adobe is just gonna eat my lunch, right, like right away. But then obviously they didn't say that and they were able to get it to the point where they were repetitive to the point where Adobe wanted to buy them. So it seems like there needs to be some kind of like distinction in the analysis to figure out whether or not you can actually reach that threshold from wondering completely. It comes down to asking the founders what their plan is. They should be aware of it. And I say the founders for last, it's probably the most important point. Investing startups versus public companies. But yeah, the first thing that would have asked Figma is how are you gonna differentiate yourself for Adobe and how are you gonna make sure they don't just roll you over and they better have a good answer or I don't want to invest. If they haven't thought about that, like forget it. So there's. So sorry, like what would their answer possibly have been that like what do you think in their view, their answer would be that that question. I think you said we're gonna focus on a niche that it doesn't make sense for Adobe to get, go for and then roll it from there. So we're gonna get people on the ecosystem and learning how to use it because we're gonna focus on something really nichey that Adobe doesn't have anywhere on their roadmap. And they're gonna use that as a beach head to roll out to other products. Like anytime you're going up a well-established data company that's kind of the only way, right? Then you start super specialized and branch out after you have existing clients. You start adding more modules, more features and you start encroaching on Adobe's core business. And then Adobe sees it as a threat and they're like oh shit, we're just gonna buy these guys. But yeah, Adobe probably could have done that, but they probably would focus on like 100 different things. And any decent founder, if you're gonna invest in them, they better have an answer for that and it better be something smart. Cool. So MasterCard, they kind of invented the market, Sean, right? Like is that my visa? So like they kind of invented the card networks, right? It's like a MasterCard visa and American Express really, but American Express isn't as widely covered. The MasterCard is huge around the world. Like you can invent a space, which is what MasterCard and Visa basically did. They also had tons of travelers checks so they're already global to start with. They had a big travelers checks business. They had relationships with banks around the world from that. And from that they transitioned it to this great part system which is annoying her merchants useful for consumers, but insanely profitable monopolies. Apple, they started with no developers users. Branding is huge, right? Like Apple was on the verge of death until Steve Jobs came back. I think it was like 97. And Apple came out with those colorful Macs, which kind of like got them off the death bed and making money. And he kind of used that and crazy, crazy branding to bring her back from death. And that actually brings me to my last point, which is like founders or with Apple and CEO with Steve Jobs as the founders. So it's important for large companies. It's even more important for startups, right? Like it's number one, two, three. But the founders must be religious about their products and believe they're gonna change the world against all odds. Because you have to be a little bit crazy to be a founder, you have to be a little bit arrogant. Most of them are gonna be crazy arrogant and wrong but the ones who are crazy arrogant and right, like their good founders. So for founders, if you just have money in the bank, it's not enough to build a successful company. If you just have good tech, it's not enough to build a successful company. Like how many times have we talked internally, there's something like really smart academic who created something really often, had no idea how to sell it, had no idea how to raise the funds, had no idea how to motivate people to come work for him. And he just died on the vine and then like a year later, someone who has those skills basically just took that idea and ran with it. They did a great job. And it was massively successful. So founders, what I'm looking for is founders who are motivated to build something. They care about money, but it's not all about money. They would be doing this for free because they believe in it. They're not looking for a quick score, they're looking to build something. Like it's a bit of a cliche, but they're looking to change the world in some way or at least have their product to have a huge impact on the world and they truly believe that. Founders need to be persuasive. They're gonna be asking other people to make like sacrifices, to make the founders dream come true, which is bringing this company to life. They're gonna ask for long hours. They're gonna ask people to give in and just work on their vision, not the employees vision, so they need to be persuasive. They're also gonna need to be able to convince investors that this company is gonna work in the future and there's their funding. So that's what I mean. They need to be charismatic and they need to be able to tell the story well or it won't grow and it'll die on the vine. They also need to understand who their customer is and what problem they're helping to solve. So that ties into the answer, the question and the conversation before, right? A founder has to be able to articulate that. What exactly is the problem I'm solving and who am I gonna help and what's the approach to do that? I don't expect founders to know every single point that I just mentioned or in this presentation today, but they should have a general understanding of all these things. Like if we talked about, oh, you are gonna compete against Adobe. What's your plan on that? Like that's question one, if I'm sitting down with the big founders, they'd better have a good answer. And let's just talk about some of the super famous CEOs and the huge companies. Just so happens, these guys are all assholes. You do not need to be an asshole, do well, but you do need to character as excited as it explains. But like Zuckerberg, he's like all these guys are zealous. Like Zuckerberg, Steve Jobs, Elon Musk, Bill Gates, Jeff Bezos, Travis Klamik, AppLike Uber. All these guys have all those characteristics and they're not bigger than their companies, but I promised you at the same time, they would build in their companies. There was like 10 of a guys in the world who probably had the exact same idea as all of them. I mean, a lot of them weren't even first. Like Facebook obviously was like, I don't know, maybe the 20th social network out there. And yet it dominated because these guys were killers and they knew all this stuff, they could tell the story, they understood what their consumer was. They had the tech and the money, but they needed all those other characteristics to actually make it work versus everyone else who failed. So that's it for my thing here. So I'm happy to have a more general discussion if anyone has questions or wants to talk like specific companies or what they're thinking. And they were lucky, you need luck too. Yeah, that's basically life. Better be lucky than good. Yeah, you need all those things, but like you control what you can't control. So look, so when you go out there after today, Max and I are gonna be trying to find companies. We're gonna take a lot of meetings. A lot of them are gonna be bad. We're gonna pass on them, like that's just a numbers game. We're pretty much happy to sit down with anyone with like a happy, it's an idea of. Hopefully after this you have like a bit of a sense of what we're looking for and what could work. And now you have more of a critical eye. So when you see a startup or talk to a founder and he's saying these things in your head like, man, this isn't gonna work because of, you know, there's no tab or there's, you know, like Amazon's gonna roll these cuts over in like two days or whatever, you know, or the man, this is really interesting because not only they're not doing it and no one else is doing this, but like they're going after a big market. It makes perfect sense. They're solving a huge pain point and like this guy can tell a story for this girl, tell a story like that's what we want. So yeah, that's it guys. Hopefully that was like kind of different and interesting. Thanks Jordan. Cool, all right, I'll see you guys later. \ No newline at end of file diff --git a/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-AGENDA.txt b/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-AGENDA.txt deleted file mode 100644 index 094a4424..00000000 --- a/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-AGENDA.txt +++ /dev/null @@ -1,4 +0,0 @@ -GitHub -Requirements -Junior Developers -Riding Elephants \ No newline at end of file diff --git a/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-Summary.txt b/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-Summary.txt deleted file mode 100644 index 146b342e..00000000 --- a/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-Summary.txt +++ /dev/null @@ -1,4 +0,0 @@ -Summary of: https://www.youtube.com/watch?v=DzRoYc2UGKI - -Small Developer is a program that creates an entire project for you based on a prompt. It uses the JATGPT API to generate code and files, and it's easy to use. The program can be installed by cloning the GitHub repository and using modalcom. The program can create projects for various languages, including Python and Ruby. You can also create a prompt.md file to input your prompt instead of pasting it into the terminal. The program is useful for creating detailed specs that can be passed on to junior developers. Overall, Small Developer is a helpful tool for quickly generating code and projects. - diff --git a/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-Transcript.txt b/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-Transcript.txt deleted file mode 100644 index c7e26689..00000000 --- a/server/reflector-local/7min-SmolDeveloper/7min-SmolDeveloper-Transcript.txt +++ /dev/null @@ -1 +0,0 @@ - As an engineer, small developer is absolutely amazing and simultaneously terrifying. It is like having a junior engineer in your pocket. This goes beyond putting a prompt into JATGPT and having it help you code. This actually creates an entire project for you. And it's super easy to use. So I'm going to show you a project that I created in two minutes. Then I'm going to show you how to install it. Let's do it. So here's the prompt that I gave it. Write a Python project that takes a JATGPT API key in an M file. And then when the main script runs, ask the user for a prompt and then use the JATGPT 3.5 turbo API endpoint to get a response from JATGPT. Then it displays that response to the user and asks for another prompt. Make sure to include a requirements.txt file. Also make sure to use the open AI module. Here's an example of what the JATGPT API call should look like. And then I gave it an example straight from the JATGPT API docs. So I ran small developer on this prompt and it output an entire project for me. So we have the requirements file. We have the main file and we also have an M file. Now I'm definitely going to rotate this API key before publishing the video. So let's run it. Let's see what happens. Enter your prompt. Tell me a joke. There it is. Why couldn't the bicycle stand up by itself because it was too tired? Amazing. When was Bill Clinton president? Bill Clinton was the 42nd president of the United States and he served two terms from January 20th, 1993 to January 20th, 2001. So again, I created this entire project from a few lines of a prompt and it actually created all the files for me. So let me show you how to install it now. So this is the GitHub page. Small-AI slash developer. It has over 7,000 stars and nearly 500 forks and it's one of the trending GitHub repos right now. Human centric and coherent whole program synthesis, aka your own personal junior developer. So it gives a bunch of information about what it does, but let's actually do it. Now this is so easy to use and they give a few examples of incredible projects built from just a few prompts and you can get extremely detailed in these prompts. You can think of it like writing a spec and then you pass it off to a junior developer to write and chat, GPT writes it for you and creates all the files. So it's amazing. So the first thing we're going to do is come down around two thirds of the way through the page. We're going to grab this line, get clone and then the GitHub repo. We're going to copy that. So I have a new VS Code window open. I'm going to come up to the top right and click for a new terminal. Now once that new terminal opens, I'm going to CD to the desktop, hit enter and now I'm going to paste that line, get clone, GitHub.com, small-AI slash developer. Then I'm going to hit enter and then that's going to clone it to my desktop. From there, I'm going to CD into that folder. So now I'm in the folder. So the next thing I'm going to do is come up to the top left, click the Explorer icon and then click open folder. Then I'm going to open the developer folder. So there it is. That's the entire small developer project. I'm going to open up the terminal again by clicking the toggle panel in top right and while that's going, I'm going to rename the dot example dot nv to just dot nv. Then we're going to click on it and we're going to enter our open AI API key. Now you can use nthropic as well if you want, but I'm just going to stick with open AI. So if you don't have an open AI API key, head on over to open AI and just generate one. Then we're going to save this file. We're going to go down to main.py and this is the main file. So the nice thing about this project is it uses modal.com. I had actually not heard of modal.com, but essentially it's like a container like Docker. And it really takes away all of the complexities of managing module versions, Python versions, which I always stumble on. And according to a lot of the comments in my videos, a lot of you stumble on too. So this is a great solution to that. Now you don't need it, but it really makes it easier. So to use modal, go to modal.com, sign up for a new account. It says here that it's in private beta, but I was able to sign up no problem. So I don't think it's in private beta anymore. Then we're going to copy this right here, pip installed modal dash client. I'm going to switch back to my terminal and I'm going to paste it. So pip installed modal dash client. Enter. I already have it installed. But if you didn't, it would have installed it there. Now if you wanted to run this without modal, all you have to do is pip install dash our requirements dot txt and then run the file Python main no modal dot pi rather than main dot pi. But we're going to stick with using modal. And so the basic usage is right here. So we're going to grab just these first commands. And then I'm going to copy it. And it's modal run main dot pi dash dash prompt, switch back to visual studio code. And I'm going to paste that in. And so let's start with something really basic, write a Ruby script that counts to 100. And then I'm going to hit enter. Now the first time you do this modal is going to ask you to authenticate. And all you have to do is click the link, open up the website, log in and then switch back to terminal. And you're done. So it says it's going to create one file counts to 100 dot RB. And you can see here that it's actually using containers with modal. And it has a pretty nice UI for being strictly through the terminal. And there it is. It actually outputs the file name and what's in the file. But the cool part is it actually generates it for me. So if I come up here to the left and click this generated folder, click the drop down. There's the file. It just created count to 100 dot RB. So dev count to 100 for I and one to 100 puts I. That is correct. Now obviously this was a very simple example. But you can get quite complex. And the nice thing is, you don't actually have to put the entire prompt in the terminal. You can create this thing called prompt dot md and put your entire prompt in there. All right. So I have my little script here, modal run main dot pi dash dash prompt. And then I output the prompt. So I'm going to highlight that whole thing. I'm going to come to the terminal. I'm going to paste it. And then when it's finished, I hit enter. And now it's going to start creating that project. Now you could just create a dot md file and put your prompt in there instead of having to paste the entire prompt directly into terminal. Now while that's going, let me show you an example of what they've done. Now here's a really detailed spec of a chrome extension. Now I don't have access to and the topic. Claud yet. But as soon as I do, I'm going to test this out. But for now, you can see they basically created an entire detailed spec that can be passed to a junior developer. And in this case, the junior developer is chachy PT. And there it's done. App completed. So let's check it out. Now we're going to go to the generated folder. We're going to click it. And we're going to look inside. So we have the project route. We have our dot m file. We have the main dot pi file. And it looks like everything is correct. And we even have the requirements dot txt file. So for here, we need an API key. So I'm going to go grab that. I'm going to double click it. Paste and now I have the API key. And I'm going to save. Then I'm going to go to main dot pi. And I'm going to push play. And let's see if it works. Okay, so it loaded. Tell me a joke. Hit enter. And there it is. Why wouldn't the bicycle stand up by itself because it was too tired. Funny. Gaming the same exact joke as before. And that's it. We've created an entire project just with a prompt. And it creates the entire file structure, all of the code for it. And it's easily done. Give it a try. Let me know what you think. If you liked this video, please consider giving me a like and subscribe. And I'll see you in the next one. \ No newline at end of file diff --git a/server/reflector-local/readme.md b/server/reflector-local/readme.md deleted file mode 100644 index c395423e..00000000 --- a/server/reflector-local/readme.md +++ /dev/null @@ -1,11 +0,0 @@ -# Record on Voice Memos on iPhone - -# Airdrop to MacBook Air - -# Run Reflector on .m4a Recording and Agenda - -python 0-reflector-local.py voicememo.m4a agenda.txt - -OR - using 30min-CyberHR example: - -python 0-reflector-local.py 30min-CyberHR/30min-CyberHR.m4a 30min-CyberHR/30min-CyberHR-agenda.txt \ No newline at end of file diff --git a/server/reflector-local/whisper_summarizer_bart.py b/server/reflector-local/whisper_summarizer_bart.py deleted file mode 100644 index 4184fafe..00000000 --- a/server/reflector-local/whisper_summarizer_bart.py +++ /dev/null @@ -1,125 +0,0 @@ -import argparse -import os -import tempfile - -import moviepy.editor -import nltk -import whisper -from loguru import logger -from transformers import BartTokenizer, BartForConditionalGeneration - -nltk.download('punkt', quiet=True) - -WHISPER_MODEL_SIZE = "base" - - -def init_argparse() -> argparse.ArgumentParser: - parser = argparse.ArgumentParser( - usage="%(prog)s [OPTIONS] ", - description="Creates a transcript of a video or audio file, then summarizes it using BART." - ) - - parser.add_argument("location", help="Location of the media file") - parser.add_argument("output", help="Output file path") - - parser.add_argument( - "-t", "--transcript", help="Save a copy of the intermediary transcript file", type=str) - parser.add_argument( - "-l", "--language", help="Language that the summary should be written in", - type=str, default="english", choices=['english', 'spanish', 'french', 'german', 'romanian']) - parser.add_argument( - "-m", "--model_name", help="Name or path of the BART model", - type=str, default="facebook/bart-large-cnn") - - return parser - - -# NLTK chunking function -def chunk_text(txt, max_chunk_length=500): - "Split text into smaller chunks." - sentences = nltk.sent_tokenize(txt) - chunks = [] - current_chunk = "" - for sentence in sentences: - if len(current_chunk) + len(sentence) < max_chunk_length: - current_chunk += f" {sentence.strip()}" - else: - chunks.append(current_chunk.strip()) - current_chunk = f"{sentence.strip()}" - chunks.append(current_chunk.strip()) - return chunks - - -# BART summary function -def summarize_chunks(chunks, tokenizer, model): - summaries = [] - for c in chunks: - input_ids = tokenizer.encode(c, return_tensors='pt') - summary_ids = model.generate( - input_ids, num_beams=4, length_penalty=2.0, max_length=1024, no_repeat_ngram_size=3) - summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True) - summaries.append(summary) - return summaries - - -def main(): - import sys - sys.setrecursionlimit(10000) - - parser = init_argparse() - args = parser.parse_args() - - media_file = args.location - logger.info(f"Processing file: {media_file}") - - # If the media file we just retrieved is a video, extract its audio stream. - if os.path.isfile(media_file) and media_file.endswith(('.mp4', '.avi', '.flv')): - audio_filename = tempfile.NamedTemporaryFile( - suffix=".mp3", delete=False).name - logger.info(f"Extracting audio to: {audio_filename}") - - video = moviepy.editor.VideoFileClip(media_file) - video.audio.write_audiofile(audio_filename, logger=None) - - logger.info("Finished extracting audio") - media_file = audio_filename - - # Transcribe the audio file using the OpenAI Whisper model - logger.info("Loading Whisper speech-to-text model") - whisper_model = whisper.load_model(WHISPER_MODEL_SIZE) - - logger.info(f"Transcribing audio file: {media_file}") - whisper_result = whisper_model.transcribe(media_file) - - logger.info("Finished transcribing file") - - # If we got the transcript parameter on the command line, save the transcript to the specified file. - if args.transcript: - logger.info(f"Saving transcript to: {args.transcript}") - transcript_file = open(args.transcript, "w") - transcript_file.write(whisper_result["text"]) - transcript_file.close() - - # Summarize the generated transcript using the BART model - logger.info(f"Loading BART model: {args.model_name}") - tokenizer = BartTokenizer.from_pretrained(args.model_name) - model = BartForConditionalGeneration.from_pretrained(args.model_name) - - logger.info("Breaking transcript into smaller chunks") - chunks = chunk_text(whisper_result['text']) - - logger.info( - f"Transcript broken into {len(chunks)} chunks of at most 500 words") # TODO fix variable - - logger.info(f"Writing summary text in {args.language} to: {args.output}") - with open(args.output, 'w') as f: - f.write('Summary of: ' + args.location + "\n\n") - summaries = summarize_chunks(chunks, tokenizer, model) - for summary in summaries: - f.write(summary.strip() + "\n\n") - - logger.info("Summarization completed") - - -if __name__ == "__main__": - main() diff --git a/server/trials/__init__.py b/server/trials/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/server/trials/finetuning/__init__.py b/server/trials/finetuning/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/server/trials/finetuning/inference_fine_tuned.py b/server/trials/finetuning/inference_fine_tuned.py deleted file mode 100644 index 4a396071..00000000 --- a/server/trials/finetuning/inference_fine_tuned.py +++ /dev/null @@ -1,24 +0,0 @@ -# Steps to prepare data and submit/check OpenAI finetuning -# import subprocess -# subprocess.run("openai tools fine_tunes.prepare_data -f " + "finetuning_dataset.jsonl") -# export OPENAI_API_KEY= -# openai api fine_tunes.create -t -m -# openai api fine_tunes.list - - -import openai - -# Use your OpenAI API Key -openai.api_key = "" - -sample_chunks = ["You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . -> ", - " We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas . - > "] - -# Give your finetuned model name here -# "davinci:ft-personal-2023-07-14-10-43-51" -model_name = "" -response = openai.Completion.create( - model=model_name, - prompt=sample_chunks[0]) - -print(response) diff --git a/server/trials/finetuning/youtube_scraping.py b/server/trials/finetuning/youtube_scraping.py deleted file mode 100644 index be8b7e41..00000000 --- a/server/trials/finetuning/youtube_scraping.py +++ /dev/null @@ -1,98 +0,0 @@ -import json -import yt_dlp as youtube_dl -from whisper_jax import FlaxWhisperPipline -import jax.numpy as jnp - -# Function to extract chapter information from a YouTube video URL -def get_youtube_chapters(video_id): - video_url = "https://www.youtube.com/watch?v=" + video_id - ydl_opts = { - 'extract_flat': 'in_playlist', - 'skip_download': True, - 'quiet': True, - } - - with youtube_dl.YoutubeDL(ydl_opts) as ydl: - video_info = ydl.extract_info(video_url, download=False) - - chapters = [] - - if 'chapters' in video_info: - for chapter in video_info['chapters']: - start_time = chapter['start_time'] - end_time = chapter['end_time'] - title = chapter['title'] - - chapters.append({ - 'start': start_time, - 'end': end_time, - 'title': title - }) - - return chapters - - -# Function to extract video transcription using yt_dlp -def get_youtube_transcription(video_id): - ydl_opts = { - 'format': 'bestaudio/best', - 'postprocessors': [{ - 'key': 'FFmpegExtractAudio', - 'preferredcodec': 'mp3', - 'preferredquality': '192', - }], - 'outtmpl': './artefacts/audio', # Specify output file path and name - } - - # Download the audio - with youtube_dl.YoutubeDL(ydl_opts) as ydl: - ydl.download(["https://www.youtube.com/watch?v=" + video_id]) - media_file = "./artefacts/audio.mp3" - - pipeline = FlaxWhisperPipline("openai/whisper-" + "tiny", - dtype=jnp.float16, - batch_size=16) - whisper_result = pipeline(media_file, return_timestamps=True) - return whisper_result["chunks"] - - - -# Function to scrape YouTube video transcripts and chapter information -def scrape_youtube_data(video_id): - transcript_text = get_youtube_transcription(video_id) - chapters = get_youtube_chapters(video_id) - print("transcript_text", transcript_text) - print("chapters", chapters) - return transcript_text, chapters - - -# Function to generate fine-tuning dataset from YouTube data -def generate_finetuning_dataset(video_ids): - prompt_completion_pairs = [] - for video_id in video_ids: - transcript_text, chapters = scrape_youtube_data(video_id) - if transcript_text is not None and chapters is not None: - for chapter in chapters: - start_time = chapter["start"] - end_time = chapter["end"] - chapter_text = chapter["title"] - - prompt = "" - for transcript in transcript_text: - if transcript["timestamp"][0] >= start_time and transcript["timestamp"][1] < end_time: - prompt += transcript["text"] - - if prompt is not None: - completion = chapter_text - prompt_completion_pairs.append({"prompt": prompt, "completion": completion}) - - return prompt_completion_pairs - - -# Add all the video ids here, the videos must have captions [chapters] -video_ids = ["yTnSEZIwnkU"] -dataset = generate_finetuning_dataset(video_ids) - -with open("finetuning_dataset.jsonl", "w", encoding="utf-8") as file: - for example in dataset: - file.write(json.dumps(example) + "\n") diff --git a/server/trials/server/__init__.py b/server/trials/server/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/server/trials/server/server_multithreaded.py b/server/trials/server/server_multithreaded.py deleted file mode 100644 index 6739fbf6..00000000 --- a/server/trials/server/server_multithreaded.py +++ /dev/null @@ -1,188 +0,0 @@ -import asyncio -import datetime -import io -import json -import threading -import uuid -import wave -from concurrent.futures import ThreadPoolExecutor - -import jax.numpy as jnp -import requests -from aiohttp import web -from aiortc import MediaStreamTrack, RTCPeerConnection, RTCSessionDescription -from aiortc.contrib.media import MediaRelay -from av import AudioFifo -from sortedcontainers import SortedDict -from whisper_jax import FlaxWhisperPipline - -from reflector.utils.log_utils import LOGGER -from reflector.utils.run_utils import CONFIG, Mutex - -WHISPER_MODEL_SIZE = CONFIG['WHISPER']["WHISPER_REAL_TIME_MODEL_SIZE"] -pcs = set() -relay = MediaRelay() -data_channel = None -sorted_message_queue = SortedDict() -CHANNELS = 2 -RATE = 44100 -CHUNK_SIZE = 256 -pipeline = FlaxWhisperPipline("openai/whisper-" + WHISPER_MODEL_SIZE, - dtype=jnp.float16, - batch_size=16) -start_time = datetime.datetime.now() -executor = ThreadPoolExecutor() -audio_buffer = AudioFifo() -frame_lock = Mutex(audio_buffer) - - -def channel_log(channel, t, message): - print("channel(%s) %s %s" % (channel.label, t, message)) - - -def thread_queue_channel_send(): - loop = asyncio.new_event_loop() - asyncio.set_event_loop(loop) - try: - least_time = sorted_message_queue.keys()[0] - message = sorted_message_queue[least_time] - if message: - del sorted_message_queue[least_time] - data_channel.send(message) - except Exception as e: - print("Exception", str(e)) - pass - loop.run_forever() - - -def get_transcription(): - while True: - with frame_lock.lock() as audio_buffer: - frames = audio_buffer.read_many(CHUNK_SIZE * 960, partial=False) - if not frames: - transcribe = False - else: - transcribe = True - - if transcribe: - print("Transcribing..") - try: - sorted_message_queue[frames[0].time] = None - out_file = io.BytesIO() - wf = wave.open(out_file, "wb") - wf.setnchannels(CHANNELS) - wf.setframerate(RATE) - wf.setsampwidth(2) - - for frame in frames: - wf.writeframes(b''.join(frame.to_ndarray())) - wf.close() - - whisper_result = pipeline(out_file.getvalue()) - item = { - 'text': whisper_result["text"], - 'start_time': str(frames[0].time), - 'time': str(datetime.datetime.now()) - } - sorted_message_queue[frames[0].time] = str(item) - start_messaging_thread() - except Exception as e: - print("Exception -> ", str(e)) - - -class AudioStreamTrack(MediaStreamTrack): - """ - An audio stream track to send audio frames. - """ - - kind = "audio" - - def __init__(self, track): - super().__init__() # don't forget this! - self.track = track - - async def recv(self): - frame = await self.track.recv() - audio_buffer.write(frame) - return frame - - -def start_messaging_thread(): - message_thread = threading.Thread(target=thread_queue_channel_send) - message_thread.start() - - -def start_transcription_thread(max_threads: int): - for i in range(max_threads): - t_thread = threading.Thread(target=get_transcription) - t_thread.start() - - -async def offer(request: requests.Request): - params = await request.json() - offer = RTCSessionDescription(sdp=params["sdp"], type=params["type"]) - - pc = RTCPeerConnection() - pc_id = "PeerConnection(%s)" % uuid.uuid4() - pcs.add(pc) - - def log_info(msg: str, *args): - LOGGER.info(pc_id + " " + msg, *args) - - log_info("Created for " + request.remote) - - @pc.on("datachannel") - def on_datachannel(channel): - global data_channel, start_time - data_channel = channel - channel_log(channel, "-", "created by remote party") - start_time = datetime.datetime.now() - - @channel.on("message") - def on_message(message: str): - channel_log(channel, "<", message) - if isinstance(message, str) and message.startswith("ping"): - # reply - channel.send("pong" + message[4:]) - - @pc.on("connectionstatechange") - async def on_connectionstatechange(): - log_info("Connection state is " + pc.connectionState) - if pc.connectionState == "failed": - await pc.close() - pcs.discard(pc) - - @pc.on("track") - def on_track(track): - log_info("Track " + track.kind + " received") - pc.addTrack(AudioStreamTrack(relay.subscribe(track))) - - # handle offer - await pc.setRemoteDescription(offer) - - # send answer - answer = await pc.createAnswer() - await pc.setLocalDescription(answer) - return web.Response( - content_type="application/json", - text=json.dumps({ - "sdp": pc.localDescription.sdp, - "type": pc.localDescription.type - }), - ) - - -async def on_shutdown(app: web.Application): - coros = [pc.close() for pc in pcs] - await asyncio.gather(*coros) - pcs.clear() - - -if __name__ == "__main__": - app = web.Application() - app.on_shutdown.append(on_shutdown) - start_transcription_thread(6) - app.router.add_post("/offer", offer) - web.run_app( - app, access_log=None, host="127.0.0.1", port=1250 - ) diff --git a/server/trials/title_summary/__init__.py b/server/trials/title_summary/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/server/trials/title_summary/api.py b/server/trials/title_summary/api.py deleted file mode 100644 index eb6a1fbb..00000000 --- a/server/trials/title_summary/api.py +++ /dev/null @@ -1,57 +0,0 @@ -import requests -import spacy - -# Enter the Machine where the LLM is hosted -LLM_MACHINE_IP = "" -# This is the URL of text-generation-webui -URL = f"http://{LLM_MACHINE_IP}:5000/api/v1/generate" - -headers = { - "Content-Type": "application/json" -} - - -def split_text_file(filename, token_count): - nlp = spacy.load('en_core_web_md') - - with open(filename, 'r') as file: - text = file.read() - - doc = nlp(text) - total_tokens = len(doc) - - parts = [] - start_index = 0 - - while start_index < total_tokens: - end_index = start_index + token_count - part_tokens = doc[start_index:end_index - 5] - part = ' '.join(token.text for token in part_tokens) - parts.append(part) - start_index = end_index - - return parts - - -final_summary = "" -parts = split_text_file("transcript.txt", 1600) - -for part in parts: - prompt = f""" - ### Human: - Given the following text, distill the most important information - into a short summary: {part} - - ### Assistant: - """ - data = { - "prompt": prompt - } - try: - response = requests.post(URL, headers=headers, json=data) - print(response.json()) - except Exception as e: - print(str(e)) - -with open("summary.txt", "w") as sum: - sum.write(" ".join(final_summary)) diff --git a/server/trials/title_summary/bert.py b/server/trials/title_summary/bert.py deleted file mode 100644 index a79bb76d..00000000 --- a/server/trials/title_summary/bert.py +++ /dev/null @@ -1,43 +0,0 @@ -import torch -from transformers import BertTokenizer, BertModel -from sentence_transformers import SentenceTransformer -from sklearn.metrics.pairwise import cosine_similarity - -# Load the pre-trained BERT model and tokenizer -model_name = "bert-base-uncased" -model = BertModel.from_pretrained(model_name) -tokenizer = BertTokenizer.from_pretrained(model_name) - -# Set the device to use -device = torch.device("cuda" if torch.cuda.is_available() else "cpu") -model.to(device) - -# Load the SentenceTransformer model -sentence_transformer_model = SentenceTransformer('average_word_embeddings_glove.6B.300d') - -# Define the input text -text = "Your input text to be summarized goes here." - -# Tokenize the text -tokens = tokenizer.tokenize(text) -input_ids = tokenizer.convert_tokens_to_ids(tokens) -input_ids = torch.tensor([input_ids]).to(device) - -# Get the BERT model output -with torch.no_grad(): - outputs = model(input_ids)[0] # Extract the last hidden states - -# Calculate sentence embeddings -sentence_embeddings = outputs.mean(dim=1).squeeze().cpu().numpy() -input_text_embedding = sentence_transformer_model.encode([text])[0] - -# Calculate cosine similarity between sentences and input text -similarity_scores = cosine_similarity([input_text_embedding], sentence_embeddings) - -# Sort the sentences by similarity scores in descending order -sorted_sentences = [sent for _, sent in sorted(zip(similarity_scores[0], sentences), reverse=True)] - -# Choose the top sentences as the summary -num_summary_sentences = 2 # Adjust as needed -summary = ". ".join(sorted_sentences[:num_summary_sentences]) -print("Summary:", summary) diff --git a/server/trials/title_summary/chat_llm.py b/server/trials/title_summary/chat_llm.py deleted file mode 100644 index 557fb531..00000000 --- a/server/trials/title_summary/chat_llm.py +++ /dev/null @@ -1,79 +0,0 @@ -""" -This is an example code containing the bare essentials to load a chat - LLM and infer from it using a predefined prompt. The purpose of this file - is to show an example of inferring from a chat LLM which is required for - banana.dev due to its design and platform limitations -""" - -# The following logic was tested on the monadical-ml machine - -import json - -import torch -from transformers import ( - AutoModelForCausalLM, - AutoTokenizer -) -from transformers.generation import GenerationConfig - -# This can be passed via the environment variable or the params supplied -# when starting the program via banana.dev platform -MODEL_NAME = "lmsys/vicuna-13b-v1.5" - -# Load the model in half precision, and less memory usage -model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, - low_cpu_mem_usage=True, - torch_dtype=torch.bfloat16 - ) - -# Generation config -model.config.max_new_tokens = 300 -gen_cfg = GenerationConfig.from_model_config(model.config) -gen_cfg.max_new_tokens = 300 - -# Load the tokenizer -tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME) - -# Move model to GPU -model = model.cuda() -print(f"Loading {MODEL_NAME} successful") - -# Inputs -sample_chunks = [ - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . ", - " We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas ."] - -# Model Prompt template for current model -prompt = f""" - ### Human: - Create a JSON object as response.The JSON object must have 2 fields: - i) title and ii) summary.For the title field,generate a short title - for the given text. For the summary field, summarize the given text - in three sentences. - - {sample_chunks[0]} - - ### Assistant: - """ - -# Inference : Chat generation -input_ids = tokenizer.encode(prompt, return_tensors='pt').to(model.device) -output = model.generate(input_ids, generation_config=gen_cfg) - -# Process output -response = tokenizer.decode(output[0].cpu(), skip_special_tokens=True) -response = response.split("### Assistant:\n") -print("TitleSummaryJsonResponse :", json.loads(response[1])) -print("Inference successful") - -# Sample response for sample_chunks[0] - -# TitleSummaryJsonResponse : -# { -# 'title': 'Google Cloud Next Conference: Simplifying AI and Machine Learning for Everyone', -# 'summary': 'Google Cloud announced a wide range of innovations and new products in the AI -# and machine learning space at the recent Google Cloud Next conference. The goal -# is to make these technologies accessible to everyone by simplifying the process -# and providing tools for data processing, cybersecurity, and machine learning. -# Google is also working on advances in AutoML and packaged solutions for certain areas.' -# } \ No newline at end of file diff --git a/server/trials/title_summary/gpt2.py b/server/trials/title_summary/gpt2.py deleted file mode 100644 index 1930a2d2..00000000 --- a/server/trials/title_summary/gpt2.py +++ /dev/null @@ -1,101 +0,0 @@ -# Approach 1 -from transformers import GPTNeoForCausalLM, GPT2Tokenizer - -model_name = 'EleutherAI/gpt-neo-1.3B' -tokenizer = GPT2Tokenizer.from_pretrained(model_name) -model = GPTNeoForCausalLM.from_pretrained(model_name) - -conversation = """ -Summarize the following conversation in 3 key sentences: - -We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . -Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . -Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . -Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . -Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . -Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas . -""" - -input_ids = tokenizer.encode(conversation, return_tensors='pt') - -output = model.generate(input_ids, - max_length=30, - num_return_sequences=1) - -caption = tokenizer.decode(output[0], skip_special_tokens=True) -print("Caption:", caption[len(input_ids):]) - - -# Approach 2 -import torch -from transformers import GPT2LMHeadModel, GPT2Tokenizer - -model_name = "gpt2" -tokenizer = GPT2Tokenizer.from_pretrained(model_name) -model = GPT2LMHeadModel.from_pretrained(model_name) - -model.eval() - -text = """ -You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . " -""" - -tokenizer.pad_token = tokenizer.eos_token -input_ids = tokenizer.encode(text, - max_length=100, - truncation=True, - return_tensors="pt") -attention_mask = torch.ones(input_ids.shape, dtype=torch.long) -output = model.generate(input_ids, - max_new_tokens=20, - num_return_sequences=1, - num_beams=2, - attention_mask=attention_mask) - -chapter_titles = [tokenizer.decode(output[i], skip_special_tokens=True) for i in range(output.shape[0])] -for i, title in enumerate(chapter_titles): - print("Caption: ", title) - -# Approach 3 - -import torch -from transformers import GPT2LMHeadModel, GPT2Tokenizer - - -def generate_response(conversation, max_length=100): - input_text = "" - for entry in conversation: - role = entry["role"] - content = entry["content"] - input_text += f"{role}: {content}\n" - - # Tokenize the entire conversation - input_ids = tokenizer.encode(input_text, return_tensors="pt") - - # Generate text based on the entire conversation - with torch.no_grad(): - output = model.generate(input_ids, pad_token_id=tokenizer.eos_token_id) - - # Decode the generated text and return it - response = tokenizer.decode(output[0], skip_special_tokens=True) - return response - - -if __name__ == "__main__": - - # Call appropriate approach from the main while experimenting - model_name = "gpt2" - model = GPT2LMHeadModel.from_pretrained(model_name) - tokenizer = GPT2Tokenizer.from_pretrained(model_name) - - sample_chunks = [ - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . " - ] - - conversation = [ - {"role": "system", "content": "Summarize this text"}, - {"role": "user", "content": " text : " + sample_chunks[0]}, - ] - - response = generate_response(conversation) - print("Response:", response) diff --git a/server/trials/title_summary/incsum.py b/server/trials/title_summary/incsum.py deleted file mode 100644 index 571af77f..00000000 --- a/server/trials/title_summary/incsum.py +++ /dev/null @@ -1,157 +0,0 @@ -import spacy -import sys - - -# Observe the incremental summaries by performing summaries in chunks -with open("transcript.txt", "r", encoding="utf-8") as file: - transcription = file.read() - - -def split_text_file(filename, token_count): - nlp = spacy.load('en_core_web_md') - - with open(filename, 'r', encoding="utf-8") as file: - text = file.read() - - doc = nlp(text) - total_tokens = len(doc) - - parts = [] - start_index = 0 - - while start_index < total_tokens: - end_index = start_index + token_count - part_tokens = doc[start_index:end_index] - part = ' '.join(token.text for token in part_tokens) - parts.append(part) - start_index = end_index - - return parts - - -# Set the chunk length here to split the transcript and test -MAX_CHUNK_LENGTH = 1000 - -chunks = split_text_file("transcript.txt", MAX_CHUNK_LENGTH) -print("Number of chunks", len(chunks)) - -# Write chunks to file to refer to input vs output, separated by blank lines -with open("chunks" + str(MAX_CHUNK_LENGTH) + ".txt", "a", encoding="utf-8") as file: - for c in chunks: - file.write(c + "\n\n") - -# If we want to run only a certain model, type the option while running -# ex. python incsum.py 1 => will run approach 1 -# If no input, will run all approaches - -try: - index = sys.argv[1] -except: - index = None - -# Approach 1 : facebook/bart-large-cnn -if index == "1" or index is None: - SUMMARY_MODEL = "facebook/bart-large-cnn" - MIN_LENGTH = 5 - MAX_LENGTH = 10 - BEAM_SIZE = 2 - - print("Performing chunk summary : " + SUMMARY_MODEL) - - from transformers import BartTokenizer, BartForConditionalGeneration - - tokenizer = BartTokenizer.from_pretrained(SUMMARY_MODEL) - model = BartForConditionalGeneration.from_pretrained(SUMMARY_MODEL) - summaries = [] - for c in chunks: - input_ids = tokenizer.encode(c, - truncation=True, - max_length=MAX_CHUNK_LENGTH, - padding="max_length", - return_tensors='pt') - summary_ids = model.generate( - input_ids, - num_beams=BEAM_SIZE, - max_length=56, - early_stopping=True, - length_penalty=1.0) - summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True) - summaries.append(summary) - - with open("bart-summaries.txt", "a", encoding="utf-8") as file: - for summary in summaries: - file.write(summary + "\n\n") - -# Approach 2 -if index == "2" or index is None: - print("Performing chunk summary : " + "gpt-neo-1.3B") - - import torch - from transformers import GPTNeoForCausalLM, GPT2Tokenizer - - model = GPTNeoForCausalLM.from_pretrained("EleutherAI/gpt-neo-1.3B") - tokenizer = GPT2Tokenizer.from_pretrained("EleutherAI/gpt-neo-1.3B") - tokenizer.add_special_tokens({'pad_token': '[PAD]'}) - summaries = [] - - for c in chunks: - input_ids = tokenizer.encode(c, - truncation=True, - return_tensors='pt') - input_length = input_ids.shape[1] - attention_mask = torch.ones(input_ids.shape, dtype=torch.long) - - max_summary_length = 100 - max_length = input_length + max_summary_length - - output = model.generate(input_ids, - max_length=max_length, - attention_mask=attention_mask, - pad_token_id=model.config.eos_token_id, - num_beams=4, - length_penalty=2.0, - early_stopping=True) - summary_ids = output[0, input_length:] - summary = tokenizer.decode(summary_ids, skip_special_tokens=True) - summaries.append(summary) - with open("gptneo1.3B-summaries.txt", "a", encoding="utf-8") as file: - file.write(summary + "\n\n") - -# Approach 3 -if index == "3" or index is None: - print("Performing chunk summary : " + "mpt-7B") - - import torch - import transformers - from transformers import AutoTokenizer - - config = transformers.AutoConfig.from_pretrained('mosaicml/mpt-7b', - trust_remote_code=True) - config.attn_config['attn_impl'] = 'triton' - config.max_seq_len = 1024 - config.init_device = "meta" - - model = transformers.AutoModelForCausalLM.from_pretrained( - 'mosaicml/mpt-7b', - trust_remote_code=True, - torch_dtype=torch.bfloat16 - ) - - tokenizer = AutoTokenizer.from_pretrained('EleutherAI/gpt-neox-20b') - - summaries = [] - for c in chunks: - input_ids = tokenizer.encode(c, return_tensors="pt") - attention_mask = torch.ones(input_ids.shape, dtype=torch.long) - output = model.generate(input_ids, - max_new_tokens=25, - attention_mask=attention_mask, - pad_token_id=model.config.eos_token_id, - num_return_sequences=1) - summary = tokenizer.decode(output[0], - skip_special_tokens=True) - summaries.append(summary) - - with open("mpt-7b-summaries.txt", "a", encoding="utf-8") as file: - for summary in summaries: - file.write(summary + "\n\n") diff --git a/server/trials/title_summary/openai_endpoint.py b/server/trials/title_summary/openai_endpoint.py deleted file mode 100644 index c92856c5..00000000 --- a/server/trials/title_summary/openai_endpoint.py +++ /dev/null @@ -1,37 +0,0 @@ -# Use OpenAI API endpoint to send data to OpenAI -# along with prompts to caption/summarize the conversation - -import openai - -openai.api_key = "" - -# to caption, user prompt used : "caption this conversation" -# max_tokens=20 - -# to incremental summarize, user prompt used : "summarize this conversation in a few sentences by taking key points" -# max_tokens=300 - -sample_chunks = [ - "You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . ", - " We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas ."] - -conversation = [ - {"role": "system", - "content": sample_chunks[1]}, - {"role": "user", - "content": "summarize this conversation in a few sentences by taking key points"} -] - -model = "gpt-3.5-turbo" -response = openai.ChatCompletion.create(model=model, - messages=conversation, - n=1, - max_tokens=300) - -# Try fine tuned model -# model = "davinci:ft-personal-2023-07-14-10-43-51" -# response = openai.Completion.create(model=model, -# prompt=sample_chunks[0] + " -> ") - -caption = response.choices[0] -print(caption) diff --git a/server/trials/title_summary/pegasus.py b/server/trials/title_summary/pegasus.py deleted file mode 100644 index 884ed3ee..00000000 --- a/server/trials/title_summary/pegasus.py +++ /dev/null @@ -1,33 +0,0 @@ -from transformers import PegasusForConditionalGeneration, PegasusTokenizer -import torch -# Load the Pegasus model and tokenizer -model_name = "google/pegasus-large" -model = PegasusForConditionalGeneration.from_pretrained(model_name) -tokenizer = PegasusTokenizer.from_pretrained(model_name) - -# Set the device to use -device = torch.device("cuda" if torch.cuda.is_available() else "cpu") -model.to(device) - -sample_chunks = ["You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . ", - " We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas ."] - - -# Define the input text for summarization -text = sample_chunks[1] - -inputs = tokenizer(text, truncation=True, padding="longest", return_tensors="pt").to(device) - -# Generate the summary -summary_ids = model.generate( - inputs["input_ids"], - attention_mask=inputs["attention_mask"], - max_length=200, - num_beams=4, - length_penalty=2.0, - early_stopping=True, -) - -# Decode and print the summary -summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True) -print("Summary:", summary) diff --git a/server/trials/title_summary/t5.py b/server/trials/title_summary/t5.py deleted file mode 100644 index 0c366ac6..00000000 --- a/server/trials/title_summary/t5.py +++ /dev/null @@ -1,27 +0,0 @@ -from transformers import T5ForConditionalGeneration, T5Tokenizer -import torch -# Load the T5 model and tokenizer -model_name = "t5-base" -model = T5ForConditionalGeneration.from_pretrained(model_name) -tokenizer = T5Tokenizer.from_pretrained(model_name) - -# Set the device to use -device = torch.device("cuda" if torch.cuda.is_available() else "cpu") -model.to(device) - -sample_chunks = ["You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . ", - " We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas ."] - - -# Define the input text for summarization -text = "Summarize the following text in 3 key points. text : " + sample_chunks[1] - -# Tokenize the input text -inputs = tokenizer.encode(text, return_tensors="pt").to(device) - -# Generate the summary -summary_ids = model.generate(inputs, max_length=1000, num_beams=4, early_stopping=True) - -# Decode and print the summary -summary = tokenizer.decode(summary_ids.squeeze(), skip_special_tokens=True) -print("Summary:", summary) diff --git a/server/trials/title_summary/transcript.txt b/server/trials/title_summary/transcript.txt deleted file mode 100644 index 316458df..00000000 --- a/server/trials/title_summary/transcript.txt +++ /dev/null @@ -1 +0,0 @@ -We 're joined next by Thomas Curian , CEO of Google Cloud , and Alexander Wang , CEO and founder of Scale AI . Thomas joined Google in November 2018 as the CEO of Google Cloud . Prior to Google , Thomas spent 22 years at Oracle , where most recently he was president of product development . Before that , Thomas worked at McKinsey as a business analyst and engagement manager . His nearly 30 years of experience have given him a deep knowledge of engineering enterprise relationships and leadership of large organizations . Thomas 's degrees include an MBA in administration and management from Stanford University , as an RJ Miller scholar and a BSEE in electrical engineering and computer science from Princeton University , where he graduated suma cum laude . Thomas serves as a member of the Stanford graduate School of Business Advisory Council and Princeton University School of Engineering Advisory Council . Please welcome to the stage , Thomas Curian and Alexander Wang . This is a super exciting conversation . Thanks for being here , Thomas . Thank you for having me . You all just came off of your incredible Google Cloud next conference where you released a wide variety of functionality and features and new products across artisan television and also across the entire sort of cloud ecosystem . You want to just first by walking through , first start by walking through all the innovations that you sort of released and what you 're excited about when you come to Google Cloud ? Now our vision is super simple . If you look at what smartphones did for a consumer , you know they took a computer and internet browser , a communication device , and a camera , and made it so that it 's in everybody 's pocket , so it really brought computation to every person . We feel that , you know , our , what we 're trying to do is take all the technological innovation that Google 's doing , but make it super simple so that everyone can consume it . And so that includes our global data center footprint , all the new types of hardware and large-scale systems we work on , the software that we 're making available for people to do high-scale computation , tools for data processing , tools for cybersecurity , processing , tools for cyber security , tools for machine learning , but make it so simple that everyone can use it . And every step that we do to simplify things for people , we think adoption can grow . And so that 's a lot of what we 've done these last three , four years , and we made a number of announcements that next in machine learning and AI in particular , you know , we look at our work as four elements , how we take our large-scale compute systems that were building for AI and how we make that available to everybody . Second , what we 're doing with the software stacks and top of it , things like jacks and other things and how we 're making those available to everybody . Third is advances because different people have different levels of expertise . Some people say I need the hardware to build my own large language model or algorithm . Other people say , look , I really need to use a building block . You guys give me . So , 30s we 've done a lot with AutoML and we announce new capability for image , video , and translation to make it available to everybody . And then lastly , we 're also building completely packaged solutions for some areas and we announce some new stuff . So , it 's a busy conference , but lots of exciting stuff going on . Yeah , it 's incredible . I mean , I want to zoom out for a second to start with , which is that this is obviously not your first time taking and packaging new technology breakthroughs for the enterprise . Both in your time at Oracle and now CEO of Google Cloud , this is something that you 've been doing for quite some time now . When you sort of zoom all the way out , what do you think are some of the things that have some of your principles , or some of your thoughts and enabling these technological breakthroughs and actually enabling the enterprise with them ? And what are the key insights that you have there ? Thank you . A lot of the work . So first of all , we 've really built out the organization the last three years . We 've seen a huge ramp up in our business , credit to all the people who joined us at one point over 70 % of organization that joined your in COVID . So they had n't met anybody . They could n't meet their managers , but they all did an amazing job together . The adoption of technology by companies , and I 'll give you just some elements , particularly in the application of AI in different domains that we 've seen . We work with a large financial institution in Hong Kong and Shanghai Bank , which uses our machine learning to detect fraud . You know , fraud detection and banking , there 's a lot of false positives , which makes it hard to really , you know , to a very expensive people doing something called anti-money laundering . And our AI algorithms are really able to be super precise on detection . Explainability is a critical thing there , right ? So people ask , why did you , why did you approve , why did you flag this one and not that one ? Because regulators are involved . So explainability becomes a big deal . We help , we help renewal , for example , monitor all of the factories . The process roughly , a billion data sets every day . Obviously , humans can process that . But making it super simple to , and you guys have given all your expertise in labeling and other things , you would get a sense . Factory floor data is not clean data . And so you have to actually clean , imagine doing a billion data sets into an environment every single day . You have to give the data pipelines really good . And so a lot of technology work happens to make that possible for companies . Third is , if you shop at IKEA , for example , behind IKEA is systems , it 's our recommendation system . find IKEA is systems , it 's our recommendation system . And the way that people shop for furniture and products is not the same in all countries . And so how are you able to one deal with the benefits you get from a global model , but also to contextually the specific elements in each country because people have different buying habits . Those are all things that we 've learned applying our AI in different contexts in different parts of the world . Yeah . You 've sort of glossed over this , but you 've led since you took over at Google Cloud , just a meteoric growth of the platform . You know , I think the past few years , you 've tripled your sales force and ending last year , you obviously ca n't come in this , but end the last year at , I believe , 20 billion of annual revenue , which is incredible and this incredible growth journey . What do you attribute your success to ? And how do you think you 've been able to drive just to an incredible growth and success ? From our point of view , every industry , virtually in the world , is now becoming a software powered technology industry . If you talk to automobile companies , they 're increasingly vehicles are more about software than mechanical systems . If you talk to telecommunications companies , the networks are commodities unless they can make them platforms to deliver applications , so they need new ways to slice , manage the network . If you look at banks at the end of the day , they 're about all the products of a bank or data , and all of that becomes how do you differentiate in the value delivering clients through a digital medium ? Because increasingly , I 'm sure all of you look at yourselves and go when was the last time I went to a branch of a bank . So a lot of our work has been pushing the technology innovation really far , but bringing that technology super easily to people in different industries . And given the demand that people have for a hair , I really want , I need the technology to help me power my industry , the change I 'm seeing in my industry , the more accessible we can make it , the easier and the faster we get adoption , and our approach has been to be completely open . And when to be completely open . And when I say completely open , we offer every part of the stack that we have from the hardware and network to the software abstractions above to things that are more packaged because different organizations have different levels at which they have expertise and want to adopt technology . Yeah . I mean it 's been , mean it 's been obviously incredible . You know going back to AI for a second , Google , Google obviously is an early mover in AI and Google Cloud has also been through , you know , starting with TensorFlow and Vertex AI and AutoML and so many incredibly innovative technologies . And AI has been obviously kind of a buzzword for some time now within the industry . And I think we see this in use as well . The adoption has maybe been a bit slower than we would expected until now . What do you think have been the barriers to greater levels of AI adoption , greater levels of enterprise that 's in value from AI ? And what do you think the future holds ? So we 've worked with a huge number of companies doing work , having them adopt AI . A lot of the lessons we 've seen and observed from it are the barriers to adoption are rarely about the algorithm itself . It 's often the barriers to adoption about very algorithm itself . It 's often the various adoption about very different things . So when we work with customers in many , many industries , take retailers an example , and you think of a very mundane example , like recommendations , to make product discovery on the web much easier for their own products . The biggest challenges standardizing the meaning of the product and the catalog . Because unless you have a standardized definition of the products and the data behind the algorithm is clean , it 's super hard to actually get to recommendation . And so in the work we did with H & M , for example , or at Macy 's , or at IKEA , or Bloomingdale 's , a huge number of these brands , the big part of the program is actually how do you label and clean the data upfront and standardize it before you get into the algorithmic phase . So that 's one part of things we see . Second part is for large organizations to adopt AI , they have to need to integrate the results of the algorithm back into their core processes . So , you know , practical example , we work with OGE , OGE is a large , large electric producer , electricity and power producer in Europe . They are probably one of the largest renewable energy producer in the world . They use wind farms . One of the things they really struggled with was , how do you predict how much wind is going to be there three days from now ? Because the power grid requires that prediction in order to capacity plan how much power is going into the grid . So they work with us and they use our AI to do that . But that needs to be tied into how they 're telling the rest of the power sources that work on the grid . Hey , if this went to wind is coming in , here 's all the other sources in each generation . So tying it back in is not as simple as people think . And so a lot of time is that the third on the people side , there 's change management you go through to get people to trust the algorithm . So one of the things we 've done work with many banks , particularly during the pandemic , when the government issued small business loans . There was a giant bottleneck in being able to get loans out to individual consumers . And frankly , because the banks did n't want to bring a huge army of loan officers in , they had to use software and algorithms to process it . Now the challenge people had is they needed to trust the algorithm was being fair in saying yes to some and no to others and that it would mirror for example the recommendations that their best mortgage bankers would do , right ? Just as a loan office as we do . So it gave them the benefit of scale because we processed literally millions and millions of mortgages through our technology , but it required them to get comfortable that things like fairness and other things were working . So often when people look at AI , they think it 's a skills issue . There 's certainly a skill issue involved . There 's not enough talent in the ecosystem . But things are getting easier and easier as the models get more and more sophisticated . Often people forget about these other issues that are important in getting adoption . Yeah . I mean , you 're preaching the choir when you mention the data challenges that all these enterprises face and how critical that is to getting working in the early days . One of the things that I think is interesting about Google Cloud strategies that you really have products at different layers of the stack and different layers of closest to the bare metal all the way up to these package solutions . In what way do you think that the enterprise world and even the broader business world is going to adopt these AI technologies ? Do you think that the end state is that a lot of them are using your lower level , more infrastructure ? Products , or do you think that many of them are going to adopt solutions ? How do you think this plays out over the next few years ? So we offer four layers of technology for people . There 's a set of people who say , look , I just need your computational infrastructure , your large systems . We build something called tens of processing unit , which is our large scale systems . We 're also working with Crossing Unit , which is our large-scale systems . We 're also working within video to build a really high-scale GPU Bay system . But many people , some customers say , look , I just need access to that . And we make that available because the TPUs are what we use within Google . And we make that available along with the compilation software to optimize models on the TPUs . Take as an example , LG , the Korean company that makes appliances , their team is built a large , I mean , multi-hundred billion parameter model , because they wanted to make that a way that people can interact with appliances without having to press buttons on them . So they built a model . They said , I just need access to your infrastructures . That 's one way we offer a peak capability . A second level is people say look , I really do n't need access to the raw infrastructure itself . What I need is the ability to build models using your platform . And so we offer a platform called Vertex and people build models and push them using our machine learning platform . And there are many , many organizations in logistics and financial services in retail and others who build their own models on top of the platform . The third is to make things even easier , we 've taken some of the core pieces , translation , documents , image processing , video . And we 've said , we can offer an auto-email based solution , which further simplifies how you use our platforms . And so for example , translation , we have a capability to handle translation in 135 languages . One of the important things that people ask when they go to many languages is if you look at the data sets that I used to train models , they are primarily , there 's a large set in English , because you have the whole internet is primarily in a very small number of languages . But once you get to more narrow languages , for instance , Swahili or some of the African languages , or even in Asia , there are many languages , even from where I grew up in India . There are languages that are not as widely represented on the internet . Can your model in translation provide equivalent fidelity in sparse languages ? Because it 's always important to those people only understand that language that they get a high fidelity result . So we 've built something called translation hub and it 's being used in very mundane places but with extraordinary impact . For example , when people announce COVID guidelines or recently monkey parks , for example , which is another thing , they needed translate many , many languages . And normally the process would take a long time . We have movie studios , for example , in a different example , saying , hey , when we launch a movie , we have a high fidelity set of languages , we 're actually going to hold the movie up and show that people do it . But for the long tail , we just need captioning . We 're not necessarily going to do voice dubbing . We 're going to do captioning . And they use our translation solutions to go to that . Even within companies , every medicine , for example , uses it to translate all their instruction manuals into many languages for their technicians . And then lastly , in some places , there are companies like retailers who tell us , look , a handful of the largest retailers may build their own software teams . But some of us who are small merchants , we 're not software companies . And telling us , you 've got to be a software company to use AI is not fair . So for some industries , we actually build fully packet solutions . If you call many telephone companies , the context center , behind it , sits our voice agent . And the rationale behind that was super simple , when a new smartphone launches like an iPhone or a Pixel , typically in the morning of the launch , some of these contact centers get three , four million calls in an hour . And it 's hard to hire that many agents to handle the phones . So we said , why would n't software be able to handle it ? We then evolved it so that the natural language interface can become actually the workflow for these organizations . But that 's a much more of a package solution so that telephone companies do n't have to have armies of data scientists to do it . So our work spans all of these because people have different needs and we find that as you improve the maturation of this and you make it more easy for people to adopt it . You will get broader proliferation and adoption of AI as a whole . Yeah , you know , you walk through so many different use cases and so many applications to the technology . I imagine one , and there 's so desperately , you know , everywhere from , you know , fraud detection to translation to translation of manuals , you know , there 's such a wide translation of manuals . There 's such a wide array of use cases . How do you all like Google Cloud think about helping businesses understand what is AI good for ? What can they use AI for ? There 's obviously such a wide diversity of different use cases , but what at a framework level do you tell them , how can I use AI within my business ? It 's a really good question . I mean , a lot of our work actually comes from clients asking us now , and that 's actually an encouraging thing . Because you know , see from up on the view , some simple things , how many of you believe in a few years ' time there 's gon na be intelligence software and non-intelligence software , right ? I mean , nobody would say in three , few years ' time , there 's going to be intelligence software and non-intelligence software . I mean , nobody would say in three , four years ' time , we 're going to write software that has not powered in some form of fashion by AI . So you know , in most companies actually , it 's really encouraging to see that they look at domain problems they 're having and say , for instance , I used to do it using a rules engine , which is an older model for defining kind of workflow within organizations . Can you apply AI to do it in a new way ? I used to do this in a specific way . I heard about image recognition . One example really fun or interesting one , US Navy , when you have corrosion on the base of ships , the old way was to lift it into dry dark and take a look at it . If you 've ever seen one of these ships , you can imagine lifting to dry dark is not an easy thing . So they said , can we fly a drone with your camera image recognition around it and detect corrosion ? And so what we 've seen is that as you lift up the capability where image , audio , text , et cetera , all these forms of input can be processed extremely accurately , most customers start figuring it out . And so they call us with , most of our work has come from customers calling us , saying , hey , I have this need . Can I apply AI to it ? And so we talk to them about how and when it makes sense to use AI . But we also talk to them about the consequences if the models are not handling things like skew in the data . How do you ensure that , for example , you 're treating fairness properly ? How do you ensure that the model is safe , etc . Yeah , I think it 's , I mean , all the use cases , the variety is incredibly exciting . It 's cool that these customers are coming to you directly with many of them . What is , again , kind of thinking bigger picture , what is machine learning an AI mean for Google Cloud on the whole over the next call 510 years ? So we feel that the boundary of what machine learning and what AI can do will change over time . When it started , it was about doing what we would call assistive things . Assistive things are where a human being is able to do it , but the computer assists the human being in some ways to do it better . Right ? So common examples people talk about is , hey , your doctor or radiologist , you used to look at x-ray images . Now , a computer is going to look at it and detect tumors , but it 's assisting you to find something that you may have done another way . So that 's the first phase and a lot of the work we see is primarily in that phase today . The second phase is to do something where you could n't do it with a human because the quantity of data you need to process or the amount of people you need would be just far too significant . And so the machine is doing something that humans could n't do , but it 's still an incremental element on top of what humans could do themselves . The third phase , I think , is where we think generative AI , for example , goes , because it 's about enabling people to express themselves in a different way , and to assist them in expressiveness . So I 'll give you a practical example . A lot of you probably use tools , slides , and things like that in your day to day job . PowerPoint was invented a long time ago and was really just about drawing things . You know , I 've got a 14 year old . And so if you look at the younger generation , if you look at what slides were , they were really tools to help people draw . And then to take what was on the slide projector and presented . Then the younger generation says , hey , I do n't want to draw things that 's really old-fashioned . I 'm going to go to the internet and copy images , right ? Because when they do class projects , they 're copying images into the slides . And then , as people observe , you know , on the social media environment , people going from text , which may have been Facebook to short images , which is Instagram to short video TikTok , we would say , hey , why would n't we be able to record short video ? And be used that as a mechanism to share . But recording short video is still capturing the real world through the lens of the camera . What people want is a more expressive way of saying , I have an idea , can I translate it ? And it may not be something I can capture . Imagine a kid in California and a school saying saying I want to capture how the landscape and outside of Paris and France is right now . I think they need to be able to generate some of the ideas that they could capture by physically being there . And so we 're working on all of this and we 're bringing some of these into our products to change what people could possibly do through the application of AI so they improve expressiveness for people . And so every boundary as the technology gets more sophisticated we think it moves from just assistance to assistance on things that human beings may not have been able to just linearly do to now things like expressiveness , which is a very different capability than people could actually do themselves . Yeah , I mean , all of this is very obviously incredibly exciting and we 're all watching it happen in real time . There 's an artist who actually described the image generation models as , he sort of image generation models as he was , he sort of said like , you kind of think about like a camera . Like it 's a new tool that allows you to create fundamentally new forms of art . That 's right . Yeah . And not just one medium of art , right ? Because if you look in the past , people said , you were a painter , you were a sculpture , you were a musician , and now these technologies allow you to blend all of it as a form of expressiveness . Yeah . You know , the last question I have for you is , you know , you obviously sit down with many of the sort of leading CEOs and business leaders of of the sort of largest organizations in the world . And I 'm sure one thing that is on many of their minds is sort of as AI technology develops and it continues to progress is potential disruption that might come from art of film intelligence . What sort of , how do you approach that conversation ? What sort of your advice to these business leaders who are looking at this powerful new technology and thinking about what that might mean for the businesses and the business landscape . When we talk to CEOs , I mean the biggest things we talk to them about number one , productivity in the long term , productivity has always been the primary driver of improving both company productivity , meaning their own companies , as well as societal benefit , things like affluence of a society , etc . And the means and equality of distribution of income to people across all spectrum society . Eventually , the most important metric , and you can look at any economic textbook is productivity . Software and technology has probably been the biggest boomer productivity over the last 30 , 40 years . This is the next evolution of that . And so we always say , if you approach it the right way , for example , labor shortages are going on right now . The biggest potential benefit is the application of some of these platforms like AI to do in that . The second , with any technological generation revolution , like artificial intelligence , but if you went back in time and looked at the industrial revolution , etc . There are always during the period of transition , anxiety about the consequences of that technology . And it does n't mean the technology by itself is good or bad . It 's the application of the technology that 's good or bad . So it 's incumbent upon both the technology providers and the users of the technology to ensure that the negative consequences of it are managed properly . Right ? The obvious example is , for instance , if you look at a very simple thing , image recognition . Image recognition can help doctors find tumors way better than having the best radiographer . It 's a system in that context and it 's like helping people with a better quality microscope than they had before . Object recognition is helping people find , for example , people who are in the ocean much more accurately so the coastguard can rescue them . At the same time , being able to use a camera and say that 's Thomas Korean has , you know , a lot of potential negative consequences . And so as a provider of technology , we at Google have chosen not to do that third part . But we also tell companies , it 's important not just to say , this is what 's regularly allowed by the legal framework , because law in many countries is not yet keeping up with how fast AI technology is moving . But to take the responsibility as a company CEO to say , here 's what I believe comfortable with , and here 's what I wo n't be comfortable with . Yeah . Well , Thomas , thank you so much for such incredible conversations . I think I 'm very heartened to hear all the incredible work that Google Cloud is doing to make artificial intelligence accessible to the entire business world and all of every enterprise around the globe . And I 'm so excited that you 're able to join us . Thank you so much . Thank you so much for having me . Thank you . Thank you . \ No newline at end of file diff --git a/server/trials/title_summary/vicuna.py b/server/trials/title_summary/vicuna.py deleted file mode 100644 index 588869c0..00000000 --- a/server/trials/title_summary/vicuna.py +++ /dev/null @@ -1,44 +0,0 @@ -from gpt4all import GPT4All - -model = GPT4All("/Users/gokulmohanarangan/Library/Application Support/nomic.ai/GPT4All/ggml-vicuna-13b-1.1-q4_2.bin") - -import spacy - - -def split_text_file(filename, token_count): - nlp = spacy.load('en_core_web_md') - - with open(filename, 'r') as file: - text = file.read() - - doc = nlp(text) - total_tokens = len(doc) - - parts = [] - start_index = 0 - - while start_index < total_tokens: - end_index = start_index + token_count - part_tokens = doc[start_index:end_index] - part = ' '.join(token.text for token in part_tokens) - parts.append(part) - start_index = end_index - - return parts - -parts = split_text_file("transcript.txt", 1800) -final_summary = [] -for part in parts: - prompt = f""" - ### Human: - Summarize the following text without missing any key points and action items. - - {part} - ### Assistant: - """ - output = model.generate(prompt) - final_summary.append(output) - - -with open("sum.txt", "w") as sum: - sum.write(" ".join(final_summary)) diff --git a/server/trials/whisper-jax/__init__.py b/server/trials/whisper-jax/__init__.py deleted file mode 100644 index e69de29b..00000000 diff --git a/server/trials/whisper-jax/agenda-headers.txt b/server/trials/whisper-jax/agenda-headers.txt deleted file mode 100644 index fd8034a2..00000000 --- a/server/trials/whisper-jax/agenda-headers.txt +++ /dev/null @@ -1,8 +0,0 @@ -AGENDA: Most important things to look for in a start up -TAM: Make sure the market is sufficiently large than once they win they can get rewarded -Product market fit: Being in a good market with a product than can satisfy that market -Unit economics: Profit for delivering all-in cost must be attractive (% or $ amount) -LTV CAC: Life-time value (revenue contribution) vs cost to acquire customer must be healthy -Churn: Fits into LTV, low churn leads to higher LTV and helps keep future CAC down -Business: Must have sufficient barriers to entry to ward off copy-cats once established -Founders: Must be religious about their product. Believe they will change the world against all odds. \ No newline at end of file diff --git a/server/trials/whisper-jax/whisjax.py b/server/trials/whisper-jax/whisjax.py deleted file mode 100644 index fb4f5e1f..00000000 --- a/server/trials/whisper-jax/whisjax.py +++ /dev/null @@ -1,183 +0,0 @@ -#!/usr/bin/env python3 - -# summarize https://www.youtube.com/watch?v=imzTxoEDH_g -# summarize https://www.sprocket.org/video/cheesemaking.mp4 summary.txt -# summarize podcast.mp3 summary.txt - -import argparse -import os -import re -import subprocess -import tempfile -from datetime import datetime -from urllib.parse import urlparse - -import jax.numpy as jnp -import moviepy.editor -import nltk -import yt_dlp as youtube_dl -from whisper_jax import FlaxWhisperPipline - -from ...utils.file_utils import download_files, upload_files -from ...utils.log_utils import LOGGER -from ...utils.run_utils import CONFIG -from ...utils.text_utils import post_process_transcription, summarize -from ...utils.viz_utils import create_talk_diff_scatter_viz, create_wordcloud - -nltk.download('punkt', quiet=True) -nltk.download('stopwords', quiet=True) - -WHISPER_MODEL_SIZE = CONFIG['WHISPER']["WHISPER_MODEL_SIZE"] -NOW = datetime.now() - -if not os.path.exists('../../artefacts'): - os.makedirs('../../artefacts') - - -def init_argparse() -> argparse.ArgumentParser: - """ - Parse the CLI arguments - :return: parser object - """ - parser = argparse.ArgumentParser( - usage="%(prog)s [OPTIONS] ", - description="Creates a transcript of a video or audio file, then" - " summarizes it using ChatGPT." - ) - - parser.add_argument("-l", "--language", - help="Language that the summary should be written in", - type=str, - default="english", - choices=['english', 'spanish', 'french', 'german', - 'romanian']) - parser.add_argument("location") - return parser - - -def main(): - parser = init_argparse() - args = parser.parse_args() - - # Parse the location string that was given to us, and figure out if it's a - # local file (audio or video), a YouTube URL, or a URL referencing an - # audio or video file. - url = urlparse(args.location) - - # S3 : Pull artefacts to S3 bucket ? - - media_file = "" - if url.scheme == 'http' or url.scheme == 'https': - # Check if we're being asked to retreive a YouTube URL, which is - # handled differently, as we'll use a secondary site to download - # the video first. - if re.search('youtube.com', url.netloc, re.IGNORECASE): - # Download the lowest resolution YouTube video - # (since we're just interested in the audio). - # It will be saved to the current directory. - LOGGER.info("Downloading YouTube video at url: " + args.location) - - # Create options for the download - ydl_opts = { - 'format': 'bestaudio/best', - 'postprocessors': [{ - 'key': 'FFmpegExtractAudio', - 'preferredcodec': 'mp3', - 'preferredquality': '192', - }], - 'outtmpl': './artefacts/audio', # Specify output file path and name - } - - # Download the audio - with youtube_dl.YoutubeDL(ydl_opts) as ydl: - ydl.download([args.location]) - media_file = "../artefacts/audio.mp3" - - LOGGER.info("Saved downloaded YouTube video to: " + media_file) - else: - # XXX - Download file using urllib, check if file is - # audio/video using python-magic - LOGGER.info(f"Downloading file at url: {args.location}") - LOGGER.info(" XXX - This method hasn't been implemented yet.") - elif url.scheme == '': - media_file = url.path - # If file is not present locally, take it from S3 bucket - if not os.path.exists(media_file): - download_files([media_file]) - - if media_file.endswith(".m4a"): - subprocess.run(["ffmpeg", "-i", media_file, f"./artefacts/{media_file}.mp4"]) - media_file = f"./artefacts/{media_file}.mp4" - else: - print("Unsupported URL scheme: " + url.scheme) - quit() - - # Handle video - if not media_file.endswith(".mp3"): - try: - video = moviepy.editor.VideoFileClip(media_file) - audio_filename = tempfile.NamedTemporaryFile(suffix=".mp3", - delete=False).name - video.audio.write_audiofile(audio_filename, logger=None) - LOGGER.info(f"Extracting audio to: {audio_filename}") - # Handle audio only file - except Exception: - audio = moviepy.editor.AudioFileClip(media_file) - audio_filename = tempfile.NamedTemporaryFile(suffix=".mp3", - delete=False).name - audio.write_audiofile(audio_filename, logger=None) - else: - audio_filename = media_file - - LOGGER.info("Finished extracting audio") - LOGGER.info("Transcribing") - # Convert the audio to text using the OpenAI Whisper model - pipeline = FlaxWhisperPipline("openai/whisper-" + WHISPER_MODEL_SIZE, - dtype=jnp.float16, - batch_size=16) - whisper_result = pipeline(audio_filename, return_timestamps=True) - LOGGER.info("Finished transcribing file") - - whisper_result = post_process_transcription(whisper_result) - - transcript_text = "" - for chunk in whisper_result["chunks"]: - transcript_text += chunk["text"] - - with open("./artefacts/transcript_" + NOW.strftime("%m-%d-%Y_%H:%M:%S") + - ".txt", "w") as transcript_file: - transcript_file.write(transcript_text) - - with open("./artefacts/transcript_with_timestamp_" + - NOW.strftime("%m-%d-%Y_%H:%M:%S") + ".txt", - "w") as transcript_file_timestamps: - transcript_file_timestamps.write(str(whisper_result)) - - LOGGER.info("Creating word cloud") - create_wordcloud(NOW) - - LOGGER.info("Performing talk-diff and talk-diff visualization") - create_talk_diff_scatter_viz(NOW) - - # S3 : Push artefacts to S3 bucket - prefix = "./artefacts/" - suffix = NOW.strftime("%m-%d-%Y_%H:%M:%S") - files_to_upload = [prefix + "transcript_" + suffix + ".txt", - prefix + "transcript_with_timestamp_" + suffix + ".txt", - prefix + "df_" + suffix + ".pkl", - prefix + "wordcloud_" + suffix + ".png", - prefix + "mappings_" + suffix + ".pkl", - prefix + "scatter_" + suffix + ".html"] - upload_files(files_to_upload) - - summarize(transcript_text, NOW, False, False) - - LOGGER.info("Summarization completed") - - # Summarization takes a lot of time, so do this separately at the end - files_to_upload = [prefix + "summary_" + suffix + ".txt"] - upload_files(files_to_upload) - - -if __name__ == "__main__": - main() diff --git a/server/trials/whisper-jax/whisjax_realtime.py b/server/trials/whisper-jax/whisjax_realtime.py deleted file mode 100644 index ec822854..00000000 --- a/server/trials/whisper-jax/whisjax_realtime.py +++ /dev/null @@ -1,151 +0,0 @@ -#!/usr/bin/env python3 - -import time -import wave -from datetime import datetime - -import jax.numpy as jnp -import pyaudio -from pynput import keyboard -from termcolor import colored -from whisper_jax import FlaxWhisperPipline - -from ...utils.file_utils import upload_files -from ...utils.log_utils import LOGGER -from ...utils.run_utils import CONFIG -from ...utils.text_utils import post_process_transcription, summarize -from ...utils.viz_utils import create_talk_diff_scatter_viz, create_wordcloud - -WHISPER_MODEL_SIZE = CONFIG['WHISPER']["WHISPER_MODEL_SIZE"] - -FRAMES_PER_BUFFER = 8000 -FORMAT = pyaudio.paInt16 -CHANNELS = 2 -RATE = 96000 -RECORD_SECONDS = 15 -NOW = datetime.now() - - -def main(): - p = pyaudio.PyAudio() - AUDIO_DEVICE_ID = -1 - for i in range(p.get_device_count()): - if p.get_device_info_by_index(i)["name"] == \ - CONFIG["AUDIO"]["BLACKHOLE_INPUT_AGGREGATOR_DEVICE_NAME"]: - AUDIO_DEVICE_ID = i - audio_devices = p.get_device_info_by_index(AUDIO_DEVICE_ID) - stream = p.open( - format=FORMAT, - channels=CHANNELS, - rate=RATE, - input=True, - frames_per_buffer=FRAMES_PER_BUFFER, - input_device_index=int(audio_devices['index']) - ) - - pipeline = FlaxWhisperPipline("openai/whisper-" + - CONFIG["WHISPER"]["WHISPER_REAL_TIME_MODEL_SIZE"], - dtype=jnp.float16, - batch_size=16) - - transcription = "" - - TEMP_AUDIO_FILE = "temp_audio.wav" - global proceed - proceed = True - - def on_press(key): - if key == keyboard.Key.esc: - global proceed - proceed = False - - transcript_with_timestamp = {"text": "", "chunks": []} - last_transcribed_time = 0.0 - - listener = keyboard.Listener(on_press=on_press) - listener.start() - print("Attempting real-time transcription.. Listening...") - - try: - while proceed: - frames = [] - start_time = time.time() - for i in range(0, int(RATE / FRAMES_PER_BUFFER * RECORD_SECONDS)): - data = stream.read(FRAMES_PER_BUFFER, - exception_on_overflow=False) - frames.append(data) - end_time = time.time() - - wf = wave.open(TEMP_AUDIO_FILE, 'wb') - wf.setnchannels(CHANNELS) - wf.setsampwidth(p.get_sample_size(FORMAT)) - wf.setframerate(RATE) - wf.writeframes(b''.join(frames)) - wf.close() - - whisper_result = pipeline(TEMP_AUDIO_FILE, return_timestamps=True) - timestamp = whisper_result["chunks"][0]["timestamp"] - start = timestamp[0] - end = timestamp[1] - if end is None: - end = start + 15.0 - duration = end - start - item = {'timestamp': (last_transcribed_time, - last_transcribed_time + duration), - 'text': whisper_result['text'], - 'stats': (str(end_time - start_time), str(duration)) - } - last_transcribed_time = last_transcribed_time + duration - transcript_with_timestamp["chunks"].append(item) - transcription += whisper_result['text'] - - print(colored("", "yellow")) - print(colored(whisper_result['text'], 'green')) - print(colored(" Recorded duration: " + - str(end_time - start_time) + - " | Transcribed duration: " + - str(duration), "yellow")) - - except Exception as exception: - print(str(exception)) - finally: - with open("real_time_transcript_" + NOW.strftime("%m-%d-%Y_%H:%M:%S") - + ".txt", "w", encoding="utf-8") as file: - file.write(transcription) - - with open("real_time_transcript_with_timestamp_" + - NOW.strftime("%m-%d-%Y_%H:%M:%S") + ".txt", "w", - encoding="utf-8") as file: - transcript_with_timestamp["text"] = transcription - file.write(str(transcript_with_timestamp)) - - transcript_with_timestamp = \ - post_process_transcription(transcript_with_timestamp) - - LOGGER.info("Creating word cloud") - create_wordcloud(NOW, True) - - LOGGER.info("Performing talk-diff and talk-diff visualization") - create_talk_diff_scatter_viz(NOW, True) - - # S3 : Push artefacts to S3 bucket - suffix = NOW.strftime("%m-%d-%Y_%H:%M:%S") - files_to_upload = ["real_time_transcript_" + suffix + ".txt", - "real_time_transcript_with_timestamp_" + suffix + ".txt", - "real_time_df_" + suffix + ".pkl", - "real_time_wordcloud_" + suffix + ".png", - "real_time_mappings_" + suffix + ".pkl", - "real_time_scatter_" + suffix + ".html"] - upload_files(files_to_upload) - - summarize(transcript_with_timestamp["text"], NOW, True, True) - - LOGGER.info("Summarization completed") - - # Summarization takes a lot of time, so do this separately at the end - files_to_upload = ["real_time_summary_" + suffix + ".txt"] - upload_files(files_to_upload) - - -if __name__ == "__main__": - main()