We are developing a data driven, land regeneration platfrom that leverages scalable, mass produced solar robots

Aigen

Whether you talk about the tech seeping through smartphones, cars having autonomous-driving features, or the tools being used by retailers to connect to their customers, AI can be said to have permeated human lives completely. While it has been around for decades now, it was with the release of ChatGPT in November 2022 that everyone finally woke up to it. Recent Generative AI applications, such as ChatGPT, Stable Diffusion, , and others have been the talk of the town for various reasons. The top one, certainly, is their broad utility. A very large chuck of the population can use these tools, for both personal and professional benefits. GitHub Copilot To grasp what the future holds, you must uncover the breakthroughs that have catalyzed the rise of Gen AI. To become a part of this revolution, or to make this revolution a part of your rise, an expert’s guide is needed. And here is one. What is Generative AI? It is an often asked question and here is an extremely simple definition for you.  Generative AI is a subdivision of machine learning that draws from leading techniques such as reinforcement learning and deep learning. It generates unique outputs, which can be text, music, video, images, etc. What is Generative AI? Gen AI can simply be defined as apps generally built by employing foundation models. The said models consist of expansive artificial neural networks (ANNs), which take inspiration from the billions of connected neurons in the human brain. These models are a segment of deep learning (DL), which refers to multiple deep layers found within neural networks. This is just the beginning of completely understanding gen AI, what it is, what are its true capabilities, and what the future holds. That said, it is a continuous learning process, especially with a technology like gen AI, which itself is still growing. Generative AI will sway all industry sectors in the coming years. High tech, life sciences & banking are expected to be the , in a positive sense. most impacted industries What is the Hype Around Generative AI? Do you remember having heard about Generative AI until a couple of years ago? Or even if you had, was it as omnipresent as it is today? I’m pretty sure the answer is no. The sudden hype around gen AI can, without any doubt, be accredited to the rise of and DALL-E. There are plenty of gen AI tools that are available today, some free to use, while some are paid. All of these have certainly made a lot of things easier for a lot of people, both personally and professionally. ChatGPT The curiosity around Gen AI has suddenly skyrocketed as it holds great potential to accelerate growth while reducing costs. It adds intelligence to data, making it easier for workers to gain a hold of previously unknown capabilities or knowledge. Generative AI Tools & Their Growing Role in Various Industries Generative AI tools have entered various industries and are judiciously stepping into roles traditionally played by humans. 1. Content Creation: Gen AI has become a favorite among writers, content creators, and marketers to streamline workflows and automate content generation, all in a much shorter time. Thus, leaving them with the time to focus on more creative and strategic tasks. The winner in this section is certainly ChatGPT. 2. IT & Software Development: A lot of change is happening around how code is written, especially since gen AI has entered the sphere. Developers can now generate code snippets, get solutions around coding challenges, and identify over 30% more defects. Leading gen AI tools for software development are GitHub Copilot, Bard, and AlphaCode. 3. Healthcare Industry: Gen AI has truly been a game-changer in the medical field, as it assists in diagnosing diseases, predicting outcomes, and analyzing medical images. Radiology has benefitted incredibly from Gen AI too. Be it drug discovery, clinical trials, or medical robots, a sharp impact has been witnessed. 4. Design & Art: Designers and artists are making the most of the impressive gen AI tools out there. Creating unique and unimaginable illustrations, designs, and visual art was never as easy as this. Midjourney and Jasper are text-to-art tools, wherein text input will result in creative output. 5. Banking & Finance: Financial institutions are making the most of advancements in Generative AI by leveraging it to predict stock movements, optimize trading strategies, and analyze market trends. Hence, a significant increase in trading profitability and better investment decisions. 6. Language Translation: People are stepping through the language barriers as many gen AI tools with the capability to translate speech and text in real-time have arrived. Seamless communications for global collaboration are no longer a distant dream. How to Learn Generative AI? When it comes to finding an answer to how to learn Generative AI from scratch, there are various things that you must consider. One of the key aspects is to judge how good your current knowledge is, and how much you still need to learn about the associated topics. Here are a few technologies, topics, tools, and aspects you need to get started in this field. 1. Learn the Fundamentals To learn anything, you must be clear about the fundamentals, and the same is true when it comes to figuring out how to learn Generative AI. Here are the key pointers to begin with. Linear algebra offers the key mathematical operations to work with tensors, matrices, vectors, and various other multi-dimensional arrays in ML. Probability accommodates techniques such as Bayesian statistics, which is employed to make probabilistic predictions and statistical distributions. and are key open-source libraries to develop and train ML models. TensorFlow PyTorch Calculus is employed to optimize ML algorithms. Gain an understanding of key concepts like curve fitting, multivariate calculus, derivatives, interaction, and gradients. Python is the widely preferred programming language for ML. This aspect includes control flow, tools like Pandas and NumPy, object-oriented design, and data structures. Statistics offers the techniques to analyze ML data. Hence, you must learn descriptive statistics, regression, statistical reasoning, inference, and significance testing. Understand Machine Learning 2. A subset of artificial intelligence, machine learning, or ML focuses on building systems with the ability to learn from data sets and improve with time. The top ML approaches to learn about are. Supervised Learning means training of the algorithm on labeled data sets, offering instances of the input and the intended output. The famous supervised learning algorithms incorporate decision trees, support vector machines, linear regression, k-nearest neighbors, and logistic regression. In addition, can help in both AI/ML. Cybersecurity skills Unsupervised Learning means analyzing data sets with just the inputs and no labeled responses. The aim is to uncover interest structure and relationships within the data. Artificial Neural Networks are inspired by biological neural networks and are made up of multiple layers of interconnected nodes like neurons. This knowledge is important to ensure fluent working. Reinforcement Learning means training algorithms for maximizing the trial-and-error interactions with the environment they are in. Training the data sets is not required as the agent learns by achieving their goals by getting feedback on their actions about which one is the better one. 3. Convolutional Neural Networks (CNNs) A specialized kind of neural network, CNNs are apt for image processing. Convolutional layers are utilized to preserve spatial relationships within pixels. It is followed by pooling layers to further reduce dimensionality. It is important to learn about CNN architectures such as YOLO, ResNet, and VGG for Generative image modelling. 4. Generative Models Generative models refer to a class of ML algorithms with the ability to generate new data samples, such as audio, text, or images. There are two popular Generative models. Generative Adversarial Networks or GANs are composed of a discriminator neural network and a generator neural network, competing against one another. The task of the generator is to create synthetic data samples, whereas the task of the discriminator is to distinguish between the real and the fake one. You must gain knowledge of Sequence GANs for text generation, WaveGANs for audio generation, and Deep Convolutional GANs for image generation. Variational Autoencoders are generative models that use autoencoding to learn latent representations of data. It comprises an encoder network and a decoder network. The former compresses data into a latent space vector, while the latter reconstructs data from the latent vector. You should learn key VAE architectures like Recurrent VAEs, VQ-VAEs, and Convolutional VAEs. 5. Master Associated Tools & Libraries There are a plethora of tools and libraries, which require your proficiency in them. Here are some of the most influential tools and libraries associated with generative AI to look into. TensorFlow is a popular, open-source ML library, offering a comprehensive framework to build and train different generative models. It is a preferred library for deep learning tasks, such as text generation, style transfer, and image generation. Keras is a neural network API running atop TensorFlow and other backends. Both experienced and beginner developers prefer it, as it simplifies building and training DL models. Hugging Face Transformers is a widely preferred library as it facilitates working with natural language processing (NLP) models, such as BERT and GPT-2. GANLib is a popular Python library, which has specifically been designed for Generative Adversarial Networks (GANs). OpenAI Gym, an open-source toolkit, is designed to both develop and compare reinforcement learning algorithms. PyTorch is a prominent DL library that provides dynamic computation graphs, highly preferred by researchers due to its flexibility. 6. Take Generative AI Courses Taking Generative AI courses is one of the best ways to get started on your journey to how to learn Generative AI from scratch. Enrolling in Generative AI training online will help you learn from those who have many years of experience in related fields. While you can get started with Generative AI tutorials, online videos, and blogs, nothing can compare to the effectiveness of a Generative AI certification program. 7. Get Hands-on Experience Once you decide to make a career in this field or expand the horizons of your career with Generative AI, then you must get your hands engaged in gaining practical experience. Here are a few ways to get started. Complete online projects, courses, and tutorials Recreate projects and seminal papers Participate in GitHub projects and other competitions Become a part of the community Build a portfolio Attend seminars and sessions 8. Stay Up-To-Date Nothing you ever learn in association with the tech world will remain the same over time. Hence, you have to constantly read, work, and practice to stay updated. Read studies and papers Attend meetups and conferences Follow forums and blogs Listen to videos and podcasts Conclusion Understanding Generative AI is an endless sea to search and learn. The more you read about machine learning, , the more curious you will get. Hence, the best way to learn Generative AI is with the aid of experts, who will be able to guide you in the best way possible. Follow this expert guide to become a pro in this emerging and growing technology. artificial intelligence, and robotics

How to 'Learn' Generative AI

Welcome to HackerNoon’s interview series. This section is curated by our Editorial team to spotlight bold and disruptive startups across the globe. If you’ve been nominated, and answer these questions Startups-Of-The-Year create a brand profile here. Hey Hackers, Vectara has been nominated in HackerNoon's annual Startup of the Year awards in Palo Alto, USA. Please : ]. vote for us here [https://hackernoon.com/startups/north-america/north-america-palo-alto-ca-usa#ljpyx4tp/ljpyx4tp Read more about us below to understand why we deserve your vote. Meet Vectara Vectara is an end-to-end platform for embedding powerful generative AI features into applications with extraordinary results. Built on a solid hybrid-search core, Vectara delivers the shortest path to a correct answer/action through a safe, secure, and trusted entry point. Vectara never trains on your data, allowing businesses to embed generative AI capabilities without the risk of data or privacy violations. My Role My name is Sean Anderson, and I am Head of Product Marketing at Vectara. I am excited to chat with developers daily who are making generative AI a reality. I want the world to know more about how Vectara solves the hallucination problem with LLMs and makes generative AI a reality for every business. How We're Disrupting/Improving the Generative AI Industry Companies need to act fast on their generative AI (Gen AI) strategies in response to ever-growing data volumes and the monumental shift in user expectations brought on by the advent of tools like ChatGPT. NLUs and large Language Models (LLMs) hold the promise of delivering business efficiencies, increasing profits, and decreasing costs. But Gen AI introduces operational complexities that need careful attention. CIOs, Product leads, and IT leaders find it hard to scale AI projects because they require deep ML expertise and maintenance of complicated LLMs, and even if they use open-source, the infrastructure is expensive to operate. There is a risk in assembling the components manually. The stakes run high because Gen AI also includes serious risks related to hallucination and data security. Enterprises that invest wisely in AI and LLMs will gain a competitive edge. Those who don’t will experience the costs, complications, and long lead times endemic to unwise DIY decisions. That competitive edge requires a proper workflow integration between internal data systems, existing solutions, expertise, and the right tools. Companies struggle to recruit, build, and implement next-generation generative AI techniques due to inexperience in data science and machine learning. The market is saturated with build-it-yourself LLM technology but many organizations can't adopt the risk and do not know where to start. Organizations need a safe entry point and end-to-end platforms to address the full spectrum of needs for Generative AI and hybrid search. Developers need to be able to easily embed Generative AI into their applications without the risk of exposing data (through sending off-site or being used as training data for the model). Standing Out From The Crowd Vectara provides an end-to-end SaaS solution abstracting the complex ML Operations pipeline (Extract, Encode, Index, Retrieve, Re-Rank, Summarize). Vectara’s “Grounded Generation” allows businesses to quickly, safely, and affordably integrate best-in-class conversational AI and question-answering into their application with zero-shot precision. Vectara never trains on your data, allowing businesses to embed generative AI capabilities without the risk of data or privacy violations. Vectara is built for product managers and developers with an easily leveraged API that gives full access to the platform's powerful features. Because Vectara indexes your own data, the risk of hallucination is all but eliminated, which is transferred and stored in encrypted form, and is never used to train Vectara models. Vectara is an end-to-end platform that handles indexing, encoding, evaluation, ranking, and summarization. Product developers can quickly build end-to-end generative solutions without maintaining and integrating individual components. Our Predictions/Thoughts on the Generative AI Industry in 2023 Since the launch of ChatGPT last year, the urgency around leveraging AI has never been greater. Many companies understand AI's value but don't fully understand the technology. Organizations seek trusted advisors to make their AI implementation successful and tools and platforms that simplify AI adoption by providing a safe entry point. These partnerships remove many barriers to adoption, including privacy and accuracy concerns. Companies struggle to recruit, build, and implement next-generation generative AI techniques due to inexperience in data science and machine learning. The market is saturated with build-it-yourself LLM technology, but many organizations can't adopt the risk and do not know where to start. Organizations need a safe entry point and end-to-end platforms to address the full spectrum of needs for Generative AI and hybrid search. Developers need to be able to easily embed Generative AI into their applications without the risk of exposing data (through sending off-site or being used as training data for the model). What word defines the state of Generative AI in 2023? At the beginning of 2023, the Vectara leadership team made some predictions of how AI will proliferate into our daily lives. Having experienced the explosion of generative AI and its feverish evolution, we know that time will continue to test these assumptions. Here are some areas we feel will advance the state of Generative AI in 2023. NLP will become so mainstream and part of everything we do, just like SW ate the world. The leaders in all segments will be applying NLP/LLM/GenerativeAI to jump their competition. The web will have its own layer of AI-powered additional information (similar to how we think about Augmented Reality); every major site will be layered with additional NLP-generated information, delivered based on user behavior. “Did you mean?”, “Check out these recommendations” and “Related Categories and Content” will become standard user interaction models in many more applications. AI-powered automation will begin to disrupt key software markets including document management, RPA, customer service, cybersecurity, legal contracts, and more. In 2 years, 80% or 90% of the world’s languages will be translated almost perfectly, allowing a regional language with few writers or speakers to be understood by someone else using a different regional language with few writers or speakers (the long tail) Why we decided to participate in HackerNoon's Startup of the Year awards We are excited and honored to participate in HackerNoon’s Startup of the Year awards and want to personally thank everyone who participates in the voting process for us and for startups in general. Vectara is a startup’s startup, founded by people like Amr, who started Cloudera and navigated through the emergence of big data and machine learning. The founders truly believe that AI can advance the state of our culture as long as the AI has all the relevance to make smart decisions. As a startup, we are uniquely positioned in the generative AI landscape to listen to our customers to shape our roadmap and features inside our platform. The HackerNoon Startup of the Year represents a stable of extremely talented startups, and our inclusion is a memorable milestone for the company. Today is another eventful day for us as we are launching our embedding model, Boomerang. We want to thank the HackerNoon community and everyone who took the time to vote for any startup. Final Thoughts 2023 very much feels like the year of emergence is generative AI. Encapsulating that interest with communities like the HackerNoon community and acknowledgment in these voting processes. Vectara is still a very young company in a fastly changing market. We look forward to working closely with early adopters of generative AI to understand the challenges they face. We are an open and inclusive community for developers of all experience levels. We invite you to continue the conversation with us here and on all our online platforms. [ ] Vote for us today! https://hackernoon.com/startups/north-america/north-america-palo-alto-ca-usa#ljpyx4tp/ljpyx4tp

Startups of the Year 2023: Vectara - The Platform for Embedding Powerful Generative AI Features

Last week, the that is testing an AI tool that enables news organizations to create content using AI. It is unclear what the exact capabilities of the product are but based on one comment from Google, it helps with tasks like headlines and modifying writing styles. Amidst fears of what impact this could have on journalist jobs, Google quickly put out a : New York Times reported Google clarifying statement Our goal is to give journalists the choice of using these emerging technologies in a way that enhances their work and productivity. Quite simply these tools are not intended to, and cannot, replace the essential role journalists have in reporting, creating, and fact-checking their articles. This has sparked a heated conversation about the future of news with generative AI. The companies making AI products argue that this technology will enable media organizations and journalists to be more effective, while critics contend that it could potentially harm journalist jobs, increase misinformation, and flood the market with low-quality AI-generated content. In this piece, we unpack the by exploring a few things: impact of generative AI on the future of news Types of news articles today by complexity The business of news The anatomy of producing a news article What generative AI can and cannot do well Generative AI and the Future of News — A tale of two worlds Types of news articles today by complexity There are several types of news content on the internet, each with varying levels of complexity in their production. The complexity typically arises from factors like timeliness, the amount of research required, and the story being told. The low-complexity articles include: Factual/data news (eg. article listing mortgage rates in San Francisco, article with numbers from a company’s earnings call) — these are relatively straightforward with minimal subjective opinions or perspectives. Interest-based/informational news (eg. summer recipes published on NY Times Cooking, an article explaining what the Fed interest rate means) — there is some creativity in selecting topics but the focus is more informational and catering to specific interests. Breaking news (eg. an article about a CEO resigning, an article describing an active weather event) — these are typically short articles about a rapidly unfolding event, with limited initial information and emerging facts. The include: high-complexity articles News coverage (with context, research, and facts) — these provide a more detailed explanation of news backed by research, additional context, and often interviews with people, thereby requiring more time and effort to produce; they are also extensively fact-checked. Interpretive news — this includes opinion pieces, op-eds, and analyses (like this one) that provide interpretations/perspectives/opinions on current issues; they are often subjective and require extensive research to back up the perspectives. Feature pieces — these are typically deep dives on topics that might not really be hot news right now but important issues, investigative journalism would fall in this category; these require extensive research and interviews spanning across months, and also creative storytelling. I bring up this categorization because the purpose behind and process involved in producing these two types of articles is very different, and consequently will evolve differently with the use of generative AI. With this in mind, let’s talk about how news makes money. The business of news The business of news is tricky — most news organizations operate on an ad-supported media model, and a small subset have successfully managed to pivot to subscriptions. This has major implications for the type of content a news organization produces. The businesses (NYTimes notably now gets from subscriptions) have a content strategy that’s tied directly to consumer value — focus on to make the subscription worthwhile. For example, of the types of content above, NYT primarily does news coverage (that is well-researched), interpretive news, and feature pieces for their core news offering, backed up by interest-based content like and . Some other publications like the Wall Street Journal and the Washington Post have made strides with subscriptions using a similar approach. subscription-based news ~70% of revenue high-quality content + diverse interest-based content NYT Cooking Wirecutter However, and will continue to be so for the foreseeable future. This means that they are focused on generating high engagement: more eyeballs → more ad inventory → more revenue. The strategy that has been more effective for them is (like news coverage, interpretive news) (like factual/data news, interest-based/informational news). most publications are still ad-supported augmenting high-complexity content with a ton of high volume, low-complexity content This strategy works because high-complexity content helps deliver long-term value to consumers, while the ton of low-complexity content helps . The SEO loop looks something like: high volume of content that gets clicks → search engines think your content is valuable → all of your content gets ranked higher → more eyeballs. get eyeballs as well as win the SEO game This is no criticism of the strategy and the need to play the SEO game is a necessity for ad-supported media. The reality is that — the internet disrupted how content was created and distributed (mostly through Google Search / today), and news organizations have not recovered from the impact of that disruption. There are attempts to fix the imbalance, which are still evolving — Canada/Australia passed laws that force Google/Meta to pay what’s essentially a “disruptor tax” to support good journalism, some organizations like NPR are partially supported by , and some newspapers like the Washington Post are . news media is a terrible business Meta federal funding subsidized by (generally well-meaning) billionaires All this to say — news media/journalism is an absolutely essential public utility for a well-functioning democracy but a not-so-great business. Therefore, the more self-reliant these news businesses can be (without having to depend on regulation or billionaires), the more effective they can be towards achieving their mission. Generative AI can’t solve all of these structural challenges (especially content distribution) but it can definitely make content creation more efficient without necessarily compromising on quality. The anatomy of producing a news article To understand what part of news production will generative AI have the most impact on, it’s helpful to understand the different steps involved in getting a news article published. We can break down the effort into a few sequential components: Research — finding a topic, gathering facts/data, talking to the right people for quotes or perspectives Storytelling — making sense of the research to come up with a compelling storyline Writing — turning the storyline into actual words that make an article, buttoning up the storyline, backing it up with the right data / adding supporting links Editing — improving the writing, coming up with catchier headlines, having better visuals, and fact-checking to meet standards Distribution — publishing, and distributing across channels like social media. Based on some research and also influenced by my writing experience, I estimate the effort of producing an article into something like: . Take the actual numbers with a grain of salt but they are directionally accurate. Research (30%), Storytelling (20%), Writing (20%), Editing (20%), and Distribution (10%) Now, note that — the low-complexity pieces might not need to rigorously go through all steps in the life cycle above but the high-complexity pieces do. not all formats of news articles require all of the steps As an example, this article took me ~8–9 hours to finish (so did most other articles on ), and I would like to think that my articles fall under interpretive news. If I wrote a low-complexity piece, I could probably get it done in ~2 hours. Another data point — says he can typically get a news article done in a couple of hours while a feature piece might take up to 6 months. my Substack newsletter this NYT reporter So, what can generative AI actually help with? It basically comes down to what steps in the news production process can current (and future) products do well. What generative AI can and cannot do well Here’s my take on each step. Spoiler — they are somewhat effective for writing and editing, they can be effective for research if the right products are built, and they will continue to be bad at storytelling. Research — surprisingly low capability Most current generative AI products (like ChatGPT, and Google Bard) have surprisingly low capability when it comes to Research: There are definitely certain capabilities that they are good at. For example, they are good at giving arguments for a particular point of view or providing inspiration for new topics. They are average-to-good at summarizing content and in particular, answering questions based on reading an article. For example, you can ask them to read this article and list out different steps involved in news production. However, they often produce (“hallucinations”), to what they claim is the truth, and have questionably sourced data with potential copyright infringement. See the screenshot below. factually incorrect information do not provide links/sources For a generic use case (say you want to draft an email to a future LinkedIn connection), these issues don’t matter. But when you are writing a news article, not fact-checking can severely impair your brand perception. Though the research tools can be used for directionally useful research (for example, you want to know how much market share Bing and Yahoo search have), writers/journalists still have to do additional work to find a new source to link because current tools neither provide links nor have fully accurate data. for anyone building a research product for writers, and it’s very likely that new companies will emerge here. Using cleanly sourced data + being able to provide answers with trustworthy links is a big opportunity Storytelling — low capability Generative AI’s storytelling capabilities are fairly weak today. Here’s an example: I did a bunch of research for this article, put together the research in an organized format, and asked ChatGPT to give me a storyline. See the screenshot below for results. At first glance, it looks like they “make sense”. But that’s literally all it is — it makes sense at a surface level. In reality, none of these storylines are very compelling and the research notes I provided had much more detail and nuance beyond the surface-level results provided here. You can make the argument that these are still decent, but in the context of writing an article, this was completely useless to me — at best, this was inspiration/ideas, not a storyline based on the research I submitted. If you have a low-complexity story, it will get the job done for you. If you are building a nuanced story, or have the data and need help constructing a story, current products do not get the job done. Given the high amount of subjectivity involved, I am not bullish it will get better and this will continue to be the part where writers can add the most value. Writing — medium capability If you provide current products with a detailed storyline of what you want to say, it can generate a barely decent v1 of the content. The output is still pretty cookie-cutter and it is pretty challenging to instruct the current models to craft the language in a way that it tells a story. See the screenshot for an example output after I fed ChatGPT detailed notes about the storyline for this article. At first glance, you might think it looks okay for a v1 draft. It is not — the tone isn’t right, the story doesn’t flow and it always looks like it’s been written by a bot. It’s very generic and does not articulate the story despite being fed a very specific narrative. If I published that article, you would not read it. And that’s the challenge with the writing capability today — it can work for low-complexity articles but for anything more complex, you essentially have to rewrite the whole draft. — writers don’t want to take terrible cookie-cutter versions and redo the whole thing line by line. What they would like is some form of a user interaction construct that lets a writer feed in a storyline and sequentially craft the article section by section, while giving active feedback to the AI tool. Cherry on top would be being able to customize the writing style by feeding past articles written by the same writer. The big product unlock here would be enabling effective human instruction I think the underlying models today do have the capability to do this and there is , which I believe will happen near-term. innovation needed at the UI layer Editing — medium capability Tools today have a good amount of capability to review articles, find errors and make corrections. These tools are also very good at cosmetic tasks, like coming up with ideas for catchy headlines or section titles that might perform well. However, there is still work needed at the UI layer to make this usable for editing — there are some partial solutions today like which lets you improve language and make sentences shorter / longer from within a Notion page, but it does not capture the full context of the page. ChatGPT does a good job of making edits across a full article but lacks the ability to take instructions to easily edit specific sections and also cannot support links (i.e. I give it a blurb with text that has hyperlinks, I get back text without any links). Notion AI I absolutely hate the editing part of the writing process, and I am sure several writers and journalists do as well — generative AI can most definitely minimize some of this editing grunt work near-term. Distribution — medium capability There are tools coming up here today that for example help you generate social snippets, or identify parts of your story that might go more viral. This will likely continue getting better in the future. Generative AI and the future of News — A tale of two worlds Based on the above analysis, you can see a clear duality emerging: , which require simpler research & storytelling, and can be written & edited easily, will start to be (or heavily AI-assisted). Low-complexity formats more and more AI-generated , which require more complex research and storytelling capabilities that don’t exist today, will but generative AI can bring a good amount of High-complexity formats continue to be primarily created by journalists efficiency by minimizing writing and editing gruntwork. While AI-generated low-complexity content seems bad at first glance, these articles are primarily written for SEO purposes or to augment existing high-quality content, and the race to commoditization had started much before the generative AI wave. For example, the Associated Press has been using since 2014. The upside here — this is not the kind of content journalists want to spend time creating and automating this will free up their time for high-complexity content. bots to publish articles reporting company earnings More and more high-complexity content will emerge. Research and storytelling capabilities with products today are limited, which means that the ability to . This, accelerated by AI writing and editing tools that take the grunt work out of publishing quality content, will be a big boon for journalists. construct a compelling story based on information and story-tell in a unique way will continue to be the biggest currency journalists have What about some of the concerns raised around using AI for news? Some are fair but I believe they are mostly solvable: A wave of junk SEO content — Google has that they won’t penalize AI-generated content and have gotten criticism for opening up a world of junk content, including factually incorrect articles. That’s a somewhat fair critique but I believe Google will crack down on this — not out of the goodness of their hearts but because the crux of a search product is to give users useful results. Google already has penalties for SEO hacking practices (like keyword stuffing, and link farming) and it would be easy to extend this framework to AI content. taken the stance Journalists will lose jobs — Some have brought up the concern that journalists might lose jobs, or be in a less advantageous position (like writers in Hollywood who are currently on strike); the big difference here is that while there are a large number of writers in Hollywood, newsroom employment in the US since 2008, not because we don’t need journalists but because news is a bad business; Generative AI can help fix the economics of the business while continuing to empower journalists do what they love. has fallen 26% Conclusion I am by no means an AI maximalist and I absolutely think there are real risks to AI that need to be addressed as we scale the technology. However, I think it’s important to analyze every AI-impacted market/situation separately and not bundle them into one big AI-impact problem. In the case of news, generative AI can tremendously improve the economics of news businesses. The products are not there today — there is a clear need for AI products that thoughtfully solve writers’ needs without a brute-force, chat-based, language model interface, but I’m confident these will emerge near-term. Low-complexity writing will be more and more AI generated and that’s okay — it can help companies run an efficient SEO machine, while journalists (assisted by AI for grunt work) bring to life a lot more high complexity content that elevates the public discourse. Thank you for reading! Also published here.

Exploring the Impact of Generative AI on the Future of News 

খবরের ভবিষ্যৎতে জেনারেটিভ এআই-এর প্রভাব অন্বেষণ করা

Erkundung der Auswirkungen generativer KI auf die Zukunft der Nachrichten

Explorando el impacto de la IA generativa en el futuro de las noticias

Explorer l'impact de l'IA générative sur l'avenir des actualités

समाचार के भविष्य पर जेनरेटिव एआई के प्रभाव की खोज

ニュースの未来に対する生成 AI の影響を探る

뉴스의 미래에 대한 생성적 AI의 영향 탐구

Explorando o impacto da IA generativa no futuro das notícias

Üretken Yapay Zekanın Haberin Geleceği Üzerindeki Etkisini Keşfetmek

Khám phá tác động của AI sáng tạo đối với tương lai của tin tức

探索生成人工智能对新闻未来的影响

Cloud computing is a dynamic and ever-evolving field, and keeping up with its advancements can be challenging. However, with the emergence of Generative AI (gen AI) tools, such as ChatGPT or Bard, and others we now have a game-changing tool that can propel your cloud computing learning experience to new heights. Gen AI tools are an invaluable resource for individuals preparing for cloud certification exams, solutions architects designing cloud solutions, software engineers building cloud applications, teachers guiding students, and AI/ML enthusiasts exploring the potential of cloud technologies. This blog will walk you through the 10 transformative ways in which you can harness the power of generative AI to accelerate your journey in mastering cloud computing. As a bonus, each sections comes with example prompts that you can use to start with your learning instantly. While generative AI tools can be valuable resources for learning, it is important to exercise caution and not rely solely on their responses. Occasionally, these tools may yield incorrect answers. Therefore, it is crucial to double-check and validate your learnings using official documentation and trusted sources. DISCLAIMER: 1) Simplifying Concepts Gen AI can break down complex cloud computing concepts into digestible and understandable explanations. Through interactive conversations and illustrative examples, gen AI tools can demystify intricate topics, making them more accessible to learners. Example prompts - Explain bastion host in networking to child Explain to me how the content delivery network works with examples 2) Summarizing Blogs and Whitepapers Gen AI can save you time by providing concise summaries of insightful blogs and comprehensive whitepapers. By distilling the key takeaways and condensing lengthy texts, generative AI empowers you to grasp the essence of cloud computing documentation efficiently. Example prompts - Summarize the blog  Google Cloud brings generative AI to developers, businesses, and governments Give me the top 3 takeaways from the blog Summarize the blog  Google Cloud brings generative AI to developers, businesses, and governments Summarize the whitepaper - Big Data Analytics Options on AWS Top 5 takeaways from the whitepaper - Big Data Analytics Options on AWS 3) Comparing 2 products Uncover the power of gen AI in comparing and contrasting different cloud products. Gen AI tools can provide you with a side-by-side comparison of two products, including their features, use cases, and performance. This can be helpful when preparing for cloud certifications. Example prompts - Difference between BigTable and BigQuery in tabular format Give me 5 key differences between EC2 and Lambda along with use cases 4) Exploring Real-World Applications of Cloud Products: Gen AI can be used to learn more about cloud products and discover how customers use them in real-world scenarios. This can be helpful when trying to understand how a particular cloud product can solve your business problems. Gen AI can provide you with customer case studies, white papers, and other resources that shed light on specific cloud products. Example prompts - Tell me what is BigQuery and give me 3 customer references who use it and how they benefit from it Could you please explain what BigQuery is and share three customer examples who are using it? I'd love to learn more about how these customers benefit from using BigQuery in their operations. 5) Solutioning design questions Gen AI can provide a solutions complex design challenges. Whether you're seeking guidance on designing real-time data processing systems or designing solutions for your business, gen AI tools can provide step-by-step solution with cloud product recommendations. The solution serves a good starting point to think about your design. Example prompts - You are Cloud Architect - Design a real-time data processing system in GCP You are Cloud Architect - How to set up monitoring for Dataproc in GCP 6) Answering coding questions Gen AI can be a valuable resource for answering coding questions related to cloud computing. It is important to be mindful that as the coding problems become more complex, there is a possibility of the AI "hallucinating" and generating inaccurate code. Example prompts - Write a step-by-step function in Python to create Cloud Functions that read from a Pub/Sub topic and write data to BigQuery Write step-by-step code in Terraform to create Python Cloud Functions that read from a Pub/Sub topic and write data to BigQuery 7) Code conversions Gen AI can assist developers in translating code from one programming language to another. It can save time and effort in the development process, helping you translate more complex code effectively. Always double-check the generated code, especially if it will be deployed in a production environment. Tip: Example prompts - Convert the following code to Python - cloud pubsub subscriptions create sub_one --topic=hello_topic Convert the following code to Python - public class HelloWorld {
    public static void main(String[] args) {
        System.out.println("Hello, World!");
    }
} 8) Creating certification/study roadmaps Gen AI can create personalized study roadmaps for mastering cloud computing. Whether you aim to become a certified cloud engineer or specialize in specific cloud platforms, gen AI can generate tailored study plans complete with recommended resources and learning materials. Use the gen AI created plan and tweak it based on your availability and schedule. Tip: Example prompts - I want to prepare for the Cloud Engineer role. Create a roadmap to learn GCP Cloud Computing in 3 months with a weekly study plan. Please provide a link to resources and YouTube videos. I want to prepare for the GCP Professional  Data Engineer certification. Create a roadmap to learn GCP Cloud in 3 months with a weekly study plan with links to resources and YouTube videos. 9) Debugging errors When encountering errors in your code, gen AI can offer solutions and resolutions. By analyzing error messages and providing troubleshooting strategies, gen AI tools offer timely and effective resolutions, saving you valuable time and effort. Example prompts - Why am I getting the following error in BigQuery "Resources exceeded during query execution: Your project or organization exceeded the maximum disk and memory limit available for shuffle operations". Tell me how to resolve it. 10) Get Books and blogs recommendations Unlock a wealth of learning resources recommended by gen AI tools. From blogs on cloud computing advancements to highly-rated books on cloud architecture and implementation, gen AI can provide curated recommendations that align with your specific interests and learning goals. Example prompts - Give me blogs recommendation for learning Data warehousing in GCP. Include links. Give me highly-rated book recommendations for learning Data warehousing in GCP. Include links. Conclusion Gen AI is a powerful technology that can expedite your cloud computing learning journey. By utilizing prompts and examples provided, you can leverage gen AI tools like ChatGPT or Bard to enhance your cloud computing learning experience today. Remember to exercise caution, validate information from official documentation and trusted sources, and make the most of generative AI's capabilities. Happy Cloud learning! 😃

How to Leverage ChatGPT/Bard to Learn Cloud Computing Fast!

In June, the young family of generative AI (gen AI) unicorns suddenly got a new member. Inflection AI ripped the chart suggested by CBInsights just a month before and got onto the third position, dethroning Cohere AI. So we have to oblige, and after covering two Gen AI leaders OpenAI and Anthropic, we dive into the precocious startup, Inflection AI. Having introduced its core product, an AI assistant named Pi, in May and published a memo about Inflection-1, a Large Language Model ( ) behind Pi in June, Inflection AI took the industry by storm the same month with an astonishing $1.3 billion in funding and a $4 billion valuation – all within its first two years of existence! LLM The company's sudden surge owes much to the attraction of high-profile investors, including technology luminaries such as Bill Gates, Eric Schmidt, and Reid Hoffman (also a co-founder), along with tech giants and . These financial ties extend beyond mere investments. Microsoft is providing its cloud computing infrastructure to strengthen Inflection's capabilities, while Nvidia, in partnership with CoreWeave, is integrating 20,000 of its H100 GPUs for LLM training. Microsoft Nvidia Let's dive deeper and discover more about the founders, their vision, the development behind the model powering Pi, and other crucial details. Finding them was not that easy! Content Overview The starting point of Inflection Financial and legal situation Mission of Inflection and its founders First product Tech behind Pi (LLM development facts) How Inflection is going to make money? Founders’ attitude toward AI risks A few not widely known facts about Inflection's founders 1. The starting point of Inflection According to the official information, Inflection AI has three founders: Karén Simonyan (Chief Scientist and Co-founder) Reid Hoffman (Co-founder) Mustafa Suleyman (CEO and Co-founder) , Inflection hasn't been very transparent with independent media. Some information available that extends beyond mere press releases or PR work stems from the podcast hosted by Reid Hoffman for Greylock Partners (a VC fund where Hoffman has been a partner since 2009). In January 2022, announcing Mustafa Suleyman as a new partner in Greylock Partners, Reid Hoffman : “I’ve known Mustafa personally for more than a decade. We first met at a pub in London. What struck me most was our inspiring and provocative discussion on the power and potential of AI to help solve some of humanity’s most urgent challenges. It marked the beginning of a deep and lasting friendship between Mustafa and I.” Operating with a low profile said More than a decade ago from 2022 was about a time when Mustafa Suleyman (along with Demis Hassabis and Shane Legg) co-founded DeepMind, one of the first AI research labs. In 2014, it was acquired by Google ( ). In January 2022, Suleyman left Google, joined Greylock Partners, and now the main AI R&D is led by the new structure called Google DeepMind in March 2022, Inflection was born (incubated by Greylock). Located in Palo Alto, California. Introducing Inflection AI on the Greylock website, Suleyman called it an "AI-first consumer products company." However, at that moment he didn't clarify exactly what that meant. 2. Financial situation and legal situation In May 2022, in equity financing, though the source of the capital wasn't immediately clear as Inflection didn't respond to the request for more information. Subsequently, when Inflection announced the $1.3 billion funding round in 2023, they clarified in that the first round was raised from Greylock, Microsoft, Reid Hoffman, Bill Gates, Eric Schmidt, Mike Schroepfer, Demis Hassabis, Will.i.am, Horizons Ventures, and Dragoneer. TechCrunch that Inflection had secured $225 million reported the official press release At some point, the information on Inflection's website changed to describe the company as an "AI studio creating a personal AI for everyone." They chose the same legal structure as did Anthropic: , a new type of corporation to balance the interests of shareholders, materially affected individuals, and the public good. public benefit corporation 3. Mission of Inflection and its founders Introducing Inflection, as always, on the Greylock podcast, , "Our mission was to build a tool that is designed to understand the user, rather than the other way around. We wanted it – whether that be connecting with other human beings, making decisions, figuring out how to navigate a problem at work, or more.” Reid Hoffman said to be more than an answerer of questions but a participant in the dialogue that helps you along your life's path Throughout other interviews about Inflection and its first product Pi, they used words such as 'reliable,' 'ever-present,' 'non-judgmental,' 'kind,' and 'supportive’. To get to the point where a chatbot can be all this sounded like a mission. 4. First product In May 2022, Inflection AI shipped its first core product – Pi, a chatbot with the primary goal to serve as a 'Personal AI'. It’s said that Pi uses emotional intelligence to aim for human-like conversations. While its use case is narrower than other services like ChatGPT, Google’s Bard, or Microsoft’s Bing chatbot, it strives to mimic human conversation patterns. Pi's training incorporates data up to November 2022, and it is committed to user privacy, refraining from using identifiable information in training. WhatsApp, Instagram PMs, Facebook Messenger, iMessage, or via the website – the ways to talk to Pi. The main difference from the most commonly used LLM products is the Pi remembers communications with you. The chatbot is also made to resist starting to flirt with a user. Named for "personal intelligence", Pi functions in several roles like teacher, coach, and confidante, designed to align with individual interests and needs. 5. Tech behind Pi Not much is known about the technology behind Pi. However, a month after its launch, the company revealing its foundation on the proprietary LLM, Inflection-1. According to the website, Pi doesn't utilize any third-party language model APIs. released a memo In the memo, the company stated: "The Inflection-1 architecture, dataset, and training procedure are proprietary, and we omit their details in this memo." What we do know is that Inflection-1 demonstrated impressive performance on the Massive Multitask Language Understanding (MMLU) benchmark, surpassing other well-known LLMs such as Meta's LLaMA, OpenAI's GPT 3.5, and Google's PaLM 540B, particularly in answering trivia questions. Additional information from Inflection's partner, CoreWeave's official website, that Inflection AI utilized 3,500 NVIDIA H100 Tensor Core GPUs and a cluster comprising over 40,000 cables and 500 miles of InfiniBand fiber cables for MLPerf Training benchmark competition submission, achieving an impressive 11-minute training benchmark. confirms , that Inflection AI is working on a cutting-edge supercomputer equipped with 22,000 NVIDIA H100 GPUs that is expected to be one of the largest In the industry, right behind AMD's Frontier. However, there's some debate regarding the true computational power of this setup. While Inflection AI claims that these GPUs could provide 22 exaflops in 16-bit precision mode, suggest that the actual performance might not meet these expectations. There are rumors calculations by The Next Platform Though the model is proprietary, the references that are mentioned in the do not list any in-house research papers. official memo 6. How Inflection is going to make money? There are many unknowns at the moment. The developers are exploring different monetization strategies, including the possibility of adding subscription options. Another option is to keep it free but use ads to earn revenue. Currently, Inflection AI is a free platform available to everyone worldwide. On our request to learn more about the business model, neither Reid Hoffman nor Mustafa Suleyman replied. In another podcast with Reid Hoffman, Mustafa Suleyman stated, “ And likewise, we sort of don’t want to drive the kind of addictive dopamine hits of outrage or anger or despair or viral misinformation. These kinds of things have really incentivized people to spend longer on the platform, and sometimes that’s because they’re getting paid for it. We are very, very conscious that this is a moment that needs to change with AI, and that new business models need to emerge to address this." It’s really important to us that we figure out a monetization mechanic that doesn’t incentivize us to keep you engaged or maybe distracted on the platform for as long as possible. 7. Founders’ attitude toward AI risks Reid Hoffman intentionally chose to be positive about AI. , “So in recent months, He has done so in blog posts, television interviews, and fireside chats. He has spoken to government officials around the world. He hosts three podcasts and a YouTube channel. And in March, he published a book, “Impromptu,” co-written with the A.I. tool . The New York Times reports Mr. Hoffman has engaged in an aggressive thought-leadership regimen to extol the virtues of A.I. GPT-4 “I’m beating the positive drum very loudly, and I’m doing so deliberately,” he said. The 55-year-old sits on the boards of 11 tech companies including , which has gone all in on AI, and eight nonprofits. His venture capital firm, Greylock Partners, has backed at least 37 AI companies.” Microsoft , Reid Hoffman said that the , He thinks that since technology 'always puts tools in the hands of good and bad actors,' throughout our history, we have learned how to figure this out. In the same interview, he said, ' . In an interview to Washington Post conversations about superintelligence divert us from the real risk which is AI in the hands of bad actors. Part of the reason why I beat the drum so firmly in the positive camp is not that I'm risk unaware, and I think we need to navigate them, but oh, my gosh, every month we delay an AI tutor and doctor, for everyone who has a smartphone or has a friend who has a smartphone, just think about the human suffering that is caused by that delay, and that is super important This story was at Turing Post. Consider to receive new episodes straight to your inbox. originally published ing subscrib 8. A few not widely known facts about Inflection Founders Reid Hoffman Reid Hoffman ( ) is among a small group of interconnected tech executives leading the A.I. charge, many of whom also led the last internet boom. on Twitter He is part of the "PayPal Mafia," a group of former PayPal employees who went on to become successful entrepreneurs and investors in the tech industry. Reid Hoffman's entrepreneurial journey has been marked by resilience and determination. After his failure with SocialNet, he became obsessed with scaling and invested heavily in the social networking phenomenon, which eventually led to the co-founding of LinkedIn, the world's largest professional network with a diversified revenue model that includes subscriptions, advertising, and software licensing. He also led the company to its successful acquisition by Microsoft. One of the first investors in OpenAI, he was sitting on its board until March 2023, stepping down to avoid any conflicts of interest with Inflection AI. He still sits on the Board of directors at Microsoft. Reid Hoffman is a proponent of effective altruism (founders of OpenAI and Anthropic have strong ties with EA also), which advocates for using wealth and resources to create the best for the most people. In 2017, he : “If you have wealth, and at least a reasonable moral compass, to ask, “How do you create the most good for the most people?” I’m not saying this is the perfect system. But what’s the better system?” said to the Atlantic Reid Hoffman is a prominent philanthropist and has invested in impact investments through charitable entities. over $1.5 billion Hoffman is an avid fan of Dungeons & Dragons, and he attributes its influence to his thinking about life as a heroic quest with various players coming together to accomplish something. He believes in the power of friendship and considers it the true center of his life. Reid views friendship as a skill that can be learned and improved through asking questions and interacting with others. Mustafa Suleyman ( ) on Twitter He dropped out of Oxford University at 19 to start the Muslim Youth Helpline, a telephone counseling service for Muslims in the UK. He worked as a policy officer on human rights for Ken Livingstone, the Mayor of London, before founding Reos Partners, a ‘systemic change’ consultancy that uses methods from conflict resolution to navigate social problems. In February 2016, Suleyman launched DeepMind Health at the Royal Society of Medicine, a project that builds clinician-led technology for the NHS and other partners to improve frontline healthcare services He received the Commander of the Most Excellent Order of the British Empire (CBE) in 2019 for his influence in the UK technology sector and also accepted the Silicon Valley Visionary Award the same year Within DeepMind, he established the DeepMind Ethics & Society research unit, aiming to study the real-world implications of AI and promote ethical practices in the technology industry. One of the founding members of the Partnership on AI — an organization set up in September 2016 to ensure that AI is developed safely, ethically, and transparently — along with Facebook's AI head Yann LeCun, Microsoft Research director Eric Horvitz, and several others. Suleyman recently proposed a modern Turing test, , “to measure whether a machine has achieved ACI, you give an AI $100,000 and see if it can turn the seed investment into $1 million. To do so, the bot must research an e-commerce business opportunity, generate blueprints for a product, find a manufacturer on a site like Alibaba and then sell the item (complete with a written listing description) on Amazon or Walmart.com.” Suleyman expects AI will pass this more practical threshold sometime in the next two years. according to Bloomberg Karén Simonyan There is much less information about Karén Simonyan. Karén is not active on social media, he has no Twitter, a barely filled , and a well-developed . LinkedIn account Google Scholar page Education & Career History: 🔬 Chief Scientist, Inflection AI (2022 – Present) 🔬 Research Scientist, DeepMind (2014 – 2022) 🔬 Postdoc, University of Oxford (2013 – 2014) 🔬 Ph.D. student, University of Oxford (2009 – 2013) While doing his postdoc in 2013, in Oxford, Karén Simonyan along with Andrew Zisserman created the VGGNet image processing framework. VGGNet won the prestigious ImageNet Challenge. To commercialize it, they created a company Vision Factory AI, which was soon acquired by a not-very-known at that time startup DeepMind Technologies, where in 2014 Simonyan became a Principal Research Scientist. There, he played a key role in various AI breakthroughs, such as AlphaZero, AlphaFold, WaveNet, BigGAN, and Flamingo. While in DeepMind, Karen authored and co-authored numerous publications, including five papers in Nature and Science, accumulating an impressive 180,000 citations. Since he co-founded Inflection AI, only one paper was published with Karén Simonyan as a co-author . HiP: Hierarchical Perceiver That’s it for now. The information about Inflection AI will be constantly updated… Thank you for reading, please feel free to share with your friends and colleagues. This story was at Turing Post. Consider subscribing to receive new episodes straight to your inbox. originally published

The Inflection AI Story: How a Secretive Unicorn Reached a $4 Billion Valuation

transformer-based language model definitely lives up to the hype. Following the natural evolution of , this generative language model drew a lot of attention by and appearing in the online text adventure game . OpenAI’s GPT-2 Artificial Intelligence (AI) engaging in interviews AI Dungeon GPT-2, built on transformer decoder blocks, essentially leverages its to train and formulate word sequences based on what it has already learned. Its ability to adapt and generate human-like text interactions with minimal prompts quickly is exceptional. natural language prediction algorithm So it’s no surprise that it’s been deployed across applications like chatbots (of course), translation platforms, (which shows its impressive cognitive abilities), , , and more. TL;DR generators music generators poetry generators Although more complex Natural Language Processing (NLP) models like and were released recently, GPT-2 continues to reign supreme. One of the primary reasons for this is its ability to run efficiently on everyday hardware. For example, the sheer number of trainable parameters, 1.5 billion to be exact, only demands 6.5 GB storage. GPT-3 Google T5 So at , we decided to put it to the test and see what else we could train it to do. Someone suggested playing a game of chess, and we ran with it. Intersog Machine Learning (ML) researcher already trained . The model is clear evidence of its ability to recognize known patterns within the game. Shawn Presser GPT-2 to play chess using Portable Game Notation (PGN) files We decided to take it a step further by training the GPT-2 model on the current board state rather than PGN sequences. This approach ensures that it can play the game based on the present without going through gaming history to predict the next best move. The code you’ll see in the steps below is inspired by and trained on . Professor Blank’s Programming a Chess Player Presser’s Cryochess – GPT-2 1.5B chess engine Step 1. Set up Your Dependencies To play chess with smart algorithms, you need a powerful ML-ready machine with a robust Graphics Processing Unit (GPU). Install , , and . It’s best to engage in this activity in a with a installed. CUDA 10.1 PyTorch TensorFlow 2 virtual environment JupyterLab extension Install python-chess [4] and aitextgen [5] modules: In [ ]: !pip install python-chess !pip install aitextgen !pip install tqdm Download PGN files to the PGN folder. You also have the option of converting databases (*.sg4) to PGN format. For this exercise, we engaged in training by leveraging 100,000 games in PGN archives. SCID Some valuable PGN resources include the following: PGN ChessBook PGN Mentor The Week in Chess import os if not os.path.exists("pgn"): os.mkdir("pgn") 2. Generate Training Data As reflected in the example below, our model uses the current board state along with the training file for the next move on each line. It follows the following format: [Result] FEN-position-and-side-only - next_move [1-0] r1bq3k/ppp2rpp/5b2/3n4/3P4/P4p2/BP1B1PPP/R2QR1K1 w - a2d5 [1-0] 2bQ4/p4kb1/6n1/q1p1p3/1rn1P3/N3BP2/1PP5/2KR2R1 w - a3c4 [0-1] 1r3rk1/p4nbp/1qppb1p1/4p3/PP2P3/4NN1P/2QB1PP1/2R1R1K1 b - f8c8 We decided to use only chess matches that were won and skipped draws. The code to generate this function is as follows: import os from tqdm.auto import tqdm import glob import chess.pgn MAX_IMPORT = 100000 def importPgn(filename, s, max_import): counter = 0 total = 0 with open(filename) as f: for line in f: if "[Result" in line: total += 1 if total > max_import: total = max_import pbar = tqdm(total=total, desc="read " + filename, unit=" games", mininterval=1) pgn = open(filename) while counter < max_import: game = chess.pgn.read_game(pgn) if not game: break board = game.board() moves = game.mainline_moves() count = sum(1 for _ in moves) # skip unfinished games if count <= 5: continue result = game.headers["Result"] # import only resultative games if result != "1-0" and result != "0-1": continue for move in moves: if board.turn == chess.WHITE and result == "1-0": line = ( "[1-0] " + " ".join(board.fen().split(" ", 2)[:2]) + " - " + move.uci() ).strip() s.add(line) elif board.turn == chess.BLACK and result == "0-1": line = ( "[0-1] " + " ".join(board.fen().split(" ", 2)[:2]) + " - " + move.uci() ).strip() s.add(line) board.push(move) counter += 1 pbar.update(1) pbar.close() return counter def convert(): games = 0 moves = 0 max_import = MAX_IMPORT s = set() # load previous state if os.path.exists("fen.txt"): with open("fen.txt") as f: for line in tqdm(f, desc="read fen.txt", unit=" moves", mininterval=1): if line: s.add(line) max_import -= 1 if max_import <= 0: break for file in glob.glob("pgn/*.pgn"): count = importPgn(file, s, max_import) games += count max_import -= count if max_import <= 0: break with open("fen.txt", "w") as f: for line in tqdm(s, desc="write fen.txt", unit=" moves", mininterval=1): f.write(line + "\n") moves += 1 print("imported " + str(games) + " games, " + str(moves) + " moves") convert() It took us about 15 minutes to import 100,000 games. If you want to use more games for training, you’ll also need more RAM. Step 4: Training the GPT-2 Model As described in the , we trained a small GPT-2 model from scratch using only the model memory. We chose a small model as we could quickly train it on basic/average hardware (rather than larger models). aitextgen documentation Larger models come with their own sets of demands and benefits, but it’s far too complex for a simple demonstration. The training function can run many times, repeat it, and archive acceptable loss as a model checkpoint. As model checkpoints are periodically saved, the training function can be run multiple times, repeat the training, and achieve acceptable losses. In this scenario, to save time, we stopped at a loss value close to 0.8. Even at this level, the GPT-2 model could predict moves with an acceptable level of accuracy. To better fit with your GPU and negate out of memory scenarios, tune-up batch_size, and num_workers. from aitextgen import aitextgen from aitextgen.utils import build_gpt2_config from aitextgen.TokenDataset import TokenDataset from aitextgen.tokenizers import train_tokenizer import os file_name = "fen.txt" model_dir = "trained_model" config_file = os.path.join(model_dir, "config.json") pytorch_model_file = os.path.join(model_dir, "pytorch_model.bin") vocab_file = os.path.join(model_dir, "aitextgen-vocab.json") merges_file = os.path.join(model_dir, "aitextgen-merges.txt") dataset_cache_file = os.path.join(model_dir, "dataset_cache.tar.gz") max_length = 100 vocab_size = 10000 def train(): if not os.path.exists(model_dir): os.mkdir(model_dir) # train tokenizer if necessary if not os.path.exists(vocab_file): print("training tokenizer, please wait...") train_tokenizer(file_name, save_path=model_dir, vocab_size=vocab_size) if os.path.exists(dataset_cache_file): # use cache data = TokenDataset(dataset_cache_file,vocab_file=vocab_file,merges_file=merges_file,block_size=max_length,from_cache=True,) else: # or create token cache if necessary data = TokenDataset(file_name,vocab_file=vocab_file,merges_file=merges_file,block_size=max_length,line_by_line=True,save_cache=True,cache_destination=dataset_cache_file) if not os.path.exists(pytorch_model_file): config = build_gpt2_config(vocab_size=vocab_size,max_length=max_length,dropout=0.0,n_embd=512,n_head=16,n_layer=16,) ai = aitextgen(config=config, vocab_file=vocab_file, merges_file=merges_file, to_gpu=True) else: ai = aitextgen(model=pytorch_model_file, config=config_file, vocab_file=vocab_file, merges_file=merges_file, to_gpu=True) ai.train(data, num_steps=150000, generate_every=1000, save_every=1000, learning_rate=1e-4, batch_size=16, num_workers=4,) train() This process takes about eight hours. However, if you want to use a well-trained model, it’s best to give it a few days. Step 5: Assessment 5.1. Introduce a Random Player (random_player) In this scenario, a random player is a simple (or novice) player who’s pretty poor at the game of chess. The function basically makes an arbitrary choice based on a list of valid moves. import random def random_player(board): move = random.choice(list(board.legal_moves)) return move.uci(), False, False 5.2. Introduce a GPT-2 Player (gpt2_player) This player uses AI-powered GPT-2 to predict the next move. The model prompt is constructed from the expected result from the current board state and side (and since we want to win, white = 1-0 and black = 0-1). From here, once prompted the model supplements the next generated move. The GPT-2 player is trained on a small ML dataset and can never evolve into a chess master. The predicted moves are based on unknown board states and not ones presented during the training phase. Whenever the model generates an invalid movement, it’s fixed by using a random valid move. import os from aitextgen import aitextgen from aitextgen.utils import build_gpt2_config import chess from tqdm.auto import tqdm model_dir = "trained_model" vocab_file = "aitextgen-vocab.json" merges_file = "aitextgen-merges.txt" max_length = 100 model_dir = "trained_model" config_file = os.path.join(model_dir, "config.json") pytorch_model_file = os.path.join(model_dir, "pytorch_model.bin") vocab_file = os.path.join(model_dir, "aitextgen-vocab.json") merges_file = os.path.join(model_dir, "aitextgen-merges.txt") dataset_cache_file = os.path.join(model_dir, "dataset_cache.tar.gz") max_length = 100 ai = aitextgen( model=pytorch_model_file, config=config_file, vocab_file=vocab_file, merges_file=merges_file, from_cache=True, to_gpu=True, # to_fp16=True ) # a set to find known states db = set() with open("fen.txt") as f: for line in tqdm(f, desc="read fen.txt", unit=" moves"): if line: db.add(" ".join(line.split(" ", 3)[:3])) def gpt2_player(board): if board.turn == chess.WHITE: prompt = "[1-0] " + " ".join(board.fen().split(" ", 2)[:2]) else: prompt = "[0-1] " + " ".join(board.fen().split(" ", 2)[:2]) isKnown = prompt in db prediction = ai.generate_one(prompt=prompt,max_length=max_length,temperature=0.9,top_k=0,) isPredicted = False try: uci = prediction.split(' - ')[1].strip() move = chess.Move.from_uci(uci) isPredicted = True except Exception as e: # print(str(e)) move = None if not move or move not in board.legal_moves: # give up and do random move move = random.choice(list(board.legal_moves)) isPredicted = False return move.uci(), isPredicted, isKnown Step 6: Play a Game of Chess Now comes the fun part! This function demands two players to play chess against each other: import time from IPython.display import display, HTML, clear_output import chess def who(player): return "White" if player == chess.WHITE else "Black" def display_board(board, use_svg): if use_svg: return board._repr_svg_() else: return "<pre>" + str(board) + "</pre>" def play_game(player1, player2, visual="svg", pause=0.1): """ playerN1, player2: functions that takes board, return uci move visual: "simple" | "svg" | None """ use_svg = (visual == "svg") board = chess.Board() known1 = 0 predicted1 = 0 total1 = 0 known2 = 0 predicted2 = 0 total2 = 0 if visual is not None: display(display_board(board, visual == 'svg')) try: while not board.is_game_over(claim_draw=True): if board.turn == chess.WHITE: uci, isPredicted, isKnown = player1(board) total1 += 1 if isKnown: known1 += 1 if isPredicted: predicted1 += 1 else: uci, isPredicted, isKnown = player2(board) total2 += 1 if isKnown: known2 += 1 if isPredicted: predicted2 += 1 name = who(board.turn) board.push_uci(uci) board_stop = display_board(board, use_svg) html = "<b>Move %s %s, Play '%s':</b><br/>%s<br/>Known/Predicted/Total moves: %s/%s/%s %s%% - %s/%s/%s %s%%" % ( len(board.move_stack), name, uci, board_stop, known1, predicted1, total1, round(predicted1 / (total1 or 1) * 100), known2, predicted2, total2, round(predicted2 / (total2 or 1) * 100)) if visual is not None: if visual == "svg": clear_output(wait=True) display(HTML(html)) if visual == "svg": time.sleep(pause) except KeyboardInterrupt: msg = "Game interrupted!" return (None, msg, board) result = "1/2-1/2" if board.is_checkmate(): msg = "checkmate: " + who(not board.turn) + " wins!" result = "1-0" if who(not board.turn) == "White" else "0-1" elif board.is_stalemate(): msg = "draw: stalemate" elif board.is_fivefold_repetition(): msg = "draw: 5-fold repetition" elif board.is_insufficient_material(): msg = "draw: insufficient material" elif board.can_claim_draw(): msg = "draw: claim" if visual is not None: print(msg) return (result, msg, board) Let’s play the gpt2_player vs. random_player: play_game(gpt2_player, random_player) pass Move 61 White, Play 'd2d7': Known/Predicted/Total moves: 2/29/31 94% - 0/0/30 0% checkmate: White wins! When you're working with a small dataset, these chess matches often end up in a stalemate. While we aren't a hundred percent certain, it could also result from not analyzing the next move and choosing the best one. Now let's play 100 games where the gpt2_player plays white: from tqdm.auto import tqdm plays = 100 white_wins = 0 black_wins = 0 pbar1 = None pbar2 = None for i in tqdm(range(plays), desc="Plays"): if not pbar1: pbar1 = tqdm(total=plays, desc="White wins") if not pbar2: pbar2 = tqdm(total=plays, desc="Black wins") result, _, _ = play_game(gpt2_player, random_player, visual=None) if result is None: break elif result == "1-0": white_wins += 1 pbar1.update(1) elif result == "0-1": black_wins += 1 pbar2.update(1) pbar1.close() pbar2.close() print("Final score: %s-%s" % (white_wins, black_wins)) Final score: 52-0 Most often, the GPT-2 controlled player wins the game or draws. In this scenario, the GPT-2 controlled white player won more than half the games. We also noticed that the current board state was almost always new to the model. The model was also able to engage in more valid moves than fails. So we can conclude that the GPT-2 model was able to learn some basic patterns from the training data to predict the next move successfully. Let’s play the gpt2_player vs. Human Player: The following function effectively handles human input into the game: def human_player(board): uci = get_move("%s's move [q to quit]> " % who(board.turn)) legal_uci_moves = [move.uci() for move in board.legal_moves] while uci not in legal_uci_moves: print("Legal moves: " + (",".join(sorted(legal_uci_moves)))) uci = get_move("%s's move[q to quit]> " % who(board.turn)) return uci, True, False def get_move(prompt): uci = input(prompt) if uci and uci[0] == "q": raise KeyboardInterrupt() try: chess.Move.from_uci(uci) except: uci = None return uci Do you want to play against gpt2_player? If you do, note that you must enter your move in the user-computer interface representing each piece and place. This means that typing something like "a2a4"moves the piece located at a2 to a4. play_game(human_player, gpt2_player) pass Move 10 Black, Play 'b7b6': Known/Predicted/Total moves: 0/5/5 100% - 2/5/5 100% Conclusion In this experiment, we deployed GPT-2 to learn and play a game of chess. Although it was nowhere near our grandmasters (at least not yet), it showed that it was capable of understanding the game's basics. With more training data and larger model size, the model can theoretically take it to the next level and possibly beat human players. An interesting observation was that the model itself behaved quite erratically when playing the random player who made arbitrary moves. However, when we added a human player, the model made more calculated and confident moves. This suggests that a better player creates board states that are similar to the training data. We also noticed that the GPT-2 model engaged in natural text generation and confidently generated different types of textual patterns. It was also able to leverage its training data to successfully handle unknown input, almost like it developed a new internal algorithm. If you like to do your own experiments, this notebook is available . HERE

How to Play Chess Using a GPT-2 Model

Generative Artificial Intelligence for Software Engineering: Research Agenda

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Kiriaki Frangias;
(2) Andrew Lin;
(3) Ellen Vitercik;
(4) Manolis Zampetakis. Table of Links Abstract and Introduction Warm-up: Agents with known equal disutility Agents with unknown disutilites Experiments Conclusions and future directions, References A Summary of notation B Omitted proofs from Section 2 C Omitted proofs from Section 3 D Additional Information about Experiments 3 Agents with unknown disutilites We now significantly expand our model to study heterogeneous agents with differing, unknown disutilities. The full proofs of all results in this section are in Appendix C. 3.1 Model and notation Some agents might experience high disutility when exerting effort, so it may not be in the principal’s best interest to incentivize all agents to exert effort. Therefore, the principal must set the payment so that the number of agents incentivized is utility maximizing for the principal. For the rest of this section, we denote by g the number of agents that the principal aims to incentivize; in Theorem 3.6, we describe an efficient optimization problem that solves for the optimal value of g, denoted g∗. The principal controls the number of incentivized agents only through the payment function. Hence, the realized number of agents that exert effort, denoted ˆg, is a random variable with a distribution that depends on the payment. Under this model, the principal will: 1. Calculate g∗, the optimal number of agents to incentivize. 2. Announce the payment pg∗ ∈ R+ (based on g∗). 3. Verify a subset of comparisons assigned to the agents. 4. Assign each comparison to a total of r agents. 5. Pay pg∗ to all agents not identified as “bad.” 3.2 Payment function 3.3 Correctness of CrowdSort Next, we bound the redundancy required to ensure that CrowdSort returns the groundtruth ordering. Proof. We know from Lemma 3.4 that if then with high probability, the algorithm CrowdSort(T, s, v, r, δ) identifies all “bad” agents. Therefore, it suffices to assign each comparison to at least one “good” agent for the true value of the comparison to be returned. By Remark 3.3, we know that for any agent ai, So the probability that all r agents assigned to a single comparison are “bad” is at most: Taking the union bound over all comparisons, we have that the probability there exists a comparison such that all agents assigned to it are “bad” is at most Upper-bounding the above probability by δ/2, we get that: Finally, notice that the above inequality and Inequality (6) hold simultaneously with probability at least 1 − δ/2. 3.4 Principal’s incentive We combine the results presented above in the following theorem, in which we additionally prove that it is possible for the principal to efficiently compute the optimal number of agents to incentivize. Based on Lemma 3.5, we use the notation and This paper is available on Arxiv under CC 4.0 license. This paper is available on arxiv under CC 4.0 license. Authors: (1) Kiriaki Frangias; (2) Andrew Lin; (3) Ellen Vitercik; (4) Manolis Zampetakis. This paper is available on arxiv under CC 4.0 license. Authors: Authors: (1) Kiriaki Frangias; (2) Andrew Lin; (3) Ellen Vitercik; (4) Manolis Zampetakis. Table of Links Abstract and Introduction Abstract and Introduction Warm-up: Agents with known equal disutility Warm-up: Agents with known equal disutility Agents with unknown disutilites Agents with unknown disutilites Experiments Experiments Conclusions and future directions, References Conclusions and future directions, References A Summary of notation A Summary of notation B Omitted proofs from Section 2 B Omitted proofs from Section 2 C Omitted proofs from Section 3 C Omitted proofs from Section 3 D Additional Information about Experiments D Additional Information about Experiments 3 Agents with unknown disutilites We now significantly expand our model to study heterogeneous agents with differing, unknown disutilities. The full proofs of all results in this section are in Appendix C. 3.1 Model and notation Some agents might experience high disutility when exerting effort, so it may not be in the principal’s best interest to incentivize all agents to exert effort. Therefore, the principal must set the payment so that the number of agents incentivized is utility maximizing for the principal. For the rest of this section, we denote by g the number of agents that the principal aims to incentivize; in Theorem 3.6, we describe an efficient optimization problem that solves for the optimal value of g, denoted g∗. The principal controls the number of incentivized agents only through the payment function. Hence, the realized number of agents that exert effort, denoted ˆg, is a random variable with a distribution that depends on the payment. Under this model, the principal will: 1. Calculate g∗, the optimal number of agents to incentivize. 2. Announce the payment pg∗ ∈ R+ (based on g∗). 3. Verify a subset of comparisons assigned to the agents. 4. Assign each comparison to a total of r agents. 5. Pay pg∗ to all agents not identified as “bad.” 3.2 Payment function 3.3 Correctness of CrowdSort Next, we bound the redundancy required to ensure that CrowdSort returns the groundtruth ordering. Proof. We know from Lemma 3.4 that if then with high probability, the algorithm CrowdSort(T, s, v, r, δ) identifies all “bad” agents. Therefore, it suffices to assign each comparison to at least one “good” agent for the true value of the comparison to be returned. By Remark 3.3, we know that for any agent ai, So the probability that all r agents assigned to a single comparison are “bad” is at most: r Taking the union bound over all comparisons, we have that the probability there exists a comparison such that all agents assigned to it are “bad” is at most Upper-bounding the above probability by δ/2, we get that: Finally, notice that the above inequality and Inequality (6) hold simultaneously with probability at least 1 − δ/2. 3.4 Principal’s incentive We combine the results presented above in the following theorem, in which we additionally prove that it is possible for the principal to efficiently compute the optimal number of agents to incentivize. Based on Lemma 3.5, we use the notation and This paper is available on Arxiv under CC 4.0 license. This paper is available on Arxiv under CC 4.0 license. Arxiv Arxiv

AIGEN

aigen.io

-- employees

Since n.d.

COMPANY RANKING

AIGEN

EVERGREEN INDEX #1564

Representing this company? Click here to claim and customize this page!

Aigen WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

AIGEN

aigen.io

-- employees

Since n.d.

COMPANY RANKING

AIGEN

EVERGREEN INDEX #1564

RELATED COMPANIESRANK

STORIES ABOUT AIGEN

Representing this company? Click here to claim and customize this page!

Aigen WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!