ChatGPT Enterprise🚀, OpenAI guide for teachers📚, The GPU-Poors💻, Autonomous Agents survey🤖, future of generative AI🌀, and How Trustworthy Are Large Language Models?🔍
AI Connections #34 - a weekly newsletter about interesting blog posts, articles, videos, and podcast episodes about AI
JOIN 🪐
Cohere for AI: Aya - An Open Science Initiative to Accelerate Multilingual AI Progress: HERE
Contribute 500 submissions in any language and get Cohere for AI swag 🚀
READ 📚
“ChatGPT Enterprise” blog post by OpenAI: READ ChatGPT Enterprise plan offers enterprise-grade security and privacy, unlimited higher-speed GPT-4 access, longer context windows for processing longer inputs, advanced data analysis capabilities, customization options, and much more.
“Teaching with AI” blog post by OpenAI: READ OpenAI is releasing a guide for teachers using ChatGPT in their classroom—including suggested prompts, an explanation of how ChatGPT works and its limitations, the efficacy of AI detectors, and bias.
“Identifying AI-generated images with SynthID” blog post by DeepMind: READ This blog post introduces SynthID, a new tool developed in collaboration between Google Cloud and Imagen for watermarking and identifying AI-generated images. SynthID embeds a digital watermark directly into an image's pixels, ensuring it remains undetectable to the human eye but can be recognized for validation. The tool aims to combat the challenges posed by realistic AI-generated content in spreading misinformation, ensuring users can confidently identify and work with such content responsibly.
“Google Gemini Eats The World – Gemini Smashes GPT-4 By 5X, The GPU-Poors” blog post by SemiAnalysis: READ This blog post discusses the advancements in AI models, highlighting Google's MEENA model which briefly outperformed OpenAI's GPT-2. The article emphasizes the divide between GPU-rich and GPU-poor entities, with Nvidia emerging as a dominant force in the GPU market, overshadowing other major players like HuggingFace and Databricks. Amidst Nvidia's dominance, the post ends by hinting at a potential entity that might challenge or change this prevailing landscape.
“What’s the future of generative AI? An early view in 15 charts” article by McKinsey: READ
“The latest canvas for Refik Anadol’s AI-generated art? The new Sphere in Las Vegas” article by Los Angeles Times: READ
“How Trustworthy Are Large Language Models Like GPT?” article by Stanford University: READ This article is about the potential pitfalls of relying on Generative AI, specifically GPT models, for critical applications. In collaboration with other institutions, Researchers Sanmi Koyejo and Bo Li examined the trustworthiness of GPT-3.5 and GPT-4, assessing them on eight trust parameters such as toxicity, bias, and privacy. While they acknowledged improvements in newer models, they discovered vulnerabilities like the potential for toxic outputs, biases, and privacy leaks and hence advised users to exercise caution and skepticism when using such models, especially in sensitive areas.
“Supporting the Open Source AI Community” blog post by a16z: READ
“Generative AI and intellectual property” blog post by Ben Evans: READ This blog post delves into the complexities of intellectual property in the context of generative AI. The author highlights how AI, like the upcoming smartphone apps that can imitate famous singers' voices, challenges our traditional understandings of copyright, especially when AI is tasked with creating content "in the style of" existing artists. Furthermore, the article raises questions about whether training AI on vast amounts of data is a form of intellectual property theft, the implications of AI-generated content, and the future of authenticity in art and music in an age where AI can replicate or mimic human creativity.
“Large language models aren’t people. Let’s stop testing them as if they were.” article by MIT: READ
READ (RESEARCH PAPERS) 📚
“A Survey on Large Language Model based Autonomous Agents” research paper by Gaoling School of Artificial Intelligence: READ The research paper provides an in-depth review of autonomous agents, emphasizing the recent advancements made possible through large language models (LLMs) that aim to achieve human-level intelligence. The paper introduces a unified framework that captures most of the existing work on LLM-based agent architectures and outlines their applications across various domains, including social science, natural science, and engineering. Furthermore, the paper discusses evaluation methods for these agents and suggests challenges and potential avenues for future research in the domain.
“Graph of Thoughts: Solving Elaborate Problems with Large Language Models” research paper by ETH Zurich: READ The paper introduces the Graph of Thoughts (GoT) framework, which enhances the prompting capabilities in large language models (LLMs). GoT represents information from LLMs as a graph, where each piece of information is a vertex and the relationships between them are edges, allowing for more intricate and interconnected thought processes. The GoT method outperforms existing models in tasks like sorting, and its design is reminiscent of human thought processes and neural networks.
“Nougat: Neural Optical Understanding for Academic Documents” research paper by Meta AI: READ The research paper introduces "Nougat," a Visual Transformer model designed for Optical Character Recognition (OCR) specifically tailored for scientific documents, converting them into a markup language. Traditional PDFs often lose semantic details, especially in mathematical expressions, hindering digital accessibility. Nougat aims to enhance this accessibility, transforming human-readable documents into machine-readable formats, with the authors releasing their models and code for further advancements in the field.
WATCH🎥
LEARN 📚
“Anti-hype LLM reading list” guide by veekaybee: LEARN
“LLM Learning Lab” guide by Lightning AI: LEARN
“AI Safety & Alignment” guide by Prinston University: LEARN
AI TOOLS 🔧
“Communicative Agents for Software Development” LLM agent simulations of multi-agent organizational structure and are united by a mission to "revolutionize the digital world through programming.": TRY
“Invideo AI” tools to generate text for YouTube shorts: TRY
COOL THINGS 😎
The Artificial Intelligence Association of Lithuania is organizing a mini-conference in Vilnius: Registration
Glitch - AI short animation by Jeff Synthesized.
AI Burger commercial by Matan Cohen-Grumi