OpenAI updates & use cases 🧠, Elon Musk xAI chatbot Grok 🤖, Aleph Alpha raises 500 million 💰, Levels of AGI 🌐, Accurate Weather forecasting 🌦️, LLMs for Chip Design 🖥️

AI Connections #40 - a weekly newsletter about interesting blog posts, articles, videos, and podcast episodes about AI

David Mataciunas

Nov 17, 2023

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 🚀

HUGE KUDOS FOR OUR PAID SUBSCRIBERS:

edita.perku ⭐️
ausra.pock ⭐️
donatas.jasas ⭐️

OpenAI Dev Day - the most awaited AI event this AUTUMN is already over!!

Is OpenAI the new Apple? 9 key highlights from DevDay 2023 - Edge Middle East

OpenAI presented a lot of new updates:

🏎 GPT-4 Turbo (faster version)

📚 New context window 128K tokens (this means that new GPT-4 can understand documents

⚙ More control with JSON, functions, and viewing logs

📃 Bring in your own knowledge

🎓 Knowledge cutoff now April 2023 (and being updated more frequency)

🎨 Dall-e 3, GPT-4 Turbo, and TTS in API

🎁 16k context in GPT 3.5

✨ Better customization / GPT-4 Fine Tuning

🧵 Higher rate limits when prompting

💎 "Copyright Shield" - OpenAI will cover legal expenses if you run into copyright issues

🎪 GPT-4 Turbo pricing 2.75x cheaper than previous pricing

💸 GPT-3.5 API pricing cheaper as well

What GPTs Can Do - Custom Versions Of ChatGPT

🦁 Announcement of "GPTs" - Create custom bots tailored for your specific purpose and then share the bot with others

🦈 Launching a "GPT Store"

🎪 Introduces "Assistants" and Assistant Playground

READ 📚

“Elon Musk company xAI release Grok” blog post by xAI: READ This blog post is about xAI's development of Grok, an advanced AI tool aimed at aiding humanity in its pursuit of understanding and knowledge. It details the journey and technical advancements of Grok-1, a state-of-the-art language model, and xAI's commitment to building reliable infrastructure with efficient resource utilization to further AI research and innovation.

“Aleph Alpha raises 500 million to build a European rival to Openai” article by CNBC: READ Congrats Jonas Andriulis!!! 🚀🌌

“GraphCast: AI model for faster and more accurate global weather forecasting” blog post by Google DeepMind: READ This blog post is about the introduction of GraphCast, a state-of-the-art AI model that significantly enhances medium-range weather forecasting. GraphCast achieves unparalleled accuracy in predicting weather conditions up to 10 days in advance, offering earlier warnings for extreme weather events, and its open-source model code is aiding scientists and forecasters globally.

“Transforming the future of music creation” blog post by Google DeepMind: READ This blog post is about the announcement of Google DeepMind's Lyria, their most advanced AI music generation model, and the launch of two new AI experiments in collaboration with YouTube.

“Adversarial Attacks on LLMs” blog post by Lilian Weng: READ

“Emu Edit: Precise Image Editing via Recognition and Generation Tasks” blog post by Meta AI: READ This blog post is Emu Edit, a multi-task image editing model that sets a new state-of-the-art in instruction-based image editing.

“The Next Billion Developers” blog post by Sequoia Cap: READ This blog post is about the debate surrounding Artificial General Intelligence (AGI) in Silicon Valley, contrasting the optimistic views of figures like Elon Musk and Sam Altman with skeptics like Yann LeCun and Jack Clark. It proposes an alternative path to the future promised by AGI, focusing on the potential of expanding the global developer population to 1 billion, which could have a transformative economic impact through the automation of various industries, assisted by advancements in AI and software engineering.

“Retool State of AI 2023” report by Retool: READ

“This AI robot chemist could make oxygen on Mars” article by Nature: READ

READ (RESEARCH PAPERS) 📚

“Levels of AGI: Operationalizing Progress on the Path to AGI” research paper by Google DeepMind: READ This research paper is about proposing a framework for classifying Artificial General Intelligence (AGI) models, introducing levels of performance, generality, and autonomy, aimed at providing a standardized language for comparing models, assessing risks, and tracking progress towards AGI.

“FinGPT: Large Generative Models for a Small Language” research paper by TurkuNLP Group & Hugging Face: READ This research paper is about the development and evaluation of large language models (LLMs) for Finnish, a less commonly spoken language, involving the creation of extensive Finnish datasets and the training of both monolingual FinGPT models and a multilingual BLUUMI model, alongside introducing a Finnish-specific evaluation benchmark, FIN-bench.

“Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena” research paper by UC Berkeley: READ This research paper is about using large language models (LLMs) as judges to evaluate chat assistants, addressing biases and limitations, and introducing benchmarks like MT-bench and Chatbot Arena to verify the agreement between LLM judgments and human preferences, demonstrating over 80% agreement and providing a scalable, explainable approach to approximate human preferences.

“ChipNeMo: Domain-Adapted LLMs for Chip Design” research paper by Nvidia: READ This research paper is about ChipNeMo, which explores using large language models (LLMs) for industrial chip design, employing domain adaptation techniques like custom tokenizers and continued pretraining, demonstrating significant performance improvements in applications like engineering assistance, script generation, and bug analysis, and suggesting potential for further advancements in domain-adapted LLM approaches.