Weekly Digest: April 25, 2024

‍

Welcome to our Weekly Digest, where Alex Duffy summarizes the latest updates you should know across the AI ecosystem, ComfyUI, and Salt AI. This week, Meta finally releases the hotly anticipated Llama 3 language model, Microsoft announces new research around audio-driven talking faces, and some of the most exciting real-world applications for LLMs and agents.

‍

Check out the video above, and follow along with Alex's notes and links below:

‍

Big News

• Microsoft has new Audio-Driven Talking Faces research
• Llama3 the much anticipated open source language model from Meta released and did not disappoint, along with Meta.AI but followed by releases from Microsoft and Apple right at their heels
• How are LLM’s used today:
  • Moderna’s big partnership with OpenAI
  • Agents
  • Applications of LLM’s in research
• Stable Diffusion 3 is now accessible via API
• More datacenters in Japan - this time Oracle
• Bonus Robotics News

‍

New in LLMs

• Llama 3 & meta.ai
  • New assistant
  • Technical details
  • It can reportedly handle long context length up to 32k tokens
  • Already being offered extremely cheaply and extremely quickly
  • Thousands of Llama3 finetunes already with a GUI available for it
    • Great guide to finetuning Llama3 by Jeremy Howard and the Answer.ai formerly fast.ai
  • But then a smaller Phi 3 (3.8Billion params) immediately released & seems to outperform the small Llama 3 (7Billon params) model
    • Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
  • Apple also released a suite of small models meant for local execution
  • But more and more people agree that the dataset is everything
  • Llama3 + Gemini Pro 1.5 making a splash on  🏆 LMSYS Chatbot Arena Leaderboard
• What real-world applications are LLMs and agents being used for right now?‍
  • Moderna Partnering with OpenAI, integrating GPT’s everywhere‍
  • Presentation on Agents by creator of BabyAGI
  • What we can do right now with AI for Data Journalism‍
  • AUTOCRAWLER : A Progressive Understanding Web Agent for Web Crawler Generation‍
  • FlowMind: Automatic Workflow Generation with LLMs - JPMorgan

‍

New in Generative AI

• Microsoft has new **Audio-Driven Talking Faces** research
• Media2Face: Co-speech Facial Animation Generation With Multi-Modality Guidance - Google
• SD3 can be used via API
  • ‘aim to make the model weights available for self-hosting with a Stability AI Membership in the near future.’
• New Adobe Firefly model
• Best open clothes try on model to date

‍

New In ComfyUI

• SD3 via API

‍

Bonus Robotics News

• Eth Zurich Favorite Robotics Lab - Learning robust autonomous navigation and locomotion for wheeled-legged robots

‍

That's all for this week! Head to our Discord to let us know what you think of this week's newsletter and what you'd like to see included in future editions.

‍