- Bot Eat Brain
- Posts
- Chai Discovery released Chai-1
TOGETHER WITH
Good morning, human brains, and welcome back to your daily munch of AI news.
Here’s what’s on the menu today:
Another AI for molecular structures? 🤖 🧬
Chai Discovery released Chai-1.
Is it fun? Ask a bot 👾 🎮
Researchers unveiled the Exploratory Agent Framework.
Sponsor Bot Eat Brain and reach over 20,000 readers.
(Now 50% off)🤯
Peep today's ‘What Would You Do?’ at the bottom. 👇
MAIN COURSE
Another AI for molecular structures? 🧬 🤖
On Tuesday, Chai Discovery released Chai-1. It's a new AI model that rivals AlphaFold3 in predicting molecular structures.
What is Chai-1?
It's a multi-modal foundation model that predicts the structures of proteins, small molecules, DNA, RNA, and other biomolecules.
How good is it?
It matches or beats top models like AlphaFold 3 on several benchmarks. For example, it has a 77% success rate on the tough PoseBusters protein-ligand test.
What makes it special?
Chai-1 can make accurate predictions from single protein sequences without needing multiple sequence alignments.
Any cool features?
It can incorporate experimental data to improve predictions in real-time, doubling accuracy in some cases.
How can I use it?
Chai-1 is freely available via a web interface for both academic and commercial use. They're also open-sourcing the model weights and code for non-commercial use.
Why is this a big deal?
Open-sourcing a frontier model is rare, especially as competitors like DeepMind are moving towards closed-source approaches.
Didn't someone else release a similar model recently?
Kind of. On Monday, we covered Google DeepMind’s AlphaProteo. Chai-1's open-source approach and multi-modal capabilities set it apart in the field.
SPONSORED BY LLM BEEFER UPPER
Instantly improve your LLM outputs
The LLM Beefer Upper automates chain of thought, using multi-agent prompt templates.
97% of users say the final post-reflection result is better than the first. Choose from popular pre-built templates or design your own, for whatever LLM process you need.
New users get 50 credits.
SIDE SALAD
Is it fun? Ask a bot 👾 🎮
On Tuesday, University of London researchers unveiled the Exploratory Agent Framework. It uses agents to evaluate how engaging procedurally-generated game levels are for human players.
What did they do?
They developed AI agents to explore video game levels, looking for high places, interesting objects, and more.
How did they test it?
The AI explored 5 “engaging” and 5 “unengaging” levels, rating them on factors like level coverage and novelty.
What were the results?
The AI consistently rated the engaging levels higher than the unengaging ones.
Why is this useful?
It could help game designers automatically test and improve levels before human playtesting.
What's the potential impact?
This could speed up game development and lead to more consistently fun levels for players to explore.
Didn't someone just use AI to make Doom?
Yup. Last week, we covered GameNGen. It’s an AI system that simulates the classic shooter Doom without using a traditional game engine.
YOUR DAILY MUNCH
Think Pieces 🧠
New Tesla patent? It’s a wireless charging system for its Robotaxis with adaptive control for coil variations, real-time charging adjustments, and more.
Startup News 💰
Getty Images released a visual dataset. It’s 3750 images on Hugging Face that’s free of duplicates, NSFW content, and other unwanted elements.
Research 👨🔬
Loopy — an end-to-end audio-conditioned diffusion model for portrait video generation that generates lifelike portrait videos from audio alone.
Open-MAGVIT2 — an open-source tool that generates images using a step-by-step prediction process.
FURRY FRIENDS
Say “hi” to Walter on our walk. 🐶
Respond to this email with your pet’s name and pic for a chance to be featured here tomorrow!
MEMES FOR DESSERT
AI ART SHOW
WHAT WOULD YOU DO?
I knitted a sweater… 🧥
Who is this going on? 👇
Sweater weather... |
Ideas? Comments? Complaints?
Respond to this email or hit me up on 𝕏.
Until next time 🤖😋🧠
What'd you think of today's newsletter? |