- AI For All
- Posts
- Simulating 500 Million Years of Evolution
Simulating 500 Million Years of Evolution
with a Large Language Model
Hello readers,
Welcome to another edition of This Week in the Future! A new AI company wants to give biologists the tools to become engineers, engineers of life, starting with ESM3, a frontier generative model that simulated 500 million years of evolution.
Simulating Evolution
A new company called EvolutionaryScale has an ambitious goal — to engineer biology in the same way we engineer machines and computer programs. Making biology programmable requires understanding the language of life and read and write to the genetic code. If possible, it could revolutionize drug discovery and a whole lot more.
The company’s first step towards this future is ESM3, a frontier language model for the life sciences that reasons over the sequence, structure, and function of proteins. The model was trained on 1x1024 FLOPS and has 98B parameters, significant for a biological model. The model is designed to follow prompts to generate new proteins.
ESM3 has already generated a new green fluorescent protein (GFP). Fluorescent proteins are resposinble for the glowing colors of jellyfish. The company is calling the protein esmGFP, and they estimate that based on the rate of GFP diversification in nature, this new GFP is equivalent to simulating 500 million years of evolution.
Why This Matters
Reverse engineering biology would have profound implications for the world. Instead of leaving it to chance whether or not the next person born is “fit” or not, fitness could be guaranteed. However, meddling with genetics has an abundance of risks, which is why EvolutionaryScale (a public benefit company) is emphasizing responsible development.
🔥 Rapid Fire
Google releases Gemma 2 and Gemini 1.5 Pro 2M context
OpenAI strikes content deal and partnership with TIME
OpenAI delays voice mode and opens desktop app to all
OpenAI acquires desktop collaboration app Multi
Anthropic adds collaboration to Claude with Projects
Anthropic expands access to Claude for governments
Hugging Face revises leaderboard amidst plateauing LLMs
AI work assistants reportedly need a lot of handholding
Major record labels sue AI music generators Suno and Udio
Toys R Us unveils commercial made with OpenAI’s Sora
Developers find major security flaw in Rabbit R1 device
Lenovo accelerates smarter AI with new AI solutions
Oracle announces industry first in-database LLMs
OpenAI blocks API access in China and Hong Kong
Apple Intelligence won’t launch in the European Union
Israel plans AI supercomputer to compete in global AI race
Google’s Gemini is coming to the side panel of Gmail
Firefox will integrate multiple AI services in Nightly test
Figma launches Figma AI with new AI features for designers
Reddit clamps down on web scrapers in new statement
📖 What We’re Reading
“LLMs have grown exponentially in size and complexity. However, this growth brings significant challenges, particularly when it comes to deploying these models on devices with limited memory and processing power. This is where the innovative field of LLM quantization comes into play.”
💻️ AI Tools and Platforms
Robust Intelligence - Enterprise AI security
n8n.io → AI-native workflow automation
Cleanlab → Pipeline for cleaning AI data
Era → AI-powered financial assistant
Juicebox → AI-powered talent sourcing
What did you think of today's newsletter?We value your feedback, so please take time to let us know what you liked and what could be improved. |