In the video titled ‘AI Model Simulates 500 Million Years of Evolution to create new Proteins! ESM3 is a LLM for Biology’ by Wes Roth, the focus is on the groundbreaking ESM3, a language model designed to simulate 500 million years of evolution to create new proteins. The video explores the significance of this development in the field of biology and its potential applications.

ESM3 is a frontier language model similar to ChatGPT and Gemini, but it is specifically designed for biological applications. It can generate new proteins by reasoning over the fundamental biological properties of proteins: sequence, structure, and function. This capability allows scientists to engineer biology from first principles, potentially transforming drug development, biotechnology, and environmental solutions.

The video explains how ESM3 tokenizes the biological properties of proteins, enabling the model to understand and generate new protein sequences, structures, and functions. One notable achievement highlighted is the creation of a new fluorescent protein, ESM GFP, which has a sequence only 58% similar to the closest known fluorescent protein. This protein was developed through a process equivalent to simulating 500 million years of evolution.

The video also discusses the implications of this technology, such as the potential to create proteins that can break down plastic waste or capture carbon. ESM3’s ability to generate proteins with specific functions and structures gives scientists unprecedented control over protein design.

Additionally, the video touches on the broader context of AI advancements, mentioning a new AI chip by a company called Etched, which claims to be the fastest AI chip available, capable of processing over 500,000 tokens per second. This chip is designed to handle large language models like ESM3 more efficiently than current GPUs.

Overall, ESM3 represents a significant leap forward in the field of synthetic biology, offering new tools for scientists to explore and manipulate the building blocks of life.

Wes Roth
Not Applicable
July 7, 2024
ESM3 Blog
PT16M14S