Large Discovery Models: LLM Proposes, Bayesian Surrogate Scores, 2.4x Lower Error

_akhaliq · x · 2026-08-19

The paper Large Discovery Models: learning where to search next proposes an iterative discovery loop: an LLM proposes candidates, a Bayesian surrogate scores uncertainty, and the loop repeats to decide where to search next.

Across programs, proteins, and molecules, the method cuts validation error by 2.4x, improves binding energy by 18%, and boosts molecular objectives by 60%+.

Related event: Large Discovery Models combine LLM proposals with Bayesian scoring for open-ended search(2 posts)→

Original post →

More from Research

Research channel →