LLM sequential decision making under uncertainty in biochemical domains
Five frontier LLMs in a Bayesian Optimization setting are benchmarked against published statistical baselines on seven combinatorial datasets spanning protein engineering, reaction optimization, molecular design, peptide self-assembly, and catalysis, indicating that priors and data must be decoupled to achieve effectiv...
Mattias Akke, Soojung Yang, Jurǵis Ruža et al.
· 0 citations