Evidence-Informed LLM Beliefs for Continual Scientific Discovery

The article addresses the limitation of AutoDiscovery's use of static "Bayesian surprise" by introducing evidence-informed LLM beliefs, where priors are updated with evidence from previous hypotheses to compute non-stationary surprisal. The authors find that embedding-based retrieval-augmented generation over prior discoveries best anticipates eventual posteriors and identify 37.5% of static surprisals as spurious.

• The method introduces belief-update filtering and diversity maximization to modify the search procedure.
• It avoids spurious rewards and prioritizes hypotheses that remain surprising under non-stationary beliefs.
• Across five discovery domains, the approach increases accumulated non-stationary surprisal by 30.62% on average compared to the original search.

The authors conclude that continual scientific discovery with LLMs requires not only better belief measurement but also search procedures that avoid redundancy and encourage diversity.