Large Discovery Models: Empirically-grounded Model-Based Open-Ended Search

Summary

Scientific discovery often involves optimizing expensive-to-evaluate objectives over vast, structured, and open-ended hypothesis spaces, such as molecules, protein sequences, and computer programs. Generative models like large language models (LLMs) provide expressive priors, but their likelihoods and self-assessments are unreliable for objectives and calibrated epistemic uncertainty, especially for novel candidates. The Large Discovery Model (LDM) couples a generative model with a Bayesian non-parametric reward surrogate, where the generative model proposes and refines candidates, and the surrogate predicts performance and quantifies uncertainty. The discovery memory and surrogate are continually updated with new observations. Evaluated on neural-network training, antibody design, and molecular optimization, LDM achieves a 2.4-fold greater reduction in validation BPB, an 18.2% relative decrease in binding energy, and over 60% relative gains in multi-objective performance compared to LLM-only reflection or traditional statistical search, suggesting LDM as a general-purpose discovery engine.