Study: LLMs propose hypotheses, but Bayesian experiments decide data efficiency
rohanpaul_ai · x · 2026-08-16
A paper finds that letting agents choose experiments and fit mechanisms is data-inefficient. The Model Discovery Agent (MDA) uses LLMs to propose hypotheses while Bayesian inference scores mechanisms and value-of-information designs experiments.
On FORCEBENCH, MDA matches the accuracy of an unthrottled Opus 4.7 agent using 8 experiments instead of 41, achieving a 93% numeric pass rate versus 31%. When predictions fail, MDA asks the LLM for new mechanisms and repeats the loop.
Related event: LLM Hypotheses Plus Bayesian Inference Speed Up Scientific Discovery(3 posts)→
More from Research
- 3rd 3D HUMANS Workshop Returns at ECCV 2026 — dimadamen · 2026-08-18
- TogetherAI open-sources XoRL: 0 train-infer mismatch for large MoE RL training — PandaAshwinee · 2026-08-18
- SegDAC: Boosting Visual RL Generalization with Dynamic Object Tokens — GlenBerseth · 2026-08-18
- Study Compacts Context, Finds Prompt Caching Makes Summarization Obsolete — AI Engineer · 2026-08-18
- Georgia Tech's 2026 LLM Course: From MoE and Self-Play RL to Diffusion LMs — cocoweixu · 2026-08-18
- New Dataset: 35k Hugging Face Model Summaries Generated for $0.43 per 1k Rows — vanstriendaniel · 2026-08-18