Reproduce “SciResearch-Eval: Evaluating Scientific Judgment in Autonomous Experimental Agents” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Brian T. Weston, Alondra Torres Contreras, Aldair E. Gongora

SciResearch-Eval tests whether AI agents can estimate thermal properties when supplied simulators omit physics. Agents design experiments, fit the simulator, and report 95% intervals scored against known truth. On 38 held-out tasks with missing physics, default coverage ranged from 20% to 62% for most models, while two reached 94% and 99%; this is a controlled simulation in one thermal domain.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.