Reproduce “Principia: Test-Time Hypotheses to Principle Construction for Scientific Discovery” 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

Dongki Kim, Sangwoo Park, Jinheon Baek, Soyeong Jeong, Sung Ju Hwang

Principia turns observed outcomes into revisable guidance for scientific prediction and optimization. It compares related inputs and outcomes to propose tentative explanations, generalizes recurring patterns into task-specific principles, and updates them when new results do not fit. Across eight scaffold-held-out molecular-property tasks, it achieves mean accuracy of 69.75% and macro-F1 of 69.03%, versus 63.95% and 62.30% for a few-shot baseline. These benchmark results do not validate causal explanations or establish universal generalization.

Summary by alphaXiv.

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