Reproduce “MaD Physics: Evaluating information seeking under constraints in physical environments” 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

Moksh Jain, Mehdi Bennani, Johannes Bausch, Yuri Chervonyi, Bogdan Georgiev, Simon Osindero, Nenad Tomašev

MaD Physics evaluates agents that choose what, when, and how precisely to measure simulated systems before predicting a later state. Its benchmark covers classical, fluid, and quantum mechanics, with noisy, costly observations and altered physical laws. In one fluid-mechanics setting, Gemini 2.5 Pro’s error falls from 2.51 to 0.70 with Strategy prompting; this result is bounded to simplified benchmark tasks.

Summary by alphaXiv.

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