Reproduce “Trust, but verify, audit and attribute: a human-in-the-loop framework for checking AI-produced research results” 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
György Eigner, Patrik P. Süli, Roland Hollós
The authors propose a workflow for checking whether tools assess artefacts that faithfully represent a paper’s claims. Humans fix claims and ambiguities, while tools return verdicts and back-translations for fidelity auditing and reporting. In a worked example, Wolfram and Lean refute an exact 0.37 maximum but confirm its rounded reading; MATLAB is inconclusive. This is proposed, not validated as a guarantee.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.