Reproduce “Automating Formal Verification with Reinforcement Learning and Recursive Inference” 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
Max Tan
The authors study language models for formal verification, training Dafny programs with verifier-based rewards and using verifier feedback to guide structured proof search in Lean. On a selected 26-task pilot, direct repair solves 12/26 tasks, while the full scaffold solves 18/26, but this small residual set costs more inference and does not establish broad gains.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.