On May 25, 2026, a new paper titled 'ImProver 2: Iteratively Self-Improving LMs for Neurosymbolic Proof Optimization' was published on ArXiv AI.
This work focuses on the need to refactor verified proofs to enhance their maintainability, as formal mathematics libraries continue to grow in complexity.
Additionally, the approach seeks to improve the quality of training data for neural provers, which is crucial for their effectiveness in handling formal proofs.