FVEL: Interactive Formal Verification Environment with Large Language Models via Theorem Proving

Neural Information Processing Systems 

Formal verification (FV) has witnessed growing significance with emerging program synthesis by the evolving large language models (LLMs). However, current formal verification mainly resorts to symbolic verifiers or hand-craft rules, resulting in limitations for extensive and flexible verification. On the other hand, formal systems for automated theorem proving, such as Isabelle, serve as another line of rigorous verification, upheld by extensive rules and theorems.