Show HN: Formal – Formal verification for AI-generated code using Lean 4 (github.com)

A new tool called "Formal" has been announced, designed for formal verification of AI-generated code using Lean 4, a proof assistant. This system extracts correctness properties from pure functions produced by any large language model (LLM), such as GPT-4 or Claude, and translates these properties into theorems that are then validated using Lean 4 and its Mathlib library. This process enables developers to obtain mathematical proofs of the logic within their AI-generated code, surpassing traditional testing methods that only ensure functionality under specific conditions. The significance of this development lies in its ability to provide a higher level of assurance for AI-generated software, particularly in critical systems where correctness is paramount. By focusing solely on pure functions and excluding side effects like database calls or I/O operations, the tool ensures that the properties verified are deterministic and reliably reproducible. This structured approach not only enhances the reliability of AI-driven code but also fosters greater trust in utilizing LLMs for programming tasks, paving the way for broader adoption in safety-sensitive applications. Key features include seamless integration with multiple LLM backends and an automated setup process, making formal verification more accessible for developers.