A Very Small SAT Solver (From Haskell) Now in Dafny, Proved Correct with LLMs (github.com)

A new SAT solver, originally written in Haskell, has been successfully translated and proved correct in Dafny, utilizing large language models (LLMs) such as ChatGPT and Claude. This development is notable as it showcases the effectiveness of LLMs in assisting with complex proofs and programming tasks. The solver is not only minimal in design but also rigorously established to be consistent, sound, and complete — fundamental properties vital for any SAT solver. It ensures that every solution returned has no contradictory literals, satisfies the problem constraints, and guarantees a solution exists if one can be found. This achievement holds significant implications for the AI/ML community as it highlights the integration of LLMs into formal verification processes, enhancing productivity and reliability in software development. The implementation also leverages Dafny's capabilities for expressing and proving properties of algorithms, offering robust guarantees about correctness in logical computations. The collective use of LLMs for tasks such as lemma decomposition and proof simplification can revolutionize how developers approach verification, paving the way for safer and more reliable AI and machine learning systems.