We have proof automation now
Summary
The article explores how large language models can automate proof tasks in dependently-typed languages like Lean, potentially making such systems practical for real-world software engineering. It surveys Zstandard’s entropy encoding (FSE), the role of proofs, and the promise and limits of combining LLMs with formal verification, including references to open-source tooling and verified assembly ideas. The piece argues that proof automation with AI is emerging as a viable capability for advanced software correctness and verification workflows.