Sign Up
OverviewAPI
Baldur, an AI-powered proof generator, aims to facilitate formal verification and bug fixing in software. Leveraging Google's Minerva LLM and fine-tuning on data about proofs and theorems, Baldur has a 41% proof rate and can repair its own proofs, leading to a 66% correct proof rate when combined with the current state-of-the-art tool Thor. The tool's potential applications include national security system verification and interactive tools for software developers.

Ask a question