Show HN: CakeML-based self-verifying, self-improving system

Category: other

Tags: formal-verification, self-modifying-systems, ai-safety

Score: 7.3/10 (Innovation: 8, Technical: 9, Documentation: 8, Utility: 4)

Svenvs is a self-verifying, self-improving system built on CakeML and Candle, allowing an AI to operate within a machine-checked safety envelope that can be upgraded without losing security guarantees. It is interesting for its rigorous approach to corrigibility and self-modification, using formal proofs to ensure safety even when the system improves its own theorem prover.

Target audience: researchers in formal verification and AI safety

Repository: https://emberian.github.io/svenvs/ · Standard ML · 3 stars

View on Hacker News