Linking AI models to formal verification methods can correct LLM shortcomings such as false assertions. Amazon's Byron Cook explains the promise of automated reasoning.

Thinking-2507, as we'll call it for short, now leads or closely trails top-performing models across several major benchmarks.