AI systems are beginning to produce proof ideas that experts take seriously, even when final acceptance is still pending.
The method has two main features: it evaluates how AI models reason through problems instead of just checking whether their ...