A credible researcher's firsthand account that frontier models are generating novel mathematical ideas faster than humans can absorb them signals where reasoning-model capability is heading and why interpretability of AI-generated proofs will become a bottleneck.
