Anthropic mathematician Levent Alpöge announced a proof of the Generalized Riemann Hypothesis (GRH) developed with the AI model Claude, formalized in the Lean programming language across 14 data packages for computer verification. Alpöge stated the AI extended a method from mathematician Stepanov, overcoming the inability to differentiate discrete integers by chaining mathematical transformations to mimic derivatives; the key insight was that a sequence of length X requires exactly √X chained transformations. The printed proof would exceed 250,000 pages due to exhaustive logical expansion. Hours later, Stanford University released a competing, entirely human draft proof of the GRH by mathematician Alessandro Masullo, vetted for months by number theorist Brian Conrad. AMS president Ravi Vakil posted the draft prematurely in response to Alpöge's claim, noting that AI-assisted review had already identified issues requiring resolution before community acceptance. Both approaches are being taken seriously: Masullo's benefits from extensive human expert review, while Alpöge's relies on Lean's step-by-step verification, though correct formalization of the initial problem statement remains a potential failure point. Alpöge acknowledged Claude surpassed his understanding in subfields and plans a human-readable version. The breakthroughs follow a rapid 2026 sequence where Claude solved the Jacobian conjecture (July) and contributed to Navier-Stokes and prime number advances, prompting an identity crisis in mathematics regarding AI's role in resolving millennium problems.

Read original