Anthropic's unreleased model pushed the Riemann hypothesis lower bound in 36 hours
An unreleased Anthropic model raised the lower bound up to which the Riemann hypothesis is known to hold — not a proof, but real progress on a 150-year-old problem — after a staff member with no mathematical training told it to "take a real stab" and left it running for about a day and a half across 650 ideas, 60 subagents, and 31 million output tokens. Two in-house mathematicians confirmed the result and it was formalized in Lean, putting it alongside OpenAI's 10 Astra proofs and Anthropic's disproof of the Jacobian conjecture in a run of machine-led results. The credit question is now live: a June declaration from prominent mathematicians argued proofs must be attributable to authors who take responsibility for correctness, while Fields Medalist Timothy Gowers countered that unattributed theorems may be no more troubling than unnamed stars.
Source: techcrunch.com ↗
If we arrive at a world where mathematical theorems are no longer associated with mathematicians, maybe that won't be any more problematic than the fact that stars aren't named after astronomers and most aren't named at all.
Why this matters
- → AI pushed a 150-year-old unsolved problem measurably forward in 36 hours
- → Raises urgent questions about attribution and authorship in mathematical discovery
- → Signals a shift in how breakthroughs happen — coordination beats individual genius