Summary
Two Minute Papers host Dr. Károly Zsolnai-Fehér covers a striking result from Anthropic: an unreleased version of Claude improved a mathematical bound related to the Riemann hypothesis beyond the existing human record. The model did not solve the Riemann hypothesis itself — one of mathematics’ most famous open problems — but pushed a closely related result further than any prior human work, a development that professional mathematicians have publicly described as a massive leap forward. A full technical paper and a formally verified, automatically checkable proof are both publicly available.
Three details make the process particularly notable. The prompting was done by a non-mathematician, and for much of the 650-attempt run, the only input was encouragement — variants of “keep going” and “believe in yourself.” Claude had internet access throughout the session but did not use it during the key breakthrough run, which came after 37 minutes of silence. When the result appeared, Claude itself flagged it as suspiciously strong — expressing doubt that something “too strong to be new” could be correct before the result was externally validated.
Zsolnai-Fehér notes this is not the first time the pattern has appeared: a similar encouragement-based prompting approach was used to help Claude disprove the Jacobian conjecture in an earlier experiment. The video draws out what the result suggests about AI capability in open-ended, long-horizon mathematical reasoning — and what it implies when even the model producing a breakthrough underestimates how far AI progress has come.
📺 Source: Two Minute Papers · Published August 14, 2026
🏷️ Format: News Analysis
