Claude AI Failed 650 Times…Then Beat The Human Record
Unreleased Claude beat human record on Riemann hypothesis-related bound after 650 failed attempts
“Throughout this process, Jared's input was mostly limited to sending Claude messages of encouragement. Mostly variants of keep going and believe in yourself.”
An unreleased Claude model, prompted by a non-mathematician using only words of encouragement, improved a bound related to the Riemann hypothesis beyond any prior human result after 650 failed attempts. The AI self-recovered from wrong paths, required no internet access during the breakthrough run, and produced a formally verifiable proof. This signals a step-change in AI mathematical reasoning capability with implications for scientific discovery workflows.