Two Minute Papers · 2026-08-14 · notable
Two Minute Papers — 'Claude AI Failed 650 Times, Then Beat The Human Record'
Two Minute Papers walks through Anthropic's Riemann zeta result, where a research version of Claude tried 650 ideas that failed before raising a longstanding lower bound from 41.6% to 67.2%.
Two Minute Papers covers the Anthropic maths result, including the 650 dead ends that came before it.
What is it?
Two Minute Papers breaks down Anthropic's Riemann zeta work, in which an unreleased research version of Claude improved the lower bound on the fraction of Riemann zeta zeros satisfying the Riemann hypothesis from 41.6% to 67.2%. The video also links Scientific American's pushback piece, which argues the model did not solve the Riemann hypothesis itself.
How does it work?
The Anthropic write-up that Two Minute Papers uses describes Claude first generating and trying 650 ideas, none of which worked. It then ran roughly 60 subagents over about a day and a half, issuing around 2,400 shell commands and thousands of numerical checks before it landed on the improved bound.
Why does it matter?
Two Minute Papers frames the result as a test of how far long-running agent orchestration can push research maths, and pairs it with the Scientific American critique so viewers see both the claim and the limits. The 650 failed attempts are the useful part for practitioners: the win came from letting the model run wide and long, not from one clever prompt.
Who is it for?
ML researchers, maths-curious developers
Try it
Watch the video, then read Anthropic's write-up at anthropic.com/research/riemann-zeta