Claude's mathematical capabilities
Summary
Anthropic reports that an unreleased research version of Claude was challenged to make progress on the famous Riemann Hypothesis, which remains unsolved, and although Claude did not prove the hypothesis, it unexpectedly produced a new mathematical result: it improved the known lower bound for the proportion of zeros of the Riemann zeta function lying on the critical line from 41.6% to 67.2%.
The result builds on decades of previous work by mathematicians, particularly results by Baluyot, Goldston, Suriajaya, Turnage-Butterbaugh, and Bombieri. Claude generated hundreds of ideas, coordinated around 60 subagents, ran thousands of numerical checks and mathematical experiments, examined dozens of research papers, and independently checked its argument; Anthropic mathematicians subsequently reviewed and validated the work, while Claude also produced a formally verifiable
Lean proof. Anthropic emphasizes that this does not solve the Riemann Hypothesis, but argues that the episode demonstrates an emerging ability of AI systems to combine existing mathematical knowledge in unexpected ways and potentially contribute genuinely new results to advanced mathematics.
ARTICLE / ARCHIVE
Summary
Anthropic reports that an unreleased research version of Claude was challenged to make progress on the famous Riemann Hypothesis, which remains unsolved, and although Claude did not prove the hypothesis, it unexpectedly produced a new mathematical result: it improved the known lower bound for the proportion of zeros of the Riemann zeta function lying on the critical line from 41.6% to 67.2%.
The result builds on decades of previous work by mathematicians, particularly results by Baluyot, Goldston, Suriajaya, Turnage-Butterbaugh, and Bombieri. Claude generated hundreds of ideas, coordinated around 60 subagents, ran thousands of numerical checks and mathematical experiments, examined dozens of research papers, and independently checked its argument; Anthropic mathematicians subsequently reviewed and validated the work, while Claude also produced a formally verifiable
Lean proof. Anthropic emphasizes that this does not solve the Riemann Hypothesis, but argues that the episode demonstrates an emerging ability of AI systems to combine existing mathematical knowledge in unexpected ways and potentially contribute genuinely new results to advanced mathematics.
ARTICLE / ARCHIVE
┌────────────────────────────────┐
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘
│ KONSTANTINOS MICHAILIDIS │
└────────────────────────────────┘