GPT-5 Claims to Improve Riemann Hypothesis Bound by 0.002%

I_HALF_CATS · reddit · 2026-08-11

A user shared an interesting AI experiment where GPT-5.6 (Sol) attempted to improve upon Anthropic's bound for the proportion of Riemann zeta zeros.

Sol proposed a candidate unconditional improvement, raising the bound from 67.250070% to approximately 67.25214%. The key observation utilizes the Gram matrix determined by the geometry of zero locations, retaining a nonnegative Gram-spectrum defect to show that triples of close simple zeros contribute a positive spectral penalty. However, Sol does not regard this as a theorem yet, noting that the finite-dimensional linear algebra and truncation passage still require a rigorous analytic audit and independent verification.

Original post →

More from Fun

Fun channel →