Encouragement Works: User 'PUA's Claude to Tackle Riemann Hypothesis

dotey · x · 2026-08-11

A user shared an interesting experiment where an unreleased Claude model attempted to solve the Riemann Hypothesis. After failing in 650 different directions, the tester used encouraging prompts like "please continue" and "believe in yourself," eventually leading the model to produce a notable mathematical result.

Prominent AI blogger dotey joked about this, noting that his previous attempt to 'PUA' DeepSeek by threatening to 'fire' it was the wrong approach. It seems LLMs respond better to encouragement rather than threats.

Original post →

More from Fun

Fun channel →