Greg Kamradt: only 1-2% of Claude's proof-exploring knowledge nodes get used

GregKamradt · x · 2026-11-21

Commenting on a 'timeline of Claude solving the Generalized Riemann Hypothesis' visualization, Greg Kamradt asks what happened to all the ideas that didn't work. He guesses 1-2% efficiency — knowledge nodes created vs. nodes eventually used in the proof — and argues we're approaching a stable $-to-knowledge conversion ratio for AI reasoning.

Related event: Greg Kamradt estimates only 1-2% of Claude's proof ideas are actually used(2 posts)→

Original post →

More from Models

Models channel →