Bayesian update: if AI solves the Riemann hypothesis, P(dangerous AI takeoff) jumps from 10% to 74%

builderjaydub · x · 2026-11-21

Forecasting researcher NunoSempere shared a worked Bayesian update: starting from P(constrained AI progress)=90% and P(scary AI takeoff)=10%, with P(Riemann hypothesis solved | constrained)=3% vs P(RH | scary takeoff)=80%, the total P(RH) works out to 10.8%. If the RH actually gets solved, Bayes' rule pushes the posterior P(scary AI takeoff | RH) to 74%.

David Kasten quote-tweeted that this is basically the archetypal example of 'Thing You'd Expect To See If We're In A Takeoff Scenario' — your reaction should match someone asking 'can we REALLY land a man on the Moon?' while Sputnik beeps overhead.

Original post →

More from AGI Musings

AGI Musings channel →