Software singularity must be a sigmoid: compute substrate caps AI takeoff curves

tszzl · x · 2026-11-18

tszzl argues the software singularity must ultimately be a sigmoid curve: performance on any given computational substrate has an upper limit, with the early phase looking exponential. An accompanying chart shows three paths under the same fixed compute budget — fast takeoff, gradual takeoff, and early plateau — with a dotted line marking frontier-lab-level AI research ability. The open question: how much smarter than humans is the ceiling, and how easy is coexistence? History is full of plateau predictors who turned out wrong.

Original post →

More from AGI Musings

AGI Musings channel →