ben_j_todd pushes back on AI plateau claims: 5 years of rapid progress, bad track record

ben_j_todd · x · 2026-11-21

benjtodd responds to claims that AI is plateauing on some task subsets: after 5 years of rapid progress across the board, he starts out very skeptical of plateau narratives, which have a terrible track record. If progress is rapid at the 50% task-completion time horizon, he assumes rapid progress at the 80% horizon too until proven otherwise. ronbodkin concedes we need better data — 50%-horizon numbers don't reveal reliability trends on harder problems.

Related event: Debate over AI task reliability plateauing near 90%: saturation or bottleneck(12 posts)→

Original post →

More from AGI Musings

AGI Musings channel →