Ben Todd: Leading Models' Task Time Horizon May Exceed 100 Hours as Everyone Underestimates AI

Benjamin Todd, former head of 80000 Hours, argues nearly everyone has underestimated AI over the past 18 months. His analysis of AECI scores suggests leading models' task time horizon has reached 100 hours at 50% reliability, with the growth rate doubling to about 50x per year.

2026-11-13 ~ 2026-11-13 · 3 related posts

• 80% METR time horizon likely already above 30h, says Ben Todd — ben_j_todd · 2026-11-13
• Leading models' time horizons hit ~100h as doubling rate doubles to ~50x/year — ben_j_todd · 2026-11-13
• Benjamin Todd: superforecasters and experts all underestimated AI progress over 18 months — ben_j_todd · 2026-11-13