Benjamin Todd: almost everyone underestimated AI progress over the past 18 months
ben_j_todd · x · 2026-10-27
Benjamin Todd (ex-80,000 Hours co-founder) scores forecasts against 18 months of real AI progress and concludes AI wasn't overhyped — nearly everyone underestimated it. Superforecasters expected AI to match a top virologist team on lab troubleshooting by 2034; it likely did in April 2025. AI experts predicted 14% on LiveCodeBench Pro (Hard) by end-2026; the best model hit 54% by May. Economists pegged the top AI company at $16bn annualised revenue by end-2026. Todd admits his own "AGI by 2030?" piece was too conservative on numbers, though his core thesis — pretraining, RL, test-time compute and agent scaffolding all keep working — has held, and agents as "the next stage of scaling" was right. Overall: the public and superforecasters underestimated most, AI experts were closer but still too slow, and only AI CEOs were slightly optimistic — assume their '6 months' means 6-18 and you'd have predicted well.
Related event: Ben Todd: AI wasn't overhyped—almost everyone underestimated it(6 posts)→
More from AGI Musings
- 'Lessons learned' papers may be worth more to AI models than 2500-citation atlas papers — owl_posting · 2026-10-27
- AI reportedly disproves the Exponential Time Hypothesis, shifting P vs NP odds — esrtweet · 2026-10-27
- The SMB process-data gold rush: 1M businesses selling know-how to RL environment companies — willcb · 2026-10-27
- Economist-spotlighted paper asks what evidence on work and wellbeing implies for AI futures — HaydnBelfield · 2026-10-27
- AI Researcher Flags Academic Post Hinting at Aerial Tracking, Zero Questions Raised — giffmana · 2026-10-27
- US Creative Industries Lost 200k Jobs in Four Years as AI Rises — emmanuelvivier · 2026-10-27