AI Futures Q2.5 update: reality running 70-90% as fast as AI 2027 predicted
ben_j_todd · x · 2026-09-22
- AI Futures Project (Eli Lifland, Daniel Kokotajlo, Brendan Halstead) releases its Q2.5 2026 timelines update: timelines slightly shortened, with notably improved modeling and evidence. They re-evaluated AI 2027's predictions, finding reality running about 70-90% as fast as predicted.
- New methods: the original model anchored the compute for an "Automated Coder" (AC — the point where a frontier lab would rather fire human engineers than give up AI coding) via METR's coding time-horizon, a method with big weaknesses (unclear mapping to AC, contested superexponential trends).
- Two new anchors: coding uplift (their favorite — extrapolate the doubling time of uplift-1 from speedups AIs give engineers at AGI labs) and revenue.
- The post also details a simpler 3-parameter uplift model and notes the update was delayed by publishing their domestic-regulation post "AI 2040: Plan A".
Related event: AI 2027 Forecast on Track: Reality Moving at 70-90% of Predicted Pace(2 posts)→
More from AGI Musings
- OpenAI, Anthropic and DeepMind job posts signal PMs must write evals and code — aakashgupta · 2026-11-12
- Architecture study sparks monitoring concerns over a multi-hop planning spike — ChowdhuryNeil · 2026-11-12
- PowerShell creator Jeffrey Snover at Ignite: AI revolution is an extinction event for those who don't adapt — dfinke · 2026-11-12
- Worst AI use: signaling you don't care about things you should — Afinetheorem · 2026-11-12
- Stanford HAI blueprint: 56 US agencies report 3,611 AI use cases, evaluation lags — StanfordHAI · 2026-11-12
- Fleuret amplifies debate: AI can't slash wet-lab experiments in unmapped biology — francoisfleuret · 2026-11-12