Gordic Aleksa Dissects Kokotajlo's 'AI 2040' Plan A: A Chain of Low-Probability Assumptions
gordic_aleksa · x · 2026-09-13
- Kokotajlo et al. (AI 2027 team) released "AI 2040: Plan A," proposing delaying superintelligence to 2040 via US-China compute declaration and inspection (<1% hidden compute), inference-only verification hardware, total research transparency, and "mutually assured compute destruction" (US datacenters in Mongolia, Chinese in Canada).
- Gordic Aleksa's critique: it's a chain of low-probability events — no historical precedent for such US-China coordination; Mongolia/Canada hosting destructible strategic datacenters is unrealistic; the plan ignores the economic shock given massive committed capex.
- He also pushes back on the timeline's "mode collapse" narrative that only offers job loss, permanent hegemony, or extinction as outcomes.
More from AGI Musings
- Ex-Google quantum lead claims big labs have achieved RSI and aim to cement an intelligence cartel — beffjezos · 2026-09-13
- Screenshot Stash: User Lists Open-Source AI Promises to Watch After 'Dystopian Fork' Fears — EddyVGG · 2026-09-13
- Marc Andreessen Revives 'Baptists and Bootleggers' Framework to Dissect AI Regulation — illscience · 2026-09-13
- tszzl predicts open-source models will be banned after a major AI disaster; Beff pushes back — beffjezos · 2026-09-13
- Ex-DeepMind Engineer: 'A Fool in Motion Beats a Genius at Rest' Sums Up Deep Learning — yacinelearning · 2026-09-13
- The AI risk Goldilocks meme: less worried = out of touch, more worried = EA-poisoned — shakoistsLog · 2026-09-13