Kokotajlo's AI 2040 Plan A: pause AI to build safety before superintelligence
Machine Learning Street Talk · youtube · 2026-09-08
Daniel Kokotajlo and Thomas Larsen of the AI Futures Project join Machine Learning Street Talk to discuss AI 2040: Plan A — a proposal to pause AI development, build safety infrastructure, then proceed cautiously up to the strongest controllable AI. Key points:
- Limits of AI 2027-style forecasting and why AI skeptics are updating
- What happens when AI can automate research and sustain an economy without human workers
- Control buys time but cannot substitute for alignment
- The case for public AI research and whether the US and China could jointly enforce a slowdown
- What evidence would change their forecasts; the 'AI as normal technology' disagreement
More from AGI Musings
- The age of the non-AI proof is over: no one will believe you didn't use AI — inductionheads · 2026-09-08
- Cost is the only thing stopping widespread AI job loss, argues tech commentator — CtrlAltDwayne · 2026-09-08
- GenAI won't kill AAA: ~75% of EA revenue is live services anchored by sports franchises — dioscuri · 2026-09-08
- New article argues robots will create more retail jobs by 2035, not fewer — broodsugar · 2026-09-08
- Podcaster Stewart Alsop III on AI psychosis, token maxxxing, and why LLMs aren't AGI — StewartalsopIII · 2026-09-08
- AI tools won't save idea guys — their bottleneck is agency, not tools — tekbog · 2026-09-08