Ex-OpenAI researcher Kokotajlo: recursive self-improvement just 0–4 years away
vkrakovna · x · 2026-10-02
In a Palisade AI interview, ex-OpenAI researcher Daniel Kokotajlo warns that companies are handing AI development to their AIs — "exactly as dangerous as it sounds." He predicts recursive self-improvement and AI better than the best humans at everything is 0–4 years away, argues nobody knows how to control superintelligence, and says the world should shut down frontier labs or block them from building stronger models.
More from AGI Musings
- Harvard physicist uses Claude to link physics calculations to ecology, genetics — AnthropicAI · 2026-10-02
- Author argues humans have just 1-2 decades before AGI destroys or transforms us — danfaggella · 2026-10-02
- Menlo's Deedy Das on Enterprise AI, Model Pricing Wars, and AI's Impact on Indian IT Jobs — HimanshiET · 2026-10-02
- Burkov on AI replacing coders: hand-crafting code is fine, but business always picks efficiency — burkov · 2026-10-02
- Researcher Mocks 'TESCREAL' as Useless Acronym for AI Debate Factions — jankulveit · 2026-10-02
- Understanding AI launches Understanding Robots newsletter, debuting with GPT-Astra 6 analysis — binarybits · 2026-10-02