Eli Lifland explains reasoning behind superhuman AI timelines and takeover odds
eli_lifland · x · 2026-10-11
Eli Lifland detailed the reasoning behind his risk estimates: TED-AI means an AI at least as good as top human experts at virtually all cognitive tasks; his takeover probability reasoning is closest to a cited long-form piece plus an analysis of takeover chance vs. amount of slowdown. He clarified that clear signs of misaligned takeover could come substantially after TED-AI — in AI 2027, extinction came over two years after TED-AI, even with a fast takeoff — though it could also come around the same time. He also linked Ryan Greenblatt's discussion of fatalities from AI takeover.
More from AGI Musings
- Patrick Collison: personal agents are a structural subsidy for good products — sujingshen · 2026-10-11
- 228 tasks, 26 models: METR's time horizon curve is nearly flat from 2-30 min — lulzxdxdxd · 2026-10-11
- AI Code Was Dismissed as Slop a Year Ago; Now It May Redefine Good Software — kevinnbass · 2026-10-11
- Complexity scientist Melanie Mitchell: "superintelligence" isn't yet a scientific concept — khademinori · 2026-10-11
- "Censoring AI is step one": a warning about the surveillance slippery slope — PierceLilholt · 2026-10-11
- Togelius: commenters keep telling pros upset by AI progress to accept obsolescence — kchonyc · 2026-10-11