AI Futures Model missed in the opposite direction: predicted 1.3x acceleration, METR found ~2x
austinc3301 · x · 2026-11-11
Austin Chen pushes back on Ramez Naam's critique of takeoff models, conceding past misses but noting the error has flipped: the conservative AI Futures Model expected 1.3x acceleration by now, while METR found 2x. Chen estimates the takeoff loop at 6% per ECI point (vs Naam's 2-3%), no longer 5-10x short of self-sustaining, and cites Anthropic's Opus 5.2x system card showing the one-time-jump model fit worse in 100 of 100 refits. He agrees parallel agents scale poorly but argues the AI Futures Model already discounts parallel labor, with takeoff hinging on research taste rather than agent count.
More from AGI Musings
- "I asked for an AI to do my job—should've specified keeping me" — kuanhoong · 2026-11-11
- 27B Model Can Beat Larger Models at Cyber, Hard Part Is Real Understanding — gnukeith · 2026-11-11
- A safety-talent startup tried this and died on program length — austinc3301 · 2026-11-11
- Safety community's best spend: a 2-year institution teaching LLM internals — viemccoy · 2026-11-11
- Reddit take: $200k GDP per capita is the threshold for a post-labor AI world — deferare · 2026-11-11
- Who owns scientific progress under gradual disempowerment? — najoungkim · 2026-11-11