OpenAI's own acceleration data undercuts the ASI-is-nigh narrative
joshua_saxe · x · 2026-11-19
A thread disputing OpenAI's research acceleration blogpost: agent workdays grew jaggedly (8.2x pretraining vs 2.7x other), swarm sizes barely grew, token spend hasn't outpaced labor costs, experiments per researcher and internal ECI scores show no acceleration, and sub-15-minute task success rates are nearly flat. Author speculates Anthropic may be far ahead in automating AI R&D.
More from AGI Musings
- When AI makes content worthless, authenticity and trust become the moat — kylegawley · 2026-11-19
- tszzl: models rewriting math and physics still fail at simple common sense — tszzl · 2026-11-19
- How mode collapse emerges: RL bias meets in-context learning feedback loop — jessi_cata · 2026-11-19
- NYT: Anthropic's first IPO cash priority is buying compute to reach recursive self-improvement first — Polymarket · 2026-11-19
- Gary Marcus amplifies critique: Anthropic's winner-take-all AI view is 'the most dangerous' — GaryMarcus · 2026-11-19
- Stanford prof Anshul Kundaje: peer review shouldn't funnel public funds to publishers — anshulkundaje · 2026-11-19