Gary Marcus: OpenAI's research-acceleration data beats Anthropic's and shows no runaway superintelligence
GaryMarcus · x · 2026-11-19
Gary Marcus boosteds Ramez's analysis: OpenAI has released AI research-acceleration data far more detailed, granular and longitudinal than Anthropic's METR-based data — and nothing in it resembles runaway superintelligence.
- Citing Theo Jaffee: the OpenAI blogpost doesn't match what you'd expect on the cusp of ASI — agent workdays grew jaggedly, swarm sizes barely grew, token spending hasn't outpaced spending on human workers, experiments per researcher haven't accelerated, and internal ECI scores haven't improved
- Most surprisingly, task success rates for <15 min tasks have barely grown
- Marcus argues the METR team was handcuffed by Anthropic, which redacted the clearest evidence; the released data is insufficient for independent conclusions
- Given Anthropic expects an upcoming model to hit early RSI stages and cross a safety threshold, the responsible move would be releasing data matching OpenAI's granularity
Related event: OpenAI's Acceleration Report Praised for Transparency, Questioned on ASI(4 posts)→
More from AGI Musings
- Prediction: in 20 years no one will use non-personalized UIs — StewartalsopIII · 2026-11-19
- "China should stop trying to build frontier models": hot take sparks debate — rosstaylor90 · 2026-11-19
- Erik Hoel: philosopher Nagel's denial of LLM consciousness isn't an obvious mistake — erikphoel · 2026-11-19
- Philosopher Thomas Nagel singles out Anthropic's approach to AI as 'abominable' — GaryMarcus · 2026-11-19
- Zvi's AI #195: Anthropic IPO day, Claude Fable 5.2 vs GPT-6.2, Google's new models — TheZvi · 2026-11-19
- Bob McGrew: Superintelligence Needs the Physical World — the Hard Takeoff Is Wrong — No Priors · 2026-11-19