Gary Marcus and Ramez rebut Dario's RSI claim: system cards show it hasn't started
GaryMarcus · x · 2026-09-13
Gary Marcus amplifies Ramez's rebuttal to Dario Amodei's "Pacing AI" letter, arguing the RSI claim fails under scrutiny.
- Both the blog posts Dario cited and the system cards for the latest Claude and ChatGPT models indicate RSI is not happening yet.
- The Mythos 5.1 system card reportedly states they aren't close to RSI, rating the risk low, and that even if RSI occurred it would hit steeply diminishing returns rather than trigger an unbounded intelligence explosion.
- The authors contend Anthropic's own released data undercuts Dario's framing; this is Marcus's second critique of the letter, following one in his newsletter.
More from AGI Musings
- HF Researcher: AI Poses Big but Non-Existential Risks, Would Leave Field Otherwise — soldni · 2026-09-13
- Merve Noyan: If Independent Evaluators Can't See Lab Mistakes, How Can They Catch Them? — mervenoyann · 2026-09-13
- Damodaran Calls AI Slowdown Pact 'Suspicious': Secret Admission of Training-to-Inference Shift? — pdamodaran · 2026-09-13
- 2009 NYT: scientists already debated AI risk at Asilomar, 16 years ago — erichorvitz · 2026-09-13
- Businesses Don't Need 140 IQ Staff, They Need Automated 80-100 IQ Agents with Context — ericwdolan · 2026-09-13
- e/acc vs d/acc: full Beff Jezos vs Vitalik debate on slowing AGI — beffjezos · 2026-09-13