Dario claims RSI is starting, but Claude and ChatGPT system cards say otherwise
paulnovosad · x · 2026-09-13
- Dario Amodei's Pacing AI letter claims RSI (recursive self-improvement) is beginning.
- ramez pushes back: the blog posts Dario links, plus the latest Claude and ChatGPT system cards, make clear labs are not close to RSI.
- Anthropic's released data suggests RSI would hit steeply diminishing returns rather than an unbounded intelligence explosion.
- HumanLevelJen adds a tldr from 6 years of working on why RSI is hard, especially for big labs.
- Takeaway: real disagreement inside the field, with evidence leaning toward "not yet".
Related event: Researchers Push Back on Dario Amodei's Claim That RSI Has Begun(8 posts)→
More from AGI Musings
- Stop dismissing Europe's AI catch-up efforts, argues founder in viral post — DominiqueCAPaul · 2026-09-14
- Anthropic researcher: we're entering a regime where alignment evals provide almost no evidence — davidmanheim · 2026-09-14
- Researchers mined 1.2M US STEM dissertations to map scientific human capital production — Afinetheorem · 2026-09-14
- Pompliano mocks AI doomerism: if it can kill humanity, why not profit first? — nikvassev · 2026-09-14
- Nobel laureate Banerjee joins ETH Zurich panel on how AI will change economics — soumitrashukla9 · 2026-09-14
- Activist heads to DC to push for a ban on superintelligence and recursive self-improvement — erikphoel · 2026-09-14