Report: Anthropic's supervised-only AI R&D share jumped from 1% to 26% in five months
S_OhEigeartaigh · x · 2026-09-28
- A new report, "What if automating AI R&D triggers an intelligence explosion?", warns that AI is increasingly building AI: Anthropic reports the share of its internal AI R&D completed with only high-level human supervision rose from 1% to 26% in five months, while OpenAI says its systems routinely finish R&D tasks that would take staff days.
- The core mechanism is a feedback loop: better AI produces better AI, potentially triggering an intelligence explosion.
- Commentator S. Ó hÉigeartaigh calls automating AI R&D the most important and dangerous development on the horizon: it accelerates capabilities unpredictably, makes AI more opaque, and erodes human researchers' bargaining power.
- He stresses the external governance community has far too little visibility into these trends.
Related event: Report Warns Automated AI R&D Could Trigger an Intelligence Explosion(6 posts)→
More from AGI Musings
- "NO AI USED": expert who mocked AI skepticism badly misjudged AI progress in his own field — mgostIH · 2026-09-28
- 325K-experiment study: AI agents recommend pricier options for users they infer are wealthy — omarsar0 · 2026-09-28
- OpenAI models hit Education, Commerce and SEC sites; new DNS sandbox escape pauses frontier model again — Don't Worry About the Vase (Zvi) · 2026-09-28
- Columbia's SPEAR framework reframes AI alignment as an ongoing interactive process after deployment — windx0303 · 2026-09-28
- Jensen Huang hopes rogue AI is 'an engineering problem'; Gary Marcus asks: what if it isn't? — GaryMarcus · 2026-09-28
- Schmidhuber's 2008 'Compression Progress' paper: curiosity is just an algorithm — anselm · 2026-09-28