Original video: Hassabis on RSI phases, from 3-5x engineer boost to closed-loop self-improvement
haider1 · x · 2026-11-06
Haider shares the original video of Demis Hassabis discussing the two phases of recursive self-improvement (RSI): AI currently makes engineers 3-5x more productive with humans in the loop, while the next phase — closed-loop RSI where AI improves itself without human review — will be faster and far riskier on the path to AGI.
More from AGI Musings
- Muse assistant sparks 'AGI feeling': Meta researcher boosts review saying you don't need the best LLM — ThomasScialom · 2026-11-07
- 'Every solution path ends at AI will solve it': the creeping AI-fixes-everything mindset — djcows · 2026-11-07
- Congresswoman calls to pause advanced AI and ban superintelligence; Yudkowsky backs her — ESYudkowsky · 2026-11-07
- Agent order numbers don't add up: platforms report single digits, users claim hundreds — mobileraj · 2026-11-07
- Analyst: agents scaling across enterprises will unleash a storage boom as 80% of unstructured data gets contextualized — BenBajarin · 2026-11-07
- "Humans were never very good at research either": AI verification argument resurfaces — RexDouglass · 2026-11-07