Ramez Naam: AI self-improvement loop needs to be 5-10x stronger to sustain a fast takeoff
ramez · x · 2026-11-09
In a guest post on Noahpinion, Ramez Naam argues that recursive self-improvement (RSI) won't automatically trigger a runaway intelligence explosion. Calibrated on the few real-world datasets available, the self-improvement feedback loop would need to be roughly 5–10x stronger just to sustain itself, let alone run away.
- Current takeoff models (e.g., AI 2027) assume feedback-loop coefficients and exponents far stronger than anything observed empirically
- Narrow superintelligence exists today, but real-world research is much harder, and diminishing returns bite harder than fast-takeoff models assume
- Naam expects remarkably rapid AI progress by any other technology's standards, but not a sudden explosion to ASI — while acknowledging forecasters have repeatedly underestimated AI, and that better data is the key gap
A substantive entry in the takeoff-vs-diminishing-returns debate.
More from AGI Musings
- Everyone is automating white collar work — the real moat is blue collar — JoeJustice · 2026-11-09
- Cathie Wood: AI Costs Fall 75-99% a Year, CPI Badly Undercounts Deflation — rohanpaul_ai · 2026-11-09
- The models *are* the product: just add enough structure without over-constraining — ivan_bezdomny · 2026-11-09
- 'You Can Get Arbitrarily Smart Without Reading Books' — But Does Anyone Have an Example? — alexisgallagher · 2026-11-09
- Future CEO of agentic orgs will be a system designer, not a McKinsey alum — GregCook2011 · 2026-11-09
- Box CEO: agent swarms turn AI productivity gains from 20-50% into unbounded — inductionheads · 2026-11-09