Forethought's Tom Davidson: Even Without Foom, RSI Still Accelerates AI Progress 2–3x
HaydnBelfield · x · 2026-10-07
- Forethought's Tom Davidson distinguishes the "foom" and "fizzle" scenarios of recursive self-improvement: people discussing RSI lazily assume foom, but that's only one possibility.
- In the fizzle scenario, the recursive loop exists but ideas are getting harder to find, so RSI is less scary — though still scary: 2x to 3x the pace of progress.
- His core claim: "RSI is just scary, fizzle or foom," noting many frontier lab staff assume foom and conflate the two.
More from AGI Musings
- AI risk discourse's totalizing nature makes some act like the power-seeking optimizers they fear — foundjuliette · 2026-10-07
- The next Einstein or Newton will be an OpenAI or Anthropic model version, quips founder — clemnt · 2026-10-07
- Cohere Labs research suggests AI has yet to impact nearly half of workforce occupations — Cohere_Labs · 2026-10-07
- Math PhD now feels like swinging a scythe against millions of autonomous harvesters — FlorianGallwitz · 2026-10-07
- Dev observation: our AI outputs are becoming data packets passed agent-to-agent — aloncarmel · 2026-10-07
- Nobel economist Daron Acemoglu: making 100 people 5% more productive beats automating 5% — cen6wkf · 2026-10-07