Chollet: science is the ultimate recursively self-improving system, yet progress stays linear
fchollet · x · 2026-10-10
Fchollet argues science itself — modeled as a goal-directed agent — is the best reference for understanding AI recursive self-improvement. Scientific inputs grow exponentially: researchers double every 15 years, global R&D spending every 13 years, science-dedicated compute every 2 years. Yet the rate of scientific impact has stayed roughly constant since the industrial revolution — progress is linear (life expectancy up 3 months/year since 1840). Root cause: the highest-impact, easiest problems get solved first; every subsequent problem is exponentially harder or lower-impact. He cites Nielsen & Collison (2018) and 'Are Ideas Getting Harder to Find?' (2020) as supporting research.
More from AGI Musings
- Liv Boeree: AI probably isn't sentient, but torturing it is still bad for you — HarperSCarroll · 2026-10-10
- Data centers could become SpaceX's biggest revenue stream, with $1T revenue on the horizon — Dr_Singularity · 2026-10-10
- Mathematicians stunned as OpenAI's Quasi-Riemann solutions rediscover their signature tricks — ctjlewis · 2026-10-10
- AI costs deflated in 3 years as much as PCs did in 15, a16z argues cheap intelligence reshapes software — a16z · 2026-10-10
- Cryptographers Urge Frontier Labs to Publish Negative Results on Code-Breaking — matthew_d_green · 2026-10-10
- Epoch AI: Frontier Models Achieve Only 15% of Human Paper's Gains in AI R&D Test — Epoch AI · 2026-10-10