Chollet: by experience-to-competence efficiency, current AI is ~6 OOMs behind humans
GregKamradt · x · 2026-09-15
François Chollet, quoting OpenAI capabilities researcher Dan Selsam's public AI-risk statement, reasserts his definition of intelligence as the efficiency of converting experience into competence — and argues current AI is roughly 6 orders of magnitude behind humans by this metric.
His evidence:
- Human evolutionary history never prepared us for programming, yet a person learns Python in a few hundred hours; an LRM needs the equivalent of 1B hours of training data (on top of non-programming data) for the same competence.
- Test-time and energy efficiency show a 3-4 OOM gap: a human solves an ARC-AGI-3 game for under $0.10 of electricity, while Astra costs about $300-$400 per game.
The quoted Dan Selsam statement: 15+ years in AI — early probabilistic programming work at MIT, Lean theorem prover co-development at Microsoft Research, pioneering neural reasoning demos in his Stanford PhD, and nearly five years at OpenAI on chain-of-thought optimization — sharing his personal views on AI risk.
More from AGI Musings
- Noam's 2-year-old bet that general models would beat humans on a benchmark pays off — GregKamradt · 2026-09-15
- 'People Who Say We Need to Nuke SF Are More Hypocritical Than OpenAI' — wordgrammer · 2026-09-15
- e/acc's Beff Jezos Argues Capability Diffusion Is Safest, Slams Anthropic's Closed Approach — beffjezos · 2026-09-15
- The User-Assistant Format Is an Illusion: Why Persona-Based AI Alignment Likely Won't Work — mayfer · 2026-09-15
- Clarifying the AI safety split: safety testing time vs long internal deployment — JacquesThibs · 2026-09-15
- a16z partner puts P(abundance) at 99.99% — nptacek · 2026-09-15