Toby Ord: Most AI capability metrics grow logistically, not exponentially
tobyordoxford · x · 2026-10-09
In a discussion with François Chollet, Toby Ord argues that while famous AI capability measures like METR's time horizon have grown exponentially — and revenue arguably qualifies too — most capability metrics do not show exponential improvement over time.
He adds that such metrics grow roughly linearly on a log-odds scale, which is essentially what ECI does — implying each metric follows logistic rather than exponential growth, slowing as it approaches its ceiling.
Related event: Oxford's Toby Ord rebuts Chollet on exponential AI progress(3 posts)→
More from AGI Musings
- OpenAI Researcher: Econ Theorists Have the Best Intuitions for Managing Agent Swarms — willcb · 2026-10-10
- Terence Tao: The Era of "Big M" Mathematics and "Big P" Physics Has Arrived — CatAstro_Piyush · 2026-10-10
- Sakana AI Researcher on Why Artificial Life Ideas Matter More as Autonomous AIs Proliferate — kenneth0stanley · 2026-10-10
- NBER: Machinery's Skill Bias Comes From Suppliers — High-STEM Demand Rises 45% Externally — soumitrashukla9 · 2026-10-10
- Quanta: as AI cracks math's hardest problems, mathematicians face existential reckoning — MacrinePhD · 2026-10-10
- When AI Writes the Code, What Replaces Design Patterns? — JnBrymn · 2026-10-10