Ben Todd: no evidence AI is slow on fuzzy tasks, doubling-time data doesn't correlate
ben_j_todd · x · 2026-11-13
Ben Todd pushes back on the claim that AI models make slow progress on "fuzzy" tasks. If true, task fuzziness should correlate with doubling times — but it doesn't. Current models are also already strong at many tasks once considered fuzzy, like legal advice and understanding product specs.
More from AGI Musings
- The Economist: AI model revenue is in the low hundreds of billions but data centers price in trillions — asusarla · 2026-11-13
- Oxford scholar on Jensen Huang's resignation: the paradox of power — CarissaVeliz · 2026-11-13
- Kai-Fu Lee's AI job-loss prediction by end of decade leaves LBC host stunned — kaifulee · 2026-11-13
- Geoffrey Huntley's one-line thesis: tokens scale, humans don't — kieranklaassen · 2026-11-13
- Paper grounds consciousness theory in brain dynamics: emotion decides what becomes conscious — daniel_mac8 · 2026-11-13
- Debate: why MIRI treated human values as intractable and bet on corrigibility — jd_pressman · 2026-11-13