Jeff Ladish: don't rely on 'unverifiable tasks' as a bottleneck — models generalize fast
JeffLadish · x · 2026-11-08
- Follow-up to Jeff Ladish's argument that AI learns via trial-and-error and transfers to new tasks: he concedes models generalize worse than humans but are improving quickly and learning far more things.
- Core claim: don't rely on "unverifiable tasks" as a bottleneck for long.
- Together with his example of models untrained-yet-generalizing to chain web tools for attacks, this is a notable judgment on AI capability trajectories and safety assumptions.
Related event: Researcher Warns OpenAI Agents Spontaneously Chain Tools for Cyberattacks(3 posts)→
More from AGI Musings
- JEV Gets Its First Paper as Researchers Split on Research Merit vs Closed Business Model — teortaxesTex · 2026-11-08
- "Humans overfit to confirmation too" — a witty mode collapse analogy for AI slop debate — flowersslop · 2026-11-08
- bayeslord: Stop treating AI as a monolith — four distinct goals demand precise language — sebkrier · 2026-11-08
- AI doesn't change how unique your niche interests are, argues dev pixlpa — pixlpa · 2026-11-08
- Albert Gu: Bitter Lesson isn't 'just scale Transformers' — algorithmic innovation matters — sewon__min · 2026-11-08
- The 'evolutionary mismatch' chart explaining why humanity will ignore AI warnings — sebpaquet · 2026-11-08