Default to jagged AI capability: path-dependence makes uneven skills the safe assumption
herbiebradley · x · 2026-11-23
Herbie Bradley argues AI's jaggedness across verifiable vs. less-verifiable tasks (or messiness in METR time horizons) is already clear. While operationalizing 'equal ability' across dimensions is hard, AI capabilities are highly path-dependent — contingent on data markets, infra, etc. — so we should default to presuming jaggedness, countering Toby Ord's 'jagged relative to humans' framing.
Related event: Oxford Philosopher and AI Researchers Debate 'Jagged Intelligence'(4 posts)→
More from AGI Musings
- Gary Marcus doubles down: Oracle stock down 60% since OpenAI deal he called peak bubble — GaryMarcus · 2026-11-23
- Delegating to AI is hard: capabilities shift fast and subtle errors are hard to spot — sqcai · 2026-11-23
- Taylor Lorenz: AI doomer crowd shows doomsday-cult-like panic that may turn violent — eigenrobot · 2026-11-23
- OpenAI and Anthropic are hiring life sciences roles, with Anthropic posting 5x more — dr_alphalyrae · 2026-11-23
- AI safety is bottlenecked by high-context senior talent, not talent — JacquesThibs · 2026-11-23
- 'AI reads well only to people who don't read' — philosopher pushes back on writing-quality takes — TuhinChakr · 2026-11-23