Toby Ord disputes METR's 1.9x slope: it ignores the pre-training jump
tobyordoxford · x · 2026-11-06
Toby Ord argues METR's 1.9x slope estimate assumes zero jump from the Mythos pre-training run, while his own 1.4x estimate accounts for both the jump and the slope change, making it a better estimate.
More from AGI Musings
- AI is coming for analysts who only wait for clean data, says data lead — Nasereliver · 2026-11-06
- 'Why learn if AI can do it' is dangerous — deep engineers stay most valuable — Abhishekcur · 2026-11-06
- AI stans can't accept informed criticism: "criticism is a narrative violation" — menhguin · 2026-11-06
- Science: cracks emerge in Lean certificates backing AI-generated math proofs — neurovium · 2026-11-06
- AI 2027 tracker claims 85% prediction accuracy, says progress is ahead of schedule — connoraxiotes · 2026-11-06
- Trivial for an ASI to serve humanity forever? The pillbug analogy says otherwise — danfaggella · 2026-11-06