Toby Ord: AI capability scaling coefficient γ sits near 0.66, but whether it has plateaued is unknown
tobyordoxford · x · 2026-10-09
Toby Ord notes we cannot yet tell whether γ — the coefficient mapping AI performance gains from more tokens onto human time-horizon gains — has plateaued at roughly 0.66 or kept rising. In theory it could even exceed 1, if AI gains more from 10x tokens than a human does from 10x time. He cautions the estimates carry large uncertainty: they come from a single experiment and rely on time-horizon, a famously tricky construct, and more experiments estimating γ are needed.
Related event: Oxford's Toby Ord Estimates AI Capability Scaling Factor γ at 0.66(3 posts)→
More from Research
- AI2's Nature Paper: Byteification Retrofits LLMs to Byte-Level for Under 1% of Pretraining Cost — TheTuringPost · 2026-10-09
- Ramanujan parallel: mathematicians call AI proofs a threat, not genius — CatAstro_Piyush · 2026-10-09
- Datology AI: Curated Pretraining Data Lifts 30B-A3B to 46.8% vs 37.7% Baseline — josh_wills · 2026-10-09
- Ai2 rebuilt its GPU scheduler: median queue wait fell from 5 minutes to 24 seconds on H100 cluster — allen_ai · 2026-10-09
- New COLM Workshop Paper Measures and Reduces Slop in Long-Horizon Coding Agents — dan_fried · 2026-10-09
- Cryptographer Lays Out a Balanced Take on AI's Impact on Cryptography — jedisct1 · 2026-10-09