Toby Ord: OpenAI's RL scaling curves show no smooth slowdown, just logarithmic gains
tobyordoxford · x · 2026-10-15
Toby Ord joined a debate on RL compute scaling: his analysis of OpenAI's scant data last year found no detectable smooth slowdown—scaling looked purely logarithmic. The attached chart shows o1/o3 AIME Pass@1 rising steadily with compute. Ramez Natarajan countered that several papers find RL follows a sigmoid on a log scale, performing even worse than log scaling.
More from Research
- Global Transport coupling lifts guided flow generation: ImageNet FID 2.06 to 1.91 — mmbronstein · 2026-10-15
- Your breathing gives you away: Wi-Fi signals can reveal your breath rhythm — TinfoilTricorn · 2026-10-15
- Beyond text correlation: attaching a geometry simulator to ground model reasoning — gerardsans · 2026-10-15
- UniForge3D turns sparse-view images into high-fidelity 3D meshes, code released — camenduru · 2026-10-15
- A Ticket-Routing Model Predicted Human Choices Across 2,416 Game Situations for Six Cents — RexDouglass · 2026-10-15
- Shallow LLM guidance beats deep injection in VQA: 39.4% vs 31.8% on OK-VQA — Aneerban Saha · 2026-10-15