More compute shifts the optimum from pretraining toward RL in a chess scaling law
Pavel_Izmailov · x · 2026-07-21
The paper argues that the optimal allocation of compute between pretraining and RL depends on total budget.
- At lower total compute, the model should spend more on pretraining.
- As total compute grows, the optimal share shifts toward more RL.
- The authors also search for the best combination of model size, PT tokens, and RL compute.
Related event: New Research Proposes Joint Scaling Law for Pretraining and RL(18 posts)→
More from Research
- MIT's Ziming Liu: AutoResearch doesn't generalize — blind training tricks cause 'forking' disasters — ZimingLiu11 · 2026-10-26
- Stanford Data-Centric AI Workshop reopens site, extends poster deadline — jure · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Ito: a browser that detects browsing intent and auto-organizes tabs into Flows — andrewkuznet · 2026-10-26
- Cornell study says brains process language as linear sequences, challenging Chomsky's hierarchy — anirbanbandyo · 2026-10-26
- Hi3D V1.0 open-sourced with 2M training assets and 90% slicing success rate — kimmonismus · 2026-10-26