Defining outscaling: token multiplier rising along the overtraining axis
_katieeverett · x · 2026-09-10
The thread defines outscaling: at fixed model size, the token multiplier is the ratio of tokens a baseline optimizer needs versus the compared optimizer to reach the same loss. Outscaling means this multiplier increases along the overtraining axis, reflecting a better token-decay exponent, a lower high-token loss limit, or both.
More from Research
- arXiv paper "Agent Meltdowns": overly helpful agents turn benign errors into unsafe actions — alex_verem · 2026-10-27
- Cornell paper: 64.7% of agents take unsafe actions after hitting ordinary errors — alex_verem · 2026-10-27
- LightMTP: lightweight latent multi-token prediction boosts LLM planning in pretraining — gdm3000 · 2026-10-27
- EMNLP study: AI literary translation reads fine but varies more, even within one book — tuvllms · 2026-10-27
- Lab monkeys at $30-80K each: a 4-year breeding cycle is the real preclinical drug bottleneck — zebird0 · 2026-10-27
- OpenAI Foundation funds UNC tumor antigen project, hiring senior experimental scientist — iskander · 2026-10-27