Skaling

Skaling — an interaction exponent bridging Chinchilla's exponents and Kaplan's coupling

Why it matters

Breaks the assumption that N and D are independent, reducing prediction error by 1.5 to 3 times and extrapolating with about 10 times less compute

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2608.07222

Added in Week 35, 2026 · See that week

Same week

xAI releases Imagine Image 2.0OfficialMirendil signs $100M+ multi-year compute deal with Google CloudPressJeff Dean, Sanjay Ghemawat, Quoc Le and Oriol Vinyals found Discovery LoopOfficialQwen3.8-Max promises open Qwen-Max-class weights, but none are out at launchPress

Nearby on the same axis

DiffusionGemmaPaperResearchers release Boundary-Calibrated Intervention Transfer (BCIT)PaperFast Weight Attention for Continual LearningPaperSliding-window beats linear attentionPaper