Analysis: Debunking 'Copycat' Claims on Chinese Labs & Deep Dive into Scaling Law

GaryMarcus · x · 2026-08-20

Gary Marcus shared an insightful analysis on optimizing model improvements and solving different bottlenecks, arguing that the notion of Chinese labs catching up solely by copying is a mistake. The referenced content delves into the nuances of Scaling Law, highlighting that parameter count is meaningless without context on data volume, compute allocation, and deployment conditions. It contrasts the findings of Kaplan et al. (2020) and Hoffmann et al. (2022), illustrating the industry's shift from parameter-heavy scaling to compute-optimal strategies.

Original post →

More from Companies & People

Companies & People channel →