Do LLMs Truly Generalize? Skepticism Arises Over the Scaling Law Myth

JacquesThibs · x · 2026-08-11

The author questions the widespread belief that LLMs will magically generalize with increased scale. He argues that people rely too heavily on the idea that "scale leads to emergent capabilities," noting that we haven't seen true out-of-distribution generalization yet, as models mostly perform interpolations of their training data.

Related event: Industry Questions LLM Generalization and Scaling Law(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →