Why the author is now more bearish on LLMs for math

JacquesThibs · x · 2026-07-20

The author says they have become more bearish on LLMs for math than they were two years ago.

Their view has shifted after watching what these models actually solve, and more importantly what they still fail at: out-of-distribution generalization remains weak, so progress in math does not look as broad or reliable as earlier optimism suggested.

Related event: Debate on LLM Math Capabilities and Generalization Shortcomings(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →