Neel Nanda: Astra's Gains Look Like a Looping Model, Not General Progress

NeelNanda5 · x · 2026-09-11

Interpretability researcher Neel Nanda argues that Astra's disproportionate improvement on no-CoT reasoning — rather than the general capabilities we actually care about — suggests something unusual, likely a looping model, rather than genuine general capability progress. He links to his full analysis.

Related event: Astra's No-CoT Reasoning Surge Raises Safety Concerns(7 posts)→

Original post →

More from Models

Models channel →