Hacker News Reader

Wait, I thought they wanted to avoid CoT monitoring, in order to avoid models learning to conceal/encrypt their thoughts.

Hopefully subterfuge would be flagged during the planning stages. Hopefully.