Rereading Nostalgebraist's excellent critique of Anthropic's research into Agentic Misalignment, it just struck me that if we want to avoid ineffectual roleplaying sessions, we *don't want to do this with a checkpoint that strongly believes it's Claude*. We need fresh minds.