x
Survival Bias in LLMs: Your Model Judges AI Self-Preservation Differently Than Human Self-Preservation — And I Found The Circuit — LessWrong