Deceptive Alignment and Homuncularity — LessWrong