Gradations of Inner Alignment Obstacles — LessWrong