A single principle related to many Alignment subproblems? — LessWrong