Work harder on tabooing "Friendly AI" — LessWrong