Multipolar AI is Underrated — LessWrong