AI alignment with humans... but with which humans? — LessWrong