Getting from an unaligned AGI to an aligned AGI? — LessWrong