x
Alignment as Constraints — LessWrong