Coherent decisions imply consistent utilities — LessWrong