Arbital postmortem — LessWrong