Closest stable alternative preferences — LessWrong