Relative likelihood — LessWrong