Ultimate ends may be easily hidable behind convergent subgoals — LessWrong