Rafael Cosman comments on Consider using reversible automata for alignment research