Joshua Clancy comments on How useful is mechanistic interpretability?