Bronson Schoen comments on Training fails to elicit subtle reasoning in current language models