I agree, though they do make the pretty strong statement of

  NNAEPR suggests that one may abandon using NNs altogether, and simply use PR instead.

in the blog post, which sounds to me like they're going for a finding.