I find this article moot. They are basically saying that applying a blank ML stamp can yield unsatisfying results on your data. That ought to be obvious and seems unnecessarily repetitive.