At times it's just a little bit too much handwaving for my taste.
I like that I can quickly learn about advanced models since I already know about / can infer the intuition / mathematical details myself most of the time.
I dislike it because it's almost all handwaving. The way things actually work aren't really explained (other than at a high level) so either you're stuck with a wishy washy understanding or you will read the paper yourself.
But for some of us "in between", we're not satisfied with the layman's explanation, yet the paper is too formal to digest. I think there's an element of survivorship bias there, where some people just give up because they're not getting the explanations they're looking for.
Those who are smart enough to understand papers on their own can do fine, because they'll take the course as-is and just use it as a guide for what's new and fresh. Those who don't know what the heck is going on (and don't care) are happy with "plain English" explanations without looking any further.
They can't do everything, but what is most obviously missing is the "80% of data science" - data preparation/feature-engineering. The data scientists we encounter that are self-trained with courses like fastai are mostly wholly unprepared for dealing with enterprise data platforms, and, in particular, engineering data at scale.