Lately everything deep learning seems to get upvoted, even rather mediocre stuff. I like deep learning and certainly think it is interesting but it is getting ridiculous.

I think the mediocre stuff is often necessary for people on the outside of relatively specialist areas - like I am - to have a way in. Not that I've been upvoting much deep learning stuff, but I'm sure I've upvoted similar.

I seem to recall that the same thing happened when HN had a fling with Bayesian statistics...last year? Two years ago?

I think it's a necessary consequence of having a large, broad technical audience that really enjoys investigating very narrow niches. Eventually even the people on the 'edges' of that topic's appeal want/need to find a way in.

A shallow treatment of a topic can still be excellent, if it picks the right stuff to present and does so in a pedagogical manner.

Going by reputation, the dragon book would be an example of the opposite: Very in-depth, but with horrible writing.