I first read about it on Reddit, followed shortly by the CANLIB-DATA Listserv, but as of today Google has a new search engine dedicated to research data sets, the cleverly-named Google Dataset Search.
The good: Surfacing this stuff is great! Google is using schema.org to discover stuff, and has a pretty extensive page on how this all works. Results link through to Google Scholar to show who has cited a dataset. Likely not comprehensive, but a good start. Oh wait, it's not very good at all - I thought it was linking to the DOI, but it's just some sort of keyword linking. I just found a declassified Los Alamos report from 1957 in the top spot that supposedly links to one of these datasets :-/ Right idea, totally the wrong approach.
The annoying: Just as with Google Scholar, there's no way to know exactly what is and isn't being indexed. Also annoying not to have a count on the number of results. I can't get the "share" button to work, but that may very well be something specific with my browser and some extension - not a huge deal right now.
The weird: Of course I did a search for MPOW, and "University of Calgary" auto-suggests to a record about our institutional repository, but nothing from within our IR. Do we not conform with schema.org? (entirely possible). Why is that link from the French version of the National Research Council Canada?
The bad: No filters or facets of any sort - boo!
I have already found a couple of datasets of interest, and one that eventually led through to a deleted dataset, making me wonder how fresh the index is.
Definitely one to watch!