[gwern]

I would say easily a third the time I need a paper, I'll have to get it through Libgen, and especially if a paper is new or published through Elsevier or Nature. (If you've read through my comments and noticed me posting Dropbox links to fulltext PDFs - Libgen in action.) It's also good for older papers pre-2000 where it's unlikely the author has put up a copy or someone else has hosted it, and finally, it's good for getting finalized copies of papers as published since preprints often change quite a bit (economics papers in particular seem to spend years hanging around going through many versions which can change conclusions dramatically).