Yep, it uses the same technique. I've pulled every comment and story from HN through the API and make a bunch of Markov chains to produce story titles and comments. The input corpus is a lot smaller than subredditsimulator though and the various subreddits have a large variation in the words used (meaning more funny comments), so it's no way near as good.

Perhaps a future enhancement might be to partition comments by the submission domain, i.e use only comments for github.com stories for fake github submissions, might generate very different text than a munge of every comment.