I noticed two major things in this article.

First, Twitter can't do what the students are doing - an informal, unreliable service going "hey , bot!" is fine for some random people to offer, but the site owner can't chance high error rates on public statements. Twitter keeps suggesting they're doing something like this internally, which would make sense.

Second, the article offers no real evidence that this service works. It's a machine learning problem with no proven classifications in the dataset, so the learner will at best reproduce the developers opinions of what a Russian bot looks like.

I'll play around with the tool, but my initial guess is that it just learned "patriotic, inflammatory language, retweets only" as "bot", which isn't a breakthrough anyone can actually act on.