Training an AI to write BuzzFeed clickbait headlines

Am I missing something here, or would a mark of chain be much easier to write and solve the problem just as well?

Seems like much less effort. Of course, maybe the author wanted to work on an RNN instead?