Do you have a link to a good dataset of english-language NK propaganda? I've been playing around with char-rnn and that sounds like a good one to play with.

I agree the output here is lower quality than I would expect from an RNN, and quite similar to what I would expect from ngram-based markov bots.