Just came across this paper/article that used character-based LSTM RNN’s to generate pretty believable fake restaurant reviews.
The paper:
And the hyped news article:
Interesting in terms of attack vectors, but please do not start a business doing this!