textgenrnn

It should be noted that per the included Jupyter Notebook, the resulting network took 24 hours to train, and the resulting network (of a 3-layer 512 cell LSTM) would be hundreds of MB. (too much to share on GitHub anyways). The approach in the repo also uses a word-based approach instead of a character-based approach. Although a word-approach avoids typoes/nonsense, it makes prediction much more difficult since the selection space is harder. (from notebook, error is high as well). There are lots of trade offs in this field. Training these networks is not for the faint of heart and one of the reasons I developed a tool to train text generators using a pretrained RNN (https://github.com/minimaxir/textgenrnn) on a much more simple network architecture that trains quickly, although it probably would not hit the depth of all of Game of Thrones

As someone who's cited by the OP and coincidentally released a tool today that easily trains a LSTM on any dataset demoed with Hacker News data (https://github.com/minimaxir/textgenrnn), I have a couple comments: 1) My Get All Hacker News Submissions script is somewhat obsolete since all HN data is now on BigQuery (will add a note to the README today) 2) There is a massive, massive selection bias in terms of the quality of the selected generated texts. If you look at the sample output (https://github.com/danhon/deep-hackernews/blob/master/sample...), there is a large disparity in quality between the chosen submissions for the blog post and a typical generated submission. (this is expected, although the need for human curation damages the "Turing test" suggestions proposed by other commenters) Compare with the sample output at similar temp