I get the feeling that the truth is that the blog just outputted a ton of "deep dream" like text fragments.
While the author makes it sound like the bot created a long and interesting text that could qualify as a blog post.
https://drive.google.com/folderview?id=1lgaXfKNTS_fm2fU9wyaS...
As you can see for each run and training step, the model seems to generate more believable content just like blog posts written by actual people.
The author has only posted chosen content from the whole array of text the bot generated, to appeal her specific audience of book bloggers, authors and readers.
I do however, have all the content with me, until 500 training steps, after which I stopped the model from running further. I think I'll share it in this thread in a while.