The main page displays prose generated by a state-of-the-art machine learning model (Open AI's GPT-2) fine-tuned on Infinite Jest, a novel by David Foster Wallace.
Part of each sequence of several paragraphs generated by the model is fed back into it as the prompt for the next original chunk of machine-generated prose.
This is a really fun use of GPT-2. Uncannily similar to DFW's own writing! And definitely got a hearty laugh out of long strings of "Jim. Jim, Jim. Jim, Jim, Jim, Jim.". Would love to see this extended further!
Thanks so much - glad to hear the Jim part, especially since it drove me nuts when I started generating text.
I wrote a paragraph for the FAQ on the over-fitting on Jim's dad monologues - not sure if it's true, but I think it's fun to ponder:
> For some reason the model picked on the monologues of Jim's (James Incandenza’s) dad. I found it funny that many sentences in some of the rambling monologues that the model is spewing end with ", Jim.". It's been a while since I last read the book, but I don't think these parts comprised more than, say, 5% of the book. It's interesting that the model fixated on them so much. My pet theory is that the model sees a rambling semi-coherent monologue it's producing, and then looks back at it and thinks "Oh, I must be inside one of these dumb Jim's dad monologues"
In the beginning I wanted to make it as close to being an Infinite Jest fanfic as possible. In hindsight - adding more of DFW works wouldn't have hurt. I might get round to it.
Part of each sequence of several paragraphs generated by the model is fed back into it as the prompt for the next original chunk of machine-generated prose.
The process repeats. The result is Entertainment.