Sure!
I didn't train it myself, though it does start to use the book in progress as context once it gets going.
I used a combination of sudowrite and inferkit. Sudowrite was better for bulk generating the entries that make up most of the book, and inferkit was better at fleshing out those entries with more detail.
Most of my labor was curation, as I threw out about 95% of what GPT-3 generated. This is important, as what you keep influences what gets generated next.
Hopefully that answers your question, but feel free to ask more!