How to fine tune gpt3 on raw text without prompt

Hi, I knew this question has been discussed a lot last year. But as thing have been changing continuously i want to start the discussion again and want to know is there a way to fine tune on a totally new book to make sure our tuned model know about that book. I am not looking into semantic search, that thing can be done via langchain, but my focus is to tune model to knew about the context of book. so when next time i fine-tuned on a relevant task with prompt, its well aware of all the context in book.

If you have fine-tuned a model that is operating as expected and responding in the way you want, you can continue to add new weights to it with new training data.

A whole book is a bit much, because every new prompt:response isn’t just knowledge, but a representation of the expected response. Putting pages of a book into fine-tuning would require synthetic questions to evoke the “book” response that don’t break the current operation, like “what is page 5 of Hound of the Baskervilles”, “where in the book does the detective do {a piece of minutae}” or other unnatural things. You’d have a chatbot that is 20% how to chat and 80% how to replay sections of a book.

One might need “AI fact extraction” to prepare the data into a form that is like the type of answers you wish to solicit.