Not the entire thing. You can, however, use GPT-3 to write some passages of a book, as I did for the book I wrote on the topic.
Some tips I found: You get the best results by just asking the right questions on DAVINCI INSTRUCT with a high temperature. You can also get more dynamic prose by using the penalties (forcing it to find new words to describe concepts, which is just good writing practice).
For instance, if you wanted to write a book about nihilism:
Are we in the middle of an nihilistic crisis?
It seems clear that the world now confronting us is one of obscure, diffuse power; shared, increasingly, by transnational corporations and global networks of government. This network, composed of techno-scientists, market analysts and corporate lawyers as much as bureaucrats carries on a conversation without becoming vocal. It seems inclined to push further ahead with the conversion of the global community into one immense, (technologically) seamless marketplace. Its head, meanwhile, having turned its back on the Enlightenment notion of progress is now moving further away from it. The high priests of this Euro-American civilization are battling any attempt to pass a viable progressive proposal in the international arena today beyond …tax cuts for big business.
Writing fiction is much harder because writing a long story requires a tremendous amount of planning and continuity, which is impossible to stuff into the 2048 token limit (I’ve tried). However, nonfiction tends to be easier to compose through multiple prompts because you can just string together many passages such as the above.
Oh you mean hundreds of separate questions and results. Yeah, you can probably fudge most of the way through a book like that. I found that, sometimes, it really struggles with new concepts. Even if you define a novel concept, it’s very likely to ignore it and just start talking about something quasi-related.
Yeah so the major trick is getting to the point where you can use a main theme and then automate the entire process of exploring different primary topics, generating questions and answers, and then follow-up questions. If you think of it like a tree data structure that might help.
And yes as Boris said, you’ll need to do a lot of editing, structuring, etc. Writing a book is no joke.
I’m the author of five very much human-written books. My short answer to this question is, I have no idea.
But I can say that I think the answer is very much the same as it is for a human writer: focus on the process, not the end product.
This means using GPT-3 to create the building blocks of the book, but not trying to complete the entire thing in one fell swoop. I see AI as a way to augment the writing process by generating sentences and paragraphs. It’s not a replacement for the writer so much as an augmentation. A human/AI hybrid writer can write a new type of book, not replicate literary forms that were developed strictly within human writers’ minds.