Raw LLM Responses

I’ve been tinkering with similar processes but haven’t dedicated the time and energy to doing a whole book yet; props to you for that. The process I’ve been gradually working out in my head is pretty similar to yours, but the first and last thing of yours is a great catch. I’ve also been discussing with CGPT (mostly GPT-3, only recently got 4) various useful ways to talk about writing. One tip I got from it that I didn’t pick up in any of my graduate programs was to ask it to increase the “texture” of the prose, which it defined as something like vividness of sensory details and word choice. I also found some success while iterating a scene by using techniques from hypnosis (of all things)—specifically, I asked it to imagine dials labeled “creativity,” “vividness,” and whatever else—I don’t remember now—each with the default setting of 1, and 2 being twice as high, etc. Then I’d iterate on a scene by asking it to turn this up to 2, that down to 0.5, crank this up to 11. Pretty neat results there. I’ve found it’s very good to give it an author to emulate stylistically, or to ask it to analyze an author’s style and then feed that description back into it as a style to emulate. Also had much success asking it to create hybrid styles of multiple authors, and my favorite—asking it to analyze the aesthetic of a musician and then adapt that aesthetic to a literary style. Other than dropping direct references to the musician at the end of each output even when I explicitly told it not to, the stuff it came up with was really good—a very original and creative voice, in my mind. (For reference if you want to try it yourself, I used Glass Animals as the musical reference)