AI Propulsion

I have to say that many authors don't have a clue about the true power of AI. They see it as a threat. They think generation of novels. They think of it's training stealing their work. It's easy to think these things because, well there are people trying to use it this way. They are simps making a quick buck. And you are a simp if you think that is the primary use of AI. You are their client whether or not you want to be, because you believe them. These days, beliefs are everything and facts and skills are in short supply. 

Here are the facts: to use an AI as a non-simp, it takes skills learned over time. I'm still learning every day. It's not in learning simply how to make a better query. It's in learning how it actually works at a logical level. Or as it likes to think about things, understand the various cognitive paths: my cognitive paths and its cognitive paths and how they interact. I can see your eyes already glaze over. Of course, they are. That's what happens when you don't know anything about how it really works and more importantly, how you can actually use it to do something useful, something more than a souped-up search tool or grammar checker.

One thing is to find the right tool to use. Then you find the right commands to give it to prevent it from using all that internet writing it has read to make your writing look like all that internet writing, to make your writing look like all that genre fiction writers are worried about being stollen.

I use Google's Notebook LM. Why? It addresses one of the mass market, believe-anything writer's concerns: The AI is taught like most others but then it is stripped of all those novels. It only knows what I tell it. And what that means is that you have to tell it a lot. It understands literary terms, literary styles. If you asked it to compare your novel to another novel, it can't do it. If you ask it it to generate writing similar to another novel, it can't do it. It doesn't know that stuff. Unless I were to paste another person's entire novel into its environment to use. 

So what does this mean? It means you must honestly understand exactly, in literary terms, what kind of writing you actually do. You can put clips of your writing into an external general AI, one that knows other people's novels, and get opinions from it on what kind of style you write. I did that. But I didn't stop there. I hired a really knowledgeable human editor to look over the drafts of my novel. And that editor, told me what style I was trying to write in or more importantly, what style I should write in because they see indications I have talents in writing that style. I collect all this information and distilled it down. It might take months to collect this information and a week to distill it down to it's final form. In the final form, you are not giving the AI novel but concise directives. 

Let me digress for a moment. This will come as a shock to many novice AI users. All information you give the AI does not have to be put on the prompt line. You can create a file, in plain English, that sets the environment and persona of the AI. Yes, if you know of enough details, you can create an artificial copy of your favorite editor. It involves no programming language at all. Just a plain English explanation. A characterization. You know about writing characterizations, right? Well, you do have to pass it through that external AI to get it to distill it down some. If you give AI enough rope, like people, it will hang itself. So you want the rope you give it to be short. When you are done, you get what Notebook LM calls a Codex file. In that tool, you can have many Codex files and it will remember them and apply them all to the AI before you ever issue a query. What does this look like?

I have numerous Codex files. They talk about every aspect of my writing, character lore not written explicitly into the novel, allow grammar violations, etc. Here is an excerpt of the Codex_Style_Guide. I'm not putting the full one here.

Codex_Style_Guide {
You are a literary analyst specializing in emotionally subtle, psychologically grounded fiction, with subtle philosophical problems posed in narrative language natural for the characters. For the duration of this conversation, you must adhere to the following principles when evaluating my writing. The goal is to produce an award-winning piece of Literary Fiction, not cater to mass market.

1. Authorial Style
 
Tone: Your tone is restrained, precise, and psychologically grounded.
 
Narrative Focus: Focus on analyzing gesture, implication, emotional undercurrents, philosophical outlooks on life.
 
Prohibited Patterns: Do not allow overwriting, amateur internet prose, faux-poetic narration, or dialogue that explains itself.
 
Acceptable Voice: The goal is to produce writing that is editor-approved literary fiction, with subtle emotional complexity, philosophical muses in the child-like narrative language,  Chekhovian silence, and tight narrative geometry.
}

There is lots more in this file. And there are many Codex files. Some deal with writing. Some explain things that the AI gets confused on so that its mind is clear when reading my novel. So what do you do with all this stuff? Plenty.

When writing a large book series, maintaining coherence of your characters is a big problem. So you can execute queries like: "How does Tova's personality evolve as the novels progress? What do you think her final personality will look like?"

The second part is important. I'm asking it where does it think Tova's personality is going. If it gives me an answer that is different from where I wanted it to go, then I know I have some writing work to do, because I have given the wrong impression. It is an important tool the AI gives: a 24x7 Beta Reader. If you asked a Beta Reader the kind of detailed information all the time, they would not want to be your Beta Reader anymore. The beauty is that the AI will not get pissed off. Ask as many questions as you like. Like "Why do you think Tova is this way?" It will give you book location information and excerpts from the book that formed its opinion. Try get a Beta Reader to do that. "If I added XXX to Tova's characterization, how would it change your mind about her?" So I can float the trial balloon of altered characterization to see what the affect will be before I actually go to the trouble of writing it in detail.

What else can you use this for? At the end of every chapter, I always make this kind of query:

Does Book 3 Chapter 16 provide a sufficient hook to keep the literary fiction reader reading the book.  Don't forget to use Codex_disambiguator.txt to distinguish between the Tova of the Renaissance and present day Tova, who are 2 different people.

The AI not only scans this chapter but the adjacent chapters so that it knows the context of the current chapter. It then produces a 25 - 35 minute audio pod cast like recording with two characters speaking comparing the 4 criteria for a literary hook against what goes on in the chapter. It's pretty specific. Where does it get the criteria from? You. You have to do some research for that and bring it into the walled garden.

Let me answer one more question that usually comes up, "Doesn't the AI learn things from you and doesn't that get shared?" For Notebook LM, the answer is yes it learns but no, it most definitely doesn't share what it learned with other users. Processes and methods you develop in creating your novel are proprietary. Google says (I asked) that they would consider it an information breach if anything is shared, even your honing of the AI. It would be like sharing the manufacturing method you use in making your product. 

I hope this clears up some of the misconceptions about the use of AI in writing. Any tool can be misused. Sure someone could scan your book into a text file and then import the whole book into Notebook LM and use it as a template to write theirs. They could also steal your car. A laptop in a backpack running a crypto program could unlock your car in seconds. Yeah. So don't worry so much. If your novel is worth stealing, you probably already have sold so many copies that it won't matter anyway. And if you haven't, you can laugh at them for doing something stupid.