AI Use
I guess this is my 'process?'
Let’s talk about how I use AI on here, because it’s becoming more and more important. This is going to be almost entirely a mechanical discussion. I think there are pages and pages to write about the philosophical and personal ramifications of LLMs, but those pieces will need to start here so I can skip the mechanics in the future. I’ll try to be brief, I guess, even if that hurts.
Artist
First, and simplest, almost every “hero” image I use on Substack is a generated image that Substack made for me. I feed in a caption, it goes and generates four pictures, and I pick the one that works (or amuses) the best. Sometimes I try multiple prompts, but this is where they come from. I’ll do that for this one too.
Yup. So bad. But funny!
Why? Because I only barely care about having an image for my pieces. But Substack works so much better with a picture. So I use their nifty and/or shitty image-generating AI service. It works for me, especially when it generates a hand with…four fingers instead of five. Heh.
Research
My brain is a deep mire of a swamp. Associations are fucking weird, at least for me. So I will frequently throw things at the AI and ask it to go find things for me. I am almost certain the imagery of a “swamp” is something I’m stealing from a quote I read decades ago, but I cannot find it. We just spent thousands and thousands of tokens together trying to find it.
I also use it to find the right “words” in Wikipedia sometimes - because I don’t organize my ideas like Wikipedia. I’ll give it a concept and say “find me the entries that go with this concept.” Essentially it’s super Google. I think this should be a thing that everyone does at this point - it understands context and personalization in a way that is completely beyond what “googling” something does.
Editor
This is probably the deepest part of my process. I layer the AIs over each other in a pseudo-variation of something called an ensemble method. To wit — I have a prompt that I will give to Google Gemini, and then Claude, and then ChatGPT. With each response I update my piece with the suggestions I agree with. Grammar errors are rare by the third of my “ensemble.”
Here is my prompt:
As a bonus I get to see how the different models analyze my work as well. It’s fascinating to see what they overlap on and what they pick up on differently. Especially for things like the “three most impactful sentences” which are frequently different for each model!
Not a Writer
It doesn’t write for me. What would be the point? Where would the fun be? The deep-spine reason I write is to order things in my head, put them out there on paper, give Squirt a lens into his dad’s mind, and maybe entertain a couple other people around the world. I’m the writer, not the calculator.
Conclusion
It’s a life-changing tool for me as a hobbyist. I’m far too prolific to have someone editing my hobby writing, and I couldn’t imagine asking someone to do so. But I’m also bad at grammar, so this second pair of eyes makes a huge difference. I have no patience for hunting down pictures and, indeed, I’m not sure an aphantasic could ever be much of an artist either. Googling is wildly inefficient for the research I need, at least as a starting point.
I cannot imagine ever doing this now without the tools that AI has put in my hands. It would be like imagining writing without having the internet’s information at my fingertips. It has become ubiquitous in my writing while, in my opinion, not impacting my creativity at all. I’m a better writer with AI, but I will always be the actual writer.