LLMs represent something I’ve wanted to do since I was about nineteen years old, when I first published Rolling Sixes. There was a thing I was reading about called Natural Language Processing. For more information about Natural Language Processing, ask your AI. It’s what your AI was built on. It’s basically just using computers to interpret language. It’s computer-assisted linguistics.
Language is communication, of which words are a major, inseparable aspect. Through words we understand worlds because in a way everything is made out of words. AI seems smart because it knows how to put words together. To a degree it understands the syntax, semantics, and less so the semiotics of language, but its intelligence is limited by its methods. AI uses NLP to build a cunning edifice of words around concepts it doesn’t understand based on mostly statistical methods. It uses words to model ideas, but it won’t always give an accurate representation. People are generally dumb in the same way. Language, or our interpretations of it, gives us a similar intellectual condition. Considering how everybody seems to be treating the capacity of LLMs as some omniscient Ask Jeeves, AI’s use of language is exposing fundamental flaws in our comprehension of pretty much everything.
I love to see it.
See Robert M. Pirsig’s Zen and the Art of Motorcycle Maintenance for more information on the nature of Quality and Truth. It’s a start. That book, by some coincidence, I read when I was nineteen.
But that is besides the point. Natural Language Processing is at a point where an LLM can detect deliberate rhythmic structures of language and that is a beautiful thing. It’s exactly what I’ve always wanted. Of course there are issues, like how it needs different questioning tactics and a lot of coaching to get a decent picture out of it, but it works.
It’s a massive achievement in technology overall. I think it’s fucking sick that you can directly build images and video out of language now. I can direct my imagination into Veo3 and have something come onscreen in seconds, which is magic. There’s an argument to be made that written prose skips the whole ‘screen’ aspect, but that’s a whole different discussion.
I also think it’s fucking sick that I can now do what I wanted to do when I was nineteen and model, chart, and graph the absolute shit out of my books. I can get full quantitative analyses of my prose, breaking down the syntactic structures I used to build it. I can find out what percentage of my sentences were short vs. long, for example. I can find out how many adverbs I used. I can break down and model the grammatical conventions of a sentence like a decision tree. When I was nineteen I tried to do exactly this for Rolling Sixes, a book you shouldn’t bother reading, by the way. I tried to install some sort of NLP processing thing on my computer but made an absolute mess of it. This was on Linux, there were probably dependencies or something, I don’t know. This was over ten years ago.
Ultimately, I don’t think if I had got it working it would have yielded anything useful besides novelty and perspective. NLP as a field is so much further than that now. Now I can get deep on the juicy stuff, like morphology (fun), and use my own work to learn more about linguistics. This is exactly what I wanted to do.
Hopefully soon I can share some of this information, but to be honest I’m not certain of the veracity of any of it. I’m working to get confirmation, as everyone should when getting results from AI.
Okay. Good-day.
This post was not written with any AI consult.