My AI stance.

It’s shit.

The End.

It’s more nuanced than that; as are most things.

I, like many got caught up a bit when it first arrived. I messed around making images, and looking into the other things it could do. I tried the Google play books text to voice service they had in beta (no idea if it still exists, and they were never made public public), and I found NotebookLLM handy in that you supply the sources, and it helped me keep an eye on my characters. Were I to tell it to use only one source (say a draft I had written), and I ask, “When so-and-so was here, what colour eyes did they have,” and it would dig up the answer and provide a link to the section in the source. But, that’s what it’s good at doing.

I always hated the name, AI, It is a clever, clever thing, but it has no intelligence, it never did. It’s not some Skynet like thing.

There is an aside here with Wilp; slightly spoilered.

Wilp is written at the same time as AGI breaks through in the world; it’s part of one of the subplots. One of the early drafts dealt with the difference between AGI and AI. I toyed with the idea of covering underwater datacenters for cooling; solar for power etc. I had it all in notes, and had read a lot of things. I cut it all, it was meaningless technobabble that added nothing to the story; the AGI stuff was important, but the other crap? Needless; interesting to me, but it wasn’t needed. So I used the term “AI” over “AGI” as well because it was what people understood and needed less explanation.

It’s a clever large language model; the tech behind it is great, what it never was, is intelligent. As said, I’ve used AI for image generation. I used Adobe at the time after reading this article – https://petapixel.com/2023/03/21/adobe-firefly-is-a-text-to-image-generator-that-didnt-steal-your-work/

We all knew from the start that the models were trained on stolen data, so I looked for ones that were different. Naïve, maybe? But that’s the truth. Wilfully blind? Again, maybe, but again, the truth.

I’ve run a few rough drafts through a LLM or two at the same time to see what it chucks out; mostly unimpressed with it if I’m honest. It looses the human touch and over-complicates the most simple things. It feels generic in a lot of ways; even seen a movie that feels like a homage that’s over the top? That is what it feels like that to my eyes; perhaps it’s different for you. The plagiarism machine really showing its sources is what I see.

Put a five hundred word draft in one LLM and ask it to edit it, and you get a completely different result to what you get in another. Give one a detailed prompt for something and it spits out something random every time. It can’t create; it doesn’t think, the emotions of a sentence elude it much like a brick trying to understand how a heart pumps, or a person looking at a cat and thinking they know better.

That’s with writing; I am sure other artists can verify if it is the same for them.

I challenge anyone to talk to it for fifteen minutes about a subject you know a great deal about. How many times have you had to correct it? Put it back on track. Tell it something it said earlier contradicts something it has just told you.

Hallucinates, it claims? I say, “lie.” Think about it, who in real life says to you, “sorry old chum, I hallucinated that.” Even its errors are corporate crap made to keep you using; it’s tobacco repackaged. The first time it needed pushing back on track, or something was a “hallucination,” if it had said “oh, yeah, sorry, I made that up,” you probably wouldn’t be using it now; I like to hope you certainly wouldn’t trust it.

And then, then the environmental aspects became clear.

That’s why I played around with local LLM’s and a few online for basic stuff when it first became mainstream. Not particularly proud of being caught up in it all, but I do that with most tech “breakthroughs.” I still think the tech is very good and will have its uses in time, but that won’t be Gen AI usage by the public; it’ll be using it for what it is good at being, what it is, a large language model. What it is being used for at the moment is abhorrent, and such a waste of resources. What makes it worse is that so much of it is unneeded crap.

Oh, and yesterday we had the “book burning” story, of course.

https://www.techspot.com/news/113277-ai-firms-quietly-buying-destroying-millions-printed-books.html