zenodo.org/records/1108...

You've convinced me.

Maybe they should, idk, but if we want to start criminalizing certain kinds of math, we should at least make an effort to understand them first. And this mistaken belief that GenAI models contain copies of the work they're trained on, that's not helping.

You can feel that way, but it doesn't change what a language model is or the fact that it learns from examples much like a human does. And current copyright laws just don't have special rules for non-human intelligences.

There likewise isn't any place in the brain of a human artist where they'd need to "store" a particular image of Pikachu in order to have the ability to draw Pikachu. Knowing what something looks like and photocopying that thing are distinctly different.

I don't know what to tell you. That's just simply not how the technology works. There's no place in a language model where the training data is stored.

That's not how LLMs work. They're statistical models, not databases. If I generate a word cloud from all your bsky posts, have I stolen your work?

Not everyone has that kind of money. And honestly $25/month or whatever is an incredible deal for a personal assistant that has encyclopedic (if not expert) knowledge of every topic known to man, as well as infinite patience. It's incredibly empowering, really.

It's not theft because it's not copying. In principle, training a language model is no different from training a human writer by showing them examples of what other people have written. Or teaching a child how to speak by letting it hear you talk. The child doesn't thereafter owe you royalties.