Hacker News Reader

Comment by bilbo0s | Hacker News Reader

Pretty sure this wouldn't pass the merit part unless the prompt was unusually long and precise.

the human still made the prompt

What I can guarantee, is that series of prompts itself would be copyright-able. (The series of prompts that ultimately created the image.) No matter how little they may weigh any one of those prompts in isolation. That is, assuming the EULA of the LLM doesn't require you to essentially place your prompts in the public domain.

And of course,

<s>

everyone reads the EULA. Right?

</s>

View on HN

Unless you can make your prompt so specific that the AI generates substantially the same image every time you run it, I think you're perpetually vulnerable to the argument that significant decision making was done without human hands and therefore the work is not primarily human created.

That’s a good test. An artist working in oils can effectively create the same image over and over. An ai kind fails there.

As kids we did an art project where you mixed colors with some yoghurt-like substance. You drop it on the paper and then fold it. This created these beautiful arrangements of colors.

Does this mean that those works are not copyrighted either since the kids didn't actually direct where each color goes? Every time you do this you'd get a substantially different picture too.

I mean, these edge cases are very sensitive to the exact facts at hand. Even an experienced copyright lawyer can't give you a definitive answer until you go to trial. That's why I said you're vulnerable to the argument, not that you'll definitely lose the copyright.

Every time you do this you'd get a substantially different picture

This is actually a bad example. It's too easy for an IP attorney to bring in an expert witness,(read: physicist), and blow it out of the water in a courtroom.

I won't go into the details, but basically, you got different arrangements every time because the human did different things every time. In the case of generative AI, you get different arrangements every time when the human does the exact same thing every time.

So, if you can find it, the counterexample you're looking for is one where the human does the exact same thing every time. (In an unassailable mathematical and physics based sense of the word "same"). But gets different results.

The human is using a seed, whether implicitly or explicitly.

You can generate the same thing every time.

> What I can guarantee, is that the prompt itself would be copyright-able.

That's non-obvious to me. Even if the prompt is extremely long and precise, if it is somehow purely functional, it seems possible for it to not be (although in practice, I agree that most prompts could be).

It is basically pseudo-code, and should have the same copyright as other code if it is sufficiently complex to pass the typical test for copyright. One might think code should not have copyright, but that is a different conversation.

Yes, I agree. I don't think I am saying anything inconsistent with that.

Code is purely functional and is copyrightable so why would a prompt not be?

A prompt has essentially the same purpose as code, especially when it's long and precise.

Code is not purely functional. If it is, it is not copyrightable (at least in the US; probably true elsewhere but I am less sure) [0]. I would not expect most prompts to be purely functional.

[0]: https://www.copyright.gov/circs/circ61.pdf

Reply on news.ycombinator.com