Hacker News Reader

Comment by faragon | Hacker News Reader

If a person can read copyrighted material and produce derivative works, why not an AI?

Because - well a person can read copyrighted material it legally obtained the rights to, for example by purchasing a hard or electronic copy of the book or magazine. Alternatively, and according to the laws worldwide, if a person were to engage in massive theft for the purpose of "reading" all available copyrighted materials in the world, by obtaining copyrighted material without permission and consent of the copyright holder, they would be at least paying heavy fines, and in most jurisdictions also spend at least a few years in jail. Why should the same not apply to corporations and their executives?

I don't think there is actually a law anywhere that says you need to obtain the rights to copyright material to read/view them. The person or organisation showing it to you, which might be yourself, needs to have a license. Otherwise things like libraries couldn't exist and you wouldn't be allowed to lend books or even have books in your house that other family members can read.

Not saying that particularly impacts your argument about OpenAI, because an LLM in training is not a person. It is transforming data from one format to another for later consumption by people. Therefore they probably would need a license.

I mean, look at it this way. Let's say you purchase a Woody Allen film on DVD. Will anyone seriously prosecute you for watching it at home together with your friends? No, that falls within normal usage. But let's say you now organise a local watching event with the same DVD for 200 people in a hall somewhere, and charge everyone, whatever, $6 - just to cover the hall expenses. Will you be prosecuted? Very likely. Libraries are probably under some sort of "fair use" regulation due to public interest and such. They don't quite generate profit with their line of work - nor should they!

Right, but those 200 people won't be prosecuted for watching it, which was my point. The example I was thinking about when posting would be putting up a copy of copyright art in a public place. The people in the public place are not breaking the law by looking at it, only the person who placed it... well even then, would the workers who put it up be liable? Probably not, it's not reasonable for someone who puts up billboards to check the copyright license.

I do agree with this example in general. But I guess from my point of view, the OpenAI comes across more like the person enabling the use of copyrighted art, and would thus be subject to copyright regulations. Their users I'd see rather as the people viewing the art in public, perhaps unaware of the copyright restrictions. But it also seems like these discussions in themselves are a bit of distraction. If the LLMs worked exactly as they are being hyped up for the third year now, I think we all would get behind the effort. Who would care about copyrights if a magic machine could lead us into the so-called post-scarcity world, right? But sadly it does not appear to be nowhere near that goal, nor will it be, based on what we know about how the technology works. So here we are, discussing if mechanical parrots should read our books :)

Sure, so, can I make and sell my own Lilo and Stitch movie now? It'll be even better than the one about to release, and all that means is I'll deviate even less.

This was settled prior to LLMs - you can't do that because the characters names are copyrighted. LLMs change nothing here.

Because the AI is not a person. It doesn't seem like we're anywhere near AGI that could be considered a person. Training an LLM is taking existing content and transforming it into another format for later consumption by a person. That person can run prompts against the LLM to create derivative work, the LLM itself doesn't run prompts or do anything at all.

I don't know much about the legal side, but it seems to me, from the above, that the laws for copyright for LLMs should apply to the company training the LLM as if they're creating a derivative work that they will later sell or license for other people to interact with.

Good then that LLMs don't reproduce content.

They produce derivative works, which is also an exclusive right of a copyright holder.

If I derive my work using multiple sources, do all the copyright holders from these multiple sources have an exclusive right on my work? How otherwise would people build a knowledge on some topic and then apply that knowledge to build a product if not by reading bunch of (book) material and studying other similar products?

If they can prove it in court. Would be much easier to do for a LLM than for a human one would think.

In case there is confusion about the law:

https://www.copyright.gov/circs/circ14.pdf

If you want to change it, good luck.

Personally, I think that most intellectual works should be able to be derived from others and shared freely after a brief period of time (like 7-14 years).

> It only restricts reproduction

and distribution.

> ... a person can read copyrighted material

Yes, after paying for it.

People have to pay for it

Reply on news.ycombinator.com