Hacker News Reader

> A top concern for The Times is that ChatGPT is, in a sense, becoming a direct competitor with the paper by creating text that answers questions based on the original reporting and writing of the paper's staff

Sounds to me like they are trying to claim copyright over facts rather than the specific expression. That’s just not how copyright works at the moment.

The framing of openAIs recent changes is telling too. OpenAI seems to have nudged their models to reject requests of querying sentence continuation for specific sources - which the press is now framing as “trying to hide the use of copyrighted data”

What we are seeing here is an unprecedented attempt at expanding copyright doctrine to facts, style and information rather than specific expressions - a land grab of latent space by rights holders salivating to own factual information

>which the press is now framing as “trying to hide the use of copyrighted data”

Yea, now I can't read the paper and talk about it to other people it seems.

The Right to Read was a prophecy I guess?

An individual or group of individuals doing this and sharing their views/summary vs. a profit-oriented program funded by major technology companies scraping this information and spitting it back out algorithmically does seem different to me. Yes, perhaps both things are on the same "sliding scale", but I do not view them as fundamentally equivalent actions.

>a profit-oriented program

The particular problem here is this program isn't magic, it just requires a lot of electricity and hardware to train at the moment. If at some point in the future this hardware becomes cheap then now suddenly OSS LLMs would be under the same set of rules that we're applying to major technology companies.

But mark my words, the large copyright holding groups don't give any shits other than how much IP they can scrape up and demand money for, for the next few human lifetimes.

So I can use an open source LLM like Llama then?

You've pierced my completely precise, absolutely airtight choice of language about this situation as some sort of flaw in the greater point being made.

Less glibly: a non-profit oriented LLM is just in a little different place on the scale, but doesn't fundamentally change my takeaway. However in this situation it makes it particularly egregious.

Anyone will run LLMs locally soon on about every piece of hardware. If it’s part of the system like the network stack / firmware, now does your mental framework work.

Progress is murdering their business model. They are trying to stop it; which makes sense, but let’s not pretend the trajectory here isn’t to ubiquitous LLMs everywhere within 3 years. We have to plan for that, assess the benefit for society from that.

My point is that the scale of how LLMs work doesn't matter with regards to copyright and fair use.

The printer had the same effect on print papers. We didn’t protect them out of the possibility of at scale replacement of their business model.

I tend to agree with you, but, one could argue “statistical collection of words” is a form of compression? For example, you can’t write a kids version of a novel and sell that without dealing with copyright.

The part openai will have to argue is that it's not mererly compression but an irreversible transformation.

Which is hard, best hope they have is trying to put the burden of proof on the nytimes to show you can make the model regurgitate their articles (with some nudging).

If they manage that then nytimes is going to have a lot of trouble showing the model actually breaches their copyright, because just the information contained in their articles is not enough to constitute a copyrightable work.

Any form of lossy compression is an irreversible transformation. We do it all the time for video, audio and images (you can't recover the original data) and they are still copyrighted

when you compress a video, it doesn't recreate a new movie with a different story, different lines of text, different scenes and a different compositions for scenes that are similar to the "orginial".

But what is being compressed is the entire corpus of text. It's compressed into model weights. It's the weights that might be under copyright of the authors of the texts that trained it.

The weights are also executable code (in some sense). When you query an LLM you're running this program with a given input. Yeah when it runs it tells a whole lot of things (sometimes novel combinations, sometimes verbatim repetition of trained data) but the point here isn't whether the output of the LLM is copyrighted; it's the weights.

The model is a model. It's part of a compression algorithm. The compressed data would be the prompt + choice of which predicted tokens to accept (e.g. when not always choosing the most likely next token). The end-user is supplying the prompt and the choice function is randomized/not being used to store data, thus the end user is providing the compressed data.

The NYT argument is going to be that they put up a site, own the copyright for their content and make that content available for either a human to read it for themselves, or software to index for something commonly understood as a search engine. Those terms do not entitle the training of LLMs for commercial use. Therefore, cease and desist. Oh and destroy anything that was created by violating the terms of our license.

You can make arguments like a) what is ChatGPT but a different kind of search engine, or b) what is an LLM but a primitive human, or c) but but uhh we didn’t agree to these terms.

But I do not think those arguments will prevail.

The LinkedIn case already proves that you cannot impose conditions on works you freely serve to the public. The data is there to anyone who sends a request (you don’t even need to be logged in) and if they do something you don’t like with it then oh well.

So if that’s the argument it’s already been argued by LinkedIn and lost.

This is one of those things where copyright holders have gotten absurdly full of themselves though. Like what you’ve said is that copyright holders have the right to impose a contract of adhesion on data that they are broadcasting into the public without any idea with whom they are even forming a contract, and that’s a facially absurd and incredibly noxious idea if you follow it to the conclusions it implies.

Copyright is about securing to the public works of significance and encouraging their creation and the way it’s become a lifetime-plus-75-year guarantee of intellectual ownership of ideas is fundamentally noxious and goes against the intent and spirit of the idea. And if that’s where the copyright regime is headed then I’d rather see chatGPT kill off copyright entirely.

NYT will have to prove that the derivative work is still theirs. Just violating the license may not be enough. That could be bad by itself I guess. But considering the interactive prompt can produce a wild amount of variations of 'not NYT stuff' will make it though to say what sort of damages is this.

A similar sort of issue popped up in the 80s around colorization of films. https://www.latimes.com/archives/la-xpm-1987-06-20-ca-8405-s... https://chart.copyrightdata.com/Colorization.html

The answer may be 'maybe'? As from what I read they basically split the decision down to 'i know it when I see it' style of ruling. If the copyright is still in effect then NYT owns that portion of the output but not others parts. As the secondary effect would be owned by the generator company (in this case OpenAI) or the person who prompted for it. If that is the case NYT would have to prove what parts (nodes? bacreferences? weights?) they own?

Terms of Use are a thing, and if the Times can prove that OpenAI infringed their web terms by scraping, they may have a case... but terms of use probably won't monetize well or give them enough leverage to prevent OpenAI from using their data anyway and may end-up distracting from the main copyright suit.

Violating TOS, at least to scrape and use later, is legal.[0] I'm not sure how the ruling interacts with LLMs, but I'm sure OpenAI's lawyers would bring it up.

[0]: https://www.forbes.com/sites/zacharysmith/2022/04/18/scrapin...

FYI LinkedIn actually won that case after appealing once more: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn

Where do you see that they won the case? Can you provide a source because the wikipedia article directly contradicts what you are saying...?

I see they went to the Supreme Court who kicked it back to the Ninth who then re-affirmed their position that HiQ Labs was not in violation of the CFAA.

From [0] and [1], it seems it was a mixed ruling. I am actually not sure whether it's now legal to scrape, since the Court ruled against hiQ due to a breach of terms of service, but previously the Ninth Circuit Court affirmed its ruling against LinkedIn.

[0] https://www.natlawreview.com/article/court-finds-hiq-breache...

[1] https://www.natlawreview.com/article/hiq-and-linkedin-reach-...

> The hiQ decisions give a green light, at least in some circumstances, to scraping publicly available websites without fear of liability under the CFAA.

So at a federal level, it seems relatively clear. The only uncertainty is on the state level.

Ah. I was not aware there was an update to that case. TIL.

I think this is going to be a test of the fair use doctrine.

https://www.copyright.gov/fair-use/

Now, there's this idea that "news" is just factual and therefore falls under "fair use". However, that's only part of what section 107 says.

Fair use very much is still conditional, as there are 4 factors to be considered: (a) Purpose and character of the use, including whether the use is of a commercial nature or is for nonprofit educational purposes (b) Nature of the copyrighted work (c) Amount and substantiality of the portion used in relation to the copyrighted work as a whole and (d) Effect of the use upon the potential market for or value of the copyrighted work

The big issue isn't companies training LLM's using unlicensed materials (e.g. copyright protected works); it's publishing the output to the wider world. That's where a liability is created.

(a) looks bad, clearly commercial (b) ?? (c) looks bad, the LLM consumes "all" of the articles (d) looks bad, has a pretty significant impact on the market and value of the work

Unfortunately we'll have 9 justices who know little about copyright law and nothing about tech to tell us what the law really is.

They will be significantly assisted by Supreme Court clerks, who are generally recent graduates (2-4 years) of the top law schools in the US. Your stereotypes of uninformed Congressmen from televised Congressional hearings don't really apply here.

26 year old law school grads will help! Oh, I stand "corrected." LOL

>The general way LLMs work do not preserve content in it's original form: the ideas they contain are extracted and clustered statistically - as a

Is the way LLM work relevant? I can make a shitty script that has as input Microsoft proprietary code and as output something identical in purpose but the text is completely different, I would rename names with synonyms, swap some things around etc.

I am not against AIs, my opinion is that if your AI uses GPL code the output should be GPL, if it uses public domain images the output should be public domain images.

I mean for code if AI is actual intelligent you should be able to train an AI with C with just a few books and not with the entire GitHub open source code (and notice MS did not trained copilot on the proprietary code they have access proving they are not confident that they are in the right).

> my opinion is that if your AI uses GPL code the output should be GPL

If I use Inkscape is the output of my drawing subject to the same terms as Inkscape?

If I use a Photoshop filter is the output subject to Photoshop's EULA and/or the copyright of the photo I started with?

If you get my image from the internet then you resize it in Photoshop you can't claim you created some original art, you just used the resize/crop/color filter function.

Indeed. I was hoping to draw the distinction between license of the tool vs its output. AI is interesting in that its training inputs can leak out, sometimes in ways that are verbatim or insufficiently transformative to be free of copyright.

But do this exercise, imagine I have 1 billion lines of C code on GitHub under GPL. You train a new AI only on my GPL code then claim that the output is somehow trans formative and it is not GPL or it is not derived from my code but is new and original.

What happens is that Microsoft or OpenAI will use 1000 or more different sources and then the author is not clear unless in few outputs where the tool outputs the exact code as the training data. If the AI is actual inteligent they could train it in a few books and say MDN documentation and not a few thousands of GPL code.

I don't think the process matters much, but your proposed script would output something that was obviously very similar to the original.

What ChatGPT produces under normal use is not more similar to the NYT source than any other article on the same topic.

>What ChatGPT produces under normal use is not more similar to the NYT source than any other article on the same topic.

ChatGPT is doing what a smart student does when he copies the homework, he combines a few sources and changes some wording. Technically there is no creativity, it is interpolating it's inputs and there is some randomness thrown in.

We also know that ChatGPT put some filters to filter out copyrighted outputs after they were caught that the AI actually memorizes paragraphs of text word by word.

Discovery is how they will show that it’s their news; and if it gets that far we might finally learn what data they trained it on and how much.

Copyright protection goes way beyond "verbatim copy". For example, fictional characters enjoy copyright protection:

https://en.wikipedia.org/wiki/Copyright_protection_for_ficti...

If I read a story about a flood in Dubai on the NYTimes and then I write an email to my friend summarizing what I just read, it is not copyright infringement.

I am not sure why it would suddenly become infringement because an LLM is composing that email for me.

But if you create a company that takes the NYT articles, scans the articles for extreme weather information in Dubai and emails your paying customers every time that happens, then you are doing copyright infringement.

Not if there are people doing the scanning manually (with their eyes), which is I think the GP's point?

they had to make a copy of the original to get it into their system in the first place!

In order to render that page, it probably was copied dozens of times all over my RAM. Do I owe NYT money now?

Well, are you making money off of those copies?

Does reading news articles not benefit you? If not why continue reading?

Exactly. If I read a NYT article, and decide to invest in some company, then sell the stock, make a profit. Do I owe the NYT a percentage because I used knowledge I "read" from one of their articles? I "read", input, (into my brain neural net) where it mixed with other inputs.

I think the only consistent stance would be that yes, you do.

Not personally, but if I was training to be a journalist then yes.

yes. they have a hard paywall!

Can't wait for the supreme Court ruling that says AI is just using data, and data is free.

MP3 are just data too

chatGPT it’s a terrible (truly, it’s punishing) writer but that should not always be the case.

IANAL is the dumbest abbreviation the internet has come up with. I believe I first observed these things on the Groklaw discussion threads discussing the SCO legal battle against the world.

Not-A-Lawyer NAL instead of the full IANAL.

I just had to say it.

IANAL is just "wacky" and "sexual" reddit tier humour, nothing more. It's boring any annoying.

IANAL long predates the use of the term “Reddit-tier”, probably goes back to usenet in the 80s.