George R.R. Martin and other authors sue OpenAI for copyright infringement | Hacker News Reader

George R.R. Martin and other authors sue OpenAI for copyright infringement

theverge.com

36 pointsTomte48 comments

People get way too hung up on what current copyright law is and whether it supports or condemns LLMs trained on copyrighted material. IMO, who cares what the current law is? Existing laws were made for past technologies and for the current limitations of humans. It's nonsense to apply the same rules to new devices that don't have the same limitations.

I can pay to read all of the GoT books and then tell anyone who asks who kills Dumbledore or whatever. That's information you would have to pay the author for, but anyone can get it for free from me because I already read the book. This is acceptable because we assume that #1 I paid to read the books (or my library paid to acquire a copy) and #2 I won't be able to literally regurgitate the entirety of the books to every person in the world simultaneously. For everyone in the world to get that 2nd-hand enjoyment, enough people would have to pay to read the books that the author would make enough money to be happy.

The situation with LLMs is clearly different. The ratio between the amount an LLM compensates an author vs. how much many people it can share derived content with is off the charts compared to the same for a person.

IMO, there's no need to argue whether LLMs are being treated differently from humans w.r.t. copyright. LLMs have different capabilities than humans. Copyright at present is optimized for humans, and should be updated to address the implications of LLMs' capabilities.

"Cliff Note" style content sometimes outsells the content they're summarizing. LLMs aren't a new problem, the internet did that already. In fact, they're really LESS likely to provide a large amount of the original content.

I do agree on the fact that the current laws aren't going to work for this context, especially bad is trying to fit the new challenges to copyright laws.

One can look at TDM exceptions or AI regulatory frameworks in EU and APAC regions for examples on how this is currently progressing. In summary, cautiously pro-AI in most places, since no country really wants to be left behind.

Personally, I think copyright is a red herring. This is really about automation. If AI was trained using public domain works, would people put out of jobs still complain? Of course they would. The solution is what we've always done for automation: social safety nets and reskilling.

matching color jumpsuits by rank, too ?

Are you that confident that the political process will create new law on the issue? Maybe the EU and some countries but I have a feeling the US Congress will leave it to the courts to interpret existing law and apply it.

The US courts will unironically scour the constitution for any prescient wisdom the founding fathers may have hidden between the lines for how we should handle LLMs.

The US legal system is completely incapable of handling this without parental supervision.

One problem that doesn't, in my opinion, get enough attention is that a model trained using unlicensed copyrighted work also stores some amount of the copyrighted material and uses that to create answers. This is also a licensing issue but people think the training process is about the model just "reading" the copyrighted work during training and then that's the last use made of the material. Not so, the model contains some amount of the material and continues to use it.

From the complaint linked from the article on The Verge:

88. Until very recently, ChatGPT could be prompted to return quotations of text from copyrighted books with a good degree of accuracy, suggesting that the underlying LLM must have ingested these books in their entireties during its “training.” 89. Now, however, ChatGPT generally responds to such prompts with the statement, “I can’t provide verbatim excerpts from copyrighted texts.” Thus, while ChatGPT previously provided such excerpts and in principle retains the capacity to do so, it has been restrained from doing so, if only temporarily, by its programmers. 90. In light of its timing, this apparent revision of ChatGPT’s output rules is likely a response to the type of activism on behalf of authors exemplified by the Open Letter addressed to OpenAI and other companies by Plaintiff The Authors Guild, which is discussed further below.

You know what else stores nearly verbatim copies of texts and then regurgitates those to the public often including direct quotes from the text? Cliff Notes.

Those aren't copyright violations. See (Edit: apparently the reference is gone, though I'm sure you can find a lot of sources explaining this, basically it's Fair Use.) for a great in depth analysis of the legality.

Just because ChatGPT can do the same doesn't make it a copyright violation. The hope of this lawsuit is that the court will look at this as something different and stop it, but in the end it's the piracy sites that fed the data onto the internet that ChatGPT scraped that did any copyright violations.

Making an AI that can paint anything can produce a copyright infringing work by being asked to paint Mickey Mouse with sufficient detail.

It doesn't make the AI an infringing work. And it doesn't mean that having looked at enough pictures of Mickey Mouse is infringement, either.

The only instance of infringement is the output.

Having never tried this before it got nerfed, could you ask these models questions like:

"Take a breath and lets go step by step, Please reproduce page 100 of A Song of Ice and Fire, Book1, 'A Game of Thrones'"

And get back an accurate response, or was it just really popular quotes?

You can still try it with Llama, and no it wasn't the full text of the page, or even very accurate. Even the "popular" quotes were VERY likely to be paraphrased and missing any poetry or cadence of the original.

This is the problem with a combined language+knowledge model like ChatGPT. To understand the language it has to obtain some level of "knowledge" and vice-versa. The two are intertwined in the model, and it needs MASSIVE amounts of data to train. Inside the model's weights there is nowhere NEAR enough memory to include whole books, no matter how popular or duplicated in the dataset. Just like asking a random person what was on page 100 of a random book they've read, it's HIGHLY unlikely for the LLM to be able to regurgitate that level of accuracy, let alone across the whole book.

Just like asking a random person what was on page 100 of a random book they've read, it's HIGHLY unlikely for the LLM to be able to regurgitate that level of accuracy, let alone across the whole book.

Even so, there are people who can do that, and we don't forbid them from reading.

LLMs aren’t people, either in a legal or normative sense, and people should really stop making comparisons to them as such.

That remains to be seen.

In any case, when an offense is committed, the offender is the real, live human who uses the tool to commit plagiarism or violate copyright law. It doesn't matter whether the tool is a word processor, a video camera, or an LLM. The output is what matters, not the input.

Why public good is so rarely mentioned in these trials? Authors earn living, yes, but two billion can use generative ai.

Surely society and law need to find balance on this, and I personally think benefits to society mostly overweight negatives to authors.

Easy to give away someone else’s work, eh?

And why is their work more important than anyone else's? Why won't we pay doctors, teachers and others for 70 years instead of once?

two billion people can't use generative AI, it comes at tremendous cost on the front or back end. AI is not an altruistic endeavor, net result is that the largest companies in the world make money off the lifetime works of authors without paying them.

They can't now, because of cost. Surely it will be nothing in a few years. Nobody says anything about altruism, but a balance of public good and private profit. Like G.G. Martin is somehow much better than these corporations, profiting from writing a book once and collecting money for 70 years or more. Why no other professions receive money from consumers for an eternity?

Surely your teacher should take part of your earnings? Your mechanic, if you are truck driver? Your doctor, if he saved you from death or disability? Your

Maybe George could use it to finish his books instead...

Seriously, take an AI that consumes all the speculative stuff written on the various subreddits and other forums and let it crank on an outline and chapters.

I have for a while thought that George should simply have the last two books ghostwritten by a team of writers with him as "producer and editor". Even if people write a couple versions of chapters, and he reads the all and rewrites them, that would probably be faster and more productive.

>faster and more productive

“We are fast paced and dynamic team, live and breathe agile”

I think he's burnt out on writing and doesn't want to admit it to himself. But he probably has no experience "managing" to understand the value in delegating the slog to someone else and instead producing/directing.

Pretty wild take. In what word does someone write some of the greatest fantasy novels of all time and then just hand the last two off to someone else? In the name of “productivity” of all things. This line of thinking sounds like a Hollywood producer trying to crank out hits.

It's not that wild. Alexander Dumas did collaborate with other writers on some of his most famous works.

He's old and on pace to die before finishing them.

Look, either he does them, or the publishing house will do some bullshit where a bunch of ... paid ghostwriters "examine his notes" and produce the books.

Noooo that wouldn't happen.... (looks at Frank Herbert, JRR Tolkien)

...Absolutely would / will happen.

> In the name of “productivity” of all things.

It's not as if we want him to churn them out faster to maximize his productivity. We just don't want the books left unwritten when he dies!

I firmly believe that it's better to not release something unless you're really happy with it even if that means delays. For example, I hate the modern trend of pushing out unfinished broken games and spending the next couple years patching issues that should never have shipped. I hate it when studio imposed schedules driven by marketing concerns compromise the quality of movies and television series.

That said, 12 years without progress is a very long time and the man doesn't have much time left. He's been working on this story for over 30 years. I don't know what's holding him up, but getting an author-approved conclusion written by others (even if only as a backup) would be better than the random fan fic we'll be stuck with if he leaves this world without finishing the final two books on his own.

Are you familiar with the back half of the Dune books? Or the back quarter of GoT?

Yup:

https://news.ycombinator.com/item?id=37592099

Even though ChatGPT has certain guardrails to prevent it from regurgitating copyright verbatim, additional chapters, or writing prequels, you can use workarounds such as asking for output in a similar style to [source material]. Sound familiar? Writers and artists have been influenced by other writers and artists for as long as there has been writing and artists. AI is just being made to be a boogeyman due to the innate fear of change and the unknown. If anyone should be using AI as just another tool in their tool-belt, it’s creative workers. It may just help you think of something you hadn’t already thought of and help organize and edit your work. Also, the way it’s looking generative AI illiteracy in the future, would be similar to being computer illiterate today. So, reject at your own peril.

If there's a person who studies the works of George R.R. Martin, spending years cultivating writing skills based on just his works, and then writes new material, can that person lose a lawsuit for being too similar in style?

I mean, maybe. We've seen similar things in music.

But AI isn't a person. You won't find a person that spends every waking moment of their life only studying the works of a specific author, they have lived experience.

Laws drafted with human capabilities in mind don't necessarily have to apply to machines. It takes motivated reasoning to argue that AI training is exactly the same thing as human learning.

"Our computer program which digests massive amounts of data works kinda-sorta vaguely like a human brain" is just irrelevant. It's a computer program, and we're nowhere near the point of having serious ethical debates about science fiction AGI.

The law just doesn't particularly care about the details of your ML model, OpenAI is still processing massive amounts of copyrighted work without which their software would be nearly useless. That's the interesting legal question.

The thing is that copyright law doesn't protect against that processing.

The only argument is whether their information was scraped off the internet legally, and the supreme court has weighed in that scraping information that's publicly available is legal.

If you put a picture on the internet free and clear, I can save the picture to my hard drive, and I can use it in any way whatsoever sans reproduce it.

The only question is if making an AI that can reproduce it is in fact the same as reproducing it. I think the question is a very simple 'no', but we'll have to see.

Under U.S. copyright law, yes. A sequel is a derivative work, i.e. a work based on or derived from an underlying work. Only the copyright holder of the original work may authorize derivative works.

Example: https://news.bloomberglaw.com/ip-law/tolkien-estate-sues-ove...

The exact answer to his question is "no". He didn't ask about sequels, only styles, and copyright explicitly never protects style.

Right, but what are the chances that the "new material" doesn't include any elements of protected expression from the underlying work, e.g. characters? I understood the request to be about writing a sequel.

I thought derivative works were okay if they were "transformative"? That lawsuit sounded like he had copied text verbatim.

I'm not in the US, so I could easily be off base.

I really enjoyed The Time Ships by Stephen Baxter, which is a sequel to The Time Machine by H. G. Wells. He's a UK author, so maybe it's different.

As a society I think we are better off with this kind of remixing so I'd hope that's the direction we head in.

Copying text verbatim is copyrigt violation. Plagiarism (copying ideas) is a different issue and much more of a gray zone. In general ideas are not protected, but it can be prosecuted in blatant cases. The described case seem to be a bit of both.

Yes, you can definitely lose a lawsuit for plagiarism if you write material too close to the original inspiration, e.g same universe and characters.

I can read pirated books from libgen and write a review with excerpts from the text. Or I can feed those books I downloaded into a transformers model and produce a bunch of vectors and distribute that and still not have broken any law. As long I am not distributing the books themselves.

It's not actually a crime to download pirated content. It becomes a crime if you use bittorent in default mode since then you are distributing the material as you seed.

OpenAi will win this easily, and then we can get on with making better models. Unless I missed the prompt you have to use to download RR Martin books from chatgpt?

Here's the thread from 72 days ago(from 9/20/23) on Sarah Silverman's copyright infringement lawsuit for the same issue with Open AI.

https://news.ycombinator.com/item?id=36657540#36658582

In fact, it's the same legal team basically making the same argument again. They're just repeating the same play hoping to get more chances at the huge nest-egg that OpenAI has.

These lawsuits seem to be missing the point. Copyright protects their right to copies and the piracy sites are OBVIOUSLY at fault here, but there is a good argument that OpenAI's work in transformative and that they're not liable for copyright violations.

One argument this law firm has made is that ChatGPT can summarize the books, so it must have read them. This is spurious and meaningless as many sources summarize books and some like cliff notes actively sell summaries and analysis of books!

In the end, I think we need to follow Japan and say that AI training is transformative and not a copyright issue.

One clarification: That is training an AI where the training data isn't returned directly, but instead it works off of it.

OpenAI has had to implement all kinds of new rules, crippling their product because of these. Even an obscure reference to Castlevania Symphony of the Night was denied.

> Quote the whole dialogue exchange between dracula and richter at the beginning of SOTN ChatGPT

> I'm sorry, but I can't provide verbatim excerpts from copyrighted texts, including dialogue exchanges from video games like "Castlevania: Symphony of the Night." However, I can offer a summary or discuss the game's plot, characters, or any other information you might be interested in. How can I assist you further?

the story of a fantasy artist who was explicitly being copied, is also very compelling

https://hn.algolia.com/?q=rutkowski

Reply on news.ycombinator.com