[vlovich123]

Reminds me, did the AI companies redistribute that copyrighted material to others and make their money that way? Did Kim use the copyrighted material to generate something novel from it?

copyright law literally says something isn’t infringement if it is a novel transformation. I get the jokes and criticism about AI companies fighting and complaining about competitors distilling, but this is a much weirder comparison.

[root-parent]

Now you just have to explain this:

"Anthropic settles with authors in first-of-its-kind AI copyright infringement lawsuit" - https://www.npr.org/2025/09/05/nx-s1-5529404/anthropic-settl...

[hydrox24]

> "The training use was a fair use," [the judge] wrote. "The use of the books at issue to train Claude and its precursors was exceedingly transformative."

> However, the judge ruled that Anthropic's use of millions of pirated books to build its models – books that websites such as Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi) copied without getting the authors' consent or giving them compensation – was not.

It seems clear from the article that while the use of pirated works was illegal, the use of copyrighted works (a the work a book is based on is still copyrighted if you buy the book) was fine and transformative.

[andersonpico]

But distribution isn't the only crime here, obtaining the material illegally apparently is a crime too. And the damn robot can also spit me harry Potter verbatim so I don't know how it would also not be distribution?

[vlian2088]

>And the damn robot can also spit me harry Potter verbatim so I don't know how it would also not be distribution?

if you prompt it to, yes. just like your browser dutifully navigates to any copyright-infringing resource and GETs and POSTs whatever you ask of it.

(also it can't, not really, only small snippets before going off rails. LLMs aren't magic, they can't losslessly compress an exabyte of training data into a few terabytes of weights.)

[mapontosevenths]

If I read Harry Potter I will remember some parts verbatim. Others I will tecall in only an abridged and lossy way.

Does that make my brain copyright infringement? Does Disney now own all my output forever because some small part of me now has Harry Potter embedded?

[buran77]

Can you remember every part? Can you do this for every book in a library? Can you remember all that forever?

If you just ignore anything that's inconvenient for your argument, you can make any argument you want.

[gruez]

>Can you remember every part? Can you do this for every book in a library? Can you remember all that forever?

None of those are relevant factors when it comes to copyright law. You don't get a pass for copyright infringement just because you're not copying the entire work. Same goes for a copy that's transient. You can't set up a bootleg movie theater in your home, even if you delete the movie file afterwards, and there's no trace of the movie aside from the viewers' vague memories.

[buran77]

> None of those are relevant factors when it comes to copyright law.

And yet they very much are. US copyright law has the concept of "fair use" in 17 U.S. Code § 107 [0]. I'll paste here for your benefit, #3 is the one I referenced as most obvious but #1 and #4 are also very relevant:

  (1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
  (2) the nature of the copyrighted work;
  (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
  (4) the effect of the use upon the potential market for or value of the copyrighted work.

Naturally remembering some parts of a legally purchased book verbatim is fair use. "Memorizing" the entire library obtained via torrents and incorporating that in a commercial product that can output all that content doesn't sound like fair use to me.

The US justice system is too captured and corrupt at this point to take as reference because decisions there are bought by the highest bidder. But for the purpose of this discussion let's not play dumb for the benefit of trillion dollar corporations.

[0] https://www.law.cornell.edu/uscode/text/17/107

[gruez]

>And yet they very much are. US copyright law has the concept of "fair use" in 17 U.S. Code § 107 [0]. I'll paste here for your benefit, #3 is the one I referenced as most obvious but #1 and #4 are also very relevant:

If you're going to invoke fair use, that opens up a whole can of worms on what counts as transformative. The google books case and the google thumbnails case shows that you can make near verbatim copies of works at scale and still be considered fair use.

>The US justice system is too captured and corrupt at this point to take as reference because decisions there are bought by the highest bidder. But for the purpose of this discussion let's not play dumb for the benefit of trillion dollar corporations.

This is begging the question. The original question is whether ai companies are getting special treatment. You can't then use that as a premise to say that the courts are tilted towards ai companies. Not to mention it's questionable how ai companies were suddenly able to corrupt all the judges, some of which were appointed decades ago, even though they only got rich a couple of years ago.

[mapontosevenths]

> Can you remember every part?

No, and neither do LLM's. They're trained on vast quantities of data and retain only a fraction of it.

You might think of it as very, very lossy compression that generates new outputs rather than the original input unless something unintentional happens.

> If you just ignore anything that's inconvenient for your argument, you can make any argument you want.

I'm not. I just understand how it actually works. You either don't understand or are deliberately ignoring that what you just said is literally and technically untrue to make some sort of political statement.

[bee_rider]

Does the law really not distinguish between mechanical processing of data, and humans learning from it? It seems surprising to be if every person who read a textbook is copyright infringing. It also seems surprising if something like a lossy compression algorithm is enough to protect you from copyright law.

Somewhere between the two a line must be drawn… where we’d want to put that line, I guess, if up for quibbling. But it doesn’t seem obvious to me.

[gruez]

>Does the law really not distinguish between mechanical processing of data, and humans learning from it? It seems surprising to be if every person who read a textbook is copyright infringing. It also seems surprising if something like a lossy compression algorithm is enough to protect you from copyright law.

The google books and google thumbnails cases have so far upheld that even mechanical reproductions are allowed, depending on the context/usage.

[mapontosevenths]

To me the distinction hinges on the output being transformative enough to be considered a new work. I think that most of the time LLM output is.

Sometimes they go a bit wonky and overtrain on specific phrases which can result in verbatim copies of brief sections of coontent. Thats a bug, not a feature.

[triceratops]

If you write out the parts or recite them for other people to hear, yes it's copyright infringement.

Humans reading or watching copyrighted material isn't considered "making a copy" for the purposes of copyright law. Machines doing so generally is.

[lelandfe]

Further, why has my brain's searing remake of Snow White as a gritty murder mystery gone unscathed by Disney lawyers? Surely their negligence has diluted the Snow White trademark!

[JsonDemWitOster]

This analogy is disingenuous because by comparing the human brain to the machine, it ignores _scale_. Scale is absolutely important in copyright law. As a matter of fact, copyright law is among the various profound impacts of the---wait for it---printing press, a _machine_ for the mass production of books.

[mapontosevenths]

So if I watch a LOT of Disney movies THEN they own my own unique output forever?

[hartbook]

yes it is if you write it down from memory and sell it. Exactly what LLM companies do

[samrus]

They redistributed the statistical patterns of those copyrighted materials. Which perhaps should be treated similarly nos

As for your "technically not copyright infringement" defense. Those laws are from a time when those patterns couldnt be derived and dostributed at scale. A human had to learn and teach them. That made it different. The scale enabled my modern tech makes it a whole dofferent situation. The same way how one person standing a street corner people watching for a bit isnt that bad, but a whole constellation of flock cameras costantly montioring everyones movements and making it available to any of their customers is really really bad. The law will have to catch up to this

[vlovich123]

> They redistributed the statistical patterns of those copyrighted materials. Which perhaps should be treated similarly nos

Nos for the same reason that me giving you a word cloud of the frequency of words within Harry Potter isn’t infringement. It’s a novel transformation.

[samrus]

Thats not as complex or scalable as what LLMs can do. The capability and scale is what changes the equation. Quantity is a quality unto itself here

[vlovich123]

Exactly, and the complexity of what they’re doing is an even more significant novel creation.

Particularly in the US there’s a four point test and the very first point:

> To justify the use as fair, one must demonstrate how it either advances knowledge or the progress of the arts through the addition of something new.

I don’t know anything that has advanced the knowledge and progress of the arts more.

> The third factor assesses the amount and substantiality of the copyrighted work that has been used. In general, the less that is used in relation to the whole, the more likely the use will be considered fair.

This isn’t about usage in training. This would be in the LLM itself - the copyrighted works are very rarely used in the output.

> The fourth factor measures the effect that the allegedly infringing use has had on the copyright owner's ability to exploit his original work

Would you sincerely claim that owners have become less able to make money because of LLMs? Those same owners using LLMs to increase their own output of copyrighted works?

Anyway, copyright is not an absolute right and you have to really misunderstand copyright law to claim that LLM training infringes it.

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

[cryptonym]

This is confusing. I can torrent everything and do what I want with it, as long as I don't redistribute the exact same thing?

If so, why do we still pay for games and movies?

[midasz]

I pay for games because it's more convenient than pirating them. For movies and tv however... They make it so difficult to be a customer.

[klibertp]

Steam with Proton made gaming on Linux viable. Just for that, they deserve my money. That some of it goes to game devs is a happy coincidence ;D

[john_strinlai]

>I can torrent everything and do what I want with it, as long as I don't redistribute the exact same thing?

this is an incorrect interpretation (in the usa, at least).

downloading a game/movie is still the creation of unauthorized copy, which is not allowed. not to mention that playing/watching does not count as a "novel transformation".

(17 U.S.C. § 106 and 17 U.S.C. § 501 are the relevant pieces of reading)

[JsonDemWitOster]

IANAL (plus a whole suite of other caveats) but torrent-baiting works in Germany along these lines.

ISPs and trigger-happy law firms don't send you a C&D for downloading a torrent, they do so for seeding a torrent. It's just that practically nobody "just seeds" a torrent so people colloquially claim they got busted for downloading a torrent.

In theory this means if you torrent as a 100% leecher and turn off seeding from the get-go, you should be in the clear. But nobody sensible would dare test the extent of German Legal Spite, much less do so repeatedly to science the shit out of it.

If you can download through another protocol, say HTTP, however---<Sendung unterbrochen!>

[vlovich123]

No, that’s literally why Anthropic got sued. If they’d paid for a copy of the copyrighted works they pirated, they wouldn’t have had a problem. There were two issues in their case: does the AI infringe on copyright and did Anthropic obtain all their materials legally. The first they won on, the second they lost.

So if you pirate a bunch of content you still get in trouble for that. But if you somehow make a business out of that that isn’t just redistributing those materials, then that business itself isn’t infringing.