If someone were to train a coding-oriented LLM on only GPL code, I would assert that any code it outputs is covered by GPL because an LLM is fundamentally storing and reproducing its input, not understanding it and generating new things based on that understanding.
The starting point of GNU was that Stallman was pissed he got in trouble when he got caught copying code from the Symbolics sources to the MIT and LMI sources, which was against the agreement Symbolics and LMI had with the AI Lab, which was that improvements could only flow one-way (AI Lab to commercial). Dan Weinreb (RIP) confirmed this publicly.
Of course, not long after starting GNU, Stallman got caught copying code from Unipress emacs sources into the then-new GNU emacs sources. Oops! That’s why it was difficult for quite a long time to find early GNU emacs sources online—they were purged from various archives because they were infringing.
Someone tried to contribute “vibe-coded” device support to a project I’m involved with, they said they did it all based on the device documentation, the code their agents spit out was copied verbatim out of a (GPL’d) project with which I’m familiar which supports that device.
LLMs are not learning things and then using that learning to construct new things. They are essentially a form of lossy compression of their training set. And you don’t need to be explicit about trying to reproduce a portion of that training set for an LLM to output one.
As it happens, all evaluations I have seen in the news were in fact explicit about trying to reproduce a portion of the training set.
I am not aware of any study attempting to measure unintentional reproduction.
With your example, I question whether you have seen this happen first hand. For all I know, the contributor could have explicitly prompted the model to reference the GPL project and had the agent clone the code from the web.
At a certain point you have to take people at their word; I’m reporting what the contributor said they did (used the documentation to generate the code).
Not really, no. There have been many settlements of lawsuits, which means they are very explicitly not precedents. The LLM companies likely prefer to settle because of the existential risk being found liable for copyright infringement would result in, whereas even an expensive settlement doesn’t upend their entire business model.
Furthermore, I don’t think you can really assume that the courts will rule a certain way on this just because of the money involved; there’s a lot of money involved when it comes to the copyright holders too, and they’ve long enjoyed a rather favorable status with the courts and legislators. (For example, in the days of P2P file sharing lawsuits and attempts to legislate P2P file sharing, the software industry was already many times the size of the media industry, but the media industry consistently won.)
> Furthermore, I don’t think you can really assume that the courts will rule a certain way on this just because of the money involved; there’s a lot of money involved when it comes to the copyright holders too, and they’ve long enjoyed a rather favorable status with the courts and legislators.
I don't think this is all that plausible, even though I agree with you that it's not settled law. The size of the AI industry is gigantic, and a ruling that they are infringing the copyright of every piece of content in their training set would essentially shut them down entirely. Such a decision, if final, would probably easily wipe out a few hundred billion dollars on the stock market. Even if any court was willing to go that far, almost certainly lawmakers would step in and modify copyright law to prevent this from happening - both in the USA and the EU.
I don't think there is any comparison to make with the file sharing battle. That was a much, much smaller industry, it was not a significant chunk of the total hardware and software industries. Plus, the software titans were not nearly as well connected politically as they are today.
A couple of things: The “AI industry” is nowhere near as large as it pretends to be, most of its valuation is in the form of IOUs being traded back and forth. However, the bubble popping probably will wipe a few hundred billion of value from the stock market, because that’s what happens when a speculative bubble pops.
The second thing is that I’m not necessarily talking about whether _a specific LLM itself_ infringes copyright, but whether _its output_ is covered by the copyright of _its training material_. Whether training an LLM is an activity that infringes copyright is not well-settled in any precedential way, whether the trained LLM as an artifact infringes copyright is even less settled, and whether the output of that LLM is either infringing or covered by copyright is also not settled. These are all still extremely open questions.
That means anyone doing reasonable risk management should not just blithely race ahead and assume that there’s no infringement, which appears to be the approach the GCC project is taking explicitly and which also appears to be the approach projects like Linux and LLVM are taking implicitly (mostly through weasel-language like accepting responsibility for code you’re submitting).
"To summarize the analysis that now follows, the use of the books at issue to train Claude and its precursors was exceedingly transformative and was a fair use under Section 107 of the Copyright Act. And, the digitization of the books purchased in print form by Anthropic was also a fair use but not for the same reason as applies to the training copies. Instead, it was a fair use because all Anthropic did was replace the print copies it had purchased for its central library with more convenient space-saving and searchable digital copies for its central library — without adding new copies, creating new works, or redistributing existing copies."
IANAL and don't know how significant this decision is, but it is, at the very least, how one judge views it.
Personally, I don't think judges will rule a certain way because of the money involved but because it seems clear that training a ML model is highly transformative.
Not quite the precedent that it may sound like. A district court judge ruled that using copyrighted materials for the training itself was not infringement, but that the materials must be obtained legally.
Anthropic is trying to settle the case with most plaintiffs with respect to obtaining their works in an infringing way, but there are still plaintiffs pursuing the case on both the grounds that the remedy is insufficient (being only about $3000/work, when it has been as high as $250K/work in other copyright infringement cases and via statutory damages) and also on the grounds that the ruling that training is fair use was an error of law on the district court judge’s part.
Notably it doesn’t cover whether the output of the trained LLM continues to attach the training set’s copyright, which is independent from whether the training itself was an infringing activity. And there’s a substantial argument that the judge erred, if it can be shown that the training works are stored in a recoverable manner (even with some loss/defredation) rather than more extensively transformed.
the path of most resistance... sometimes i wonder if apple's insistence on a mac + paid membership is a kind of intentional filter, like they want someone who's devoted enough to invest 100 bucks + a mac etc
Because it would cost Apple a substantial amount of money to make it possible to build your code for iOS on other platforms in any sort of supported fashion. So all the stuff they invest very substantial amounts of money into creating and maintaining are only going to be licensed for use on the hardware they sell under the latest operating system it runs. Asking for more is asking them to spend more money without any evidence of a commensurate return.
If you can produce a properly structured submission to the App Store ingestion pipeline without using a Mac, then I doubt Apple will care. But that doesn’t mean they’re obligated to help you, and especially not to spend millions of dollars to do so.
The Apple developer terms of service require all app submissions to come from a Mac.
Apple could've chosen to offer no tools and leave it to the community to build their own. I think that's reasonable. I don't expect Microsoft to bring MSVC to Linux.
But Apple take the explicit position that you must purchase a Mac in order to ship apps for iOS.
It’s not really that onerous for a platform to require that, to develop for it, you need to actually take part in it. For Apple platforms that means becoming a Mac user. For IBM i and z that also means getting substantial hardware or paying for accounts and licenses (once you want to go beyond learning).
You can develop for Linux and BSD and Haiku and all sorts of other platforms for free. If the cost is what matters to you, then just do that. If you’re doing it as a business, though, consider that you’re extremely lucky to be able to just buy a low-end device and a low-end system and that’s all the capital investment needed (plus the US$100/year membership as recurring opex).
Xcode’s a lot more than a wrapper for UNIX utilities. For example, it has an entire build system. It doesn’t just invoke xcodebuild under the hood—and even if it did, xcodebuild uses the exact same code to build your projects as the IDE.
Most of the problems that I saw people have with Xcode over the two decades that I worked on it were the result of one fundamental problem: They wanted to work a different way than Xcode is designed to work, and jumped through a ton of hoops to do so, and then blamed Xcode for not accommodating their intentional working at cross-purposes.
A good example: An app that considered itself “large” because it had a thousand tiny static library targets each with only one to three or so ObjC classes. Far smaller than Xcode itself, but because they insisted on working that way—for no real articulable benefit besides “that’s how we want to work”—their project took hours to build from scratch instead of minutes like much larger projects that have many fewer targets.
In my experience, it’s quite “crashy.” It often locks up so bad, that I need to force-quit (or force-restart the Mac). I end up doing that once a day, or so. I suspect that many of the performance and stability issues, have to do with resource usage. It’s a damn big app.
It can also end up in strange states, where portions of the UI fall “out of sync,” and the app needs to be restarted. Some operations (like opening a storyboard, or stopping at a breakpoint) introduce massive delays.
It’s not a denigration, calling something a UI shell. I’m working on one, right now. The engine and backend were done in April (started in February), and all the work since, was on the UI shell. I probably could have shipped something that “ticked all the boxes,” in May. It’s almost done, but there’s still a couple of months of testing, ahead.
reply