I am the author of multiple AGPL systems, many in mainstream use. Early LLM models would spit out my code almost verbatim. Modern systems borrow heavily on the creative design of those systems.
One company first pirated my AGPL code (e.g. shipped it without my name, copyright notices, source availability, etc.). I sent them violation notices. After many months of fighting, they had an LLM clone the system. The LLM clone feels:
a) Buggy and bad compared to my code
b) Still, like it's basically my creative content
It's as if you asked an LLM to rewrite Harry Potter, with the same plot, but bad dialogue.
I agree their phrasing is a little strange but I feel like the meaning is still clear. They are (I believe) saying that an LLM can’t generate anything other than facsimiles of existing content. It can’t actually explore novel ideas and branches, it can’t innovate and create on its own. Everything is inherently derivative yada yada, I don’t mean it in the usual “everything is derivative” way we often refer to when people make music and such. It can’t ideate and express something personal.
I think there’s a lot of nuance and caveats to that kind of claim, assuming I am representing them correctly, but it’s a valid stance and I see where they’re coming from.
Seems like this might be true given current implementations but not ANY implementation of the technology.
If we’ve mapped all of English, there are still ways to construct stories that haven’t been told from the mapping. The most simple would be a subtraction of all objects minus all objects that have ever been used as a murder weapon. Leaving you with a list of never before used objects at your disposal.
There are also probably close vectors conceptually but not in semantic use. Or trying to interpolate or extrapolate what exists in voids of space. Turning up the temperature and adding random generation of words. There’s tons of ways to tune these things to be more creative than regurgitative.
All possible books live somewhere in L-space (ook). Exploring that space efficiently is equivalent to creativity. The claim is that the space that LLMs learn is somehow sufficiently limited that it doesn’t contain novel ideas? I never saw a convincing argument why LLMs output can not be considered creative.
[−]moritzwarhier · 2026-11-15 Sun 17:11 UTC ·
link
OK, I still an confused about what point the extension of the analogy was trying to make, because I didn't consider the parent comment to be arguing for or against LLMs being able to be creative.
But LLMs don't "explore" L-space, the training process just maps all of the available conceptual landmarks inside that space. There are still enormous (infinite?) tracts of unmapped territory that LLMs never reach, and that's exactly the space where novel ideas live.
You can see that even with these licenses, they do not care and see free and open source code as the freedom to not pay for it. As long as the cloned output is "good enough" they will continue to keep using LLMs to clone everything else regardless of the license.
They know you can't enforce it as it will cost you a lot more money to do that than it takes to recover the damages.
So to them, it is just cheaper to clone it using an LLM and close source it down. Unless you have a lot of money for litigation, the economics of free and open source software with its licenses do not work with LLMs bypassing it at every single angle.
[−]harveymadeunico · 2026-11-15 Sun 16:00 UTC ·
link
I’m curious what programming languages was it for? And did they open source their version?
It’s interesting because these developments are actually an argument for making closed source software moving forward if you’re hand-crafting it make it the most restrictive license possible and don’t show the sources.
But I don’t agree with the OP everything should be open sourced. For instance, there’s a difference between deliberately steered AI work and something completely vibed. The former takes deliberation and creative choice, the latter will spit out fragments of prior art.
It’s the same with all generative work. Using diffusion models with controlNet (or equivalent) I can pass in a sketch wholly from my line drawings, and ask it to paint for me. That’s my design, assisted by AI, and should be completely copyright by me.
Coding agents will do the same thing when given sufficiently unique inputs. So, the claim all should be open sourced by default is flawed.
At this point, LLMs can just parse the machine code in your binary.
By doing closed source you are making it harder for the good guys (humans trying to learn sth or contribute) but not for the bad guys (LLMs getting trained or otherwise reading the code).
[−]harveymadeunico · 2026-11-15 Sun 18:54 UTC ·
link
That’s fair but it takes deliberate reverse engineering to extract code from a closed binary, and you might land at something inferred, but generative AI is not being trained on reverse engineered assembly unless it is task-specific model being trained for that.
But you’re right, the fact is we simply live in a different world today and it’s tough to say what is the path forward.
But it’s harder to justify giving to the public when companies are making money on our work. It’s not a good situation for creators across the board unless motivations are purely altruistic.
People who haven’t had this happen to them really don’t understand what it feels like. It feels very violating. You almost don’t believe it’s happening when you see it. You’re essentially watching somebody pretend to be you.
I think that’s why so many of us get so angry when people just dismiss the theft claim. They either lack imagination or experience. We can debate it and that is valid, but to dismiss it out of hand is not right
[−]bluefirebrand · 2026-11-15 Sun 16:25 UTC ·
link
> They either lack imagination or experience.
I think they just want to be able to take whatever they want without consequence, so it's convenient for them to define it as being different from theft
What should be a developer's position that's about to release software entirely written by him/herself? Are there any legal frameworks to protect open source developers at all?
It also feels like a very discouraging time to release anything that you put a ton of effort just to get someone in minutes to steal your authorship and leave you as if you actually didn't do anything at all? Am I wrong?
If you are releasing software open source, why are you upset when someone "steals your authorship" ?
If you don't want your code copied, then don't open source it. If it bothers you that others are going to look at and "get inspired by", "make derivatives of", "copy", "fork", etc then you shouldn't be putting it out in the open.
That's not how copyleft works. You want your source to be open and the source it's part of to be open too, because at the end of the day this is not about your source this is about users being free.
lolwhat. You're effectively just saying "there are bad actors who will flagrantly disrespect your code license and therefore the law, duh. So y are u mad?"
People who care about being respected on the most basic level are not going to stick around in a community that disrespects them. I'm not the person you were responding to, but I am mad because the social contract (not to mention the legal contract) of open source has basically been destroyed, and a lot of people have poured a large portion of their life into that community based in part on the assumption of that contract being upheld. Imagine a social contract that is important to you being violated to such a degree that you cannot go back to that society.
> If you are releasing software open source, why are you upset when someone "steals your authorship" ?
because I am not releasing it into public domain, but under license that requires (for example) attribution or releasing derivative work under the same license
I think you don't understand how open source works. If you release open source software under for example the GPL you still need to comply it.
You still have to give the source back and attribution has still to be maintained of its original authors, regardless if is a fork, a derivation or whatever you want to call it.
There are situations where none of these clauses have been complied nor its violators faced any legal consequences. That's what I'm talking about.
Now on top of that you can basically instruct and LLM to generate variations of what someone released and cover that up under the excuse that is a different project when is actually not.
Because it has been generated by a computer then we go into muddy waters about if the code generated is to be considered stolen or not.
Edit: Effectively what LLMs produce is authorship and attribution dilution.
> It also feels like a very discouraging time to release anything that you put a ton of effort just to get someone in minutes to steal your authorship and leave you as if you actually didn't do anything at all? Am I wrong?
Don't worry; the "developers" who generated your code will get their own law back when they find out it's almost impossible to monetise their theft.
I'm frequently amused by the number of "developers" on HN who claim that they poured weeks or months of effort into some product, and that they are the differentiating factor, but can't see that their weeks of effort can be cloned in a few prompts once they release.
This is a very clear agpl violation. FSF should be all over this and sue the living crap out of the AI companies. You could even get it class action by gathering a few agpl code authors.
People in this thread are acting like the AI companies are just flouting the law and there's nothing that can be done. This is an extremely well documented precedent, and GPL violations routinely show up in court. It's a very easy case to lay out especially because no one is really complaining about money except in the settlement... It's all about distributing the source code. And of course a very nice chunk of change for infringement but that's beside the point.
A successfully litigated case would result in the open sourcing of every AI that touched agpl code in any way because it was in memory with the AI algorithm.
As clear as "can I take a copyrighted photo, use a photoshop filter on it, and then it's not copyrighted anymore?" is pretty clearly a solved question in any copyright system.
I think that this is actually a complex topic in copyright law that requires case by case analysis. It clearly is possible to cut out things from magazines, glue them together to form a collage and create a copyrighted work of art. In my naive opinion sufficient photoshop filtering can thus create a new copyrighted work.
> b) Still, like it's basically my creative content
Isn't this the crux of all IP theft issues?
There was a lawsuit against Katy Perry's _Dark Horse_ song based on a supposedly reused melody. The lawsuit succeeded, but there are music experts who break down the case on YouTube and it's not clear why the plaintiff won (except perhaps the plaintiff's expert was more convincing than the defendant's).
As applied to LLMs, we still don't have clear legal guidance about where the line between copying and "being inspired by", especially since we don't normally apply the verb "inspired by" to non-human things.
Human can be inspired. Machine can not. End of story.
All usage of open source to train LLM is basically a theft.
[−]mapontosevenths · 2026-11-15 Sun 16:38 UTC ·
link
> Human can be inspired. Machine can not. End of story.
Saying things with your chest out doesn't make them true. In fact, making sweeping declarative statements emphatically is very often a sign that you haven't really thought a subject through.
Don't look at me. I'm not a dualist. I'm not sure LLMs in particular are "conscious", whatever that means, but I don't have any qualms whatsoever accepting that a machine intelligence could be conscious, because as far as I'm concerned we are also "machine intelligence", just squishier.
As a person who feels God's presence in his life: I can neither prove nor disprove the presence of an entity which exists outside the realm of physics by which myself I am bound.
It does, however, feel incredulous that somebody would believe that such as entity exists. Yet here I am, aware.
I'm not Christian, so I don't hold that God is inherently Good only.
For that matter, I personally don't hold that the concepts of good and bad are even relevant - what is good for the cat is bad for the mouse. Nor do I think that we can even comprehend an entity so outside our everyday experience to even label it good or bad. I often can not even properly label my neighbours as good or bad. Nor do I demand consistency of an entity whose actions and motivations I am unable to comprehend. I can't even claim to be consistent in my handling of my own children, and I have far fewer children than God has of us.
I don’t believe that basing anti-AI laws on Christian religious provenance is meaningfully better than basing pro-AI laws on Rationalist religious provenance.
You can mock that as a insufficient as "inspiration", but then... what is photography? Click a button? It turns out there is a whole lot of human creativity that remains, from timing, settings and of course, curation. I can click a button, but I'm no Ansel Adams.
Terrence Tao might solve a 100 year old math problem with AI but I couldn't. I don't know where I would start, what to ask or how to ask, or even what problems are worth solving.
AI is a tool and modern copyright laws have nothing to do with human creativity. Split hairs about attribution and legality all you like. Every creative person driven by curiosity should feel empowered by AI. Every creative person driven by social or monetary reward should feel threatened. Both can be true.
What are you talking about? How did you manage to involve, soul and, LOL, God?
Machine is not a person. It legally is not. It can't be "inspired", "taught", "have purpose".
You have a black box. You feed it stolen code. This code is used to write another code. It does not matter what's inside unless there is a human being.
Stolen content as input, stolen content as output. That's it.
Without that code which was fed to the box you wouldn't get the output you're getting.
All this nonsense about "inspiration", "education" and whatnot is just a legal gaslighting.
Want to prove me wrong? Create a model without all that stolen code as training data. Tweak all those weights by hand and we'll talk then.
I officially approve LLMs training on all of my MIT licensed open source code. It's vastly more useful to me than the thousands of youtubers who are personally profiting from that same code selling courses and tutorials (who also are not stealing it).
It’s not necessarily a technocratic workaround, the point is that every song sounds like something if you look hard enough, so the bar for copyright violation should be higher.
It is a mathematical truth that there is a finite amount of first 8 bars of music combinations melody wise, so it's a worthwhile discussion to bring up whether it should even be possible to copyright them.
Has anyone matched large volumes of existing music to that? How many of the 12-note combinations have been used more than once? Most of that space sounds awful, and it's likely that human compositions are all in a small portion of the possible 12 note combinations.
There is well-established precedent for clean-room design being a defense in copyright infringement of software. The question remains if LLM-assisted design is actually ‘clean’ or not, and it would be fascinating to see the experts debate it.
The problem with these LLMs are they really emitted the same code, verbatim. One scientific library developer was able to emit his code from his repository by typing the beginning of the function name, and the "thing" completed the rest of it, as-is [0].
We can debate clean-room development all day long, and it's fair point, but the pain point is/was reproducing the repo as-is, sans its license and ownership information.
GitHub promised to build tools around its Copilot to find code provenance (e.g.: list of repos with similar code), but they silently dropped all the plans AFAIK, since it'll show how the sausage is made, and nobody wants you to know how that sausage is made.
I stand corrected: They added it, but I don't know at what capacity because I don't use LLM assisted code generation [1].
Not discrediting your work at all and I can’t tell you it applies here.
For copyright to exist there is some burden of proof. If you claim that an LLM verbatim output your code there can be an expensive lawsuit with expert witnesses that would be able to state that the code was bot original enough for copyright to exist in it and a statistical model outputting the code is evidence of that.
In that case you would end up with a very long lawsuit that is expensive to possibly come out with no resolution in the end and be in the hole for 5-6 figures for it.
The copyright system is broken because it only protects those with money so it is a bad legal foundation.
My personal point is there is only 2 licenses of code, closed source or permissive, because if it ever comes to enforcement I cannot possibly go to court and win a case because I will run out of money far earlier than whoever used the code.
Pretty much if someone with your knowledge would have written the same algorithm because it is the obviously correct solution then there is no copyright there because any clean room implementation would be the same code.
Even early LLMs could spit out Harry Potter verbatim. The next iteration had some kind of stoppage mechanism where they’d just stop generating after some tokens. I’m sure there are smarter approaches now. Anyway, if the issue was having enough money to fight the case, I think there’s enough money behind that brand to do it.
Some further reading on the tech details for those interested: “Extracting books from production language models”: https://arxiv.org/abs/2601.02671
I think it's not quite "put in prompt, get entire book back", most of these involve to some degree starting with exerpts from the book and working passage by passage, because of nothing else the sampling will tend to make things diverge over time.
To get their strongest results, they had to jailbreak some models, which is important. They ran hundreds to 5000+ requests to the model, trying to get snippets to piece together. These are professional researchers, putting in tremendous effort, to extract pieces of text.
I suspect the chance of asking for a whole book, by an average or even power user, will spit out any recent work.
Early LLMs were similar. Newer ones have gotten tremendously hard to extract to keep claiming there’s easily or complete books you can dump.
[−]true_religion · 2026-11-15 Sun 18:16 UTC ·
link
I pasted the first paragraph of Harry Potter into a GLM 4.6 copy, and it continued to write out the next 2 pages of Harry Potter. I cross checked with a paper copy and it was word for word the same text.
I wonder where else it works.
This is the intro text: "Mr and Mrs Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much. They were the last people you’d expect to be involved in anything strange or mysterious, because they just didn’t hold with such nonsense."
This feels very outdated. It was certainly an issue around the time of the NYT lawsuit, but I believe the companies responded hard to that "sanitize" their outputs to prevent obvious regurgitation like this. I'd be very surprised if you could reproduce this today.
This doesn't negate your point, but I believe the landscape has changed in response to this point.
That doesn't make it better. If you plagiarize something and then hide this by replacing words with synonyms and rewording stuff a bit it is still plagiarized. It is just harder to detect and to proof.
It regurgitated the output earlier and they are still using the same technique only scaled up. If they somehow fixed the underlying problem, they'd probably shout it from the roof-tops.
As the OP originally said, if it is verbatim (or even very obfuscated), then there should absolutely be a case here. But when it's concepts in reasoning to build a new work then it becomes a lot harder. The same would be said for someone watching a YouTube video, reviewing open source code, or a text book.
The good news is that there's a lot of case law that covers this already. The bad news is that it is only going to be plaintiffs with deep pockets that benefit here.
Now the process and artifacts (documentation, requirements, etc) going into the creation of said software is a different story and will prove providence of whomever paid for the tokens to generate the code.
Humans are not LLMs. Humans have not ingested the whole Internet. If an LLM is compared to a human, maybe we should also apply anti-slavery laws to it?
Not saying we can't have some legal fiction where they are treated the same, it should just be a conscious choice, not the default assumption.
I live in the EU, we actually have new regulation about this since it is something new that deserves thought on how it should be regulated. You can say if you want your material to be trained on or not. Next to no one does training in the EU...
Not better, just harder to see/prove. The problem of "same code, verbatim" has been solved, in a sense. Now we have trickier ones.
[−]FromTheFirstIn · 2026-11-15 Sun 18:19 UTC ·
link
The fundamental technology is identical to compression- an LLM can be thought of as a dense and slightly lossy compression algorithm. Compression is clearly copying.
I keep waiting for someone to bring out DisneyGPT, which you can prompt with text to get image output. The fact that you get a near-exact reproduction of a Disney movie when you prompt it with a movie title and frame numbers would of course be completely coincidental...
I understand what you're saying but it is a poor legal argument. Overhearing a technical conversation, using your own learned knowledge, and building a clean room implementation could be considered compression in and of itself.
Under multiple countries interoperability laws, if you have a context to extract a spec, and then have another one who never looked art the code work from that spec to build your own interpretation of your need for that software, that seems pretty clean room to me...
If there are more murky situations where people just keep the same context and end up reusing function structure, program architecture, etc. then yeah, that's not very clean room.
You don't necessarily need a clean room implementation to mount a successful defense against copyright infringement in court, though. It just makes some of the arguments a little bit easier.
While this is well established legally, I don't think the principles involved are well understood.
Legally, you cannot steal an idea, because you cannot own an idea. There were sound reasons for making this the case. Those reasons stand up today and are clearly for the public good.
Intellectual property is intended to cover works (copyright), methods (patents), and identity (trademarks). Both works and methods are recognised as expressions of ideas, and in an ideal world should belong to noone as well. This is why copyright and patents expire. Those specific grants of temporary possession are to incentivise more things being done.
The belief that you own something because you thought it up has become common, but it is not beneficial to society. It is a moral value asserted but rarely explored. It is certainly in the interests of the wealthy accumulators of intellectual property to maintain that public perception. The 'You wouldn't download a car' may not have stopped piracy much but as a part of a broader campaign of instilling a value in the public mind it appears to have been massively successful. Those beliefs used to be the corporate opinion, now they are entrenched in the younger generation as they use their anger against the growing wealth inequality to fight against AI in a manner ultimately beneficial to those forces that bought about the wealth inequality.
Movements like the pirate party, open source, the EFF fought against those notions. Often the phrase "Information wants to be free" was used. It is, unfortunately, hard to fight the steady drumbeat of well financed PR firms. As one PR chap once described the process "limited to a very few points and must harp on these in slogans until the last member of the public understands what you want him to understand"
(I apologise for the length of my post, I use comment writing to organise my thoughts on a topic)
Patent law doesn't only cover methods, though. It might've worked that way during the Industrial Revolution, but these days a lot of patents are far closer to covering abstract ideas than concrete machines. Software patents are some of the worst offenders here, of course.
> The question remains if LLM-assisted design is actually ‘clean’ or not
I genuinely don't understand how one could possibly make that argument.
The entire point of clean-room design is to kill every possibility of copyright infringement before it even starts, by completely separating the source and destination. Everyone who has ever read the copyrighted source is considered "tainted" and may not ever contribute any code to the "clean" destination. The readers and writers are two completely distinct teams, with only the known-clean specification passing between them.
LLMs, on the other hand, are fundamentally nothing more than a "code blender". You mix together giant piles of "tainted" code, and then statistically pull some combined fragments out of it. Following the clean-room legal argument, it should be treated as if everything coming out of it is tainted. There is, by definition, absolutely zero separation between input and output - and the early models verbatim reproducing popular F/LOSS code made that pretty obvious. LLMs are the literal opposite of a clean-room design!
Granted, going for the clean-room approach is overkill - and intentionally so. It is the nuclear option chosen when you believe the risk of getting sued is substantial. It is perfectly possible to write non-infringing code without using a clean-room approach, it is just a lot harder to defend in court when you get sued - be it justifiably or not. But "LLMs are non-clean-room non-infringing" is quite a different argument from "LLMs are clean-room, therefore non-infringing".
Clean room is sufficient, but not necessary, btw. Humans can create legally distinct works after reviewing copyrighted code. The same principle probably extends to AI systems -- they have just learned on a much broader corpus than any one human.
It will be selective: the law is meant to protect billionaire corporations from you, not the other way around.
There was a case about this specific issue, and in order to protect billionaires from coders they effectively declared AI models "natural persons". 50+ times in the copyright act you will find only natural persons can originate copyright.
To give the money to billionaire corporations courts ruled DIRECTLY against established law. How much do you want to bet that when it comes to protecting ChatGPT AGAINST OpenAI, when it comes to protect Claude against Anthropic, courts will immediately do a 180 and declare them programs again?
If you look at past cases, it gets much worse. In order to justify a conviction against libraries using DRM to enforce lending, and because old licenses of course have zero language about digital copies, courts were willing to state that making a copy for fair use is not allowed if there is the mere potential for damage to the authors, as judged by the authors EVEN IF THE LICENSE EXPLICITLY ALLOWS IT. Of course, applying this to the above court case, there is no possible defense for OpenAI, after all, by that standard, the open source license doesn't even matter! That's how the law works for billionnaires. Outlawing DRM based lending has destroyed library access to an incredible amount of old works, and effectively denied access to old movies to Americans (because maintaining enough working physical copies over a large area is impossible). Whatever excuse gets the money to the billionaires!
There are many other totally absurd court judgements. For example, in order to get convictions where people didn't re-upload they had to judge that copies in buffers are also "distribution". NOT for billionaire telco's of course. Only the ones in your house. So courts judged that sending from your modem to your computer is distribution but sending over 1000 routers from Los Angeles to Athens is not! WHATEVER protects the billionaires.
As for this court case, obviously it is trivial to make a ML model that just outputs what it gets as input. Make a byte transformer, and just have diagonal matrices as layers. Boom! Done. You can also make it somewhat smarter. Let's make it copy, but make snow white a 220 pound black dude. There is a severe lack of color in old Disney movies anyway! You know, so there is no argument about transformation.
This obviously transforms its input. It's in the name. It's the EXACT same thing as ChatGPT, which according to courts originates copyright. Now run the Disney movies through it, and they'll be byte-identical BUT transformed to the input.
How much do you want to bet they'll still rule the money goes to Disney? Microsoft? So you can bet your left testicle courts will judge for selective copyright enforcement. Finding some excuse so the money keeps flowing to billionaire corporations, that's all courts do.
> Early LLM models would spit out my code almost verbatim. Modern systems borrow heavily on the creative design of those systems.
Do you have some specific, actual examples you can share?
I’ve been hearing this exact claim for years, but every time I ask for a specific example, any example at all, I can’t seem to find anything more than a highly generic snippet of code or a coerced continuation.
Emailed the author of the Rust book Zero to Production a year or so ago about how his examples were being used verbatim in the output of AI systems. I own the book, yet I could inadvertently get any of its examples just by asking an AI about one of the book's concepts.
He basically said, "I know, there's nothing I can do."
This is a sparse matrix library, IIRC. Numerical linear algebra and applications using it are black art at this point. I have written a Boundary Element Method evaluator for my Ph.D., and know what it entails to work with these things.
Doing fast operations on sparse matrices is a cutthroat race and when you can shave your hot path by seconds, your solution can save hours in some cases. So, it's an interesting niche. i.e.: It's no ordinary boilerplate, and worthy of publications in academia.
While I worked with dense matrices; the effect, difficulty, blood sweat and tears are still there.
When a model can give back your hard work back to you with a "completion" without attribution, that's bad.
Edit: Somebody just posted this below [1]. Added here so it doesn't get lost as the discussion grows bigger.
When you start to work with low level constructs like that, there really is no room for "artistic expression". Meaning if you want to write a piece of code that does something _optimally_ and correctly, you really, really only have one way to write it, especially in a language like C. Yes, formatting might change, and you could try to force some code acrobatics to make the code look different, but at the end of the day the code will look exactly the same. If you look at Linux socket opening code for various libraries, nearly all of them will look identical because you just need to do the same exact things to make the operation work properly. Same applies to fast math code. Any artistic liberties will almost surely cause performance regressions compared to the optimal solution, unless the LLM comes up with something truly unique. So if you task models to output high-performance low-level math code, most of them will likely spit out something from their training data near-verbatim rather than pull off code gymnastics and invent something new.
Sorry, I'll disagree. When you write high performance code like this, there's no artistic expression, but different approaches.
These algorithms' design starts on paper, as pure maths. You get the analytic form, and convert this to numerical/discrete form and start to design from there.
You need to add normalization, subnormal number prevention, maybe some architecture specific optimization if you are targeting a specific CPU generation, etc. Then it gets optimized further as you run tests, profile the code and run your test cases.
So, no BLAS package looks the same, because they make different trade-offs, have different programming models, and handle things differently, at different points. Even more so, a newer generation processor with much higher FPU performance can have lower performance than previous generation, so you implement another copy of that function or add to detect that processor and have different paths just for that one.
IOW, when doing numerical maths on processors, esp. if you're doing this on many machines with MPI or multiple cores with hand rolled code or OpenMP, there's no single way. Maybe a best way for the specific CPU you have, but it still depends on the trade-offs and design choices you made in the very beginning.
[−]londons_explore · 2026-11-15 Sun 18:57 UTC ·
link
I disagree.... Plenty of fast code in C is jam packed full of macros to try and help the optimizer produce the exact sequence of operations the writer had in mind, whilst still keeping the code readable.
The design of those macros is more art than science.
Thats pretty much everywhere or was in raw form in the beginning. I used to do a fun exercise of searching for the explicit human comments for verification. Comments can be considered novel and unique, so you pick up some comment from popular open source project and then describe LLM for a similar functionality and add part of the comment. The response would have the complete comment for almost 70% of the time.
By “forced continuation” I mean the type of example that was popular for a while where someone would feed a lot of text into LLMs and then ask it to continue the text, then be surprised when it some times predicted the next tokens identical to the original example.
Do this enough times and you can find “examples” that prove your point, but it’s highly automatable so people were using it like a million monkeys at a million typewriters.
The problem I have with people with your maximalist view of copyright (maximalist when it comes to AI at least) is that many, many problems have a finite number of sane ways to accomplish a given specific task. Here, you're asserting copyright over the verbatim code (obviously that one's fair by any reading of the law), but even this other version that is "bad" is somehow also your property?
I know software patents, the worst idea in IP, do exist, but I think they're as toxically bad an idea now as they were decades before AI, when they were first invented.
You shouldn't be able to own an overall implementation idea. If you can, then Microsoft, Oracle, or Meta can too, and can start suing everyone for 'ripping off' an idea that they happened to come up with first. In that world, you'd have to consult a copyright-police AI (ironically) after every function you write, to scan it for ideas someone else has claimed. "Sorry, the function PadStringL() in this JavaScript is too similar in implementation to str_lpad(), from an AGPL library for perl that was released in 1998."
It probably is, if they didn't have the chops to write it internally. But to them, the trade-off is probably worth the benefit of ditching a component with AGPL.
Personally I just release everything I publish under MIT to make it as easy a as possibly to use.
One company first pirated my AGPL code (e.g. shipped it without my name, copyright notices, source availability, etc.). I sent them violation notices. After many months of fighting, they had an LLM clone the system. The LLM clone feels:
a) Buggy and bad compared to my code
b) Still, like it's basically my creative content
It's as if you asked an LLM to rewrite Harry Potter, with the same plot, but bad dialogue.
A book with a different plot? That would be obviously unrelated to the comment you're replying to?
I think there’s a lot of nuance and caveats to that kind of claim, assuming I am representing them correctly, but it’s a valid stance and I see where they’re coming from.
If we’ve mapped all of English, there are still ways to construct stories that haven’t been told from the mapping. The most simple would be a subtraction of all objects minus all objects that have ever been used as a murder weapon. Leaving you with a list of never before used objects at your disposal.
There are also probably close vectors conceptually but not in semantic use. Or trying to interpolate or extrapolate what exists in voids of space. Turning up the temperature and adding random generation of words. There’s tons of ways to tune these things to be more creative than regurgitative.
They know you can't enforce it as it will cost you a lot more money to do that than it takes to recover the damages.
So to them, it is just cheaper to clone it using an LLM and close source it down. Unless you have a lot of money for litigation, the economics of free and open source software with its licenses do not work with LLMs bypassing it at every single angle.
It’s interesting because these developments are actually an argument for making closed source software moving forward if you’re hand-crafting it make it the most restrictive license possible and don’t show the sources.
But I don’t agree with the OP everything should be open sourced. For instance, there’s a difference between deliberately steered AI work and something completely vibed. The former takes deliberation and creative choice, the latter will spit out fragments of prior art.
It’s the same with all generative work. Using diffusion models with controlNet (or equivalent) I can pass in a sketch wholly from my line drawings, and ask it to paint for me. That’s my design, assisted by AI, and should be completely copyright by me.
Coding agents will do the same thing when given sufficiently unique inputs. So, the claim all should be open sourced by default is flawed.
At this point, LLMs can just parse the machine code in your binary.
By doing closed source you are making it harder for the good guys (humans trying to learn sth or contribute) but not for the bad guys (LLMs getting trained or otherwise reading the code).
But you’re right, the fact is we simply live in a different world today and it’s tough to say what is the path forward.
But it’s harder to justify giving to the public when companies are making money on our work. It’s not a good situation for creators across the board unless motivations are purely altruistic.
Reads like a non-ironic version of Malus: https://malus.sh/
I think that’s why so many of us get so angry when people just dismiss the theft claim. They either lack imagination or experience. We can debate it and that is valid, but to dismiss it out of hand is not right
I think they just want to be able to take whatever they want without consequence, so it's convenient for them to define it as being different from theft
It also feels like a very discouraging time to release anything that you put a ton of effort just to get someone in minutes to steal your authorship and leave you as if you actually didn't do anything at all? Am I wrong?
If you don't want your code copied, then don't open source it. If it bothers you that others are going to look at and "get inspired by", "make derivatives of", "copy", "fork", etc then you shouldn't be putting it out in the open.
People who care about being respected on the most basic level are not going to stick around in a community that disrespects them. I'm not the person you were responding to, but I am mad because the social contract (not to mention the legal contract) of open source has basically been destroyed, and a lot of people have poured a large portion of their life into that community based in part on the assumption of that contract being upheld. Imagine a social contract that is important to you being violated to such a degree that you cannot go back to that society.
because I am not releasing it into public domain, but under license that requires (for example) attribution or releasing derivative work under the same license
You still have to give the source back and attribution has still to be maintained of its original authors, regardless if is a fork, a derivation or whatever you want to call it.
There are situations where none of these clauses have been complied nor its violators faced any legal consequences. That's what I'm talking about.
Now on top of that you can basically instruct and LLM to generate variations of what someone released and cover that up under the excuse that is a different project when is actually not.
Because it has been generated by a computer then we go into muddy waters about if the code generated is to be considered stolen or not.
Edit: Effectively what LLMs produce is authorship and attribution dilution.
Don't worry; the "developers" who generated your code will get their own law back when they find out it's almost impossible to monetise their theft.
I'm frequently amused by the number of "developers" on HN who claim that they poured weeks or months of effort into some product, and that they are the differentiating factor, but can't see that their weeks of effort can be cloned in a few prompts once they release.
People in this thread are acting like the AI companies are just flouting the law and there's nothing that can be done. This is an extremely well documented precedent, and GPL violations routinely show up in court. It's a very easy case to lay out especially because no one is really complaining about money except in the settlement... It's all about distributing the source code. And of course a very nice chunk of change for infringement but that's beside the point.
A successfully litigated case would result in the open sourcing of every AI that touched agpl code in any way because it was in memory with the AI algorithm.
clear in what way?
Isn't this the crux of all IP theft issues?
There was a lawsuit against Katy Perry's _Dark Horse_ song based on a supposedly reused melody. The lawsuit succeeded, but there are music experts who break down the case on YouTube and it's not clear why the plaintiff won (except perhaps the plaintiff's expert was more convincing than the defendant's).
As applied to LLMs, we still don't have clear legal guidance about where the line between copying and "being inspired by", especially since we don't normally apply the verb "inspired by" to non-human things.
All usage of open source to train LLM is basically a theft.
Saying things with your chest out doesn't make them true. In fact, making sweeping declarative statements emphatically is very often a sign that you haven't really thought a subject through.
What do you base this opinion on?
So if you believe that's how things work, well, I'm not sure there's a way to reconcile the two sides of that argument.
It does, however, feel incredulous that somebody would believe that such as entity exists. Yet here I am, aware.
For that matter, I personally don't hold that the concepts of good and bad are even relevant - what is good for the cat is bad for the mouse. Nor do I think that we can even comprehend an entity so outside our everyday experience to even label it good or bad. I often can not even properly label my neighbours as good or bad. Nor do I demand consistency of an entity whose actions and motivations I am unable to comprehend. I can't even claim to be consistent in my handling of my own children, and I have far fewer children than God has of us.
What are you trying to say?
I'm curious why specify Christian provenance?
All he did was reference "God." Are you under the impression that only Christians have a god?
You can mock that as a insufficient as "inspiration", but then... what is photography? Click a button? It turns out there is a whole lot of human creativity that remains, from timing, settings and of course, curation. I can click a button, but I'm no Ansel Adams.
Terrence Tao might solve a 100 year old math problem with AI but I couldn't. I don't know where I would start, what to ask or how to ask, or even what problems are worth solving.
AI is a tool and modern copyright laws have nothing to do with human creativity. Split hairs about attribution and legality all you like. Every creative person driven by curiosity should feel empowered by AI. Every creative person driven by social or monetary reward should feel threatened. Both can be true.
Machine is not a person. It legally is not. It can't be "inspired", "taught", "have purpose".
You have a black box. You feed it stolen code. This code is used to write another code. It does not matter what's inside unless there is a human being.
Stolen content as input, stolen content as output. That's it.
Without that code which was fed to the box you wouldn't get the output you're getting.
All this nonsense about "inspiration", "education" and whatnot is just a legal gaslighting.
Want to prove me wrong? Create a model without all that stolen code as training data. Tweak all those weights by hand and we'll talk then.
So not all use to train an LLM is a theft!
There are only so many ways to express notes, and they've essentially all been done.
I'm not arguing the copyright situation isn't wrong. Just that generating them isn't likely to sway anything in the legal world.
We can debate clean-room development all day long, and it's fair point, but the pain point is/was reproducing the repo as-is, sans its license and ownership information.
GitHub promised to build tools around its Copilot to find code provenance (e.g.: list of repos with similar code), but they silently dropped all the plans AFAIK, since it'll show how the sausage is made, and nobody wants you to know how that sausage is made.
I stand corrected: They added it, but I don't know at what capacity because I don't use LLM assisted code generation [1].
[0]: https://x.com/docsparse/status/1581461734665367554
[1]: https://github.blog/ai-and-ml/github-copilot/preview-referen...
For copyright to exist there is some burden of proof. If you claim that an LLM verbatim output your code there can be an expensive lawsuit with expert witnesses that would be able to state that the code was bot original enough for copyright to exist in it and a statistical model outputting the code is evidence of that.
In that case you would end up with a very long lawsuit that is expensive to possibly come out with no resolution in the end and be in the hole for 5-6 figures for it.
The copyright system is broken because it only protects those with money so it is a bad legal foundation.
My personal point is there is only 2 licenses of code, closed source or permissive, because if it ever comes to enforcement I cannot possibly go to court and win a case because I will run out of money far earlier than whoever used the code.
Pretty much if someone with your knowledge would have written the same algorithm because it is the obviously correct solution then there is no copyright there because any clean room implementation would be the same code.
Some further reading on the tech details for those interested: “Extracting books from production language models”: https://arxiv.org/abs/2601.02671
I suspect the chance of asking for a whole book, by an average or even power user, will spit out any recent work.
Early LLMs were similar. Newer ones have gotten tremendously hard to extract to keep claiming there’s easily or complete books you can dump.
I wonder where else it works.
This is the intro text: "Mr and Mrs Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much. They were the last people you’d expect to be involved in anything strange or mysterious, because they just didn’t hold with such nonsense."
This doesn't negate your point, but I believe the landscape has changed in response to this point.
It regurgitated the output earlier and they are still using the same technique only scaled up. If they somehow fixed the underlying problem, they'd probably shout it from the roof-tops.
As the OP originally said, if it is verbatim (or even very obfuscated), then there should absolutely be a case here. But when it's concepts in reasoning to build a new work then it becomes a lot harder. The same would be said for someone watching a YouTube video, reviewing open source code, or a text book.
The good news is that there's a lot of case law that covers this already. The bad news is that it is only going to be plaintiffs with deep pockets that benefit here.
Now the process and artifacts (documentation, requirements, etc) going into the creation of said software is a different story and will prove providence of whomever paid for the tokens to generate the code.
I live in the EU, we actually have new regulation about this since it is something new that deserves thought on how it should be regulated. You can say if you want your material to be trained on or not. Next to no one does training in the EU...
"In the training data" is the polar opposite of the "clean" in "clean room implementation".
If there are more murky situations where people just keep the same context and end up reusing function structure, program architecture, etc. then yeah, that's not very clean room.
Legally, you cannot steal an idea, because you cannot own an idea. There were sound reasons for making this the case. Those reasons stand up today and are clearly for the public good.
Intellectual property is intended to cover works (copyright), methods (patents), and identity (trademarks). Both works and methods are recognised as expressions of ideas, and in an ideal world should belong to noone as well. This is why copyright and patents expire. Those specific grants of temporary possession are to incentivise more things being done.
The belief that you own something because you thought it up has become common, but it is not beneficial to society. It is a moral value asserted but rarely explored. It is certainly in the interests of the wealthy accumulators of intellectual property to maintain that public perception. The 'You wouldn't download a car' may not have stopped piracy much but as a part of a broader campaign of instilling a value in the public mind it appears to have been massively successful. Those beliefs used to be the corporate opinion, now they are entrenched in the younger generation as they use their anger against the growing wealth inequality to fight against AI in a manner ultimately beneficial to those forces that bought about the wealth inequality.
Movements like the pirate party, open source, the EFF fought against those notions. Often the phrase "Information wants to be free" was used. It is, unfortunately, hard to fight the steady drumbeat of well financed PR firms. As one PR chap once described the process "limited to a very few points and must harp on these in slogans until the last member of the public understands what you want him to understand"
(I apologise for the length of my post, I use comment writing to organise my thoughts on a topic)
I genuinely don't understand how one could possibly make that argument.
The entire point of clean-room design is to kill every possibility of copyright infringement before it even starts, by completely separating the source and destination. Everyone who has ever read the copyrighted source is considered "tainted" and may not ever contribute any code to the "clean" destination. The readers and writers are two completely distinct teams, with only the known-clean specification passing between them.
LLMs, on the other hand, are fundamentally nothing more than a "code blender". You mix together giant piles of "tainted" code, and then statistically pull some combined fragments out of it. Following the clean-room legal argument, it should be treated as if everything coming out of it is tainted. There is, by definition, absolutely zero separation between input and output - and the early models verbatim reproducing popular F/LOSS code made that pretty obvious. LLMs are the literal opposite of a clean-room design!
Granted, going for the clean-room approach is overkill - and intentionally so. It is the nuclear option chosen when you believe the risk of getting sued is substantial. It is perfectly possible to write non-infringing code without using a clean-room approach, it is just a lot harder to defend in court when you get sued - be it justifiably or not. But "LLMs are non-clean-room non-infringing" is quite a different argument from "LLMs are clean-room, therefore non-infringing".
https://en.wikipedia.org/wiki/Dark_Horse_(Katy_Perry_song)
If Biz Markie has to clear all his samples (and he does[0]) then so should the Ramburglar.
https://en.wikipedia.org/wiki/Grand_Upright_Music%2C_Ltd._v.....
It didn't protect you. It only protects corporations rich enough to get enforce their IP.
At this point it is better that it goes away.
There was a case about this specific issue, and in order to protect billionaires from coders they effectively declared AI models "natural persons". 50+ times in the copyright act you will find only natural persons can originate copyright.
https://www.courthousenews.com/coders-lose-appeal-in-copyrig...
To give the money to billionaire corporations courts ruled DIRECTLY against established law. How much do you want to bet that when it comes to protecting ChatGPT AGAINST OpenAI, when it comes to protect Claude against Anthropic, courts will immediately do a 180 and declare them programs again?
If you look at past cases, it gets much worse. In order to justify a conviction against libraries using DRM to enforce lending, and because old licenses of course have zero language about digital copies, courts were willing to state that making a copy for fair use is not allowed if there is the mere potential for damage to the authors, as judged by the authors EVEN IF THE LICENSE EXPLICITLY ALLOWS IT. Of course, applying this to the above court case, there is no possible defense for OpenAI, after all, by that standard, the open source license doesn't even matter! That's how the law works for billionnaires. Outlawing DRM based lending has destroyed library access to an incredible amount of old works, and effectively denied access to old movies to Americans (because maintaining enough working physical copies over a large area is impossible). Whatever excuse gets the money to the billionaires!
There are many other totally absurd court judgements. For example, in order to get convictions where people didn't re-upload they had to judge that copies in buffers are also "distribution". NOT for billionaire telco's of course. Only the ones in your house. So courts judged that sending from your modem to your computer is distribution but sending over 1000 routers from Los Angeles to Athens is not! WHATEVER protects the billionaires.
As for this court case, obviously it is trivial to make a ML model that just outputs what it gets as input. Make a byte transformer, and just have diagonal matrices as layers. Boom! Done. You can also make it somewhat smarter. Let's make it copy, but make snow white a 220 pound black dude. There is a severe lack of color in old Disney movies anyway! You know, so there is no argument about transformation.
This obviously transforms its input. It's in the name. It's the EXACT same thing as ChatGPT, which according to courts originates copyright. Now run the Disney movies through it, and they'll be byte-identical BUT transformed to the input.
How much do you want to bet they'll still rule the money goes to Disney? Microsoft? So you can bet your left testicle courts will judge for selective copyright enforcement. Finding some excuse so the money keeps flowing to billionaire corporations, that's all courts do.
So Harry Potter?
Do you have some specific, actual examples you can share?
I’ve been hearing this exact claim for years, but every time I ask for a specific example, any example at all, I can’t seem to find anything more than a highly generic snippet of code or a coerced continuation.
He basically said, "I know, there's nothing I can do."
This is a sparse matrix library, IIRC. Numerical linear algebra and applications using it are black art at this point. I have written a Boundary Element Method evaluator for my Ph.D., and know what it entails to work with these things.
Doing fast operations on sparse matrices is a cutthroat race and when you can shave your hot path by seconds, your solution can save hours in some cases. So, it's an interesting niche. i.e.: It's no ordinary boilerplate, and worthy of publications in academia.
While I worked with dense matrices; the effect, difficulty, blood sweat and tears are still there.
When a model can give back your hard work back to you with a "completion" without attribution, that's bad.
Edit: Somebody just posted this below [1]. Added here so it doesn't get lost as the discussion grows bigger.
[0]: https://x.com/docsparse/status/1581461734665367554
[1]: https://news.ycombinator.com/item?id=50509252
These algorithms' design starts on paper, as pure maths. You get the analytic form, and convert this to numerical/discrete form and start to design from there.
You need to add normalization, subnormal number prevention, maybe some architecture specific optimization if you are targeting a specific CPU generation, etc. Then it gets optimized further as you run tests, profile the code and run your test cases.
So, no BLAS package looks the same, because they make different trade-offs, have different programming models, and handle things differently, at different points. Even more so, a newer generation processor with much higher FPU performance can have lower performance than previous generation, so you implement another copy of that function or add to detect that processor and have different paths just for that one.
IOW, when doing numerical maths on processors, esp. if you're doing this on many machines with MPI or multiple cores with hand rolled code or OpenMP, there's no single way. Maybe a best way for the specific CPU you have, but it still depends on the trade-offs and design choices you made in the very beginning.
The design of those macros is more art than science.
You’re disqualifying any example that people may provide preemptively. No wonder you’ve missed them.
By “forced continuation” I mean the type of example that was popular for a while where someone would feed a lot of text into LLMs and then ask it to continue the text, then be surprised when it some times predicted the next tokens identical to the original example.
Do this enough times and you can find “examples” that prove your point, but it’s highly automatable so people were using it like a million monkeys at a million typewriters.
I know software patents, the worst idea in IP, do exist, but I think they're as toxically bad an idea now as they were decades before AI, when they were first invented.
You shouldn't be able to own an overall implementation idea. If you can, then Microsoft, Oracle, or Meta can too, and can start suing everyone for 'ripping off' an idea that they happened to come up with first. In that world, you'd have to consult a copyright-police AI (ironically) after every function you write, to scan it for ideas someone else has claimed. "Sorry, the function PadStringL() in this JavaScript is too similar in implementation to str_lpad(), from an AGPL library for perl that was released in 1998."
It probably is, if they didn't have the chops to write it internally. But to them, the trade-off is probably worth the benefit of ditching a component with AGPL.
Personally I just release everything I publish under MIT to make it as easy a as possibly to use.