What do people think the future looks like where publicly available resources on the Internet (art, code, etc) aren't fair use for training ML models? Where you have to opt into models or can opt out (and many wind up doing so)?
OpenAI, Microsoft, Google, et al will STILL train such models that can do all the same things, but it will be much harder for non-industry-backed individuals to navigate the legal minefield where you must ensure you properly attribute your model outputs, only train on opt-in data, etc, etc. Surely no one really thinks that a court case against Microsoft/OpenAI (even if they lose) would stop CoPilot?
Most of these complaints seem to be extremely emotional and cherry-picked. "People's legal rights are being violated!" (you definitely don't know that, no one knows that, the article is 100% right about that), "look I prompted CoPilot for this piece of code that I already knew about and it spit it right out" (that's not how it's going to be used in practice).
It seems to me that the longer-term implications of the outcome of a lawsuit like this are far more interesting, yet almost all the comments I see are nitpicking and whining about how the world isn't the way they want it to be. I wish the conversations around generative AI could be...just better.
An english description plus three characters of a function name is enough to coax CoPilot into distributing LGPL-licensed code out of context, without a proper license. That's neither "emotional" nor "cherry-picked", it's a clear-cut license violation.
I mean, he clearly knew about that code in advance and used his prior knowledge to coax Copilot into spitting it out, yeah? Three characters can get you pretty far, that's 1 combination out of 125,580 (considering all english letters, upper and lower, along with most of the numbers and symbols on my keyboard), plus the description of a fairly complex algorithm.
Also, this code is really just executing a mathematical operation in what I would assume is fairly standard, so it may not even fall under copyright. IANAL.
Even if it is a copyright violation, that is one out of, IDK, millions, maybe billions already of Copilot completions?
That sounds cherry-picked. Just because some high profile, highly-retweeted person says something doesn't make it not cherry picked.
Of course it is cherry picked. The idea is that it allows you to INTENTIONALLY void any copyright you want.
So let's say I obtain an illegal copy of microsoft windows' source code. Under this precedent, what stops me from just (overfitting) training a neural network to produce the source code verbatim, sans any license notice?
But it doesn't end there. What stops me from making a neural network that exactly reproduces the bytes of Illegally_Ripped_Disney_Movie.mp4 that I obtain from the pirate bay? Copyright need not apply.
At what point can the neural network I've described (which is intentionally designed to violate copyright) distinguishable from a neural network like Copilot and others which violate copyright extrinsically?
> The idea is that it allows you to INTENTIONALLY void any copyright you want.
It doesn't void copyright.
Anyone that uses code that Copilot spits out which infringes on someone else's copyright is still liable. There's no requirement for intent. That may be a mitigating factor in terms of remediation, but it cannot void the copyright itself.
It can, however, produce a plague of completely ignorant copyright infringement, and since the user of Copilot has no idea where the code is coming from, there's no way to check if it was trained on infringing code.
If I used Copilot I would be real worried about the legal implications for me and that I could easily be accused of copyright infringement or plagiarism[*]. Of course people seem to not really care these days if they can cheat to get ahead so this is probably a feature, and 99.9% of user won't have their reputations ruined by using it.
[*] Although I'm personally more worried about the fact that most of the code will be wikipedia/blogs-quality and filled with bugs, edge cases and performance issues.
Large companies don't worry about this because they already have tooling that scans code and matches it against other public code out there (to avoid legal troubles if some dev quietly copies some GPL'd code or something like that). Everybody else basically has to manually check every snippet produced, which makes all the supposed convenience moot.
Between that and the typical quality of Copilot snippets, it's pretty obvious who the real beneficiary of this technology is: large sweatshops like Infosys.
Buggy code can easily come from copying Stack Overflow or even from Github open-source repos so I doubt CoPilot would contribute to it anymore than it already is.
It doesn't void anything. If you use Copilot to copy some licensed code illegally, you are the person in breach, not the tool. People using Copilot are possibly littering their code-base with future copyright liabilities, and they'd have no idea about it. Until someone writes an AI to find infringing software and automatically sue them...
Yes, that's the other outcome, which is the industry getting GPL'd to death for using this. But I don't think there is an established precedent when it comes to this.
I don't know much about AI/SL but I think the logical outcome is that using a transformer to generate source code is going to result in an over-trained system that produces sections of code verbatim because of the relative size of the space of all valid programs vs the space of all possible programs. It's not like art where you get soft failure if a single pixel or word is wrong: the inclusion, exclusion, or replacement of a single instruction or symbol is enough to introduce fatal bugs into a computer program. If the system doesn't have the capacity to understand programs generally (which it probably won't if it's just a transformer), then you're going to end up with a system that spits out samples from training data that do work.
If you told the legal team at any mid-sized or larger company "we're pretty sure only 1 in 1000 lines of code our developers write breaches someone else's copyright" there'd be some serious hell to pay.
No, no. 1 out of every 1000 lines of code has the potential to breach some form of license.
If someone is motivated to search through our entire (proprietary, private) codebase. They match it with repositories that are freely available. They’re properly motivated to make a problem out of it (some twitter randos?), and most importantly they gain some benefit out of spending hundreds of thousands of dollars engaging with our legal team.
By the time you satisfy all the conditions required for it to be an issue you are talking nation-state actors.
Yes but if you have a large team you'll be using it hundreds of times a day. I will not be surprised if Copilot indemnity insurance is a thing in M&A in a five years.
Yep - it would be useful if more people had literacy of using the tool for these conversations. I don't blame them, that shouldn't be expected or required, but there is a large gap between how bad this looks and how materially bad it is when you take into account the actual way Copilot is usually used.
I don't think there will be a ruling like "anything from a neural net is yours", that'd be a bit ridiculous for very obvious reasons.
I'm no copyright law expert and I'm certainly not a lawyer, but it seems to me that in your examples you're setting out to violate copyright as a goal, which seems like it would be a factor in a court case.
To answer your last question, your examples are pretty clearly distinguishable from Copilot in their final states that you describe. IDK exactly *when* during overfitting that line is crossed, maybe it's crossed the moment you personally decide to knowingly publish copyrighted content and has nothing to do with the neural network itself?
This doesn't make sense, a byte identical copy of other work is obviously not transformed. So anyone claiming fair use relying heavily on transformation would fail.
But the only thing that does is make step 3 of fair use harder to clear. Not impossible.
There are fair uses of copyright that use the entire identical work as is.
> Even if it is a copyright violation, that is one out of, IDK, millions, maybe billions already of Copilot completions?
If you, only once, steal lines of code that you don't have license to do so and use them to make money, that's the same exact thing. "Trusting the algo" and saying "whoops I'm sorry" doesn't make a strong legal defense.
In a company of 1000 programmers, what are the odds that copilot increases the risk of using improperly licensed code because "well because microsoft gave it to us it has to be legit!"
The SCOTUS decision in Oracle v. Google didn't rule on API copyrightability. It merely assumed that the code in question is copyrightable, then showed how it's still fair use even if so, thus making the first question irrelevant to the decision. And this is very much intentional; they spell it all out:
"Google’s petition for certiorari poses two questions. The first asks whether Java’s API is copyrightable. It asks us to examine two of the statutory provisions just mentioned, one that permits copyrighting computer programs and the other that forbids copyrighting, e.g., “process[es],” “system[s],” and “method[s] of operation.” Google believes that the API’s declaring code and organization fall into these latter categories and are expressly excluded from copyright protection. The second question asks us to determine whether Google’s use of the API was a “fair use.” Google believes that it was.
A holding for Google on either question presented would dispense with Oracle’s copyright claims. Given the rapidly changing technological, economic, and business-related circumstances, we believe we should not answer more than is necessary to resolve the parties’ dispute. We shall assume, but purely for argument’s sake, that the entire Sun Java API falls within the definition of that which can be copyrighted. We shall ask instead whether Google’s use of part of that API was a “fair use.” Unlike the Federal Circuit, we conclude that it was."
The tragic irony here is that the understanding of copyright that those that do not like Copilot put forth would indeed make things like Java's API copyrightable and to the obvious detriment of innovation.
I don't see how this is related. The question wrt Copilot can be distilled down to "what constitutes a derived work", but there's no doubt that the original source code that Copilot was trained on is copyrightable. Conversely, with Java APIs, there was no doubt that Google's use of them produced a derived work - the question was whether the original is copyrightable and/or whether that is fair use.
> the SCOTUS did rule that it was copyright infringement.
In the US, "[T]he fair use of a copyrighted work ... is not an infringement of copyright." (17 USC 107, Oracle at 14). Whether there is a difference between a finding of no infringement or infringement with no liability is academic.
Practically, the copyright on the Java API is commercially worthless because after Oracle anyone may freely copy any and all of it and use it to compete with its creator.
Any other software vendor who thinks it has platform "lock in" because its customers built to their API should take notice. (e.g., Amazon AWS).
> That sounds cherry-picked. Just because some high profile, highly-retweeted person says something doesn't make it not cherry picked.
It's his code. This "high profile, highly-retweeted" crap is an appeal to emotion. He has a specific and legitimate interest in protecting his own intellectual property. It's not "cherry-picking" to report a crime being committed on your front lawn.
So why is he complaining about copilot and not the thousands of GitHub repositories redistributing his code with improper license? Typing the little code snippet he showed into copilot is analogous to typing it into the GH search bar and grabbing a properly-licensed result.
> It's not "cherry-picking" to report a crime being committed on your front lawn.
And as soon as he sees someone actually take the chair from his front lawn he can report it as a crime. He cannot report the people walking past because they could potentially steal his chair.
One could even argue that if he didn’t want his chair taken, maybe he should have locked it in his shed.
Of course, these chairs duplicate, so it’s not as if he loses his own chair.
Sure, duh, this discussion does not constitute a legal ruling. Nobody here is a lawyer nor a judge presiding over the case. We're potential subjects of a class action suit discussing grievances and merits of the case.
Microsoft Copilot is easily the greatest theft of intellectual property in the history of man.
You want to use my code, without ever knowing I wrote it? You want to use my hard work, regurgitated anonymously, stripped of all credit, stripped of all attribution, stripped of all identity and ancestry and citation? FUCK YOU
There's no need to defend something so obviously harmful, so why do you do it?
The law should be amended to make this kind of theft illegal.
It is a code laundering tool so corporations can steal open source code rights . It's the plainest thing I've ever seen. They charge for it on top of it. The deniers are out of their minds if they can't see what's coming.
This has the potential to severely damage open source - I would not host my open source project on GitHub, especially if it was copyleft. I'm sure many others wouldn't either. Some of these authors make amazing software that we might not see because of this.
Every artist, every creative individual, must EXPLICITLY OPT IN to having their hard work regurgitated anonymously by Copilot or Dall-E or whatever.
If you want to donate your code or your painting or your music, so it can be used ("written", "painted"), in whole or in part, by everyone else, without attribution, then go ahead and opt in.
Otherwise, you can't use the artist's or author's creative work for training.
All these code/art washing systems, that absorb and mix and regurgitate the hard work of creative people must be strictly opt in.
How is me building a novel application consisting of manually linked libraries and source code any different than building a novel application out of what Copilot generates? The difference is that Microsoft is pretending attribution and source licenses don't apply to the code it generates, even though it would in any other context.
The difference is that you could be unwittingly taking on liability in the copilot case. So that's fun. But also in the copilot case, microsoft distributed the code to you without a license, in violation of the license.
> Even if it is a copyright violation, that is one out of, IDK, millions, maybe billions already of Copilot completions?
Are you saying that because there are millions of copyright violations, Copilot is too big to fail? Or are you’d saying that Copilot is too big to be held accountable for flagrant violations of the law?
I guarantee Copilot’s developers knew it was spitting out verbatim code. It’s too obvious, and probably would result in a perfect rating for the prompt.
Yeah, degree generally matters. Airplane flights rarely kill people, so we allow them. If 50% of flights resulted in death, we would not. Google searches rarely illegally return copyrighted content, so it's allowed. The Pirate Bay searches often return copyrighted content, so regulators keep shutting it down.
Stealing a car is a large degree of crime for an individual. If they had stolen a penny, it would be a small degree of crime and we'd be more willing to let it slide.
I'm sure you see you've been downvoted. But I want to say I agree with you on the millions / billions bit.
While I understand the rub about licenses, the fact is the vast majority of code is not all that original or unique. Some fringe amount is, and those edge cases are worth discussing.
But the rest? Likely not all in all all that special. Yes, we get paid good money to do it. But is that a function of what it takes to do the work, or the demand for the skill (relative to supply of that skill)?
Frankly, I think some ppl just plain ol' fear Copilot. And either don't want ro admit it, or they have buried that fear. I'm not advocating ignoring the law / licenses. But putting a licence and lipstick on what is an everyday pig doesn't make that pig a unicorn. Does it?
This example is one of those rare pieces of code that is special though. It's the product of years of deliberate work by professor-level academics. This is exactly the kind of person who would have the least to fear from copilot if it really was just automating the boring plumbing parts and not shamelessly copying high-value, creative, insightful code.
But that's not the type to fear Copilot. Yes, they might object to the license violation. I get that. I acknowledge that. But when you're that intelligent and that creative you don't fear being replaced - displaced? - by something like Copilot. Nah. That's a fear for the mundane and the common. That's a fear for the rest of us.
Or that the professor has a legitimate concern, and a lot of people in the comments here don't have any code being stolen and are just afraid of Copilot.
It can be both things. (I'm not endorsing either view, just trying to clarify.)
Suppose you wanted to do what some code does, then you see this LGPL code. What can you do? Adjust variable names and play with line spacing and comments until it feels different?
First off, that's a library of pedagogical implementations, so I wouldn't even want to copy it -- I'd prefer a library focused on performance. Second, it's linear algebra, there are alternative implementations and libraries out there. Third, it's covered by the LGPL, so I'd be perfectly happy to link to the library. Fourth, I'd look up a pseudocode description and go from there. In no case would I sit down with another person's implementation and give it the undergrad treatment to pretend that I'm not copying.
"In computer programs, concerns for efficiency may limit the possible ways to achieve a particular function, making a particular expression necessary to achieving the idea. In this case, the expression is not protected by copyright."
So if it's fine to copy the pseudocode implementation of a non public use codebase, then you couldn't possibly object to recreating a codebase into a different language then, right?
But I don't care about the expression of the idea. I just want the idea to work. And I don't know how to do the idea myself. And I've seen how you've done it.
If I want to describe life with a nature metaphor, and then see you do it with a waterfall, I can probably get away with using a waterfall to the same metaphorical effect in my story.
It's a violation but this examples doesn't jive with how I use Copilot, for whatever that is worth. Usually Copilot takes into account context and integrates the various bits into a whole that fits in with the surrounding code. The "empty file, type function signature" use case feels somewhat dubious to try to understand actual, not theoretical, harm. Though I respect the right for this person to defend their rights.
It’s a contrived example. This is not how people use copilot in the real world. Copilot is designed to be used in context. The completions it provides in context are specialized to your project.
Artificially starving copilot of context and then showing that it recites parts of its training dataset is mundane.
From what I've seen on the art side of things, the more a certain work has been copied in the real world (and thus in the training set multiple times), the more likely it is you're able to get a very close copy out of the model with the right prompts.
For example, I'm pretty sure this is why some models turn up a near exact version of The Girl With The Pearl Earring.
Jumping into this but I've honestly lost my train of thought, lol
That said... is that any different from someone copying and pasting into their code vs copilot doing it?
If someone randomly pastes code that has a copyright, and people use it, how are they supposed to know they shouldn't be using it?
I imagine we're talking functions here though. Not sure if Copilot would reproduce entire libraries unprompted if they're not open-source, anyone have an answer?
Alternative theory: you're overconfident and wrong, and haven't even read the tweet I shared. CoPilot is observed to reproduce entire files (less two lines), down to variable names and comments, verbatim, from a repo that's covered by the LGPL.
If verbatim copies are not covered by copyright, then software can effectively never be copyrighted. And that there would be an extreme departure from "reality."
regarding your edit: this is a verbatim copy. The process you're describing is actually strengthening, not weakening, my argument. It allows material dissimilarity in the face of an overall similarity. Whatever process is meant to be applied to the licensed code and the allegedly infringing code, it will produce identical output given identical input. You're attempting to claim that the entire file would be deleted by the process and that the remainder would be an empty comparison which would (granted) be covered by fair use. And since CoPilot has been shown to redistribute multiple files, your "case" would hinge upon the entire repository not being covered by copyright. This is beyond absurd.
edit 2:
> Or don't! Just keep downvoting in support of your fantasies! Wheeeeee!
Please review the site guidelines. You're both whining about downvotes and sneering at the community with this. Probably time to take a break from the keyboard.
> In a computer program, the lowest level of abstraction, the concrete code of the program, is clearly expression, while the highest level of abstraction, the general function of the program, might be better classified as the idea behind the program.
Copilot is alleged to be reproducing blocks of code verbatim, which fall into the expression side of the idea/expression distinction, which by your own links and statements appears to show that the allegations against Copilot are not false.
I'm not going to say you're necessarily disagreeing with reality (I'm not really sure what that's supposed to mean) but you're certainly contradicting the evidence you've brought.
> In a computer program, the lowest level of abstraction, the concrete code of the program, is clearly expression, while the highest level of abstraction, the general function of the program, might be better classified as the idea behind the program.
Seems to be trying to force everything associated with programming into a false “expressive/abstract idea” divide. Abstract ideas are distinct from expression and not subject to copyright, but not everything outside of the scope of copyright is abstract rather than detailed: notably, functional elements.
No, it's an illustration of two ends of spectrum, and only one of three parts of AFC. All I'm saying is that the original comment gave us three links and that they didn't strictly agree with what the comment said without further elaboration.
I will say, though, that the allegations of basically copying and pasting exact instances of substantive portions of copyrighted codebase appear to be at one end of the spectrum.
Things like code comments are clearly expressive and not utilitarian (you don’t need comments to compile code and you can express the ideas of these comments with different verbiage without loss of efficiency).
For what it's worth I've found the links you've provided very interesting and insightful! It's bringing back bits and pieces of some of the software license training I've got at various jobs in the past.
> What do people think the future looks like where publicly available resources on the Internet (art, code, etc) aren't fair use for training ML models?
A better future to me. I don't want pictures of my face training ML models, nor do I want my art, or my code. I don't want my face to be more recognizable to AI, and I don't want my work to contribute to the consolidation of power to a few big firms. And for what, what can ML models even bring me besides surveillance? Cool art? Text-to-speech?
If they hire photographers to take photos of people in public and use them for training, there’s no law stopping them really. Your only real hope would be to always walk around in a burqa.
IDK how generative models can really be used for surveillance?
Certainly facial recognition models etc, can be, but those would seem to be appropriately covered by the Google Books ruling dealing with discriminitive models: https://en.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google,.... They've also been a thing for quite a while, so I think that cat escaped the bag a long time ago.
Remember clip art in Microsoft Word back in the day? Now there is an infinite supply of that. Stock images? Infinite supply. Solo filmmakers are going to have a much easier time creating their own films that can rival the best movie studios in the world. Any text anywhere will be read to you in any voice or voices you like, with tone and setting appropriate sound-effects. Smaller things will just be better too, noise cancellation on microphones? Easy and free. Image editing? Trivial to remove, relight, reposition, etc, etc, etc.
So many other things too. It's going to be magnificent. If you're not into then I guess to each their own, but I do think we are looking at something that can be a net good for everyone in the world, so long as it's available and cheap for everyone in the world.
Such a service would make a great and fantastic service for stalkers to bypass the usual difficulties in locating someone who has done the best to excise them from their life.
What if a service could scan surveillance videos and show everywhere you have been ever?
Who here is doing a startup to secure licensing rights to every companies surveillance camera videos to make the AI/Surveillance version of Equifax Worknumber? Maybe you offer to give them the surveillance system for free in return for the rights?
>OpenAI, Microsoft, Google, et al will STILL train such models that can do all the same things, but it will be much harder for non-industry-backed individuals to navigate the legal minefield where you must ensure you properly attribute your model outputs, only train on opt-in data, etc, etc. Surely no one really thinks that a court case against Microsoft/OpenAI (even if they lose) would stop CoPilot?
I don't fucking care, I'm not in the business of competing with OpenAI or whatever. If you want to launch and AI startup but you can't that's your fucking problem, not mine. I just don't want them violating the licenses of the open-source programs I have created.
> "look I prompted CoPilot for this piece of code that I already knew about and it spit it right out" (that's not how it's going to be used in practice).
It proves that copilot has the capacity to copy existing code without fulfilling the requirements of the license. I don't care if it's "cherrypicked", this shouldn't happen under any circumstances.
> I wish the conversations around generative AI could be...just better.
I wish that these people making all these complicated language-comprehension machine-learning systems could read the fucking license statement at the top of the file and copy that license statement along with the code. this ought to be a solvable problem. I'm pretty sure i could write a bash script that does it if M$ is looking to hire.
The removal of the license where the code was learned is the real tidbit that everyone needs to focus on. This is where the laundering comments come from.
The product would be useless if it prompted you with license approvals. They didn't care and removed them. They consciously decided to prioritize their paid-for product over the rights of their users. I'm amazed that MS's lawyers allowed it out the door. That's the even scarier part.
> What do people think the future looks like where publicly available resources on the Internet (art, code, etc) aren't fair use for training ML models? Where you have to opt into models or can opt out (and many wind up doing so)?
A future where technology is developed in accordance with longstanding law? Also, maybe a future where my copyrighted works are compensated for when they're being used to automate my job away? If the music industry can deal with royalties, maybe software can, too?
> OpenAI, Microsoft, Google, et al will STILL train such models that can do all the same things, but it will be much harder for non-industry-backed individuals to navigate the legal minefield where you must ensure you properly attribute your model outputs, only train on opt-in data, etc, etc. Surely no one really thinks that a court case against Microsoft/OpenAI (even if they lose) would stop CoPilot?
I'd expect injunctions against Microsoft/OpenAI from further training CoPilot with inappropriately-licensed code. I'd expect damages for all of the instances of copyrighted material that CoPilot regurgitates.
> Most of these complaints seem to be extremely emotional and cherry-picked. "People's legal rights are being violated!" (you definitely don't know that, no one knows that, the article is 100% right about that), "look I prompted CoPilot for this piece of code that I already knew about and it spit it right out" (that's not how it's going to be used in practice).
How are these emotional? They're opinions, like all legal claims, supported by facts. It doesn't really matter if they're "cherry-picked" or not, only whether CoPilot actually violated copyright. If it gives seemingly novel results 999 times out of 1000, but in the other case verbatim generates copyrighted material without proper permission, then that is a copyright violation. At scale, that's a copyright violation with some modest damages even.
1. I think if you don't want your code re-used in CoPilot you should have that right
2. I think if CoPilot gets smart enough that it can read your open source code and then reproduce the algorithms without copying your code that should be fair use. It's the same thing a human would do. AFAIK CoPilot can not do that but I can certainly imagine it's not too many years away from that.
3. I think I would opt into sharing all my open source code mostly unrestricted with services like CoPilot. I think the group of people that choose to share their code with AI will do better over all than those that lock their code behind licenses.
Note that I'm referring to snippets of code. I don't know what a good definition of snippet is. In other words, if AI helps me write chunks of 10-100 lines at a time I don't see a problem. Effectively, S.O. answer level of snippets. Whereas, if I tell AI "create LibreOffice" and it clones the millions of lines of code, I think that is a problem. I don't know where the cut off is.
What you describe, is what co-pilot already does most of the time.
The examples where people can show it reproducing snippits of code are more the exception than the rule. And they are usually done by people who are trying to manuliption into proving that it can reproduce copyrighted code.
Some people tend to think of it as a search engine. Looking though it's database for relevant snippets for the current situation and regurgitating them unmodified.
But that's really not what it's doing. It's more like the AI autocomplete on your phone, but for code.
It might not be able to understand the algorithms. But it seems to be able to adapt simple algorithms that it's seen multiple times in it's training data to match the surrounding code (in style, naming conventions, and actually using the variable/functions you already have).
I don't have number, but from my experience, I would say it generates uniqu(ish) non-copyrighted code at least 95% of the time.
The only question is what to do about the other times when it does occasionally output potentially copyright infringing code, either by accident, or when it's forced.
Any penalty less than the profit, isn't an effective penalty, and won't act as a deterrent.
The Securities and Exchange Commission (USA) has a history of giving million-dollar fines for crimes that produced billions in profit and/or took billions away from victims. And the lack of deterrence has been reflected in the actions of the US financial industry.
Its not popular to say it, but I agree to some extent here too. We may need a wholesale reimagining of copyright/patents in many places to accept the new reality of both of building the tools (data to train) and in accepting the occasional bad output (copyright/patented function appears in output). I think watching the law evolve with the tech is going to have a lot of ups and downs.
I think you hit the nail on the head. Our laws and rules were created for a cultural context that is quickly becoming outdated. I feel there are many valid criticisms of AI today, but demonizing the technology doesn't allow for fruitful discussions. We need to evolve our thinking and we need to be open minded to do so first.
If they continue that path, the future will be that OpenAI, Microsoft, Google etc. will pay larger and larger fines at least in the EU, until they are blocked entirely.
Earlier HN thread today on a large chunk of OS code pasted almost verbatim by the CoPilot engine into a project, but stripped of any licensing references.
Within the last few days, another thread on artists who have spent decades developing a unique and valuable style are making parallel complaints about Dall-E/SD/etc., where inputting "Xyz in the style of [Artist]" produces exactly a copy of [Artist]'s unique style, barely distinguishable from the original.
These engines are fairly literally giant collage engines, able to parse language inputs and output a collage of the input works. Maybe some are small snippets so it could be fair use, but they are also evidently capable of outputs of a far larger scope, amounting to wholesale ripoff.
Opting out or not posting on Github or whatever prevents nothing, as stuff is posted everywhere by many, and with code, it's totally legit posting a fork under OS licensing.
Is there a solution analogous to a <NoRobots> flag? How do we verify it? Will there soon be HaveIBeenUsedAsTraining adversarial systems to probe these output engines?
Not sure of the solution, but this seems to rather rapidly overstepping boundaries of creators.
Huh, I wonder if they decided to remove the licenses and other comments from code before training on it. That would almost be necessary to avoid comments ending up inside of code.
> "People's legal rights are being violated!" (you definitely don't know that, no one knows that, the article is 100% right about that)
I'm not sure what the argument being made here is. If you make something opaque enough that no one can tell if it's violating legal rights, no one is allowed to say anything about it? This seems uncomfortably close to "it's only a crime if you get caught"
> look I prompted CoPilot for this piece of code that I already knew about and it spit it right out" (that's not how it's going to be used in practice).
So what? I’m not being snarky: does that actually make any difference, legally?
Well, what does the future where those materials are free use look like?
You argue that desiring ownership of works that you created is “extremely emotional and cherry-picked”, but you do not provide a compelling argument why artists, photographers, and indeed programmers should be excluded from the conversation when it is their art, photography and code that is being appropriated in the first place.
I think that future has a lot of really cool and cheap tools that will help all those artists, photographers, programmers, etc get even more out of what they love doing. I think there will wind up being some job markets that shrink (not with the tech we have now though) for small-medium things in all of these fields (think logo design, stock photos, client libraries, simple out-of-the-box applications) and my hope is that those job markets that shrink cause others to grow due to the increased levels of productivity that these tools will give us.
Ultimately my argument is that these tools will allow human beings to accomplish more things with less and that these tools should be distributed to as many people as possible for as little cost as possible. Part of that belief comes from the fact that I think these tools are coming no matter what and I'm slightly concerned about the potential (although unlikely-looking) future where a small number of large corporations are the only ones controlling these tools and they just rent-seek on them.
The job market for cheap rehashed garbage will grow and quality productions will suffer.
For programmers, the job market for loud-mouthed posers and plagiarizers will grow and quality will suffer. But those programmers will be fluent in marketing speak.
In a world where the cost of producing cheap rehashed garbage approaches 0 why would the job market for cheap rehashed garbage grow at the expense of the expensive unique gems market?
There will definitely be more cheap rehashed garbage online and we will be forced to invent tools to wade through it. I actually look at that as a bright side because there's already a lot of cheap rehashed garbage, we just don't have good tools for wading through it yet because it hasn't become completely intolerable yet.
It's a bit of a bait and switch (obviously not literally, these things didn't exist so nobody was ever promised they wouldn't be used).
But in terms of user behavior, it's rather the same. I used to make more stuff publicly available online than I do now, and the mass-scale surveillance and data modeling that big companies do off of publicly available stuff is a big part of that.
Generally that's how you get walled gardens - by abusing the commons - but here you'd need not just a walled garden but a TINY TINY invitation only one if you don't want people doing mass surveillance and data modeling (CoPilot is really more of the latter than the former, but any of this "scrape the whole internet" stuff is just a tiny little sidestep away from being used for more blatantly evil surveillance purposes - here we're training a generative model, they're we're de-annonymizing everything you've written anywhere...).
Is there a good solution to "BigCos are gonna do whatever they want with the shit you make" other than invite-only, paid-content type models?
I think I agree with this comment[1] from the other thread; never previously thought that a process being transformative means input and output datatypes do not coincide, but maybe that is it.
That's an interesting take, the whole level of indirection thing between Microsoft-OpenAI and StabilityAI and that research group is certainly another dimension to this that sort of muddies the waters.
I personally expect the law to end up with a "safe harbor"-like situation. Consider YouTube. Occasional copyrighted content does not make YouTube illegal, or able to be held liable for slip-ups. See the DMCA as well, which requires takedowns upon notice, but otherwise provides near total indemnity for user-generated content.
Because of this, if the law looks at GitHub Copilot, I would expect that they would find Copilot to be A-OK despite the occasional regurgitation of copyrighted material that isn't fair-use, as long as it is removed upon request.
There are plenty of regulations that only apply to companies with more than X employees, etc. What makes you think any way to improve the law would necessarily harm individuals?
> It creates a body of knowledge, everyone can use and can't be sued for since it would be the industry standard way to do things.
That's just not true. If the "industry standard way to do things" is to violate other peoples' copyright, then everyone doing that absolutely can be sued.
And while it's not clear if using these AI tools constitutes copyright infringement, it looks to me like there's at least a very strong case that could be made.
And at up to $10,000 per copy (register your code with the copyright office if you care about this issue!), that starts to add up very quickly. Even for a company like Microsoft.
There does seem to be more heat and noise then substantive discussions here.
Though that doesn't justify such a dismissive attitude towards ordinary HN commenters. As the way it's written implies that most are too stupid and overly emotional, which is more likely to fuel complaints instead of dousing them.
The WAY it was done with copilot is the problem: no attribution, just shoving all legal liability off on the end “programmer” without providing the attribution required TO COMPLY WITH LICENSES as the diligent programmer tries to clear all the code copilot handed it without meta data.
Go read the article before arguing further, please.
Otherwise you are wasting all of our time.
I did read the article first. Start-to-finish. As others have pointed out, it's a very visually appealing website.
I hope that when a case on generative models hits the courts that it's found that training on data from the Internet counts as fair use. I hope that for the reasons I laid out in my comment, because I think that if it isn't then we are all in trouble since these tools will STILL EXIST, but they will be in the hands of the few instead of the many. My main reaction is to how short-sighted it seems like the authors and many others are being about this technology in general. They seem to think they can just wish it away.
I also think that training on data from the Internet is fair-use, but I'm not a lawyer and I haven't studied the law extensively, so who cares what I think about that.
OpenAI, Microsoft, Google, et al will STILL train such models that can do all the same things, but it will be much harder for non-industry-backed individuals to navigate the legal minefield where you must ensure you properly attribute your model outputs, only train on opt-in data, etc, etc. Surely no one really thinks that a court case against Microsoft/OpenAI (even if they lose) would stop CoPilot?
Most of these complaints seem to be extremely emotional and cherry-picked. "People's legal rights are being violated!" (you definitely don't know that, no one knows that, the article is 100% right about that), "look I prompted CoPilot for this piece of code that I already knew about and it spit it right out" (that's not how it's going to be used in practice).
It seems to me that the longer-term implications of the outcome of a lawsuit like this are far more interesting, yet almost all the comments I see are nitpicking and whining about how the world isn't the way they want it to be. I wish the conversations around generative AI could be...just better.