Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

GitHub Copi­lot inves­ti­ga­tion

githubcopilotinvestigation.com

Post reply on HN
Print view View original

561–570 of 1001 posts

Re: GitHub Copi­lot inves­ti­ga­tion

#561

Post by BeefWellington »

mark_l_watson wrote: post #291

First, the author’s book Beautiful Racket is very cool, recommended. I largely disagree with this article, at least for MIT, BSD, etc. training code examples. The small autocompletions, even if they are several lines long, sort of seems like fair use to me. I do think that CoPilot should have an option to use a smaller model just trained in code that has very liberal use licenses, because I think the use of GPL, etc.…

MIT requires attribution, which copilot does not seem to include in the cases where it fully reproduces existing code.

Re: GitHub Copi­lot inves­ti­ga­tion

#562

Post by bergenty »

burkaman wrote: post #515
woah wrote: post #491

It would be sad if someone succeeded in shutting down CoPilot for this kind of copyright stuff. It is genuinely useful. I don't care that it reproduces copyrighted content. The only way you can get it to do that is to bait it with the function names of functions that have already been copy and pasted thousands of times onto GitHub without proper licenses. Luckily, someone will probably come out with a "renegade" vers…

I don't think most people are concerned that Copilot is going to be reproducing verbatim copyrighted code, it's more that it sucks that a giant corporation is going to make a billion dollars from a tool that is entirely built off of millions of peoples' work who were never asked permission and will never be compensated.

So what? Anti progressive luddites, it makes my blood boil.

Re: GitHub Copi­lot inves­ti­ga­tion

#563

Post by nonbirithm »

johnfn wrote: post #535
burkaman wrote: post #515

Earlier quoted context omitted.

I don't think most people are concerned that Copilot is going to be reproducing verbatim copyrighted code, it's more that it sucks that a giant corporation is going to make a billion dollars from a tool that is entirely built off of millions of peoples' work who were never asked permission and will never be compensated.

That's hardly a new thing! For instance, Google search makes billions of dollars by indexing content that other people make.

My guess is the act of production that is passed as original content tends to have more avenues to harm producers than consumption.

Alarm bells of this magnitude haven't been rung about people torrenting films for decades; it's a given that some people are just going to do it and there's little that can be done to stop it.

Producing new data from original data of questionable lineage makes the questionable acts visible. Copilot and the like actively encourage this creation.

If it were possible to peek into the rooms of everyone who downloaded a torrent to admonish them then maybe pirating would have been made a modicum more taboo. But those consumers never intended to leave their rooms. Copilot forces them to leave their rooms if they want their derivative work to be used.

Re: GitHub Copi­lot inves­ti­ga­tion

#564

Post by abigail95 »

endisneigh wrote: post #112

Copilot is trained on and returns AGPL code verbatim. It’s game over. If these licenses are not enforced it defeats the entire purpose.

That's a problem of the licensers, not for Microsoft or the CoPilot users.

If you released AGPL code but never intended to ever sue anyone. Why did you release it like that?

If you did and if someone is able to use your code without any damage to you, without reputation loss, and via a way they have access to the innocent infringer defense after you overcome fair use, after you sue them.

How is that game over?

Re: GitHub Copi­lot inves­ti­ga­tion

#565

Post by BeefWellington »

naikrovek wrote: post #451
gjkqfc wrote: post #400

Earlier quoted context omitted.

> but you all were already mad anyway ... This seems to be the line of argumentation agreed upon by several waffling pro-GitHub posters. Many comments have some variation on that diversion from the issue. > GitHub gets their own rights to your code. This is preposterous and false. GitHub has the right to display the entire work, properly attributed and licensed, to others. No new licenses are given, no dual-licensing…

> No new licenses are given, no dual-licensing takes place, no code-laundering is permitted. I suggest you read the terms of service again. here, I'll link directly to the license grant: https://docs.github.com/en/site-policy/github-terms/github-t...

I suggest you take your own advice:

    This license does not grant GitHub the right to sell Your Content. It also does not grant GitHub the right to otherwise distribute or use Your Content outside of our provision of the Service, except that as part of the right to archive Your Content, GitHub may permit our partners to store and archive Your Content in public repositories in connection with the GitHub Arctic Code Vault and GitHub Archive Program.

Re: GitHub Copi­lot inves­ti­ga­tion

#566

Post by bamboozled »

mdswanson wrote: post #178
stevage wrote: post #142

Earlier quoted context omitted.

It isn't really learning, if it's just regurgitating whole function bodies. I use Copilot a lot, and definitely see whole functions being spit out, that were presumably written by a person somewhere.

I also use Copilot a lot, and while it does suggest large function bodies, I'm not sure that it's "regurgitating" them (though it could be...I don't know). I suspect that it's seen so many function bodies that are similar that it generates another similar output. Like autocomplete in a word processor has seen so many similar chunks of text that it reproduces them based on past experience. I don't know this as a fact,…

The fact you don't know is the problem.

I can't use co-pilot because if I am stealing someone else's copyrighted code I'm in trouble from a legal standpoint.

Re: GitHub Copi­lot inves­ti­ga­tion

#567

Post by SantalBlush »

dmix wrote: post #520
jakelazaroff wrote: post #506

Earlier quoted context omitted.

> It would be sad if someone succeeded in shutting down CoPilot for this kind of copyright stuff. It is genuinely useful. I don't care that it reproduces copyrighted content. Sure — in the same way that hacking into a competitor's GitHub account and copying their private source code is "genuinely useful" to you. As the person benefitting from unlawfully using their source code, of course you wouldn't care that it rep…

> in the same way that hacking into a competitor's GitHub account That's like comparing grand-theft auto to someone stealing a pack of gum from a convenience store. It's not a useful analogy. The latter is still a problem, but we don't need to be FUDy about it. And OPs right, this will keep happening until we come up with better ways of solving this problem. Whether that's educating companies on the legal (and moral)…

>That's like comparing grand-theft auto to someone stealing a pack of gum from a convenience store.

They're both more like grand-theft auto, but one involves the valet driver leaving with your car, and the other involves smashing a window.

Re: GitHub Copi­lot inves­ti­ga­tion

#568

Post by kmeisthax »

If Copilot itself is infringing then so is GPT-3, DALL-E 2, NovelAI, and Stable Diffusion. There's no legal argument that would solely target one application of this technology, and you can't build generative AI using current ML tools without relying on a very large corpus of public data. All AI is built on free-riding[0].

While there is no US case law that explicitly says "training AI is fair use", the Second Circuit says that scanning books to make a search engine for them is. And the absolute worst interpretation of AI is that it's just a very well-compressed search engine index for its training set data[1]. So I'm not entirely sure if we can even thread the needle to only ban Copilot or AI training as a whole without also creating harmful precedent for search engines. Actual judges may try, I'm not sure if they'll succeed.

Internationally, the EU already legalized training AI on copyrighted works[2]. So if we do win against Copilot in court, all we've really done is shift AI research over to the EU where laws are already more favorable.

I fully agree that Microsoft is shoving too much liability onto their users, though. And this, again, also applies to all generative AI. My personal opinion with generative AI is that it's a nice curio, but not anywhere close to "production-ready", and Microsoft and OpenAI are trying to sell us on a lie that it's better than it really is.

[0] This also implies that all y'all playing around with image generators are just as much of a freeloader as Microsoft is.

[1] This viewpoint is also called "compressionism".

[2] This was part of the most recent EU Copyright Directive update - the one that added a de facto upload filtering requirement. It also added a copyright exception for museums and historical preservation.

Re: GitHub Copi­lot inves­ti­ga­tion

#569

Post by muzani »

HeavyStorm wrote: post #452
muzani wrote: post #432

Earlier quoted context omitted.

It says why in the linked post. People aren't doing open source for free; they do it for the community. But Copilot is there to extract value from it, giving nothing back, not even credit.

If that's the only issue, I can't see the difference when I search for something on the web, copy the code and paste into my solution. There's no attribution, there's no giving back, nothing. Because I'm the community that you are saying the code is supposed to benefit.

Yeah, that sounds like a good definition of someone who is not part of the community. I copy code off Stack Overflow too, but often provide attribution in a comment. But like piracy, it's easier to hunt the whales than the small offenders.

Stack Overflow facilitates the same thing too, so it's an interesting comparison, but SO makes attribution easy and clear, and it actually made it effortless to contribute back.

Re: GitHub Copi­lot inves­ti­ga­tion

#570

Post by Eisenstein »

SilasX wrote: post #546
schoen wrote: post #486

Here are a few thoughts I haven't formulated before: It seems clear enough to me that training AIs on copyrighted works is typically or commonly a fair use under existing law, because the AIs can and commonly do learn non-copyrightable elements and aspects of those works. It's very obvious from enormous numbers of examples that current AI systems are capable of learning much more abstract features of human culture (g…

>If human beings had much more accurate memories, copyright would be quite a bit more intrusive (and/or quite a bit less effective) because, following any exposure to some kinds of works, we could use our own memories to reproduce those entire works from scratch for our own use or pleasure without obtaining authorized copies from elsewhere. I don't know the name, but I remember some sci-fi story about some academy wh…

There is a short story by Orson Scott Card which has that theme called 'Unaccompanied Sonata'.

You can read it here: https://b-ok.cc/book/4395497/b2fb2e

Post reply on HN