All public GitHub code was used in training Copilot • Hacker News Board

Re: All public GitHub code was used in training Copilot

#311

Post by ssivark » Thu, Jul 08, 2021, 4:41 PM UTC

qayxc wrote: post #302

hellcow wrote: post #239

Earlier quoted context omitted.

Not when it outputs large sections of unique code verbatim, as it's been shown to do.

If it's large sections, that can be fixed by either licence attribution or result filtering. That's at best a technical issue. What way too many people claim, however, is that the machine isn't even allowed to look at GPL'ed code for some reason, while humans are. I'd like to learn the reasoning behind that.

I think result-filtering (based on license of search results) is gnarly enough, and likely computationally intensive, so as to break the whole feature. But it would be interesting to see if that can be crafted to fix the shortcomings of the ML model.

Re: All public GitHub code was used in training Copilot

#312

Post by tshaddox » Thu, Jul 08, 2021, 4:42 PM UTC

_Understated_ wrote: post #218

Hamuko wrote: post #208

Earlier quoted context omitted.

> If I take your hard work that you clearly marked with a GPL license and then make money from it, not quite directly, but very closely, how is that fair use? Or legal? You can wipe your ass with the GPL license if your use of the product falls within Fair Use. You can actually take snippets from commercial movies and post them onto YouTube if your YouTube video is transformative enough for your usage to be considere…

> You can wipe your ass with the GPL license if your use of the product falls within Fair Use. Is Copilot fair use? It's reading code, generating other code (some verbatim) and making money from it all while not having to release its source code to the world? > That doesn't mean that GitHub has to redistribute Copilot under GPL I wasn't saying that was the case: some of the code that Copilot used may not allow redist…

> Is Copilot fair use? It's reading code, generating other code (some verbatim) and making money from it all while not having to release its source code to the world?

Again, fair use is an exception to copyright protection. If something is fair use, the license does not apply. The fact that Copilot does not release its source code is related only to a specific term of a specific license, which does not apply if Copilot is indeed fair use.

Re: All public GitHub code was used in training Copilot

#313

Post by taywrobel » Thu, Jul 08, 2021, 4:43 PM UTC

fchu wrote: post #307

If a company built a tool like Copilot to help students write essays, is that considered plagiarism? Probably yes, and the reason is that regurgitating blobs of text without actually thinking like a human and writing them anew doesn't feel like actual work, just direct re-use. Same thinking probably applies to GitHub Copilot and copyright

It’s already fairly commonplace for news agencies to generate articles using ML solutions such as https://ai-writer.com/

So by your logic ABC, CBS, Fox, and NBC have all been plagiarizing and violating copyright for doing so? I’m not sure if there’s been a legal challenge/precedent set in that case yet, but that seems like a more apples to apples comparison than the Google Books metaphor being used.

Disclosure: I work at GitHub but am not involved in CoPilot

Re: All public GitHub code was used in training Copilot

#314

Post by SCLeo » Thu, Jul 08, 2021, 4:45 PM UTC

mikehearn wrote: post #66

ML novice question: is this atypical when training models? Wasn't GPT-3 trained on a lot of copyrighted data? My gut instinct, which is based on very low-information, is that it would be pretty hard to train models if you could only use open-licensed material.

I am not a lawyer but I do believe GPT-3 as a commercial product trained using copyrighted data constitutes infringement. I also think GPT-2 does not because it is for research purposes, which made it fair use.

Re: All public GitHub code was used in training Copilot

#315

Post by qayxc » Thu, Jul 08, 2021, 4:46 PM UTC

skinkestek wrote: post #56

Yoric wrote: post #13

Out of curiosity, how do we define license violation in that case? I, as a human being, have trained by reading code, much of which is covered by licenses that are somehow not compatible with code I'm writing. Am I violating licenses? Asking seriously. It's really unclear to me where law and/or ethics put the boundaries. Also, I'd guess it's probably country dependent.

> I, as a human being, have trained by reading code, much of which is covered by licenses that are somehow not compatible with code I'm writing. Am I violating licenses? As someone who has taught students in ICT a quick rule of thumb was that I picked a piece of text that I suspected, wrapped it in doublequotes and put it into a search engine. 9/10 times - possibly more - of the times I had that feeling it was true.…

> But in this case it is obvious that the AI isn't writing the code - at least not all the time, it is instead choosing what to copy - verbatim.

I still don't see any problem with that. If it's larger sections (e.g. entire NON-TRIVIAL function bodies), those can be filtered or correctly attributed after inference. So that's just a technicality.

Smaller snippets and trivial or mechanical implementations (generated code, API calls, API access patterns) aren't subject to any kind of protection anyway.

  int main(int argc, char* argv[]) {

Lines like that hold no intellectual value and can be found in GPL'ed code. It can be argued that that's a verbatim reproduction, yet it's not a violation of any kind in any reasonable context.

Where do you draw the line and how would you be able to - automatically even! - decide what does and does not represent a significant verbatim reproduction?

Re: All public GitHub code was used in training Copilot

#316

Post by jrm4 » Thu, Jul 08, 2021, 4:47 PM UTC

extra88 wrote: post #296

jrm4 wrote: post #261

Earlier quoted context omitted.

You're strongly and incorrectly implying that "Fair Use" is a clear (and relatively immutable) concept within copyright law, which couldn't be further from the truth. Even if this or that particular case sets out what appears to be solid grounds, one shouldn't take that as gospel by any means. This mostly has to do with the nature of the wishy-washy nature of the 4 part Fair Use test, which, unlike decent legal tests…

> Is it Novel? Is it Non-Obvious? Those questions for patents are barely more clear-cut than copyright fair use tests, there is lots of room for disagreement. It's definitely true that a fair use defense against copyright infringement varies a lot by the field of work and norms can develop which are relevant to court cases. The music field is a mess, the "Blurred Lines" judgement was total bullshit. But the software…

But there's no reason not to either - I suppose my point is, don't take too much as gospel and think about everybody's best "end-goals" and push or pull with or against the law as needed.

Re: All public GitHub code was used in training Copilot

#317

Post by kroltan » Thu, Jul 08, 2021, 4:48 PM UTC

PieUser wrote: post #241

_Understated_ wrote: post #223

Earlier quoted context omitted.

> Google did not scan those books and use it to build new books with different titles. The comparison doesn't hold up at all. Not sure if you meant to reply to me but I agree with you: you can't compare what Google did to what Copilot does.

Copilot just suggests code.

And someone accepts it. Even if suggesting derivatives of licensed code is not a license infringement, then Copilot sure is a vector for mass license infringement by the people clicking "Accept suggestion". And those people are unable to know (without doing extensive investigation that completely nullifies the point of the tool) whether that suggestion is potentially a verbatim copy of some existing work in an incompatible license.

Re: All public GitHub code was used in training Copilot

#318

Post by ralph84 » Thu, Jul 08, 2021, 4:49 PM UTC

> I've reached out to @fsf and @EFF's legal teams regarding this. Please also reach out if you would be interested in participating in a class action.

I think she's barking up the wrong tree here. If she's looking for organizations interested in eliminating fair use, RIAA, MPA, and AAP are more likely allies.

Re: All public GitHub code was used in training Copilot

#319

Post by jrm4 » Thu, Jul 08, 2021, 4:49 PM UTC

MAGZine wrote: post #298

jrm4 wrote: post #261

Earlier quoted context omitted.

You're strongly and incorrectly implying that "Fair Use" is a clear (and relatively immutable) concept within copyright law, which couldn't be further from the truth. Even if this or that particular case sets out what appears to be solid grounds, one shouldn't take that as gospel by any means. This mostly has to do with the nature of the wishy-washy nature of the 4 part Fair Use test, which, unlike decent legal tests…

I think it would be pretty easy to stake opinions on those "boolean questions." Is (was?) a swipe gesture novel? Is it non-obvious?

Oh, absolutely. Kind of furthers my point. Patent is a silly mess in a lot of ways, but at least there's something like Booleans in it. "Fair use" doesn't even have THAT.

Re: All public GitHub code was used in training Copilot

#320

Post by Luker88 » Thu, Jul 08, 2021, 4:52 PM UTC

So...

Putting the (imho) big licensing problems aside, what about the software patents?

Apache and GPL have patent protection clauses.

Does this mean that anyone using copilot might somehow get code that implements something patented, but protected by license, except they did not get proper permission through the Apache/GPL license?

...I kind of hate myself for saying this, but... Patent trolls to the rescue?