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

Re: All public GitHub code was used in training Copilot

#451

Post by sp332 » Thu, Jul 08, 2021, 7:04 PM UTC

Hamuko wrote: post #183

ssivark wrote: post #64

Calling it “public” code feels like doublespeak. It’s most definitely NOT public domain code — it only happens to be hosted on GitHub and browsable ( but not copyable ) by people. “Source available for viewing” is very different from “public property” as the phrase is commonly understood: https://en.m.wikipedia.org/wiki/Public_property

> it only happens to be hosted on GitHub and browsable (but not copyable) by people. So would you say that it's publicly visible?

Publicly visible, yes. Publicly available, yes. Public code, no.

Re: All public GitHub code was used in training Copilot

#452

Post by sp332 » Thu, Jul 08, 2021, 7:06 PM UTC

habitue wrote: post #353

I am really confused by HN's response to copilot. It seems like before the twitter thread on it went viral, the only people who cared about programmers copying (verbatim!) short snippets of code like this would be lawyers and executives. Suddenly everyone is coming out of the woodworks as copyright maximalists? I know HN loves a good "well actually" and Microsoft is always suspect, but let's leave the idea of code la…

You don't have to be a copyright maximalist to worry about a company taking snippets of code that used to be under an open license and using them in a closed-source app.

Re: All public GitHub code was used in training Copilot

#453

Post by sp332 » Thu, Jul 08, 2021, 7:07 PM UTC

e12e wrote: post #381

akerro wrote: post #340

Earlier quoted context omitted.

Just found Intel leaks and Gwent on github without any effort. Intel has a few repositories in different formats, plain copy of .svn directory or converted to git. TF2/Portal leak is there as well. All but 2 I found were made by throwaway accounts.

Now, was the leaked nt kernel source ever published on github?

https://github.com/PubDom/Windows-Server-2003 Second result on DDG.

Re: All public GitHub code was used in training Copilot

#454

Post by taywrobel » Thu, Jul 08, 2021, 7:13 PM UTC

yccs27 wrote: post #328

taywrobel wrote: post #313

Earlier quoted context omitted.

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 wor…

The big question here is: On what data was the model trained? Presumably the news stations trained theirs on public-domain works and their own backlog of news articles, so even with manual copying there would be no infringement. In contrast, Copilot was trained on other people's code with active copyright.

That’s quite a big presumption IMO. Training sets need to be quite large in order to produce reasonable output. My understanding is that these companies provide the model themselves, which seems like it’d be trained on more than one company’s publications. But I get your point, and understand both sides of the argument here.

I think this will end up with a large class action lawsuit for sure, tho I really think it’s a toss up as to who would win it. This conversation was bound to happen eventually and we’re in uncharted territory here.

I think it’s going to hinge on whether machine learning is considered equivalent in abstraction to human learning, which will be quite an interesting legal, technological, and philosophical precedent to set if it goes that way.

Re: All public GitHub code was used in training Copilot

#455

Post by ksec » Thu, Jul 08, 2021, 7:17 PM UTC

All I know is this Copilot opens a whole can of worms. And doesn't or possibly never will have a right answer without court settling.

Obviously most ( I think ) lawyers seems to be siding with Microsoft on fair use. But most owner of the code seems to think they are infringing on their work.

Then there is the international issue because one court cant decide for everyone else.

I think the issue is important enough I wonder if we could somehow crowdfund it for a court trial or something.

Re: All public GitHub code was used in training Copilot

#456

Post by jeremyjh » Thu, Jul 08, 2021, 7:19 PM UTC

ghoward wrote: post #371

jeremyjh wrote: post #321

Earlier quoted context omitted.

What if I made a few tweaks to Copilot so that it is very likely to reproduce large chunks of verbatim code that I would like to use without attribution, such as the Linux kernel. Do you really think you can write a computer program that magically "launders" IP? A compiler is run on original sources. I don't see any analogy here at all.

* They both process source code as input. * They both produce software as output. * They both transform their input. * They both can combine different works to create a derivative work of each work. (Compilers do this with optimizations, especially inlining with link-time optimization.) They really do the same things, and yet, we say that the output of compilers is still under the license that the source code had. Wh…

> Why not Copilot?

Because the sources used for input do not belong to the person operating the tool.

If you say that doesn't matter, then you are saying open source licenses don't matter because the same thing applies - I could just run a tool (compiler) on someone else's code, and ignore the terms of their license when I redistribute the binary.

Re: All public GitHub code was used in training Copilot

#457

Post by neonihil » Thu, Jul 08, 2021, 7:24 PM UTC

I think it’s pretty easy to defeat MS in court.

We just need to bring the music industry into this!

For example: Let’s train a network on Beatles music to generate new Beatles songs. I’m pretty sure music lawyers will find a way to prove that the trained network is violating the label’s copyright, as they always manage to do that.

And then we just need to use the precedent and argue that music is the same thing as code.

Re: All public GitHub code was used in training Copilot

#458

Post by TaylorAlexander » Thu, Jul 08, 2021, 7:27 PM UTC

shagie wrote: post #392

mysterydip wrote: post #187

Earlier quoted context omitted.

Makes me wonder what would happen if a similar thing was done with books. If I train an AI on all the texts of Tom Clancy, or Stephen King, or every Star Wars novel, and the books it generates every so often produce paragraphs verbatim from one of those sources, would copyright owners be up in arms? What would the distinction be between the code case and the text case?

I am not a lawyer. I do photography and have a more than passing interest in copyright as it applies to the photographs I take and the material I photograph. Copyright on art gets more interesting / fuzzier. The key part is substantial similarity - https://en.wikipedia.org/wiki/Substantial_similarity and https://www.photoattorney.com/copyright-infringement-for-sub... Rather than text, my AI copyright hypothetical...…

What would be interesting in that case would be how the transformed image would look if photos from that location were removed from the training set. That would help reveal whether it was just copying what it had seen or it actually remembered what sunsets looked like and transformed the image using its memory of sunsets in general.

Re: All public GitHub code was used in training Copilot

#459

Post by gradys » Thu, Jul 08, 2021, 7:28 PM UTC

adamdusty wrote: post #445

habitue wrote: post #353

I am really confused by HN's response to copilot. It seems like before the twitter thread on it went viral, the only people who cared about programmers copying (verbatim!) short snippets of code like this would be lawyers and executives. Suddenly everyone is coming out of the woodworks as copyright maximalists? I know HN loves a good "well actually" and Microsoft is always suspect, but let's leave the idea of code la…

I'm more surprised that people don't care about the telemetry aspect. It's an extension that sends your code to an MS service, and MS promises access is on a need-to-know basis. I don't care if MS copies my hobby projects exactly, but I'm not sure my employer(defense contractor) would even be allowed to use a tool like this. I think it looks cool though. I will probably try it out if it is ever available for free and…

It's quite possible to do this on-prem and even on-device. TabNine, a very similar system with a smaller model (based on GPT-2 rather than 3), has existed for years and works on-device.

Re: All public GitHub code was used in training Copilot

#460

Post by jefftk » Thu, Jul 08, 2021, 7:29 PM UTC

laurowyn wrote: post #430

jefftk wrote: post #295

Earlier quoted context omitted.

GitHub used code that wasn't under any license at all, just publicly visible. Their claim is not that the license allows what they're doing, but that they do not need a license.

which is a different issue to my point, but still very valid. what terms are implied if no license is specified? I would argue attribution should be expected if used, but I also wouldn't go near any code without a specific license attached as there's no express permission given - just because a license isn't disclosed doesn't mean it isn't there. you can't go copying anything and everything just because nobody has to…

It's stronger than that: if GitHub is correct that they don't need a license then they are allowed to train on publicly visible code even if it is labeled with "no one has any provision to use this for anything at all, especially training models"