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

Re: All public GitHub code was used in training Copilot

#381

Post by e12e » Thu, Jul 08, 2021, 5:43 PM UTC

akerro wrote: post #340

sailingparrot wrote: post #304

Earlier quoted context omitted.

No, it's not trained on all public code as the title suggests, it's trained on all GitHub public code (so public repos hosted on GH), none of the things you enumerate are hosted on GH.

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?

Re: All public GitHub code was used in training Copilot

#382

Post by codehawke » Thu, Jul 08, 2021, 5:43 PM UTC

Most likely they are using the private stuff too.

Re: All public GitHub code was used in training Copilot

#383

Post by AaronFriel » Thu, Jul 08, 2021, 5:45 PM UTC

shakna wrote: post #370

AaronFriel wrote: post #361

Earlier quoted context omitted.

A $0 settlement means there is no binding precedent and signals to me that Oracle's attorneys felt they didn't have a strong argument and a potential for more. If they felt the nine line function made Google's entire library an unlicensed derivative work, they would have pressed their case.

> A $0 settlement means there is no binding precedent and signals to me that Oracle's attorneys felt they didn't have a strong argument and a potential for more. That's not the case. It wasn't an out-of-court-settlement, but an agreement about the damages being sought, the court had already found it to be infringing, and that was part of the ruling. But none of that changes that 9-lines is substantial enough to be in…

I think we agree then, and appreciate the correction on the lower court settlement.

If Copilot is infringing, I suspect it's correctable (by GitHub) by adding a bloom filter or something like it to filter out verbatim snippets of GPL or other copyleft code. (And this actually sounds like something corporate users would want even if it was entirely fair use because of their intense aversion to the GPL, anyhow.)

Re: All public GitHub code was used in training Copilot

#384

Post by rapind » Thu, Jul 08, 2021, 5:47 PM UTC

zoomablemind wrote: post #348

rapind wrote: post #329

Earlier quoted context omitted.

This is pretty interesting for AI in general. Should you be able to train with material you don't own? Can your training benefit from material that has specific usage licenses attached to it? What about stuff like GameGAN?

> ...Should you be able to train with material you don't own? If relating this to how humans learn, books and other sources are used to inform understanding and human knowledge. One can purchase or borrow a book without actually owning the copyright to it. Indeed, a given passage may be later quoted verbatim, provided it is accompanied with a reference to its source. Otherwise, a verbatim use without attribution in a…

"If relating this to how humans learn" seems like a big IF though right? Are we going to treat computer neural nets as human from a legal standpoint?

At some point Neural Nets like GameGAM might be good enough to duplicate (and optimize) a commercial game. Can you then release your version of the game? Do you just need to make a few tweaks? Are we going to get a double standard because commercial interests are opposed depending on the use case?

It would be pretty funny if Microsoft as a game publisher lobbies to prevent their IP being used w/ something like GameGAN, but then takes the opposing stand point for something like their CoPilot! Although I'm sure it'll be spun as "These things are completely different!".

Re: All public GitHub code was used in training Copilot

#385

Post by e12e » Thu, Jul 08, 2021, 5:48 PM UTC

loeg wrote: post #368

slownews45 wrote: post #349

Earlier quoted context omitted.

Developers have lost the plot here. The number of people browsing stack exchange and copying code is huge. The number of people who have read GPL'ed code to learn from (from the kernel to others) is huge. The number of people who learned from code they had to maintain -> huge. This idea that a snippet of a code is a work seems crazy to me. I thought we went through this with SCO already.

Stack exchange code is explicitly permissively licensed.

Unfortunately wrongly licensed for its precieved use-case - it'd be better if so used mit or bsd :/

https://stackoverflow.com/help/licensing

I'm guessing most uses of stack overflow snippets are violating the license (no attribution, no share alike of the "remix" - which would probably be the entire program).

Re: All public GitHub code was used in training Copilot

#386

Post by shakna » Thu, Jul 08, 2021, 5:48 PM UTC

nonfamous wrote: post #380

shakna wrote: post #301

Earlier quoted context omitted.

Transformed output _may_ fall under fair use. However - Copilot directly recites code. That is _very unlikely_ to fall under fair use. Redistributing the exact same code, in the same form, for the same purpose, probably means that Copilot, and thus the people responsible for it, are infringing.

> However - Copilot directly recites code. You make that statement as an absolute, but in the interests of clarity, all evidence so far shows that it directly recites code very rarely indeed. Even the Quake example had to be prompted by the specific variable names used in the original code. In practice, the output code is heavily influenced by your own context — the comments you include, the variable names you use, e…

> all evidence so far shows that it directly recites code very rarely indeed.

_Once_ is enough for it to be infringing. The law is not very forgiving when you try and handwave it away.

Re: All public GitHub code was used in training Copilot

#387

Post by Arelius » Thu, Jul 08, 2021, 5:48 PM UTC

AaronFriel wrote: post #221

_Understated_ wrote: post #202

Earlier quoted context omitted.

> By comparison, Copilot is even more obviously fair use. Not sure I see it that way. 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? Copying and storing a book isn't recreating another book from it. Copilot is creating new stuff from the contents of the "books" in this case. Edit: I misunderstood fa…

I've edited my comment with examples and a clarification. Fair use is an exception to copyright and, by definition, copyright licenses.

I don't think that's an accurate description...

Fair use is a defense for cases of copyright infringement, which means you're starting of from a case of copyright infringement, which sort-of muckys up the whole "innocent until proven guilty" thing. And considering it's a weighted test, it's hardly very cut-and-dry at that.

Re: All public GitHub code was used in training Copilot

#388

Post by iseethroughbs » Thu, Jul 08, 2021, 5:49 PM UTC

I treat Copilot as literally a programmer in pair programming. Which means that if it's trained, i.e. it has "seen" GPL code, then it's tainted, and we should treat resulting code as GPL code.

Replace "GPL" with the most restrictive license that's on GitHub, but you get the point.

They're kinda shooting themselves in the foot, because this reduces the commercial potential of the tool to almost nothing.

Re: All public GitHub code was used in training Copilot

#389

Post by shakna » Thu, Jul 08, 2021, 5:50 PM UTC

AaronFriel wrote: post #383

shakna wrote: post #370

Earlier quoted context omitted.

> A $0 settlement means there is no binding precedent and signals to me that Oracle's attorneys felt they didn't have a strong argument and a potential for more. That's not the case. It wasn't an out-of-court-settlement, but an agreement about the damages being sought, the court had already found it to be infringing, and that was part of the ruling. But none of that changes that 9-lines is substantial enough to be in…

I think we agree then, and appreciate the correction on the lower court settlement. If Copilot is infringing, I suspect it's correctable (by GitHub) by adding a bloom filter or something like it to filter out verbatim snippets of GPL or other copyleft code. (And this actually sounds like something corporate users would want even if it was entirely fair use because of their intense aversion to the GPL, anyhow.)

It may be correctable... It doesn't change that Copilot is probably infringing today, which may mean that damages against GitHub may be sought.

Re: All public GitHub code was used in training Copilot

#390

Post by papito » Thu, Jul 08, 2021, 5:52 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…

In addition, this is extremely hard to enforce. I think the amount of code running in closed systems that does not exactly respect the original license is shocking. What was the last case you know where this was a "scandal"?

It only happens at boss level when tech giants litigate IP issues.