Hacker News Reader

Comment by NoraCodes | Hacker News Reader

I'll tell you right now, having my GPL-licensed code stolen has a pretty negative impact on my happiness.

What snippets were stolen from your code and how has it affected your happiness?

What if I read your code? Now I have memory of it in my brain and it will influence my next projects. Is it stealing?

Actually it could be! If you read GPL code and remember it well, then write a near exact copy of it into your proprietary software project, that's a violation of the GPL.

I have no opinion on whether there's infrigement or not, as I am not a lawyer, but I found the argument that the terms of use specify that you allow them to analyze your code pretty convincing:

> We need the legal right to do things like host Your Content, publish it, and share it. You grant us and our legal successors the right to store, archive, parse, and display Your Content, and make incidental copies, as necessary to provide the Service, including improving the Service over time. This license includes the right to do things like copy it to our database and make backups; show it to you and other users; parse it into a search index or otherwise analyze it on our servers; share it with other users; and perform it, in case Your Content is something like music or video.

Copilot seems to go further than all that. It can suggest verbatim, nontrivial, copyrighted work with no means of attribution. And the sources are so broad no human alive could say with certainty any given output is not infringing, unless it's so trivial it cannot be copyrighted.

I've always had trouble understanding intellectual property when it comes to code, because if I read open source code, remember everything, and write my own version (even if slightly different) 10 years later, I am not 100% sure whether it's copyright infrigement or not.

I see it as something akin to a painter studying someone else's work, then reproducing some of the techniques invented by the original artist... except that their techniques didn't have a license I guess?

I think I am not equipped to fully understand the legal boundaries between inspiration and theft, and I think most programmers are in the same situation.

Painter analogy falls down when you consider Copilot can output verbatim chunks. So more like a photo of a painting or a stroke-for-stroke copy, even if portions changed or only parts taken.

Now if changes are so significant it becomes impossible to recognize the reference then that could be legit. Though based on how Copilot works I don't think that can be assumed, or even proven. When I studied art the teachers usually taught us to begin with our own original photograph, ideally without trademarked items, for reference.

And even in literature or journalism one must quote sources, even if paraphrased. I was taught to put down any inspiring work and only begin my own work after taking a break, to reduce the likelihood of unintentionally copying the original.

Whether or not I agree with your interpretation of that clause (I am choosing not to analyze that too deeply), the vast majority of my code isn't on GitHub because I uploaded it there... it is on GitHub because it was open source and someone else--someone who uses GitHub to manage their projects--uploaded it, whether as a mere fork or as a legitimately derived work. If you were right, and this did matter to their legality, this would thereby mean 1) that Copilot's database is already tainted and 2) that I guess GitHub isn't actually capable of being used to host GPL projects at all (which I doubt is the intention).

Yeah this case is interesting actually, I had never thought of a legitimate derivative landing on github, not something the original author intended to happen...

My guess is that you're right and it does mean 1) and 2), but again, it's probably a matter of interpretation and actual ruling on the matter.

Maybe you should switch to the Unlicense. I'm so much happier not giving a damn or punishing my users with pettiness.

Licenses don't matter when Microsoft decides that AI training is fair use.

I'll take "I didn't read the GitHub TOS" for 500 Alex.

Well not everyone is in contractual privity with GitHub — like the poster a few up whose code was included in someone else’s repo. GitHub’s TOS have no effect in that circumstance.

"wants to participate in and use open source, but is not happy when others use their code"

*Use their code without complying with the license

Copilot is transformative work.

Objection, legal conclusion.

I couldn't care less about whatever restrictive license people choose and decide to call open source

Well then dont use their code you have to legally follow the codes license.

I wonder if you could just ruin these systems by having some bots constantly upload bugged code into github

I think my biggest worry about AI is that we hit a big wall right about now because training data is now contaminated with AI outputs. That's why OpenAI puts watermarks on images, I'm guessing, but Stable Diffusion doesn't, and the free one is probably going to get a lot more usage than the paid-for one.

But, maybe not, because humans liked the outputs enough to share them and put them on the Internet, so maybe they don't have much of an effect quite yet.

I think the training limits other applications like self driving cars a lot too. Training is frequently done in contrived scenarios. Now, imagine if GM or Ford put sensors and telemetry for future self driving car use on their models decades ago. You'd have practically every road mapped with ten thousand observations a piece by now. It wasn't for lack of forsight to not collect these data, it was really for lack of profit. Sensors and other data collecting systems are not new technology. Neither is machine learning, its also decades old. The only thing different these days is that the compute required to run these models on these data has gotten cheaper, but you could have been collecting data the whole time with anticipation of cheap compute to run these models in the future. Then you'd have a leg up on everyone in the business with your massive dataset of decades of real world, not test tracks and other such situations utilizing an employed driver in all likelihood.

"The GPL is a free software license, and therefore it permits people to use and even redistribute the software without being required to pay anyone a fee for doing so."

How is GHC breaking this license?

GPL requires that derivative works - which, I argue, Copilot is - also be distributed under the GPL. Microsoft disagrees with my interpretation, and wins by default because I can't afford to sue them.

The GPL has requirements, and Copilot fails to adhere to those requirements. Here's an important one:

>You may convey verbatim copies of the Program's source code as you receive it, in any medium, provided that you conspicuously and appropriately publish on each copy an appropriate copyright notice

Say I drew a lovely picture and hung it outside my house for passers by to look at. I say “sure you can take a copy, no problem”, they then sell a billion prints, and don’t credit me at all.

Legally, not stealing. Ethically? I’d call that stealing.

Except that's not exactly what they are doing, is it? Following on your example, it'd be like selling the advice of someone who has seen a lot of these pictures passing by windows, with a very good memory and understanding of pictures.

Github Copilot is not spitting out code verbatim, it's learning what code is, what shape it usually has, and trying to retrofit your own code onto the shape it thinks code should have.

It's not like it's doing a query like `SELECT * FROM COPIED_CODE WHERE CODE STARTS_WITH "def my_func("`.

Legally that actually is stealing. They must have documented reproduction rights. Misunderstanding a verbal comment will not fly in a court.

Reply on news.ycombinator.com