Art history is literally full of transformation of prior works, technology is clearly transformative (if you understand it), copyright law clearly allows for it, and indexing publicly available data is also legal — there’s no reasonable basis for debating the topic being illegal or unethical.

Legal and ethical are two wholly different concepts. There's quite a few things which are legal and not at all ethical.

Today in many states it's perfectly legal to shoot a native american if you're in a wagon circle. Is that ethical?

Fair use is also a positive defense, which means that only the courts can really say if something is fair use or not.

First line, from the post you replied to:

>> “Art history is literally full of transformation of prior works”

There’s no reasonable ethical basis to take issue with. You’re examples and counter points to me add substantive your claims. What am I missing?

This isn't really transformation though. It's compositing pixel-accurate bits and pieces of multiple pieces of art to make a new piece (as proven by stock photo watermarks being replicated in some pieces).

That difference is enough to assert that there is a wholly reasonable ethical basis. Especially when they're obviously not just training it on images with relaxed copyrights.

A collage of prior works is protected as a transformative work, regardless of if you’re able to recognize the seams between the pieces. Even use of trademarks is not protected in works of art if there is no clearly defined trademark violation; for example, putting fries in a box that looks identical to McDonald fries box and offering them for sell would not be legally justifiable by just saying it is art.

Ethically speaking, for the second time, you ignored that within art world, or real world for that matter, making derivative works is completely ethical if done within any related legal constraints; as such, to me, you’re not making a good faith effort, which is actually unethical if intentionally done, so I will not be replying any further.

  > Today in many states it's perfectly legal to shoot a native american if you're in a wagon circle.

Could you please clarify this?

Those laws are illegal and anyone using them as a basis for murder would ultimately be found guilty of it:

https://skeptics.stackexchange.com/questions/25222/is-it-leg...

And yet, every time GitHub Copilot comes up, people on HN are in arms about their code being used to train a model.

Already responded to comparison to GitHub CoPilot in this thread here:

https://news.ycombinator.com/item?id=33079278

Someone else can skip training with copyrighted images, so this copyright argument would be unusable there.

Competent creators will not be replaced, they will use the tools to do their job faster/better. I remember soem tiem ago artists having reservations on "digital art". Someone said it better, a person with a camera is not a photographer.

the problems I have with Microsoft Copilot is that the fuckers did not used MS proprietary code train but only trained with other people GPL/MIT code, this was shitty and implies that their implementation is "dangerous" to use with proprietary code but their expensive lawyers can handle a bunch of open source small developers.

GitHub’s terms of service explicitly allow them to do what they did, if anyone is in violation of the licenses you’re referencing it’s the people that loaded the code into GitHub.

That is not how copyright law works. If someone uploads an image as public domain, or code as a BSD license, without having the right to do so, that doesn't give others the right to use those things.

Unless I am missing something, nothing I said prior conflicts with your point, nor shows that GitHub is willfully violating the law. Are you aware of any copyrighted material that was illegally loaded to GitHub that has been proven to be used by CoPilot, it’s use was not transformative — AND - GitHub has not removed it from the model?

The code used in GitHub copilot is all copyrighted with GPL, MIT,BSD. Also lots of the code was not uploaded to GitHub by the authors so GitHub shity ToS won't apply so you can't claim that the authors agreed to some vague terms that we can now intepret as allows only GitHub to create copilot.

You did

1 MS did not violate copyright since is open source code, andfor soem reason you think that only proprietary code should be removed from copilot

2 even if MS violated the license the guilty person is the dude that uploaded the code to GitHub and click I Agree not he ToS, implying that somehow the GitHub ToS has the ability to change open source license(but t proprietary ones, since you claim MS "removed" such code when it was revealed).

As I mentioned above, MS are cowards, iof their thing is real creating new stuff then put their ass on the line and put their own code in too, put all proprietary code in GitHub in copilot too and then have their well paid lawyers defend their tech.

As written, GitHub users are fully responsible for abiding by the terms of service, which supersede any legal terms uploaded unilaterally by a user.

Name calling is though is very clearly against HN’s guidelines.

Are you sure you understand what you are writing here?

Let me try to use an example, maybe we are both bad at expressing this in english

  I did no say that the user that uploads has no responsibility, but if I upload the Windows code, or FreeBSD code and mark it as GPL and click I agree on the ToS , does this mean that the original license is invalid and GitHub can do whatever they want just because a random person click an Html button?

Prior statement by me within thread you are responding to:

>> “Are you aware of any copyrighted material that was illegally loaded to GitHub that has been proven to be used by CoPilot, it’s use was not transformative — AND - GitHub has not removed it from the model?”

Speculative claims to me are do not merit any response, especially when accusing party of something for which you have no related evidence; which actually might be defamation.

If you have any creditable, independently verifiable, material, etc - evidence of a violation of a specific reproducible CoPilot output, which is documented as having been a valid copyright violation by an unauthorized party, and it has not been removed from the model — please feel free to share links.

As is, GitHub has the right to use code unless they receive what I just requested multiple times from you and appears possible you’re intentionally ignoring for reasons I do not understand.

If I have missed something, it’s not intentional, happy to address any errors or omissions on my party if they exist; assuming you’re not skipping over my points.

>If you have any creditable, independently verifiable, material, etc - evidence of a violation of a specific reproducible CoPilot output, which is documented as having been a valid copyright violation by an unauthorized party, and it has not been removed from the model — please feel free to share links.

Snippets of code were already found, and MS did not removed them from the model but put an IF check to make sure is not appearing again.

So now I need to find specially for you a new snippet and also somehow ensure MS won't add it in the IF checks too ... then I would have proven to you that all that list of conditions are broken.

The proof was already shown, the code was spiting exactly snippets of code, I am not the one that needs to prove that the MS "quick fix" is working correctly now.

Dude I think this kind of AI can be done right, I am not convinced MS did it correct because they already outputed copyrighted code and on top of that they avoided to put their ass on the line. I don't hate the idea of copilot, just the bad implementation of it.

>if anyone is in violation of the licenses you’re referencing it’s the people that loaded the code into GitHub.

MS can't use this excuse, if some dude uploads Windows code or say GTA6 code on GitHub I can't just use it and point you to the GitHub repo and ask you to figure it out with that person.

MS should either

1 tell you that the code is BSD,GPL, MIT depending on what is derived from

2 tell you that is not derived but some new code, but at the same time put his fat ass on the line and also put proprietary code in the mix, like their code and private code on GitHub since they claim the output is not derived from the input.

But they did it half ways, they did not had the courage for 2 and they did not want to respect the MIT,GPL either

Please don’t repost same comment to same thread in different places — already replied here:

https://news.ycombinator.com/item?id=33081744