(along with all other licenses that require attribution)
as it will allow you to launder code automatically through an LLM to remove copyright
however if the suit is successful: every company/individual that has used it is likely suddenly liable for millions of claims of copyright infrigement
I look forward to sending out demands for settlement to everyone that's ever publicly admitted using copilot
$150,000 per infringement with wilfulness, less without
(i) actually produced code which is verbatim the same as a block of GPL code
(ii) got caught
>I look forward to sending out demands for settlement to everyone that's ever publicly admitted using copilot
Feel free, they'll tell you to leave them alone. Then what? Might as well ask every fortune 500 company for a pony instead.
1. LLMs are learning like a human, so it's fair use -> GPL dead
2. anything LLMs output are a derivative work (in the copyright sense) of what went into it -> all Copilot output is infringing the GPL
in the second case: anyone that's used it is now liable (even if they didn't intend to be)
No it won't, obviously if it copies code exactly then you can't use that. The question is whether Microsoft is liable for the fact that Copilot has the ability to output copyrighted code sometimes or whether people using it just need to check that it hasn't done that before using the code (Copilot can also do this automatically)
Google can also show you GPL code in its results, but people aren't trying to sue Google and the user is responsible for checking the license before using it (though Copilot makes this harder)
Disclaimer: I haven't read much about the actual lawsuit and I'm not a lawyer but I assume this would be the case
Not really because the GPL can be updated with a clause that allows GPLv5 (or whatever the version is going to be) to be used to train public LLM models, but explicitly forbidden to be used to train private models.
I somehow don't think this is the end of the GPL... Yet!
Microsoft's position on Copilot is that it's fair use:
> When questioned, former GNOME developer and (at the time of writing) GitHub CEO, Nat Friedman, declared publicly “(1) training ML systems on public data is fair use (2) the output belongs to the operator”.
https://www.fsf.org/licensing/copilot/if-software-is-my-copi...
I don't know how GPL (or copyright in general) can survive in the long run with these technologies.
I don't see why you couldn't do the same thing to e.g. the binary of the Windows kernel
you're unaffected if you only offer an saas though, as the end user never has any code/binary to launder
Forget the binary; there have been Windows code leaks every now and again over the years. Feed one of those into a model, start generating code for ReactOS, and see how long until MS decides that actually AI is infringing...
Only if you can prove that you are the copyright owner of the original work.
That might be a challenge for many open source projects. Even projects that require copyright assignment might not have sufficient paperwork to prove this in a court of law. The copyright might not even have been the persons to assign in the first place.
You would also face the burden of proving that the fragment that Copilot generated was sufficient to be copyrightable in the first place. The limited grammar of most programming languages would probably make proving that something was copyrightable at the function level hard. Just because the entire work was licensed under the GPL, it doesn't necessarily follow that all the individual fragments when separated out are.
Outside of sampling, this is an area that the courts have largely punted on for good reason. It's a rabbit hole nobody wants to go down.
Either outcome opens up a huge can of worms that I suspect nobody really wants to touch because it likely ends in mutual destruction.
Humm, then perhaps should be trained LLMs with leaked Microsoft code. Protocols, controllers or any kind of stuff that could contribute advances for executing Windows things within Linux.
Microsoft would react establishing their own limits, whichever option they choose to take.
I very much doubt that is a threat to Microsoft.
It is technically very straightforward to run Windows “things” under Linux thanks to virtual machines and/or RDP to a server and some UI trickery to make it seamless and facilitate interoperability between the two OSes. Parallels does quite a bit of that on macOS for example. A similar solution would be developed for Linux if there was enough demand for it.
I think the problem here is: by auto completing GPL code to developers it might open the opportunity of your company getting sued for using GPL illegally
I would also imagine those companies whose business is built around the open source development they do -- open core, SaaS, or otherwise -- would have a claim to financial damages as a result of stolen code.