161–170 of 212 posts
Here's the brutal and ugly truth: why isn't our personal data treated as private property? It's because those who write the laws governing its status either lack the requisite understanding or else practice a form of, to put it mildly, motivated reasoning.
Microsoft of course will implement compliance standards as necessary (they genuinely do not want to break the law), but what does this mean for smaller companies and individuals training models?
Earlier quoted context omitted.
Sure, but there's still the license at play here. It's not like they trained it only on public domain/CC0 code. What happens when it verbatim outputs a significant amount of code that was originally MIT, or BSD, or GPL licensed without the appropriate attribution. It can create unintended copyright violations and potentially open people using it up to liability.
I suppose it really depends on if they spit out verbatim reproductions of code or whether it is the equivalent of a 10-year experienced programmer who has just seen a lot of code but isn't reproducing anything verbatim. We shall see, by Googling some of the code it spits out. FWIW GPT-3 doesn't really tend to spit out verbatim reproductions of copyrighted books.
But it does spit out whole paragraphs at a time. This is easy to test by going to any of the GPT-2/3 playgrounds on-line (e.g. AI Dungeon), and playing with prompts. Very specific prompts work best, but sometimes even with a generic prompt, if you let the model continue on its own past the first output, it might just shunt itself into a path where following the most probable continuations happens to reproduce a substantial portion of some work verbatim.
How does Copilot avoid training on malicious code? There might be bad actors who would love to have their code scraped for this...
Earlier quoted context omitted.
> Can't you host code on GitHub that is not "free" for commercial use? If GitHub scraped these projects then it's a problem. It did. It's spitting up the AGPL in empty files, and AGPL'd code isn't free for commercial use. It requires people who use it to make changes available under the same license.
My non-expert reading of the AGPL seems to indicate that using AGPL code in a commercial project is find as long as you don't change it . GitHub wouldn't necessarily have needed to change it to include it in a data set, so I'm not sure there's a license violation. However the gray area is that the massive data set of which it is a part will spit out new code that has, in some way big or small, been influence by the A…
When you combine some [A]GPL code with said commercial project, you create a derivative work. And that work has to be compatible with [A]GPL's conditions as well. Which is usually not great for a commercial (closed-source) project.
Earlier quoted context omitted.
If large company violated my open source license, I'd certainly have a problem with it. Use my code all you want, but follow the basic rules. There's entire websites dedicated to GPL violations; people do care.
If the license allows commercial use, what would the GPL violation be in this case?
How does Copilot avoid training on malicious code? There might be bad actors who would love to have their code scraped for this...
Earlier quoted context omitted.
> Can't you host code on GitHub that is not "free" for commercial use? If GitHub scraped these projects then it's a problem. It did. It's spitting up the AGPL in empty files, and AGPL'd code isn't free for commercial use. It requires people who use it to make changes available under the same license.
My non-expert reading of the AGPL seems to indicate that using AGPL code in a commercial project is find as long as you don't change it . GitHub wouldn't necessarily have needed to change it to include it in a data set, so I'm not sure there's a license violation. However the gray area is that the massive data set of which it is a part will spit out new code that has, in some way big or small, been influence by the A…
Any restriction imposed by the GPL is also true for the AGPL. In particular, if you reuse some code from a (A)GPL project, even if you don't change the code, you must release your whole project under this license too, and give attribution to the author.
For the LGPL, the same thing applies, except you can reuse LGPL code in your code without releasing it under a *GPL license if someone can replace this code at execution time by another code (static linking does not allow this).
To me it seems, the whole subject requires additional consideration in licensing. It is a little like applying telephone based law to the internet. It will not 100% fit. If the creators interests are not clearly expressed anymore with a license, we need updates to the license texts. Let's look at MIT: ____________________ "Permission is hereby granted, free of charge, to any person obtaining a copy of this software a…
Let's handle the simplest case first: Copilot can and does regurgitate large pieces of its training dataset verbatim. This is a well-known and trivially demonstrable property of all ML models in this family. Would such exact copy fall under the license of the code being copied? This of course needs to be tested in courts, but my gut says "yes". The problem now is, if you're using Copilot, you may end up with such copied code in your codebase without ever knowing, and this might open you to liability.