5 ms·

> Is copilot really a violation of GPLv3 (our license)? How is co-pilot different from someone lifting sections of code? If it's reproducing vebratim your code

by zvr 4y ago

> Is copilot really a violation of GPLv3 (our license)? How is co-pilot different from someone lifting sections of code? If it's reproducing vebratim your code, yes it's a violation. And this is no different from others copying sections of your code. The license that you chose says that people are allowed to do many things (e.g. copy and modify the code), provided they also fulfil some obligations. If they don't do that, they are violating the license. Please note that this is not specific to GPLv3 (your chosen license). Other licenses, BSD-3-Clause, for example, also give rights (e.g. to copy and modify) and have their own set of obligations to be fulfilled (e.g. "Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer"). If someone redistributes source code (not the "complete" source code, even a "section of code" to use your words), they have to fulfill the obligations, or they violate the license.

seadan83 4y ago
It sounds like Github needs to index the license of where it obtained code, and check the license of a project using co-pilot so that it only suggests GPLv3 code for compatible projects. I'm not sure that it is really enough, enforcement is tough, it could be like one of those "are you 18" checkboxes. I get the impression that this article and it's reaction is largely a lot of people wanting to cry big brother and mostly just shit on Github & Git. IMO it would be more appropriate to look for solutions and post a public letter to github and/or just pursue actual litigation before advocating FOSS projects spend inordinate amount of effort across the board to vacate Github.
- zvr 4y ago
  I think the article starts from the position "everyone should respect the software license" and obviously promoting the SFC views. You are correct that the problem is hard, because adding licensing info to training data and subsequently use it is something not yet accomplished by anyone (who at least has spoken publicly about it). It might be the only way out would be to have different training sets according to license, but then you lose the advantage of scale. On the other hand, I believe it's a little much to ask SFC to do the cutting-edge research and propose solutions to Microsoft. Let's not forget that, when faced with the problem, their chosen path was "completely disregard licenses (for now), ask legal later". Many legal opinions are that there would not be any issue if copies of the input code were not reproduced verbatim; unfortunately this is not the case. Looking forward to more exciting developments!
  - skjoldr 4y ago
    Microsoft could afford to train several models for every significant class of licenses. So if you don't want the model which included e.g. GPL code in its training set, you would opt for the model that was trained on the dataset which didn't include any. It's a bit of a hassle for MS, but that would partially resolve this issue.
  - pabs3 4y ago
    Amazon claim their Copilot equivalent knows about license info.

by zvr 4y ago

seadan83 4y ago

It sounds like Github needs to index the license of where it obtained code, and check the license of a project using co-pilot so that it only suggests GPLv3 code for compatible projects. I'm not sure that it is really enough, enforcement is tough, it could be like one of those "are you 18" checkboxes. I get the impression that this article and it's reaction is largely a lot of people wanting to cry big brother and mostly just shit on Github & Git. IMO it would be more appropriate to look for solutions and post a public letter to github and/or just pursue actual litigation before advocating FOSS projects spend inordinate amount of effort across the board to vacate Github.

zvr 4y ago
I think the article starts from the position "everyone should respect the software license" and obviously promoting the SFC views. You are correct that the problem is hard, because adding licensing info to training data and subsequently use it is something not yet accomplished by anyone (who at least has spoken publicly about it). It might be the only way out would be to have different training sets according to license, but then you lose the advantage of scale. On the other hand, I believe it's a little much to ask SFC to do the cutting-edge research and propose solutions to Microsoft. Let's not forget that, when faced with the problem, their chosen path was "completely disregard licenses (for now), ask legal later". Many legal opinions are that there would not be any issue if copies of the input code were not reproduced verbatim; unfortunately this is not the case. Looking forward to more exciting developments!
- skjoldr 4y ago
  Microsoft could afford to train several models for every significant class of licenses. So if you don't want the model which included e.g. GPL code in its training set, you would opt for the model that was trained on the dataset which didn't include any. It's a bit of a hassle for MS, but that would partially resolve this issue.
- pabs3 4y ago
  Amazon claim their Copilot equivalent knows about license info.