3 ms·

Tell HN: Linux developers were actually working for MS thanks to Copilot X

12 points by behnamoh 4y ago

I just can't get my head around the fact that a big tech company can buy Github, a website that almost all the open-source world relies on, and then use the code on that website to train an LLM. Were there Terms of Service on Github in 2010 that allowed using my code for training LLMs? I don't think so. And yet we all have accepted the fact that MS can benefit from Linux and open-source developers in an indirect way.

bityard 4y ago
Your question doesn't make sense. The right question is, "Which open source licenses prohibit a corporation from using their source code as AI training data?" And the answer is none of them.
- ipaddr 4y ago
  The answer is most as most code on github doesn't have a license. Copying a song from the radio to a digital file is a copyright issue and it doesn't matter in what format it's store. In this case it's stored as tokens. Same as copying a song to a computer.
  - techdragon 4y ago
    Except the terms of service you agreed to in order to make an account and upload code to GitHub at all, regardless of any licensing you choose to put on it to enable others to use your code… that Terms of Service you agree to with GitHub allows them the right to keep a copy of your code, to use it for a bunch of explicitly described things like indexing it for search, displaying it to other people if it’s not a private repo, and there’s a little block of language in there that also covers them for future services they may develop that are not yet explicitly named. Future services like training CoPilot. We have them permission, to do the training that’s not something we can argue on a legal grounds… the only question we can argue is the derived works licensing question.
jfoster 4y ago
Perhaps what happens with Getty v Stability AI will have some bearing on how this eventually plays out. https://www.theverge.com/2023/1/17/23558516/ai-art-copyright-stable-diffusion-getty-images-lawsuit https://www.theverge.com/2023/1/17/23558516/ai-art-copyright...
salawat 4y ago
You agree to allow your data to be used for internal research purposes... ...is generally the approximate clause in question.

12 points by behnamoh 4y ago

bityard 4y ago

Your question doesn't make sense. The right question is, "Which open source licenses prohibit a corporation from using their source code as AI training data?" And the answer is none of them.

ipaddr 4y ago
The answer is most as most code on github doesn't have a license. Copying a song from the radio to a digital file is a copyright issue and it doesn't matter in what format it's store. In this case it's stored as tokens. Same as copying a song to a computer.
- techdragon 4y ago
  Except the terms of service you agreed to in order to make an account and upload code to GitHub at all, regardless of any licensing you choose to put on it to enable others to use your code… that Terms of Service you agree to with GitHub allows them the right to keep a copy of your code, to use it for a bunch of explicitly described things like indexing it for search, displaying it to other people if it’s not a private repo, and there’s a little block of language in there that also covers them for future services they may develop that are not yet explicitly named. Future services like training CoPilot. We have them permission, to do the training that’s not something we can argue on a legal grounds… the only question we can argue is the derived works licensing question.

jfoster 4y ago

Perhaps what happens with Getty v Stability AI will have some bearing on how this eventually plays out. https://www.theverge.com/2023/1/17/23558516/ai-art-copyright-stable-diffusion-getty-images-lawsuit https://www.theverge.com/2023/1/17/23558516/ai-art-copyright...

salawat 4y ago

You agree to allow your data to be used for internal research purposes... ...is generally the approximate clause in question.