3 ms·
>So, we either go to Github, where our licenses are abused for their shitty M As long as it's on the internet MS can scrape the data so not being on Github is
by marcinzm 4y ago
>So, we either go to Github, where our licenses are abused for their shitty M
As long as it's on the internet MS can scrape the data so not being on Github is only a temporary defense.
- jacooper 4y agoBut GitHub's TOS can only protect them from GitHub repos. If it has impact on CoPilot or not, is not clear.
- marcinzm 4y agoExcept not all contributors to projects signed the TOS since you can upload an existing project with existing non-Github contributors. One person uploading a project does not override or change the license of that project. So MS is already not relying on the TOS for legal protection which means they feel legally they can train on any OSS projects.
- kmeisthax 4y agoWell, at least in the US training an AI will probably fall under Fair Use. In the EU there is an explicit copyright exception for data mining. So I don't think there's a legal obligation for Microsoft to only train within the bounds of public GitHub repos.
- jacooper 4y agoAI fair use is not really clear anywhere. I think SFconservancy's articles about CoPilot are very helpful.
- gaius_baltar 4y ago> Well, at least in the US training an AI will probably fall under Fair Use. Provoking question: so, will it be Ok to train an AI on leaked Microsoft code the publish the models (not the code)? Of course, MSFT won't accept this, but w know that big corps are hypocrite by default.