And is there a way to tell GitHub _not_ to analyse repos that I own?
back
2 comments
Yes, there is an account setting checkbox labeled "Allow GitHub to use my code snippets for product improvements" - https://github.com/settings/copilot
I'm a bit confused how this will work in practice, because I don't see how they can retrain their entire model every time someone toggles that setting.
You can't access that page or toggle that checkbox unless you pay for Copilot first. This might be the most literal case I've ever seen of "if you're not paying, you're the product"
Perhaps I'm in some A/B testing cohort, because I certainly haven't paid for anything.
Were you part of the Copilot demo/preview? It appears that doing so leaves your github account in some sort of mostly enrolled state for Copilot that lets you tweak these settings compared to an account that wasn't part of the preview that just takes you to a "start your free trial" page.
That could be it. I've never actually used it, but maybe I applied for the preview at some point.
If your code is e.g. MIT or GPL licensed then someone else could still upload a fork to another GitHub repo.
We might really need a new clause in the license.
The MIT license states:
> The above copyright notice and this permission notice > shall be included in all copies or substantial portions > of the Software.
If you are using Copilot and it produces a MIT-licensed snippet of code, you are in plain violation of the license. It is wilful copyright violation.
A snippet by definition is not a substantial portion.
'Substantial portion' is not interpreted by you, it is interpreted by courts and juries. How it is interpreted depends on jurisdiction and more[0], the commonly accepted 'safe bet' is any single line of the software.
>Substantial portion' is not interpreted by you, it is interpreted by courts and juries. How it is interpreted depends on jurisdiction and more[0]
Yes and like standard your cite says none of the interpretations would cover a snippet.
>the commonly accepted 'safe bet' is any single line of the software.
Github/MS aren't going for the 'safe bet'. They're going for the maximum that they can get away with.
Which would seem to be an excellent hook for a wilful copyright violation lawsuit. This is not an accident.
Is there a way to prevent GitHub from doing this with forks of repos of my code?
It will be interesting to see the next version of robots.txt that begins to appear in public repos to prevent this type of data scraping.
If somebody starts going around neighborhood shooting paint balls, requiring everybody to wear "please don't shoot paint balls at me" t-shirts seems like a suboptimal solution.
If somebody left a code with some license 10 years ago, then that's how he published it and it should be respected. If there are some additional files, changes or checkboxes required on the author side (who may be dead btw), then it is a theater not a solution.
I could totally go with let's abandon licenses (but would prefer starting with patents), what's online is ours. But if we want to keep them then the license itself is enough of a statement. It's not like we have a hard time writing software which interprets them... it's more like "but we would have to do a big refactoring".