Can we please get an official answer on exactly which private repos get used for AI training?

[trusktr]

Select Topic Area

Question

Copilot Feature Area

Copilot in GitHub

Body

This question seems to be closed without an official response:

• Does Microsoft/GitHub include individual's private repos as training data for Copilot?

It would be really nice if GitHub would reply, and select an answer (even if that's a link to docs that explains exactly what people want to know).

As of now, it seems that anyone wanting free private repos may need to look for alternatives to GitHub without a clear answer.

[oxygen]

By making your question highly specific to Copilot you’re not going to get a straight answer.

You should ask them to disclose which ML training uses data or data derived from private github repos.

The longer they keep mum about it the easier it will be to destroy past private models and claim it never happened.

What we need to ask is: if ever in the past, if doing it right now and to publicly state each they time will do it in the future.

Notifying private repo owners should be mandatory.

[ell1e]

I don't think it's okay to train on public repos either, until AIs learn to attribute code. So far it doesn't seem like they can, yet they seem to produce longer verbatim snippets taken from licensed code. (That's at least what I heard.)

[oxygen]

You’re diverging from the more important issue: private not being private

[ell1e]

Why just ask for private repos? Since Co-Pilot seems to admit to not adhere to attribution, I think public repos are a big question too.