Copilot (again)
Assuming Copilot will continue to be trained, I wonder if people can upload deliberately garbage repos on GitHub. You know, take an existing code base (appropriately licensed, of course), run it through git-filter-repo reversing all the lines or other nonsense things. Just leave them out there.
(Of course, if enough people did this, I assume it wouldn't be long before this became against the TOS of GitHub or scabs started reporting repos. What a dumb present we've created.)
re: Copilot (again)
@cstanhope Wondering if they're using the stars s a filter, so if you have a repo with few stars then it probably will fly under their radar. They're likely looking for large codebases with multiple contributors, not small one-off projects.
(Speculation, of course)
re: Copilot (again)
@craigmaloney Copilot points to a future of paying for the privilege of creating and solving captchas for commercial ML.
(Although that's not really new given Tesla's approach to developing self-driving vehicles.)
(Oh geez, I'm bein' a downer today.)
re: Copilot (again)
@craigmaloney Yeah, and it just occurred to me that any additional training may be heavily slanted around people actually using the system now. So the opportunity to poison the training set has been reduced I suspect.