Tim Bray1d ago

Here's where you go to tell GitHub that it can’t train CoPilot on your code & interactions: https://github.com/settings/copilot/features

Just sayin’

#GenAI

Build software better, together

GitHub is where people build software. More than 150 million people use GitHub to discover, fork, and contribute to over 420 million projects.

GitHub
Show threadsoc1d ago
@timbray It's silly to pretend that this toggle has any actual effect.
Show threadslembcke

@soc @timbray I'm ~90% certain that setting just disables training on your copilot sessions. They absolutely 100% have already trained on all of your public repos by declaring it fair use. I suspect most of the noise is people who don't use copilot and are upset about the latter.

GitHub is managed by Microsoft's CoreAI group. They couldn't be clearer about what they expect in return for free hosting at this point.

8:08pmWeb
Show threadBoyd Stephen Smith Jr.11h ago

@slembcke @soc @timbray "The Pile" is a common training set that includes a snapshot of everything publicly visible on GitHub. "The Stack" is a less common training set that includes most permissively licensed things (non-GPL, e.g.) on GitHub, subject to opt-out.

One judge roughly said training is "fair use", but the next judge rejected that claim. It's quite uncertain right now, legally. (Both were U.S. judges; I'm OOtL for other countries.)

There are a lot of people that are going to scrape anything publicly available, and then come up with a ethical / legal justification later, and being off the biggest hosting sites might help avoid that.