Raw LLM Responses

I guess it depends what they're doing with it. It's one thing to trust Github to just store your source code and give it back to you when you ask for it. It's a bit different if they're also training AI models on your source code and then let other companies use those models, and you have to trust those models not to accidentally leak stuff when they can barely even pass the [apple test](https://www.reddit.com/r/singularity/comments/1alwn8h/gemini_ultra_fails_the_apple_test_gpt4_response/). That said, I don't think they're doing this. IIRC they say they only train on publicly-available repos -- it may access your private repo when talking to you, but your repo isn't part of the general training corpus. And there's good reason for them not to train on private data, because it used to be easy to convince Copilot to [output its training data](https://www.reddit.com/r/programming/comments/oc9qj1/copilot_regurgitating_quake_code_including_sweary/). That's a problem even if it's only trained on open source -- at least it's not a leak, but you don't want to *accidentally* be copying some open source code and violating its license. In their defense, they've basically [promised to defend you in court if this comes up again](https://blogs.microsoft.com/on-the-issues/2023/09/07/copilot-copyright-commitment-ai-legal-concerns/).