[didibus]

Have you not seen the recent HN about Codepilot?

Developers had a similar response where they don't think that training a model using their copyrighted work is fair use, especially for commercial use.

They'd expect that the model authors would need to comply with the open source license like for any other use.

This is basically the debate here, it's same for artists or developers.

The question is if it makes sense to allow someone to use your copyrighted work to train a model that they'd then use for commercial purposes, without needing a license agreement.

[orbital-decay]

The debate over Copilot is not new either: at launch, it kept suggesting Carmack's fast square root code. (which got patched a day or so after). The recent shenanigans were about the code that was published under wrong license by third parties, and then picked up by Copilot.

Any overfitting (reproducing the training data verbatim) is not intended and is avoided as much as possible. But it's inevitable sometimes, and is a worse problem in code than art, as you have no way of knowing if it reproduced some private code verbatim.

And no, it's certainly not the entire debate, many people are also afraid that their job will be automated. Which is a silly assumption in case of Copilot as well as with image gen models - it stems from the misunderstanding of what it can and can't do, what's fundamentally possible with it in the future etc.

The moral panic is strong, but I have the feeling that it's amplified by social media without really being substantiated. The advent of CGI was the same, it just was more quiet due to the lack of social media at the time.