Post

Given that whatever LLM they're using under the hood is almost certainly trained on the Git source code, even if they didn't also make the Git source code available to the LLMs (which they well may have), and the test code is anyway also copyrighted and GPL-licensed... Good luck, buddies.

This is like me deciding that my AI upscale of INCEPTION which also recuts it in the order suggested by that one Reddit post "isn't a derivative work" just because I say so

Big red flag here: "especially given the pretty massive and widespread architectural changes needed to make the implementation libified and memory safe". "Magnitude of changes" does not affect copyright. If you start from the original it's always a derivative work no matter how much you change it

> "Magnitude of changes" does not affect copyright. If you start from the original it's always a derivative work no matter how much you change it False. BSDi v. AT&T.

9/6/26, 11:09 pm0 reposts10 likes

Or. Okay, right. I think I misread your post? But the point is that starting from or having knowledge of the original codebase doesn't prevent relicensing if you've changed the code enough. I STRONGLY doubt that's what happened in this case, but it's not quite that open and shut.

Having knowledge of the original codebase doesn't necessarily prevent relicensing AIUI. Starting from the original codebase—or even the compiled binary—generally would, I believe, that's why clean-room reimplementation needs to work the way that it does.

en.wikipedia.orgUNIX System Laboratories, Inc. v. Berkeley Software Design, Inc. - Wikipedia

This one? en.wikipedia.org/wiki/UNIX_Sy....

That is what I was thinking of. Have I been misunderstanding the case?

It looks like it settled out of court

...Yeah that's an incredibly important detail for me to not recall. My recollection is that prior to the case being settled USL was not very successful, hence the settlement for "well you only need to remove only six files"