Malus, which is a piece of "satire" but also fully functional, performs a "clean room" clone of open source software, meaning users could then sell, redistribute, etc. the software without crediting the original developers. But I have a hard time with the "clean room" argument since the LLM doing the behind-the-scenes work has already ingested the entire corpus of open source software -- and somehow the output of the LLMs isn't considered a derivative work.

you are viewing a single comment's thread
view the rest of the comments

[–] madnificent@lemmy.world 5 points 1 week ago (1 children)

Could link to your reasoning and/or summarise it here? Thanks.

[–] grue@lemmy.world 15 points 1 week ago* (last edited 1 week ago)

The GPL requires that derivative works must also be licensed under the GPL.
LLMs are trained on GPL code.
LLM output is a derivative work of the training data (especially if it's asked to replicate one of the works it's trained on!).
Therefore, all LLM output is either also GPL, or if it's also been trained on stuff with conflicting licensing, just straight-up copyright infringement to use at all no matter what.

Laundering copyright is what LLMs do. It is fundamental to how they function, which means that they are a fundamentally illegal technology.