New AI copyright laws says weights aren't covered?? #207
Labels
miscellaneous
does not fit an existing category, useful to determine whether we need further categorization
question
General questions about using Llama2
The USA has some new laws on what is copyright:
https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence
If we include weights from a trained model as 'generated works' then I believe this means weights are not copyrightable.
My reasons is that
(1) The model is open to everyone
(2) The dataset is open to everyone
(3) The methods of training is available to everyone
Therefore everyone (with enough money and time) could create trained weights. Each of these different trained weights would be a 'generated work' from an AI model without creative human input.
A 4bit model would then be a 'derivative work' of an uncopyrighted work.
The question then becomes, does the non-disclosure agreement somehow overrule the lack of copyright? Or does it become invalid? Would this also apply to derivative works?
Once the weights get into public domain is the agreement still binding or are they now essentially free to use commercially?
Any lawyers care to weigh in (pun intended) on this discussion?
The text was updated successfully, but these errors were encountered: