Greg Hogan <[email protected]> writes: > On Mon, Dec 15, 2025 at 3:21 AM pelzflorian (Florian Pelz) > <[email protected]> wrote: >> >> U.S. courts have cared about paying authors of books used as training >> data. [2] >> >> [2] https://en.wikipedia.org/wiki/Artificial_intelligence_and_copyright > > Which cited case are you referring to, the case ruled in favor of Meta > or the case where the judge "granted a summary judgment for Anthropic > that affirmed that their use of purchased books for training was > within fair use"?
The one in favor of Meta, at first sight devalued by the ruling in favor of Meta, but it does show that what was used for training may have an impact on whether Guix and its users are allowed to use and share. These Intel Openimagedenoise weights maybe incorporate CC-BY-SA photos, which maybe, like the Meta ruling, is okay to use for training and all, even though it invalidates the intention of -BY, but we do not really know. If we do not know, do not put the data in Guix. Can the data be left out and downloaded from elsewhere like --tessdata-dir option in Tesseract OCR? Regards, Florian
