@ben Too bad CC argues that AI training is "fair use" and thus none of the license applies https://creativecommons.org/2023/02/17/fair-use-training-generative-ai/
I was hoping CC would be the one to come up with lawyer signed license wording to prohibit use of LLM training but it doesn't look like that will happen :-(
@schelling @ben hm, let me see if I understand correctly what this means.
I think it means this: if something, call it a file, is licensed with a CC license that requires attribution, then it is okay to ignore attribution and just copy the file and feed all of it bit-by-bit it into a machine as long as you call it "learning" instead of "copying"? Even if this means the machine can later produce an exact copy (without attribution) because it "learned" the contents of the file?
Is that right?
@eliasr @ben As far as I understand that is what the entire modern "AI" industry is built on. Basically arguing that training models on data is "fair use" and thus copyright does not apply.
The first time I heard about it was NVidia feeding the entirety of Flickr (and other image sites?) into a model to generate faces. It seemed iffy then because most on Flickr is either fully copyrighted or not usable commercially, yet NVidia didn't care. And so did most of the world.