Thank you! I love thinking about this stuff

Thanks Nimi for the RT!

Background painted by me from photograph by Yuliia Sereda via Unslpash

... every artwork, every scenery they've EVER seen in their life? Yes. Is that the same as the diffusion model having "seen" billions of images? Personally, I don't think so. AI models can't/won't disrupt their existing biases/weights from being fed one image. Humans sometimes do. And that matters.

Whether the training data that genAI gets fed is "inspiration" is hard. Diffusion is "generative," but it can't be stimulated like humans or have an "Eureka" moment. A human can be awestruck by ONE piece of artwork and create something. Does that creation require their subconscious knowledge of...

As it stands I believe the best way is that no one owns copyright to generated AI content. I also think it's highly unlikely that US courts would set precedent for AI trainers to seek permission & compensate for training data, but I hope places like within the EU can offer countering verdicts.