[janef0421]

This seems poorly argued. There's a lot of emphasis on technical minutiae, but the question is whether the way the model embeds and reproduces the features of the input images infringes on copyright. I would think that is more dependent on the capabilities and functions of the models, not the specific mechanism of implementation.

Also, the Jevon's paradox bit is a tangent. AI image tools may become widespread, and their use may expand the art market, benefitting artists who make use of them. That is not a relevant point, because the lawsuit is not about the legality of image-generating neural networks; It is about whether using an unlicensed image to train such a model infringes on its copyright. If it was found that it does, then AI art could still transform the art market; All that would change would be that images would need to licensed for training.