“Dan Shipper argues that because AI models are trained on the same corpus of past work, their default output creates undifferentiated 'slop' or 'visible sameness'.”