@mapto@j3j5@blogdiva@KathyReid
Not sure what the relevance of corrupt-and-train is to the legal argument being made here. Wolfson claims "they do not piece together new images from bits of images from their training data" but one could argue that neither is transcoding a Disney movie into a lossy MPEG format. Each frame is regenerated from discrete cosine transforms and motion vectors. Error correction happens during storage. Does that make it fair use?