Authors need to come up with the literary equivalent of “trap streets”. I can’t help but think the “proprietary” secret nature of the training data is just an

Authors need to come up with the literary equivalent of “trap streets”. I can’t help but think the “proprietary” secret nature of the training data is just an attempt to sidestep the copyright issue. Like the “national security secret” line that the US government likes to use. Crap training data = crap model. Only licensed training data = base model not as good as open models. Stolen training data plus RLHF to suppress this fact = great model!