And if they lose, big companies get a signal that they can train on copyrighted data, and serve the resulting models, as long as they dont redistribute the models or reveal their "secret" training dataset. They will probably work to obfuscate the copyrighted parts of the datasets.<p>Meanwhile, open source models and datasets are left out to dry, as they could still be sued. Either legitimately for openly including the copyrighted datasets, or illegitimately when the defendant can't afford to take it to court.<p>For users, this is a lose-lose lawsuit.