Licensing Training Data and Attributing Copyright of Derivative Content From Large Language Models Can Resolve Up- and Downstream Copyright Issues.