Comment by encomiast

For sure, that could be an instance of infringement depending on how it is used. But that's a minuscule percentage of the output and still might be fair use (read the decision in Authors Guild, Inc. v. Google, Inc.). But even if that instance is determined to be infringement, it doesn't mean the process of training models on copyrighted work is also infringement.