It's going to falsely label any recent innovation in language as generated, because the training set will skew human-written as old and GPT generated as new language.