Very interesting... I never bought the idea of GPT-2 as a threat to humanity, since it's not able to produce meaningful original new content (it just produces t

Very interesting... I never bought the idea of GPT-2 as a threat to humanity, since it's not able to produce meaningful original new content (it just produces thematically and grammatically correct filler), but if your primary goal is to plagiarize, then maybe the threat is real. Problem statement: given context C and sentence X, find a sentence X' that communicates the same idea as sentence X but in different words. This is not exactly GPT-2, but it seems doable. This kind of plagiarism would be difficult to detect. The plagiarizing party could say something like "I didn't copy or reproduce, I just used as input to a NN."