Wouldn't gpt-2 be much harder to detect. This could just be markov chains which were trained with all their comments from the past. Still pretty cool