This isn't exactly bot sumissions, and the process is not really scalable: > To quickly weed out inappropriate comments, I handpick from generated comments tho

This isn't exactly bot sumissions, and the process is not really scalable: > To quickly weed out inappropriate comments, I handpick from generated comments those that ensure a high coherence and high relevance sample for submission. So basically it's a validation of GPT-2 making sense with small amounts of text. Judging from the demo test page, they are pretty good texts, but he said it himself that larger texts betray the bot. So, i m not sure what he's trying to prove by using MTurkers, since this does not attack the problem mentioned in his introduction: the fake FCC comments were weeded out through text analysis, not via human work. In all, i'm not sure if this is something that people didn't know about gpt-2. The title is certainly not justified, perhaps "Curated bot comments can't be distinguished by humans to be obviously fake" would be better, but also more banal.