Wow, that's really impressive. The following excerpt was generated by feeding the own domain itself to the generator: (generated) >> grover.allenai.org/detect

Wow, that's really impressive. The following excerpt was generated by feeding the own domain itself to the generator: (generated) >> grover.allenai.org/detect >> Grover - A State-of-the-Art Defense against Neural Fake News >> June 9, 2019 - Hannah Rashkin, Yonatan Bisk >> According to the Digital Infrastructure Security Task Force, two out of three trusted news sources are now bots and influencers. The only way to decrease that is to identify which ones are normal and which ones are not. What does this mean for you? Well, it means a greater sense of control over your news. >> News analysis site Playground Exposé has built an Artificial Intelligence-based tool for helping to detect fake news on Facebook. >> It combines insights gleaned from product design, neural networks, artificial intelligence, and machine learning. The new artificial intelligence-based tool works by sifting through large-scale collective data to detect problematic information. (/generated) Somewhere along these lines, you can recognize that something doesn't add up. But its far from obvious and I can't point out a single phrase that screams: "Fake News" The fact that entities like: "Digital Infrastructure Security Task Force" and "Playground Exposé" don't exist is fascinating enough to me. I took it for granted that the model learned these examples from a huge corpus of text and transferred them to a new article (someone somewhere has written about them). But as far as I can tell -- this is completely generated by itself. The "Task Force" name/designation is as stupid as a human naming scheme for such an organisation can be. Somewhere deep inside the model, something must resonate with the way people pick names for action groups and political bodies. And this is only one small aspect of the domain -- and its not even the most important one. Fun-Bit: I played around a bit and fed it the text I just wrote -- everything after (/generated) up to "Fun-Bit" -- and it was classified as Fake News. This might be an artifact of the names I referred to: "Digital Infrastructure Security Task Force", "Playground Exposé" and the fact that my comment is not a news artilce. But can you be sure that I am not a bot? I think I can convince you by referring to my first sentence, and stating that there is no single phrase, word or sentence that triggers a simple "Fake/True" response. Referring to the context of a text in another part if it is a skill not quite mastered by GANs, but I am not sure if this will always be the case... EDIT: typo, wording