I'm trying to find where one can actually "play." In the Ars article, they show entire paragraphs spit out by the AI, purportedly from stories on this site. The researchers evidently found the AI so convincing that they held back releasing it.
The Ars people got to play and it turned into a surreal Infocom-like adventure, using words that had been written by authors here. I'm surprised more of the community aren't tearing off their eyelids in response.
Here's an example of the output:
https://openai.com/blog/better-language-models/#sample2