Please note that the first generation of any AI Dungeon prompt is GPT-2, and that the AI Dungeon model has been tweaked in various ways from the base GPT-3 model (including fine-tuning on adventure stories)
Note: GPT-3 was trained on common crawl ending in 2019, and so could not have "obtained most of its word patterns" from articles published in 2020. The human quality experiment used stories GPT-3 hadn't trained on (from the "future"). It was given only titles and subtitles.
I've noticed a number of people using AI Dungeon to test GPT-3's abilities. While it's a great way to see how GPT-3 can power an interesting application. It's a poor test of GPT-3's abilities in general.
The first generation of any custom prompt is actually GPT-2.
AI Dungeon is doing something more than submitting input to GPT-3. Unknown, but best guess now is fine-tuning on adventure books and some sort of wrapper on the input (e.g. "What follows is a story...") Discussion in comments here: lesswrong.com/posts/EKP4HtaH…