Perhaps the biggest blessing
and curse of those smaller models they were using from OpenAI was how susceptible to over-fitting they were when it came to Latitude training them on the CYOA text stories they swiped from ChooseYourStory. So you had the same names/characters popping up in totally unrelated stories and people in those communities posting about how baffled they were that this enigmatic Count Gray was continuously showing up to be a nuisance. It was surprisingly charming, in a way, almost like the stories really
were part of some kind of shared setting. Nowadays, the large models and lack of overfitting helps prevent that, though you do have
certain idioms and phrases that tend to show up repeatedly.