Thanks for the link. That sounds more like prompt engineering? If I understand that correctly it is providing short journal entries (1K words) and GPT3 is imitating the vibe of that (or whatever it does). But it is not “training” a Modell on all the journal text.