It's a fairly standard technique, dating to one of the previous AI summers, there's loads of them. When they're small (~"differential diagnosis between 5 common illnesses"), they're basically normal computer programs; when they get too big (well before Wikipedia sized) they have a scaling problem, which is kinda why LLMs happened.