Precisely, the model is just regurgitating and pattern matching using a large enough training set where the outputs happen to look factual or logical. Meaning that we're just anthropomorphizing these concepts onto statical models, so it's not much different than Jesus Toast.