Occam's Razor says that between the model having somehow developed a persistent internal world model representation and can now reason about it - or is it just

Occam's Razor says that between the model having somehow developed a persistent internal world model representation and can now reason about it - or is it just big enough in terms of training data, knobs and memory allocation that it can capture more and more language corner cases... Well likely the latter is true.