the larger model doesn't have a notably larger dataset to my understanding. It's just got more parameters, so learns higher-order abstractions about the dataset