Is the answer not just simply "the generalisation is encoded in the data"? And that is exactly why deep learning models need huge amounts of data.

The model, size, hyperparameters, optimiser, etc - all they do is convert the data into a form that can be used to make predictions.