Wouldn't the model start overfitting at some point? Degrading generalization for accuracy on the training set.