Overfitting: why the model that fits its data best predicts worst

Afterwards you can say why a model that fits its training data perfectly predicts new data worse, and choose a model's complexity with held-out data.