Overfitting

Training error decreases, but test error increases. This happens when a model is too large relative to its dataset. The scaling laws implicitly address overfitting by showing that large models trained on proportionally large datasets don't overfit.