Overfitting occurs when a model memorizes the noise and quirks of the training data instead of generalizing. It shows up as an excellent score on training and a clearly lower one on validation data. It is reduced by regularization, more data, or a simpler model.