When a model memorizes quirks of its training data instead of learning patterns that hold on new data. Great on what it studied, poor on anything else.

ExampleThe unlimited decision tree, or the student who learns that 8 divided by 4 is infinity by turning the 8 on its side.