The U-shaped test error of polynomial fits

You fit polynomials of increasing degree $d$ to a scatter of noisy points and track both the training error and the error on a fresh test set.

Describe how each error behaves as $d$ increases, and what the shapes reveal about bias and variance.