Overfitting from oversized coefficients, the L2 penalty that pulls weights down, and lambda as the dial between ordinary regression and underfitting.