A lunch-time talk given at Flatiron CCA. I describe under-parameterized and over-parameterized models, and how regularization can improve linear regression. Based on joint work with Soledad Villar (JHU) and Teresa Huang (JHU).