They tend not to overfit ... when there are way more data points than parameters.