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

Or when there are way more parameters than data points

https://en.wikipedia.org/wiki/Double_descent