Groups
Scaling laws say that model loss typically follows a power law that improves predictably as you increase parameters, data, or compute.
The bias–variance tradeoff explains how prediction error splits into bias squared, variance, and irreducible noise for squared loss.