Scaling laws
Definition: Scaling laws describe how a model's performance improves predictably as data, parameters and compute increase.
They guided the race toward ever-larger models. They remain empirical and do not promise that scaling alone solves everything.