Category
Level
Scaling laws say that model loss typically follows a power law that improves predictably as you increase parameters, data, or compute.