A scaling law is an observed relationship between model performance and factors such as model size, dataset size, compute, or training duration.
Scaling laws help AI teams estimate tradeoffs, plan training runs, and understand when more data or compute may improve results.