The broad phenomenon where model performance improves predictably with increases in parameters, data, or compute. Scaling laws formalize these relationships mathematically.
See also: scaling_laws, chinchilla_scaling, attention_residuals