Machine learning models are characterized by a rich variety of parameters that impart them their functional versatility of modeling vast domains of data with minimal handcrafted features. These models obviate the need for researchers to craft a different set of features, adjust and recalibrate the model, and reexamine the functional adequacy of the underlying model in the face of evolving data. However, this versatility often comes at a price – the pitfall of overfitting on training datasets but poor performance on test datasets. Notwithstanding the fact that some machine learning models, such as random forests, are not prone to overfitting, the vast majority of machine learning models featuring deep neural networks must be scrupulously tested to make sure they are not overfitting. This is particularly true of some of the large language models that have billions of free parameters.