When working with machine learning models, we care more about how our data performs with the training data than we do about how it performs with instances it has never seen.