The scale and quality of the data set used for training is often overlooked, but increasing these factors can lead to greater breakthroughs in deep learning and surpassing human level performance.