 Today, the AI scientists and the data scientists go through a highly iterative process. They try to learn and teach a computer many, many times before they get to the results that they want. And our goal is to make AI algorithms, and in this case, especially deep learning and other machine learning technologies, go as quickly as possible, go as quickly as we can provide through systems design and software innovations. Some of the types of innovations that we're uncovering to help speed up AI technologies could eventually end up into hardware in the future to make systems that are overall better at executing cognitive work. All of those things, the faster you can do them, means the faster that the consumer experiences, more accuracy in their pictures being labeled, more accuracy in speech recognition in the case of noisy environments or if you have an accent. We are able to accelerate the time that it takes to get innovations from researchers' hands into the hands of customers. And so this software is going to be available now to customers on a trial basis. In some sense, the AI over time will grow up. It'll become more intelligent. It will become able to do more things. It'll become higher functioning. And that also increases computational complexity usually. And so we want to be able to handle more data faster and we also want to be able to tackle more challenging AI problems.