 I was thinking from the point of view of what I experienced during the bombings, which is if I was, as I was reading these stories on Twitter, I would have liked to have a tool that would have given me a sense of how likely it is for what I'm reading to be true or false just so that I can sort through all the data and at least make sense of some of the chaos out there. And as I worked on my PhD, I realized that there are three categories of features that are very predictive of veracity of information and rumors on Twitter. One category is basically looking at the people who are spreading the information, the rumors. The other category is looking at the linguistic content of what is actually being said. And then the third category of features, which turned out to be the most predictive, is actually the propagation features. How is it that the rumor is spreading? It turns out that false anti rumors have very distinct propagation characteristics. And these characteristics could be used by our machine learning algorithms to distinguish, to basically separate false and true news. It turned out to be 75% accurate. This is so if you randomly guess the veracity of a rumor, you would get 50% accuracy. So we do 25% better than random chance. Not perfect, but that's still removing a lot of the confusion and chaos in the data for the consumers of news. And this is predictions made before anyone has verified whether the information that's spreading is actually true or false, which is pretty fantastic. And what we found through recent analysis, and as Sinan mentioned, was very comprehensive analysis because we had unprecedented access to the Twitter firehose and the Twitter historical archives. We could go back to the beginning of basically the first tweet made in late 2006, all the way to the end of 2016 when we finished the study and look at all rumor cascades that we could find on Twitter and do this study very comprehensively. What we found was that false news spreads further, faster and broader than true news in every category of information. And this was really pronounced for political rumors actually. So if you look at political rumors as their own category, the difference is even more pronounced. And this was really interesting, something that I didn't personally expect.