The document summarizes Jake Hofman's presentation on learning from web activity data. It discusses a case study analyzing demographic diversity on the web using anonymized browsing data from 265,000 US users. The study found some sites have highly skewed audiences along attributes like gender, race, education and income, though most sites' audiences more closely match the general population. It also compared demographic skew between online audiences and offline neighborhoods, finding websites generally have more racially diverse audiences than neighborhoods have residents. The presentation explored differences in how demographic groups distribute their time online.