Most of what politics leaves behind is text, and my methods follow the arc of what we can do with it: count it, classify it, and now interrogate the machines trained on it. The counting-and-classifying line runs through 96,000 Turkish tweets and a million congressional speeches, and the pipelines built to label them. The interrogation line treats the model itself as the object of study: whether its annotations carry the construct, and how it organizes ideology inside.
The two lines meet in measurement & validity — and in the classroom: the AIDE bootcamp, a lecture I called the anthropology of machines, and LLMs for Social Scientists, the book growing out of both.