A lot of research stalls at the same point: thousands of posts, letters or survey answers, and a question that needs every one of them labelled.
Language models can do the labelling at scale. Without proper validation, there's no way to know whether to trust what comes back.
