Detecting Patterns of Rhetoric

Some styles of writing can be recognised by patterns of similarity, even when different words are used. Think of physical actions, where any one kick or wave might be different, but the general characteristics of kicking or waving can be learned, given enough training data. Microsoft's Kinect was trained in part by CGI-generating a huge number of synthetic examples of kicks, waves and other gestures. Your task is to make a detector, not for physical gestures, but for certain ways of writing, by training a detector using huge amounts of synthetic writing generated by an LLM. Wouldn’t it be amazing if we had an automatic bullshit detector, which could be trained to detect human BS by showing it enough examples of artificial BS from ChatGPT?