Post

It takes substantially longer to read and respond to a series of student essays written with LLMs, than the kind of essays I used to receive, written by students.

Unpicking the knotted bramble of LLM 'discourse', in order to feedback something productive to the student (even if that's just a comment about how I determined their essay was written by AI) is laborious, especially when the essay makes use of hallucinated quotes.

Trying to determine whether a quotation is hallucinated is often a psychotic journey, since the quotes 'feel' right, and are often bric-a-brac compositions of that writer's ideas fused with people the LLM has determined are their contemporaries.

The majority of LLM essays contain no quotations at all, relying on broad sweeping summaries of arguments & concepts, with no relationship to examples, case studies, or actual experience. These are at least easier to pick out.