Yeah, but this is one of the flaws of the study methodology. Many of the sentences being given to the LLM to classify aren't really meaningful on their own so t

Yeah, but this is one of the flaws of the study methodology. Many of the sentences being given to the LLM to classify aren't really meaningful on their own so the LLM/human labeler then invents some secondary assumed meanings that aren't strictly speaking there. It's totally understandable why they do that but then it makes it hard to interpret the results.