Comment by mrob

>If an LLM was used in any fashion, then this statement is simply a lie.

While I don't believe the article was created this way, it's possible to use an LLM purely as a classifier. E.g. prompt along the lines of "Does this paragraph contain any errors? Answer only yes or no." and generate only a single set of token probabilities, without any autoregression. Flag any paragraphs with sufficient probability of "yes" for human review.