>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.
I agree that there is a difference between entirely LLM-generated, and LLM-reworded. But the statement is unequivocal to me:
> The Zigbook intentionally contains no AI-generated content—it is hand-written
If an LLM was used in any fashion, then this statement is simply a lie.