LLMs "declines sharply for modal particles and function words whose meanings depend heavily on discourse-level context or manuscript-specific philosophical styles", often down to 50%. Not sure, whether this is bad or good news for us non-artificial readers at this point... (2/2)