Study identifies six language patterns that make AI text harder to read

Researchers at NYU and UT Austin compared 120 dialogues, 45 coding sessions and 980 articles generated by 14 language models against human text, The Decoder reported on October 27. They found the models overuse six language structures that raise the burden on human readers. The patterns include rare word combinations, sentence connectives placed too far apart, dropped conjunctions, noun stacking, nouns turned into verbs, and animate action verbs applied to inanimate subjects. The report so far covers only the identification of these patterns and the reading burden they create.

Written by AI from the coverage · updated 45 minutes ago

Not enough continuous observations to draw a trend yet.