Study: why AI-generated text is harder to understand

A new arXiv paper uses psycholinguistic features to show all nine frontier and near-frontier models use more inanimate subjects with animate verbs and low-frequency n-grams than human writing, and some use longer dependencies. The authors argue AI text is optimized for math, code and reasoning rather than communication, testing this with a length-penalty fine-tuning experiment.

• Researcher Without Institution Seeks arXiv Endorsement for VELM Paper2026-10-15
• Event2Vec: additive event embedding library released2026-10-15
• Researcher says Google Scholar citations lag for months2026-10-12
• ArXiv Caps Submissions at Two per Month2026-10-11
• arXiv limits submissions to two per month2026-10-02