A new study reveals that AI-generated text has become increasingly difficult for humans to comprehend. Researchers investigated psycholinguistic features linked to processing difficulty, comparing AI output against human writing and round-trip translations. The findings show that top AI models consistently use more inanimate subjects with animate verbs and low-frequency n-grams.
Many models also employ longer syntactic dependencies and stacked nouns. The research suggests this misalignment occurs because AI is optimized for goals like math, code, and reasoning — prioritizing information density over successful communication. To test this, researchers fine-tuned a model under a length penalty, observing a shift toward more human-like behavior.
The study highlights a growing gap between how humans process language and how large language models generate it.
Source: Hacker News · Summarized by HeadlinesBriefing