研究揭示让AI文本读起来费力的六种语言模式
原标题：Researchers find six patterns that make AI text so exhausting to read
AI 摘要
纽约大学与得克萨斯大学奥斯汀分校的研究分析了14个语言模型（包括Claude、GPT-5.6、GLM-5、Kimi K2.5及Llama、Olmo等开源模型）在对话、编程和议论文场景下生成的文本，总结出六种让AI文本读起来费力的语言模式，如异常词组搭配、句法成分距离过远、名词堆叠、名词动用等。研究发现小模型更接近人类写作，而训练从RLHF转向RLVR、奖励信息密度而非清晰度，可能是导致前沿模型文风不自然的原因。研究还通过惩罚项实验和Anthropic员工的说法佐证了这一机制，并指出针对性微调（如Opus 5.5）可改善可读性。
正文节选
Researchers find six patterns that make AI text so exhausting to read A study from New York University and the University of Texas at Austin shows that frontier language models systematically overuse linguistic patterns that make their output harder to read. Anyone who regularly works with ChatGPT, Claude, or other language models knows the feeling. The answers are often factually correct, but reading them feels like work, full of nested sentences, odd word combinations, and metaphors no human w