产品与商业 3.0 · 值得看 2026-07-30 · 文章

The Economist: How to Spot AI Writing

经济学人拿自己的稿子训了四个 LLM,发现 AI 写作的指纹和你以为的不一样研究方法很扎实(55,940 句120 万词四个主流模型跨三家媒体做基线),结论有几条反直觉比如最新版本后只有 Claude 用 em-dash 比人多,ChatGPT 用得反而最少AI 偏拉丁词缀长句低标点少引语;和 AI 满篇 em-dash 的民间印象相反,识别 AI 写作更好的办法是看几乎没有标点的段落,而不是找破折号

打开原文回到归档

The Economist: How to Spot AI Writing

中文摘要

经济学人拿自己的稿子训了四个 LLM,发现 AI 写作的指纹和你以为的不一样研究方法很扎实(55,940 句120 万词四个主流模型跨三家媒体做基线),结论有几条反直觉比如最新版本后只有 Claude 用 em-dash 比人多,ChatGPT 用得反而最少AI 偏拉丁词缀长句低标点少引语;和 AI 满篇 em-dash 的民间印象相反,识别 AI 写作更好的办法是看几乎没有标点的段落,而不是找破折号

English Summary

The Economist trains four LLMs on its own copy to find AI-writing fingerprints - and several findings cut against folk wisdom. Methodology is solid: 55,940 sentences, 1.2M words, four frontier models, cross-benchmarked against three other publications. Counter-intuitive result: only Claude uses em-dashes more than humans in its latest version; ChatGPT actually uses them least. AI prose skews toward Latinate morphology, long sentences, low punctuation density, and few quotes. Better tell than em-dashes, the article argues, is a near-punctuationless paragraph.

来源 / Obsidian 引用

  • 本机 Obsidian 同日 digest 落地(ClawFeed 24h / AK-RSS-Digest 89 源 / X-Hot-Brief / 内容选题编排)
  • 评分依据:原文正文(不靠标题/摘要/源声誉)