📄 中文摘要
实验中，研究人员向六个大型语言模型（LLMs）提供了相同的提示，要求其撰写一篇约800个字符的技术博客文章。参与的模型包括商业模型Claude Sonnet 4和GPT-4o，以及开源模型Qwen 3.5-4B和Qwen 3。通过对生成的文章进行评估，研究发现人类撰写的文章在质量上显著优于AI生成的内容，结果与预期存在较大差异。这一发现引发了对AI生成内容质量的深入思考，尤其是在技术领域的应用潜力和局限性方面。
我们测量了180篇AI生成的日本文章，结果出乎意料
📄 English Summary
We Measured 180 AI-Generated Japanese Articles. The Results Were Not What We Expected.
In the experiment, researchers provided the same prompt to six large language models (LLMs), asking them to write a technical blog article of approximately 800 characters. The models included commercial ones like Claude Sonnet 4 and GPT-4o, as well as open-source models such as Qwen 3.5-4B and Qwen 3. The evaluation of the generated articles revealed that human-written content significantly outperformed AI-generated articles in quality, leading to results that differed greatly from initial expectations. This finding raises important considerations regarding the quality of AI-generated content, particularly in its potential applications and limitations within the technical domain.