LLM 写文档可读性差的根源：追求简洁而非易懂
Why we picked it
HF 的 Merve 吐槽 LLM 写文档只顾简短不管好不好懂，连 ASD-STE100 简化英语提示都救不了，写文档的人看了会点头。
作者 Merve Noyan 认为 LLM 写出的技术文档普遍可读性差，因为模型优化的是简短有力的表达，而不是读者理解的流畅度。她提到即使要求模型按 ASD-STE100 简化技术英语规范写作，效果依然不理想。她还指出读取文档的 agent 常在指代关系上产生混淆，而它们本身大部分场景已经在读代码了。
Original · Merve Noyan
hear me out. no matter what, LLMs write horrible docs because they don't aim for readability they aim for being punchy
even if you prompt the model to write with AST-STE100 simplified english it just doesn't work
I'm pretty sure even your agents who read the docs get confused over coreferences and they already read the code for the most of it anyway