开发者Dodaios在Hacker News社区发布了名为STE80 Humanizer的新工具,旨在通过消除人工智能写作痕迹并应用ASD-STE100简化技术英语标准,使生成的文本更接近人类书写风格且更具清晰度 1。该工具的核心理念是提升文本的可读性而非单纯规避检测器,这一设计灵感来源于Andrej Karpathy于2026年10月2日提出的关于大语言模型输出可读性的建议 1。
在具体实现上,STE80 Humanizer采取了双重处理机制:首先移除维基百科“AI写作迹象”列表中列出的26个典型特征,随后重构句子以符合“80%的ASD-STE100”标准 1。其所采用的ASD-STE100规则起源于20世纪80年代的航空航天行业,主要限制包括每句不超过20个单词、使用主动语态以及坚持一词一义(基于约900个批准词汇) 1。
该工具支持多种编程代理环境,用户可通过npx skills add Dodaios/ste80-humanizer命令或在插件市场中添加进行安装 1。为了辅助合规性检查,项目还提供了scripts/ste_check.py脚本,用于标记未遵守Pass 1和Pass 2规则的文本内容 1。需要注意的是,该项目明确声明未获得欧洲航空航天安全与国防工业协会(ASD)的认可,且ASD-STE100为该协会的商标 1。
A new tool named STE80 Humanizer has been released on Hacker News, designed to make AI-generated text appear more human-written and clearer. The tool achieves this by removing artificial writing traits and adopting the principles of ASD-STE100 Simplified Technical English 1. Its core methodology involves eliminating 26 specific features identified in Wikipedia’s "Signs of AI writing" list and restructuring sentences to reach an "80% ASD-STE100" compliance level 1. This approach draws inspiration from advice provided by Andrej Karpathy regarding the readability of large language model outputs, which was posted on October 2, 2026 1.
The underlying ASD-STE100 rules originate from the aerospace industry in the 1980s. They mandate strict constraints such as a maximum of 20 words per sentence, the exclusive use of active voice, and a one-word-one-meaning policy limited to approximately 900 approved vocabulary items 1. Users can install the tool via the command npx skills add Dodaios/ste80-humanizer or through plugin marketplaces. The tool supports various programming agent environments 1. To assist with quality control, the package includes a script named scripts/ste_check.py that flags non-compliance with Pass 1 and Pass 2 style rules 1. The developers explicitly state that the project is not endorsed by the Aerospace and Defence Industries Association of Europe (ASD), noting that ASD-STE100 remains their trademark 1.
评论
还没有评论,欢迎留下第一条。