不只属于实验室——我用 OpenAaaS 给妻子搭了一个审校助手 / Beyond the Lab: I Built My Wife a Copy Editor with OpenAaaS

[Wolido]

日常审校的难题 / The Challenges of Daily Copy Editing

我的妻子是一名媒体记者。审校稿件是她日常工作的一部分，既包括审自己的稿子，也包括审别人的稿子。

My wife is a journalist. Copy editing is part of her daily work—both her own articles and those written by others.

这个工作并不轻松。常见的问题至少有三类：人名容易写错，有错别字，还有事实错误。

The job is harder than it looks. Common problems fall into at least three categories: misspelled names, typos, and factual errors.

人名错误通常出现在同音字或形近字上。记者写稿时凭记忆输入，事后自己重读一遍往往也发现不了，因为人在阅读自己写的内容时会自动"脑补"成正确的字形。错别字的情况类似，一个逗号位置的偏差、一个量词的误用，单独看可能不起眼，但在正式报道中就是硬伤。事实错误则更难发现，因为字面上看起来没问题，甚至读起来通顺，但内容本身就是错的。有些事实错误涉及数据、时间、地点，有些则涉及行政区划、机构名称、政策范围等需要专门核实的信息。

Name errors usually involve homophones or visually similar characters; when a reporter types from memory, even a self-review often fails to catch them, because people tend to autocorrect what they read in their own writing. Typos behave similarly—a misplaced comma, a misused measure word—each may seem minor in isolation, yet in formal reporting it is a flaw. Factual errors are the hardest to catch because the text reads smoothly and looks fine on the surface, yet the content itself is wrong. Some involve dates, figures, or locations; others concern administrative divisions, institutional names, or policy scopes that must be verified against specific sources.

传统的人工审校依赖编辑的经验和记忆，面对海量稿件时，很难对每一个地名、每一个人名职务、每一项政策细节都逐一核实。传统智能审校方案大多基于规则或通用语言模型，对拼写和语法有一定识别能力，但对需要结合具体政策文件、地理知识、行政划分才能判断的事实错误，识别效果并不理想。

Traditional human copy editing relies on an editor's experience and memory. Faced with a large volume of articles, it is impractical to verify every place name, job title, or policy detail one by one. Conventional AI-assisted proofreading, mostly rule-based or built on general-purpose language models, handles spelling and grammar reasonably well, yet performs poorly on factual mistakes that can only be judged by combining policy documents, geographic knowledge, and administrative classifications.

---

搭一个网页审校助手 / Building a Web-Based Copy Editor

我基于 OpenAaaS 给她搭了一个网页审校助手。用法很简单：打开网页，把稿件粘贴进去，系统自动完成审校。不需要编程背景，也不需要配置复杂的客户端——它就是一个网页。这背后的思路是，OpenAaaS 可以连接各种智能体，实现 Agent as a Service；而客户端的形态完全可以根据场景来设计，不一定非要是编程智能体，一个普通的网页就足够好用。妻子不需要理解什么是智能体，也不需要关心系统后端的连接逻辑，她只需要像使用任何一个普通网站一样，粘贴内容、等待结果、查看修改建议。对用户来说，操作门槛降到了最低，核心能力则通过后端连接的审校智能体来完成。整个流程几乎不需要额外的学习成本。

I built her a web-based copy editor on OpenAaaS. The workflow is simple: open the page, paste the article, and the system reviews it automatically. No programming background is needed, and no complex client setup is required—it is just a web page. The idea behind it is that OpenAaaS can connect to various agents to deliver Agent as a Service, and the form of the client can be designed for the scene at hand: it does not have to be a programming agent; an ordinary web page works perfectly well. She does not need to understand what an agent is or care about the back-end connection logic; she simply pastes content, waits for the result, and reads the revision suggestions, just as she would use any ordinary website. For the user, the barrier to entry is minimized, while the core capability is handled by the copy-editing agent connected at the back end. The whole workflow requires almost no extra learning cost.

---

一次印象深刻的检出 / A Memorable Catch

这个工具在实际使用中检出了一个让我印象深刻的错误。

In actual use, the tool caught an error that stuck with me.

一篇稿件的原文写道："广州、佛山、清远、中山等粤港澳大湾区城市"。系统建议将其改为："广州、佛山、中山等粤港澳大湾区城市及清远等地"，并将此问题标记为"严重"。

An article originally read: "Guangzhou, Foshan, Qingyuan, Zhongshan and other Guangdong–Hong Kong–Macao Greater Bay Area cities." The system suggested revising it to: "Guangzhou, Foshan, Zhongshan and other Guangdong–Hong Kong–Macao Greater Bay Area cities, as well as Qingyuan and other places," and flagged the issue as "severe."

系统的审校说明原文是："本报告基于网络公开信息搜索核实，对涉及行政区划、机构名称、人名职务等关键信息进行了一一比对。稿件整体事实准确、表述规范，但存在1处涉及国家重大区域发展战略的地理范围表述错误，需修改后方可发布。"

The review note stated: "This report was produced by searching and verifying publicly available online information, with key details such as administrative divisions, institutional names, and personal titles checked one by one. The article is generally factually accurate and properly worded, but it contains one geographic scope error related to a major national regional development strategy and must be corrected before publication."

这个错误的隐蔽性在于：清远地理上接近粤港澳大湾区，但行政上并不在《粤港澳大湾区发展规划纲要》所确定的 9+2 城市范围内。把清远直接列为"粤港澳大湾区城市"，属于事实性错误。对普通读者来说，这句话读起来没有任何异常；对忙碌的人工审校编辑来说，逐一核对每个城市是否属于大湾区范围，工作量巨大，且容易因为"看起来合理"而放过。传统的智能审校方案同样难以精准识别这类问题，因为它不仅需要语言理解能力，还需要检索和比对特定的政策文件内容，判断行政区划的归属。

What made the error hard to spot is that Qingyuan is geographically close to the Greater Bay Area, yet administratively it is not among the 9+2 cities defined in the Outline Development Plan for the Guangdong–Hong Kong–Macao Greater Bay Area. Listing Qingyuan directly as a "Greater Bay Area city" is therefore a factual mistake. To an ordinary reader, the sentence raises no alarm. To a busy human copy editor, checking every city against the official list is a heavy burden, and the claim is easy to let through simply because it sounds reasonable. Conventional AI proofreading tools face the same limitation: catching this error requires not only language understanding, but also the ability to retrieve and compare specific policy documents to determine administrative membership.

---

不只属于实验室 / Beyond the Lab

这个案例让我重新认识了 OpenAaaS 的能力边界。它不限于科研场景。科研确实是 OpenAaaS 的一个重要应用场景，但它的本质是一种可以连接各类智能体、按需提供智能服务的架构。客户端可以是编程智能体，也可以就是一个网页，用户直接粘贴内容就能完成工作。只要场景需要调用智能能力来完成具体任务，OpenAaaS 都可以接入。一个媒体记者的审校需求，和一个研究人员的计算需求，在底层逻辑上是相似的：都是让合适的能力在合适的场景里发挥作用，而接口形式完全可以根据使用者的情况来调整。

This experience changed how I see the scope of OpenAaaS. It is not limited to research settings. Research is an important use case for OpenAaaS, yet its essence is an architecture that can connect diverse agents and deliver intelligence on demand. The client may be a programming agent, or it may simply be a web page where users paste content and get the job done. As long as a scene calls on external intelligence to accomplish a concrete task, OpenAaaS can be plugged in. A journalist's copy-editing need and a researcher's computing need are alike at the bottom: both put the right capability in the right context, and the interface form can be tailored to the user.

从这个意义上说，审校助手只是一个开始。任何需要结合外部信息核实、需要跨领域知识比对、需要自动化处理复杂内容的领域，都可以使用类似的思路来搭建工具。媒体出版、法律咨询、金融合规、医疗文献整理——这些领域的共同点并不是它们都用到了最前沿的科研技术，而是它们都需要把专业判断能力嵌入到日常的工作流里，都需要有人去做大量重复但关键的核实工作。OpenAaaS 的价值不在于它只能在实验室里跑科研任务，而在于它能把智能体的能力以最简单的方式交到需要的人手中。

In this sense, the copy editor is only a beginning. Any domain that requires verification against external information, cross-domain knowledge comparison, or automated handling of complex content can adopt a similar approach. Media publishing, legal consulting, financial compliance, medical literature review—these fields do not share the same cutting-edge research technology, but they do share a common need: embedding professional judgment into daily workflows, and having someone perform large amounts of repetitive yet critical verification work. The value of OpenAaaS lies not in running research tasks inside a lab, but in putting agent capabilities into the hands of those who need them, in the simplest way possible.

任何领域都可以使用。

Any field can use it.