5步教你用 WorkBuddy × ima 打造可进化的第二大脑5 Steps to Build an Evolving Second Brain with WorkBuddy × ima


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/uBWzLauBPwVs7mQRZ0JyNQ
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

之前分享过我用 ima 打造个人智能助手的故事,没了解的朋友不妨回顾一下:

作为26岁的独角兽公司创始人,我借助ima构建了个人智能大脑系统:

这款工具确实能通过知识库帮你检索信息、提供答案,能节省大量时间。不过实际使用中我发现,单纯搭建好知识库还远远不够。

智能系统需要不断进化迭代。你输入的资料可能会过时,但它无法自动识别。比如你的知识库全是6月份的内容,没有7月份的新文章,当你询问最近写了什么时,它仍会基于旧数据给出答复。

观察仔细的人很快就会发现,这种回答明显不符合你最新的实际情况。

 ㅤ

真正的问题不在于ima系统本身,而在于我自身。知识库我只是从中提取信息,却从未向其回流最新内容。

建完就完事,从来不把我的过程物投喂回去,所以它越用越笨。说白了知识库是要定期维护的,我之前也没真理解它的定义。

知识库的学名,叫 RAG(检索增强生成),说白了就是给 AI 接上你的资料库,答题前先查你的资料。但光懂这个还不够,关键在下面这张图。

Image

今天把当下用了一个多月,用 WorkBuddy(了解更多) × ima 构建可持续进化第二大脑的方法论分享给你,当下用下来觉得还不错,希望对你有协助。

01 建立个人知识库

ima 知识库分三类:个人知识库、共享知识库、订阅知识库。咱们一人公司用,主要是个人知识库。

5步教你用 WorkBuddy × ima 打造可进化的第二大脑

它兼容的格式挺全的,本地文件、文件夹、微信文件、网页链接、笔记、腾讯文档、录音这些都能直接上传,不用你重新整理一遍。

Image

我自己建了一个阿坚知识库,下方依据账号与用途划分了几个子文件夹,具体的分类方式仅供参考,你可以根据自己的使用习惯灵活调整。

Image

02 构建知识库目录

如果知识库文章数量庞大,且未预先建立索引,AI 每次都需要重复执行相同的检索任务,从头开始处理,这会导致效率低下并消耗大量算力。因此,可以提前创建索引,这种做法同样符合AI的工作逻辑。

Image

关于如何构建索引,你可以参考我接下来提供的方法说明。

“请你将个人知识库-阿坚知识库-阿坚AIGC(已发)中的文章,按照发布时间降序排列并建立索引,以md格式输出。同时,每一篇具体文章也要以md格式提供,即索引和文章均以md格式呈现,最终打包成压缩文件并提供下载链接。”
Image

执行ima任务后生成一个下载链接,点击后可获取压缩包,其中包含索引文件及每篇文章的md文档。后续若有新增文章,只需让其按相同逻辑进行增量更新,无需重新处理全部内容。

Image

索引我也会让它以笔记的形式保存在知识库中,方便随时调取对照。导出的这些 Markdown 文件会存入内容 OS 的 10_已发归档目录,与文章文件放在一起。

Image
Image

正如之前提到的:一项任务如果需要重复3次以上,就值得将其规范化。每次都要重新向 ima 说明如何建立索引、如何按相同逻辑增量更新,而且它未必能记住之前的处理方式。将其做成 Skill 后,下次可直接调用,无需重复操作。

03 将内容回流至知识库ㅤ

许多人实际上没有这个意识:今天的工作虽然完成了,但过程中的产出成果既没有回流到知识库,也没有沉淀为组织记忆。以今天的工作为例,最终的交付成果就是这篇公众号文章。如果不将其回流到 ima 知识库,那么知识库就无法得到进化。因此,作者的所有过程性产出都必须回流存档。

 ㅤ

所谓回流,其实就是:你今天完成的稿件、产出的思考、遇到的问题,不要让它们留在本地电脑的文件夹里积灰。应当将这些内容回传至 ima 对应的知识库中。

 ㅤ

回流有几种方式。

手动进行回流操作:一篇文章写完后,点击右上角的“上传到云端”,选择 ima 知识库并确认,整个过程仅需几秒钟。

Image
Image

让 WorkBuddy(了解更多) 自动完成回流任务。你只需直接对 WorkBuddy(了解更多) 指令“把今天的过程物料上传至 ima 的指定知识库”,由于 WorkBuddy(了解更多) 已连接 ima 知识库,可以执行此操作。当它反馈已回流成功时,记得去 ima 知识库核实。

 ㅤ

构建 Skill 并调用其回传功能。同时,你可以设计一个归档 Skill,每次任务完成后只需说“归档”,它便会自动识别正确的回传路径。

 ㅤ

设定自动化任务,实现定时回传。我配置了一个定时任务,定期检查是否有遗漏的产出文件,包括记忆体和文档,一旦发现便会自动上传至对应的 ima 知识库。

这样一来,你每日的产出物都会保存在知识库中,而知识库将随着时间的推移不断进化。

Image

04 ima分析

知识库有文件了,分析谁来做?我倾向在 ima 里做。

原因很简单,ima 自带的 Copilot 写出来的东西文风更合适,不像 WorkBuddy(了解更多) 直接调底层那种容易跑偏。所以要分析语气、要生成那种我自己的文 vs 对标的文的对照表,让 ima 自己做就行。

Image

这一张表怎么出。我把自己最近一周写过的稿子全丢进 ima 的同一个知识库,再丢几篇我订阅的几个头部公众号近期文章进去,让 Copilot 出三样东西。

 ㅤ

一,我哪些段落写得不错,风格和头部一致。

二,有哪些表达是头部有而我未使用的。

三,我哪些固有的写作方式实际上一直不恰当。

Image

出完这份表格并非终点。每次撰写新稿件前,我都会取出这份表格,参照着进行创作。这样,哪些要点未领会或偏离了方向,就能迅速发现。

Image

当所有过程资料都汇总到位后,我会固定使用它来帮助自己梳理:如果以标杆为参照来写这篇文章,应该如何构思。我的语气、公众号选题都可以从对标中学习。说到底这也是我持续学习的一个过程。

仅这一遍还不足够。我每周会将公众号已发表的内容也上传,让 Copilot 基于之前的索引和分析报告进行增量更新,检查这周的风格是否比上周有进步,再同步回 ima,也同步回本地。

Image

05 知识库不断迭代

如果你一直坚持执行前述的四个步骤,你会逐渐发现一种感觉:它对你越来越了解。

 ㅤ

这实际上正是桌面 Agent 最显著的优点。你之前进行的建库、索引、回流、分析等工作,本质上就是在为 WorkBuddy(了解更多) 提供一个持续认识你的档案。每一次的回流与分析,都会让它更加了解你,进而下次为你提供的交付结果也会更贴近你。

 ㅤ

但归根结底,它只是一个流程。真正让它在这个过程中持续进化的,是你自己持续不断地去实践、去探索。因为说实话,对于所有写作者来说,终究还是需要自己去分析:今天的选题是否受欢迎,文章的语气该如何掌握,针对性地去弥补不足。它为你提供一面镜子,但成长这件事终究得靠自己。

Image

 ㅤ

06 我的内容OS

有些读者看完前面的内容后可能会想搭建一套这样的系统,但不知道从何处开始。我绘制了一张自己工作区分工的示意图,仅供参考,并非标准模板。具体的每个目录用途我就不一一列出了,有兴趣深入了解的可以参考图表。

Image

 ㅤ

写在最后ㅤ

这套工作流是我实际使用的,并非仅仅停留在概念层面,也不是在介绍某个具体工具。工作流听起来可能很简单,但要真正见到效果,需要依靠日常的坚持执行和贯彻。

如果你想搭建一套系统,建议先将你当前使用的内容,按照这套方法构建一个能够不断更新的知识库。

 ㅤ

最后,用一张图表来概括整篇文章的内容。

Image

 ㅤ

 - END -

我是阿坚,公众号拥有超过2万粉丝,目前全职从事公众号运营和AI写作工作,正在打造一人公司,并持续分享关于AI、WorkBuddy(了解更多)、一人公司等领域的实操经验。

Source: https://mp.weixin.qq.com/s/uBWzLauBPwVs7mQRZ0JyNQ. For copyright issues, contact: 13951431913#139.com or leave a comment on the right.

Previously, I shared my experience of creating a personal intelligent assistant using ima. If you haven't read it, I recommend checking it out:

As a 26-year-old founder of a unicorn company, I used ima to build my personal intelligent brain system:

This tool can indeed search for information and provide answers based on your knowledge base, saving you a lot of time. However, in actual use, I found that simply setting up the knowledge base is far from enough.

Intelligent systems need continuous evolution and iteration. The data you input may become outdated, and the system cannot automatically identify this. For example, if your knowledge base only contains content from June and none from July, when you ask what you wrote recently, it will still respond based on the old data.

Observant people quickly realize that these answers do not match their current reality.

ㅤ

The real problem is not with the ima system itself, but with me. I extract information from the knowledge base but never feed the latest content back into it.

I just set it up and forgot about it, never adding new material. This is why the system becomes dumber over time. In fact, a knowledge base requires regular maintenance; I didn't fully understand its definition before.

The technical term for a knowledge base is RAG (Retrieval-Augmented Generation), which means connecting AI to your data repository so it can search your data before answering. But knowing this isn't enough. The key is shown in the image below.

Image

Today, I'm sharing my methodology for building a sustainable, evolving second brain using WorkBuddy × ima, which I've been using for over a month and find quite effective. I hope it helps you.

01 Building a Personal Knowledge Base

There are three types of ima knowledge bases: personal, shared, and subscribed. For a one-person company, we primarily use the personal knowledge base.

5 Steps to Build an Evolving Second Brain with WorkBuddy × ima

It supports many formats, including local files, folders, WeChat files, web links, notes, Tencent documents, and recordings. You can upload them directly without reformatting.

Image

I created a "Jiang Knowledge Base" and organized it into several sub-folders based on account and purpose. This specific classification is just for reference; adjust it according to your preferences.

Image

02 Constructing the Knowledge Base Directory

If your knowledge base contains many articles without a pre-built index, AI must scan the entire repository every time, which is inefficient and consumes resources. It's better to create an index first.

Image

To build the index, follow this prompt:

"Please sort the articles in 'Personal Knowledge Base - Jiang Knowledge Base - Jiang AIGC (Published)' by publication date in descending order and create an index in .md format. Also, output each individual article as a .md file, and package the index and articles into a compressed file with a download link."

Image

After executing the ima task, you'll get a download link. The compressed file will contain an index file and the .md articles. If new articles are added, just apply the same logic to update the index incrementally.

Image

I store the index in the knowledge base as a note for easy reference. The exported Markdown files are saved in the '10_Published Archives' directory of the Content OS, alongside the article files.

Image

Image

As mentioned earlier, if a task needs to be repeated more than three times, it should be standardized. Instead of manually asking ima to build the index every time, create a Skill for it so you can call it directly.

03 Returning Content to the Knowledge Base

Many people forget that today's work products (drafts, thoughts, problems) need to be archived. If you don't return them to the ima knowledge base, the system won't evolve. Everything I produce must be stored.

ㅤ

Returning means uploading your work to the ima knowledge base instead of leaving it on your local hard drive.

ㅤ

There are several ways to do this: - Manually upload by clicking "Upload to Cloud" and selecting the ima knowledge base. - Use WorkBuddy to automate the upload (it's connected to ima). - Build a Skill for archiving that automatically identifies and uploads new content. - Set up a scheduled task to periodically check for and upload missing files.

Image

Image

Image

04 ima Analysis

Once the files are in the knowledge base, you can analyze them. I prefer to do this in ima because the Copilot generates text that matches my style better than direct API calls.

Image

To get a comparison table, I upload the last week's articles and a few top-performing articles from other accounts, then ask Copilot to analyze three things: - Which paragraphs of mine match the top performers? - What expressions do they use that I don't? - Which of my habitual writing styles are incorrect?

Image

This isn't a one-time thing. Before writing new articles, I review this table to adjust my tone and content. I also update the analysis weekly to track progress.

Image

Image

05 Continuous Iteration of the Knowledge Base

If you consistently follow these steps, you'll notice that the system gets to know you better over time.

Image

This is the strength of a desktop agent. By returning data and analyzing it, you're creating a profile for WorkBuddy, allowing it to provide more personalized results.

ㅤ

However, it's still just a process. The real improvement comes from your own effort to practice and explore based on the feedback. The agent is a mirror; you must do the work to grow.

Image

ㅤ

Image

For those who want to build a similar system but don't know where to start, here is a diagram of my work divisions. It's not a strict template but rather an idea of how to organize your workflow.

Image

ㅤ

Conclusion

Image

ㅤ

Here is a summary of the entire process:

Image

ㅤ

- END -

I am Jiang, a full-time public account operator and AI writer with over 20,000 followers. I'm working on a one-person company and regularly share my experiences with AI, WorkBuddy, and solo entrepreneurship.

Image

ㅤ

06 我的内容OS

Some readers may want to build a similar system but don’t know where to start. I’ve drawn a diagram of my personal workflow to show how I organize things — this is not a standard template, just my setup. I won’t list every directory’s purpose; those interested can reference the chart for details.

Image

ㅤ

写在最后ㅤ

This workflow is actively used in my work and isn’t just theoretical. While the idea seems simple, real results come from consistent execution and commitment to the process.

If you’re considering setting up something similar, start by organizing your current content into a continuously updatable knowledge base using this method.

ㅤ

Here’s a summary chart capturing the entire article’s content.

Image

ㅤ

- END -

I’m 阿坚 (Ah Jian), with 20K+ followers on WeChat, full-time in公众号 operation and AI writing. Building a one-person company while sharing practical insights on AI, WorkBuddy (learn more), and solo entrepreneurship.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry