第二大腦／知識管理

核心觀念

• 分層架構：將資料分為「原始資料層」（Clipping、個人創作）與「AI 整理層」（知識庫），確保原始筆記不被污染。
• 自動成長：透過排程讓 AI 定期消化新資訊，自動更新知識庫，使知識系統具備生命力，隨時間增長。
• 跨平台同步：利用 GitHub、Google Drive 等雲端服務作為中介，實現 Obsidian、NotebookLM 等多個工具間的資料無縫同步與備份。
• 對話式探索：將知識庫當作可對話的專家，透過向 AI 提問來探索、連結與生成新知識，並將對話本身也納入知識庫。

整合工作流

1. 建立基座：使用 Claude Code 或 Codex 的懶人包，一鍵安裝並設定 Obsidian、GitHub、Google Drive 的連接，建立 Vault（筆記倉庫）。
2. 設定資料結構：在 Obsidian Vault 中建立三層資料夾：Clipping (存放網路剪輯內容)、創作庫 (存放個人原創筆記)、知識庫 (由 AI 整理生成)。
3. 自動化擷取：安裝 Obsidian Web Clipper 瀏覽器擴充，將網頁文章、YouTube 字幕等一鍵存入 Clipping 資料夾。
4. AI 處理與串聯：
   • 將 Clipping 與 創作庫 的內容透過 Claude/Codex 上傳至 NotebookLM 進行初步分析與摘要。
   • 設定排程任務，讓 AI 每週自動讀取新資料，整理成結構化的筆記存回 Obsidian 的 知識庫 資料夾。
   • 指示 AI 將 NotebookLM 的分析報告、Obsidian 的筆記摘要同步存檔至 GitHub，完成版本控制與備份。
5. 查詢與應用：透過 Claude Code 或直接在 Obsidian 中，以自然語言向 AI 詢問知識庫內容，並利用 Obsidian 的關聯圖視覺化知識連結。

最佳金句

「為什麼你的筆記總是沒用？打造會自動成長的 AI 第二大腦！」 「你每一次問問題就在知識庫這邊問問題，你問完問題的資料會再重新丟回知識庫，所以你的知識庫的內容會越來越多、越來越多。」 「超級方便一句話跨三個工具啊，這個就是 web 版做不到的事情。」

教學切入建議

此主題適合 AI 應用課程的中高階部分，學員需對個人知識管理有需求，並具備 Claude/Codex、Obsidian 的基礎操作經驗。教學時可從「解決筆記混亂」的痛點出發，先介紹單一工具（如 Obsidian）的優點，再逐步展示如何透過 AI Agent（Claude/Codex）將其串聯，建立自動化流程。可設計一個專案，讓學員為特定主題（如準備一場考試）建立自己的 AI 第二大腦。

常見誤區

• 過度收集，疏於整理：只專注於用 Web Clipper 擷取資料，但沒有建立後續的 AI 自動化整理流程，導致資訊過載。
• 工具迷思：花太多時間在比較與設定各種新工具，而忽略了知識管理的核心是「思考」與「產出」。
• 權限設定錯誤：在連接 GitHub 或雲端硬碟時，權限設定不當，導致 AI Agent 無法讀寫檔案，流程中斷。

---en---

Core Concepts

• Layered Architecture: Structure data into a “raw data layer” (Clipping, personal creations) and an “AI processed layer” (knowledge base) to ensure raw notes remain untainted.
• Autonomous Growth: Implement scheduled AI tasks to regularly process new information, automatically updating the knowledge base, enabling the system to evolve and expand over time.
• Cross-Platform Synchronization: Utilize cloud services like GitHub and Google Drive as intermediaries for seamless data synchronization and backup between various tools such as Obsidian and NotebookLM.
• Conversational Exploration: Treat the knowledge base as an expert you can converse with, using AI to ask questions, explore connections, generate new insights, and integrate these conversations back into the knowledge base.

Integrated Workflow

1. Establish Foundation: Use Claude Code or Codex “lazy packs” (pre-configured setups) to effortlessly install and configure connections for Obsidian, GitHub, and Google Drive, thereby setting up your Vault (note repository).
2. Define Data Structure: Within the Obsidian Vault, create a three-tiered folder structure: Clipping (for web clippings), Creation Hub (for personal original notes), and Knowledge Base (for AI-processed and generated content).
3. Automated Capture: Install the Obsidian Web Clipper browser extension to instantly save web articles, YouTube transcripts, and other content directly into the Clipping folder with a single click.
4. AI Processing and Connection:
   • Upload content from Clipping and Creation Hub to NotebookLM via Claude/Codex for initial analysis and summarization.
   • Set up scheduled tasks for AI to automatically read new data weekly, organize it into structured notes, and save them back into Obsidian’s Knowledge Base folder.
   • Instruct AI to synchronize NotebookLM analysis reports and Obsidian note summaries to GitHub for version control and backup.
5. Query and Application: Engage Claude Code or directly use Obsidian to query the knowledge base with natural language, and visualize knowledge connections using Obsidian’s graph view.

Best Quotes

“Why are your notes always useless? Build an AI second brain that grows automatically!” “Every time you ask a question in the knowledge base, the data from your questions will be thrown back into the knowledge base, so your knowledge base will grow more and more.” “It’s super convenient to bridge three tools with one command; this is something the web version cannot do.”

Teaching Entry Points

This topic is suitable for the intermediate to advanced sections of AI application courses. Students should have a need for personal knowledge management and basic operational experience with Claude/Codex and Obsidian. When teaching, start from the pain point of “solving messy notes,” introduce the advantages of a single tool (like Obsidian), then gradually demonstrate how to connect them via AI Agents (Claude/Codex) to establish an automated workflow. A project can be designed where students build their own AI second brain for a specific topic (e.g., preparing for an exam).

Common Pitfalls

• Over-collection, under-organization: Focusing solely on capturing data with Web Clipper without establishing subsequent AI automated organization processes, leading to information overload.
• Tool obsession: Spending too much time comparing and setting up various new tools, neglecting that the core of knowledge management is “thinking” and “output.”
• Incorrect permission settings: Improper permission configurations when connecting GitHub or cloud drives, resulting in AI Agents being unable to read/write files and disrupting the workflow.