Why you should take notes if you use AI

Key Concepts

• Context Engineering: The practice of providing LLMs with relevant information (notes, documents, frameworks) to improve output quality and enable deeper thinking.
• Tacit Knowledge vs. Explicit Knowledge: Tacit knowledge is internal, difficult to articulate; explicit knowledge is documented and shareable. Converting tacit to explicit knowledge is crucial for effective AI interaction.
• Prompt Engineering vs. Context Engineering: Prompt engineering focuses on crafting effective instructions; context engineering focuses on providing the LLM with the necessary background information. The trend is shifting towards the latter.
• Source of Truth: Designated documents or information that the LLM should prioritize over general knowledge.
• Frameworks for Judgment: Explicit criteria or models used to evaluate the quality of LLM outputs.
• LLM Token Limit: The maximum amount of text an LLM can process at once (mentioned as 39,000 tokens in the example).

The Critical Role of Note-Taking for AI Utilization

The core argument presented is that effective use of AI, particularly Large Language Models (LLMs) like ChatGPT, Claude, and Gemini, requires a robust note-taking system. The speaker emphasizes a shift from solely focusing on “prompt engineering” – crafting the perfect instructions – to “context engineering” – providing the AI with the right information to work with. Having used these tools for over two years, the speaker asserts that the quality of both the AI’s output and the user’s own thinking are dramatically improved with a well-maintained knowledge base. This isn’t about outsourcing thought, but about augmenting it.

From Tacit to Explicit Knowledge: The Evolution of Communication

The speaker draws a parallel to the historical development of communication. Initially, knowledge resided solely within the individual (“tacit knowledge”). The invention of writing allowed for the externalization of this knowledge (“explicit knowledge”), enabling collaboration and progress. Now, AI represents a third layer in this communication process. Effective communication with AI necessitates translating our internal understanding into a format the AI can access and utilize. The purpose of this communication, particularly for knowledge workers, isn’t about performance but about sharing tacit knowledge to facilitate collective action with minimal friction.

Building a Context-Rich Note-Taking System: Six Key Categories

The speaker outlines six categories of information to include in notes for optimal context engineering, drawing directly from principles applicable to AI communication:

1. Role: Defining the persona or perspective the LLM should adopt.
2. Goal: Clearly stating the desired outcome of the interaction.
3. Audience: Specifying who the output is intended for.
4. Constraints: Setting boundaries regarding style, format, or other parameters.
5. Input: Defining the data sources the LLM can access (e.g., an entire Obsidian vault, excluding specific tags). This includes specifying a “source of truth” – documents that should override general knowledge.
6. Judgment: Establishing criteria for evaluating the quality of the LLM’s output. This is where frameworks are crucial.

Leveraging Frameworks for Improved LLM Output

The speaker highlights the importance of frameworks for both guiding the LLM and evaluating its results. LLMs often default to generic frameworks (e.g., heading structures, “it is not this, it is that” comparisons). However, providing a specific framework – whether a well-defined model like a copywriting formula (e.g., AIDA) or a set of examples demonstrating desired qualities – significantly improves output quality. If a pre-existing framework isn’t available, the LLM can derive a framework from a collection of “good” and “bad” examples, demonstrating its ability to distill patterns. This process allows users to consciously inject their preferred criteria into the AI interaction.

Case Study: Multi-Passionate Individuals and Career Path Exploration

The speaker provides a personal example to illustrate the power of context engineering. They used Claude, connected to their Obsidian vault (containing approximately 39,000 tokens of notes), to explore a common challenge: being multi-passionate and feeling scattered.

• Without Context (ChatGPT): A generic list of potential career paths was generated, lacking specific relevance to the speaker’s unique interests. The output felt uninspired and difficult to prioritize.
• With Context (Claude & Obsidian): Claude, accessing the speaker’s notes, identified a core theme: the rejection of the false dichotomy between logic and creativity. It articulated this better than the speaker could initially, highlighting the importance of frameworks enabling creativity and viewing creativity as a proxy for intelligence. The AI then suggested relevant resources and even proposed a career path focused on bridging the gap between theory and practice in knowledge work.

This example demonstrates how a note-taking system allows the LLM to engage in a more intelligent and personalized conversation, leading to deeper insights and more actionable recommendations. The speaker emphasizes that the value isn’t just in the AI’s output, but in the process of having the AI synthesize and articulate their own thoughts.

The Risk of “Dumbing Down” vs. Enhanced Thinking

The speaker acknowledges the concern that AI can lead to intellectual laziness. However, they argue that this outcome isn’t inevitable. The key is to actively download context into the AI, rather than simply consuming its output. A robust note-taking system facilitates this process, allowing for more engaging and productive interactions. If you feel like AI is making you “dumber,” the speaker suggests you’re likely not providing it with enough context.

Notable Quotes

• “Everyone’s complaining that you know we have to outsource thinking. It doesn’t have to be that way. But if you don’t have notes then yeah there’s no choice.”
• “Communication is not about performance… but to get the tacit knowledge into explicit knowledge and to share with other people.”
• “What is holding us back from being more engaged in the thinking process is have we downloaded the context out from our brain and into the LLM.”
• “My incognito conversation is never going to get me to this level.”

Technical Terms

• LLM (Large Language Model): A type of artificial intelligence that uses deep learning to understand and generate human language. Examples include ChatGPT, Claude, and Gemini.
• Obsidian: A popular note-taking application that uses a “networked thought” approach, allowing for interconnected notes and knowledge management.
• Tokens: The basic units of text that LLMs process. LLMs have a limited token window, restricting the amount of text they can handle at once.
• Via Negativa: A method of defining something by stating what it is not.
• Maslow's Hierarchy of Needs: A motivational theory in psychology comprising a hierarchy of five innate human needs, often depicted as hierarchical pyramids.

Conclusion

The video strongly advocates for the adoption of a comprehensive note-taking system as a prerequisite for effectively leveraging the power of AI. The shift from prompt engineering to context engineering is presented as a crucial trend. By converting tacit knowledge into explicit knowledge and providing LLMs with rich, structured context, users can unlock deeper insights, enhance their own thinking, and avoid the pitfalls of intellectual outsourcing. The example provided demonstrates the tangible benefits of this approach, showcasing how a well-maintained knowledge base can transform AI interactions from generic outputs to personalized, actionable recommendations.