See the trees. See the forest.
In the age of LLMs—large language models, AI systems that can generate natural language and other text through conversation—knowledge has become remarkably easy to obtain. Almost anyone using an LLM can access nearly all the knowledge that human beings can express through language. At the same time, some of the needs once served by internet search engines and blogs are being absorbed by the LLM chat window. So in this age, does it still mean anything to keep and maintain personal notes, or to build and publish a personal blog?
From my student years through six years of working, I have always kept notes. Over time, I have maintained more than four thousand note documents. I wanted to preserve the knowledge that I found valuable and useful in my studies and work, and to share some of the better material on the internet through my blog. In the old age, keeping notes was, in essence, a way of collecting knowledge and giving myself the feeling that I had mastered it—something that helped me resist the anxiety of facing challenges at work. Publishing a blog, meanwhile, was partly about helping others by sharing my research and implementations, and partly about building a personal identity and introducing myself to other people. But in the age of LLMs, where is the value of notes and blogs? Are all the things I recorded before simply going to “rot in a box”? The real knowledge network of See the Forest: public content is no longer just a list of articles, but an interconnected forest.
Some time ago, I learned about Karpathy’s LLM Wiki. Inspired by it, I wanted to build a personal note library of my own, maintained with the help of LLM agents. The core of Karpathy’s approach is this:
Let an LLM agent build and maintain a wiki continuously and incrementally. Instead of waiting until a question is asked and then searching and assembling information on the spot, it processes material as it arrives—extracting, organizing, indexing, and continuously updating it—so that high-value results become persistent artifacts capable of compounding over time. He divides the LLM Wiki into a raw content layer, a wiki layer, and a schema layer that maintains the contract. I wanted to learn from his approach and build a note library of my own, maintained by agents. Karpathy also wrote that “LLM Wiki describes an idea, not an implementation.” I wanted to learn from the essence of that idea, rather than imitate its appearance or copy its form. At the same time, I wanted to improve the approach in my own way. While building this knowledge-base system, I followed these ideas:
In short, by collecting, classifying, connecting, and drawing inspiration from knowledge, we can build it into a network—and allow wisdom to emerge as it ferments. What the LLM agent changes in this process is the amount of work and the efficiency of doing it: it makes a complex undertaking easier and genuinely feasible. In other words, compared with the age before LLM agents, they have not changed the nature of knowledge. They have made something that was once too difficult to sustain over the long term practical.
That is why I built a knowledge-network system called See the Forest. Its name comes from a saying: “See the trees. See the forest.” It means that when we look at things, problems, or knowledge, we should not see only isolated points. We should also see the connections within and between them, and the network formed by those connections—not only the trees, but the entire forest they form, its ecosystem, and the relationships among its living things.
The first step was to reorganize and reasonably classify the notes I had accumulated, and to build a directory tree that could be maintained over the long term, so that every note had a basic place to belong. But a directory can only decide where a note is stored. It cannot fully express the relationships between pieces of knowledge.
One difference between the See the Forest knowledge-network system and Karpathy’s approach is that Karpathy separates the raw material layer from the wiki layer. I believe that the network formed by knowledge is itself part of the knowledge; they are one whole. I also believe that the original notes themselves should be maintained and improved, rather than maintaining only a separate wiki layer. So although See the Forest also distinguishes between a “knowledge foundation” and a “knowledge evolution layer,” it does not maintain them in two fundamentally different ways. Instead, the knowledge and the network produced through its evolution form a single whole, and agents help that whole network evolve and ferment.
The knowledge foundation preserves original materials and earlier understanding. The knowledge evolution layer maintains semantic relationships, synthesized judgments, evidentiary boundaries, contradictions, open questions, and the evolution of understanding. This does not mean allowing an agent to rewrite my words at will. The history of my thinking belongs to me. The agent improves efficiency, but the final judgment remains mine.
I want this to be a living, intelligent knowledge network. And the approach itself demonstrates something important: the knowledge network does not exist because LLM agents changed the nature of knowledge. It exists because their gains in efficiency make such a complex, previously impractical undertaking possible.
Select a topic and the content truly related to it lights up. From there, you can continue into the topic itself.
I did not tear down and rebuild more than four thousand notes merely to fit a new approach. Markdown and my existing directories remain the knowledge foundation, and Obsidian remains my main interface for reading and maintaining it. Through classification, bidirectional links, and Graph View, I have gradually connected content that was once scattered. But more links are not automatically better. What matters is why two things are related, and how new material changes what I previously understood.
I now maintain the system together with agents. Agents help me discover relationships across a large body of material, compare new knowledge with old, and maintain the network. But whether a question matters, which ideas are worth preserving, and whether a change should ultimately be accepted are still judgments I make. I call this process “knowledge compilation”: locate the relevant earlier knowledge, compare the old with the new, write back the changes that matter, and finally validate the result. There are more specific mechanisms behind relationship judgment, permissions, and engineering, but I will not discuss those implementation details here.
Select a node and see its place in the network and its real relationships. Once the basic structure of my personal knowledge base had gradually taken shape, I wanted to uncover more of its value—this network should be visible. If we could actually see its structure with our own eyes, first observing the topics, relationships, and spaces as a whole, and then entering the details of a specific piece of content, that would be very cool.
Obsidian already has Graph View. Its way of displaying and interacting with a knowledge network inspired me, but only I could browse it locally. Why not build a visual and interactive knowledge-network system that could be presented on a personal blog?
I did not want to add a decorative graph to the homepage of a traditional blog. Every node in the graph should correspond to real public content, and every relationship should come from real bidirectional links and semantic structure. When readers open the homepage, the first thing they see should not be a list of “latest articles,” but the topic clusters, bridges, and spaces that have formed through the long-term growth of the knowledge itself. From there, they can enter a particular piece of content.
After a period of design, development, and iteration, I built a visual and interactive knowledge-network system that can be deployed as a personal blog. In its visible behavior, it follows physical and mathematical principles: nodes move naturally, relationships remain clear, and browsing and dragging stay smooth even when a large amount of content is shown at once. I will not discuss how these effects are implemented here.
This network is a field of stars made of knowledge. Every node is a topic or an article. The stars are connected to one another, and the strength of those connections changes with the strength of the relationships. You can drag them with the mouse, or select a node to enter its topic or article.
The knowledge network is not decoration. It is the primary entrance to the blog. If exploring through the star map is not your habit, you can still find content through search, text-based topic entrances, and ordinary links.
See the relationships as a whole, then enter the content itself.
There is another change in the age of LLM agents: humans are no longer the internet’s only readers. Agents have, in practice, become users of the internet too. So this blog should not only be easy for people to read; it should also be possible for agents to discover, read, and understand it.
But I did not want simply to add a chat box to the homepage. Rather than having a model answer in my place, I care more about whether an agent can read my original words, understand the relationships between pieces of knowledge, and bring its citations back to the actual content. Humans and agents may enter through different doors, but what they ultimately see should be the same content and the same knowledge network. Material without evidence should not be presented as my view, and private knowledge will not enter the public blog.
People can enter an article from the knowledge network. The same public content is also available in representations that agents can read.
I have not open-sourced the complete blog and knowledge-star-map implementation. In the future, I may offer the whole approach as a capability and service package, helping others who also want to build a personal knowledge network and a corresponding blog—from organizing their knowledge base and evolving their knowledge, to presenting and deploying the public network.
If you simply want to see what the system looks like today, visit:
[https://www.pystone.net/notes/see-the-forest-knowledge-network-evolution-blog-system/](https://www.pystone.net/notes/see-the-forest-knowledge-network-evolution-blog-system/)
Do not search at first. Choose any topic you recognize, follow its real relationships for two steps, and then enter an article. You may understand more directly what I mean by “See the trees. See the forest.”
If you would also like to build a knowledge network and blog of your own, you can contact me by email to reserve a place in advance. At this stage, I provide the approach, deployment, and support as a paid service.