Karpathy's LLM Wiki - Full Beginner Setup Guide  🔗 **Source:** [Karpathy's LLM Wiki - Full Beginner Setup Guide](https://www.youtube.com/watch?v=iXd0t60YmMw) > [!note]+ Document Notes > tags: #ai #knowledge-management #obsidian #rag #productivity > author: Teachers Tech > source: YouTube > reference: https://www.youtube.com/watch?v=iXd0t60YmMw > date created: 2024-12-19 14:30 > > # 🏷️ Title: Karpathy's LLM Wiki - Full Beginner Setup Guide > <!-- deep-read:sections start --> ## 🧰 Action Inbox - [H1] Set up Obsidian vault with three-folder structure — 05:23 #setup - [H2] Install Obsidian web clipper extension for article capture — 08:15 #tools - [H3] Create claude.md schema file with wiki rules — 06:45 #configuration ## 🧭 Table of Contents - [[#📝 Summary]] - [[#📖 Main Content]] - [[#🗒️ Notes]] - [[#📚 References]] - [[#🗣️ Paraphrase]] - [[#📰 Newsletter]] - [[#✅ Action Items]] - [[#🧪 Hypotheses]] - [[#📊 Methodology]] - [[#❓ Questions]] - [[#🔮 Future Research]] - [[#💡 Implications]] - [[#🧠 Learning Styles Assessment]] - [[#📊 Accessibility Assessment]] - [[#🌍 DEIB Assessment]] - [[#✅ Logical Validity]] - [[#🛡️ DARVO Analysis]] - [[#🎯 Conclusion]] ### 🔗 References - [Obsidian](https://obsidian.md) - [Claude Projects](https://claude.ai) ## 📝 Summary - Andre Karpathy's LLM Wiki concept solves the problem of traditional RAG systems that start from scratch with every query - Instead of searching raw documents repeatedly, an AI builds and maintains a persistent knowledge base of interconnected markdown files - The system uses three layers: raw sources (read-only), AI-maintained wiki pages, and a schema file with rules - Over time, the wiki accumulates knowledge, makes connections, and enables more sophisticated queries across multiple sources - Bottom line: This creates a personal knowledge base that gets smarter with each addition rather than forgetting everything between sessions ## 📖 Main Content Traditional RAG (Retrieval Augmented Generation) systems have a fundamental flaw: they treat every query as if encountering the documents for the first time. When you ask a question, the AI searches through files, extracts relevant pieces, and generates an answer—but nothing persists for future queries. Karpathy's LLM Wiki flips this approach. Instead of searching raw documents repeatedly, you have the AI read documents once and build a structured, persistent knowledge base. When new sources are added, the AI doesn't just store them—it reads, extracts key ideas, integrates them into existing wiki pages, creates new concept pages, links related ideas, and flags contradictions. The three-layer architecture consists of: (1) Raw sources folder containing original documents that remain read-only, (2) Wiki folder with AI-maintained markdown files including index pages, concept pages, and entity summaries, and (3) Schema file (claude.md) containing rules for how the AI should structure and maintain the wiki. The system works best at personal scale (around 100 articles) and requires a coding agent like Claude Code or Cursor to function. Quality depends entirely on source curation, and the AI can make mistakes requiring periodic "linting" reviews. ## 🗒️ Notes #### ⏱️ [00:00](https://youtu.be/iXd0t60YmMw?t=0) — [H1] The fundamental problem with current AI document handling - Key idea: Traditional RAG systems start from scratch with every query, never accumulating knowledge - Evidence: "Ask a similar question tomorrow, and the AI does all of that work again from scratch. Nothing was saved. Nothing was built up." #### ⏱️ [01:45](https://youtu.be/iXd0t60YmMw?t=105) — [H2] Introducing Karpathy's LLM Wiki solution - Key idea: Build a persistent knowledge base where AI reads documents once and creates structured wiki - Evidence: "Andre Karpathy, one of the biggest names in AI, co-founder of OpenAI, former AI director at Tesla, recently shared an idea that fixes this problem." #### ⏱️ [02:30](https://youtu.be/iXd0t60YmMw?t=150) — [H3] The limitations of RAG explained through researcher analogy - Key idea: RAG forces AI to re-read everything as if it's never seen the documents before - Evidence: "Imagine you're a researcher and you've been reading papers on a topic for weeks. With RAG, every time you ask the AI a question, it's like it's never read any of those papers before." #### ⏱️ [03:15](https://youtu.be/iXd0t60YmMw?t=195) — [H1] How LLM Wiki integrates knowledge over time - Key idea: AI reads documents once, extracts ideas, integrates into existing pages, creates new concepts, and links related ideas - Evidence: "It updates existing pages. It creates new pages for new concepts. It links related ideas together. And if the new sources contradict something already in the wiki, it flags that too." #### ⏱️ [04:00](https://youtu.be/iXd0t60YmMw?t=240) — [H2] Karpathy's development environment analogy - Key idea: Think of Obsidian as IDE, LLM as programmer, wiki as codebase - Evidence: "Think of Obsidian as the IDE, the LLM as the programmer, and the wiki as the code base. You rarely write the wiki yourself. The AI does the writing and organizing." #### ⏱️ [04:30](https://youtu.be/iXd0t60YmMw?t=270) — [H1] Three-layer system architecture - Key idea: Raw sources (read-only), AI-maintained wiki pages, and schema rules document - Evidence: "Layer one, your raw sources... Layer two is the wiki itself... Layer three, the schema." #### ⏱️ [05:23](https://youtu.be/iXd0t60YmMw?t=323) — [H1] Setting up Obsidian vault structure - Key idea: Create new vault with three folders: raw, wiki, and templates - Evidence: "I'm going to create three folders. The first one's going to be raw... The second folder is going to be wiki... And the third folder is going to be called templates." #### ⏱️ [06:45](https://youtu.be/iXd0t60YmMw?t=405) — [H1] Creating the schema file (claude.md) - Key idea: Schema file contains rules for AI operation including purpose, folder structure, ingest workflow, formatting rules, and Q&A behavior - Evidence: "We need to create the schema file, the rules document that tells the AI how to operate the wiki." #### ⏱️ [08:15](https://youtu.be/iXd0t60YmMw?t=495) — [H2] Installing Obsidian web clipper extension - Key idea: Web clipper converts articles into markdown files for easy ingestion - Evidence: "I'm also going to add this Obsidian extension here. It's a web clipper... what it's going to do is convert any web articles into a markdown file." #### ⏱️ [10:30](https://youtu.be/iXd0t60YmMw?t=630) — [H1] First document ingestion demonstration - Key idea: AI reads article, creates structured wiki pages with neighborhoods, summaries, and connections - Evidence: "Claude is reading the article. It's creating the wiki pages... Here's the pages for different neighborhoods." #### ⏱️ [12:15](https://youtu.be/iXd0t60YmMw?t=735) — [H1] Graph view visualization of knowledge connections - Key idea: Obsidian's graph view shows how concepts link together across documents - Evidence: "And if I go over to graph view, take a look at this. You can see the connections forming. This is one document. Imagine what this looks like after 20 sources." #### ⏱️ [13:30](https://youtu.be/iXd0t60YmMw?t=810) — [H1] Second source integration and wiki updating - Key idea: Adding new sources updates existing pages and creates new connections rather than just adding separate content - Evidence: "Claude isn't just creating new pages. It's actually updating the neighborhood pages as well that it already made." #### ⏱️ [14:45](https://youtu.be/iXd0t60YmMw?t=885) — [H1] Cross-document query demonstration - Key idea: AI answers complex questions by pulling from multiple wiki pages rather than searching raw documents - Evidence: "Claude's not searching the raw articles. It's pulling from the neighborhood pages, the food pages, the temple page. It's connecting dots that were spread across completely different sources." #### ⏱️ [16:00](https://youtu.be/iXd0t60YmMw?t=960) — [H2] Wiki linting for maintenance and quality control - Key idea: Periodic AI review identifies contradictions, outdated claims, orphan pages, and missing concept pages - Evidence: "Just like how a code linter checks your code for problems, you can periodically ask the AI to edit the whole wiki." #### ⏱️ [17:30](https://youtu.be/iXd0t60YmMw?t=1050) — [H3] Practical use cases across domains - Key idea: Students build research wikis, teachers organize curriculum materials, businesses structure meeting notes and customer calls - Evidence: "If you're a student or a researcher, build a wiki as you read papers and articles... If you're a teacher, feed in curriculum documents... If you're a business, feed in meeting notes." #### ⏱️ [18:45](https://youtu.be/iXd0t60YmMw?t=1125) — [H2] System limitations and scale considerations - Key idea: Works best at personal scale (~100 articles), requires quality source curation, needs coding agent, AI can make mistakes - Evidence: "This works best at personal scale. Karpathy talks about having wikis of around 100 articles... The wiki is only as good as the sources you feed it." ## 📚 References - [Obsidian](https://obsidian.md) - [Claude Projects](https://claude.ai) - [Obsidian Web Clipper Extension](https://chrome.google.com/webstore) ## 🗣️ Paraphrase - Traditional AI document systems forget everything between questions, forcing repeated analysis of the same materials without building cumulative knowledge - Karpathy's approach creates a persistent knowledge base where AI reads documents once, extracts concepts, and builds interconnected wiki pages that grow smarter over time - The system uses a simple three-layer architecture with raw sources, AI-maintained wiki pages, and a schema file defining operational rules ## 📰 Newsletter **Transform Your Knowledge Management with AI-Powered Wikis** Tired of AI systems that forget everything between questions? Andre Karpathy's LLM Wiki concept changes the game entirely. Instead of repeatedly searching through raw documents, this approach builds a persistent, interconnected knowledge base that gets smarter with every addition. The setup is surprisingly simple: Obsidian for visualization, Claude Code for AI processing, and a three-folder structure (raw sources, wiki pages, schema rules). As you add documents, the AI doesn't just store them—it reads, synthesizes, creates concept pages, and links related ideas across all your sources. Perfect for educators building curriculum wikis, students organizing research, or anyone accumulating knowledge over time. Your data stays local in plain markdown files you control. **Resources:** [Obsidian Download](https://obsidian.md), [Claude Projects](https://claude.ai), [Web Clipper Extension](https://chrome.google.com/webstore) ## ✅ Action Items #### DevRel - Developer advocacy: Demo LLM wiki for technical documentation workflows - Education: Create workshop on AI-powered knowledge management systems - Developer experience: Explore integration with existing documentation tools - Community management: Survey developer preferences for knowledge base tools #### WordPress - Documentation: Pilot LLM wiki for organizing WordPress development resources - Training: Develop lesson plans for AI-assisted content organization - Community: Create templates for meetup organizers to build local knowledge bases ## 🧪 Hypotheses - LLM wikis will outperform traditional RAG systems for personal knowledge management tasks requiring cross-document synthesis - Users with 50+ documents will see significantly better query results compared to standard document search after 30 days - Wiki-based systems will reduce time-to-insight for research tasks by 40% compared to traditional file searching - Personal knowledge workers will prefer persistent wikis over session-based AI interactions for complex domains ### 📊 Methodology - A/B test comparing LLM wiki vs traditional RAG for research synthesis tasks - Longitudinal study tracking knowledge base growth and query accuracy over 90 days - User experience research on cognitive load differences between search-based and wiki-based knowledge retrieval ### ❓ Questions - How does wiki quality degrade with sources containing contradictory information? - What's the optimal document-to-concept ratio for maintaining coherent wiki structure? - How can version control be implemented for collaborative wiki development? ### 🔮 Future Research - Integration with collaborative platforms for team-based knowledge building - Automated source quality scoring and bias detection - Cross-wiki federation for domain-specific knowledge networks ### 💡 Implications - Practice: Shifts knowledge work from search-and-synthesize to curate-and-query workflows - Education: Enables personalized curriculum development and adaptive learning pathways - Security: Requires consideration of data sovereignty and local-first architectures - Funding: May reduce enterprise search infrastructure costs while improving knowledge retention ### 🧠 Learning Styles Assessment Reading level estimate: 9th-10th grade #### 👀 Visual learners - Present: Obsidian graph view screenshots, folder structure diagrams - Improve: Architecture diagrams showing data flow between layers #### 🔊 Aural learners - Present: Full video tutorial with clear narration - Improve: Audio-only summary highlighting key concepts #### 🗣️ Verbal learners - Present: Clear analogies (IDE/programmer/codebase), step-by-step explanations - Improve: Glossary of technical terms and conceptual comparisons #### 🧩 Kinesthetic learners - Present: Complete hands-on setup walkthrough, downloadable templates - Improve: Interactive checklist and troubleshooting guide #### 🧠 Logical learners - Present: Clear three-layer system architecture, workflow sequences - Improve: Decision trees for when to use different approaches #### 🧑‍🤝‍🧑 Social learners - Present: Use case examples across different roles and domains - Improve: Community templates and collaboration patterns #### 🧘‍♀️ Solitary learners - Present: Self-contained setup guide, personal knowledge management focus - Improve: Reflection prompts on knowledge organization preferences #### 🌿 Naturalistic learners - Present: Organic growth analogies for knowledge base development - Improve: Ecosystem thinking about information relationships ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Clear heading hierarchy, descriptive video narration - Improve: Alternative text for screenshots, structured content outline #### Keyboard navigation - Present: Standard YouTube video controls - Improve: Keyboard shortcuts guide for Obsidian setup #### Contrast and readability - Present: High contrast video interface, clear typography - Improve: Adjustable playback speed recommendations #### Text resizing and reflow - Present: YouTube's responsive video player - Improve: Downloadable text transcript with proper formatting #### Media and captions - Present: Auto-generated captions available - Improve: Human-reviewed captions and downloadable transcript #### Forms and interactions - Present: Simple file creation and folder management - Improve: Error handling guidance for setup issues ### 🌍 DEIB Assessment #### Cultural breadth - Present: Travel planning example accessible to global audience - Improve: Multiple domain examples reflecting diverse professional contexts #### Inclusive language - Present: Plain language explanations, avoids unnecessary jargon - Improve: Consider accessibility of required tools and subscriptions #### Neurodiversity support - Present: Step-by-step visual guidance, multiple learning modalities - Improve: Alternative organization strategies for different cognitive styles #### Representation - Present: Focus on practical utility across roles and domains - Improve: Diverse use case examples beyond academic/research contexts #### Psychological safety - Present: Encouraging tone, acknowledges limitations honestly --- ## Full Transcript [00:00] There's a problem with the way most of us use AI right now. And once you see it, you're not  <br>[00:04] going to be able to unsee it. When you upload documents to something like ChatGpt or Notebook LM  <br>[00:10] and ask a question, the AI searches through your files, pulls out some relevant pieces, and gives  <br>[00:15] you an answer. That works. But here's the thing. Ask a similar question tomorrow, and the AI does  <br>[00:21] all of that work again from scratch. Nothing was saved. Nothing was built up. Every single  <br>[00:26] question starts from zero. Andre Karpathy, one of the biggest names in AI, co-founder of OpenAI,  <br>[00:31] former AI director at Tesla, recently shared an idea that fixes this problem. He calls it the LLM  <br>[00:37] wiki. And on honestly, once you understand what it does, the old way of working with documents  <br>[00:42] starts to feel broken. In this video, I'm going to walk you through exactly what the LLM wiki is,  <br>[00:48] why it matters, and then we're going to build one together from scratch, step by step. You  <br>[00:52] don't need to be technical. If you can create a folder on your computer, you can do this.<br>[01:00] Hi, I'm Jamie and welcome to Teachers Tech. So,  <br>[01:03] let me explain the problem a bit more clearly because this is important.  <br>[01:07] The way most AI tools handle your documents right now is called RAG, retrieval augmented generation.  <br>[01:13] You upload some files, you ask a question, the AI searches through those files,  <br>[01:18] grabs the chunks that seem relevant, and generates an answer. And that's fine for simple questions,  <br>[01:23] but what if your questions require connecting ideas across five different documents? The AI has  <br>[01:28] to find all those pieces and stitch them together every single time. There's no memory. There's no  <br>[01:34] accumulation. Nothing compounds. Think about it like this. Imagine you're a researcher and you've  <br>[01:40] been reading papers on a topic for weeks. With Rag, every time you ask the AI a question, it's  <br>[01:45] like it's never read any of those papers before. It starts fresh every time. That's the bottleneck.  <br>[01:51] Garpathy's idea flips this completely. Instead of searching raw documents every time you ask a  <br>[01:56] question, you have the AI read your documents once and build a structured wiki out of them.  <br>[02:02] A real persistent knowledge base made of interlink markdown files. So when you add a new source,  <br>[02:07] say a PDF or an article, the AI doesn't just store it for later. It actually reads it, extracts the  <br>[02:13] key ideas, and integrates them into the wiki. It updates existing pages. It creates new pages for  <br>[02:19] new concepts. It links related ideas together. And if the new sources contradict something already  <br>[02:24] in the wiki, it flags that too. So over time, the wiki keeps growing and getting richer. The  <br>[02:30] connections are already there. The synthesis is already done. When you ask a question,  <br>[02:35] the AI is not starting from scratch. It's actually working from a pre-built organized knowledge base.  <br>[02:40] Here's how Kaproy describes it. He says, "Think of Obsidian as the IDE, the LLM as the programmer,  <br>[02:47] and the wiki as the code base. You rarely write the wiki yourself. The AI does the writing and  <br>[02:53] organizing. You focus on what goes in and what questions to ask. Now, the whole system has three  <br>[02:59] layers and they're very simple. Layer one, your raw sources. These are your original documents  <br>[03:05] like PDFs, articles, meeting notes, whatever you're working with. The important thing is  <br>[03:10] that these are read only. The AI reads them but never changes them. This is your source of truth.  <br>[03:16] Layer two is the wiki itself. This is a folder of markdown files that the AI creates and maintains.  <br>[03:22] It's going to have things like an index page, concept pages, entity pages, summary comparisons,  <br>[03:28] all interlin, all maintained by the AI. Layer three, the schema. This is basically a rules  <br>[03:34] document. It tells the AI how to structure the wiki, how to handle new sources, how to format  <br>[03:40] everything. If you're using Claude Code, this would be your claw.md file. If you'd followed  <br>[03:45] my other Claude Code series, you already know what that is. If you're new to Claude Code,  <br>[03:49] I'll put the link to my beginner's video right up there. All right, let's get into the setup. Here's  <br>[03:54] what we're going to need. First, Obsidian. This is a free note-taking app that works with plain  <br>[03:59] Markdown files. It's going to be our viewer. You can download it at obsidian.md. I'll put the link  <br>[04:04] down below in the description. And don't worry if you've never used Obsidian before. I'll walk  <br>[04:09] you through the parts that matter. Second, an AI coding agent. I'm going to be using clawed code  <br>[04:14] for this because this is what I've been using in my series and it works really well for this,  <br>[04:19] but you could also use OpenAI codeex cursor or other tools that can read and write files on  <br>[04:24] your computer. Now, I just want to point out I'm using Obsidian because it has the graph view that  <br>[04:29] makes the connections really visual, but this is just a folder of markdown files. You could use  <br>[04:33] VS Code or any text editor, whatever you're most comfortable with. Once you got Obsidian installed,  <br>[04:39] just go ahead and open it. And the first thing what I'm going to do is just go and create a  <br>[04:44] new vault. You'll see it right here. And this is just a fancy name for folder. So, I'm going to go  <br>[04:49] create and I'm going to call this one LLM wiki. And I'm just going to save it somewhere simple.  <br>[04:55] I'm just going to put it into my documents here. You'll see there. And you can put it where you'd  <br>[05:00] like. I'm going to go and hit create. Now, we need to set up a folder structure. I'm going to  <br>[05:04] create three folders. The first one's going to be raw. I'm just clicking right up here, new folder,  <br>[05:10] and I'm going to call it raw. The AI will read from this, but never change anything in here. The  <br>[05:16] second folder is going to be wiki. This is where the AI will build and maintain all of its pages.  <br>[05:23] And the third folder is going to be called templates. This templates folder is optional.  <br>[05:29] If you wanted to manually create notes in Obsidian with a consistent format, you could put a template  <br>[05:34] right in here. But since Claude is going to be creating all of our wiki pages for us,  <br>[05:38] we don't need it for this tutorial. It's just here as a point to tell you about. So here's what our  <br>[05:44] structure looks like. We have our wiki templates and raw. Nothing too complicated with this.  <br>[05:50] Now, here's the important part. We need to create the schema file, the rules document that tells  <br>[05:54] the AI how to operate the wiki. If you're using Claude Code, you're going to create a file called  <br>[06:00] claude.md in the root of your vault. So, this is the file that Claude Code reads automatically when  <br>[06:06] it opens a project. So, I'm going to give you a starter template that you can copy. It's linked  <br>[06:11] down below in the description, but let me walk you through what's in it. Now, first of all,  <br>[06:15] I'm just going to bring the claw.md file into the root right here. So, I'm just going to drop  <br>[06:20] it so we can have it here. You can see my other folders are here, but here's the claw.md file. So,  <br>[06:26] if I click on it, you're going to be able to see what's in it. Now, first the purpose right here.  <br>[06:31] So, this is the purpose of the wiki. What's the knowledge base about? So, in our template,  <br>[06:36] I've set this to planning a trip to Japan because this is what we're going to do in the demo today.  <br>[06:41] But, uh, you need to when you download this file, this is the one line you can change to  <br>[06:46] match whatever you're going to be building a wiki about. If you're researching renewable energy,  <br>[06:51] change it to that. If you're wanting to track books that you want to learn from,  <br>[06:55] change it to that. Everything else in the template works as is. The purpose of the line is the only  <br>[07:00] thing that you really need to customize to get started. Second, the folder structure. Where are  <br>[07:06] the raw resources? Where's the wiki output? What goes where? Third, the ingest workflow.  <br>[07:12] When you add a new source document, what should the AI do? The basic steps are read the document,  <br>[07:18] extract key concepts, create the update wiki pages, update the index, and log what changed.  <br>[07:24] Fourth, page formatting rules. things like every page should have a summary at the top.  <br>[07:28] Every claim should reference its source. Pages should link to related concepts. And fifth,  <br>[07:35] the question answering behavior. When you ask the AI a question, it should consult the wiki first,  <br>[07:42] cite its sources, and tell you when something is uncertain. Now, don't overthink this. The template  <br>[07:48] I'm giving you gives you a solid starting point. You can always refine it as you go.  <br>[07:52] That's actually part of the process. The schema evolves as the wiki grows. I'm also going to add  <br>[07:58] this Obsidian extension here. It's a web clipper. So, I'm going to go ahead and add this to Chrome.  <br>[08:03] It's free to do and what it's going to do is convert any web articles into a markdown file. So,  <br>[08:08] it's super handy. All right. Now, here comes the fun part. Let us feed the wiki with its  <br>[08:12] first document. Now, I'm going to drop an article into the RAW folder. And for this demo,  <br>[08:17] I'm going to be planning a trip to Japan. So, I'm going to start with a travel blog  <br>[08:21] post about visiting Tokyo. things to do like neighborhoods to explore, that kind of thing.  <br>[08:26] I I'm going to save this as a markdown with the extension that we just installed. So,  <br>[08:31] if I go up top, you can see I have the extension here and I could go directly to Obsidian or I can  <br>[08:38] download it. So, I am going to just or I could copy paste it over too. I'm going to hit save as.  <br>[08:42] Now, I'm going to hop back over to Obsidian here. So, that's just in my downloads folder right now.  <br>[08:49] So, I can see it. I can drag it over. And where do I want it? I want to put it in my raw. So, if  <br>[08:56] I click here, it is now. Here's that article. All that information right in here into my raw folder.  <br>[09:03] And by the way, your sources don't have to be markdown files. If you have PDFs, just drag them  <br>[09:08] straight into the raw folder. Claude Code can read PDFs natively. Same with text files. Same  <br>[09:13] with Markdown. Whatever your format documents are in, just drop them in and Claude will handle it.  <br>[09:19] Now, I want to go and open Claude. But before I do that, I need to make sure that I'm pointing  <br>[09:23] towards where we have this all set up. So, I'm going to just change my directory  <br>[09:28] to this right through here. You can see I put it in my documents and this is what it's called.  <br>[09:34] So, we have our directory changed. Now, I'm going to go ahead and open up Claude.  <br>[09:40] Okay. So, now I'm going to tell it to ingest the new source. So, I'm just going  <br>[09:45] to say I just added a new source to the raw folder. Please read it and update the wiki.  <br>[09:51] And watch what's happening. Claude is reading the article. It's creating the wiki pages. And there's  <br>[09:57] the summary of the article. Here's the pages for different neighborhoods. Like all through here,  <br>[10:03] you can see all the different wiki pages here that it's planning to create. And if this looks good,  <br>[10:09] I'm going to tell it to go ahead. But you can see how I can adjust the scope as well.  <br>[10:12] I'm just going to say go ahead. Okay. You can see after about 3 minutes, it's all done. But  <br>[10:17] let's go check out Obsidian and what's happening over there. All right, let's open up the wiki.  <br>[10:23] See, look at this. We have structured pages. If we click on any of these here,  <br>[10:28] we have links to all of these. And if I go over to graph view, take a look at this. You can see  <br>[10:36] the connections forming. This is one document. Imagine what this looks like after 20 sources.  <br>[10:42] Now, this is where it gets really interesting. Let me add a second source. We're going to do this  <br>[10:47] food guide here to Japan. I'm going to do what I did last time. I'm just going to go ahead and  <br>[10:51] make sure that I save it and I'll bring it back over obsidian. Then I'll tell Claw to ingest it.  <br>[10:59] Let's say the exact same thing. I just added new sources to the raw folder. Please read it and  <br>[11:03] update the wiki. Now look at this. So Claude isn't just creating new pages. It's actually updating  <br>[11:10] the neighborhood pages as well that it already made. And you can kind of look specifically at  <br>[11:15] the details how they're making those adjustments to each of these. This is the wiki doing its job.  <br>[11:20] Now look at the graph view now. More nodes, more connections. The wiki is getting smarter with  <br>[11:26] every source we add. Now let me ask a question that requires information from both sources.  <br>[11:31] What neighborhood should I stay at if I want to be close to the best food and still near the major  <br>[11:37] temples? And look at the answer. Claude's not searching the raw articles. It's pulling from  <br>[11:43] the week from the neighborhood pages, the food pages, the temple page. It's connecting dots  <br>[11:47] that were spread across completely different sources. It's citing specific wiki pages. This  <br>[11:52] is completely different from what you get with basic rag setup. One more thing I want to show  <br>[11:57] you that I think is really clever. Karpathy talks about this idea of linting your wiki. Just like  <br>[12:03] how a code llinter checks your code for problems, you can periodically ask the AI to edit the whole  <br>[12:09] wiki. It will look for things like contradiction between pages, claims that might be outdated,  <br>[12:14] pages that have no links pointing to them or like orphan pages and concepts that are mentioned but  <br>[12:20] don't have any of their own page yet. We can just say something like this. Please lint the wiki.  <br>[12:26] Now Claude's going to go through everything and give you a report.  <br>[12:30] This is how you keep your wiki healthy as it grows. And look what it gives me back here.  <br>[12:37] the different checks from orphan pages to broken links. This is telling me that it's structurally  <br>[12:41] sound, which I expected since we only have two different articles in this. And it points out the  <br>[12:46] biggest gap here is the unested food source, and even makes the offer to fix the citation issues.  <br>[12:52] So, what would you actually use this for? Here are some ideas. If you're a student or a researcher,  <br>[12:57] build a wiki as you read papers and articles on a topic. By the end, you have this structured  <br>[13:02] knowledge base, just not a pile of highlighted PDFs. If you're a teacher, feed in curriculum  <br>[13:07] documents, professional development materials and articles. Build a personal teaching wiki  <br>[13:12] that grows over time. If you're a business, feed in meeting notes, customer call transcripts,  <br>[13:18] and project documents. So, this allows new team members to browse this organized wiki instead of  <br>[13:23] digging through Slack history. If you're just a curious person who reads a lot, use it to track  <br>[13:29] what you learn from books, podcasts, and articles. It's like building your own personal encyclopedia.  <br>[13:34] The pattern works anywhere you're accumulating knowledge over time and you want it organized  <br>[13:39] rather than just scattered. Okay, let me be straight about the limitations because this  <br>[13:43] isn't magic. First, this works best at personal scale. Karpathy talks about having weeks of around  <br>[13:50] 100 articles. If you're trying to build something with tens of thousands of pages, you're going to  <br>[13:54] want more infrastructure than just some markdown files. Second, garbage in, garbage out. The wiki  <br>[14:00] is only as good as the sources you feed it. You still need to curate what goes in. Third,  <br>[14:06] you do need a coding agent to make this work. Obsidian by itself doesn't do any of this. The AI  <br>[14:11] is the engine. So, you need to access something like Claude Code, codeex, or a similar tool.  <br>[14:16] And fourth, the AI can make mistakes. It might miscatategorize something or misconnection. That's  <br>[14:22] why the lint feature exists. If you want to review what it builds, especially earlier on. But with  <br>[14:27] all that said, this is genuinely one of the most practical AI workflows I've seen. It solves real  <br>[14:33] problems. It's free to set up and your data stays on your computer in plain text files that you own.  <br>[14:39] And that's the LLM Wiki, a personal knowledge base that the AI builds and maintains for you  <br>[14:44] that actually gets better over time instead of starting from scratch on every question.  <br>[14:49] I'll have the schema template and all the links you need in the description below. If you want to  <br>[14:54] learn Claude Code so you can use it for yourself, my beginner's guide is down there. Also, thanks  <br>[14:59] for watching this time on Teachers Tech. We'll see you next week with more tech tips and tutorials. <!-- deep-read:sections end --> 📄 **Full text:** ![[Full Document Contents/Articles/Karpathy's LLM Wiki - Full Beginner Setup Guide.md|Karpathy's LLM Wiki - Full Beginner Setup Guide by YouTube]] ## Highlights - document. It tells the AI how to structure the  wiki, how to handle new sources, how to format   everything. If you're using Claude Code, this  would be your claw.md file. If you'd followed   my other Claude Code series, you already know  what that is. If you're new to Claude Code,   I'll put the link to my beginner's video right up  there. All right, let's get into the setup. Here's   what we're going to need. First, Obsidian. This  is a free note-taking app that works with plain   Markdown files. It's going to be our viewer. You  can download it at obsidian.md. I'll put the link ([View Highlight](https://read.readwise.io/read/01kpm4fwkgd97je8a11gsv25km)) - Tags: 👻 ai highlighted - So, we have our directory changed. Now,  I'm going to go ahead and open up Claude.   Okay. So, now I'm going to tell it to  ingest the new source. So, I'm just going   to say I just added a new source to the raw  folder. Please read it and update the wiki.   And watch what's happening. Claude is reading the  article. It's creating the wiki pages. And there's   the summary of the article. Here's the pages for  different neighborhoods. Like all through here, ([View Highlight](https://read.readwise.io/read/01kpm4fwmghp688y3ke21yeyyk)) - Tags: 👻 ai highlighted - you can see all the different wiki pages here that  it's planning to create. And if this looks good,   I'm going to tell it to go ahead. But you  can see how I can adjust the scope as well.   I'm just going to say go ahead. Okay. You can  see after about 3 minutes, it's all done. But   let's go check out Obsidian and what's happening  over there. All right, let's open up the wiki.   See, look at this. We have structured  pages. If we click on any of these here,   we have links to all of these. And if I go over  to graph view, take a look at this. You can see ([View Highlight](https://read.readwise.io/read/01kpm4fwnfssvyygg643bwee1f)) - Tags: 👻 ai highlighted