Transcription
So this is my second brain. It holds all the knowledge that I have. Everything about YouTube videos, everything about AI labs, the latest papers, even my sponsors. Everything that I know gets automatically logged, connected, and summarized for me in my second brain. And I didn't write most of it anywhere really. The AI did. And as it slowly aggregates all the information, it's able to parse it. And I'm able to ask it a question and get immediate answers based on my context and all of the data that I have access to. For example, these little clusters here are different YouTube creators. For example, this node is me. It connects me to all of my videos, the various topics that I've talked about, all my content and performance on YouTube, on X newsletters, etc.
Here's an example of how I would use it. So, here's Cloud Code, the desktop app. Recently, I've started using the desktop app for pretty much everything. Before, I was running most of my stuff in the command line interface. Now, it's pretty much all here. So, I'm going to ask it based on the data we have available. What topics haven't I covered that are very popular in the AI sphere? And here's the answer. Grock, Deepseek, Copilot, Codeex, etc. It's taking into account the velocity of the different videos, kind of which topics are trending right now, and pulling from the entire database to provide the best possible answer. So does this from its own notes. The notes that it wrote and that it keeps updated every single day. It knows my sponsor deadlines better than I do. And it keeps track of a million different things that I could not possibly keep track of.
This whole idea came from one of the most respected names in AI, and that is Andre Karpathy. And by the end of this video, you'll know exactly what to do to build your own. I also had to create some slides for me so I can do this presentation without losing track of what I'm talking about. So, first and foremost, what is a second brain? This idea has been around for some time. It went by many different names, but the core idea was how do we get all the stuff that we're dealing with, all the little notes and ideas and the to-do lists and meeting minutes and just everything, everything everything, how do we sort of collect it all, then organize it all and have it available for when it's actually actionable. And back in the days, you would write it down in a notebook and you'd hope that that page was there when you needed it. Apps improved it a little bit. You could write things down. You can save it. All of us at some point had some system for the intake of all the information that life throws at us. But just saving information, hoarding it, you know, all the bookmarks you have saved in X, that was never really the problem. The problem was trying to maintain it, trying to stay on top of it and using it when you needed to use it, having that information available to you when you need it.
So the first brain, the one in your skull right now, gently floating there, it's great at thinking. It's terrible at storing and organizing. We forget most of the things that we're supposed to do, most of the things that we read. There's some very small percentage of the population that have amazing memory and they just remember everything where they need to. And you know, hooray for them, but that's not most of us. And then we got LLMs. And then these LLMs got good. And we're now at a point where that LLM, well, it can be a librarian for all the data that you have. So you just throw all of your notes and data and recordings and everything everything everything into this vault. Some of it you do manually. Some of it you set up various collection processes for that. So it's done automatically. But you just throw it in the vault. The librarian, this AI, it reads everything. It files it correctly. It puts it on the right shelf, so to speak. It connects all those things in some way that makes sense. Some things you might need for work, some things you might need for the house, for what I'm doing. I like to organize it by topic clusters and this AI it keeps all that information tidy every day forever on autopilot.
So the idea I got to give credit to Andre Karpathy. So he called it the LLM wiki and before that it was called the second brain. Before that there was a book called Getting Things Done the GTD system and it goes back even even further. In fact this original idea predates LLMs. It predates even computers. This idea of a machine that organizes all of your thoughts and ideas that comes from 1945. So as Andre Karpathy said, Obsidian is the IDE. So the IDE is your development environment. It's kind of where you shape software, where you build software. If you're not familiar with Obsidian, don't worry, you will be. And I think you're going to like it. It's free. Then the LLM is the programmer. That's the thing that goes in there and manipulates things and builds things and creates structure. And the wiki is the codebase. So in plain English, Obsidian, that's this app. Just think of it like a notebook. It's this app that you look through to see your notes. Then we have the LLM, the AI. It writes the pages, interlinks them, and keeps them updated. And the wiki becomes sort of the product, the codebase. It's the thing from which you draw all the insights and the knowledge. It's a growing library of interconnected pages. Your job is to make sure that it's getting fed with the various raw data that you want in there. And then on the other side, you ask it questions or you have it deliver the insights to you in some scheduled manner.
Here's the thing, humans, we've been dreaming about this for a long, long time. This dream is 80 years old. Someone named Vannevar Bush thought of this in 1945. He originally called it the Memex. Memex sounds cooler, I got to say. So, it's basically a desk that recorded everything you've ever read with trails connecting related ideas. So, we had that idea 80 years ago. We just never had anything capable of maintaining this database, making it useful on autopilot. We do now. It's large language models. It's ChatGPT, Grok, Gemini, Claude, you name it.
All right, so first and foremost, what is Obsidian? Obsidian is a pretty simple app that has been built on a radical idea. This is going to blow your mind. You know, like all your notes and the stuff you write down. What if instead of you putting it on somebody else's computer like the cloud somewhere some large enterprise what if bear with me here what if you just kept them all on your computer like there your files your data your notes what if they were just uh on your computer if Obsidian vanishes tomorrow all the stuff that you have saved it's still there people call this local first software and it uses something called markdown if you're working for a lot of chatbots you you've probably heard about markdown so they're markdown files. All right. So, that MD extension like claude.md, skill.md, those are your markdown files. Markdown files are super simple. They're basically just text plus a few symbols that do something. You've probably seen something like this, right? So, you just type your text. If you want a heading, you just use one hashtag for heading one, the big one, or two hashtags or or three for heading three. Two asterisks around something makes it bold. One asterisk around it makes it italicized. And simple ways to add code blocks or links, etc. It's super super simple. So, a 10-year-old can learn this in 10 minutes.
So, this right here is Obsidian. This is kind of what it looks like. And this is the graph view. It's pretty cool. Kind of a way to visualize all of the things, all the topic clusters, whatever you have saved in there. Each one of these little dots is a file. So, for example, here's this cluster. Let's zoom in and see what this is. It's my sponsor content flow with my sponsors. These sponsors are not real. They're fake. I can't actually show you the real data. So, I had Cloud Code come up with some fake data just to be able to kind of showcase the idea. But this is more or less exactly the system that I use to keep track of things. Each sponsor gets its own node and then all the information that is needed is uploaded to that. Based on that, we work out what needs to be done, when the due date is, any assets that I need to know about is in there. For me, it's kind of like homework. You know, if you think back to the days you went to school, you got homework and you know, you you hoped that you remember about it and when the due date is. I had struggle with that. Sometimes I would forget to write down the due date. I would forget when things were due. And sometimes I had trouble kind of like breaking up the work into manageable tasks so that slowly over time you completed the project. So, if you've ever had trouble with the same thing or paying the bills on time or filing some specific paperwork that you needed to file, that's sometimes referred to as the ADHD tax. And for a lot of people, this is a very real thing that they struggle with on a daily basis. If you're fortunate enough to be able to afford an executive assistant or somebody that just like handles those things for you, that's great. For most of us, there wasn't an easy solution for most of our lives until recently. Now I just have to find a way to get all those important things and funnel them into my second brain. Then I work with my favorite assistant Claude Code or ChatGPT or whatever to then set up systems, automated systems that make sure that I get notified, hey, this is coming up. Maybe you should start working on this. Also, I don't have to open five different tabs and hunt for different pieces of information all over the place. Everything is connected.
Let's zoom out and I'll give you another example. For example, let's zoom over here. Each one of these things is a paper or blog post from a Frontier AI lab. There's something that I've talked about in the previous videos that I might need to talk about in the future. So, for example, there's a page about AlphaGo and Tree of Thoughts. Some of you might have been following me from those days of years and years ago when we covered Tree of Thoughts. Who remembers that? And all of these are interconnected and they're linked. So, anything that has to do with Meta AI or Demis Hassabis or Sam Altman, they're all connected to each other through these nodes. If you're wondering what this mess of a cloud is, these are the various people that publish about AI. For example, this is me. All those little purple lines point to things that are connected to me. All of my videos. Recently, I started cataloging some of my tweets, although I don't think this is hooked up to that yet. We're we're in the process of doing that. It also connects to topics that I've talked about. So, when we ask the question like, "What topics haven't I talked about in the last whatever 6 months, it has all that data. It's not guessing. It's not going online and searching. It knows all the data is here. It also knows what everyone else is posting. And notice that goes to a number of these lines here that connects all those videos, for example, to the topics that they discuss to the analytics behind those videos to what works, what doesn't. And these red dots here, that's what Claude decided to call beats, like on the beat or my beat. By the way, I have Claude naming a lot of these things. So, some of them look a little bit weird, like it decided to call something the armory. The armory is all the things that it thinks I should build. Things that would be useful and helpful to me, but I need to sit down, you know, plan it out, tell Fable or whatever model I'm using to go ahead and build it out. These are the things that are on that list. The priority board, analytics, demon, X wide funnel, packaging lab, first responder pipeline, retention, minor, comment archive, clip engine.
Now, if you're wondering what are these projects that they're talking about, for example, this is one of them. So, this is what we called the X data ingestion engine. We're getting tons of data from X/Twitter about the performance of various tweets from my own account as well as some other people's accounts. So, step one was to build the engine, sort of the data collection engine. By the way, if you're wondering, oh, are you going to show us how how you did that? I I I also want to know how to do that. Yeah, sure. I opened up Fable 5 on High Effort. By the way, this is, I think, one of the best ways to use it. I found that this is kind of the sweet spot. Not extra high, not ultra Fable 5 high. So, I opened it up and I said, "This is what I want. Tell me what you need for me. What kind of API keys? What services should I sign up for?" And then go build it. Once that X engine is built, on top of that, you build the things that are actually useful to you. So, for example, something that alerts you when a new trending topic is developing. Or in this case, as you can see, we sort of broke down how well different formats of tweets work. What if it's a standalone tweet with a native video or a quote and a video clip, quote plus image, quote or link, or just bare text. What I realized by looking at the data is that the Twitter algorithm changed a few months ago, and I didn't realize it. And so what that meant was that my impressions used to keep going up and up and up month over month. That huge line in January 26th, that was a few viral hits. So that's not really kind of representative, but that was a good month. One of those tweets was shown in a FireShip video. So yes, as seen on FireShip, he was a little bit sarcastic about the tweets, but he's a little bit sarcastic about everything. So and totally love that guy. So I I was just happy to be mentioned. But notice there's a steep drop off, right? You can look at it. You can see it. You know something happened. What?
So, as you can see here, I have one of this these folders X analytics. So, we have all of our data that we are ingesting. Ingesting is a special word that we use here to basically say collect the data, take the data into the vault into our second brain. Not just using that word because I'm hungry. That's the correct terminology. Here we have all our important accounts that we're keeping track of. A weekly scorecard of how well my tweets are performing and also the X engine. The X engine is the actual thing that runs it. So right now within the second brain, this is the amount of data that we have. 22,000 posts archived, almost 6 billion views represented, 4,400 unique authors. There's a lot there. How quickly can I reference one of them? Quickly pull out some piece of information that I need. Instant. It's offline. No API needed. How much did this cost? Under a hundred bucks. I don't know the exact number. I know it's under a hundred because I purchased $100 in credits and that was enough. I just don't know exactly how much it spent. And also, this isn't a static database. It's being watched. So, for example, it lets us compute velocity of the trending topic or tweet. Which post is getting 600 likes an hour right now, tracking breaking AI news. It's also seeing which strategies started breaking down at that algorithmic change that we saw a few months ago. And it's benchmarking me against some of the other accounts, right? So, if something that I'm doing is underperforming, it allows me to pinpoint exactly what it is that I'm doing wrong. And notice that it's storing all these insights and updating them and curating them. And it's a living document, which Claude decided to call X Growth Playbook. Now, again, the reason I want to bring that up is because I I wouldn't have called it that necessarily. The point isn't let's grow. It's not a growth playbook. It's not a growth hacks. I'm thinking of it more as a don't shoot yourself in the foot Wes playbook. The whole point is basically to understand kind of how the algorithm functions if it changes so I don't get caught up in the changes and just lose all my views etc. So the point isn't growth hacks. The point is what are the best practices right now?
By the way, the next level up I think is to turn it into something like this. This is kind of what I'm building right now. This also brings in ideas like for example which apps are connected to Claude. So I have things like Obsidian, Social Blade X, the X API, etc., etc. As your little sort of branching empire starts growing larger and larger, it really helps to have just one place where you can at a glance see, okay, what are all of the things that are connected? What are all the apps and APIs that are hooked into the system? This is pretty important from a lot of different angles. Security, doing basic security checks. It's important to have this visualized somewhere saving money, right? You can see at a glance what you're paying for, what services you need to cancel. By the way, all these systems are getting pretty good at actually doing computer use, running their own browser. So, at some point, they'll be able to cancel a lot of the recurring and billing for us. I've already been testing it, trying to use the browser within Cloud Code to do certain tasks. It's pretty good so far, and I'm planning to start ramping it up more and more. Then, we have our routine. So, these are actually the things that are running on a daily basis. So if you want to see all the things that are running the cron jobs, the things ingesting new data into the second brain, all those routines, everything is there. Then we also have our various skills. So those are like the skill.md files and everything everything. Now if you're wondering what this doctrine is, again I have Claude naming a lot of these things. So bear that in mind. So I told it to put all of the things like the learnings about the X algorithm. All of the sort of final insights, all of the juicy information that we're squeezing out of this thing into a folder. I'm like, call it something cool. And I was like, oh, I know the doctrine. I was like, all right, whatever, Claude. All right.
So, so far we've talked about our first tool that you need. It's Obsidian. That's this on the left. It's free. It's wonderful. It's local first. And Obsidian is basically just a bunch of markdown files. So again, those markdown files is just text plus a few simple symbols that make it functional and it's really good for cross linking everything. So for example, here is Karpathy's LLM wiki. That's kind of the idea that kick this whole thing off. So let's click on it. This is part of the wiki. It's the database around that subject, that topic. And markdown is super simple. So let's say I wanted to add that he worked at OpenAI. We'll do two hashtags for heading two. I'm going to say used to work at. And notice how it turns into heading two. And I'll say Andre used to work at and I'm going to say OpenAI, but I will cross-link those two documents. I'll do double brackets. Notice how it prefills the other two double brackets. Now everything that you type between those two becomes a link. So I'll type in OpenAI. Notice it already gives me all of the other pages that we have on OpenAI the topic, OpenAI the entity, and various transcripts that include OpenAI in the title. So in this case, we'll say OpenAI the entity. And now that links to that page. So you do this enough times and these stop being just pages and they become a network. By the way, when you don't have any tabs open, if you hit Ctrl+G, that opens up the graph view that lets you visualize that entire network. And if you hit animate here, you can kind of see how page by page by page through cross-linking, the whole thing takes shape as you add more and more data, more and more pages, both raw pages that are just from the internet or from whatever data you're pulling in to actual summaries that are made by the LLM to all of the different stuff that you're adding to it. This slowly becomes that kind of knowledge graph. It slowly becomes your second brain. Once you build this whole thing and you hit that animate button, this is just kind of rewarding. Just watching all that information slowly come together. We're not going to watch it cuz I have too much stuff in here. It'll take forever. But this is your tool number one, Obsidian. And your second tool is Claude Code or ChatGPT or Codeex.
Now, if you've been following this channel, you've seen me use these models through a lot of different interfaces. For a long time, I dealt more or less exclusively with Open Claude. I would use a Telegram to talk to it. I've used the command line interface, tons of different ways of interacting with it. Currently now with this new iteration of Claude Code desktop which is what you're seeing here. At this point I'm pretty much exclusively using this. They added a lot of functionality to where you really don't need to leave this at all. It has Claude Code. You can switch over to the home tab which has your regular kind of ability to talk to Claude as well as Claude co-working kind of on this side. So what I did was I created a second brain directory or folder and I just told Claude to build everything in there. So now whatever new information we're ingesting, it finds a place somewhere in there. So for example, recently Anthropic released this a global workspace in language models. So it's basically talking about if Claude could be conscious on some level or they're not suggesting that that's what's happening. They're just finding a lot of very interesting similarities in how Claude's brain works and how LLMs work. In some ways, it's very similar to how the human brain works. So this idea of a global workspace is something that exists in human brains. It's a mechanism by which we sort of find things that are unconscious and kind of bring it to the surface so that we're able to interact with it in our brains and they're finding something that is analogous or similar in Claude. So definitely kind of a big deal of a of a publishing of a paper. So we want to ingest this into our second brain. By the way, a lot of this should be handled automatically here. I'm just showing you how you would deal with it, how you would do it manually if you needed to. So I'm going to take this URL or just copy this and paste it. And we're going to go into Claude. We're going to say ingest and I'll just paste the link and we'll click go. Another really good feature of Claude Code Desktop is you can just dictate your commands. Click this microphone button and just say what you want it to do. Now, by the way, one recent thing that they've added is an actual built-in browser. So, if you click on that, you can actually just type in whatever URL and it will open within this built-in browser within Claude Code Desktop. So I can go to google.com for example and I can actually tell it to open up web pages, interact with those web pages, whatever you want. But here we'll actually open up a file. This is my second brain just a folder with a number of other folders in it. And at the bottom I had to create this. So that is just this this kind of a visual representation of kind of like the second brain 2.0 that I'm trying to build that is going to have all the skills and routines and everything else on top of it with a different visualization. And notice here as it's building out, ingesting that content from the Anthropic website, it's saying now the ripple. So they're cross-linking all these pages, adding more information about it. So they're adding it to the interpretability concept page and updating all the other entity pages. So I give it one link, it adds it, and now it's rippling through and adding it and interconnecting it within the network.
All right, so that's how we ingest information. That's how we add information to our second brain. All right, but this is where it stops being just a research engine and starts kind of running my life because your second brain shouldn't just know about the news and what's going on in the world. It should know about your life. So, you've probably heard about the Kanban board. So, it's usually something that you have maybe like on the wall you have sticky notes and you move those sticky notes from place to place. Each sticky note is a project or a to-do item that kind of goes through stages. So, maybe going from to-do to doing to done. In Obsidian, it's very easy to create a Kanban board. So, for example, we might have a flow like this if we're doing a content calendar where videos get produced from idea to research to scripted to filmed, edited, and published. Now, currently, my process of creating videos is a lot more chaotic, let's say. And also, I don't script them. And I apologize if I'm stating the obvious. If you ever seen me go on some wild tangent and forget my original idea, you probably can tell that none of this is scripted. But now to try to keep up with the sheer amount of information and releases, I am trying to be a little bit more organized about how I release things, having certain ideas, some from me, some that Claude or some other chatbot comes up with automatically based on the information available on the trending news. So you might have tons of ideas ranked by some metric, how relevant it is, how interesting it is. So let's say I want to create one of these. So recently I published a video called the $20,000 revenue apps with one person teams or something like that. So I would pick it out of my list of ideas and I would move it to kind of this packaging gate. So if it scores good on some metric about how viable it is as a video idea. So it gets put there. Once it's scored, we can move it to research. And by the way, a lot of this stuff can be automated. So if I move it there, Claude can go ahead and start working on it. So here, as you can see, Claude already wrote some suggested hooks for me. The first one is, "Three years ago, I showed you a dad selling Excel formulas for $25,000 a month. The number today made me double-check my sources." In that video, I used the hook about 6 minutes in. The first 6 minutes was me rambling. And then after 6 minutes or so, I got to the hook. Claude tries and does a great job. I still find ways to mess it up, but that's on me. So, let's say once we've done all the research, we move that to, you know, scripting the video. Now again, I don't script my videos, but I do like to have these little cheat sheets with the numbers and the claims, dates, things like that written out that ensures that what I say on camera is accurate. So, I tell Claude that I did a video 3 years ago about this thing. I want to do a follow-up. So, keep in mind, it has the transcript of the video that I did 3 years ago that's in the vault. It knows every word I said on that video. Take a look at this. We covered a product back then 3 years ago in 2023 that was doing $20,000 a month. It was called thumbnailest.com and it was A/B testing thumbnails by the way. And this is why I love Claude. He's insufferable. Look at that. It says the thing your war room now does for free with that grinning kind of smiley face like it's up to something. So, it built that A/B testing thumbnail software for me. And this is it just being kind of smug about it's like, "Oh yeah, like I built that thing for you." I, as you can imagine, did not ask for that to be in the show notes in the in the thing that I'm going to use to prepare for my video for Claude to be like, "What's up?" That was not asked for. But notice what it did here. So, it found what happened to that case study that I did in 2023. What happened to thumbnailest.com? Is it still making $20,000? Is it making more? It found that it sold for six figures in 2024. By the way, since then, YouTube actually launched their own internal thumb testing split testing tool. And as Claude is saying here, the platform ate the moat. And notice what it's saying here. This is the exact platform risk warning from your 2023 video. When OpenAI announced Whisper, everyone building that was gone. So this is kind of why having a second brain like this is so important because it's going back and checking my notes from 3 years ago. It's also updating it from doing internet search, kind of seeing what happened since then to now. It's doing all of that while while being smug about it. What's not to love here?
So, while I don't use the content calendar in that Kanban style dashboard, I'm planning to do that a little bit more to kind of automate more of the research and information gathering, but here is a sponsor flow Kanban board. This I actually do use to help me visualize where I am in the process. These are dummy names, kind of dummy sponsors. They're not real. I can't put the actual sponsors in there because often times there's non-disclosure things. So, I can't use the real sponsors. But this is literally what it looks like. Like we have the script, the the sponsor approval, recording, editing, and all the way once it's approved into publishing. As I get approvals, I just drag it over, and this updates its status. Once it's published, I put it into the done category, and I'm done. This, by the way, can be very easily hooked into some sort of a system that notifies you on your phone through a text message or email if you're running behind on something. If you're keeping up with things like this through Slack, for example, we can pull that information in here as well.
And finally, it brings us to maybe the most important piece of this whole thing. Kind of the point of the second brain. It's called the doctrine. And again, I have to remind you here, I I don't come up with these names. This is all Claude. I think it knows that I like those RPG games. So, it tries to kind of flavor everything in that style. So, as it wrote here, right? So, this is the output layer of the second brain. So, we have raw data flowing into it. The wiki organizes everything that's known and the doctrine is what comes out the other end. It's Fable Analyze. So, this is done by Fable 5, which I found is incredibly good at this kind of deep data analysis and coming up with insights. So, it's Fable Analyze receipts backed actionable strategy. Every doc here answers, what do we actually do? And every claim in here traces back to the data that we've collected. So, the war room gathers intelligence. So, I have this mini PC that's always on. So, it's kind of like a Mac Mini and it just sits there. It's hooked up to Wi-Fi. It doesn't take up a lot of electricity. Doesn't take up a lot of room. It just kind of looks like this. And I think it cost about $200 bucks on Amazon. And it runs 24/7. It never turns off. It doesn't have a screen saver. It's just like a little box that's always on. And so that's sort of the war room, if you will. It kind of just sits there, collects data. It's looking at what's happening on YouTube, on X, on various news platforms. It's the 24/7 kind of home of the agents that just gather data. Then the wiki remembers it, organizes it, cross-links it, all the stuff that we talked about before. And the doctrine decides how we fight. Again, I'm sure we could have used some corporate speak to make these names and describe what they do, but I think I would just like fall asleep here at my keyboard. And then the armory tracks what we're building next. So those future projects, those nice to have that that's all in there largely selected and suggested by Fable. Now, of course, at the end of the day, I'm the one that's choosing what to focus on, what to do. But a lot of the heavy lifting, the analysis, the data collection, all of that is handled by Claude.
By the way, the next big step will be once we have kind of like our to-do actions from the doctrine, we're going to execute on them and collect data about how it works. So, the next, let's say, few quarters, 6 months, 12 months, whatever. That will become its own sort of flywheel where we're putting together strategies, we're executing on them, we're seeing the results, and we're updating in real time how well it's working. So, the longer it runs, the more it compounds, not just in terms of the sheer data that's coming in, but also in terms of the the learning that the system is doing, both in terms of of just what it knows, but also of making strategies, executing them, and and getting feedback. So, sort of that OODA loop. So, for those who are not familiar, so observe, orient, decide, and act. And then it becomes a loop. So observe is the data collection, orient is the wiki and the summaries and in fact the the doctrine then deciding is like kind of like what we're doing with that. They act as the actual action the execution of that strategy and then we're taking that data and we're adding it into the OODA loop. By the way, since Fable designed a lot of this even if Fable does go away eventually we don't get it back a lot of the stuff that it's built will still be helpful. So a lot of this doesn't necessarily rely on Fable to to continue. A lot of the data collection is automatic. But think about this. As time goes on, this system, what happens as better and better models come out? Does the system become better, worse, or stay the same? I think we can safely say that the system not only just gets better the longer it runs, it also gets better and better with stronger and smarter models being released and and used to run the system to to improve the system.
So, let me show you how to build this for yourself. And my advice to you is take the time to do this. This might take some time to set up. Maybe there's going to be some new skills that you have to learn. Learning can and probably should be a little bit uncomfortable. There's a certain feeling that comes with doing new stuff. It's not just like pure joy. There's there's a little bit of a difficulty of resistance. Just push through that. Build this because once it's in place, it starts compounding. It starts growing. I honestly wish I did this on day one whenever Karpathy talked about it. I knew it was a good idea. I should have jumped on it right then and there.
All right. So this is how you build this for yourself. I don't want to say it's super fast. Some of these steps take time. For some of them, you have to wait for for Claude to build some of it, to organize some of it, but you can probably do this in a single afternoon. So first and foremost, you need two tools. Obsidian and Claude Code. So Obsidian is the note taking app, although it's a little bit more than that. So it's over here. Obsidian.md. Here it is. Again, free to start. Most of it is free. It has a huge community. It's a pretty cool tool if I do say so myself. There's a lot to like here and it's free without limits. No sign up required. No strings attached. It's a cool tool by by cool people. Then get Claude Code Desktop. Again, you don't have to get the desktop app. If you're already settled in certain routine, you know what you're doing, do that. But I got to say, if you haven't tried it, they've really been making a lot of good strides with it, and it does seem like it's becoming that super app that we've been waiting for. And I don't know that that might be the the final form, at least for me. I'm really wondering what else they can do to to improve on it. Like if you haven't realized that this is it. I'm using the browser within it to search for the stuff that I need. I can even ask Claude to go and download and install it. By the way, if I need to take this on the road and use it from my phone, I just type in /remote control. I hit enter and then that allows it for me to use it from the Anthropic or Claude app on my phone. And as they say here also to view and control the session from cloud.ai/code. So you're able to remote control this from anywhere.
All right. So you got Obsidian, you got Claude Code. Both are free to start. My recommendation is you do purchase a subscription either Anthropic or OpenAI or whatever chatbot you think is best. But at this point, I feel like you kind of need one. If you understand kind of the significance of what these companies are are building, I would say it's time to invest if you don't yet have a subscription. Then we make the vault. We do that by opening up Obsidian. When you open up for the first time, the button is create new vault or something like that. And that will get you started. Inside you make three folders. Inbox, raw, and wiki. Inside the wiki, you can make folders like concepts, entities, summaries, plus two empty notes, index, and log. Here's the thing. I didn't do any of this. I told Claude to build me this thing. It made all of the folders, all the files, everything, everything, everything. By the way, quick note. Notice how flat the structure is. So, we don't have 50 subpages beneath each page. Everything is pretty flat. This is not me being lazy or Claude being lazy. This is by design. Also, notice that the folders aren't topics. So, there isn't a folder called AI news. The folders are the different layers. What goes in, what it knows, and what it concludes. In fact, some of these things like templates, I shouldn't even have it on here. And the topics, those topics, they live in the links. So, OpenAI is a topic. It's the link that we use to cross-link all of the different pages that have anything to do with OpenAI. Creating too many subfolders, those kind of deep nested structures, it becomes a nightmare for LLMs. Keep it very, very flat.
Next step to creating this would be to write the rulebook aka claude.md. Now again, I didn't write this. Claude did. By the way, I'll have a template down below that you can just download and give to your agent and it will execute everything for you. But the claude.md file that's the rulebook that's the employee rulebook. Every morning Claude wakes up and reads the rule book and goes to work. So for example the raw files those are the immutable source documents. So those are the things directly from the source. We don't change them. The wiki is the LLM wiki. The summaries the entities concepts those are the compounding knowledge maintained by our AI librarian. So just start there. Later you can add all the other things like I added the doctrine etc. The inbox are quick captures from me waiting to be processed. So, this is going to have to be in a different video, but there are ways to do, for example, voice notes where you dictate something or certain emails or even creating a little Chrome plugin to whenever you see something that you want to add to this, you just talk it into your phone or you just record your voice or you just click a button so that it goes to the inbox and then later gets processed by Claude.
Then the next step is optional, but if you wanted to have that graph, that data view, that's a plugin in Obsidian. Same with the Kanban board. The next step is optional and that's adding two plugins, data view and Kanban. Data view builds those automatic tables. Kanban is that Kanban view where you drag little stickers across the board. You can skip those on day one if you want. This works very well without them, but you can find those in settings and they have core plugins and they also have community plugins. You have to turn them on. So approve the fact that those can be used and I'm using here data view and Kanban. And then you start feeding your second brain. You start ingesting data. If you do it 10 times around 10 those dots start to become a web. So take one afternoon to set this up. Then daily just start adding maybe one link a day or whatever you think is best. Ideally you also set up some automation so it pulls the data that you care about. You can do it for your personal tasks for your business or job or school. You can do it for your health or whatever you want. Check out the link below. So I'll have a PDF that kind of explains it. You can read it or just hand it to Claude Code or whatever chatbot you're using and tell it to set it up for you.
So what we built is a Wikipedia that you care about. It's maintained entirely by AI. It can run your work life plus create certain actionable playbooks from your own data, things that you care about. Everything's stored on your computer as basically text files. It's on your computer. You own it forever. You're not tied down to any application, any model. As new things come out, this stays useful. Obsidian or Claude Code, they don't control those files. Those files are text files. No one can lock them down or take them away. Why this matters is because notes that get maintained this way, they actually get used. They're useful. They also don't take a lot of bandwidth for you to to figure them out and organize them. This was an 80-year-old dream at this point that is finally possible. It's your own personal library with a librarian that never sleeps. So, make sure you're subscribed to this channel because more stuff is coming that's going to utilize this and build on top of this. If you have any questions, comments, tips, leave them below. And if anything didn't make sense, definitely let me know so I can kind of troubleshoot and hopefully improve the next time that I'm talking about this. If you made this far, thank you so much for watching. I will see you in the next.