Everyone Asked About My Second Brain. Here Is How It Works.

0:00Hello, my name's Echelo Harrod and

0:01welcome to YAI Matters. And today we're

0:04going to be showing you my second brain.

0:08This is YAI Matters.

0:18Welcome to YAI Matters and everybody

0:21everybody everybody has been asking me

0:23in the past 24 hours what in the world I

0:27use for a second brain and how I get

0:29that thing going. And I thought, "Well,

0:32I might as well if I'm going to say it

0:33once, I might as well say it to

0:35everybody." This video is about my

0:37second brain, how I've compiled it and

0:40what it is, how I use it and does it do

0:44me any good? All right, let's jump right

0:46in. Any good second brain starts out

0:49with a folder where this thing is going

0:52to live, okay? My second brain is mostly

0:55made up of markdown files. There are

0:59some repositories and stuff like that

1:01that kind of seep in there and stuff

1:03like that, but mostly it's made up of

1:06markdown files. Now you say to yourself,

1:08"What is a markdown file?" A markdown

1:10file is pretty simply put a actual file

1:15that is like a word processor file, you

1:17know, that you would get in Word or

1:19something else like that except it uses

1:21symbolization. So it will put two

1:23asterisks and two asterisks and it'll

1:25make it bold or something like that in

1:28order to denote. So it uses markdown to

1:31create these things. It's great for us

1:33because we can read it when they bold

1:35things and when they make things bigger

1:37and stuff like that. But it's really

1:39good for AIs when they have to read what

1:43they need what we need from them. That's

1:45why most skills are made with markdown,

1:47okay? So my second brain is just a

1:51folder, but I use this dude's second

1:55brain as the basis. I found it

1:58interesting. He called it Co-work OS, I

2:01believe he did. Claude Co-work. And here

2:03it is right here. He His name is Paul

2:08Lipsky right there. So just look him up

2:11and on his actual video you will be

2:15>> Then hover over productivity after you

2:17watch his video. It's about a 32-minute

2:19video and if you want to use this second

2:22brain as your basic second brain because

2:25I have three. If you want to use this as

2:27your basic second brain, this is a great

2:30video and a great place to start. And

2:33right down here, if you click the more

2:35right here, you'll find the Google Drive

2:38link to this plugin for Claude. Now this

2:42only works in Claude right here, but you

2:45can set up something very similar that

2:48works with Codex or anything else by

2:51making a instead of a Claude.md file,

2:54which is what this would set up in every

2:56file that's in every folder that is

2:59under this second brain. Instead, right,

3:02of doing that, what you would do is have

3:06the whole second brain as a different

3:09thing, okay? So let me back up. Let me

3:12back up really quick cuz I'm doing this

3:14all off the cuff and kind of explaining

3:15it to you. You want to download this and

3:18plug it into Co-work. You then, if you

3:22want to, you can have Codex analyze this

3:26and see if it can use the same

3:28structure. It will say yes. In this

3:30particular version, it makes Claude.md

3:32files that go in every folder to kind of

3:35inform it about things and it creates a

3:37memory file in there too so that it can

3:40know what has happened in there. And I

3:42would do these things like say update

3:44the memory or update the Claude.md file

3:47to tell them about our progress and

3:49stuff like that. Well, that particular

3:51process worked really well at in the

3:54beginning. As a matter of fact, it

3:55worked really well for a very long time.

3:57And then I found out about Mem Palace.

4:01Let's move on to Mem Palace. Now here we

4:03are inside of Claude Code. I'm in the

4:06desktop application. I did some stuff so

4:09that my permissions would all be turned

4:11off and stuff like that because I like

4:13to live dangerously. What I did was the

4:15reason I started this app was because I

4:17wanted to show y'all the usage of Mem

4:20Palace. This is what I want you to do. I

4:22want you to tell the people what Mem

4:24Palace is. Instead of me reading it for

4:27you, I'm going to have it tell you what

4:30Mem Palace is. Yes, I had to put my

4:31headphones on so that y'all could hear

4:34so that I could hear it too.

4:35>> Here's what Mem Palace is. Mem Palace is

4:38a personal knowledge storage and

4:40retrieval system. Think of it as a

4:42searchable structured archive of

4:44everything you've ever thought, said, or

4:47built across AI sessions. It's organized

4:50like an actual memory palace from the

4:52classical mnemonic technique. Palace,

4:55the top-level container, you can have

4:56multiple. Wings, major categories. For

4:59example, Claude sessions, D second brain

5:02content, Codex sessions. Rooms,

5:05subtopics or projects within a wing.

5:07Drawers, individual memory units, the

5:10actual stored knowledge. EJ currently

5:12runs two palaces on an external drive,

5:14the archive. 10,000 plus drawers across

5:17AI sessions, text messages, Apple notes,

5:21historical imports. The working palace,

5:23current D second brain content. You

5:25interact with it via CLY or through a

5:27custom-built web GUI at the core use

5:30case. Instead of losing everything when

5:32an AI session ends, you mine those

5:34sessions into the palace and can query

5:37across all of it later. What did I

5:38decide about X last month? Becomes

5:41answerable. Yes. As a matter of fact, it

5:43not only becomes answerable, it gives

5:46you infinite ability to look back into

5:48your past. You can look at your text

5:50messages from the past. You can look at

5:52your Apple notes from the past. You can

5:54look at so many things from the past.

5:56Mem Palace. That's what I like to call

5:59long-term memory.

6:01So I used to start I I started out with

6:04Co-work OS and that was kind of working,

6:07but it's a little cludgy because you

6:09have to constantly tell it to update the

6:13memory, update the memory, update the

6:15memory of wherever you are in that

6:17thing. It may not trickle or reach the

6:19top level of the agents. It may not do

6:22that because it's stuck in a subfolder.

6:25So I was having these moments where I

6:26was missing things that I had talked

6:29about specifically. That was one of the

6:31main reasons why I got attracted to the

6:33Mem Palace because the Mem Palace you

6:35could it would mine all of my

6:37information, everything I'm talking

6:39about. Take all of that information,

6:41make it infinitely available to me,

6:43infinitely easily available to me to

6:46kind of mine and do anything I want with

6:48it. So that's what I did. I was working

6:52with the Mem Palace and then I realized

6:55it is such a as far as second brains go,

6:59it works really great for long-term

7:00memory. As a matter of fact, when you're

7:02constantly mining mining things into the

7:05Mem Palace, it takes on all of those

7:07sort of extra searches and stuff like

7:10that. So that's my long-term memory. But

7:13I was looking for some type of memory

7:16that could be like better. And

7:18short-term memory, something that I

7:20could easily

7:21add things to and all those things. And

7:24in steps Mr. Andrej Karpathy. If you

7:28don't know what this is, this is GitHub.

7:31And if you wanted to, you could go to

7:33GitHub and you could look up fufikfak.

7:36That's right, that's me. I'm fufikfak.

7:39So f u f i c f a c. That's fufikfak, all

7:44right? And basically future fiction

7:47factory.

7:48I have 16 public things that are out

7:52there in the public that you could use,

7:54that you could actually use and build

7:56yourself, that I've built, that I've

7:58given away to everyone. These

8:00repositories are great way to save

8:03somebody else's repository. What we're

8:05going to do now is we're going to go and

8:06we're going to look for. And this is

8:09Andrej Karpathy. He basically built the

8:12entire artificial intelligence over at

8:15Tesla. He was a founding member of

8:17OpenAI. He came he went back to OpenAI.

8:19He's now doing his own thing. And one of

8:22the new things that he did was introduce

8:24something called a wiki. He's got oh so

8:28many stars on everything that he

8:30basically introduces. You can see he's

8:33got nano something Karpathy, his blog,

8:37which he obviously produces very little

8:40actually.

8:41>> [laughter]

8:41>> Well, actually he probably produces a

8:43lot. But what I'm looking for is oh,

8:45he's got a LLM Council from 2025. And

8:49this is his LLM wiki. It's more like a

8:52concept, more like a thought

8:55>> [laughter]

8:55>> than an actual GitHub repo. It's under

8:59the gist, GitHub gist. So you can it's a

9:02easy way to share files and stuff. And

9:04this particular gist right here, LLM, a

9:07pattern for building personal knowledge

9:11based using LLMs. This is an idea file.

9:14It is designed to be copy-pasted into

9:16your own LLM, whatever LLM it is. It's

9:19goal is to communicate the high-level

9:21idea, but the agent will build out the

9:24specifics in collaboration with you.

9:26Core idea, most people's experience with

9:29LLMs and documents look like rag. You

9:31upload a collection of files, the LLM

9:34retrieves relevant chunks at query time

9:37and generate an answer. This works, but

9:39the LLM is is rediscovering knowledge

9:42from scratch on every question. There's

9:45no accumulation. Ask a subtle question

9:48that requires synthesizing five

9:49documents and the LLM has to find and

9:52piece together the relevant fragments

9:54every single time. Another reason why

9:56people have been building a lot of

9:57skills. Nothing is built up. Notebook

10:01LM, Chat GPT file uploads, and most rag

10:04systems work this way. The idea here is

10:07different. Instead of just retrieving

10:09from raw documents at query time, the

10:12LLM incrementally builds and maintains a

10:15persistent wiki structured interlink

10:16collection of markdown files. Remember

10:19we talked about markdown files earlier?

10:20That sits between you and the raw

10:23sources. When you add a new source, the

10:26LLM doesn't just index it for later

10:29retrieval, it reads it, extracts the key

10:31information, and integrates it into the

10:34existing wiki, updating in- entity

10:37pages, revising topic summaries, noting

10:41where new data contradicts itself. All

10:44of that stuff happens inside of this

10:46wiki. Now, I can't zoom in on this, but

10:48I sure as heck can have 11 Labs explain

10:52to you what this is. First, it's going

10:55to check the wiki. And let me explain to

10:57you one real quick thing. I made an

10:59innovation. I want you to hear what that

11:02innovation is. Here's what the wiki has

11:04on EJ Palace wiki. This is me.

11:07>> The key innovations you've added on top

11:09of Karpathy's pattern. So, what I did

11:11was I changed the narrative a little

11:13bit. I liked what I liked Karpathy's

11:16wiki. I liked the I used to like the

11:19Co-worker OS, but it was cumbersome. I

11:21really like Mem Palace. I like it for

11:24how easy it is to look up information

11:27about everything about my Every time I

11:30go to work on something and I need to

11:32think about the history of the moment, I

11:34try to weave that into my stories. This

11:37is perfect for all of that stuff. But,

11:40this is my innovation.

11:41>> What Karpathy built, a curated

11:43approximately 5% synthesis layer, a wiki

11:46maintained by an LLM from your raw

11:48sources. What you added? Your

11:51innovations, Mem Palace integration.

11:53Karpathy's pattern has no long-tail

11:55memory. You plugged in Mem Palace,

11:57Chroma DB plus SQLite as a searchable

12:00100% vector layer underneath the curated

12:04wiki. The wiki stays the smart 5%. Mem

12:07Palace catches everything else.

12:08>> Infrastructure is the prose for an LLM.

12:12It's the it's what it reads. It's not

12:14code. It's it's it's it's sweet sweet

12:17release. That is what you are doing is

12:21read is writing these beautiful

12:24wonderful markdown files. And those

12:27markdown files in turn are going to do

12:29you just fine in creating Karpathy's

12:33wiki, which sits on top of the Mem

12:35Palace, and then I use the Mem Palace to

12:39mine the sessions of the brain, and then

12:43inject those into the inbox so that they

12:47in turn get spread amongst all of the

12:49agents. Very complicated, I know, but I

12:52thought this would be a good opportunity

12:55to kind of explain how my brain has been

12:59built, my second brain has been built.

13:01And that's a look at the other brain. I

13:05will try to make sure that I have links

13:07to Karpathy's LLM wiki, to Milla

13:11Jovovich's Mem Palace, and to my very

13:15own EJ's Palace wiki. If you want to use

13:18any of those things, there'll be links

13:20down in the description below. All

13:22right, ladies and gentlemen, that's it

13:24for me. And if you're really really

13:26interested in what GPT-5 has to offer,

13:31there's a video

13:33right there.

13:34>> [laughter]

13:35>> I wish you put that. There's a video

13:37literally video right there. There we

13:39go. Right right there. That's it. That's

13:41video. All right. Bye.