YT YouTube2Text

YouTube2Text → Transcripts → Comment avoir un SECOND CERVEAU parfait sur Claude ? (le système simple)

Comment avoir un SECOND CERVEAU parfait sur Claude ? (le système simple) — Transcript

by Melvynx · 2,871 words · 411 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00This video is very special and it will
0:02surely change your life. I am speaking
0:05to everyone who uses Claude, Codex, or
0:08other agents, who isn't technical, and
0:11who wants a kind of second brain. So,
0:15I'm talking here about a place where
0:17you can chat with AI, and the AI knows
0:19what you do, what you like, what you're
0:21working on right now, your projects; it
0:23has access to your Gmail, your YouTube,
0:25can edit your videos, and so on. It’s
0:29a dream many people want to achieve,
0:32and it's what we can call a second
0:34brain, a place where you can offload
0:36your mind and have your entire body of
0:39work. It's also what we can call a work
0:42folder. And by doing calls with quite a
0:45few non-technical people, I realized
0:47that 80%of people had, uh, work folders
0:49—what I just mentioned—that were
0:51completely awful, riddled with errors,
0:53with complexity where they had copied
0:56and pasted systems from dozens and
0:58dozens of different videos into their
1:00thing. And this video is dedicated to
1:03those people, adapted to the models we
1:05are working with now. And it's going to
1:08completely change how you work with
1:11extraordinary simplification. So I
1:15suggest we get started on this topic of
1:17the second brain. The goal here is not
1:21to build an overcomplicated mess, and I
1:23know many of you either have nothing at
1:24all or have an overly complex system.
1:28So, where does this come from? It comes
1:31from 1945 with the Memex, an augmented
1:33memory where you could store your books
1:35, your letters, your notes, and be able
1:38to retrieve everything in an instant.
1:41That was the objective. But now, the
1:44dream is to have a "brain" in English,
1:46or a brain in French, with Claude or
1:48any AI agent that reads, searches, and
1:51summarizes all your information into
1:53one folder. And what you're sold on
1:57YouTube is often a 3D brain, often with
1:59an Obsidian graph, a vector database,
2:02or multiple databases and a 10-step
2:04pipeline to create something. Let me
2:08tell you that in my opinion, all that
2:10is an overcomplicated mess, something
2:12that isn't sustainable in the long run.
2:14So today, we are going to talk about
2:16the simplest yet most effective system
2:18that I have implemented for a huge
2:20number of people, and which will
2:22simplify your life. What you need is a
2:25folder. A folder with your info. Offer,
2:28client, decision. When an AI goes to
2:32search for information, it will simply
2:34go into this folder, which must be
2:36well-organized to retrieve the right
2:38information, and it will be able to
2:41answer you using that information. That
2:43is what we call a RAG. Often, when we
2:47talk about RAG, it's a term that has
2:49been used, reused, and used over and
2:51over again. We often talk about a
2:54vector database, but here we are going
2:56to do a much, much simpler RAG, because
2:59a RAG is just retrieval-augmented
3:01generation. We often talk about vector
3:04databases, but in fact, a vector
3:06database is just the most common way to
3:08do it. But with RAG, the idea is just:
3:10search first, and then we have the
3:12result. And so, that is exactly what
3:14Claude now allows you to do. You open a
3:17new conversation, you tell it: "write
3:18the launch email for my course"—an
3:20example that relates to me—and then
3:22it says: "With pleasure, which course,
3:24what price, for whom, and what tone?".
3:27So, in fact, it doesn't know who you
3:28are. In every conversation, it has
3:30different things. You are using the
3:32application, so it will ask you
3:33questions and you will have to clarify,
3:35and so on. And so you will have to
3:38manually re-paste the course you are
3:40selling, the course you want, and so on
3:42and so forth, with a prompt for every
3:44single thing. The idea here is to keep
3:48pasting everything in to get Claude to
3:51answer. The problem is that it is
3:53expensive and slow, less accurate, and
3:55worse, to it, everything has the same
3:57value. So your old 2024 price or your
3:592026 price, well, to it, both are
4:01useful pieces of information. When you
4:04work like that, you are using the
4:05desktop. You have to understand that
4:07the AI has a kind of desktop. That is
4:09its context window. It's a bit like
4:10what you see on your screen; you see,
4:12when you work on your computer, you
4:13have your desktop. It's big, but it's
4:15not infinite. And then, we will have a
4:17filing cabinet. A filing cabinet, that
4:18will be our folder, that will be our
4:20brain. And there, we will have all our
4:22projects stored. And this cabinet will
4:25be read on our desktop. So when you, on
4:28your computer, come to work on your
4:30YouTube channel, you don't open all the
4:32pages at once. You don't open your
4:35YouTube channel, your Instagram account
4:37, and so on; you only open YouTube.
4:38Well, that is exactly what you can
4:40imagine right here. So, the goal is for
4:44Claude to first search using a brain.
4:47When you ask Claude what my current
4:49price is, it will look for "price" in
4:52the database to find the 2026 offer
4:54with the "about" file, and it will read
4:56the files. So clearly here, reading
5:00files is actually superior to a
5:03database. According to Boris, creator
5:05of Cloud Code. Searching for words in
5:08files is more optimized than bothering
5:10to store dozens of pieces of
5:12information in a standard database. A
5:15file, you read it. So you can read it,
5:18meaning both a human can read the file
5:19and an agent can read the file. So it
5:22is agent and human-friendly. For a
5:24database, you generally need code and a
5:26server, whereas a text file is just a
5:28folder. To correct information, you
5:30need scripts. Whereas with a text file,
5:33you can even manually change the info.
5:36And if you change AI, well, with text
5:38files, it works with any agent. Files
5:41are superior to a database right here.
5:44So when you ask how much my training
5:46costs, it will look for the word price,
5:48it will look for the word rate, it will
5:51look for the word euros, and that will
5:53allow it to answer that Editing AI Pro
5:55is X according to your 2026 offer.
5:57Right here. How do we do that? Well, we
6:00manage a folder. Now, this interface
6:03might look scary, but it's just a file
6:05directory. It's exactly like when you
6:08open the Finder. So here we can see
6:11that we have folders with thumbnails,
6:13stuff, and so on. Well, it's the same
6:16view, except that here, we display it
6:18in one dimension right here. So we'll
6:21have our brain right here, which is
6:23right there with an agent.md file. If
6:27you don't know what an agent.md file is
6:29, MD is simply a file read at every new
6:31conversation that explains to it who
6:33you are and our project rules. We will
6:36have project folders with, for example,
6:39the training, your YouTube channel,
6:41your real estate project, etc. We will
6:43have notes with journals, IDs, and
6:45to-dos. And we will have the archives.
6:47So that's for once a folder or project
6:49is finished, we move it to the archive
6:51right here. No database, no
6:53subscription, just text. And so in this
6:57about.md file in each project, it will
6:59describe where the project stands, the
7:01decisions, etc. So the agent.md file,
7:03as I mentioned just before, is read at
7:05the start of every conversation right
7:08here. It will be read, which allows you
7:10to provide your basic information. The
7:12goal is not to give it too much
7:14information. The goal is to give it
7:17just enough information. And then, when
7:20you open a new conversation and ask it
7:22a question, it will generally read the
7:24"about" file, which is what we call an
7:26entry point, so that when you talk
7:28about a project—for example, "work on
7:30my course"—I'll use this example
7:32because it's the simplest one. Editing
7:35AI Pro, a course on how to do video
7:38editing with AI. Well, it reads the "
7:42about" file and tells you, "Here is a
7:44draft for your course." In this "about"
7:49file, we will have a summary of what's
7:52happening in the course. We will have
7:56where you currently stand with your
7:58current information. We will have the
8:01decisions you might have made right
8:02here. And we will have the next step,
8:05with the launch email sequence, for
8:07example, that you need to do here. So
8:09here, we have all the basic data, but
8:11it's more of an overview. Then, we will
8:13have other files inside that describe
8:15the other steps. Summarized by an AI,
8:19about.md. So, the overly complex
8:21version involves files that summarize,
8:23repeat themselves, and so on. The "
8:25brain" version means having a clean
8:27about.md file that is always updated
8:29and follows specific rules. We will
8:31keep track of every decision you make
8:33along with its reason so the AI doesn't
8:35repeat the same mistakes. Instead of
8:37complicated searches, we will simply
8:39have Claude perform the search on its
8:40own. And we will see that later.
8:44Instead of copying tons of content, we
8:46will just have tools connected to our
8:48AI so it can use everything via what we
8:50call skills. And we will have automatic
8:53backups with "don" and automatic
8:55backups that allow us to manage all of
8:57this. No need to put everything in,
8:59right? In your brain, we will have your
9:02offer, your prices, your target clients
9:03, your processes, your decisions, your
9:05transcripts, your notes, your ideas,
9:06and so on. But we won't repeat Stripe
9:10info, calendar info, or email info; all
9:14of that will go through tools. To get
9:17started on your end, you can go to the
9:20first link under this video at
9:22mlv.sh/fx to receive a masterclass on
9:24everything you can do with this brain
9:27system and ultimately everything you
9:29can automate with AI. I really invite
9:33you to click on this first link under
9:34the video to sign up immediately. And
9:36otherwise, you can simply use the
9:38interview right here. So with an "
9:39interview" prompt, tell it to interview
9:41me to fill in the project name. Ask me
9:43one question at a time, client offer,
9:45price, goal, then write all that in
9:47process non about. Right here, you will
9:50then be able to put in your
9:52documentation and do what we call voice
9:54dumps. So you do a 5-minute voice
9:56recording and it saves everything. Your
9:58conversations are kept on Claude and on
10:01our tool for only 30 days. And that's
10:04why we want to save and extract the
10:06essentials into our brain. We are going
10:09to have daily notes. So every day, we
10:11will have the projects you're working
10:13on for your memory. This is actually
10:15your brain, it will save all of that on
10:17its own. You don't need to do much at
10:19all. And it will save your ideas and
10:22to-dos right here. And when you ask it,
10:24"What did I do this week?" It will be
10:26able to answer you. It will be able to
10:28read the notes to make decisions, and
10:30so on. You'll have the/done commands,
10:33but you can also have by default here,
10:36with the setup I have, you can tell it
10:38that if you haven't continued a
10:40conversation for 20 minutes, it will
10:42automatically save and that will allow
10:45you to create about files, roadmap
10:47files, journal files that then allow
10:50you to save everything on a big hard
10:52drive online called GitHub. We're going
10:55to have rules everywhere. So the first
10:58point is agent/haagent.md, which is how
11:01it talks to you and your constraints,
11:03always read. Next, we have the brain
11:06notes with the agent.md, which contains
11:08the details of who you are and your
11:09organization. Then after that, we have
11:11about files which are only read on
11:14demand. The advantage is that all of
11:16this consists of files that belong to
11:18you. Your brain is just files that are
11:20on your computer. So, you can use them
11:23with CL.code, Codex, or even Hermes on
11:26a VPS. So if tomorrow you want to
11:29change AIs, if tomorrow you want to use
11:31anything else, everything works
11:33perfectly without any modifications. We
11:36will have several recipes and that
11:39follows with what we call skills.
11:41Create project, project, cleanup, and
11:44skill manager. So, skills are simply
11:47processes that we will be able to
11:48repeat. And instead of saving prompts
11:52that you copy 100 times, you will
11:54create skills so that the AI finds them
11:56on its own. Now, a second very
11:59important point is the tools. Twelve
12:02months of emails copied and recopied
12:04every 15 minutes is complicated.
12:06Actually, what we want to do instead is
12:08copy nothing. It's about not
12:09duplicating information because your
12:11projects are what are normally in your
12:13head. Here, we are going to put them
12:15into the AI. But then, when we build
12:18tools, we will connect Claude directly
12:19to Gmail, Notion, and Stripe so it has
12:21access. This way, Claude will access
12:24your data on demand, and we won't be
12:26copying it. If you have a data source,
12:28we won't duplicate that data source.
12:30This way, when you ask it where your
12:32YouTube channel is, it can read your
12:34YouTube file and read the info from
12:36your YouTube channel. And how many
12:38sales this week? Well, it will read the
12:41file and go get the information from
12:43Stripe to have up-to-date data. It is
12:45very important. Right here, Claude
12:47manages to automatically link to a
12:49cloud or ChatGPT, etc. One piece of
12:51info in one single place. So if the
12:53price changes, you just modify one line
12:56. And you have to be careful because AI
12:58tends to duplicate information. So here
13:00, we will try to ensure that
13:02information is atomic. All decisions,
13:05along with the reason why you chose to
13:07decide that, will also be saved. Slash
13:10allows you to save, and cleanup allows
13:13you to tidy up. So your brain will have
13:16many files; it will have your "brain"
13:19folder with all the connectors here. It
13:22is actually very important to have all
13:23the connectors. So here, I only put
13:25three with Notion, Mail, and Stripe.
13:27But in reality, I have dozens and
13:29dozens of connectors. I have my
13:32connector for my emails, the connector
13:34for my training platform, and a
13:36connector that allows the AI to publish
13:38my modules, do video editing, and so on
13:40. Now, I'm going to finish this video.
13:43There is a connector that will take the
13:45video I am currently recording, edit it
13:47, and publish it on YouTube without me
13:49needing to do anything. Next, the step
13:52after that, which we won't discuss just
13:54yet, is to put all of this on a VPS so
13:56your brain can be synchronized online
13:59and you can chat with it directly via
14:01WhatsApp, Telegram, Discord, and all
14:03messaging apps. And right here, it will
14:06then allow you to always be able to
14:08chat and have your data even there. So
14:10the steps to take now are, first,
14:12create your brain. For that, you can go
14:14to mlv.sh/fx to get a mini-training
14:16course to help you get started, and
14:18above all, a masterclass to understand
14:20everything you can automate right now,
14:22regardless of your industry, with AI.
14:25Second point, get yourself interviewed,
14:27and third point, connect all the tools
14:29you have immediately to the AI to get a
14:31real masterclass in results. Your
14:33second brain is a folder with an
14:36agent.md and an about file. You let it
14:38search on its own and you keep it clean
14:40. If I show you my folder, it's exactly
14:44what I showed you on the other slide,
14:46but we're not going to have 12,000
14:48folders, you see, we'll keep only a few
14:50. It's very important. So here, when
14:53you look at the number of folders we
14:56have, we have a project folder, a notes
14:58folder, that's all. And I see a huge
15:00number of people with code and
15:02information that becomes complex and so
15:03on. Let's imagine I want to do video
15:06editing. If I want to do video editing,
15:09I won't store that in my brain because
15:11my brain is my workspace, it's my space
15:13for reflection. I don't want to mix the
15:16skills for video editing with the
15:17skills for managing my thoughts, my
15:19data, and my business. So what we're
15:22going to do here is create another
15:23folder. So here, we'll have our brain
15:25folder which will, for example, be in
15:27our documents folder on our computer.
15:29And then, we're going to do video
15:30editing, we'll create a video editing
15:32folder. In this video editing folder,
15:34we'll put our rushes and so on, as well
15:36as the video editing skills that we
15:38will always store in that location.
15:41This way, we'll separate the
15:43information and we'll optimize our AI's
15:45results, which will allow for much
15:47better outcomes, I assure you. So, the
15:50masterclass is at mlv.sh/fx, my first
15:52link below this video. I invite you to
15:54watch it now. Talk to you soon. Bye bye
15:56.

About this transcript

This page contains the full transcript of Comment avoir un SECOND CERVEAU parfait sur Claude ? (le système simple) by Melvynx, generated from the public captions YouTube serves with the video. The transcript has 2,871 words across 411 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.