YouTube2Text

He Creado un Segundo Cerebro con IA… y Es una Locura! — Transcript

by Javier DV - Inversión y Finanzas · 4,995 words · 708 segments · language en · Watch on YouTube

Full transcript

  1. 0:00This is crazy. Imagine for a moment
  2. 0:03that you could create a second brain, a
  3. 0:06complete neural network of all the
  4. 0:09concepts that you want the AI to have
  5. 0:11and that they feed off each other so
  6. 0:14that every time the AI needs to work
  7. 0:17with you, it is an expert who has all
  8. 0:19the information at its disposal. And
  9. 0:22this is not a method that I have
  10. 0:25invented, it is the step-by-step method
  11. 0:27shared by the employee of Antropic
  12. 0:30Carpati, who was the co-founder of Open
  13. 0:33AI and who has provided the method to
  14. 0:35train the AI and create this second
  15. 0:38brain. And this has many implications,
  16. 0:41from being able to have your entire
  17. 0:43business to helping you understand a
  18. 0:45company with all its concepts, how they
  19. 0:48align with each other, all the
  20. 0:50information that we have available,
  21. 0:53digested and ready to be used every
  22. 0:55time we use artificial intelligence.
  23. 0:57Well, one of the things that I was
  24. 1:00doing was using the most important
  25. 1:02academic literature in finance so that
  26. 1:05it could synthesize it all and create
  27. 1:07an expert with all the important
  28. 1:10knowledge of the financial world and
  29. 1:12how all these concepts are related to
  30. 1:15each other. Now, academic papers are a
  31. 1:17bit boring and we are going to try to
  32. 1:20create a kind of second brain with all
  33. 1:22the knowledge of the last decades of
  34. 1:25the best investor of all time and that
  35. 1:27way we will have all the folders with
  36. 1:29all the concepts organized that we can
  37. 1:32read through Obsidian. I am going to
  38. 1:34explain to you step by step and the
  39. 1:36concept of Carpaty's method so that you
  40. 1:38can replicate it in anything you want.
  41. 1:41Industry analysis, company analysis,
  42. 1:43anything you need to study, everything
  43. 1:46to be able to create the right
  44. 1:48architecture to work with AI. Look, the
  45. 1:51main problem with how we use artificial
  46. 1:54intelligence today is that surely you
  47. 1:56and I and everyone we do is that we end
  48. 1:58up with the one we use. For example, I
  49. 2:00usually use the cloud, we have several
  50. 2:02videos on the channel of practical
  51. 2:03things that I have done and we give it
  52. 2:05all the information. For example,
  53. 2:06imagine that we are analyzing a company
  54. 2:09and we upload the latest reports of the
  55. 2:11company, the 10K, the 10Q, an analysis
  56. 2:13of experts and different things and we
  57. 2:15ask it to analyze. The problem with
  58. 2:17this is that the context remains in
  59. 2:18this conversation and when the
  60. 2:20conversation ends, that context tends
  61. 2:22to disappear. It is true that the AI
  62. 2:24can save memory and recover certain
  63. 2:26concepts in other conversations, but
  64. 2:28all those files that you have given it
  65. 2:30have not been synthesized and,
  66. 2:31therefore, the next time you want to
  67. 2:33ask it something you would have to
  68. 2:35upload these files again so that it
  69. 2:36could work with them again, which
  70. 2:38involves a great consumption of tokens
  71. 2:40and, obviously, repeating everything,
  72. 2:42uploading the files again; or you could
  73. 2:44make a project with all the files
  74. 2:46inside, but each time you ask it about
  75. 2:48something, the AI is going to have to
  76. 2:49look at those files over and over again
  77. 2:51. And that is the problem that Carpati
  78. 2:54raised a few months ago. He said that
  79. 2:55the problem is that everyone gives it a
  80. 2:57collection of files, different things
  81. 2:59and that this works, but that each time
  82. 3:01you have to start from scratch. There
  83. 3:03is no accumulation. Every time you ask
  84. 3:05it a new question, the AI has to find
  85. 3:07the relevant pieces and fragments, and
  86. 3:09here he emphasizes that nothing is
  87. 3:11being built, but proposes a different
  88. 3:14idea. Instead of just giving it the raw
  89. 3:17documents each time, what we can do is
  90. 3:20create an architecture that constantly
  91. 3:23builds on top of it and updates pages,
  92. 3:26reviews topics, looks for contradictory
  93. 3:29data to fix it or even improve the
  94. 3:32system. All this sounds a bit technical
  95. 3:35, but let's bring it down to earth. And
  96. 3:37the solution that Carpat proposes is
  97. 3:38the one that he himself is implementing
  98. 3:40, since he shared this in a tweet as
  99. 3:42something personal, saying, "Hey, this
  100. 3:44is how I work." And it went extremely
  101. 3:46viral. That led Carpaty to share the
  102. 3:49guide, the structure of how he does it.
  103. 3:51He published a pattern that is not a
  104. 3:53product or code. It is a way of working
  105. 3:56, especially for those of us who work
  106. 3:58with information, with many documents.
  107. 4:00Instead of having the AI reread your
  108. 4:02documents for every question, you build
  109. 4:04a persistent wiki once, a complete
  110. 4:06three-layer structure that I'm going to
  111. 4:08show you now and we're going to do it
  112. 4:10step by step, and you maintain it
  113. 4:11forever. It says, for example: what
  114. 4:14gets interesting is that once your wiki
  115. 4:17, that brain, is so big (for example,
  116. 4:19mine has about 100 articles and over
  117. 4:22400,000 words), you can ask your AI
  118. 4:24agent any complex question about that
  119. 4:27subject, because it has access to that
  120. 4:29entire directory. And this basically
  121. 4:32works with a three-layer architecture
  122. 4:34and it may seem very technical, but I'm
  123. 4:36going to explain it very simply.
  124. 4:37Observe layer 3, where we would have
  125. 4:39our sources, our documents, and all of
  126. 4:41our raw files. Here what I'm going to
  127. 4:44do is dump all of Warren Buffett's
  128. 4:47letters. Those are raw files in PDF, in
  129. 4:51Word, in whatever, that are not
  130. 4:54modified. That folder is ours, it's
  131. 4:56where our files are saved. Now, this is
  132. 4:58where it gets interesting. That local
  133. 5:01folder will have a cloud.md, MD, which
  134. 5:03is basically going to be a file that
  135. 5:06tells Cloud how to work within that
  136. 5:08folder, within that brain, if you want
  137. 5:11to call it that. Cloud MD will know
  138. 5:13exactly your tastes, your way of
  139. 5:15working, your way of analyzing, in this
  140. 5:17case, the market, your investment
  141. 5:19philosophy and will adapt its work to
  142. 5:21that context to get to know you and
  143. 5:24know how it can best help you. And in
  144. 5:26layer two we have the wiki. The wiki
  145. 5:28belongs to the AI. That's where it
  146. 5:30creates, updates and maintains. In
  147. 5:32other words, what the AI is going to do
  148. 5:34is that each raw file that we give it
  149. 5:36is going to be ingested, it is going to
  150. 5:38be synthesized and it is going to be
  151. 5:40able to create pages that then allow it
  152. 5:42to have that information always
  153. 5:44available to it quickly and how that
  154. 5:46information relates to all the rest.
  155. 5:48Look, here we have the diagram in a
  156. 5:49little more detail and now when we do
  157. 5:51it step by step you are going to
  158. 5:52understand it much better. Folder one
  159. 5:53contains all our files. We create this
  160. 5:56on our computer and we can create the
  161. 5:58raw files we want. Layer number two
  162. 6:00belongs to artificial intelligence.
  163. 6:02This is where the artificial
  164. 6:03intelligence will create those pages,
  165. 6:05those concepts and summarize the
  166. 6:07information. And it will also have an
  167. 6:09index and a log. And Carpati says this
  168. 6:10is very useful for the following. The
  169. 6:12index will basically be a file that is
  170. 6:14basically like the index of a book. It
  171. 6:16tells the artificial intelligence
  172. 6:18everything that is in this layer two
  173. 6:19and it basically organizes it. Each
  174. 6:21page is listed with its link, a quick
  175. 6:23summary of what each page shows,
  176. 6:25different concepts, sources, etc. What
  177. 6:27the artificial intelligence will do is
  178. 6:29first read the index and then search
  179. 6:31within all the files for the relevant
  180. 6:33pages. And then the log is basically a
  181. 6:35file that is kept up to date, which is
  182. 6:38chronological and where it will put
  183. 6:39everything that has been done,
  184. 6:41everything that has been changed within
  185. 6:43this brain. That way we always have a
  186. 6:45history of all the changes that have
  187. 6:47been made. With this, what we achieve,
  188. 6:49people, is a brutal context, that every
  189. 6:51time we work with AI, it has at its
  190. 6:53disposal all that brain, those
  191. 6:55connections that we have seen to be
  192. 6:57able to gather information or help us
  193. 6:59in the best possible way. According to
  194. 7:01Carpati, what are the applications of
  195. 7:03use? Well, it can be personal to have
  196. 7:05your goals, health, personal
  197. 7:06development topics, you can keep track,
  198. 7:08for example, if you keep a diary,
  199. 7:10articles, podcasts. It also helps you
  200. 7:12with research and this is the most
  201. 7:14interesting part in my opinion because
  202. 7:16it allows you to create a brain on a
  203. 7:18topic that you want to investigate and
  204. 7:20to be able to gather a lot of
  205. 7:21information and synthesize it. For
  206. 7:23example, if you are studying something
  207. 7:25specific or you are learning about a
  208. 7:27specific industry, about a sector,
  209. 7:29about a topic, well, you can create a
  210. 7:31kind of brain focused on this. The same
  211. 7:33with books, the same with teamwork or
  212. 7:36even business. So now we are going to
  213. 7:38see what application we can give it. As
  214. 7:40I was telling you, I've been
  215. 7:42experimenting with academic papers to
  216. 7:44get the concepts down, to have detailed
  217. 7:46information from all the academic
  218. 7:47studies out there and the real
  219. 7:49foundations of finance, but now we're
  220. 7:51going to do it with something a little
  221. 7:53cooler because I know all this is very
  222. 7:55boring. Let's do it with Warren Buffett
  223. 7:57. We're going to try to create a second
  224. 7:59brain that compiles all those years of
  225. 8:01teachings that Buffett has written in
  226. 8:03his annual letters. And how do you work
  227. 8:05? What platforms do you work with? Well
  228. 8:07, Carpati typically uses Obsidian,
  229. 8:09which is, at the end of the day, just a
  230. 8:12reader for MD notes files, isn't it?
  231. 8:14Markdown. The good thing about it is
  232. 8:16that if Obsidian ever disappears, those
  233. 8:18Markdown files are text files that any
  234. 8:20computer can read, so you could use
  235. 8:22them in another application. And
  236. 8:23Obsidian is basically just that note
  237. 8:26file reader that also makes that
  238. 8:28connection map for us that we can then
  239. 8:30have. But Obsidian basically only reads
  240. 8:33Bolt, which is this architecture. And
  241. 8:35then with Cloud Code, personally,
  242. 8:36that's who I'm going to work with to be
  243. 8:38able to update all this. To use Cloud
  244. 8:41Code, you need at least a Pro account.
  245. 8:43So I'm going to look for the Versai
  246. 8:45Hataway shareholder letters. I'm going
  247. 8:46to go to the Versai Hataway page. And
  248. 8:48as you can see, here we have all the
  249. 8:51years. If I click, it takes me directly
  250. 8:53to the PDF and I can download it with
  251. 8:55the current year. And here, as you can
  252. 8:57see, each letter with Warren Buffett's
  253. 8:59lessons for that year. This is
  254. 9:01obviously going to take us some time,
  255. 9:02guys, it's not something we can do
  256. 9:04quickly and that's it. We have to have
  257. 9:05it well structured, well organized, but
  258. 9:07once we have it, I think we can give it
  259. 9:09many uses. Okay, so I've already
  260. 9:11downloaded the letters from 2024 to
  261. 9:132008. For now, we're going to stop
  262. 9:14there and then I'll continue feeding it
  263. 9:16, because you can continue feeding it
  264. 9:18with many more raw files later. That's
  265. 9:20the great thing, that you scale this
  266. 9:22structure up more and more. Then, I
  267. 9:24could add the missing letters from
  268. 9:27previous years. Okay, for now we're
  269. 9:29going to leave these files here and
  270. 9:30what we're going to do is go to Cloud
  271. 9:32Code and then to use Cloud Code you can
  272. 9:34do it in different ways. You can do it
  273. 9:36through your computer's terminal if you
  274. 9:37want, you can do it through the
  275. 9:39application, which I think I'm going to
  276. 9:41do with that in this video because it
  277. 9:42would be the easiest. Or you can
  278. 9:44install, for example, a tool like
  279. 9:46Visual Studio Code, which is the one
  280. 9:48many people use, and there integrate
  281. 9:50Cloud Code and use it through this tool
  282. 9:52. In my case, I'm going to use the
  283. 9:54cloud application on my computer, where
  284. 9:56I'm going to go directly, here we have
  285. 9:57chats, cowork and code. I'm going to go
  286. 9:59to the code section and here we would
  287. 10:01be in cloud code. What I also have to
  288. 10:03do is create a new folder on my
  289. 10:05computer. That's going to be the folder
  290. 10:06where we're going to create the entire
  291. 10:07structure. For example, I'm going to
  292. 10:09call it brain buffet. And now what I'm
  293. 10:11going to do is select within Cloud Code
  294. 10:13, give access to this folder that we
  295. 10:15just created. Basically here you can
  296. 10:17select by opening a folder and select
  297. 10:20the folder that we created, brain
  298. 10:22buffet. Trust the workspace, we click
  299. 10:23on it and we would already have the
  300. 10:25folder assigned. Before creating the
  301. 10:27Cloud MD file under Carpati standards,
  302. 10:29what we are going to do is create a
  303. 10:31subfolder and we are going to call it
  304. 10:33row. And this is where our raw files
  305. 10:35are going to go. So, here in the row we
  306. 10:37are going to dump all of Warren
  307. 10:39Buffett's letters. Basically, I am
  308. 10:41going to take them from the other
  309. 10:43folder and dump them here. And now we
  310. 10:45would have another subfolder inside the
  311. 10:47brain buffet folder, which is row, with
  312. 10:49all the letters in PDF. And here we
  313. 10:51could continue adding files, since this
  314. 10:53folder belongs to us and these files
  315. 10:54are not going to be touched. And now
  316. 10:56inside Cloud Code and with the folder
  317. 10:58already selected, what I am going to do
  318. 10:59is give it a little context. I am going
  319. 11:01to tell you that here below you have
  320. 11:03Carpaty's LM Wiki pattern. I am going
  321. 11:05to attach it. Now I am going to tell
  322. 11:07you where to download it. Build me a
  323. 11:09second Warren Buffett brain. This is
  324. 11:11kind of my goal. And you give it a
  325. 11:13little context. Here you can explain
  326. 11:15whatever you want. The more precise you
  327. 11:17are, the better, because that way the
  328. 11:19cloud MD, which is the file that Cloud
  329. 11:21is going to use to follow a series of
  330. 11:22rules within this entire architecture,
  331. 11:24will obviously be more detailed and
  332. 11:26will work better. And here I told him,
  333. 11:28since I am an analyst, my focus is
  334. 11:30long-term investment, second order,
  335. 11:32create the cloud.md, the structure and
  336. 11:34then read all the row files and build a
  337. 11:36structured and interconnected financial
  338. 11:39brain on Puffet. My idea is to have at
  339. 11:41his disposal all the knowledge he used
  340. 11:43in his letters so that he can use it in
  341. 11:45my investments. Stop and ask me
  342. 11:46whatever you need before starting. This
  343. 11:48is important, okay? So that he can ask
  344. 11:50us questions about things that he wants
  345. 11:52to know how we want to do. This is key
  346. 11:54to putting it. And now down here I am
  347. 11:56going to copy Carpaty's pattern. For
  348. 11:59this I am going to leave the website in
  349. 12:00the description. As you can see, he
  350. 12:01made it public 3 months ago. And here
  351. 12:03he basically explains the idea of how
  352. 12:05this architecture works. This is simply
  353. 12:08the idea of how it works, it is not
  354. 12:10code or says how to do anything, simply
  355. 12:12how to structure the idea. Therefore,
  356. 12:14if we give the idea to Cloud Code, it
  357. 12:17will know how to approach this
  358. 12:19architecture to follow Carpati's steps
  359. 12:21exactly. Therefore, we are going to
  360. 12:23click on row here to be able to copy it
  361. 12:25. And what I am going to do is copy all
  362. 12:27this. Control C, I go back to cloud and
  363. 12:30I am going to paste it down here. And
  364. 12:33so we have all this. I press enter.
  365. 12:36Here we can select the language model.
  366. 12:37For now I am going to go with opus,
  367. 12:39although Fabel is now available again,
  368. 12:40but for now I am going to go with opus
  369. 12:42because I do not think it needs a lot
  370. 12:43of capacity to do this. It is going to
  371. 12:45ask us for access to the folder, in
  372. 12:47this case we have created a new one,
  373. 12:48the Buffett brain folder, which is the
  374. 12:50one we have given it access to and it
  375. 12:51is going to start working. Now what it
  376. 12:53is doing is creating the Cloud MD file,
  377. 12:54which is going to be the series of
  378. 12:56guides that it is going to use. Each
  379. 12:57time Cloud accesses a file within this
  380. 12:59folder, note, we have only given it
  381. 13:01access to this folder, it is going to
  382. 13:02ask us for different permissions. There
  383. 13:04, if you want, you can enable the
  384. 13:06bypass so that it doesn't ask for your
  385. 13:07permission and does it directly, or you
  386. 13:09can simply confirm permission each time
  387. 13:11. See? Extract information from the PDF
  388. 13:13, well, you allow it and that's it. In
  389. 13:15this case, what it's doing is reading
  390. 13:16the PDFs. Look, it's already asking us
  391. 13:18the first questions. In what language
  392. 13:20do you want to build the wiki? The
  393. 13:21summary pages, the principles, and the
  394. 13:24analysis. Do you want it in Spanish
  395. 13:25with the citations in English? Do you
  396. 13:27want it in full English? Do you want it
  397. 13:29in full Belgian? Each page with its
  398. 13:31version in both languages. In my
  399. 13:32specific case, I'm going to go with
  400. 13:33English, but the good thing here is
  401. 13:35that you can configure it however you
  402. 13:36want. If you want everything in Spanish
  403. 13:38, well, in Spanish, I'm going to choose
  404. 13:40English by pressing two or by clicking,
  405. 13:42but by selecting two and pressing next.
  406. 13:44Okay. It tells me, "What should the
  407. 13:46brain be optimized around?" Choose
  408. 13:47those that provide most value.
  409. 13:49Principles and mental models,
  410. 13:51application layer, actionable
  411. 13:52checklists, framework, Buffett-style
  412. 13:55decisions, business cases, transaction
  413. 13:57and stake records that Buffett analyzes
  414. 14:00as live case studies, accounting and
  415. 14:02financial concepts, well, we could tell
  416. 14:04you a little bit about all of them
  417. 14:07because they all interest me. Third
  418. 14:09question, how much freedom do I have
  419. 14:11regarding sources? Only the letters
  420. 14:13with citation. Everything stated in the
  421. 14:15wiki must come from the 17 letters and
  422. 14:17be cited by year plus topic, maximum
  423. 14:18rigor and traceability. This way we
  424. 14:20have traceability of what is being
  425. 14:22cited from each document. And we have
  426. 14:23the document in raw file. This would be
  427. 14:26the cleanest option and it is the one I
  428. 14:27recommend the most. Then we could use
  429. 14:29it, allowed to supplement with
  430. 14:30established knowledge from outside the
  431. 14:32letters if you want it to use external
  432. 14:33information. But when we want to work
  433. 14:36with good information that only we
  434. 14:38provide, and this is especially for
  435. 14:40analyzing companies, my recommendation
  436. 14:42would be to opt for one. That way it
  437. 14:44will always use the information in the
  438. 14:46raw file. I'm going to click on one in
  439. 14:48this case. Next. I only want what the
  440. 14:50letters say. How do I run the initial
  441. 14:52build? Everything now. I process the 17
  442. 14:54cards in one go with subagents. I build
  443. 14:56the wiki and show you the result. First
  444. 14:58, I do a pilot. I process two or three
  445. 14:59cards. I show you and, after your
  446. 15:01approval, I continue with the rest. My
  447. 15:03opinion is to first do a pilot to see
  448. 15:04how it is synthesizing each of the
  449. 15:06cards. We click send and it will start
  450. 15:08giving us a pilot in case we want to
  451. 15:10correct things. Well, there you have it
  452. 15:12, the pilot is now built. Here is the
  453. 15:15summary for your review. In this case,
  454. 15:17it took about 20 minutes because we
  455. 15:19gave it a lot of files. My
  456. 15:21recommendation is to first use a
  457. 15:23slightly lower language model, perhaps,
  458. 15:25for example, with Sonet it would be
  459. 15:27enough to consume fewer tokens. And the
  460. 15:30second thing is that we have to do this
  461. 15:33progressively. That is to say, if at
  462. 15:35first we try to give it all the brain
  463. 15:37files and have it ingest them all at
  464. 15:39once, in my opinion it will be too much
  465. 15:41work. We could have told it at the
  466. 15:43initial prompt to first go step by step
  467. 15:45, to only create the cloud MD and to
  468. 15:47only ingest one of the letter files.
  469. 15:50And each day what we do is that we ask
  470. 15:52it to ingest more row files that are
  471. 15:54missing and in this way it would
  472. 15:56gradually work and we can do it
  473. 15:57progressively without having to consume
  474. 15:59our entire session in a single session.
  475. 16:02Having said that, here it tells us that
  476. 16:04it has already created the pilot brain,
  477. 16:06the pages that it has created, all the
  478. 16:08concepts, all the business cases. In
  479. 16:10this case it has used a few, well, I
  480. 16:13have used 33 interconnected pages, okay
  481. 16:15? There would be many letters that it
  482. 16:17has not created, but what we are going
  483. 16:19to do is that then little by little
  484. 16:21each day we could tell it to ingest
  485. 16:22some letters. And this will again be
  486. 16:24built layer by layer by layer. We
  487. 16:27already have the structure, therefore
  488. 16:29everything that we add to it and do
  489. 16:31more will add up. So, let's see how the
  490. 16:33folder has turned out. Look, here I
  491. 16:35have Buffett's brain, he created Cloud
  492. 16:37MD, he didn't touch the row folder
  493. 16:39because those are our files and he
  494. 16:41created the wiki folder. And look at
  495. 16:44everything he's done. He's created the
  496. 16:46concepts, he's created the cases, he's
  497. 16:49organized everything by the
  498. 16:50applications, the people involved in
  499. 16:52the letters, the principles, the
  500. 16:54sources he's used, the summary, a lot
  501. 16:56of things. But of course, reading this
  502. 16:58this way isn't useful. This is the
  503. 17:00architecture. With this, every time we
  504. 17:03work in Cloud Code and have our
  505. 17:05Buffett's brain folder selected, we can
  506. 17:08ask him questions or we can add more
  507. 17:11things or again have all the context
  508. 17:13because now Cloud can find it very
  509. 17:16quickly. But now we can use an MD file
  510. 17:19reader to see a little bit about how
  511. 17:21this whole brain turned out and we can
  512. 17:23also have access to those notes. For
  513. 17:26this we can use the Obsidian
  514. 17:27application. You have it for Mac and
  515. 17:30also for other platforms. I think it's
  516. 17:32available for Windows too. Look, you
  517. 17:35download Obsidian, it's free, and when
  518. 17:37you download it, it's going to tell you
  519. 17:39this. When you open Obsidian, it's
  520. 17:41going to tell you either to create a
  521. 17:43new vault or to open a folder as a
  522. 17:44vault. In this case, since we already
  523. 17:46have the folder, we already have that
  524. 17:48brain created, we want to open it. So,
  525. 17:49we're going to open the Buffet brain
  526. 17:51folder. When I open the folder, I
  527. 17:53already have all the subfolders here.
  528. 17:55First, the row files, which we said,
  529. 17:57which are all the letters in PDF.
  530. 18:00Perfect, we're not going to touch those
  531. 18:01. And then all the wikis that it has
  532. 18:03created. Look, it has created a
  533. 18:05checklist to look at the quality of the
  534. 18:08business that we can use. Hey, it has
  535. 18:10pricing power, it has incremental
  536. 18:12returns on capital, the balance sheet,
  537. 18:14the management, it has created
  538. 18:16different red flags, important points
  539. 18:18that we have to review when we analyze
  540. 18:20red flags, right? When we analyze a
  541. 18:22company and it shows us everything
  542. 18:24linked together. Note that each concept
  543. 18:27and each year is linked so that we can
  544. 18:29access other notes on these topics.
  545. 18:32Then he has made us a folder of cases
  546. 18:34with, for example, some companies, for
  547. 18:36example, Geiko, the insurance company,
  548. 18:38where he talks to us about different
  549. 18:39things, he links us to all the things
  550. 18:41that we want to have. Siscandis also,
  551. 18:44the candy company. Then he has also
  552. 18:46made us the concepts, for example, the
  553. 18:49concept of economic goodwill, what it
  554. 18:51means, how the concept of float works,
  555. 18:53the concept of intrinsic value, who it
  556. 18:55relates to. And we can click here with
  557. 18:58Obsidian and it makes us a graph of the
  558. 19:00concepts that he has created and we can
  559. 19:02see how everything is linked to each
  560. 19:04other. For example, Mr. Market
  561. 19:07redirects us to other important
  562. 19:08concepts. For example, in 2008 he
  563. 19:10talked to us about all this, in 2011 he
  564. 19:12talked to us about all these concepts
  565. 19:15and we could click and go to the
  566. 19:16concept that we created. For example,
  567. 19:19if we create a concept such as the
  568. 19:20margin of safety, we see everything
  569. 19:22that is related to the margin of safety
  570. 19:24. Then it also creates those important
  571. 19:27pages, that structure that Carpaty
  572. 19:29mentioned, which will basically always
  573. 19:31be updated. For example, the index
  574. 19:33basically tells us how everything is
  575. 19:35structured. The log tells us the latest
  576. 19:37changes. Here it tells us that it has
  577. 19:39created the schema and then it tells us
  578. 19:41that it has ingested the letters from
  579. 19:432008, 2011 and 2014 only. So you can
  580. 19:45see little information here, but look,
  581. 19:47it is already quite a few concepts and
  582. 19:48we have only extracted three letters.
  583. 19:50What I could do now day by day is ask
  584. 19:52it to ingest more letters so that it
  585. 19:55gradually creates this entire database,
  586. 19:57this entire brain that we can then use
  587. 19:59for the context. And here is the Cloud
  588. 20:01MD file, which we could review and make
  589. 20:03changes to. The Cloud MD is basically
  590. 20:05the entire architecture and all the
  591. 20:07steps of how Cloud is going to read
  592. 20:08these files, how it is going to use
  593. 20:10them, what rules it is going to apply,
  594. 20:12why it does it, what the context is and
  595. 20:14we would have everything. So whenever
  596. 20:16we want, we can upload more files to
  597. 20:18the row folder, raw files, and have it
  598. 20:20ingest them and build those wikis,
  599. 20:22those pages, and all those concepts,
  600. 20:25and that way, little by little, we're
  601. 20:27building that brain. For example, let's
  602. 20:30open a new window in Cloud Code. Let's
  603. 20:33tap Buffet's brain again and ask it to
  604. 20:36ingest, let's tell it the charts for
  605. 20:412020, 2021, and 2022. If I had added a
  606. 20:44new file, I could have told it directly
  607. 20:47, "Ingest this file." And that way,
  608. 20:50little by little, we're building that
  609. 20:52brain. The utility of this is that
  610. 20:55every time we use Cloud Code with that
  611. 20:57folder selected, Cloud Code will
  612. 20:59already have access to all that context
  613. 21:01, to that entire database, to all those
  614. 21:03pages created with all the concepts
  615. 21:06that are related to each other. Imagine
  616. 21:08that we are studying to make a video or
  617. 21:11a script about Warren Buffett, the next
  618. 21:13time I work on a script about Warren
  619. 21:15Buffett for my YouTube channel, I could
  620. 21:18work on this folder so that I have all
  621. 21:20the references. And then, in the
  622. 21:22YouTube script, it will cite what he
  623. 21:24said in the 2012 letter, what he said
  624. 21:26in the 2010 letter. This is related to
  625. 21:28the concept that Geiko explained in the
  626. 21:312020 letter. Do you understand? We are
  627. 21:33creating a kind of neural network so
  628. 21:35that every time we want to work on it,
  629. 21:37we always have the context, we don't
  630. 21:40have to teach it again. Obviously, at
  631. 21:42first it is hard to build, it takes
  632. 21:43time, it requires spending tokens, it
  633. 21:45requires making the whole structure,
  634. 21:47dedicating a few minutes to it. As you
  635. 21:49can see, it took me 20 minutes to make
  636. 21:50the whole structure. But then little by
  637. 21:52little, feeding that structure, well,
  638. 21:54that's what takes the least time and
  639. 21:56dedicating a little time each week to
  640. 21:58feeding this kind of brain. I can think
  641. 22:01of many applications. For example, for
  642. 22:03those of you who are analysts and cover
  643. 22:05a company in great detail and
  644. 22:07constantly monitor this company,
  645. 22:09analyze all its results, see all the
  646. 22:11news it publishes, have all the
  647. 22:13necessary information, you could create
  648. 22:15an independent brain with all the
  649. 22:17information from this company, all the
  650. 22:19raw data, the latest 10K, the annual
  651. 22:21reports, all the financial data, and
  652. 22:23have Cloud only use those raw documents
  653. 22:26. That way it won't have to go out and
  654. 22:27look for anything to fill in, because
  655. 22:29sometimes AI has a bit of amnesia and,
  656. 22:31when it can't find information in a
  657. 22:33document, it basically gives you
  658. 22:34information from a context it has from
  659. 22:36a web page, or from a blog, or from the
  660. 22:38consensus that currently exists on Wall
  661. 22:39Street. So, this way of working seems
  662. 22:41much more correct to me because you're
  663. 22:43basically going to pull from the files
  664. 22:45of this database that you're building
  665. 22:47and then, in addition, you'll have it
  666. 22:49for yourself so that you can access all
  667. 22:51that information and have it well
  668. 22:53organized and well connected to each
  669. 22:55other. In fact, I have to show you
  670. 22:57something before I go, and it's the
  671. 22:58following. The good thing is that we
  672. 23:00can also feed these Obsidian files with
  673. 23:02its extension. We can download the
  674. 23:04Obsidian Chrome extension and, when we
  675. 23:07are in an article that we think can
  676. 23:09feed our brain, our architecture, we
  677. 23:12could go to the extension, click on
  678. 23:14Obsidian Webclipper and, basically, we
  679. 23:17could make a clipping of this article
  680. 23:19and directly add it to Obsidian. We
  681. 23:22click on open Obsidian and look, it can
  682. 23:24make a text file of the article note
  683. 23:27with the source, the author, the date
  684. 23:29it was published and with the summary
  685. 23:31of this article so that we also have it
  686. 23:34available in that architecture, in that
  687. 23:36database. It is also important that
  688. 23:38with Cloud, each time we work, we are
  689. 23:40going to give it context and feedback.
  690. 23:41And that will mean that the
  691. 23:43conversations will be adjusted.
  692. 23:44Everything will be recorded in the log:
  693. 23:46all the changes we requested, all the
  694. 23:48feedback we gave, "Hey, this wasn't put
  695. 23:50so well," or "this concept didn't get
  696. 23:52the terms right," and you can go about
  697. 23:54correcting that entire file, that
  698. 23:56entire architecture, so that we have it
  699. 23:58much cleaner. Again, this isn't magic,
  700. 24:01it obviously requires time to maintain,
  701. 24:04create, and also give it a useful use.
  702. 24:06So if you liked it, don't forget to
  703. 24:08support the video. And now, I'll see
  704. 24:10you in the next one. And if you want to
  705. 24:12start investing, I'll leave you a
  706. 24:13complete free course below where I show
  707. 24:15you the step-by-step instructions to
  708. 24:17get started. See you in the next video.

About this transcript

This page contains the full transcript of He Creado un Segundo Cerebro con IA… y Es una Locura! by Javier DV - Inversión y Finanzas, generated from the public captions YouTube serves with the video. The transcript has 4,995 words across 708 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.