YouTube2Text

Cómo usar DeepSeek Mejor que el 99% de las personas (GRATIS) — Transcript

by Alejavi Rivera · 4,660 words · 722 segments · language en · Watch on YouTube

Full transcript

  1. 0:00It’s not always the one with the best
  2. 0:01cards who wins, but the one who learns
  3. 0:03how to play the hand they have.
  4. 0:05Even when it seemed they had the worst
  5. 0:08hand, DeepSeek found a way to be more
  6. 0:10efficient and optimized in the
  7. 0:11artificial intelligence market. Today,
  8. 0:14they don't just have the best AI models
  9. 0:16in terms of price-to-performance, but
  10. 0:19also a complete ecosystem capable of
  11. 0:21anything.
  12. 0:25To prove it, we’ll cover everything
  13. 0:27about DeepSeek, from zero to expert,
  14. 0:29with everything you need to know, like
  15. 0:32their models, apps, key functions, and
  16. 0:34running it locally or uncensored.
  17. 0:37The best part is that everything
  18. 0:39DeepSeek developed is open, meaning we
  19. 0:41can start from cheap to free, private,
  20. 0:44and limitless.
  21. 0:45Sound good?
  22. 0:47Well, let’s start at the beginning.
  23. 0:51Let’s properly place DeepSeek. This is a
  24. 0:54Chinese company founded in 2023 and
  25. 0:56based in Hangzhou. Behind it is Liang
  26. 0:59Wenfeng, an entrepreneur and engineer
  27. 1:01who worked in artificial intelligence.
  28. 1:04This project was born linked to High
  29. 1:06Flyer, a quantitative investment fund
  30. 1:08that used AI to trade in financial
  31. 1:10markets.
  32. 1:11Its goal was very clear: to research and
  33. 1:14develop artificial general intelligence
  34. 1:16to compete against the big labs.
  35. 1:19Their great revolution arrived in
  36. 1:21January 2025 with DeepSeek R1, which
  37. 1:23shocked everyone with its reasoning
  38. 1:25capacity and efficiency, causing the US
  39. 1:28artificial intelligence market to
  40. 1:30plummet.
  41. 1:32They are committed to working with
  42. 1:34highly optimized, open-source models
  43. 1:36that can be downloaded, modified, and
  44. 1:38executed.
  45. 1:40Looking at their business model, it
  46. 1:42might seem like everything is free,
  47. 1:44since we really have a free chat,
  48. 1:46low-cost APIs, and open-source models
  49. 1:49you don't have to pay for.
  50. 1:51But here is the key: even though their
  51. 1:53models are low-cost, they can support it
  52. 1:56because they are very efficient, and
  53. 1:59that open-source code also helps the
  54. 2:01project improve future features. Let’s
  55. 2:04start by getting to know their models a
  56. 2:06bit more in depth.
  57. 2:07Basically, we have two main ones: the
  58. 2:09Flash models, which are the fastest,
  59. 2:12cheapest, multimodal, and best for
  60. 2:14day-to-day tasks; and the Pro model,
  61. 2:16which is larger, geared toward complex
  62. 2:19tasks and deeper reasoning.
  63. 2:22Although as of the making of this video
  64. 2:24, the latest update is the 4.1 Flash
  65. 2:25version, which, besides being cheaper,
  66. 2:27is also more powerful.
  67. 2:29Historically, we also find the R1 models
  68. 2:32, the family I mentioned before, focused
  69. 2:34on mathematical, logical, and
  70. 2:36programming reasoning. And from this,
  71. 2:39distilled models were born, also known
  72. 2:41as R1 Distils, which are smaller
  73. 2:44versions designed to run on fewer
  74. 2:46resources. And from these, many
  75. 2:48community variations have also emerged,
  76. 2:51even releasing them to be free of any
  77. 2:53type of restriction. DeepSeek also
  78. 2:56develops other types of models like OCR
  79. 2:58and UL, but we will focus mainly on the
  80. 3:01AI ones that we can use through its
  81. 3:03ecosystem.
  82. 3:05To make use of this model, the easiest
  83. 3:07way would be to jump to their official
  84. 3:08cloud platform with their DeepSeek chat
  85. 3:10.
  86. 3:11From here, keep in mind that we will be
  87. 3:13able to use their latest model, and also
  88. 3:15, no matter where you look, you won't
  89. 3:16find a paid subscription since they
  90. 3:18offer it completely for free.
  91. 3:21Operation is very simple; we simply
  92. 3:22input any prompt here, as in this case
  93. 3:25where I'm going to ask for a
  94. 3:26mini-application to show companies how
  95. 3:28much each meeting is costing them by
  96. 3:30calculating it based on the number of
  97. 3:32attendees, hourly rate, duration, and so
  98. 3:34on.
  99. 3:36When I click send, this tool starts
  100. 3:38working, and a minute later, after all
  101. 3:40the reasoning we can see here, a little
  102. 3:42further down we can see that it has
  103. 3:44developed the entire code for the
  104. 3:46application, which we can also preview
  105. 3:48right here.
  106. 3:50We just have to click Run, and from
  107. 3:52there we can find this meeting cost
  108. 3:55calculator, where we can input the
  109. 3:57number of attendees—for example, I'll
  110. 4:00put seven attendees—the hourly cost,
  111. 4:03I'll use 75 as an example, and an
  112. 4:05approximate duration of an hour or so.
  113. 4:09It shows us very clearly how the cost
  114. 4:11would be about 525 €, and it also
  115. 4:14developed a very simple timer so that
  116. 4:16when we hit play, we can see in real
  117. 4:19time the cost that meeting is incurring
  118. 4:22, something I think could be very
  119. 4:24interesting for more than one company.
  120. 4:27And from this platform, the truth is
  121. 4:29that we have a very capable model via
  122. 4:31chat with which we can do many things
  123. 4:33with practically no limits at all.
  124. 4:36However, if we want to get the most out
  125. 4:38of DeepSeek, we should start with the
  126. 4:40next section.
  127. 4:44In this second section, we are going to
  128. 4:46talk about one of the main pillars that
  129. 4:48DeepSeek currently has, which is its
  130. 4:50application known as DeepSeek Harness.
  131. 4:53And this harness is climbing the ranks
  132. 4:55and positioning itself among the top,
  133. 4:57comparing it with the main tools like
  134. 5:00Codex, Hermes, Open Code, Cloud Code,
  135. 5:02and others.
  136. 5:04And I'm not surprised, because not only
  137. 5:06has it managed to keep up in just a few
  138. 5:08months and in record time, but it has
  139. 5:10also incorporated certain features that
  140. 5:12none of the previous ones have. To
  141. 5:14verify this, I went to the official
  142. 5:16DeepSeek Harness page, and by the way,
  143. 5:18all the links I’m mentioning throughout
  144. 5:20this video will be provided in the
  145. 5:22description below.
  146. 5:24From here, keep in mind that as of today
  147. 5:26, we have a very simple way to download
  148. 5:28its application.
  149. 5:30We just have to click here, and if for
  150. 5:31some reason it doesn't show your
  151. 5:33operating system correctly, if we click
  152. 5:35this arrow, we can see that it is
  153. 5:36available for both Mac and Windows.
  154. 5:39Once downloaded and installed, we will
  155. 5:42find an interface very similar to any
  156. 5:44harness like ChatGPT, only this one is
  157. 5:47completely open source.
  158. 5:49Before looking at the key points of this
  159. 5:51desktop application, in case you are
  160. 5:53wondering, yes, we also have an app for
  161. 5:56mobile devices.
  162. 5:57To do this, go to your mobile phone's
  163. 5:59app store, click on the search bar, and
  164. 6:01type "DeepSeek."
  165. 6:03From here, you'll find this official app
  166. 6:06, which you can install for free.
  167. 6:09Once opened, after logging in or
  168. 6:11creating an account, it will confirm
  169. 6:13this is the official free app, and by
  170. 6:16clicking "start chat," you can start
  171. 6:18using DeepSeek right away.
  172. 6:20But you should keep in mind that the
  173. 6:23mobile app is more similar to the web
  174. 6:25version with its chat mode, rather than
  175. 6:28what we see on the computer where we
  176. 6:30actually work with agents.
  177. 6:32If we return to the application, it is
  178. 6:35true that we have some additional
  179. 6:36features, such as the fact that by
  180. 6:38holding down this button, we can dictate
  181. 6:41the prompt more easily.
  182. 6:43Now, if I let go, we can see how it
  183. 6:45responded instantly thanks to the
  184. 6:47transcription of what I just said.
  185. 6:50Let's return to the desktop application
  186. 6:52now, since this is really where we have
  187. 6:54a ton of possibilities.
  188. 6:56To discover all of this, let's look at
  189. 6:58the next block: the key points to keep
  190. 7:00in mind to get the most out of it.
  191. 7:02However, I will end this on a relatively
  192. 7:04negative note, which is the fact that we
  193. 7:07don't have that free DeepSeek artificial
  194. 7:09intelligence model linked here, but we
  195. 7:11will have the possibility to connect it
  196. 7:14officially via API at a very low cost
  197. 7:16and with all the agentic possibilities
  198. 7:18that we will be seeing in the next block
  199. 7:21. But beyond this, thanks to the fact
  200. 7:23that it is open source, we will be able
  201. 7:24to integrate any provider, even those
  202. 7:26that offer us the ability to use it for
  203. 7:28free.
  204. 7:30From here, notice how I actually have a
  205. 7:31bunch of artificial intelligence models
  206. 7:33that I can use.
  207. 7:35Many of them, moreover, if I type the
  208. 7:37word free, I will be able to use for
  209. 7:39free, and for that, keep in mind that if
  210. 7:41it is the first time you download and
  211. 7:43install this application, it will
  212. 7:44probably ask you to add some credit
  213. 7:46during the installation process.
  214. 7:48This is not necessary, you can skip it
  215. 7:50since I am now going to tell you all the
  216. 7:52ways to connect artificial intelligence
  217. 7:54. To do this, what we have to do is
  218. 7:56click on the sidebar on the left and
  219. 7:58from here, if we go to the plus sign and
  220. 8:00then enter the settings section, we will
  221. 8:02find an option called models.
  222. 8:04From here, we will be able to connect
  223. 8:06any AI provider, whether it is the
  224. 8:08official one itself or any other tool
  225. 8:10that offers us artificial intelligence
  226. 8:12models. These can be in the cloud or
  227. 8:14even installed locally, as we will be
  228. 8:16seeing later.
  229. 8:18To be able to add a provider, we simply
  230. 8:20have to click on this section here,
  231. 8:21select some from this list in the
  232. 8:23providers section, but be careful,
  233. 8:25because if you don't find a provider,
  234. 8:26you can also add it in a custom way, as
  235. 8:28we will see later with local models.
  236. 8:31To see the official case, I am going to
  237. 8:33select the DeepSeek one and from here
  238. 8:35all we have to do is paste our API key.
  239. 8:38I already have it pasted, but keep in
  240. 8:40mind that you will be able to do this by
  241. 8:42jumping to the official DeepSeek
  242. 8:44platform, which I will also be leaving
  243. 8:45below in the description, and by adding
  244. 8:47some credit here, even if it is just
  245. 8:49about 2 dollars, which actually goes a
  246. 8:51long way, then by entering the API Key
  247. 8:53part and clicking on create a new API
  248. 8:55key, by entering a name and selecting
  249. 8:57create, we will only have to copy, paste
  250. 8:59from here and when we click apply we
  251. 9:01will have the model available.
  252. 9:03However, the best thing about having
  253. 9:05this open-source model is the ability to
  254. 9:07go beyond a specific provider. To do
  255. 9:10this, notice how I have Nvidia installed
  256. 9:12here, I even have my own ChatGPT codec
  257. 9:15and others like OpenRouter.
  258. 9:18I've done this by following the exact
  259. 9:20same process, but using these other
  260. 9:21providers that also offer free APIs.
  261. 9:24With what we have just seen, we are
  262. 9:26ready to do a lot with all these
  263. 9:27artificial intelligence models. In large
  264. 9:30part also due to the information found
  265. 9:32on the internet to train their AI models
  266. 9:34. Now, have you ever wondered how much
  267. 9:36of your personal information is
  268. 9:38circulating on the internet without you
  269. 9:40knowing?
  270. 9:41To find out, you should know about
  271. 9:43Incogni, a service that takes care of
  272. 9:45requesting data deletion for us, because
  273. 9:48there are companies that collect and
  274. 9:50sell our personal data, like our phone
  275. 9:52number, email, or even our address.
  276. 9:56I don't know if it has happened to you,
  277. 9:57but I keep getting calls from
  278. 9:58telemarketers and scammers, but we can
  279. 9:59avoid this this way.
  280. 10:02Look, if we simply create an account and
  281. 10:04authorize it to act on our behalf, from
  282. 10:06that point on it will connect with more
  283. 10:08than 420 data brokers requesting its
  284. 10:10removal.
  285. 10:11In addition, it repeats requests
  286. 10:13periodically and we can check the
  287. 10:15progress at any time.
  288. 10:17I did it with mine and it showed me
  289. 10:19everything easily, even with a privacy
  290. 10:21score, but although it looked good, I
  291. 10:22discovered a bunch of brokers that were
  292. 10:24selling my personal information.
  293. 10:27Only now Incogni is removing it
  294. 10:29automatically. And hey, thanks to
  295. 10:31Incogni being sponsors of this video,
  296. 10:33they have left me a special link that
  297. 10:35also has a bigger discount for Black
  298. 10:37Friday, which I will leave for you in
  299. 10:38the description below, with which by
  300. 10:40also using the code Ale Javi you will
  301. 10:42get an even bigger discount. Take
  302. 10:44advantage because these are the best
  303. 10:45prices of the year and this is one of
  304. 10:47the tools that is worth trying. And once
  305. 10:49we know this, let's move on to the next
  306. 10:50block.
  307. 10:54I have just returned to the DeepSeek
  308. 10:56application to really see those keys
  309. 10:58that make it different from other
  310. 11:00platforms. One of the things to keep in
  311. 11:02mind is that the main thing that
  312. 11:04differentiates it is this plugins
  313. 11:06functionality, since it literally allows
  314. 11:08us to develop any function we want or
  315. 11:10even integrate one from the community
  316. 11:13that we like.
  317. 11:14To see this, I'm going to click on
  318. 11:16plugins and from here notice how we will
  319. 11:18have a lot of functionalities, both
  320. 11:20available and some experimental ones
  321. 11:22that we could activate or deactivate
  322. 11:24simply by clicking here.
  323. 11:26But beyond all these functionalities
  324. 11:29with sub-agents and so on, we will also
  325. 11:31be able to integrate the ones we need.
  326. 11:34Now imagine that you like to use some
  327. 11:36free tokens offered by some artificial
  328. 11:38intelligence platforms.
  329. 11:40For example, Open Code offers some free
  330. 11:42models directly on its own platform,
  331. 11:44like the one from DeepSeek itself.
  332. 11:47But even if we can't do this directly
  333. 11:49from here, because it has a large
  334. 11:51community and is open source, we can
  335. 11:53integrate it easily.
  336. 11:55In fact, notice how we already have
  337. 11:56repositories right here that can do
  338. 11:58exactly that.
  339. 12:00And to integrate any functionality we've
  340. 12:03seen in a repository, it's as easy as
  341. 12:05clicking "add plugin," pasting the
  342. 12:08GitHub URL here, and simply clicking "
  343. 12:10install" to start processing it so it
  344. 12:13later appears here, ready for us to
  345. 12:15activate.
  346. 12:18Now, if I start a new session and from
  347. 12:20the model section, select those from the
  348. 12:23Open Code provider, notice that if I
  349. 12:25search for "DeepSeek," beyond the
  350. 12:28official ones I have via API, I can also
  351. 12:30find free versions like DeepSeek 4.1
  352. 12:33Flash.
  353. 12:35If I click on this "free" one and ask it
  354. 12:37if it can hear me, as you can see, after
  355. 12:40about two seconds, it replies "yes"
  356. 12:42perfectly.
  357. 12:44The plugins really allow us to add
  358. 12:46whatever we want, not just free
  359. 12:47artificial intelligence—and by the way,
  360. 12:50if you’d like to know all this in detail
  361. 12:52, this video here talks exactly about
  362. 12:54all the options DeepSeek gives us to
  363. 12:56connect for free—but beyond that, if we
  364. 12:58click on plugins again, we could install
  365. 13:00anything.
  366. 13:02Notice how I installed another
  367. 13:04repository here that allows me to use my
  368. 13:06own ChatGPT models from my subscription
  369. 13:09directly on DeepSeek, and along with
  370. 13:11this, beyond integrating existing
  371. 13:14repositories, we could create our own
  372. 13:16plugins.
  373. 13:18To do that, we can simply click the plus
  374. 13:20sign, select "create your own plugin,"
  375. 13:22and by writing what we want it to create
  376. 13:23here, it will be able to do it.
  377. 13:26And with this, we can really do things
  378. 13:29like create a timer to easily see how
  379. 13:32long each task takes, keep track of all
  380. 13:34spending in terms of both tokens and
  381. 13:37cost, completely change the interface,
  382. 13:40customize it, create interactive 3D
  383. 13:42applications like this world, or even
  384. 13:45change the application name, like I did
  385. 13:47here when I named it "Aleja Bijarnes."
  386. 13:51I already talked about all these "
  387. 13:52creator mode" examples in this other
  388. 13:54video here, so if you want to dive a
  389. 13:55little deeper, you have that video there
  390. 13:57, but let's look at other key features
  391. 13:59of the application.
  392. 14:01The thing is, focusing on a new chat
  393. 14:03session, one of the lesser-known aspects
  394. 14:05is the small features geared toward
  395. 14:07specific tasks, as well as the types of
  396. 14:09formats we can preview from here.
  397. 14:12Let’s start with the basics; keep in
  398. 14:14mind that through our harness, we can
  399. 14:16grant access to any folder on our
  400. 14:18computer, whether for it to review or
  401. 14:20execute tasks, such as reorganizing
  402. 14:22everything, moving files, creating new
  403. 14:25ones, and more. We will also have
  404. 14:26different modes available right here.
  405. 14:28By default, we have the standard mode,
  406. 14:31but there are others like PTC, focused
  407. 14:33on massive tasks; the minimal mode,
  408. 14:35where we force users to utilize the
  409. 14:37terminal to test and verify performance
  410. 14:40more efficiently; and the creator mode,
  411. 14:42which I mentioned earlier, allowing us
  412. 14:44to change the application to our liking
  413. 14:47and add any features we want.
  414. 14:50Let’s start with a simple task, so I’ll
  415. 14:53go back to the standard section, and
  416. 14:55before sending it, notice we also have
  417. 14:57this last option that allows read-only
  418. 14:59access—very useful if we want to give
  419. 15:01access to a folder with a lot of data
  420. 15:03that shouldn't be modified, even if we
  421. 15:05want it to use that information.
  422. 15:08By clicking on read-only, we gain that
  423. 15:10control and avoid any modification
  424. 15:13errors. Another option allows it to
  425. 15:15execute changes.
  426. 15:16We also have a third one that grants
  427. 15:18full access, allowing it to make all
  428. 15:20changes without constantly asking us for
  429. 15:23confirmation.
  430. 15:24And finally, a beta self-review mode so
  431. 15:26it can iterate on itself before giving
  432. 15:29us the final result. Knowing this, I’m
  433. 15:31going to write the following prompt,
  434. 15:33asking it to create a visual
  435. 15:34presentation about six artificial
  436. 15:36intelligence scams and how we can
  437. 15:38protect ourselves from them.
  438. 15:40To do this, I will ask it to include
  439. 15:42eight 16: 9 slides with all those key
  440. 15:44points and real statistics.
  441. 15:46I’ll click send, and it will start
  442. 15:48processing it right away.
  443. 15:50In this case, we are using the free Open
  444. 15:53Code model, and about seven minutes
  445. 15:55later, it would have created a
  446. 15:57presentation and even a PDF that we can
  447. 15:59preview here, so let’s click on the
  448. 16:01presentation first, and see how the
  449. 16:03presentation appears for us to expand,
  450. 16:06giving us all the details we requested.
  451. 16:10Actually, you can skip through the video
  452. 16:12if you want to see those common AI scams
  453. 16:14we're encountering, but let's focus on
  454. 16:17the features.
  455. 16:18If I close from here, notice that if I
  456. 16:20go to the PDF section now, it’s also
  457. 16:22something I’ll be able to preview and
  458. 16:25even download.
  459. 16:26And we aren’t just talking about
  460. 16:28creating applications here; we can
  461. 16:30actually interact with any type of
  462. 16:32documentation, application, or folder
  463. 16:34with control and a very simple way to
  464. 16:36interact. Returning to the presentation
  465. 16:39, I want to show you another thing that
  466. 16:40sets this DeepSeek application apart,
  467. 16:42which would be the entire traceability
  468. 16:44and transparency section.
  469. 16:47Notice that when we go to any chat, like
  470. 16:49this one here, by clicking on the
  471. 16:51trajectory, we can see the breakdown of
  472. 16:54everything that happened in the
  473. 16:56background, what tools were used, what
  474. 16:58the assistant was doing, and we can
  475. 17:00truly expand and detail this entire
  476. 17:03timeline to see exactly what was going
  477. 17:05on.
  478. 17:07Because yes, beyond being able to
  479. 17:08customize it to our liking, we can also
  480. 17:10have control over everything that is
  481. 17:12happening.
  482. 17:13Additionally, notice that if we go back
  483. 17:16to the settings section, we also have an
  484. 17:18AI model usage section where we can see
  485. 17:21all the costs incurred according to the
  486. 17:23different providers, like these free
  487. 17:25models I'm using from Open Code, Codex
  488. 17:28itself, and ultimately a breakdown of
  489. 17:30everything happening at any given moment
  490. 17:33.
  491. 17:38In this fourth block, we’re going to
  492. 17:39focus more on local usage, because while
  493. 17:41what we’ve seen so far is great and
  494. 17:43gives us a lot of control, what happens
  495. 17:44when we want greater privacy or to
  496. 17:46simply use models that are totally
  497. 17:47unlimited?
  498. 17:49Well, we can do this too, not only
  499. 17:51because we can install local AI models
  500. 17:54in our own DeepSeek application, but
  501. 17:56because there are also open-source
  502. 17:59DeepSeek models that we can integrate
  503. 18:01with their own harness very easily.
  504. 18:04Before learning about any model or
  505. 18:06wasting time installing something that
  506. 18:08might slow down your computer or just
  507. 18:10not work well, I’m going to recommend a
  508. 18:12site that will tell you if it will
  509. 18:14actually work before you do all of that
  510. 18:16, so you can save all that time. The
  511. 18:19platform is this one here, which I’ll
  512. 18:20also be leaving down in the description
  513. 18:22. And from here, when we enter, the
  514. 18:24first thing it will do is analyze our
  515. 18:26hardware.
  516. 18:27In my case, it says I have an Apple M3
  517. 18:29Pro with 18 GB of VRAM, with the speed
  518. 18:32we have and the cores it contains.
  519. 18:34Knowing all this, it can then recommend
  520. 18:37which models will run best on my device
  521. 18:39. Notice how it recommends, for example
  522. 18:41, small models like Alibaba's Qwen,
  523. 18:43models like Melun, and some from other
  524. 18:45companies like Google, Mistral, and
  525. 18:47others.
  526. 18:49But here, if we wanted to use a specific
  527. 18:52model, like DeepSeek, we can go to the
  528. 18:54search section and, by clicking on the
  529. 18:57search bar and typing the model we want
  530. 18:59to install—in our case, DeepSeek—all the
  531. 19:02available ones we can use will appear.
  532. 19:06Notice how the ones we had via API, like
  533. 19:08Flash and Pro, actually require so much
  534. 19:11hardware that practically no computer,
  535. 19:13even if it's open source, will be able
  536. 19:16to run them at home. And that is why, if
  537. 19:18we want much more power, it might be
  538. 19:20interesting to rely on the previous API
  539. 19:22blocks. However, for slightly simpler
  540. 19:25tasks, we can use smaller models. For
  541. 19:27this, we have R1 and distilled R1, which
  542. 19:30have fewer parameters and allow us to
  543. 19:32run them on our device easily.
  544. 19:34In my case, I could run more or less up
  545. 19:36to the 14-billion parameter version, and
  546. 19:39it shows me how easily my device will
  547. 19:41handle it. So, once we know this, we are
  548. 19:44ready to install artificial intelligence
  549. 19:46models locally. To do this, we have many
  550. 19:48possibilities. We have platforms like
  551. 19:51Jan, LM Studio, or the one I like best,
  552. 19:53Ollama.
  553. 19:55Being this platform right here—which, by
  554. 19:57the way, I'm going to go a bit faster
  555. 19:59since I talked about it in detail in
  556. 20:01this video here—I'll give you some key
  557. 20:03points: click to download and install
  558. 20:05this application, because once you have
  559. 20:07it, even if it looks similar to DeepSeek
  560. 20:09, just in a more basic way, what we are
  561. 20:12really interested in is installing the
  562. 20:14models locally and then transferring
  563. 20:16these models to DeepSeek itself.
  564. 20:18Let's see how to do this. For that, I've
  565. 20:20returned to Ollama, and we'll start by
  566. 20:22seeing how we can download an artificial
  567. 20:23intelligence model. We will do this by
  568. 20:25clicking on the models section and
  569. 20:27searching from there for the model we
  570. 20:28want to install. To do this, I am going
  571. 20:31to search for DeepSeek and go to a
  572. 20:33slightly older model, the R1, since it
  573. 20:35has different variations depending on
  574. 20:37the number of parameters it has.
  575. 20:40In my case, since you can use it with up
  576. 20:42to 14 billion parameters, I will stick
  577. 20:44with this last one with 8 billion
  578. 20:46parameters, so I will click on copy
  579. 20:48model and from a terminal—and pay
  580. 20:50attention, because even though it may
  581. 20:52seem a bit technical, it is very simple
  582. 20:54—just by typing the word "Ollama pull,"
  583. 20:56we add a space and paste the ID we just
  584. 20:58copied for the model we want to download
  585. 21:00; upon pressing enter, it will start
  586. 21:02downloading.
  587. 21:04Notice how it directly shows us the
  588. 21:05download speed and how long it will take
  589. 21:07; in my case, approximately a minute and
  590. 21:09a half, but this will depend a lot on
  591. 21:11your computer and the type of model you
  592. 21:13are downloading. After that time, it
  593. 21:15will tell me here that it has been
  594. 21:16successfully downloaded, and to test it
  595. 21:18, the best thing would be to return to
  596. 21:20the Ollama application, where that model
  597. 21:22should appear.
  598. 21:23See how I have it here, DeepSeek R1 with
  599. 21:258 billion parameters, and if I send it
  600. 21:27any test like "hello, how are you?"
  601. 21:30And after having reasoned the answer
  602. 21:31thoroughly, we can see that it has
  603. 21:33responded well, and we will also be able
  604. 21:35to use this without any internet
  605. 21:37connection, as everything will stay
  606. 21:38between the computer and us.
  607. 21:41So now we are going to see how we can
  608. 21:43connect this model to our own DeepSeek
  609. 21:44application, where we have more advanced
  610. 21:46features.
  611. 21:47To do this, notice that if we go to the
  612. 21:50left sidebar and then to applications,
  613. 21:52we will find a bunch of options here to
  614. 21:54connect local models to many artificial
  615. 21:57intelligence tools. Among them, we have
  616. 21:59the DeepSeek harness, and to do this,
  617. 22:01all we have to do is click on this
  618. 22:04command named "Ollama launch dsh," so
  619. 22:06here in a new terminal, I will paste
  620. 22:08that command and press enter. From here
  621. 22:10, it asks us which model we want to
  622. 22:12connect to.
  623. 22:14In fact, here we can find some models
  624. 22:16that they offer on their own cloud, or
  625. 22:18others that we can install locally or
  626. 22:19those we already had before, and if we
  627. 22:21also wanted to install one directly from
  628. 22:23here, like Qwen, for example, just hit
  629. 22:25enter and click that we want to download
  630. 22:28it; after doing so, our DeepSeek
  631. 22:29application would reload, only with that
  632. 22:31model connected.
  633. 22:33I just canceled it because I want to
  634. 22:35show you another way, since it's really
  635. 22:36not the only method. If we go to the
  636. 22:39more section and enter the settings menu
  637. 22:41again, remember that we can also add a
  638. 22:44custom provider for models. From here,
  639. 22:47we can enter the data to connect to our
  640. 22:49Ollama.
  641. 22:51To do this, for the name I'll enter
  642. 22:53Ollama, for the base URL we'll put this
  643. 22:55one here, and by clicking on search
  644. 22:57available models, all the ones we have
  645. 22:59installed locally will appear, such as
  646. 23:01Cuen, DeepSeek R1 with 8 billion
  647. 23:03parameters, and Gemma 4.
  648. 23:06If we wanted to use all three, we could
  649. 23:08leave them checked, but if instead we
  650. 23:10only wanted to use some of them, like
  651. 23:12I'm going to leave these two here,
  652. 23:14DeepSeek and Gemma, I can click on
  653. 23:16update list.
  654. 23:17They would then appear linked here, and
  655. 23:20if I click on create provider, I'll have
  656. 23:22this custom provider. Then, when I click
  657. 23:25on the models section, beyond all those
  658. 23:27available via API, if I scroll to the
  659. 23:30bottom, I can find these two models that
  660. 23:32we have installed locally. And with this
  661. 23:35, we not only get free use, but also
  662. 23:37unlimited and private access.
  663. 23:43We will finish with the final block,
  664. 23:44knowing that we can install DeepSeek
  665. 23:46models locally, which have also been
  666. 23:48modified by the community and completely
  667. 23:50unlocked without restrictions.
  668. 23:53Use it responsibly because it is
  669. 23:55dangerously powerful, as it can
  670. 23:56reverse-engineer applications or answer
  671. 23:58anything we need.
  672. 24:00Before seeing how we can use these fully
  673. 24:03unlocked DeepSeek models, I want to show
  674. 24:05you some use cases so you can see the
  675. 24:07opportunity this offers. For one, you
  676. 24:09can use it to research controversial
  677. 24:11topics, such as historical or political
  678. 24:14ones, that some models avoid answering.
  679. 24:16We can also perform reverse engineering
  680. 24:18to study how our own applications work,
  681. 24:20or those we are authorized to analyze,
  682. 24:23and learn to build alternatives. We can
  683. 24:25use it for research and cybersecurity,
  684. 24:27to analyze vulnerabilities in our own
  685. 24:29systems and understand how to protect
  686. 24:31them. And we can also use it with total
  687. 24:33privacy for more sensitive matters,
  688. 24:35because since it runs on our computer,
  689. 24:37we can consult confidential documents,
  690. 24:39discuss personal issues, or explore
  691. 24:41health questions without sending our
  692. 24:44conversation to external servers or
  693. 24:46having it tell us it can't answer that.
  694. 24:48Regarding the unlocked version, keep in
  695. 24:50mind that although I recommend it
  696. 24:51locally, it could also be used in the
  697. 24:53cloud.
  698. 24:54If we go to tools like Luteos or
  699. 24:56Alliteration, from there they will
  700. 24:58really let us run models that are open
  701. 25:01source and uncensored.
  702. 25:03Keep in mind that we will be able to use
  703. 25:05them with private servers, but there
  704. 25:07will always be some cost, since we are
  705. 25:09theoretically running them with more
  706. 25:11security.
  707. 25:12However, in many repositories we can
  708. 25:14find, for example on Hugging Face, we
  709. 25:17can also find models like DeepSeek
  710. 25:19version 4 Pro based on GPT-3.5, which is
  711. 25:21completely uncensored.
  712. 25:23These also don't require much hardware
  713. 25:25and we can install them in a similar way
  714. 25:27or as we have seen in previous steps,
  715. 25:29only instead of Ollama, we have to do it
  716. 25:32with other providers like Hugging Face.
  717. 25:34With all this, DeepSeek shows us that it
  718. 25:36is capable of doing a lot and that, on
  719. 25:38many occasions, it is not necessary to
  720. 25:40have the best hand to win, since
  721. 25:42sometimes moving forward with what we
  722. 25:43have can change the course entirely.

About this transcript

This page contains the full transcript of Cómo usar DeepSeek Mejor que el 99% de las personas (GRATIS) by Alejavi Rivera, generated from the public captions YouTube serves with the video. The transcript has 4,660 words across 722 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.