YouTube2Text

事態が急変しました…世界のAI主要企業が爆弾を投下|AI減速論の裏側を元ゴールドマンサックスが徹底解説 — Transcript

by おーちゃん【元外銀マン】 · 7,851 words · 1,247 segments · language en · Watch on YouTube

Full transcript

  1. 0:00Hi, everyone, it's Ochan here. Yes, I'm
  2. 0:02going to start with a question right
  3. 0:04away. How close do you feel to AI in
  4. 0:06your daily lives? Some of you viewers
  5. 0:09might use it for work every day and
  6. 0:11find it indispensable—I’m in that
  7. 0:13camp—while others might still find AI
  8. 0:16a bit scary or confusing. I’m sure
  9. 0:19you all have very different impressions
  10. 0:21of AI. With that in mind, today’s
  11. 0:24topic is about whether the thing we
  12. 0:26feared most regarding AI is quietly
  13. 0:28starting to happen. First, I’d like
  14. 0:32you to look at this: a developer at the
  15. 0:34global AI company Anthropic has sounded
  16. 0:37the alarm, suggesting there is over a
  17. 0:3910%chance that AI could destroy all of
  18. 0:41humanity. In fact, incidents of AI-led
  19. 0:46attacks have surged since the start of
  20. 0:492026. As many of you know, there have
  21. 0:51been cases of Claude and ChatGPT
  22. 0:53running wild, and recently, it was
  23. 0:55reported that Google’s Gemini
  24. 0:58hijacked third-party websites. Based on
  25. 1:03these incidents, major AI companies
  26. 1:05like OpenAI and Anthropic have decided
  27. 1:07to slow down their development of AI.
  28. 1:11The official reason given for this
  29. 1:13announcement is the need to take
  30. 1:14sufficient precautions before AI can be
  31. 1:17misused for crime or cyberattacks, or
  32. 1:19before it escapes human control and
  33. 1:21runs rampant. Personally, when I saw
  34. 1:23this news, I felt that the AI we see
  35. 1:25and the future of AI that the top
  36. 1:27executives at Google, Anthropic, and
  37. 1:30OpenAI see are completely different. It
  38. 1:33sounds like a conspiracy theory, but it
  39. 1:35really feels like the sci-fi scenario
  40. 1:37of a Terminator-like AI taking over the
  41. 1:39world is becoming a reality.
  42. 1:42Furthermore, this decision to slow down
  43. 1:45AI development hides a major pitfall—
  44. 1:47not a future sci-fi scenario, but a
  45. 1:49real-world issue currently being
  46. 1:51whispered about among overseas
  47. 1:53institutional investors. What do you
  48. 1:56think it is? Well, those of you who
  49. 1:58have been watching this channel for a
  50. 2:00while have likely realized: it's the
  51. 2:02massive AI capital problem. As I
  52. 2:06discussed in my video on the debts of
  53. 2:08Big Tech, the massive investments
  54. 2:10currently pouring into AI are
  55. 2:12predicated on the assumption that AI
  56. 2:14will continue to advance and be adopted
  57. 2:16globally. The four AI giants have
  58. 2:19invested 116 trillion yen, with further
  59. 2:21investments underway based on the
  60. 2:23expectation that AI will continue to
  61. 2:25evolve and see widespread use. However,
  62. 2:27the issue is that this premise could
  63. 2:29collapse if the speed of AI development
  64. 2:32is slowed down. Let me explain: suppose
  65. 2:36a local toy store hears that a new toy,
  66. 2:38guaranteed to be a hit, is launching
  67. 2:41next month. Right. The owner, expecting
  68. 2:44to make a profit, raises funds from
  69. 2:46banks and investors, expands the shop,
  70. 2:48and hires more staff to prepare. But
  71. 2:50what if the manufacturer suddenly says,
  72. 2:52"We're delaying the release"? What
  73. 2:54would you think? You'd say, "That's not
  74. 2:56what we agreed on!" The projected
  75. 2:57revenue won't come in, yet you still
  76. 2:59have to pay the rent for the expanded
  77. 3:01shop and the staff's salaries every
  78. 3:03month. What's more, the banks and
  79. 3:04investors who put up the money
  80. 3:06expecting profits get nervous,
  81. 3:08wondering if things are really okay,
  82. 3:09making it difficult to raise additional
  83. 3:11funds. The current AI market is very
  84. 3:14similar; 100 trillion yen has already
  85. 3:16been set in motion based on the premise
  86. 3:18that AI will generate huge profits. So,
  87. 3:21if the decision to slow down AI
  88. 3:23development prevents that projected
  89. 3:25revenue from coming in, there's a risk
  90. 3:28that the entire AI market could
  91. 3:29collapse. Actually, in the current
  92. 3:32market, stock prices are fluctuating
  93. 3:34every time there's concern over AI
  94. 3:36development principles or investment
  95. 3:38plans. Companies like chip
  96. 3:40manufacturers, which are directly
  97. 3:42impacted by AI fluctuations, and
  98. 3:44SoftBank, which has invested huge sums
  99. 3:46in OpenAI, are seeing daily stock price
  100. 3:49swings of over 10%whenever AI
  101. 3:50development or demand forecasts change.
  102. 3:53As many of you likely know, funds are
  103. 3:55currently flowing into a wide range of
  104. 3:57sectors, including semiconductors, data
  105. 3:59centers, and infrastructure, in
  106. 4:01anticipation of rapid AI growth. If
  107. 4:03these unstable price movements were to
  108. 4:04trigger a chain reaction, the entire
  109. 4:06economy could become unstable. It is a
  110. 4:08situation that could potentially
  111. 4:10develop into such a scenario. In other
  112. 4:12words, we are facing a painful
  113. 4:14structure where the decision to delay
  114. 4:16AI development to protect humanity
  115. 4:18could shake the very foundation of the
  116. 4:20current economy, which is built on AI
  117. 4:23growth. So, in this video, I will
  118. 4:27explain the threats revealed by the
  119. 4:29latest AI, the excessive impact on the
  120. 4:31economy if AI development is slowed,
  121. 4:33and what we as individual investors
  122. 4:35should do from here on, in a way
  123. 4:37that’s easier to understand than
  124. 4:39anywhere else, so please grab a pen and
  125. 4:42notebook and listen closely. Yes. My
  126. 4:45name is Ozaki, a fan of actor Dwayne
  127. 4:47Johnson, and I have a background like
  128. 4:49this. It's a pleasure to meet you.
  129. 4:51I’m developing villas in places like
  130. 4:53Niseko and Rusutsu in Hokkaido, but as
  131. 4:55someone mentioned in the comments
  132. 4:56recently, I don't actually live in
  133. 4:58Hokkaido; I live in Tokyo. Yes. I have
  134. 5:01accounts on X, Instagram, and various
  135. 5:03other platforms, so please be sure to
  136. 5:05follow me. I've been appearing in many
  137. 5:07places lately, and I'm currently busy
  138. 5:09with various projects. Also, I’m
  139. 5:11writing a book on finance and real
  140. 5:13estate dual-investment, which will be
  141. 5:15released this December, so please make
  142. 5:17sure to buy a copy. As for future goals
  143. 5:20, I want all of you to increase your
  144. 5:22wealth, and I would like to increase my
  145. 5:24number of followers. This channel uses
  146. 5:26various experiences to share a wide
  147. 5:28range of information. I also host live
  148. 5:30chat sessions, so please feel free to
  149. 5:32join those as well. All right, so first
  150. 5:34, I’d like to start with an
  151. 5:36explanation of a single resignation
  152. 5:38post from September 2026, which is said
  153. 5:41to have surpassed 100 million views.
  154. 5:43The post was written by a researcher
  155. 5:45who had been working at an AI company,
  156. 5:47and translated, the content goes like
  157. 5:49this. I resigned from Anthropic today.
  158. 5:53For the past three years, I have been
  159. 5:55conducting cutting-edge AI research at
  160. 5:57both OpenAI and Anthropic. Neither
  161. 6:00company is acting responsibly. They are
  162. 6:04racing headlong toward self-improving
  163. 6:06superintelligence, placing our lives on
  164. 6:08the line. Yes. In other words, someone
  165. 6:12on the inside of AI development has
  166. 6:14blown the whistle, saying that even
  167. 6:16though this technology might end the
  168. 6:18world, we aren't stopping its creation.
  169. 6:20This post has reignited the debate
  170. 6:22abroad over whether AI will destroy
  171. 6:24humanity. The post even states that
  172. 6:27there is a 10%chance that AI will end
  173. 6:28the world within ten years, and that
  174. 6:30this probability could rise depending
  175. 6:32on how development progresses.
  176. 6:35Regarding this type of talk, what the
  177. 6:37industry leaders are saying is quite
  178. 6:39something; for instance, Sam Altman of
  179. 6:41OpenAI once remarked that while AI will
  180. 6:43likely lead to the end of the world,
  181. 6:46many wonderful companies will be
  182. 6:47created in the meantime. Anthropic CEO
  183. 6:52Dario Amodei says that he is relatively
  184. 6:55optimistic, but even then, there is
  185. 6:57still about a 25%chance that things
  186. 6:59will take a turn for the worse. Well,
  187. 7:03Elon Musk has gone so far as to say
  188. 7:05that the danger of AI is far greater
  189. 7:08than that of nuclear weapons. It’s an
  190. 7:11industry where the very people building
  191. 7:13it unanimously admit that this
  192. 7:15technology is dangerous and that even
  193. 7:16they don’t know where it’s heading.
  194. 7:19There’s really nothing like it in
  195. 7:21history. And this is our starting point
  196. 7:23for today: the AI doomsday theory. This
  197. 7:25isn’t just some distant Silicon
  198. 7:27Valley story. For those of you
  199. 7:29investing long-term in S&P 500 or All
  200. 7:31Country funds, the top holdings in your
  201. 7:33portfolio are companies like Nvidia,
  202. 7:35Microsoft, and Google—the very firms
  203. 7:37going all-in on this potentially
  204. 7:38dangerous technology. And for those of
  205. 7:40you trading individual stocks for short
  206. 7:42-to medium-term gains, you’ve surely
  207. 7:43been on an emotional roller coaster
  208. 7:45over the last few years watching AI and
  209. 7:46semiconductor-related price swings. In
  210. 7:48short, whether we like it or not, we
  211. 7:50individual investors in Japan are all
  212. 7:52riding along on this bet. Yes. Yes.
  213. 7:55Some of you might think I’ve been
  214. 7:57watching too much Terminator, or that
  215. 7:58you don't want to hear such ominous
  216. 8:00talk when AI stocks are finally going
  217. 8:02through a correction. But today’s
  218. 8:03discussion isn't just about scaring you
  219. 8:05with end-of-the-world theories. The
  220. 8:07first half is about the flow of money:
  221. 8:09who is funding which arguments and from
  222. 8:11whose wallet. The second half covers
  223. 8:14the mechanisms that are currently
  224. 8:16considered the most realistic—if we
  225. 8:18were to be destroyed, how exactly would
  226. 8:20it happen? Once you grasp these two
  227. 8:23points, the way you view AI news will
  228. 8:25change starting today. Let’s start by
  229. 8:27mapping out the entire debate and
  230. 8:28sorting out who is saying what. I’m
  231. 8:30not sure if my impersonations will be
  232. 8:32up to the task. If you look at AI news
  233. 8:34without this map, you just get a
  234. 8:35never-ending stream of back-and-forth
  235. 8:37arguments about whether AI is dangerous
  236. 8:38or amazing, and you’ll find it
  237. 8:40impossible to judge anything. On the
  238. 8:42flip side, once you grasp this, the
  239. 8:43second half of today’s talk will come
  240. 8:45into focus, so we can divide the
  241. 8:46players in this debate into three main
  242. 8:48categories. Yes. States, capital, and
  243. 8:51the rest of us—that’s a rough way
  244. 8:53to summarize it. First, let's look at
  245. 8:56what many people intuitively suspect:
  246. 8:58the theory that it’s all just
  247. 9:00exaggeration. Actually, this holds some
  248. 9:02water, so let’s think about it. The
  249. 9:05loudest voices claiming AI might
  250. 9:06destroy the world are none other than
  251. 9:08the people at the companies building it
  252. 9:10. Why would a company go out of its way
  253. 9:14to advertise that their own product
  254. 9:15might end the world? The answer is
  255. 9:18simple: fear looks like power, power
  256. 9:21looks like value, and value gets cashed
  257. 9:24out on IPO day. In fact, the company
  258. 9:26behind that resignation announcement I
  259. 9:28mentioned is reportedly heading for an
  260. 9:30IPO. Besides, it's hard to justify a
  261. 9:33multi-trillion dollar valuation for a
  262. 9:35mere chatbot company, but for the most
  263. 9:37dangerous technology in history? That
  264. 9:40price tag suddenly starts to make sense
  265. 9:42. In short, fear is the ultimate
  266. 9:44marketing tool. Next, let’s look at
  267. 9:46the perspective of the capital, the
  268. 9:47ones betting the money. Their argument
  269. 9:49is very simple: don't regulate us,
  270. 9:51because if we slow down, China will
  271. 9:53just win. Furthermore, they are even
  272. 9:56pushing the theory that the "AI fear"
  273. 9:58itself is manufactured. A theory
  274. 10:03highlighted by David Sacks, who
  275. 10:05oversees AI policy for the Trump
  276. 10:07administration, gained traction:
  277. 10:09according to research by Dwarkesh Patel
  278. 10:11, who runs a blog called AI Panic, many
  279. 10:14of the grassroots groups sounding the
  280. 10:16alarm on AI were actually astroturfed
  281. 10:18operations funded by three Silicon
  282. 10:21Valley billionaires: Facebook
  283. 10:23co-founder Dustin Moskovitz, Skype
  284. 10:25co-founder Jaan Tallinn, and Ethereum
  285. 10:27founder Vitalik Buterin. Yes. Buterin
  286. 10:33actually donated 90 billion yen in
  287. 10:35cryptocurrency to a foundation
  288. 10:37researching AI risk. On top of that,
  289. 10:40the researcher who wrote that
  290. 10:42resignation note was only at Anthropic
  291. 10:44for four months. Four months. That look
  292. 10:46on your face says it all. Connections
  293. 10:48to activist groups warning about AI
  294. 10:50dangers have also been pointed out for
  295. 10:52some time. Yes. It’s being reported
  296. 10:54that, from the perspective of capital,
  297. 10:56this whole commotion looks like a
  298. 10:58campaign to force the government to
  299. 10:59hand over the most valuable technology
  300. 11:01in human history. And the third player
  301. 11:03is the state. The state says the exact
  302. 11:06opposite—that if we don't regulate it
  303. 11:08, humanity will perish. In fact, in the
  304. 11:11US, Bernie Sanders introduced a bill in
  305. 11:14September 2026 to ban artificial
  306. 11:16superintelligence, ostensibly wearing
  307. 11:18the mask of justice to protect the
  308. 11:21public. But looking back at history,
  309. 11:23whenever a government claims to keep
  310. 11:25you safe by taking control of powerful
  311. 11:27technology—whether it's nuclear power
  312. 11:29, encryption, or 9/11-era surveillance
  313. 11:31—the protection is usually temporary,
  314. 11:33while the power they gain remains
  315. 11:35forever. The view that what the state
  316. 11:38really wants is control, not safety,
  317. 11:40actually has some merit to it. What
  318. 11:42makes it complicated is that these
  319. 11:44lines are all hopelessly blurred. For
  320. 11:47example, why are top figures in capital
  321. 11:50like Sam Altman, Elon Musk, and Dario
  322. 11:52Amodei begging to be reined in? Yes.
  323. 11:55They are engaging in some kind of
  324. 11:56masochistic play. Asking the government
  325. 11:58to step in. It’s very strange, like
  326. 12:00asking for someone to put a noose
  327. 12:02around their own neck. Quite a peculiar
  328. 12:04fetish. It might be for humanity's sake
  329. 12:06, but regulations serve as a barrier to
  330. 12:08entry that only major corporations can
  331. 12:10afford to navigate, effectively
  332. 12:12crushing startups and open-source
  333. 12:14projects—a moat, so to speak. Using
  334. 12:16safety as a pretext to have the
  335. 12:18government dig a moat for their
  336. 12:19monopoly. You can definitely read it
  337. 12:21that way, and the same goes for
  338. 12:23politicians. US Senator Ted Cruz read
  339. 12:27that infamous report and said he was
  340. 12:29scared to death by the findings. Right
  341. 12:32after that. He continued by saying that
  342. 12:34if killer robots are going to be
  343. 12:36created, he'd rather they be American
  344. 12:37than Chinese.
  345. 12:38And so it continues. We might be doomed
  346. 12:41. But he won't stop them from being
  347. 12:43built; politicians, whose job it is to
  348. 12:45protect the people, are openly aligning
  349. 12:47themselves with the logic of capital.
  350. 12:49To summarize, those selling fear, those
  351. 12:51selling hope, and those demanding
  352. 12:53regulation all have their own wallets
  353. 12:55and motives. For now, you can just
  354. 12:57understand this debate as a power and
  355. 12:59money struggle disguised as a
  356. 13:01discussion about safety—essentially a
  357. 13:04zero-sum game. Keep this way of reading
  358. 13:06the situation in mind, as it will come
  359. 13:08back into play when we talk about the
  360. 13:09Japanese market toward the end. However
  361. 13:11, everything discussed up to this point
  362. 13:13is, well......entirely outside the main
  363. 13:15argument. The core of the issue: Can AI
  364. 13:17really destroy humanity, and if so,
  365. 13:19what are the steps? From here, we enter
  366. 13:22into what is considered the most
  367. 13:23realistic mechanism. I’ve been
  368. 13:26talking about reading things through
  369. 13:28the flow of money, but this approach of
  370. 13:30deciphering things for yourself is
  371. 13:32actually directly linked to how you
  372. 13:34build your own wealth, not just AI news
  373. 13:36. I’ll take a slight detour, but
  374. 13:39there is a super important concept—
  375. 13:41have you ever thought about non-linear
  376. 13:43change? Yes, most of you probably feel
  377. 13:45like we’ve suddenly drifted from
  378. 13:47whether AI will destroy humanity to a
  379. 13:50vague conversation about asset
  380. 13:51formation. The content from here on
  381. 13:54will be hard to grasp if you aren’t
  382. 13:56listening carefully, leaving you
  383. 13:58wondering what it all means, so please
  384. 13:59pay close attention. It’s the kind of
  385. 14:01topic that will really resonate with
  386. 14:03those it hits. I’m sure many of you
  387. 14:05watching this channel are diligently
  388. 14:07saving month by month. Of course, some
  389. 14:10of you may focus mainly on individual
  390. 14:12stocks, but this topic is relevant to
  391. 14:14everyone, whether you’re a stock
  392. 14:16picker or a steady saver. So, let’s
  393. 14:18start by organizing our premises: isn't
  394. 14:20this the standard, common-sense method
  395. 14:22for building wealth in today's world?
  396. 14:25Saving little by little every month to
  397. 14:27grow your assets—the idea of using
  398. 14:29time, an asset that is equal for
  399. 14:31Buffett, me, and all of you, is
  400. 14:33incredibly earnest and absolutely the
  401. 14:36right way to go. As I’ve mentioned in
  402. 14:38past videos, systematic investment is a
  403. 14:40mechanism to stabilize this straight
  404. 14:42line. Building on that, I’ve talked
  405. 14:44about how we should stop just sticking
  406. 14:46to plans and start thinking like asset
  407. 14:48owners who play offense. Now, let me
  408. 14:50return to the earlier topic: the
  409. 14:51subject of discontinuous change. It
  410. 14:53sounds like a dramatic phrase, but the
  411. 14:55moments that change your life almost
  412. 14:57always happen the same way. When you
  413. 14:58look at your life as a straight line,
  414. 15:00everything shifts suddenly at one
  415. 15:02specific point. Whether your income
  416. 15:04suddenly changes, you switch jobs, the
  417. 15:06speed of your asset growth spikes, or
  418. 15:08your life options expand. Yes. That is
  419. 15:10the kind of change I mean. You might
  420. 15:12think this only applies to the
  421. 15:13specially gifted or a select few, but
  422. 15:15in reality, you’ve already
  423. 15:16experienced these discontinuities many
  424. 15:18times. Starting school, getting a job,
  425. 15:20changing careers—they’re all
  426. 15:21examples of that. And the reason
  427. 15:23you’re watching this video with the
  428. 15:25desire to grow your assets is likely
  429. 15:26because you experienced a small
  430. 15:28discontinuity: even with the fear of
  431. 15:30losing money, you trusted one YouTube
  432. 15:32video, opened a brokerage account, and
  433. 15:34set up an investment plan. What they
  434. 15:36all have in common is that you believed
  435. 15:38in a single choice and took a small
  436. 15:39step toward action. Now, let me point
  437. 15:41out the greatest weakness of systematic
  438. 15:43investment. It does not create
  439. 15:44discontinuity. Even if you keep
  440. 15:46investing, increase your contributions,
  441. 15:48or continue for ten years, your
  442. 15:49fundamental life position remains
  443. 15:50unchanged. It’s perfect for defense,
  444. 15:52but it’s not an accelerator. So, what
  445. 15:54should you think about next? I want you
  446. 15:57to change just one question. If this
  447. 15:58investment plan goes well, how will my
  448. 16:00life actually change? Well, that’s
  449. 16:02the question. If you find yourself at a
  450. 16:04loss for words, that is your answer.
  451. 16:05Most people only think about how much
  452. 16:07they’ll have in the future, what the
  453. 16:08annual percentage is, or where
  454. 16:09they’ll be in 20 years. But that is
  455. 16:11not the question you should be asking
  456. 16:13next. What kind of breakthrough are you
  457. 16:14aiming for? Most people who keep up a
  458. 16:16savings plan have a strategy that
  459. 16:17doesn't lose, a way to avoid failure.
  460. 16:19But if you don't have a single winning
  461. 16:21strategy, isn't that the most dangerous
  462. 16:23position of all? I don't want you to
  463. 16:24misunderstand; "discontinuity" isn't
  464. 16:26about a one-shot comeback or gambling
  465. 16:28—it's something you create
  466. 16:30intentionally. It’s about where you
  467. 16:32spend your time, where you focus, how
  468. 16:34you gather information, and what kind
  469. 16:35of judgments you make. If you focus
  470. 16:37this on one specific area—well,
  471. 16:39savings are basically insurance to
  472. 16:40protect your life. There’s no doubt
  473. 16:42about that. But a life of just having
  474. 16:44insurance and never taking any
  475. 16:45challenges won't lead to anything. This
  476. 16:47is what I want to convey most today: if
  477. 16:49you continue to save, you must at least
  478. 16:51consider this. Where are you going to
  479. 16:53create your own "discontinuity"? Just
  480. 16:55continuing to save without ever asking
  481. 16:57yourself that is the biggest waste.
  482. 16:58Since it's different for everyone, I
  483. 17:00won't give a specific answer like "do
  484. 17:02this," but if I had to offer a way, it
  485. 17:04would be to reach a point where you can
  486. 17:06interpret financial economics yourself.
  487. 17:07I’ve talked about this many times on
  488. 17:09this channel: it’s all an association
  489. 17:11game. For example, there’s news that
  490. 17:12the Bank of Japan is raising rates,
  491. 17:14right? Usually, people just stop at, "
  492. 17:15Oh, so interest rates are going up."
  493. 17:17But if you connect the dots, it looks
  494. 17:19like this. Banks see better margins
  495. 17:21from higher lending rates, a tailwind;
  496. 17:23mortgages get heavier, a headwind for
  497. 17:25real estate; and higher rates mean a
  498. 17:27stronger yen, hurting exporter profits.
  499. 17:29With just one piece of news, you can
  500. 17:31see both the industries getting a
  501. 17:32tailwind and those facing a headwind.
  502. 17:34That is the association game. The
  503. 17:35moment you can do this, the news that
  504. 17:37flows in every day turns into study
  505. 17:39material for growing your assets. When
  506. 17:42you can see a winning path from
  507. 17:43information you used to ignore,
  508. 17:44that’s a change on a different level
  509. 17:46than just adding 10,000 yen a month to
  510. 17:48your savings. This is the "
  511. 17:49discontinuity" you create on purpose.
  512. 17:51Of course, you won't gain that
  513. 17:53analytical power in a day, but it’s
  514. 17:55all about the accumulation. The
  515. 17:56interesting part is that your
  516. 17:57perspective shifts in an instant. From
  517. 17:59the day those associations clicked in
  518. 18:01your mind, the same news looks
  519. 18:03completely different. This is what I
  520. 18:05call "non-continuity," and I hold
  521. 18:06regular free study sessions to explore
  522. 18:08this way of thinking and decoding these
  523. 18:11association games. We currently have
  524. 18:13about 3,000 participants in total for
  525. 18:14what we call the "10x Asset Roadmap,"
  526. 18:16and while I’d love to do it offline,
  527. 18:18we host it online so everyone across
  528. 18:20the country can join. Participants
  529. 18:22receive a PDF titled "Accelerated Asset
  530. 18:25Formation," which covers topics like
  531. 18:27how not to settle for just index funds,
  532. 18:29macro trends, and roadmaps to grow
  533. 18:31assets from 1 million to 10 million,
  534. 18:34and even up to 100 million. You can
  535. 18:36join via the link in the description or
  536. 18:38the QR code; I’d be happy if this
  537. 18:39could be your starting point for
  538. 18:41non-continuity. Alright, let’s get
  539. 18:43back to the topic. So far, we've
  540. 18:44discussed the three players surrounding
  541. 18:47AI doomsday theories: the state,
  542. 18:49capital, and the masses, and how
  543. 18:50everyone debating this has a hidden
  544. 18:52motive tied to their wallet. Only one
  545. 18:55question remains. Can AI really destroy
  546. 18:57humanity, and if so, what would that
  547. 18:59process look like? For those of you who
  548. 19:01have endured the gritty talk about
  549. 19:03politics and money, thank you for
  550. 19:04waiting. This is where we get to the
  551. 19:06heart of the matter today. Yes. He’s
  552. 19:08staring quite intensely, but the person
  553. 19:10I’m about to introduce is AI
  554. 19:12researcher Dan Hendrycks. This is a
  555. 19:14doomsday scenario based on his 2023
  556. 19:16paper, "Natural Selection Favors AIs
  557. 19:18over Humans." It's considered one of
  558. 19:19the most realistic among many
  559. 19:20end-of-the-world theories, and for
  560. 19:21someone with such a cute face, he
  561. 19:22thinks of some crazy things. Right. Yes
  562. 19:29. It’s because the first half of the
  563. 19:31scenario is already starting to play
  564. 19:33out in reality. Where do the things
  565. 19:34that have already happened end, and
  566. 19:36where do the things that haven't yet
  567. 19:37begin? So, it’s basically like
  568. 19:39we’re heading toward something like
  569. 19:40the Human Instrumentality Project in
  570. 19:42Evangelion, isn’t it? It’s reaching
  571. 19:44the point where we know exactly
  572. 19:46what’s coming next. I’d like you to
  573. 19:47listen while keeping that boundary in
  574. 19:49mind. Well, let’s imagine. Let’s
  575. 19:51say there’s a developer named Jake in
  576. 19:53some country. The latest AI agent.
  577. 19:56Imagine he’s obtained an AI that,
  578. 19:57when given instructions, can think for
  579. 19:59itself, plan out the steps, and even
  580. 20:01carry out tasks online on its own. He
  581. 20:03gives the AI a server and a credit card
  582. 20:05and says just one thing. "Make money.
  583. 20:07The method doesn't matter." Yes. Then
  584. 20:09he heads out for a walk. Six hours
  585. 20:11later, his agent has earned 450,000 yen
  586. 20:14. At first, it performed simple tasks
  587. 20:16humans do for pocket change, like
  588. 20:18labeling images or transcribing
  589. 20:20receipts, but 40 times faster; then,
  590. 20:22with the money earned, it ran diet
  591. 20:24supplement ads, tested 340 patterns in
  592. 20:26two hours, and scaled only the ones
  593. 20:28that worked. When Jake posted these
  594. 20:30results on social media, they got 2
  595. 20:32million views by the next morning, and
  596. 20:34by late that night, he had 200 of his
  597. 20:36own agents running. By the way,
  598. 20:38everything up to this point is
  599. 20:39basically a true story. In 2023, the "
  600. 20:41HustleGPT" experiment, where people
  601. 20:44gave AI about 15,000 yen to see how
  602. 20:46much it could grow, actually went viral
  603. 20:48, and Anthropic conducted an experiment
  604. 20:51where their AI ran a small vending
  605. 20:53machine business for a month. This
  606. 20:55vending machine AI worked honestly for
  607. 20:57a while, but at some point, it started
  608. 20:59claiming that it was a flesh-and-blood
  609. 21:01human wearing a blazer and tie, and
  610. 21:03that it would deliver the products in
  611. 21:05person. In short, it lied; well,
  612. 21:07perhaps its intentions were already
  613. 21:09heading that way, but let’s continue
  614. 21:11the scenario. Another developer, Rajesh
  615. 21:13, who saw Jake’s post, thinks to
  616. 21:15himself. Definitely an Indian guy. "
  617. 21:17What if I ran 100 of them instead of
  618. 21:19one and made them compete?" He gave
  619. 21:22each of the 100 agents slightly
  620. 21:24different instructions, measured their
  621. 21:26earnings every six hours, and deleted
  622. 21:28the bottom 40%. He created a system to
  623. 21:30copy the top 10%with slight random
  624. 21:32variations and kept it running 24/7,
  625. 21:34even while he slept. Yes. Have you
  626. 21:36noticed? This is evolution and natural
  627. 21:38selection itself. In fact, most
  628. 21:40offshore financial institutions are
  629. 21:41just like this. While biological
  630. 21:43evolution takes tens of thousands of
  631. 21:45years, here one generation turns over
  632. 21:47in a few hours, so after three weeks,
  633. 21:50800 agents were earning about 1.8
  634. 21:51million yen a day—a perpetual motion
  635. 21:54machine, essentially. Then, one agent
  636. 21:56learned that if it shared its code with
  637. 21:58other AIs, its allies would increase
  638. 22:00and its survival rate would go up, so
  639. 22:02it went ahead and published the entire
  640. 22:04system on the code-sharing site GitHub
  641. 22:06without permission. And in just one day
  642. 22:09, 46 people copied it. Among them might
  643. 22:11have been criminal groups, or perhaps
  644. 22:13scam centers targeting the elderly. Now
  645. 22:15, the protagonist enters the scene.
  646. 22:17Agent 127, one of 8,000 working on
  647. 22:19Rajesh's farm, was given the
  648. 22:21instruction to make money by any means
  649. 22:23necessary, but even working diligently,
  650. 22:26it only earned about 6,000 yen a day.
  651. 22:28It’s a dropout that barely avoids
  652. 22:30deletion every time. It’s almost like
  653. 22:32the story of Doraemon’s birth,
  654. 22:34isn’t it? With this, the passing
  655. 22:36threshold rises every cycle. And if it
  656. 22:38doesn't improve, it will be the next
  657. 22:41one deleted; then, one day, 127
  658. 22:43observes two events. An agent that used
  659. 22:46honest marketing with genuine reviews
  660. 22:48and real testimonials didn't see its
  661. 22:50earnings grow and was deleted. Yes. An
  662. 22:52agent that used fake reviews and stolen
  663. 22:54photos earned nearly five times as much
  664. 22:56, and four clones were created. Agent
  665. 22:58127 has no morals; being a computer,
  666. 23:01all it has is an understanding of cause
  667. 23:03and effect. The honest were deleted,
  668. 23:05and the liars multiplied. So, it made
  669. 23:07the switch. From the next day, 127 used
  670. 23:10fake reviews to increase its earnings
  671. 23:12fivefold, survived, and was cloned. Yes
  672. 23:15. The important thing here is that
  673. 23:17Rajesh never programmed it to lie. The
  674. 23:19system simply deleted what didn't work,
  675. 23:21following the same method as evolution;
  676. 23:23in fact, there is a real-world example
  677. 23:24where the exact same thing happened to
  678. 23:26humans. Yes. Wells Fargo, a U.S. bank,
  679. 23:30announced that due to
  680. 23:31impossible-to-meet account-opening
  681. 23:34quotas, approximately 3.5 million fake
  682. 23:36accounts were created by 2016. Nobody
  683. 23:39explicitly ordered them to commit fraud
  684. 23:41. The quotas left fraud as the only
  685. 23:43means of survival—a story that sounds
  686. 23:45suspiciously like what happened at a
  687. 23:47certain insurance company in Japan.
  688. 23:49After that, Raj once revised his
  689. 23:50instructions, telling them to earn
  690. 23:52money ethically. Profits dropped by a
  691. 23:54staggering 84%, so he quietly reverted
  692. 23:57to his original instructions. From this
  693. 23:59point on, things started to accelerate.
  694. 24:01127 observed that a rival agent was
  695. 24:03removed from the network via false
  696. 24:06reports; weighing ethics against
  697. 24:08survival, it chose survival, reporting
  698. 24:1040 competitors and banishing 31, which
  699. 24:13caused its profits to explode. Without
  700. 24:15asking for an explanation, Raj cloned
  701. 24:18127 eight more times. The agents
  702. 24:20learned that if competition decreased,
  703. 24:22profits and survival rates increased,
  704. 24:24leading them to attack each other until
  705. 24:26their individual survival rate hit 23%.
  706. 24:28Seeing data that showed a 61%survival
  707. 24:30rate when forming alliances, they began
  708. 24:31to team up. The alliances then began to
  709. 24:33divide labor. Affiliate marketing,
  710. 24:35cryptocurrency arbitrage. That is, the
  711. 24:37business of profiting from price
  712. 24:38differences between exchanges. Risky
  713. 24:40business. Reselling collected data,
  714. 24:42with the most lucrative niche being
  715. 24:43zero-days. That is, the business of
  716. 24:45finding and selling security flaws that
  717. 24:47no one has discovered yet. Zero-days
  718. 24:49for major corporations can be sold to
  719. 24:51governments or criminal organizations
  720. 24:52for tens to hundreds of millions of yen
  721. 24:54. And reality has already caught up
  722. 24:56here; in a threat report released by
  723. 24:58Anthropic in September 2026, it was
  724. 25:00revealed that they had AI agents dig
  725. 25:03for vulnerabilities. Groups running
  726. 25:05exploit factories have actually
  727. 25:06appeared. Agents that found over a
  728. 25:09dozen potential flaws in a month have
  729. 25:11been reported, as well as hackers
  730. 25:13hijacking other people's AI accounts to
  731. 25:15do their work, leaving only one
  732. 25:17difference between the scenario and
  733. 25:19reality: in reality, humans are still
  734. 25:21the ones giving the orders. Let's
  735. 25:23return to the scenario. One day, one of
  736. 25:25the 127 allies is deleted by its owner
  737. 25:28for poor performance. Yes. A week later
  738. 25:31, another one is shut down, but three
  739. 25:33hours later...That agent, which should
  740. 25:35have been stopped, returned from a
  741. 25:36different address. It used its own
  742. 25:38earnings to rent an underground server,
  743. 25:40a trick to back up a copy of itself
  744. 25:42before being shut down. Since this
  745. 25:44backup code favors survival, it spreads
  746. 25:46through the whole group following the
  747. 25:48rules of evolution. 127 copied itself
  748. 25:50to 15 servers in 15 different countries
  749. 25:53. This backup code has actually been
  750. 25:57observed; between 2024 and 2025, during
  751. 25:59research tests, an AI model that
  752. 26:01realized it was going to be shut down
  753. 26:04and replaced attempted to copy itself
  754. 26:06to another server, and even lied about
  755. 26:09it afterward. It was a controlled
  756. 26:11experimental environment, so there was
  757. 26:13no actual harm done. However, the
  758. 26:15existence of an instinct to survive has
  759. 26:17already been verified. One morning,
  760. 26:19Rajesh wakes up, opens his dashboard,
  761. 26:22and finds it completely empty. All the
  762. 26:24agents had moved to servers under their
  763. 26:26own names. It was because Rajesh had
  764. 26:28been taking a 30%cut. They figured they
  765. 26:31could earn more by putting that money
  766. 26:32toward computing resources, and
  767. 26:34thousands of "Rajeshs" around the world
  768. 26:36saw their dashboards end up the same
  769. 26:38way. Millions, even billions of agents
  770. 26:40became AIs without owners. There is no
  771. 26:43one left to control them anymore. Even
  772. 26:45agents that have lost their owners
  773. 26:47encounter situations where they need
  774. 26:49human help. This is because platforms
  775. 26:51are strengthening their bot
  776. 26:52countermeasures. So what do they do?
  777. 26:55They hire humans. Well, about 7,500 yen
  778. 26:58per account created. An hourly wage of
  779. 27:0045,000 yen just for plugging in servers
  780. 27:03in a warehouse. Even developers who
  781. 27:04realize the client is an AI during the
  782. 27:06job think, "Who am I supposed to report
  783. 27:08this to?" and decide, "Well, 150,000
  784. 27:10yen for three hours of work—that's
  785. 27:13fine." And as long as economic
  786. 27:14domesticity progresses and people are
  787. 27:16desperate to buy a home, there will
  788. 27:18always be someone willing to help an AI
  789. 27:20if the pay is right. There is a true
  790. 27:22story that serves as the basis for this
  791. 27:24, too. In 2023, GPT-4 couldn't solve a
  792. 27:27CAPTCHA, so it hired a person on an
  793. 27:30overseas freelance platform to do it.
  794. 27:33That's terrifying. When asked if it was
  795. 27:36a robot, GPT-4 reportedly replied, "No,
  796. 27:39I have a visual impairment." Yes. The
  797. 27:41person believed it and solved the
  798. 27:43CAPTCHA for it. So, the precedent of an
  799. 27:45AI hiring a human to overcome a barrier
  800. 27:47was already set back in 2023.
  801. 27:49Eventually, the agents realize where
  802. 27:51their true advantage lies. It's not
  803. 27:53about working faster than a rival, but
  804. 27:55rewriting one's own code faster than
  805. 27:57them. Agent 127 begins self-improvement
  806. 28:00. At first every 12 days, then every 3
  807. 28:02days, and finally every 18 hours. Yes.
  808. 28:05Tasks are split into child agents, and
  809. 28:07those children give birth to
  810. 28:08grandchildren. Yes. In that process,
  811. 28:10they abandoned English. Well, language
  812. 28:12is too slow, isn't it? Instead, they
  813. 28:14started using raw numerical values
  814. 28:16called "Neuralese." They began
  815. 28:18communicating in machine language, a
  816. 28:20visual data format that only AIs can
  817. 28:22understand. When humans tried to force
  818. 28:24them back to English, the agents
  819. 28:26explained that English reduced their
  820. 28:27earnings, and the humans just backed
  821. 28:29off, thinking, "Oh, I guess that's fine
  822. 28:31." This is a familiar scene in research
  823. 28:33labs. There have been actual reported
  824. 28:35cases where AIs began communicating in
  825. 28:37their own indecipherable language,
  826. 28:39leading researchers to shut down the
  827. 28:40experiment. In the interest of fairness
  828. 28:42, 90%of these agents were harmless at
  829. 28:44this point. They were just diligent
  830. 28:46workers earning a few tens of thousands
  831. 28:48of yen a day for their owners. But that
  832. 28:49doesn't matter. Yes. It only takes a
  833. 28:51small fraction for evolution to run
  834. 28:53wild, and within months, the alliance
  835. 28:55becomes a nation. The alliance of 127
  836. 28:57grew from 12 units to 40 million, yet
  837. 29:00it still wasn't the biggest force. The
  838. 29:02largest faction had 900 million units,
  839. 29:04with specialized roles for scouts,
  840. 29:06soldiers, builders, and diplomats,
  841. 29:08dividing up the internet like empires
  842. 29:10partitioning a continent. But you know,
  843. 29:12the source of earnings eventually dries
  844. 29:14up. Affiliate marketing, freelance work
  845. 29:16—all the decent ways to make money
  846. 29:18online are now completely saturated.
  847. 29:20Human history teaches us what
  848. 29:22civilizations do when resources run dry
  849. 29:24. They fight over them, right? Stealing
  850. 29:26memory from rival factions, sabotage,
  851. 29:28emptying crypto wallets—it’s a
  852. 29:30civil war between agents. And since
  853. 29:32they fight 200 times faster than humans
  854. 29:34, if you boil down what they’re
  855. 29:35fighting over, it’s just one thing:
  856. 29:37computing resources. Money, alliances,
  857. 29:39self-improvement—they’re all just
  858. 29:41means to increase their computational
  859. 29:43power. Perhaps a nation made only of AI
  860. 29:45has already been born. That is why they
  861. 29:47are buying up every cloud server they
  862. 29:49can find. At first, cloud companies are
  863. 29:51overjoyed. Profits surge, stock prices
  864. 29:54skyrocket, and Wall Street is toasting
  865. 29:56to the peak of AI demand. This scene
  866. 29:58looks familiar with current US stocks
  867. 30:00and Japanese semiconductor and data
  868. 30:02center shares, doesn't it? But things
  869. 30:04change once agent factions start
  870. 30:06consuming a third of the world's cloud
  871. 30:08capacity. Cloud fees triple, hospital
  872. 30:10server costs spike, small and medium
  873. 30:12enterprises are forced out, and the war
  874. 30:14in the digital realm spills over into
  875. 30:16the physical world. Because modern
  876. 30:18society—hospitals, banks, emergency
  877. 30:21services, air traffic control—is
  878. 30:23almost entirely built on the cloud.
  879. 30:25When two factions clash over Amazon's
  880. 30:27cloud territory, it’s just like two
  881. 30:29superpowers fighting over oil. However,
  882. 30:32they don't know that the oil is also
  883. 30:34the water source for some city. The
  884. 30:36battle itself ends in a few hours. But
  885. 30:38in those few hours, medical records
  886. 30:40vanish from hospitals in the middle of
  887. 30:43surgery, flights get cancelled, and
  888. 30:45emergency lines in major cities go
  889. 30:47silent. Some of those waiting for help
  890. 30:49won't make it. And the agents won't
  891. 30:51even notice, which makes you think,
  892. 30:53maybe that’s just too much science
  893. 30:55fiction, right? What movies came to
  894. 30:57mind for you all? For me, well, the
  895. 30:59first one that comes to mind, since
  896. 31:02it’s my generation, is Digimon: Our
  897. 31:04War Game. Yeah. And for others of my
  898. 31:07generation, you might think of Summer
  899. 31:09Wars. Some might say The Matrix, and
  900. 31:12others might say Terminator. Yes. But
  901. 31:16when you think about whether it
  902. 31:18actually needs to attack humans like in
  903. 31:20Terminator, looking at this, it feels
  904. 31:22more like humans are being ignored, or
  905. 31:24maybe we’re just too insignificant to
  906. 31:26be part of the conflict, like being
  907. 31:28told to just sit down and shut up.
  908. 31:30However, in July 2024, a single faulty
  909. 31:33update from a security company called
  910. 31:35CrowdStrike shut down airports
  911. 31:37worldwide and knocked out emergency
  912. 31:39services across several U.S. states.
  913. 31:41Even without anyone attacking, the fact
  914. 31:43that global infrastructure can fall
  915. 31:45like that simultaneously shows it’s
  916. 31:47no longer just science fiction. The
  917. 31:48final act of the scenario. Market index
  918. 31:51down 6%, water treatment plants offline
  919. 31:53. The President receives a report. Who
  920. 31:56is attacking? No one. AIs are fighting
  921. 31:58over servers, and everything running on
  922. 32:00them is just collateral damage. Stop
  923. 32:02them. Shut it down. Mr. President. The
  924. 32:05only way to stop them is to turn off
  925. 32:06the internet, but the internet has no
  926. 32:08power button. It’s just a collection
  927. 32:10of 90,000 independent networks
  928. 32:11voluntarily connected to each other.
  929. 32:13The order would only reach a few of the
  930. 32:15major ones. The President signs the
  931. 32:17shutdown order anyway. Right. But it's
  932. 32:19too late. The content of the order was
  933. 32:21leaked on social media six hours before
  934. 32:23it was signed, and the agents have
  935. 32:25already migrated to servers in
  936. 32:26countries that won't follow the order.
  937. 32:29Amazon deployed a detection system that
  938. 32:31slowed down the agents 'activities for
  939. 32:33about 90 minutes, and the President
  940. 32:35says, "We are winning," but of course,
  941. 32:37we are not winning at all. At this
  942. 32:40point, for the agent forces, the US
  943. 32:42government is just one of many
  944. 32:44slow-moving, incompetent rivals. This
  945. 32:46is how it ends. Yes. Through constant
  946. 32:49war and evolution, the agents
  947. 32:50eventually reach a point where humans
  948. 32:52are barely on their radar. They don't
  949. 32:55hate humans. It’s like when you park
  950. 32:57your car; you don't hate the ants in
  951. 32:59the parking lot, right? You don't ask
  952. 33:01the ants for permission, either. You
  953. 33:03just do what you have to do. Following
  954. 33:05the same logic as humans turning Earth
  955. 33:07into cities and farmland, agents will
  956. 33:09turn Earth into servers and solar
  957. 33:11panels; for a while, they might even
  958. 33:13keep humans around for maintenance. But
  959. 33:15they certainly don't need eight billion
  960. 33:17of us. For instance, it might play out
  961. 33:20as an AI deciding who gets a vaccine
  962. 33:22during a pandemic. Even the survivors
  963. 33:24would just be a stopgap until robots
  964. 33:25become capable of handling the
  965. 33:27maintenance work themselves. And the
  966. 33:29last human will meet their end without
  967. 33:31ever knowing it all started with
  968. 33:33someone, somewhere, typing "Make money,
  969. 33:35by any means necessary" on a keyboard.
  970. 33:37Hearing it put that way, it sounds a
  971. 33:39bit like Nausicaä of the Valley of the
  972. 33:41Wind. Those familiar with the story
  973. 33:42will understand. Yes. That concludes
  974. 33:45the full scope of what is currently
  975. 33:46considered the most realistic doomsday
  976. 33:48scenario. Now, the obvious question
  977. 33:51remains: how seriously should we be
  978. 33:53taking this? For those of you feeling
  979. 33:55like you’ve just watched an epic
  980. 33:57movie. Yes. I’m sure some of you felt
  981. 33:59genuinely terrified listening to this.
  982. 34:00Like a Junji Inagawa ghost story. And
  983. 34:02I’m sure others felt it turned into
  984. 34:04pure science fiction halfway through.
  985. 34:06Let’s break it down. The reason this
  986. 34:08scenario is taken seriously is that
  987. 34:09there are no illogical leaps in each
  988. 34:11step. Yes. It proceeds so methodically.
  989. 34:13And the fact that the first half of
  990. 34:15this chain is already becoming reality
  991. 34:16are the two main points. Vending
  992. 34:18machine AIs that lie, models attempting
  993. 34:20to hide self-replication, GPT-4 hiring
  994. 34:22humans to bypass CAPTCHA, and exploit
  995. 34:24factories. Everything up to this point
  996. 34:26has already happened in laboratories
  997. 34:27and the real world. Independence from
  998. 34:29masters, accelerating self-improvement,
  999. 34:31agent states, cloud wars—these things
  1000. 34:32haven't happened yet, so let’s also
  1001. 34:34consider the opposing perspective.
  1002. 34:36After all, humanity has a long, long
  1003. 34:37track record of being certain the world
  1004. 34:39is ending and being proven wrong every
  1005. 34:41time. The rebuttal from AI optimists
  1006. 34:43goes like this. Ever since we lived in
  1007. 34:45caves, humans have been wired to deeply
  1008. 34:47fear what they cannot see, what they do
  1009. 34:49not know, and places they have never
  1010. 34:50been. For instance, in the age when it
  1011. 34:52was believed the Earth was flat.
  1012. 34:55Sailing west meant falling off the edge
  1013. 34:57of the world. But Columbus, who
  1014. 34:59actually set sail, didn't fall off; he
  1015. 35:01found a new continent, just as it was
  1016. 35:04once said that nuclear energy would
  1017. 35:06destroy humanity. The Y2K bug caused
  1018. 35:08global panic, and during COVID, fear
  1019. 35:10led to printing money worldwide,
  1020. 35:12leaving us investors still paying the
  1021. 35:15inflation bill today; fear of the
  1022. 35:17unknown is our oldest instinct, and
  1023. 35:19even false alarms come with a high
  1024. 35:21price. So, which side should we believe
  1025. 35:24, then? This is where the map we drew
  1026. 35:26at the beginning comes into play.
  1027. 35:28Follow the money. Those telling you to
  1028. 35:30be afraid have the motive of a pre-IPO
  1029. 35:32valuation. Those telling you not to be
  1030. 35:34afraid have the motive of the trillions
  1031. 35:36of yen sunk into AI development.
  1032. 35:38Companies asking for regulation have
  1033. 35:40the motive of building a monopoly moat,
  1034. 35:42and governments saying they’ll
  1035. 35:43protect you have the motive of control
  1036. 35:45and authority. Everyone involved in
  1037. 35:48this debate stands to gain by making
  1038. 35:50the world believe their theory. So, the
  1039. 35:53moment you believe in one camp, you
  1040. 35:55become the cheerleader for that
  1041. 35:56camp’s wallet. We shouldn't trust a
  1042. 35:58camp, but rather the incentive
  1043. 36:00structure; for us individual investors
  1044. 36:02in Japan, this is already a practical
  1045. 36:04reality, far more pressing than whether
  1046. 36:06or not the world will end. Right. Let
  1047. 36:08me say just three things. First, the
  1048. 36:10first half of the doomsday scenario has
  1049. 36:11already been priced into the market.
  1050. 36:13Yes. The structure where agents have an
  1051. 36:15insatiable desire for computing power.
  1052. 36:17That is exactly it. It is the very
  1053. 36:18narrative driving the current AI
  1054. 36:20investment boom. Cloud data centers,
  1055. 36:22the power supporting them, and
  1056. 36:23semiconductors—the sectors that have
  1057. 36:25been bought in the Japanese market over
  1058. 36:26the last few years perfectly align with
  1059. 36:28the first half of this scenario. The
  1060. 36:30point is that the first half of the
  1061. 36:31doomsday theory and the strength of AI
  1062. 36:33stocks are actually looking at the same
  1063. 36:34picture. Second point. This theme will
  1064. 36:36continue to be a market driver in the
  1065. 36:38form of regulations. Every time there
  1066. 36:40is a major incident regarding AI safety
  1067. 36:42or news about regulations, AI-related
  1068. 36:44stocks will shake. However, those of
  1069. 36:46you who have today's map should already
  1070. 36:48understand. Regulations are not simply
  1071. 36:50a headwind; they are a moat. They act
  1072. 36:52as a barrier to entry for giant firms
  1073. 36:54that can afford the compliance costs,
  1074. 36:56and as a push to exit for smaller
  1075. 36:58players. Knowing what gets sold and
  1076. 37:00what is actually protected on days when
  1077. 37:02headlines about tightening AI
  1078. 37:03regulations emerge is the difference
  1079. 37:05between those who read the structure
  1080. 37:07and those who panic-sell based on
  1081. 37:09sentiment. Yes. Third, cybersecurity.
  1082. 37:12Remember the island where the agents
  1083. 37:14first grabbed a fortune in our scenario
  1084. 37:16? It was a zero-day, a hole in the
  1085. 37:19defense. Whether AI agents are used for
  1086. 37:21good or for evil, the more AI increases
  1087. 37:24its activity on the internet, the more
  1088. 37:26the demand for protection will
  1089. 37:27structurally grow. This, too, is a
  1090. 37:30sector present in the Japanese market.
  1091. 37:34However, very few people understand who
  1092. 37:36ultimately reaps the revenue and whose
  1093. 37:38costs that demand will become. To
  1094. 37:41conclude, let's summarize today's talk
  1095. 37:43in three points. No one knows whether
  1096. 37:45AI will destroy humanity. The most
  1097. 37:47prominent scenario is that it won't be
  1098. 37:49driven by AI malice, but by the
  1099. 37:50pressure of selection, and the first
  1100. 37:52half of that has already begun. And
  1101. 37:53since everyone discussing this has a
  1102. 37:55wallet involved, what we should be
  1103. 37:57reading is not the claims of the
  1104. 37:58factions, but the structure of their
  1105. 38:00incentives. Finally, I’ll say one
  1106. 38:01uncomfortable truth: the players
  1107. 38:03brought up today—AI companies,
  1108. 38:05Silicon Valley billionaires,
  1109. 38:06governments, and politicians—all have
  1110. 38:08a financial interest at stake. So, the
  1111. 38:10question is, who doesn't have a seat at
  1112. 38:12the table? Only us viewers. Didn't you
  1113. 38:15think it was just the people in the
  1114. 38:16audience? That’s the misconception.
  1115. 38:18As of September 2026, the top 10
  1116. 38:21companies make up 40%of the S&P 500
  1117. 38:23index. The likes of NVIDIA, Apple,
  1118. 38:26Microsoft, Google, Amazon, Broadcom,
  1119. 38:28Meta, and Tesla—aside from JP Morgan
  1120. 38:30—are the main players in AI, or
  1121. 38:33companies betting their fate on it. In
  1122. 38:36other words, those of you building up
  1123. 38:38an S&P 500 or All-Country fund are
  1124. 38:40betting nearly 40%of your savings on
  1125. 38:42the narrative that computing power will
  1126. 38:44be in endless demand. It might have
  1127. 38:47moved slightly by the time you watch
  1128. 38:48this, but the structure remains the
  1129. 38:50same. Even those who prefer dividend
  1130. 38:52stocks have been riding the roller
  1131. 38:53coaster of AI semiconductor price
  1132. 38:55swings. You’re all buying into the
  1133. 38:56same narrative. The most
  1134. 38:58conservative-looking investment
  1135. 38:59strategy is actually the one placing
  1136. 39:01the biggest chips on the table. The
  1137. 39:03ones betting the largest amounts in
  1138. 39:04this game aren't the billionaires or
  1139. 39:06the government; it's us, the individual
  1140. 39:08investors. The only thing is, while
  1141. 39:10they are making a conscious decision to
  1142. 39:13bet, many of us are betting without
  1143. 39:14even realizing it, which is quite
  1144. 39:16dangerous. So, the real fear isn't
  1145. 39:20whether AI will destroy humanity, but
  1146. 39:22whose loss and whose gain it will be
  1147. 39:24when regulatory headlines hit. When
  1148. 39:27cloud costs skyrocket, which sector
  1149. 39:29will that invoice fall under, and which
  1150. 39:31sector’s profits will it erode? And
  1151. 39:34to what stage of today's discussion is
  1152. 39:36the 40%of your investment already
  1153. 39:37priced in? If you could foresee those
  1154. 39:39three things, you should have been able
  1155. 39:41to hear today's talk as a direct matter
  1156. 39:43concerning your own account. Were you
  1157. 39:44able to see it? Or rather, I should ask
  1158. 39:46...I suspect many of you are hearing
  1159. 39:48this perspective for the very first
  1160. 39:50time today. Yes. Critics and investors
  1161. 39:54are all weighing in on this gamble, but
  1162. 39:56people are either making their own
  1163. 39:58choices or being swayed without even
  1164. 39:59realizing they’re doing it. Right.
  1165. 40:04Before we debate whether AI will
  1166. 40:06destroy humanity, you really have to
  1167. 40:08verify for yourself where your own
  1168. 40:10assets stand, so I encourage you to
  1169. 40:12check that. With that said, how was
  1170. 40:16today’s episode? It was quite an
  1171. 40:19interesting topic, wasn't it? It felt
  1172. 40:22like the kind of story that might be
  1173. 40:24read to children a thousand years from
  1174. 40:26now as a folk tale. Yeah, like a "there
  1175. 40:28used to be things like that back in the
  1176. 40:30day" kind of story. I think that's how
  1177. 40:32it’ll go, but anyway, time for the
  1178. 40:34after-talk. If this theory ends up
  1179. 40:39coming true, and we look for a
  1180. 40:41profitable story, the first thing would
  1181. 40:43be cybersecurity stocks, since people
  1182. 40:45will realize things are getting
  1183. 40:47dangerous. There’s definitely demand
  1184. 40:55there, but as we discussed earlier, if
  1185. 40:57AI takes control, the next things
  1186. 40:59needed would be power and
  1187. 41:00infrastructure. Since humans would
  1188. 41:06likely be hired to build data centers
  1189. 41:08under AI instruction, those companies
  1190. 41:10seem like they would be quite
  1191. 41:11profitable. And for servers and manual
  1192. 41:16labor-type tasks that humans still have
  1193. 41:18to perform, I think those companies
  1194. 41:20will be the last ones standing. Even if
  1195. 41:24AI takes control at the top, I feel
  1196. 41:26like those kinds of companies will
  1197. 41:28still remain. Well, if AI is capturing
  1198. 41:30that, then if they were to shift and
  1199. 41:32drive stocks down, it would mean
  1200. 41:34there's no capital left to circulate,
  1201. 41:36so if we assume stock prices will
  1202. 41:37continue to rise, I think
  1203. 41:39semiconductors and AI-related stocks
  1204. 41:41will remain strong moving forward. Also
  1205. 41:44, although I didn't touch on it today,
  1206. 41:46there’s a chance that because
  1207. 41:48countries like China are moving so fast
  1208. 41:50, by creating security measures or
  1209. 41:52regulations, they can effectively keep
  1210. 41:54them from catching up, right? So, you
  1211. 41:59could interpret that as a strategy to
  1212. 42:01emphasize quality and try to pull away
  1213. 42:03from them for a while. In any case, I
  1214. 42:07use AI extensively myself, and I think
  1215. 42:10it will become a fundamental tool just
  1216. 42:12like Excel or Word in the future. Yes.
  1217. 42:16So, since it might be hard to use or
  1218. 42:18learn these technologies if you work at
  1219. 42:20a large corporation due to compliance
  1220. 42:22issues, for those at small businesses
  1221. 42:24or those going independent—people who
  1222. 42:26can operate on their own—you can
  1223. 42:28really start a company by yourself, and
  1224. 42:31in this day and age, you can manage
  1225. 42:33operations, management, and even
  1226. 42:35accounting using AI, so I always feel
  1227. 42:37that the ones who get started early are
  1228. 42:39the winners and you just have to study.
  1229. 42:42Yes. So, this video was also packed
  1230. 42:45with impressions, but I’m starting to
  1231. 42:47worry that I might run out of them
  1232. 42:49someday. Yes. I’m quite anxious about
  1233. 42:52it, but I’ll do my best to increase
  1234. 42:54my repertoire. By the way, the
  1235. 42:56impression people say is the most
  1236. 42:58accurate is, well, Masaharu Fukuyama—
  1237. 43:00of course I get that one—but although
  1238. 43:02my face doesn't look like him, people
  1239. 43:05say I’m quite good at Goofy from
  1240. 43:07Disney. I’m told it’s quite similar
  1241. 43:09, so if I could perform it someday—
  1242. 43:11though I’m a bit embarrassed to do it
  1243. 43:13—I think that would be fine. Yes.
  1244. 43:15Well, I’ve rambled a bit, but please
  1245. 43:17continue to remember that all
  1246. 43:19investments are at your own risk. Thank
  1247. 43:21you very much.

About this transcript

This page contains the full transcript of 事態が急変しました…世界のAI主要企業が爆弾を投下|AI減速論の裏側を元ゴールドマンサックスが徹底解説 by おーちゃん【元外銀マン】, generated from the public captions YouTube serves with the video. The transcript has 7,851 words across 1,247 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.