How AI Companies Created a Fake Arms Race — Transcript

Full transcript

1. 0:00It's happening again. In his farewell
2. 0:02address, President Eisenhower warned us
3. 0:05of the rise of the military-industrial
4. 0:07complex. In Biden's farewell address, he
5. 0:10warned us of the tech industrial
6. 0:12complex. And something super weird
7. 0:14happened recently. Poll after poll shows
8. 0:17the vast majority of Americans agree AI
9. 0:19needs regulation. I mean, currently AI
10. 0:22companies are regulated less than a taco
11. 0:24cart, which is just insane. And yet,
12. 0:28despite this, bizarrely, big tech almost
13. 0:31successfully lobbyed Congress to ban
14. 0:34states from regulating AI entirely for
15. 0:3710 years. 10 years. So, what on earth
16. 0:41happened? How did they almost pull off
17. 0:44something so wildly unpopular? Well, the
18. 0:46answer may surprise you. First of all,
19. 0:48they're spending a fortune in dark money
20. 0:51lobbying, but that's obvious. The AI
21. 0:53industry has more lobbyists in
22. 0:55Washington and Brussels than the fossil
23. 0:57fuel industry and the tobacco industry
24. 1:00combined. Zuckerberg alone is
25. 1:02outspending all of civil society 10 to1.
26. 1:06And what narrative are they pushing with
27. 1:07all that money? Fear of China, an AI
28. 1:09arms race. But why? Well, it's an old
29. 1:13trick to manipulate public opinion. Fear
30. 1:15cells. Throughout the Cold War, the
31. 1:17military-industrial complex spent a
32. 1:19fortune pushing the false narrative that
33. 1:21the Soviets were far more advanced than
34. 1:24they actually were, which ensured the
35. 1:25money from Congress kept flowing. It was
36. 1:27called the missile and the bomber gap.
37. 1:30They lied and lied and lied again to get
38. 1:33bigger and bigger defense contracts.
39. 1:37Now, obviously, there is some amount of
40. 1:39competition between the US and China,
41. 1:41but big tech is lying and exaggerating
42. 1:43like crazy to terrify Congress into
43. 1:46giving them whatever they want. And what
44. 1:48they want is fat government contracts
45. 1:50and zero democratic oversight. Now, it
46. 1:53seems like not a day goes by without
47. 1:55another big AI company announcing a
48. 1:57giant contract with the Department of
49. 1:59Defense. But, are we actually in a race?
50. 2:02Let's go back in time. In January 2025,
51. 2:05China challenged years of American AI
52. 2:08dominance with the release of DeepSeek
53. 2:10R1, wiping more than a trillion dollars
54. 2:12off the American stock market.
55. 2:14Billionaire Mark Andre was everywhere,
56. 2:16claiming that China was now ahead of the
57. 2:18US in AI. Well, are they? According to
58. 2:22the AI researchers behind AI 2027, there
59. 2:25are three main pillars that would decide
60. 2:26a US China AI arms race.
61. 2:30China, previously thought to be years
62. 2:33behind, had potentially closed the gap
63. 2:35to mere months. So, if China supposedly
64. 2:37had their foot on the accelerator, why
65. 2:39wouldn't they completely catch up or
66. 2:41even surpass the US? China has amazing
67. 2:44AI talent, a big infrastructure
68. 2:46advantage, and top-notch companies. The
69. 2:49answer is that the country that holds
70. 2:50the real power over who wins the AI race
71. 2:53isn't China or the US.
72. 2:57It's Taiwan. The most valuable company
73. 2:59in the world right now isn't Microsoft,
74. 3:02Google or Apple. It's Nvidia because
75. 3:04they are supplying the GPUs or AI chips
76. 3:08fueling the entire AI race. According to
77. 3:10the researchers, the whole AI race comes
78. 3:13down to one thing, computing power, also
79. 3:15known as compute. AI companies are
80. 3:17spending as much as they possibly can on
81. 3:19computing power. Hundreds of billions of
82. 3:22dollars. Because the bottleneck to
83. 3:23scaling increasingly smarter AI systems
84. 3:26isn't money. It's cuttingedge chips.
85. 3:29But here's the catch. Nvidia doesn't
86. 3:31actually manufacture those chips. That
87. 3:33job falls to Taiwan Semiconductor
88. 3:35Manufacturing Company or TSMC. They've
89. 3:38been working together since 1998, which
90. 3:40shows just how long the two companies
91. 3:42have relied on each other. TSMC depends
92. 3:44on equipment produced by America and its
93. 3:47allies, putting it within the US sphere
94. 3:49of influence. Taiwan and China have been
95. 3:51geopolitical adversaries ever since the
96. 3:53Chinese civil war between the PRC and
97. 3:56the ROC ended in 1949 with the ROC
98. 3:59retreating from mainland China to the
99. 4:01island of Taiwan. So since 2022, the US
100. 4:04has banned companies like Nvidia from
101. 4:07selling top tier chips to China. This
102. 4:09left China with almost no good options
103. 4:11to obtain the best chips. Crazily enough
104. 4:14roughly 20% of all of China's chips come
105. 4:16in from smuggling. You could get paid $1
106. 4:19million per suitcase of GPUs you bring
107. 4:22to the country. But the rabbit hole
108. 4:23keeps going. The same band has also
109. 4:26torpedoed China's own chipmaking
110. 4:28industry. Before 2022, all the best
111. 4:31Chinese-made chips were made using
112. 4:33special silicon chips called wafers made
113. 4:35by TSMC. Without access to those,
114. 4:38Chinese chip makers have scrambled to
115. 4:40come up with replacements. China's best
116. 4:42domestic chip manufacturer, SMIC, lags a
117. 4:45few years behind TSMC. This leads to a
118. 4:48massive difference. The researchers
119. 4:50estimated that the United States had 75%
120. 4:53of the world's supply of the advanced
121. 4:55chips needed for AI development with
122. 4:58China having 15% and the rest of the
123. 5:00world combined having 10%. Google alone
124. 5:03has probably 15% of all the world's
125. 5:05compute while DeepS has maybe 1%. This
126. 5:09is a huge deal. The US has five times
127. 5:11the compute as China. So claims that
128. 5:14America is losing the AI arms race feels
129. 5:16strikingly similar to what US military
130. 5:18contractors were saying about the Soviet
131. 5:20Union during the Cold War to get larger
132. 5:23defense contracts. In fact, while tech
133. 5:25billionaires like Mark Andre claimed
134. 5:27that Deepseek was a Sputnik moment,
135. 5:29showing that China was actually ahead of
136. 5:31the US, Deepseek's own CEO accidentally
137. 5:34revealed last year just how much the
138. 5:36chip ban had hurt them. mid last year,
139. 5:38the CEO of Deep Seek, they were working
140. 5:40in in total obscurity. And what he says
141. 5:43is he's like, "Yeah, so honestly, like
142. 5:46we're really excited and doing this AGI
143. 5:49push or whatever." And like honestly,
144. 5:50like money's not the problem for us.
145. 5:52Talent's not the problem for us, but
146. 5:54like access to compute, like these
147. 5:57export controls, man, that's a real
148. 6:00problem for us. Oh boy.
149. 6:02>> Nothing can stop us on the path to AGI
150. 6:05except access to compute. He who
151. 6:07controls the chips controls the future
152. 6:09of AI. So when it comes to the first
153. 6:11pillar of AI development, compute, the
154. 6:14US is winning by far. But there are two
155. 6:16more major potential bottlenecks that
156. 6:18could actually tell us if we're in an AI
157. 6:20arms race. Because it doesn't just take
158. 6:23advanced chips to train smarter AI
159. 6:25models. You also need energy. China has
160. 6:28surpassed the US in energy, but that
161. 6:31power means nothing if China doesn't
162. 6:33have the AI chips they need. And the US
163. 6:35has enough potential energy sources that
164. 6:37it could unlock that energy won't become
165. 6:40a bottleneck for AI. Besides the new
166. 6:42solar installations and revived nuclear
167. 6:44reactors, for the last decade, US gas
168. 6:47power plants have only been running at
169. 6:49around half of their total capacity,
170. 6:51meaning that the US can grow energy
171. 6:53production even without building new
172. 6:55plants. Energy, while critical, won't be
173. 6:57the bottleneck that slows the US down
174. 6:59enough for China to catch up. Okay, so
175. 7:02what about people? China has four times
176. 7:05as many people as the US and twice as
177. 7:07many STEM PhDs. But the US gets to draw
178. 7:09in the sharpest minds from the rest of
179. 7:11the world. Take OpenAI founder Ilia
180. 7:13Sutsk who was born in Russia or Google
181. 7:16DeepMind founder Demis who was born in
182. 7:18London. The talent battle isn't USA
183. 7:20versus China. It's China versus the rest
184. 7:23of the world. But even if China had more
185. 7:26talent, it wouldn't matter because the
186. 7:28specific goals of AI companies is to
187. 7:30create an intelligence explosion
188. 7:32scenario by having AIs themselves do AI
189. 7:35research. This would increase the total
190. 7:37amount of talent on both sides by 10 or
191. 7:40100 times. NVIDIA's CEO thinks that the
192. 7:43company will have 100 million AI workers
193. 7:46in the next few years. AIS will recruit
194. 7:48other AIs to solve problems. AIS will be
195. 7:50in, you know, Slack channels with each
196. 7:52other. As AI workers increasingly
197. 7:54contribute to research and to
198. 7:55algorithmic advances, talent ceases to
199. 7:58become a bottleneck. The only remaining
200. 8:01bottleneck is compute. And there the US
201. 8:03is already crushing China with the US
202. 8:05having five times as much compute. And
203. 8:07the US would pull even further ahead
204. 8:09after closing the few remaining
205. 8:10loopholes that China has been using to
206. 8:12circumvent the chip ban. So if the US
207. 8:15already controls the chips, the talent,
208. 8:17and has the energy it needs, why the
209. 8:20panic? Why are senators pounding the
210. 8:22table, warning that we can't afford to
211. 8:24lose the AI race? It's simple. Big tech
212. 8:26is pouring millions into lobbying
213. 8:28Congress to spread the narrative that
214. 8:30China is sprinting ahead and that even
215. 8:32the tiniest amount of regulation would
216. 8:35hand China the race. But in reality,
217. 8:38China's AI industry is the most
218. 8:40regulated in the entire world. In fact,
219. 8:43even a Xi Jinping himself says he
220. 8:45doesn't want an AI arms race, saying
221. 8:48that China should abandon unhited growth
222. 8:50that comes at the cost of sacrificing
223. 8:52safety. Let that sink in. So, if the
224. 8:55idea of an AI race is so unpopular, both
225. 8:58in the Chinese leadership and among the
226. 9:00American public, who are the people
227. 9:02arguing that we need a race to beat
228. 9:04China? Andre Horowitz or A16Z is a
229. 9:07venture capital firm with 42 billion in
230. 9:10assets and is deeply invested in AI.
231. 9:13Shockingly, they don't want regulation
232. 9:15on AI. And what better way to sell
233. 9:17American lawmakers on the need to go
234. 9:19fast than convincing everyone that the
235. 9:22US is in a new global arms race. Andre
236. 9:25Horowitz is a backer of the American
237. 9:27Innovators Network, which has plans to
238. 9:29spend hundreds of thousands of dollars
239. 9:31on opposing a New York State AI safety
240. 9:34bill. The opposition is spreading
241. 9:36outright lies like the bill would kill
242. 9:38all of New York's AI startups when in
243. 9:41reality, the regulation would only apply
244. 9:43to big tech companies spending over
245. 9:45hund00 million on training frontier
246. 9:47models. This is the classic playbook.
247. 9:49But who else is trying to sell US
248. 9:51lawmakers on the idea of an AI brace?
249. 9:54The tech company CEOs themselves. The
250. 9:56biggest fear-mongerer in the space is
251. 9:58Zuckerberg. The playbook is simple. Big
252. 10:01Tobacco used it to cast doubt that
253. 10:03smoking was actually bad for you. Big
254. 10:05oil uses the same playbook to cast doubt
255. 10:07on climate change. There's a fantastic
256. 10:09book on the topic called Virgins of
257. 10:10Doubt that I highly recommend except in
258. 10:13AI it's more like they are merchants of
259. 10:15fear.
260. 10:17For example, Zuckerberg has pumped at
261. 10:19least $40 million that we know of into a
262. 10:22front group called American Edge.
263. 10:24American Edge spent millions of dollars
264. 10:27blasting out Facebook ads to millions of
265. 10:29people, declaring that banning states
266. 10:31from regulating AI will protect American
267. 10:33innovation and jobs. But these AI
268. 10:36companies explicit goal is to use AI to
269. 10:38replace all human labor, which yeah,
270. 10:41might have an effect on jobs. Yeah,
271. 10:43okay. Big shocker that this is what
272. 10:45Zuckerberg is up to, but it's not just
273. 10:47him.
274. 10:48>> We must continue to run faster. Uh, this
275. 10:50is a race.
276. 10:51>> That woman is Lisa Sue, AMD CEO,
277. 10:55Nvidia's biggest competitor at selling
278. 10:57AI chips. Right next to her is OpenAI
279. 10:59CEO Sam Alman, who wants Congress to
280. 11:02ensure that AI has very loose and
281. 11:04permissive regulations. Sam Alman claims
282. 11:06to support regulating AI because of how
283. 11:09dangerous future models might be. But
284. 11:11unsurprisingly, like most industry
285. 11:13monopolists, whenever journalists or
286. 11:15senators ask him what specific
287. 11:17regulation he would like, he dodges the
288. 11:20question and an extremely light touch AI
289. 11:22safety bill was passed in both chambers
290. 11:24in California with an overwhelming
291. 11:26majority of support from civil society.
292. 11:28But vetoed by Governor Gavin Newsome,
293. 11:31big AI companies heavily lobbied against
294. 11:33the bill. And who does Nuome have deep
295. 11:35donor ties to? You guessed it. And the
296. 11:38stakes here aren't just escalation
297. 11:40toward World War II, but potentially
298. 11:42much worse. These companies and VCs have
299. 11:45the explicit goal of creating artificial
300. 11:48super intelligence, an AI system that
301. 11:50could outperform humans at everything.
302. 11:53But according to the Godfathers of AI
303. 11:55and the company's CEOs themselves, if we
304. 11:57create this super intelligence as
305. 11:59quickly as possible without spending the
306. 12:01time to make sure it's safe, the results
307. 12:03could be literal human extinction. And
308. 12:06the bad case, and I think this is like
309. 12:07important to say, is like lights out for
310. 12:09all of us. And the risk of this is
311. 12:11frighteningly high. Jeffrey Hinton,
312. 12:13who's a Nobel laureate, thinks that the
313. 12:15odds of humanity going extinct is a coin
314. 12:18flip.
315. 12:20We have to face the possibility
316. 12:22that unless we do something
317. 12:25soon,
318. 12:27we're near the end.
319. 12:29Even the CEO of Google himself said that
320. 12:32the odds of extinction from AI are
321. 12:34actually pretty high. Sit with that for
322. 12:37a moment. He told Lex Friedman he's an
323. 12:39optimist because he thinks humanity will
324. 12:42rally to prevent catastrophe. In other
325. 12:44words, that humanity will rally to stop
326. 12:47him from building the thing he is
327. 12:49building. This timeline is wild. Okay,
328. 12:51but there may be a solution to all this
329. 12:53madness. The head of DeepMind believes
330. 12:55that this is a problem that nations
331. 12:56can't solve by competing. but by
332. 12:59cooperating. In the same way that we
333. 13:01have CERN, he believes that we need a
334. 13:03CERN for AGI. Right now, the US is by
335. 13:07far the leader in AI and could use that
336. 13:09leverage over China to negotiate a
337. 13:11bilateral treaty regulating AI
338. 13:13development. And instead of creating
339. 13:14fake AI arms races, humanity could
340. 13:17actually work together.
341. 13:20Hey, I'm Drew. If you want to see a
342. 13:22concrete, deeply researched scenario of
343. 13:24how actually AI could take over written
344. 13:27by AI researchers, watch this video
345. 13:29next. Thanks for watching. This video
346. 13:31took a long time to make and I
347. 13:33appreciate all your