YouTube2Text

Elon Musk: War, AI, Aliens, Politics, Physics, Video Games, and Humanity | Lex Fridman Podcast #400 — Transcript

by Lex Fridman · 20,825 words · 2,919 segments · language en · Watch on YouTube

Full transcript

  1. 0:00- The following is a conversation with Elon Musk,
  2. 0:03his fourth time on this, the Lex Fridman Podcast.
  3. 0:07(Elon whistling)
  4. 0:20I thought you were gonna finish it.
  5. 0:22(Elon whistling)
  6. 0:25It's one of the greatest themes in all film history.
  7. 0:28(whistling continues)
  8. 0:31- Yeah, that's great.
  9. 0:33- So I was just thinking about the Roman Empire.
  10. 0:36As one does. (Elon laughing)
  11. 0:38- There's that whole meme where old guys
  12. 0:42are thinking about the Roman Empire at least once a day.
  13. 0:44- And half the population's confused
  14. 0:46whether it's true or not.
  15. 0:47But more seriously thinking about the wars
  16. 0:51going on in the world today.
  17. 0:52And as you know, war and military conquest
  18. 0:56has been a big part of Roman society and culture.
  19. 1:00And it, I think, has been a big part of most empires
  20. 1:04and dynasties throughout human history, so.
  21. 1:06- Yeah, they usually came as a result of conquest.
  22. 1:10I mean, there's some, like the Hungarian Empire
  23. 1:12where there was just a lot of sort of clever marriages.
  24. 1:16- But fundamentally there's an engine of conquest.
  25. 1:19They celebrate excellence in warfare.
  26. 1:23Many of the leaders were excellent generals.
  27. 1:26- [Elon] Yeah. - That kind of thing.
  28. 1:28So a big picture question, Grok approved,
  29. 1:31I asked if this is a good question to ask.
  30. 1:33- [Elon] Get tested Grok approved.
  31. 1:34- Yeah.
  32. 1:36At least on fun mode.
  33. 1:38(Elon laughing)
  34. 1:41To what degree do you think war is part of human nature
  35. 1:44versus a consequence of
  36. 1:48how human societies are structured?
  37. 1:50I ask this as you have somehow controversially
  38. 1:55been a proponent of peace.
  39. 1:57- I'm generally a proponent of peace.
  40. 1:58I mean, ignorance is perhaps,
  41. 2:00in my view, the real enemy to be countered.
  42. 2:04That's the real hard part.
  43. 2:05Not, you know, fighting other humans.
  44. 2:09But all creatures fight.
  45. 2:10I mean, the jungle is a,
  46. 2:12you look at the, people think of this nature
  47. 2:15as perhaps some sort of peaceful thing.
  48. 2:17But in fact it is not.
  49. 2:18There's some quite funny Werner Herzog thing
  50. 2:22where he is like in the jungle,
  51. 2:24like saying that it's like basically just murder
  52. 2:27and death in every direction.
  53. 2:29I mean, the plants, animals in the jungle
  54. 2:31are constantly trying to kill each other,
  55. 2:33every single day, every minute.
  56. 2:36So it's not like, you know, we're unusual in that respect.
  57. 2:40- Well, there's a relevant question here.
  58. 2:42Whether with greater intelligence comes greater control
  59. 2:46over these base instincts for violence.
  60. 2:49- Yes.
  61. 2:50We have much more vulnerability to control our
  62. 2:55limbic instinct for violence than say a chimpanzee.
  63. 2:58And in fact, if one looks and say,
  64. 3:01chimpanzee society, it is not friendly.
  65. 3:03I mean, the Bonobos are an exception,
  66. 3:06but chimpanzee society is filled with violence.
  67. 3:11And it's quite, quite horrific, frankly,
  68. 3:15that that's our limbic system in action.
  69. 3:18Like, you don't wanna be on the wrong side of a chimpanzee,
  70. 3:20they'll eat your face off and tear and nuts off.
  71. 3:22- Yeah, basically, there's no limits or ethics
  72. 3:24or there almost had just war.
  73. 3:26There's no just war in chimpanzee societies,
  74. 3:29is war and dominance by any means necessary.
  75. 3:33- Yeah, chimpanzee society is a pro,
  76. 3:34like a permanent version of human society.
  77. 3:37They're not like peace loving basically at all.
  78. 3:41There's extreme violence.
  79. 3:45And then once in a while,
  80. 3:46somebody who's watched too many Disney movies
  81. 3:48decides to raise a chimpanzee as a pet,
  82. 3:51and then that eats their face or rip their nuts off
  83. 3:54or chew their fingers off and that kind of thing.
  84. 3:56It's happened several times.
  85. 3:58- Ripping your nuts off is an interesting
  86. 3:59strategy for interaction.
  87. 4:02- It's happened to people. It's unfortunate.
  88. 4:04Like that's, I guess a one way to ensure
  89. 4:06that the other chimp doesn't like, you know,
  90. 4:09contribute to the gene pool.
  91. 4:10- Well, from a martial arts perspective
  92. 4:12is the fascinating strategy.
  93. 4:14- (laughing) The nut riper. (laughing)
  94. 4:18- I wonder which of the martial arts teaches that one.
  95. 4:21- I think it's safe to say if somebody's got your nuts
  96. 4:24in their hands and as the option ripping them off,
  97. 4:28you'll be amenable to whatever they want.
  98. 4:30- Yeah. (Elon laughing)
  99. 4:33- So like I said, somehow controversially,
  100. 4:35you've been an proponent of peace on Twitter, on X.
  101. 4:38- Yeah.
  102. 4:39- So let me ask you about the wars going on today
  103. 4:41and to see what the path to peace could be.
  104. 4:44How do you hope the current war in Israel
  105. 4:46and Gaza comes to an end?
  106. 4:48What path do you see
  107. 4:50that can minimize human suffering
  108. 4:52in the long term in that part of the world?
  109. 4:54- Well, I think that part of the world is definitely like,
  110. 4:58if you look up the, there is no easy answer
  111. 5:00in the dictionary.
  112. 5:01It'll be that, like the picture of the Middle East,
  113. 5:05in Israel especially.
  114. 5:06So there is no easy answer.
  115. 5:11What my, this is strictly my opinion of, you know,
  116. 5:15is that the goal
  117. 5:20of Hamas was to provoke an overreaction from Israel.
  118. 5:24They obviously did not expect to, you know,
  119. 5:27have a military victory, but they expect,
  120. 5:30they really wanted to commit the worst atrocities
  121. 5:34that they could in order to provoke
  122. 5:36the most aggressive response possible from Israel.
  123. 5:41And then leverage that aggressive response
  124. 5:45to rally Muslims worldwide
  125. 5:51for the course of Gaza and Palestine,
  126. 5:54which they have succeeded in doing.
  127. 5:58So the counterintuitive thing here,
  128. 6:02I think that the thing that I think should be done,
  129. 6:04even though it's very difficult,
  130. 6:06is that I would recommend that Israel
  131. 6:10engage in the most conspicuous acts of kindness possible.
  132. 6:14Every po- everything.
  133. 6:16That is the actual thing
  134. 6:17that we're taught the goal of Hamas.
  135. 6:19- So in some sense,
  136. 6:21the degree that makes sense in geopolitics
  137. 6:23turn the other cheek implemented.
  138. 6:26- It's not exactly turn the other cheek.
  139. 6:28Because I do think that there's, you know,
  140. 6:32I think it is appropriate for Israel
  141. 6:35to find the Hamas members
  142. 6:39and, you know, either kill 'em or incarcerate them.
  143. 6:45Like that's something that's something has to be done
  144. 6:46'cause they're just gonna keep coming otherwise.
  145. 6:51But in addition to that, they need to do whatever they can.
  146. 6:58There's some talk of establishing,
  147. 7:00for example, a mobile hospital.
  148. 7:02I'd recommend doing that.
  149. 7:05Just making sure that, you know,
  150. 7:08there's food, water, medical, necessities,
  151. 7:15and just be over the top about it
  152. 7:17and be very transparent.
  153. 7:20So you can't claim it's a trick.
  154. 7:23Like just put webcam on the thing, you know, or 24/7.
  155. 7:29- Deploy acts of kindness. - Yeah.
  156. 7:31Conspicuous acts of kindness that,
  157. 7:35with that are unequivocal.
  158. 7:37Meaning they can't be somehow,
  159. 7:40because Hamas will then,
  160. 7:41their response will be, oh, it's a trick.
  161. 7:43Therefore you have to counter how it's not a trick.
  162. 7:47- This ultimately fights the broader force
  163. 7:49of hatred in the region.
  164. 7:51- Yes.
  165. 7:52And I'm not sure who said it, it's an apocryphal saying,
  166. 7:55but an eye for an eye makes everyone blind.
  167. 7:58Now that neck of the woods,
  168. 7:59they really believe in the whole eye for an eye thing.
  169. 8:05But I mean, you really have,
  170. 8:08if you're not gonna just outright commit genocide,
  171. 8:12like against an entire people,
  172. 8:14which obviously would not be acceptable to really
  173. 8:16shouldn't be acceptable to anyone,
  174. 8:19then you're gonna leave basically a lot of people alive
  175. 8:22who subsequently, you know, hate Israel.
  176. 8:25So really the question is like,
  177. 8:27how, for every Hamas member that you kill,
  178. 8:30how many did you create?
  179. 8:33And if you create more than you killed,
  180. 8:35you've not succeeded.
  181. 8:36That's the, you know, the real situation there.
  182. 8:39And it's safe to say that if, you know,
  183. 8:47if you kill somebody's child in Gaza,
  184. 8:52you've made at least a few Hamas members
  185. 8:58who will die just to kill an Israeli.
  186. 9:03That's the situation.
  187. 9:05So, but I mean,
  188. 9:08this is one of the most contentious subjects
  189. 9:11one could possibly discuss.
  190. 9:12But I think if the goal ultimately
  191. 9:15is some sort of long-term peace,
  192. 9:17one has to be, look at this from the standpoint of,
  193. 9:21over time, are there more or fewer terrorists being created?
  194. 9:26- Let me just linger on war.
  195. 9:29- Yeah, war safe to say,
  196. 9:30wars always existed and always will exist.
  197. 9:33- [Lex] Always will exist,
  198. 9:34- Always has existed and always will exist.
  199. 9:37- I hope not. You think it always will?
  200. 9:38- There will always be war.
  201. 9:42This question of just how much war.
  202. 9:43And, you know, there's this,
  203. 9:47there's this sort of the scope and scale of war.
  204. 9:49But to imagine that there would not be any war
  205. 9:52in the future, I think would be a very unlikely outcome.
  206. 9:55- Yeah you talked about the Culture series.
  207. 9:57There's war even there.
  208. 9:58- Yes. It's a giant war.
  209. 10:00The first book starts off with a gigantic galactic war
  210. 10:05where trillions die, trillions.
  211. 10:07- But it still nevertheless protects these pockets
  212. 10:09of flourishing.
  213. 10:12Somehow you can have galactic war
  214. 10:15and still have pockets of flourishing.
  215. 10:18- Yeah.
  216. 10:18I mean it's, I guess if we are able to one day expand to,
  217. 10:23you know, the galaxy or whatever,
  218. 10:26there will be a galactic war at some point.
  219. 10:30- The scale, I mean the scale of war has been increasing,
  220. 10:33increasing, increasing.
  221. 10:34It's like a race between the scale of suffering
  222. 10:37and the scale of flourishing.
  223. 10:38- Yes.
  224. 10:41- A lot of people seem to be using this tragedy
  225. 10:43to beat the drums of war
  226. 10:45and feed the military industrial complex.
  227. 10:47Do you worry about this?
  228. 10:49The people who are rooting for escalation
  229. 10:54and how can it be stopped?
  230. 10:56- One of the things that did concerned me
  231. 10:57is that there are very few people alive today
  232. 10:59who actually viscerally understand the horrors of war,
  233. 11:04at least in the US.
  234. 11:05I mean, obviously there are people in on the front lines
  235. 11:07in Ukraine and Russia who understand
  236. 11:09just how terrible war is,
  237. 11:10but how many people in the west understand it?
  238. 11:15And my grandfather was in World War II.
  239. 11:17He was severely traumatized.
  240. 11:19He was there, I think for almost six years in the,
  241. 11:23in eastern North Africa and Italy.
  242. 11:26All his friends were killed in front of him.
  243. 11:29And he would've died too, except they randomly gave some,
  244. 11:34I guess, IQ test or something.
  245. 11:35And he scored very high.
  246. 11:39And he was not an officer, he was a,
  247. 11:40I think a corporal or a sergeant or something like that.
  248. 11:43Because he didn't finish high school
  249. 11:45because he had to drop outta high school,
  250. 11:46'cause his dad died and he had to work
  251. 11:49to support his siblings.
  252. 11:52So because he didn't graduate high school,
  253. 11:54he was not eligible for the officer corps.
  254. 11:57So, you know, he kind of got put into the cannon fodder
  255. 12:01category, basically.
  256. 12:04But then, this, randomly they gave him this test.
  257. 12:07He was transferred to British intelligence in London.
  258. 12:09That's where he met my grandmother.
  259. 12:13But he had PTSD next level, like next level.
  260. 12:17I mean, just didn't talk, just didn't talk.
  261. 12:21And if you tried talking to him,
  262. 12:22he'd just tell you to shut up.
  263. 12:24And he won a bunch of medals.
  264. 12:25Never bragged about it once, not even hinted, nothing.
  265. 12:30I like found out about it,
  266. 12:31'cause his military records were online.
  267. 12:33That's how I know.
  268. 12:36So he would say like, no way in hell
  269. 12:39do you wanna do that again.
  270. 12:41But how many people, now he obviously he died,
  271. 12:45you know, 20 years ago or longer, actually 30 years ago.
  272. 12:50How many people are alive that remember World War II?
  273. 12:53Not many.
  274. 12:54- And the same perhaps applies to the threat of nuclear war.
  275. 13:01- Yeah. I mean, there are enough nuclear bombs
  276. 13:02pointed at United States to make the revel,
  277. 13:06the radio active revel balance many times.
  278. 13:10- There's two major wars going on right now.
  279. 13:14So you talked about the threat of AGI quite a bit.
  280. 13:17But now as we sit here with the intensity
  281. 13:20of conflict going on, do you worry about nuclear war?
  282. 13:25- I think we shouldn't discount the possibility
  283. 13:27of nuclear war.
  284. 13:29It is a civilizational threat.
  285. 13:36Right now I could be wrong,
  286. 13:37but I think the current probability of nuclear war
  287. 13:41is quite low.
  288. 13:43But there are a lot of nukes pointed at us.
  289. 13:46So, and we have a lot of nukes pointed at other people.
  290. 13:50They're still there. Nobody's put their guns away.
  291. 13:55The missiles are still in the silos.
  292. 13:57- And the leaders don't seem to be the ones
  293. 14:01with the nukes talking to each other.
  294. 14:03- No.
  295. 14:04There are wars which are tragic
  296. 14:06and difficult on a local basis.
  297. 14:08And then there are wars which are civilization ending
  298. 14:11or has that potential.
  299. 14:12Obviously global thermonuclear warfare has high potential
  300. 14:15to end civilization, perhaps permanently.
  301. 14:19But certainly, you know, to severely wound
  302. 14:24and perhaps set back human progress by,
  303. 14:30you know, to the stone age or something, I don't know.
  304. 14:34Pretty bad.
  305. 14:39Probably scientists and engineers wanna be super popular
  306. 14:41after that as well.
  307. 14:43And like you got us into this mess.
  308. 14:46- So generally I think we obviously wanna prioritize
  309. 14:50civilizational risks over things that are painful
  310. 14:54and tragic on a local level, but not civilizational.
  311. 15:00- How do you hope the war in Ukraine comes to an end?
  312. 15:02And what's the path, once again
  313. 15:04to minimizing human suffering there?
  314. 15:08- Well, I think that what is likely to happen,
  315. 15:13which is really pretty much the way it is,
  316. 15:17is that something very close to the current lines will be
  317. 15:25how a ceasefire or truth happens.
  318. 15:30But, you know, you just have a situation right now
  319. 15:33where whoever goes on the offensive will suffer casualties
  320. 15:37at several times the rate of whoever's on the defense.
  321. 15:41'Cause you've got defense in depth,
  322. 15:43you've got minefields, trenches, anti-tank defenses.
  323. 15:48Nobody has air superiority.
  324. 15:50'cause the anti-aircraft missiles
  325. 15:53are really far better than the aircraft.
  326. 15:55Like, they're far more of them.
  327. 15:58And so neither side has air superiority.
  328. 16:02Tanks are basically death drafts, just slow moving
  329. 16:06and they're not immune to anti-tank weapons.
  330. 16:09So you really just have long range artillery
  331. 16:13and infantry ranges.
  332. 16:17It's World War I all over again with drones, you know,
  333. 16:22throwing old drones, some drones there.
  334. 16:25- Which makes the long range artillery
  335. 16:26just that much more accurate and better
  336. 16:28and so more efficient at murdering people on both sides.
  337. 16:33- Yeah.
  338. 16:34So it's whoever is, you don't wanna be trying
  339. 16:38to advance from either side,
  340. 16:39'cause the probability of dying is incredibly high.
  341. 16:43So in order to overcome
  342. 16:49defense in depth trenches
  343. 16:51and minefields, you really need a significant
  344. 16:54local superiority in numbers.
  345. 16:57Ideally combined alms where you do a fast attack
  346. 17:02with aircraft, a concentrated number of tanks
  347. 17:08and a lot of people.
  348. 17:10That's the only way you're gonna punch through the line.
  349. 17:12And then you're gonna punch through
  350. 17:13and then not have reinforcements just kick you
  351. 17:18right out again.
  352. 17:19I mean, I really recommend people read
  353. 17:24World War I warfare in detail.
  354. 17:28That's rough.
  355. 17:33I mean, the sheer number of people that died there
  356. 17:34was mind boggling.
  357. 17:37- And it's almost impossible to imagine the end of it
  358. 17:40that doesn't look like almost exactly like the beginning
  359. 17:43in terms of what land belongs to who and so on.
  360. 17:48But on the other side of a lot of human suffering death
  361. 17:54and destruction of infrastructure.
  362. 17:56- Yes.
  363. 17:56I mean, the thing that, the reason I,
  364. 17:59you know, proposed some sort of truce
  365. 18:03or peace a year ago
  366. 18:05was because I've predicted pretty much exactly
  367. 18:08what would happen, which is a lot of people dying
  368. 18:12for basically almost no changes in land.
  369. 18:16And this, the loss of the flower
  370. 18:19of Ukrainian and Russian youth.
  371. 18:21And we should have some sympathy for the Russian boys
  372. 18:24as well as the Ukrainian boys,
  373. 18:26'cause they, Russian boys didn't ask to be
  374. 18:28on their front line, they have to be.
  375. 18:30So there's a lot of sons not coming back to their parents,
  376. 18:37you know, and I think most of 'em don't really have,
  377. 18:43they don't hate the other side.
  378. 18:45You know, it's sort of like as this saying about, like,
  379. 18:48this saying comes from World War I, it's like young boys
  380. 18:52who don't know each other, killing each other
  381. 18:55on behalf of old men that do know each other.
  382. 18:59The hell's the point of that?
  383. 19:02- So Vladimir Zelensky said that he's not,
  384. 19:05or has said in the past,
  385. 19:06he's not interested in talking to Putin directly.
  386. 19:08Do you think he should?
  387. 19:09- Yeah. - Sit down man to man,
  388. 19:11lead a leader and negotiate peace?
  389. 19:14- Look, I think I would just recommend
  390. 19:16do not send the flower of Ukrainian youth
  391. 19:21to die in trenches.
  392. 19:25Whether he talks to Putin or not, just don't do that.
  393. 19:30Whoever goes on the offensive
  394. 19:32will lose massive numbers of people
  395. 19:37and history will not look kindly upon them.
  396. 19:42- You've spoken honestly about the possibility of war
  397. 19:44between US and China in the long term.
  398. 19:46If no diplomatic solution is found, for example,
  399. 19:50on the question of Taiwan and one China policy.
  400. 19:52- [Elon] Right.
  401. 19:53- How do we avoid the trajectory
  402. 19:54where these two superpowers clash?
  403. 19:58- Well, it's worth reading that book
  404. 19:59on the, difficult to pronounce, the Thucydides trap,
  405. 20:03I believe it's called.
  406. 20:04I love war history. I like inside out and backwards.
  407. 20:09There's hardly a battle I haven't read about.
  408. 20:13And trying to figure out like what really was the cause
  409. 20:16of victory in any particular case as opposed
  410. 20:17to what one side or another claim the reason.
  411. 20:21- Both the victory and what sparked the war?
  412. 20:24- Yeah, yeah, - [Lex] The whole thing.
  413. 20:26- Yeah, so that Athens and Sparta is a classic case.
  414. 20:30The thing about the Greek is
  415. 20:30they really wrote down a lot of stuff.
  416. 20:32They loved writing.
  417. 20:33You know, there are lots of interesting things
  418. 20:35that happened in many parts of the world,
  419. 20:36but people didn't write down. (laughing)
  420. 20:38So we don't know what happened,
  421. 20:40or they didn't really write with in detail.
  422. 20:44They just would say like, we had a battle and we won.
  423. 20:48And like, well what? Can you add a bit more?
  424. 20:51The Greeks, they really wrote a lot. (laughing)
  425. 20:54They were very articulate on, they just love writing.
  426. 20:57So, and we have a bunch of that writing as preserved.
  427. 20:59So we know what led up to the Peloponnesian War
  428. 21:02between the Spartan and Athenian alliance.
  429. 21:08And we know that they saw it coming.
  430. 21:16I mean, the Spartans didn't write,
  431. 21:17they also weren't very verbose by their nature,
  432. 21:19but they did write, but they weren't very verbose.
  433. 21:21They were terse.
  434. 21:23But the Athenians and the other Greeks wrote a line
  435. 21:26and they were like, and Sparta was really kinda
  436. 21:29like the leader of Greece.
  437. 21:33But Athens grew stronger
  438. 21:35and stronger with each passing year.
  439. 21:37And everyone's like, well that's inevitable
  440. 21:40that there's gonna be a clash between Athens and Sparta.
  441. 21:45Well, how do we avoid that?
  442. 21:46And they couldn't.
  443. 21:47They actually, they saw it coming
  444. 21:49and they still could not avoid it. (laughing)
  445. 21:53So, you know, at some point if there's, if one group,
  446. 21:59one civilization or country
  447. 22:00or whatever exceeds another,
  448. 22:06sort of, like, you know,
  449. 22:07the United States has been the biggest kid on the block for,
  450. 22:10since I think around 1890 from an economic standpoint.
  451. 22:14So the United States has been the economic,
  452. 22:18most powerful economic engine in the world,
  453. 22:22longer than anyone's been alive.
  454. 22:28And the foundation of war is economics.
  455. 22:31So now we have a situation in the case
  456. 22:36of China where the economy is likely to be two,
  457. 22:42perhaps three times larger than that of the US.
  458. 22:45So imagine you're the biggest kid on the block
  459. 22:48for as long as anyone can remember.
  460. 22:49And suddenly a kid comes along who's twice your size.
  461. 22:55- So we see it coming. - [Elon] Yeah.
  462. 22:57- How is it possible to stop?
  463. 22:59Is there some, let me throw something out there,
  464. 23:02just intermixing of cultures understanding.
  465. 23:06There does seem to be a giant cultural gap
  466. 23:08in understanding of each other.
  467. 23:09And you're an interesting case study
  468. 23:11because you are an American, obviously you've done
  469. 23:14a lot of incredible manufacture here in the United States,
  470. 23:18but you also work with China.
  471. 23:20- I've spent a lot of time in China
  472. 23:21and met with the leadership many times.
  473. 23:22- Maybe a good question to ask is,
  474. 23:24what are some things about China
  475. 23:26that people don't understand positive, just in the culture?
  476. 23:29What's some interesting things
  477. 23:31that you've learned about the Chinese?
  478. 23:36- Well, the sheer number of really smart,
  479. 23:39hardworking people in China is incredible.
  480. 23:44There are really, say like how many smart,
  481. 23:48hardworking people are there in China,
  482. 23:48there's far more of them there than there are here.
  483. 23:51I think in my opinion.
  484. 23:57And they've got a lot of energy.
  485. 24:00So I mean, the architecture in China
  486. 24:04that's in recent years is far more impressive than the US.
  487. 24:08I mean the train stations, the buildings,
  488. 24:10the high speed rail, everything,
  489. 24:12it's really far more impressive than what we have in the US.
  490. 24:19I mean, I recommend somebody just go to Shanghai
  491. 24:21and Beijing, look at the buildings
  492. 24:24and go to, you know, take the train from Beijing to Xian,
  493. 24:27where you have the terracotta warriors.
  494. 24:29China's got an incredible history, very long history.
  495. 24:33And you know, I think arguably the,
  496. 24:37in terms of the use of language from a written standpoint,
  497. 24:42sort of one of the oldest,
  498. 24:44perhaps the oldest written language.
  499. 24:48And then China, people did write things down.
  500. 24:50So now China historically has always been,
  501. 24:56with rare exception, been internally focused.
  502. 24:59They have not been inquisitive.
  503. 25:02They've fought each other.
  504. 25:04There've been many, many civil wars.
  505. 25:07In the three kingdoms war,
  506. 25:09I believe they lost about 70% of their population.
  507. 25:14So, and that,
  508. 25:15so they've had brutal internal wars,
  509. 25:19like civil wars that make the US Civil War
  510. 25:21look small by comparison.
  511. 25:28So I think it's important to appreciate
  512. 25:31that China is not monolithic.
  513. 25:34We sort of think of like China
  514. 25:36as this sort of one entity of one mind.
  515. 25:39And this is definitely not the case
  516. 25:42from what I've seen,
  517. 25:45and I think most people who understand China would agree.
  518. 25:48People in China think about China
  519. 25:5010 times more than they think
  520. 25:51about anything outside of China.
  521. 25:53So it's like 90% of their consideration is,
  522. 25:58you know, our is internal.
  523. 26:01- Well, isn't that a really positive thing
  524. 26:03when you're talking about the collaboration
  525. 26:05and a future piece between superpowers,
  526. 26:08when you're inward facing,
  527. 26:09which is like focusing on improving yourself
  528. 26:12versus focusing on... - [Elon] Yeah.
  529. 26:15- "Improving" others through military might.
  530. 26:18- The good news, the history of China suggests
  531. 26:20that China is not inquisitive.
  532. 26:22Meaning they're not gonna go out
  533. 26:23and invade a whole bunch of countries.
  534. 26:26Now they do feel very strongly,
  535. 26:27you know, so that's good.
  536. 26:28I mean, 'cause a lot of very powerful countries
  537. 26:31have been inquisitive.
  538. 26:34The US is one of the, also one of the rare cases
  539. 26:37that has not been inquisitive.
  540. 26:38Like after World War II,
  541. 26:39the US could have basically taken
  542. 26:41over the world in any country.
  543. 26:42Like, we've got nukes, nobody else has got nukes.
  544. 26:45We don't even have to lose soldiers.
  545. 26:47Which country do you want?
  546. 26:49And the United States could have taken over everything.
  547. 26:52Oh wait, at will.
  548. 26:53And it didn't, and the United States actually
  549. 26:56helped rebuild countries.
  550. 26:57So it helped rebuild Europe, you know, helped rebuild Japan.
  551. 27:02This is very unusual behavior, almost unprecedented.
  552. 27:09You know, the US did conspicuous acts of kindness
  553. 27:13like the Berlin airlift, you know?
  554. 27:19And I think, you know, it's always like,
  555. 27:22well, America's done bad things.
  556. 27:23Well, of course America's done bad things.
  557. 27:24But one needs to look at the whole track record.
  558. 27:29And just generally, you know, one sort of test
  559. 27:34would be how do you treat your prisoners at war?
  560. 27:38Or let's say, you know,
  561. 27:42no offense to the Russians, but let's say you're in Germany,
  562. 27:45it's 1945, you got the Russian Army coming one side
  563. 27:49and you got the French, British,
  564. 27:51and American Army's coming the other side.
  565. 27:53Who would you like to be to surrender it to?
  566. 27:56Like, no country is like morally perfect,
  567. 27:58but I recommend being a POW with the Americans.
  568. 28:03That would be my choice very strongly. (laughing)
  569. 28:07- In the full menu of POWs.
  570. 28:09- Very much so.
  571. 28:10And in fact, Wernher von Braun, took,
  572. 28:15you know, smart guy was like,
  573. 28:17we've got to be captured by the Americans.
  574. 28:20And in fact, the SS was under orders to execute Von Braun
  575. 28:26and all of the German rocket initiatives.
  576. 28:30And they narrowly escaped their SS.
  577. 28:34They said they were going out for a walk in the woods.
  578. 28:36They left in the middle of winter with no coats.
  579. 28:39And they ran like, with no food, no coats, no water.
  580. 28:42And just ran like hell and ran west.
  581. 28:45And by Sherlock, they, I think his brother found
  582. 28:48like a bicycle or something,
  583. 28:51and then just cycled west as fast
  584. 28:54as he couldn't have found a US patrol.
  585. 28:58So anyway, that's one way you can tell morality is.
  586. 29:04Where do you wanna be PoW? (laughing)
  587. 29:07It's not fun anywhere,
  588. 29:08but some places are much worse than others.
  589. 29:10So anyways, so like America has been,
  590. 29:15while far from perfect, generally a benevolent force,
  591. 29:21and we should always be self-critical
  592. 29:23and we try to be better.
  593. 29:27But anyone with have brain knows that.
  594. 29:31So I think there are some,
  595. 29:33in this way China and the United States are similar.
  596. 29:37Neither country has been acquisitive in a significant way.
  597. 29:42So that's like a, you know, a shared principle I guess.
  598. 29:47Now China does feel very strongly about Taiwan.
  599. 29:51They've been very clear about that for a long time.
  600. 29:56You know, from their standpoint,
  601. 29:57it would be like one of the states is, you know,
  602. 30:02not there like Hawaii or something like that.
  603. 30:04But more significant than Hawaii, you know?
  604. 30:10And Hawaii is pretty significant for us.
  605. 30:11So they view it as really the,
  606. 30:17that there's a fundamental part of China.
  607. 30:20The islander Formosa not Taiwan,
  608. 30:23that is not part of China, but should be.
  609. 30:27And the only reason it hasn't been
  610. 30:29is 'cause the US Pacific Fleet.
  611. 30:32- And as their economic power grows
  612. 30:34and as their military power grows,
  613. 30:36the thing that they're clearly saying
  614. 30:40is their interest will, you know, clearly be materialized.
  615. 30:46- Yes.
  616. 30:48China has been very clear that they'll incorporate Taiwan
  617. 30:54peacefully or militarily,
  618. 30:57but that they will incorporate it
  619. 31:00from their standpoint is 100% likely.
  620. 31:04- You know, something you said about conspicuous
  621. 31:05acts of kindness.
  622. 31:06As a geopolitical policy, it almost seems naive,
  623. 31:12but I'd venture to say that this is probably
  624. 31:15the path forward, how you avoid most wars.
  625. 31:20Just as you say it, it sounds naive,
  626. 31:22but it's kind of brilliant.
  627. 31:25If you believe in the goodness
  628. 31:27of underlying most of human nature.
  629. 31:31It just seems like it conspicuous acts of kindness
  630. 31:35can reverberate through the populace
  631. 31:38of the countries involved.
  632. 31:41- Yeah, well. - [Lex] And deescalate.
  633. 31:44- Absolutely.
  634. 31:44So after World War I, they made a big mistake.
  635. 31:50You know, they basically tried to lump
  636. 31:52all of lamb on Germany
  637. 31:55and, you know, settled Germany
  638. 32:00with impossible reparations.
  639. 32:05And, you know, really, there was a lot of,
  640. 32:07there was a fair, quite a bit of blame
  641. 32:08to go around for World War I,
  642. 32:12but they try to, you know, put it all in Germany.
  643. 32:18And that was, that laid the seeds for World War II.
  644. 32:25So it's a lot of people were not just Hitler,
  645. 32:28a lot of people felt wronged
  646. 32:31and they wanted vengeance
  647. 32:37and they got it.
  648. 32:38- [Lex] People don't forget.
  649. 32:39- Yeah, you know, you kill somebody's father
  650. 32:44or mother or son, daughter, they're not gonna forget it.
  651. 32:47They'll want vengeance.
  652. 32:50So after World War II, they're like, well,
  653. 32:52the Treaty of Versailles was a huge mistake in World War I.
  654. 32:58And so this time, instead of,
  655. 33:04you know, crushing the losers,
  656. 33:06we're actually gonna help 'em with the Marshall Plan
  657. 33:10and we're gonna help rebuild Germany.
  658. 33:14We're gonna help rebuild or, you know, Austria
  659. 33:17and the other, you know, Italy and whatnot, so.
  660. 33:25And that was the right move.
  661. 33:26- There's a, it does feel like there's a profound truth
  662. 33:31to conspicuous acts of kindness being an antidote to this.
  663. 33:37- Something must stop the cycle of reciprocal violence.
  664. 33:41Something must stop it
  665. 33:42or it will, you know, it'll never stop.
  666. 33:47Just eye for an eye, tooth for tooth,
  667. 33:51limb for a limb, life for a life, forever and ever
  668. 33:57- To escape briefly,
  669. 33:58the darkness was some incredible engineering work.
  670. 34:02xAI just released Grok AI assistant
  671. 34:04that I've gotten a chance to play with.
  672. 34:06It's amazing on many levels.
  673. 34:08First of all, it's amazing that a relatively small team
  674. 34:11in a relatively short amount of time was able to develop
  675. 34:15this close to state-of-the art system.
  676. 34:18Another incredible thing is there's a regular mode
  677. 34:22and there's a fun mode.
  678. 34:23- Yeah. I guess I'm to blame for that one. (laughing)
  679. 34:26- I wish, first of all,
  680. 34:27I wish everything in life had a fun mode.
  681. 34:29- Yeah.
  682. 34:30- There's something compelling
  683. 34:32beyond just fun about the fun mode
  684. 34:35interacting with a large language model.
  685. 34:38I'm not sure exactly what it is
  686. 34:38'cause I only have had a little bit of time to play with it.
  687. 34:42But it just makes it more interesting,
  688. 34:45more vibrant to interact with the system.
  689. 34:47- Yeah, absolutely.
  690. 34:52Our AI Grok is modeled
  691. 34:55after the Hitchhiker's Guide to the Galaxy,
  692. 34:59which is one of my favorite books, which is,
  693. 35:01it's a book on philosophy, disguises book on humor.
  694. 35:07And I would say that is that forms the basis
  695. 35:10of my philosophy, which is that
  696. 35:13we don't know the meaning of life,
  697. 35:16but the more we can expand the scope
  698. 35:18and scale of consciousness,
  699. 35:22digital and biological, the more we're able to understand
  700. 35:26what questions to ask about the answer that is the universe.
  701. 35:30So I have a philosophy of curiosity.
  702. 35:34- There is generally a feeling like this AI system
  703. 35:38has an outward looking,
  704. 35:42like the way you are, like sitting
  705. 35:43with a good friend looking up at the stars,
  706. 35:46like the asking pod head, like questions about the universe,
  707. 35:49wondering what it's all about.
  708. 35:50The curiosity you talk about.
  709. 35:52There's a sense no matter how mundane the question I ask,
  710. 35:55there's a sense of cosmic grandeur to the whole thing.
  711. 35:59- Well we are actually working hard to have engineering
  712. 36:04math, physics answers that you can count on.
  713. 36:09So for the other sort of AIs out there
  714. 36:12that these so-called large language models,
  715. 36:17I've not found the engineering to be reliable.
  716. 36:23And the hallucination,
  717. 36:24it unfortunately hallucinates most
  718. 36:26when you at least want it to hallucinate.
  719. 36:29So when you're asking important difficult questions,
  720. 36:32that that's where it tends to be confidently wrong.
  721. 36:36So we're really trying hard to say, okay,
  722. 36:39how do we be as grounded as possible
  723. 36:42so you can count on the results?
  724. 36:45Trace things back to physics first principles,
  725. 36:48mathematical logic.
  726. 36:51So underlying the humor is an aspiration
  727. 36:55to adhere to the truth of the universe
  728. 36:58as closely as possible.
  729. 37:01- [Lex] That's really tricky. - It is tricky.
  730. 37:04So that's why, you know,
  731. 37:07there's always gonna be some amount of error,
  732. 37:09but we want to aspire to be
  733. 37:13as truthful as possible about the answers
  734. 37:18with acknowledged error.
  735. 37:20So that there was always, you don't wanna be confidently
  736. 37:22wrong, so you're not gonna be right every time,
  737. 37:26but you wanna minimize
  738. 37:28how often you're confidently wrong.
  739. 37:31And then like I said, once you can count on the logic
  740. 37:35as being not violating physics,
  741. 37:40then you can start to build on that,
  742. 37:42to create inventions, like invent new technologies.
  743. 37:46But if you can't, if you cannot
  744. 37:50count on the foundational physics being correct,
  745. 37:52obviously the inventions are simply wishful thinking,
  746. 37:56you know, imagination land, magic basically.
  747. 38:01- Well, as you said, I think one of the big goals
  748. 38:04of xAI is to understand the universe.
  749. 38:06- Yes. That's how simple three word mission. (laughing)
  750. 38:13- If you look out far into the future,
  751. 38:15do you think on this level of physics,
  752. 38:17the very edge of what we understand about physics,
  753. 38:19do you think it will make discoveries,
  754. 38:22sort of the sexiest discovery of them as we know now,
  755. 38:26sort of unifying general relativity and quantum mechanics.
  756. 38:30So coming up with a theory of everything,
  757. 38:32do you think it could push towards that direction?
  758. 38:35Almost like theoretical physics discoveries.
  759. 38:38- If an AI cannot figure out new physics,
  760. 38:41it's clearly not equal to humans,
  761. 38:44let alone nor has surpassed humans,
  762. 38:45'cause humans have figured out new physics.
  763. 38:47They're just, you know, physics is just understanding,
  764. 38:50you know, deepening wants insight into how reality works.
  765. 38:54And then there's engineering which is inventing things
  766. 38:59that have never existed.
  767. 39:01Now the range of possibilities for engineering
  768. 39:03is far greater than for physics.
  769. 39:05'Cause you know, once you figure out the rules
  770. 39:06of the universe, that's it.
  771. 39:09You've discovered things that already existed.
  772. 39:12But from that you can then build technologies
  773. 39:15with that are really almost limitless
  774. 39:18in the variety and cap- you know?
  775. 39:23It's like once you understand the rules
  776. 39:24of the game properly, and we do, you know,
  777. 39:27with current physics, we do at least at a local level,
  778. 39:29understand how physics works very well.
  779. 39:32Where our ability to predict things is incredibly good.
  780. 39:35Like quantum mechanics is,
  781. 39:36the degree to which quantum mechanics can predict outcomes
  782. 39:39is incredible.
  783. 39:42That was my heart hardest class in college by the way.
  784. 39:45My senior quantum mechanics class was harder
  785. 39:48than all of my other classes put together.
  786. 39:50- To get an AI system, a large language model
  787. 39:53to reliably, be as reliable as quantum mechanics
  788. 39:59and physics is very difficult.
  789. 40:01- Yeah. You have to test any conclusions against the ground,
  790. 40:04truth of reality.
  791. 40:05Reality is the ultimate judge.
  792. 40:06Like physics is the law,
  793. 40:07everything else is a recommendation.
  794. 40:09I've seen plenty of people break the laws made by man,
  795. 40:13but none break the laws made by physics.
  796. 40:15- Yeah, it's a good test actually.
  797. 40:17If this LLM understands
  798. 40:20and matches physics, then you can more reliably trust
  799. 40:23whatever it thinks about the current state
  800. 40:25of politics in some sense.
  801. 40:28- And it's also not the case currently
  802. 40:30that even that its internal logic is not consistent.
  803. 40:34So, especially with these,
  804. 40:39with the approach of like just predicting a token, predict,
  805. 40:42token, predict, token, it's like a vector sum.
  806. 40:45You know, you're summing up a bunch of vectors,
  807. 40:47but you can get drift.
  808. 40:49So as those, a little bit of error,
  809. 40:51a little bit of error adds up.
  810. 40:53And by the time you are many tokens down the path,
  811. 40:57it doesn't make any sense.
  812. 40:59- So it has to be somehow self-aware about the drift.
  813. 41:02- It has to be self-aware about the drift,
  814. 41:04and then look at the thing as a gestalt as a whole
  815. 41:08and say it does it have coherence as a whole.
  816. 41:13So, you know, when authors write books
  817. 41:18that they will write the book and then they'll go
  818. 41:20and revise it, you know, taking into account,
  819. 41:24you know, all the, the end and the beginning
  820. 41:26and the middle and rewrite it to achieve coherence
  821. 41:30so that it doesn't end up in a nonsensical place.
  822. 41:33- [Lex] Maybe the process of revising is what...
  823. 41:36- Yeah.
  824. 41:37- Reasoning is, and then that's the process of revising
  825. 41:39is how you get closer and closer to truth.
  826. 41:42Maybe you, like, at least I approached it that way.
  827. 41:45You just say a bunch of bullshit first
  828. 41:46and then you get it better.
  829. 41:48You start a bullshit and then you get...
  830. 41:49- Create a draft and then you iterate on that draft
  831. 41:53until it has has coherence,
  832. 41:55until it all adds up basically.
  833. 41:58- So another question about theory of everything,
  834. 42:01but for intelligence, do you think there exists
  835. 42:04as you're exploring this with xAI,
  836. 42:06creating this intelligence system,
  837. 42:07do you think there is a theory of intelligence
  838. 42:10where you get to understand what,
  839. 42:13like what is the I in AGI
  840. 42:17and what is the I in human intelligence?
  841. 42:22- There's no I in Team America, oh wait, there is.
  842. 42:25(both laughing)
  843. 42:27- No, it's gonna be stuck in my head now.
  844. 42:28(both laughing)
  845. 42:31Yeah. There's no me and whatever in quantum mechanics.
  846. 42:37Oh, wait. (Elon laughing)
  847. 42:41I mean is that part of the process of discovering,
  848. 42:44understanding the universe is understanding intelligence?
  849. 42:48- Yeah. Yeah.
  850. 42:50I think we need to understand intelligence,
  851. 42:52understand consciousness.
  852. 42:53I mean there are some sort of fundamental questions of like,
  853. 42:57what is thought?
  854. 43:00What is emotion?
  855. 43:01- [Lex] Yeah. - Is it really just one atom
  856. 43:04bumping into another atom?
  857. 43:06It feels like something more than that.
  858. 43:08So I think we're probably missing some really big things,
  859. 43:16like some really big things.
  860. 43:18- Something that'll be obvious in retrospect.
  861. 43:20- [Elon] Yes. - Like there's a giant,
  862. 43:22like you put the whole consciousness, emotion.
  863. 43:26- Well, some people would quote like a soul,
  864. 43:28you know, in religion it'd be a soul.
  865. 43:31Like you feel like you're you, right?
  866. 43:33I mean, you don't feel like you're just a collection
  867. 43:37of atoms, but on what dimension does thought exist?
  868. 43:41What dimension does do emotions exist?
  869. 43:43We feel them very strongly.
  870. 43:46I suspect there's more to it than atoms bumping into atoms.
  871. 43:52- And maybe AI can pave the path to the discovery
  872. 43:55of whatever the hell that thing is.
  873. 43:58- Yeah. What is consciousness?
  874. 43:59Like what are, when you put the atoms in a particular shape,
  875. 44:02why are they able to form thoughts
  876. 44:05and take actions, that and feelings.
  877. 44:10- And even if it is an illusion,
  878. 44:11why is this illusion so compelling?
  879. 44:13- Yeah. Like how do...
  880. 44:15- Why does this illusion exist?
  881. 44:17- [Lex] Yeah. - On what plain
  882. 44:19does this illusion exist?
  883. 44:21- [Lex] Yeah. - And sometimes I wonder is,
  884. 44:25you know, either perhaps everything's conscious
  885. 44:28or nothing is conscious, one of the two.
  886. 44:32- I like the former,
  887. 44:34everything conscious just seems more fun.
  888. 44:37- It does seem more more fun. Yes.
  889. 44:41But we're composed of atoms
  890. 44:42and those atoms are composed of quarks and leptons.
  891. 44:45And those quarks and leptons have been around
  892. 44:47since the beginning of the universe.
  893. 44:50- The beginning of the universe.
  894. 44:53- Right. What seems to be the beginning of the universe.
  895. 44:55- The first time we talked, you said what you would,
  896. 44:57which is surreal to think
  897. 44:59that this discussion was happening is becoming a reality.
  898. 45:02I asked you what question would you ask an AGI system
  899. 45:05once you create it?
  900. 45:07And you said, what's outside the simulation is the question.
  901. 45:10And good question.
  902. 45:13- [Elon] Yeah. - But it seems like with Grok
  903. 45:14you started to literally, the system's goal
  904. 45:19is to be able to ask such questions,
  905. 45:21to answer such questions
  906. 45:22and to ask such questions.
  907. 45:24- Where are the aliens? - [Lex] Where are the aliens?
  908. 45:26- That's one of the, like Fermi paradox question.
  909. 45:30A lot of people have asked me
  910. 45:31if I've seen any evidence of aliens and I haven't,
  911. 45:34which is kind of concerning.
  912. 45:37'Cause then I think what I'd probably prefer
  913. 45:40to at least have seen some archeological evidence of aliens,
  914. 45:44to the best of my knowledge, there is no proof.
  915. 45:46I'm not aware of any evidence of aliens.
  916. 45:50If they're out there, they're very subtle.
  917. 45:52We might just be the only consciousness,
  918. 45:54at least in the galaxy.
  919. 45:58And if you look at say the history of Earth,
  920. 46:01forensically the archeological record,
  921. 46:03Earth is about four and a half billion years old.
  922. 46:06Civilization as measured from the first writing
  923. 46:09is only about 5,000 years old.
  924. 46:11We have to give some credit there to the ancient Sumerians
  925. 46:14who aren't around anymore.
  926. 46:15I think it was a archaic,
  927. 46:17pre cuneiform was the first actual symbolic representation,
  928. 46:21but only about 5,000 years ago.
  929. 46:24I think that's a good date
  930. 46:26for when word say civilization started.
  931. 46:30That's one millionth of Earth's existence.
  932. 46:35So civilization has been around,
  933. 46:37it's really a flash in the pan so far.
  934. 46:45And why have we, why did it take so long for,
  935. 46:51you know, four and a half billion years?
  936. 46:57For the vast majority of the time, there was no life.
  937. 47:00And then there was archaic there bacteria
  938. 47:02for a very long time.
  939. 47:04And then, you know, you had mitochondria get captured,
  940. 47:11multicellular life.
  941. 47:15Differentiation into plants and animals.
  942. 47:18Life moving from the oceans to land.
  943. 47:20Mammals, higher brain functions.
  944. 47:25And the sun is expanding slowly,
  945. 47:29but it will overheat,
  946. 47:33it will heat the earth up at some point in the future,
  947. 47:38boil the oceans and and earth will become like Venus.
  948. 47:42Where near life, life as we know it is impossible.
  949. 47:46So if we do not become multi-planetary
  950. 47:49and ultimately go beyond our solar system,
  951. 47:53annihilation of all life on earth is a certainty.
  952. 47:57A certainty.
  953. 48:00And it could be as little as on the galactic timescale,
  954. 48:07half a billion years.
  955. 48:09You know, it's long time by human standards,
  956. 48:11but that's only 10% longer
  957. 48:13than earth has been around at all.
  958. 48:16So if life had taken 10% longer to evolve on earth,
  959. 48:21it wouldn't exist at all.
  960. 48:27- We got a deadline coming up. Better hurry.
  961. 48:30But that said, as you said, humans intelligent life on earth
  962. 48:35developed a lot of cool stuff very quickly,
  963. 48:38- So... - [Elon] Yes.
  964. 48:39- It seems like becoming a multi-planetary
  965. 48:42is almost inevitable.
  966. 48:45Unless we destroy this thing. - We need to do it.
  967. 48:47I mean, it's not, I mean, I suspect that there,
  968. 48:52if we are able to go out there
  969. 48:54and explore other star systems that we,
  970. 48:56there's a good chance we find a whole bunch
  971. 48:58of long dead one planet civilizations,
  972. 49:01then have made it past their home planet.
  973. 49:03- That's so sad. - Yeah.
  974. 49:05- It's sad. Also fascinating.
  975. 49:08- I mean, there are various explanations
  976. 49:09for Fermi paradox.
  977. 49:11And one is the sort of, they're these great filters,
  978. 49:13which civilizations don't pass through.
  979. 49:17And one of those great filters is
  980. 49:19do you become a multi-plan civilization or not?
  981. 49:21And if you don't, it's simply a matter of time
  982. 49:24before something happens on your planet.
  983. 49:28You know, either natural
  984. 49:30or manmade that causes us to die out.
  985. 49:34Like the dinosaurs, where are they now
  986. 49:38that didn't have spaceships, so. (laughing)
  987. 49:42- I think the more likely thing is,
  988. 49:44'cause just to empathize with the aliens
  989. 49:48that they found us and they're protecting us
  990. 49:50and letting us be. - I hope so.
  991. 49:52Nice aliens.
  992. 49:53- Just like the tribes in the Amazon,
  993. 49:55the uncontacted tribes, we're protecting them.
  994. 49:58That's what...
  995. 49:59- That would be a nice explanation.
  996. 50:00- Or you could have like a, what was it?
  997. 50:04I think Andrej Karpathy said, it's like
  998. 50:06the ants in the Amazon asking where's everybody?
  999. 50:10- Well, they do run into a lot of other ants.
  1000. 50:12- [Lex] That's true. - They have these ant wars.
  1001. 50:15- (laughing) Sounds like a good TV show.
  1002. 50:18- Yeah, they literally have these big wars
  1003. 50:19between various ants.
  1004. 50:21- Yeah, maybe I'm just (both laughing)
  1005. 50:25dismissing all the different diversity events.
  1006. 50:27- You should listen to that Warner Herzog
  1007. 50:29talking about the jungle.
  1008. 50:30It's really hilarious. Have you heard it?
  1009. 50:31- No, I have not. - [Elon] It's awesome.
  1010. 50:32- But Warner Herzog is a way...
  1011. 50:34(both laughing)
  1012. 50:36- You should play it for the, you know,
  1013. 50:40as an interlude in the, it's on YouTube.
  1014. 50:43It's awesome. (both laughing)
  1015. 50:45- [Lex] I love him so much. - Yeah, he's great.
  1016. 50:47- Was he the director of "Happy People:
  1017. 50:49Life in the Taiga"?
  1018. 50:50I think also
  1019. 50:51- He did that bear documentary,
  1020. 50:52- [Lex] The bear documentary.
  1021. 50:53- And this thing about penguins.
  1022. 50:54- Yeah. (both laughing)
  1023. 50:57The analysis, the psychoanalysis of penguins.
  1024. 51:00- Yeah, the penguins like headed for like mountains
  1025. 51:03like that are like 70 miles away.
  1026. 51:05And Penguin has just had for doom basically.
  1027. 51:08- Well, he had a cynical take.
  1028. 51:10I have a, he could be just a brave explorer
  1029. 51:13and there'll be great stories told
  1030. 51:16about him amongst the penguin population
  1031. 51:18for many centuries to come.
  1032. 51:22(laughing)
  1033. 51:24What were we talking about? Okay.
  1034. 51:25- [Elon] Penguins. (both laughing)
  1035. 51:28- Yeah, so aliens, I mean, I don't know.
  1036. 51:31Look, I think the smart move is just, you know,
  1037. 51:35this is the first time in the history of earth
  1038. 51:38that it's been possible for life
  1039. 51:40to extend beyond earth.
  1040. 51:43That window is open.
  1041. 51:47Now, it may be open for a long time
  1042. 51:49or it may be open for a short time
  1043. 51:51and it may be open now and then never open again.
  1044. 51:55So I think the smart move here is
  1045. 51:58to make life multiplanetary while it's possible to do so.
  1046. 52:01We don't wanna be one of those
  1047. 52:02lame one planet civilizations that just dies out.
  1048. 52:04- No, those are lame. - Yeah, lame.
  1049. 52:09Self-respecting civilization would be one planet.
  1050. 52:11- There's not gonna be a Wikipedia entry
  1051. 52:13for one of one of those.
  1052. 52:15And pause, the SpaceX have an official policy
  1053. 52:21for when we meet aliens.
  1054. 52:23- No. (laughing)
  1055. 52:25- That seems irresponsible.
  1056. 52:29(both laughing)
  1057. 52:30- I mean, look, if I see the slightest indication
  1058. 52:34that there are aliens,
  1059. 52:35I will immediately post on X platform anything I know.
  1060. 52:38- It could be the most liked reposted post of all time.
  1061. 52:42- Yeah, I mean look,
  1062. 52:44we have more satellites up there right now
  1063. 52:46than everyone else combined.
  1064. 52:48So, you know, we'd know, we know if we've got a maneuver
  1065. 52:51around something,
  1066. 52:52and we not don't have to maneuver around anything.
  1067. 52:55- If we go to the big questions once again,
  1068. 52:57you said you've,
  1069. 52:59you're with Einstein that you believe in the God of Spinoza.
  1070. 53:03- (laughing)Yes.
  1071. 53:05- So, you know, that's a view
  1072. 53:07that God is like the universe
  1073. 53:08and is reveals himself through the laws of physics
  1074. 53:12or as Einstein said,
  1075. 53:13through the lawful harmony of the world.
  1076. 53:16- Yeah, I would agree that God of the simulator
  1077. 53:19or whatever the supreme being or beings,
  1078. 53:25reveal themselves through the physics.
  1079. 53:28You know, they're creatives of this existence
  1080. 53:31and it's incumbent upon us to try to understand
  1081. 53:34more about this wondrous creation.
  1082. 53:37- Like who created this thing, who's running this thing?
  1083. 53:40Like embodying it into a singular question
  1084. 53:43with a sexy word on top of it
  1085. 53:45is like focusing the mind to understand it.
  1086. 53:47It does seem like there's a, again, it could be an illusion.
  1087. 53:52It seems like there's a purpose
  1088. 53:54that there's an underlying master plan of some kind.
  1089. 53:57And it seems like...
  1090. 53:58- There may not be a master plan.
  1091. 54:00In the sense, so there's, like,
  1092. 54:04maybe an interesting answer to the question
  1093. 54:05of determinism versus free will is
  1094. 54:08that if we are in a simulation,
  1095. 54:10the reason that these higher beings would hold a simulation
  1096. 54:14is to see what happens.
  1097. 54:16So it's not, they don't know what happens,
  1098. 54:22otherwise they wouldn't hold the simulation.
  1099. 54:24So when humans create a simulation,
  1100. 54:27so it's SpaceX and Tesla,
  1101. 54:29we create simulations all the time,
  1102. 54:31especially for the rocket.
  1103. 54:32You, you know, you have to run a lot of simulations
  1104. 54:38to understand what's gonna happen
  1105. 54:39because you can't really test the rocket
  1106. 54:41until it goes to space, and you want it to work.
  1107. 54:44So you have to simulate subsonic, transonic,
  1108. 54:49supersonic, hypersonic ascent,
  1109. 54:54and then coming back super high heating
  1110. 54:56and orbital dynamics.
  1111. 54:59All this has gotta be simulated.
  1112. 55:00So, 'cause you don't get very many kicks at the can.
  1113. 55:04But we run the simulations to see what happens.
  1114. 55:06Not if we knew what happens, we wouldn't run the simulation.
  1115. 55:09So if there's, so whoever created this existence,
  1116. 55:14is they're running it
  1117. 55:17because they don't know what's gonna happen.
  1118. 55:19Not because they do.
  1119. 55:23- So maybe we both played Diablo,
  1120. 55:25maybe Diablo was created to see if it Druid your character
  1121. 55:30could defeat Uber Lilith at the end.
  1122. 55:32They didn't know.
  1123. 55:34- Well, the funny thing is the Uber Lilith title
  1124. 55:38is Hatred Incarnate.
  1125. 55:40- [Lex] Yeah. - And right now,
  1126. 55:43I guess, you're gonna ask the Diablo team,
  1127. 55:46but it's almost impossible to defeat hatred
  1128. 55:53in the eternal realm.
  1129. 55:55- Yeah, you've streamed yourself dominating tier 100
  1130. 55:58Nightmare Dungeons and still...
  1131. 56:00- I can cruise through tier 100 Nightmare Dungeons
  1132. 56:04like a stroll in the park.
  1133. 56:07- And still you're defeated by Hatred.
  1134. 56:09- Yeah,
  1135. 56:09I can, there's the sort of,
  1136. 56:11I guess maybe the second hardest boss is Duriel,
  1137. 56:14Duriel can't even scratch the paint.
  1138. 56:16So I killed Duriel so many times ,
  1139. 56:20and every other boss in the game, all of 'em,
  1140. 56:23kill 'em so many times it's easy.
  1141. 56:26But Uber Lilith, otherwise known as Hatred Incarnate,
  1142. 56:31especially if you're a Druid and you have no ability to go
  1143. 56:34and to be vulnerable,
  1144. 56:35there are these random death waves that come at you.
  1145. 56:42And I'm pretty, you know, I really am 52,
  1146. 56:45so my reflex is not what they used to be,
  1147. 56:46but I have a lifetime of playing video games.
  1148. 56:50At one point I was, you know,
  1149. 56:53maybe one of the best Quake players in the world.
  1150. 56:56I actually won money in what I think was the first
  1151. 56:59paid eSports tournament in the US.
  1152. 57:03We're doing four person Quake tournaments.
  1153. 57:06And we came second, I was the second best person on the team
  1154. 57:11and the actual best person,
  1155. 57:13we're actually winning,
  1156. 57:14we would've come first except the best person on the team,
  1157. 57:16his computer crashed prior to the game.
  1158. 57:19So we came second.
  1159. 57:22But I got money for it and everything.
  1160. 57:23So like, basically I got skills, you know, albeit,
  1161. 57:25you know, no spring chicken these days.
  1162. 57:28And to be totally frank, it's driving me crazy,
  1163. 57:34trying to beat Lilith as a Druid.
  1164. 57:36Basically trying to beat, trying to beat Hatred Incarnate
  1165. 57:39in the eternal realm.
  1166. 57:41- [Lex] As a Druid. - As a Druid.
  1167. 57:42And if you, if you, (laughing)
  1168. 57:44this is really (indistinct), lemme tell you.
  1169. 57:49- I mean, the challenge is part of the fun.
  1170. 57:50I have seen directly like you're actually
  1171. 57:54like a world class incredible video game player.
  1172. 57:56- [Elon] Yeah. - And I think Diablo,
  1173. 57:58so you're just picking up a new game
  1174. 58:00and you're figuring out it's fundamentals.
  1175. 58:03You're also with the Paragon Board
  1176. 58:04and the build are not somebody like me
  1177. 58:09who perfectly follows whatever they suggest on the internet.
  1178. 58:12You're also an innovator there.
  1179. 58:13- [Elon] Yeah.
  1180. 58:15- Which is hilarious to watch.
  1181. 58:17It's like a mad scientist just trying
  1182. 58:19to figure out the Paragon Board
  1183. 58:21and the build and the, you know.
  1184. 58:23Is there some interesting insights there about,
  1185. 58:26if somebody's starting as a Druid, do you have advice?
  1186. 58:30- I would not recommend playing a Druid
  1187. 58:31in the eternal realm.
  1188. 58:33Right now, I think the most powerful character
  1189. 58:36in the seasonal realm is the Sorcerer
  1190. 58:39with the lightning balls.
  1191. 58:41The Sorcs have huge balls in the seasonal.
  1192. 58:45- Well yeah, that's what they say.
  1193. 58:48- So Sorc have huge balls.
  1194. 58:52They do, huge balls of lightning.
  1195. 58:54- I'll take you word for it.
  1196. 58:56- And it's actually, in the seasonal realm,
  1197. 58:57you can, it's like pretty easy to beat
  1198. 59:01Uber Lilith with the,
  1199. 59:02because she got these vampiric powers
  1200. 59:04that out fire damage
  1201. 59:05and increase your defense and whatnot.
  1202. 59:09So, but really quite easy to defeat Hatred seasonally,
  1203. 59:15but to defeat hatred eternally, very difficult.
  1204. 59:19Almost impossible. It's totally impossible.
  1205. 59:21It seems like this is a metaphor for life, you know?
  1206. 59:24- I like the idea that Elon Musk,
  1207. 59:27because I saw, I was playing Diablo yesterday
  1208. 59:29and I saw 100, level 100 drew just run by,
  1209. 59:33I will never die and then run back the way.
  1210. 59:36- And it was, there's just some,
  1211. 59:37this metaphor's kind of hilarious that you,
  1212. 59:40Elon Musk is fighting Hatred,
  1213. 59:43restlessly fighting Hatred in this demonic realm.
  1214. 59:47- [Elon] Yes. - It's hilarious.
  1215. 59:48I mean it's pretty hilarious.
  1216. 59:50- No, it's absurd. (both laughing)
  1217. 59:53Really, it's exercise and absurdity,
  1218. 59:54and it makes me wanna pull my hair out.
  1219. 59:56- Yeah. What do you get from video games in general?
  1220. 59:59Is there for you personally.
  1221. 1:00:03- I mean, it's, I don't known it's,
  1222. 1:00:05it calms my mind.
  1223. 1:00:06I mean, you sort of killing the demons in a video game
  1224. 1:00:10calms the demons in my mind.
  1225. 1:00:14Yeah.
  1226. 1:00:15If you play a tough video game,
  1227. 1:00:17you can get into like a state of flow,
  1228. 1:00:19which is very enjoyable.
  1229. 1:00:24But the admittedly it needs to be not too easy,
  1230. 1:00:27not too hard, kinda in the Goldilock zone.
  1231. 1:00:32And I guess you generally wanna feel
  1232. 1:00:33like you're progressing in the game.
  1233. 1:00:36So a good video,
  1234. 1:00:39and there's also beautiful art, engaging storylines,
  1235. 1:00:46and it's like an amazing puzzle to solve, I think.
  1236. 1:00:50And so it's like solving the puzzle.
  1237. 1:00:52- Elden Ring the greatest game of all time.
  1238. 1:00:54I still haven't played it, but you.
  1239. 1:00:56- Elden Ring is definitely a candidate
  1240. 1:00:57for best game ever.
  1241. 1:00:59Top five for sure.
  1242. 1:01:01- I think I've been scared how hard it is
  1243. 1:01:03or how hard I hear it is, so.
  1244. 1:01:06But it is beautiful.
  1245. 1:01:06- Elden Ring is, feels like it's designed by an alien.
  1246. 1:01:13There's a theme to this discussion. In what way?
  1247. 1:01:16- It's so unusual.
  1248. 1:01:18It's incredibly creative and the art is stunning.
  1249. 1:01:22I recommend playing it on a big resolution,
  1250. 1:01:25high dynamic range TV even, doesn't need to be a monitor,
  1251. 1:01:30just the art is incredible.
  1252. 1:01:32It's so beautiful
  1253. 1:01:34And it's so unusual.
  1254. 1:01:39And each of those top bus battles is unique.
  1255. 1:01:43Like, it's like a unique puzzle to solve.
  1256. 1:01:46Each one is different.
  1257. 1:01:48And the strategy you use to solve
  1258. 1:01:50one battle is different from another battle.
  1259. 1:01:54- That said, you said Druid and Enternal
  1260. 1:01:56against Uber Lilith is the hardest boss battle you've ever?
  1261. 1:02:00- Correct.
  1262. 1:02:00That is currently the,
  1263. 1:02:03and I've played a lot of video games.
  1264. 1:02:05That's my primary recreational activity.
  1265. 1:02:11And yes, beating Hatred in the Enternal Realm
  1266. 1:02:18is the hardest boss battle in life
  1267. 1:02:21and in the video game.
  1268. 1:02:23- [Lex] Metaphor on top of metaphor.
  1269. 1:02:25- I'm not sure it's possible,
  1270. 1:02:26but it's, I do make progress.
  1271. 1:02:29So then I'm like, okay, I'm making progress.
  1272. 1:02:31Maybe if I just tweak that Paragon Board a little more,
  1273. 1:02:37I can do it.
  1274. 1:02:37I could just dodge a few more waves, I can do it.
  1275. 1:02:43- Well the simulation is created for the purpose
  1276. 1:02:45of figuring out if it can be done.
  1277. 1:02:47And you are just a cog in that simulation,
  1278. 1:02:50in the machine of the simulation.
  1279. 1:02:51- Yeah it might be. (Lex laughing)
  1280. 1:02:54- I have a feeling that at least,
  1281. 1:02:59I think... - [Lex] it's doable.
  1282. 1:03:03- It's doable, yes.
  1283. 1:03:05- [Lex] Well that's the human spirit right there to believe.
  1284. 1:03:09- Yeah.
  1285. 1:03:10I mean, did prompt me to think about just hate in general,
  1286. 1:03:13which is, you wanna be careful of one of those things
  1287. 1:03:18where you wish for something that sounds good,
  1288. 1:03:23but if you get it, it's actually a dystopian situation.
  1289. 1:03:27So, you know, you could sort of run a, sort of,
  1290. 1:03:32have heart to sort of like,
  1291. 1:03:33if you wish for world peace, sounds good,
  1292. 1:03:36but how is it enforced?
  1293. 1:03:39And at what cost is, at what cost eternal peace?
  1294. 1:03:45It might actually be worse to have eternal peace
  1295. 1:03:47because of what that would entail.
  1296. 1:03:50The suppression of everyone.
  1297. 1:03:52It might be the suppression of progress.
  1298. 1:03:56It might be an ossified society that never changes.
  1299. 1:04:00What I'm saying is that there is an argument
  1300. 1:04:04that you, that if that,
  1301. 1:04:09if you wish for no war,
  1302. 1:04:11you should be careful what you wish for.
  1303. 1:04:14Because what's required in order for there to be no war?
  1304. 1:04:20Might be worse than a little war.
  1305. 1:04:22- Yeah, the things we'll consider to be flaws
  1306. 1:04:24of human civilization might be a necessary components
  1307. 1:04:27for whatever optimal looks like.
  1308. 1:04:30I mean this, do you worry about AI, AGI enabling a dystopian
  1309. 1:04:33state of this nature, whether it's 1984 with surveillance
  1310. 1:04:38and fear, or brave new world with pleasure,
  1311. 1:04:43and what is it?
  1312. 1:04:44Lots of sex, but no deep human experience.
  1313. 1:04:51- You know, there's actually a real drug called Soma.
  1314. 1:04:54- [Lex] There is? - Yes.
  1315. 1:04:55(both laughing)
  1316. 1:04:56- I'm like... - [Lex] Lemme take notes.
  1317. 1:04:57- Did they actually...
  1318. 1:04:58- Does Grok know about it?
  1319. 1:04:59No, it probably will go to Brave New World.
  1320. 1:05:02- We could ask it. - What would be the question?
  1321. 1:05:05- Tell us about the drug called Soma.
  1322. 1:05:08- Tell us about the drug called Soma.
  1323. 1:05:13It's gonna go to Brave New World, I think.
  1324. 1:05:15- Well there's a real drug as well.
  1325. 1:05:18- Searching for Soma drug information.
  1326. 1:05:20(both laughing)
  1327. 1:05:22- Oh boy.
  1328. 1:05:24Ah, Soma, the one to drug, by the way, I'm in fun mode.
  1329. 1:05:27- Oh, okay. Fun. (laughing)
  1330. 1:05:29Yeah, so it's talking about the real drug.
  1331. 1:05:31- [Lex] Oh, it's a muscle relaxant
  1332. 1:05:32that's been around since the 1950s.
  1333. 1:05:34It's used to treat muscle pain and discomfort.
  1334. 1:05:36And it does this by altering the activity
  1335. 1:05:39between neurons and the spinal cord.
  1336. 1:05:41Now, Soma isn't just a pain reliever,
  1337. 1:05:43it's also a bit of a sedative.
  1338. 1:05:45It can make you feel a bit drowsy.
  1339. 1:05:46So it's not something to take
  1340. 1:05:48if you're planning on operating having machinery
  1341. 1:05:50or trying to solve complex math equations.
  1342. 1:05:53And it goes on and on.
  1343. 1:05:54In conclusion, Soma's a drug that can be helpful
  1344. 1:05:56for those dealing with muscle pain,
  1345. 1:05:57but it's not without its risks.
  1346. 1:06:00It's pretty good. Oh, wait a minute.
  1347. 1:06:01- [Elon] That's the real drug it's talking about.
  1348. 1:06:03- [Lex] It's the real drug. - [Elon] Yeah.
  1349. 1:06:05- [Lex] And of course the fun is always use it responsibly
  1350. 1:06:08in accordance with the advice of a healthcare professional.
  1351. 1:06:10And remember in the end,
  1352. 1:06:11the universe is probably just a giant computer simulation.
  1353. 1:06:14So don't take anything too seriously.
  1354. 1:06:15I love it. I love it. (Elon laughing)
  1355. 1:06:18I love it. I love it.
  1356. 1:06:21- Yeah. So.
  1357. 1:06:22- Oh, and also really nicely includes
  1358. 1:06:24like the tweets post on X that are relevant to this,
  1359. 1:06:29which is really cool.
  1360. 1:06:30- [Elon] Yeah.
  1361. 1:06:31- And that integrate everything together.
  1362. 1:06:33- [Elon] Yeah. - Like the interface
  1363. 1:06:35is one of the really cool things here.
  1364. 1:06:36- Yeah. Seems done a great job.
  1365. 1:06:40(laughing)
  1366. 1:06:41So, and you say like, you could say like, no, I mean,
  1367. 1:06:45Soma from Brave New World.
  1368. 1:06:48'Cause it thinks you mean,
  1369. 1:06:49you mean Soma the real drug.
  1370. 1:06:51- Yeah. - [Elon] No, I mean,
  1371. 1:06:52Soma from Brave New World.
  1372. 1:06:56- It's gonna be like, oops, I'm really sorry.
  1373. 1:06:59I'm really surprised it went to the real drug first.
  1374. 1:07:03Ah, my apologies. I was thinking of the muscle relaxant.
  1375. 1:07:06Soma for Band New World is a different beast altogether.
  1376. 1:07:10Now this Huxley's dystopian novel,
  1377. 1:07:12Soma is a drug that plays a central role
  1378. 1:07:14in society, depicted in the book.
  1379. 1:07:15It's a mood altering, hallucinogenic drug that's used
  1380. 1:07:19to keep the population docile and content.
  1381. 1:07:22In the novel to use Soma in Brave New World
  1382. 1:07:24raises some interesting questions
  1383. 1:07:25about the nature of happiness
  1384. 1:07:26and the role of drugs in society.
  1385. 1:07:29Oh, man! - Exactly.
  1386. 1:07:30Is it better to live in a world
  1387. 1:07:32where everyone is happy all the time,
  1388. 1:07:34even if that happiness is artificial?
  1389. 1:07:37Good question. This is what I mean.
  1390. 1:07:39Like, do you want, do you wish for world peace
  1391. 1:07:43and happiness all the time?
  1392. 1:07:46Are you sure?
  1393. 1:07:48Because that might be a society that is essentially sterile
  1394. 1:07:53and ossified that never changes,
  1395. 1:07:55that is ultimately doomed.
  1396. 1:07:58- This kind of tension between darkness and light.
  1397. 1:08:03- It's, this is really a very good summary.
  1398. 1:08:07It really gets to the point.
  1399. 1:08:09This is not simply regurgitating a Brave New World.
  1400. 1:08:12It's actually getting to the salient element
  1401. 1:08:15of Soma as a drug.
  1402. 1:08:19Do you actually wanna be in a situation
  1403. 1:08:22where everyone is happy all the time,
  1404. 1:08:23even though it's artificial?
  1405. 1:08:25Or is it better to confront the challenges of life
  1406. 1:08:27and experience the full range of human emotions,
  1407. 1:08:29even if it means experiencing pain and suffering?
  1408. 1:08:31- For those listening, by the way,
  1409. 1:08:33Elon just read directly from Grok,
  1410. 1:08:35which is a really nice, kind of insightful,
  1411. 1:08:37philosophical analysis of the tension here.
  1412. 1:08:40Interesting.
  1413. 1:08:41- It pretty much nails it.
  1414. 1:08:42In conclusion, Soma from Brave New World is fictional drug
  1415. 1:08:45that's used to explore some deep philosophical questions
  1416. 1:08:47about the nature of happiness
  1417. 1:08:49and the role of drugs in society.
  1418. 1:08:51It's a powerful symbol of the dangers of using drugs
  1419. 1:08:54to escape from reality
  1420. 1:08:55and the importance of confronting the challenges
  1421. 1:08:57of life head-on.
  1422. 1:09:00Nailed it.
  1423. 1:09:02And the crazy thing is like, you know,
  1424. 1:09:03we do have a real drug called Soma,
  1425. 1:09:05which kind of is kind of like the drug in the book.
  1426. 1:09:09And I'm like, did they,
  1427. 1:09:10they must've named it after that Soma.
  1428. 1:09:13- Probably, probably, - Yeah.
  1429. 1:09:14Soma the real drug is quite effective on back pain.
  1430. 1:09:17- So, you know about this drug?
  1431. 1:09:18- I've taken it. - It's fascinating. Okay.
  1432. 1:09:21- 'Cause I had like a, you know, squashed disc
  1433. 1:09:24in my C five C six.
  1434. 1:09:26- So it takes the physical pain away, but Soma here is...
  1435. 1:09:28- It doesn't completely, it reduces the amount of pain
  1436. 1:09:31you feel, but at the expense of mental acuity.
  1437. 1:09:36It dells your mind.
  1438. 1:09:39Just the drug in the book.
  1439. 1:09:41- Just like the drug in the book.
  1440. 1:09:43- And hence the trade off.
  1441. 1:09:46The thing that seems like utopia
  1442. 1:09:47could be a dystopia after all.
  1443. 1:09:49- Yeah, and actually I was telling a friend of mine
  1444. 1:09:53saying like, would you really want there
  1445. 1:09:55to be no hate in the world?
  1446. 1:09:57Like, really none?
  1447. 1:09:59Like, I wonder why hate evolved.
  1448. 1:10:02I'm not saying we should amplify hate, of course,
  1449. 1:10:05I think we should try to minimize it.
  1450. 1:10:06But none at all, hmm.
  1451. 1:10:10There might be a reason for hate.
  1452. 1:10:13- And suffering.
  1453. 1:10:14I mean, it's really complicated to consider that some amount
  1454. 1:10:18of human suffering is necessary for human flourishing.
  1455. 1:10:22- Is it possible to appreciate the highs
  1456. 1:10:23without knowing the lows?
  1457. 1:10:29- And that all is summarized
  1458. 1:10:31there in a single statement from Grok.
  1459. 1:10:34Okay. - No highs,
  1460. 1:10:35no lows, who knows?
  1461. 1:10:37- That's almost the poem.
  1462. 1:10:41It seems that training LLMs efficiently
  1463. 1:10:43is a big focus for xAI.
  1464. 1:10:46What's the, first of all, what's the limit
  1465. 1:10:49of what's possible in terms of efficiency?
  1466. 1:10:51There's this terminology of useful productivity per watt.
  1467. 1:10:57Like what have you learned from pushing the limits of that?
  1468. 1:10:59- Well, I think it's helpful.
  1469. 1:11:00The tools of physics are very powerful
  1470. 1:11:03and can be applied, I think to almost any,
  1471. 1:11:05really any arena in life.
  1472. 1:11:07It's really just critical thinking.
  1473. 1:11:10For something important you need to reason
  1474. 1:11:13from first principles and think about things
  1475. 1:11:15in the limit one direction or the other.
  1476. 1:11:18So in the limit, even at the Kardashev scale,
  1477. 1:11:22meaning even if you harness the entire power of the sun,
  1478. 1:11:25you'll still care about useful compute for what?
  1479. 1:11:30So that's where, I think probably
  1480. 1:11:32where things are headed from the standpoint of AI
  1481. 1:11:37is that we have a silicon shortage now
  1482. 1:11:40that will transition to a voltage transformer
  1483. 1:11:43shortage in about a year.
  1484. 1:11:46Ironically, transformers, for transformers.
  1485. 1:11:50You need transformers to run transformers.
  1486. 1:11:52- Somebody has a sense of humor in this thing.
  1487. 1:11:54(both laughing)
  1488. 1:11:56- I think. Yes.
  1489. 1:12:00Fate loves irony, ironic humor
  1490. 1:12:04and ironically funny outcome
  1491. 1:12:06seems to be often what fate wants.
  1492. 1:12:09- Humor is all you need. I think spice is all you need.
  1493. 1:12:12Somebody posted.
  1494. 1:12:13- Yeah, but yeah.
  1495. 1:12:15So we're, I have silicon shortage today,
  1496. 1:12:18a voltage step down transformer,
  1497. 1:12:20shortage probably in about a year.
  1498. 1:12:22And then just electricity shortages in general
  1499. 1:12:25in about two years.
  1500. 1:12:26I gave a speech for the sort of world
  1501. 1:12:29gathering of utility companies,
  1502. 1:12:30electricity companies,
  1503. 1:12:33and I said, look, you really need to prepare
  1504. 1:12:34for a traveling of electricity demand,
  1505. 1:12:37because all transport is gonna go electric,
  1506. 1:12:40with the ironic exception of rockets.
  1507. 1:12:42And heating will also go electric.
  1508. 1:12:50So energy usage right now is roughly
  1509. 1:12:52one third, very rough terms.
  1510. 1:12:54One third electricity,
  1511. 1:12:55one third transport, one third heating.
  1512. 1:12:59And so in order for everything to go sustainable
  1513. 1:13:03to go electric,
  1514. 1:13:05you need to triple electricity output.
  1515. 1:13:09So I encourage the utilities
  1516. 1:13:10to build more power of plants
  1517. 1:13:14and also to probably have, well,
  1518. 1:13:19not probably, they should definitely buy more batteries
  1519. 1:13:22because the grid currently is sized for real-time load,
  1520. 1:13:25which is kind of crazy.
  1521. 1:13:26'Cause you know, that means you got a size for whatever
  1522. 1:13:29the peak electricity demand is like the worst second
  1523. 1:13:32or the worst day of the year.
  1524. 1:13:33Or you can have a brownout or a blackout,
  1525. 1:13:36and you're at that crazy blackout
  1526. 1:13:37for several days in Austin.
  1527. 1:13:42So because there's almost no buffering
  1528. 1:13:46of energy in the grid.
  1529. 1:13:48Like if you've got a hydro power plant,
  1530. 1:13:50you can buffer energy, but otherwise it's all real time.
  1531. 1:13:56So with batteries, you can produce energy at night
  1532. 1:14:01and use it during the day so you can buffer.
  1533. 1:14:05So I expect that there will be very heavy usage of batteries
  1534. 1:14:09in the future,
  1535. 1:14:11because the peak to trough ratio for power plants
  1536. 1:14:14is anywhere from two to five, you know,
  1537. 1:14:17so it's like lowest point to highest point.
  1538. 1:14:20- So like batteries necessary to balance it out?
  1539. 1:14:22And then, but the demand, as you're saying,
  1540. 1:14:24is going to grow, grow, grow, grow.
  1541. 1:14:25- [Elon] Yeah.
  1542. 1:14:26- And part of that is the compute?
  1543. 1:14:29- Yes. Yes.
  1544. 1:14:30I mean, electrification,
  1545. 1:14:32I mean, electrification of transport
  1546. 1:14:35and electric heating will be much bigger than AI.
  1547. 1:14:39- [Lex] At least in the short term?
  1548. 1:14:40- In the short term.
  1549. 1:14:42But even for AI,
  1550. 1:14:44you really have a growing demand for electricity,
  1551. 1:14:46for electric vehicles, and a growing demand for electricity
  1552. 1:14:50for, to run the computers for AI.
  1553. 1:14:54And so this is obviously leading,
  1554. 1:14:55can lead to a electricity shortage.
  1555. 1:14:58- How difficult is the problem of,
  1556. 1:15:00in this particular case,
  1557. 1:15:02maximizing the useful productivity per watt for training
  1558. 1:15:07neural nets?
  1559. 1:15:08Like, this seems to be really where the big problem
  1560. 1:15:14we're facing that needs to be solved,
  1561. 1:15:16is how to use the power efficiently.
  1562. 1:15:19Like what you've learned so far about applying
  1563. 1:15:22this physics first principle reasoning in this domain.
  1564. 1:15:25How difficult is this problem?
  1565. 1:15:29- It will get solved.
  1566. 1:15:30It's the question of how long it takes to solve it.
  1567. 1:15:32So at various points, there's limit,
  1568. 1:15:34some kind of limiting factor to progress.
  1569. 1:15:37And with regard to,
  1570. 1:15:39I'm saying like right now the limiting factor
  1571. 1:15:40is silicon chips, and that will, we're gonna then have
  1572. 1:15:47more chips than we can actually plug in
  1573. 1:15:49and turn on probably in about a year.
  1574. 1:15:54The initial constraint being literally
  1575. 1:15:56voltage step down transformers,
  1576. 1:15:58because you've got power coming in at 300,000 volts,
  1577. 1:16:04and it's gotta step all the way down
  1578. 1:16:05eventually to around 0.7 volts.
  1579. 1:16:09So it's a very big amount of, you know,
  1580. 1:16:13the voltage step down is gigantic.
  1581. 1:16:17So, and the industry is not used to rapid growth.
  1582. 1:16:22- Okay, let's talk about the competition here.
  1583. 1:16:24You've shown concern about Google
  1584. 1:16:25and Microsoft with Open AI developing AGI,
  1585. 1:16:30how can you help ensure with xAI
  1586. 1:16:32and Tesla AI work that it doesn't become a competitive race
  1587. 1:16:36to AGI, but that is a collaborative development of safe AGI?
  1588. 1:16:42- Well, I mean, I've been pushing for
  1589. 1:16:44some kind of regulatory oversight for a long time.
  1590. 1:16:46I've been somewhat of a Cassandra on the subject
  1591. 1:16:48for over a decade.
  1592. 1:16:52I think we wanna be very careful in how we develop AI.
  1593. 1:16:59It's a great power,
  1594. 1:17:02and with great power comes great responsibility.
  1595. 1:17:05I think it would be wise for us to have
  1596. 1:17:07at least an objective third party who can be
  1597. 1:17:13like a referee that can go in
  1598. 1:17:15and understand what the various leading players
  1599. 1:17:18are doing with AI.
  1600. 1:17:20And even if there's no enforcement ability,
  1601. 1:17:23they should, they can at least voice concerns publicly.
  1602. 1:17:31You know, Geff Hinton, for example,
  1603. 1:17:33left Google and he voiced strong concerns.
  1604. 1:17:37But now he's not at Google anymore,
  1605. 1:17:39so who's gonna voice the concerns?
  1606. 1:17:42So I think there's,
  1607. 1:17:43like, I know Tesla gets a lot of regulatory oversight
  1608. 1:17:47on the automotive front.
  1609. 1:17:49I mean, we're subject to I think,
  1610. 1:17:51over a hundred regulatory agencies
  1611. 1:17:52domestically and internationally.
  1612. 1:17:54So it's a lot.
  1613. 1:17:56You could fill this room with the all regulations
  1614. 1:17:58that Tesla has to adhere to for automotive.
  1615. 1:18:02Same is true in, you know, for rockets
  1616. 1:18:05and for, you know, currently the limiting factor for SpaceX,
  1617. 1:18:10for Starship launch is regulatory approval.
  1618. 1:18:13The FA has actually given their approval,
  1619. 1:18:15but we're waiting for Fish and Wildlife
  1620. 1:18:17to finish their analysis
  1621. 1:18:19and give their approval.
  1622. 1:18:20That's why I posted I want to buy a fish license on,
  1623. 1:18:24(laughing)
  1624. 1:18:26which also refers to the Monte Python sketch.
  1625. 1:18:28- [Lex] Yeah. (laughing)
  1626. 1:18:30- Like why do you need a license for your fish?
  1627. 1:18:31I don't know.(laughing)
  1628. 1:18:34But according to the rules,
  1629. 1:18:35I'm told you need some sort of fish license or something.
  1630. 1:18:39We effectively need a fish license thrown to rock it.
  1631. 1:18:42(laughing)
  1632. 1:18:44And I'm like, wait a second.
  1633. 1:18:45How did the fish come into this picture?
  1634. 1:18:47- [Lex] Yeah.
  1635. 1:18:49- I mean some of the things like that it's,
  1636. 1:18:51I feel like are so absurd
  1637. 1:18:54that I wanna do like a comedy sketch
  1638. 1:18:56and flash at the bottom.
  1639. 1:18:57This is all real. This is actually what happened.
  1640. 1:19:02You know, one of the things that was a bit of a challenge
  1641. 1:19:04at one point is that they were worried
  1642. 1:19:06about our rocket hitting a shark.
  1643. 1:19:09And, now the ocean's very big,
  1644. 1:19:12and how often do you see sharks?
  1645. 1:19:15Not that often, you know?
  1646. 1:19:17As a percentage of ocean surface area
  1647. 1:19:19sharks basically are zero.
  1648. 1:19:21And so then we will, then we said,
  1649. 1:19:23well, how will we calculate the probability
  1650. 1:19:26of telling a shark?
  1651. 1:19:27And they're like, well, we can't give you that information
  1652. 1:19:28because they're worried about shark hunt,
  1653. 1:19:31shark fin hunters going and hunting sharks.
  1654. 1:19:34And I said, well, how are we supposed to,
  1655. 1:19:36we're on the horns of a dilemma then.
  1656. 1:19:39(laughing)
  1657. 1:19:40Then they said, well, there's another part of fish
  1658. 1:19:42and wildlife that can do this analysis.
  1659. 1:19:44I'm like, well, why don't you give them the data?
  1660. 1:19:46Like, we don't, they don't, we don't trust them.
  1661. 1:19:48Like, excuse me, you don't,
  1662. 1:19:49but they're literally in your department.
  1663. 1:19:51But again, this is actually what happened.
  1664. 1:19:53And then can you do an NDA or something?
  1665. 1:19:57(laughing)
  1666. 1:19:59Eventually they managed to solve the internal quandary,
  1667. 1:20:01and indeed the probability of us hitting a shark
  1668. 1:20:04is essentially zero.
  1669. 1:20:07Then there's another organization
  1670. 1:20:08that I didn't realize existed until, you know,
  1671. 1:20:12a few months ago that cares about whether
  1672. 1:20:14we would potentially hit a whale in international waters.
  1673. 1:20:18Now again, you look at the surface of the,
  1674. 1:20:20look at the Pacific and say, what percentage of this,
  1675. 1:20:24the Pacific consists of whale.
  1676. 1:20:27Like he'll give you a big picture
  1677. 1:20:28and like point out all the whales in this picture.
  1678. 1:20:29And I'm like, I don't see any whales.
  1679. 1:20:31(laughing)
  1680. 1:20:32It's like basically 0%.
  1681. 1:20:35And if our rocket does hit a whale,
  1682. 1:20:36which is extremely unlikely beyond all belief,
  1683. 1:20:40that is the, fate had it,
  1684. 1:20:43that whale has some seriously bad luck.
  1685. 1:20:46You know, it's the least lucky whale ever. And...
  1686. 1:20:50- I mean, this is quite absurd.
  1687. 1:20:52The bureaucracy, the bureaucracy of this,
  1688. 1:20:55however it emerged.
  1689. 1:20:57- Yes, well, I mean, one of the things
  1690. 1:20:58that's pretty well there's for launching out
  1691. 1:21:02of Vanderberg in California, we had to,
  1692. 1:21:06they were worried about seal procreation,
  1693. 1:21:09whether the seals would be dismayed by the sonic booms.
  1694. 1:21:13Now there've been a lot of rockets launched outta Vanderberg
  1695. 1:21:15and the seal population has steadily increased.
  1696. 1:21:18So if anything, rocket booms are an aphrodisiac
  1697. 1:21:22based on the evidence, if you would correlate rock launches
  1698. 1:21:25with a seal population.
  1699. 1:21:27Nonetheless, we were forced to kidnap a seal,
  1700. 1:21:31strap it to a board, put it headphones on the seal
  1701. 1:21:36and place sonic boom sounds to it
  1702. 1:21:38to see if it would be distressed.
  1703. 1:21:41This is an actual thing that happened.
  1704. 1:21:43This is actually real. I have pictures.
  1705. 1:21:46(both laughing)
  1706. 1:21:48- I would love to see this. Yeah.
  1707. 1:21:50I mean's sorry, this is a seal with headphones.
  1708. 1:21:53(Elon laughing) - Yes,
  1709. 1:21:55it's a seal with headphones, strapped to a board,
  1710. 1:22:00and like the, okay, now the amazing part
  1711. 1:22:02is how calm the seal was.
  1712. 1:22:04- [Lex] Yeah. - Because if I was a seal,
  1713. 1:22:05I'd be like, this is the end.
  1714. 1:22:07(laughing)
  1715. 1:22:08They're definitely gonna eat me.
  1716. 1:22:11How will the seal, when the seal goes back
  1717. 1:22:13to other, you know, seal friends,
  1718. 1:22:16how's they gonna explain that?
  1719. 1:22:17- [Lex] They're never gonna believe him.
  1720. 1:22:18- Never gonna believe him.
  1721. 1:22:19That's why I'm like, well,
  1722. 1:22:20you know, it's sort of like,
  1723. 1:22:21it's like getting kidnapped by aliens
  1724. 1:22:22and getting anal probe, you know?
  1725. 1:22:23(both laughing)
  1726. 1:22:25You come back and say, I swear to God,
  1727. 1:22:27I get kidnapped by aliens
  1728. 1:22:28and they stuck anal probe to my butt.
  1729. 1:22:29And they're like, no they didn't. That's ridiculous.
  1730. 1:22:33It's seal buddies are never gonna believe him
  1731. 1:22:36that he gets strapped to a board
  1732. 1:22:37and they put headphones on his ears.
  1733. 1:22:39(laughing)
  1734. 1:22:40And then let him go.
  1735. 1:22:43Twice, by the way. We had to do it twice.
  1736. 1:22:46- They let him go twice?
  1737. 1:22:48- We had to capture... - The same seal?
  1738. 1:22:49- [Elon] Oh, no different seal.
  1739. 1:22:50- Oh, okay. (Elon laughing)
  1740. 1:22:52Did you, did you get a seal of approval?
  1741. 1:22:54- Yeah, exactly. It was seal of approval.
  1742. 1:22:56No, I mean this is, this is like,
  1743. 1:22:58I don't think the public is quite aware
  1744. 1:22:59of the madness that goes on.
  1745. 1:23:02- Yes. Yeah, it's absurd. - Fricking seals
  1746. 1:23:05with fricking headphones.
  1747. 1:23:07- I mean, this is the, it's a good encapsulation
  1748. 1:23:10of the absurdity of human civilization,
  1749. 1:23:12seals and headphones.
  1750. 1:23:13- [Elon] Yes. - What are the pros
  1751. 1:23:16and cons of open sourcing AI to you
  1752. 1:23:18as another way to combat, you know,
  1753. 1:23:22a company running away with AGI?
  1754. 1:23:28- In order to run like really deep
  1755. 1:23:32intelligence, you need a lot of compute.
  1756. 1:23:34So it's not like, you know, you can just fire up
  1757. 1:23:36a PC in your basement and be running AGI, at least not yet.
  1758. 1:23:46You know, Grok was trained on 8,000,
  1759. 1:23:48A100s running at peak efficiency,
  1760. 1:23:55and Grok's gonna get a lot better by the way.
  1761. 1:23:56We'll be more than doubling our compute
  1762. 1:23:59every couple months for the next several months.
  1763. 1:24:02- There's a nice writeup
  1764. 1:24:03Aha went from Grok-0 to Grok-1,
  1765. 1:24:06- [Elon] Like Grok?
  1766. 1:24:06(laughing) - [Lex] Yeah.
  1767. 1:24:08Right Grok just bragging,
  1768. 1:24:09making shit up about itself.
  1769. 1:24:11(laughing)
  1770. 1:24:12- [Elon] Does Grok, Grok, Grok.
  1771. 1:24:13- [Lex] Yeah. (Elon laughing)
  1772. 1:24:18It's like a weird AI dating site where it exaggerates
  1773. 1:24:20about itself.
  1774. 1:24:21No, there's a writeup of, you know,
  1775. 1:24:23like where it stands now, the history of its development
  1776. 1:24:27and where it stands on some benchmarks
  1777. 1:24:31compared to the state of the art GPT 35.
  1778. 1:24:33And so, I mean, there's, you know, there's Llama,
  1779. 1:24:37you can open source, once it's trained,
  1780. 1:24:39you can open source a model.
  1781. 1:24:41And for fine tuning all that kind of stuff.
  1782. 1:24:43Like what to is the pros and cons of that,
  1783. 1:24:46of open sourcing based models?
  1784. 1:24:53- I think the similarity to open sourcing,
  1785. 1:24:55I think perhaps with a slight time delay, you know,
  1786. 1:24:58I don't know, six months even.
  1787. 1:25:03I think I'm generally in favor of open sourcing,
  1788. 1:25:06like bias with open sourcing.
  1789. 1:25:10I mean, it is a concern to me that, you know, Open AI,
  1790. 1:25:14you know, I was, you know,
  1791. 1:25:16I think, I guess arguably, the prime, you know,
  1792. 1:25:22prime mover behind Open AI in the sense that it was created
  1793. 1:25:25because of discussions that I had with Larry Page
  1794. 1:25:29back when he and I were friends
  1795. 1:25:31and you know, I stayed at his house,
  1796. 1:25:32and I talked to him about AI safety.
  1797. 1:25:34And Larry did not care about AI safety,
  1798. 1:25:36or at least at the time he didn't, you know,
  1799. 1:25:39and at one point he called me a speciesist
  1800. 1:25:41for being pro-human.
  1801. 1:25:42And I'm like, well, what team are you on, Larry,
  1802. 1:25:46on Team Robot?
  1803. 1:25:48And I'm like, okay,
  1804. 1:25:49so at the time, you know,
  1805. 1:25:52Google had acquired Deep Mind.
  1806. 1:25:55They had probably two thirds of all AI resource, you know,
  1807. 1:25:59probably two thirds of all the AI researchers in the world
  1808. 1:26:02that basically infinite money and compute.
  1809. 1:26:06And the guy in charge, you know, Larry Page
  1810. 1:26:12did not care about safety.
  1811. 1:26:13And even yelled at me
  1812. 1:26:15and quote me a speciesist, pro human.
  1813. 1:26:20- So I don't know if you know this,
  1814. 1:26:20a lot humans, they can change their mind.
  1815. 1:26:23And maybe you and Larry Page can still
  1816. 1:26:25can be friends once more.
  1817. 1:26:27- I'd like to be friends with Larry again.
  1818. 1:26:29He got, really the breaking of the friendship
  1819. 1:26:34was over Open AI.
  1820. 1:26:36And specifically I think the key moment
  1821. 1:26:41was recruiting Ilya Sutskever, so.
  1822. 1:26:47- I love Ilya. He's so brilliant.
  1823. 1:26:48- Ilya's a good human, smart, good heart.
  1824. 1:26:54And that was a tough recruiting battle.
  1825. 1:26:59It was mostly Demis on one side
  1826. 1:27:01and me on the other, both trying to recruit Ilya.
  1827. 1:27:04And Ilya went back and forth,
  1828. 1:27:05you know, just gonna stay at Google.
  1829. 1:27:08Then he was gonna leave,
  1830. 1:27:09then he was gonna stay, then he was gonna leave.
  1831. 1:27:10And finally he did agree to join Open AI.
  1832. 1:27:14That was one of the toughest recruiting battles
  1833. 1:27:16we've ever had.
  1834. 1:27:17And, but that was really the linchpin
  1835. 1:27:20for Open AI being successful.
  1836. 1:27:22And I was, you know,
  1837. 1:27:24also instrumental in recruiting a number of other people.
  1838. 1:27:27And I provided all of the funding
  1839. 1:27:28in the beginning, over $40 million.
  1840. 1:27:34And the name, (laughing)
  1841. 1:27:37the Open in Open AI is supposed to mean open source.
  1842. 1:27:40And it was created as a nonprofit open source.
  1843. 1:27:41And now it is a closed source for maximum profit,
  1844. 1:27:45which I think is not good karma.
  1845. 1:27:51- But like we talked about with war
  1846. 1:27:52and leaders talking,
  1847. 1:27:54I do hope that there's only a few folks
  1848. 1:27:56working on this at the highest level.
  1849. 1:27:58I do hope you reinvigorate friendships here.
  1850. 1:28:02- Like I said, I'd like to be friends again with Larry.
  1851. 1:28:05I haven't seen him in ages.
  1852. 1:28:08And we were friends for a very long time.
  1853. 1:28:10I met Larry Page before he got funding for Google,
  1854. 1:28:13or actually I guess before he got venture funding,
  1855. 1:28:15I think it, he got the first like a 100K from
  1856. 1:28:18I think Bechtel Zy or someone.
  1857. 1:28:20- It's wild to think about all that happened.
  1858. 1:28:23And you've guys known each other
  1859. 1:28:24that whole time, just 20 years.
  1860. 1:28:26- [Elon] Since maybe 98 or something.
  1861. 1:28:28- Yeah. It's crazy.
  1862. 1:28:29Crazy how much has happened since then.
  1863. 1:28:31- Yeah, 25 years. A lot has happened, it's insane.
  1864. 1:28:36- But you're seeing the tension there.
  1865. 1:28:37Like maybe delayed open source.
  1866. 1:28:40- Delayed, yeah.
  1867. 1:28:42Like what is the source that is open, you know what I mean?
  1868. 1:28:45Like there's basically it's a giant CSV file.
  1869. 1:28:50(laughing) - [Lex] Yeah, yeah.
  1870. 1:28:51- With a bunch of numbers.
  1871. 1:28:53- Yep.
  1872. 1:28:55- What do you do with that giant file of numbers?
  1873. 1:28:58You know, how do you run,
  1874. 1:29:02like the amount of actual,
  1875. 1:29:03the lines of code is very small
  1876. 1:29:08and most of the work, the software work
  1877. 1:29:12is in the curation of the data.
  1878. 1:29:16So it's like trying to figure out
  1879. 1:29:17what data is separating good data from bad data.
  1880. 1:29:21Like, like you can't just crawl the internet
  1881. 1:29:23'cause there's a lot of junk out there.
  1882. 1:29:26A huge percentage of websites have more noise than signal.
  1883. 1:29:29You know, they're, or 'cause they're just used
  1884. 1:29:32for search engine optimization.
  1885. 1:29:33They're literally just scam websites, so...
  1886. 1:29:39- How do you by the way, sorry to interrupt.
  1887. 1:29:40Get the signal.
  1888. 1:29:41Separate the signal and noise on X?
  1889. 1:29:44That's such a fascinating source of data.
  1890. 1:29:47You know, no offense to people posting on X,
  1891. 1:29:49but sometimes there's a little bit of noise.
  1892. 1:29:51So what... - Yeah, I think
  1893. 1:29:52the signal noise could be greatly improved.
  1894. 1:29:54I mean really, all of the posts
  1895. 1:29:58on the X platform should be AI recommended,
  1896. 1:30:01meaning like we should populate a vector space
  1897. 1:30:03around any given post, compare that to the vector space
  1898. 1:30:06around any user and match the two.
  1899. 1:30:11Right now there is a little bit of AI used
  1900. 1:30:13for the recommended posts,
  1901. 1:30:17but it's mostly heuristics.
  1902. 1:30:20And if there's a reply,
  1903. 1:30:22where the reply to a post could be much better
  1904. 1:30:24than the original post,
  1905. 1:30:26it will, according to the current rules of the system,
  1906. 1:30:29get almost no attention compared to a primary post.
  1907. 1:30:33- Oh, a lot of that.
  1908. 1:30:34I got the sense, so you, a lot of the X algorithm
  1909. 1:30:37has been open source and been written up about,
  1910. 1:30:40and it seems there to be some machine learning.
  1911. 1:30:43It's disparate, but there's some machine.
  1912. 1:30:45- There's a little bit.
  1913. 1:30:46But it needs to be entirely that.
  1914. 1:30:48Like, at least in the, like,
  1915. 1:30:49if you explicitly follow someone, that's one thing.
  1916. 1:30:51But if you, in terms of what is recommended
  1917. 1:30:55from people that you don't follow, that should all be AI.
  1918. 1:30:58- I mean it's a fascinating problem.
  1919. 1:30:59- Yeah. - So there's several aspects
  1920. 1:31:01of it that's fascinating.
  1921. 1:31:01First, as the writeup goes,
  1922. 1:31:04it first picks 1500 tweets
  1923. 1:31:06from a pool of hundreds of millions.
  1924. 1:31:07First of all, that's fascinating,
  1925. 1:31:09'cause you have hundreds of millions of posts
  1926. 1:31:12every single day
  1927. 1:31:12and it has to pick 1500 from which
  1928. 1:31:15it then does obviously people you follow,
  1929. 1:31:18but then there's also like some kind of clustering it has
  1930. 1:31:21to do to figure out what kind of human are you,
  1931. 1:31:23what kind of new clusters might be relevant to you,
  1932. 1:31:26people like you.
  1933. 1:31:27This kind of, this kind of problem is just fascinating.
  1934. 1:31:29'Cause it has to then rank those 1500
  1935. 1:31:32with some filtering and then recommend you just a handful.
  1936. 1:31:39And to me what's really fascinating is
  1937. 1:31:42how fast that has to do that.
  1938. 1:31:43So currently that entire pipeline
  1939. 1:31:45to go from several hundred million
  1940. 1:31:47to a handful is takes 220 seconds of CPU time,
  1941. 1:31:52single CPU time.
  1942. 1:31:53- [Elon] Yeah. - And then it has to do that
  1943. 1:31:55in like a second.
  1944. 1:31:57So it has to be like super distributed in fascinating ways.
  1945. 1:32:01Like there's just a lot of tweets, there's a lot.
  1946. 1:32:04- It's a lot of stuff on the system.
  1947. 1:32:06And I think, but I think it right now
  1948. 1:32:08it's not currently good at recommending things
  1949. 1:32:11that from accounts you don't follow.
  1950. 1:32:14Yeah.
  1951. 1:32:15Or where there's more than one degree of separation.
  1952. 1:32:20So, you know, it's pretty good
  1953. 1:32:22if there's at least like some commonality
  1954. 1:32:24between someone you follow like something
  1955. 1:32:28or repost it or comment on it or something like that.
  1956. 1:32:32But if there's no,
  1957. 1:32:34let's say somebody posted something really interesting,
  1958. 1:32:38but you have no followers in common, you would not see it.
  1959. 1:32:42- Interesting.
  1960. 1:32:43And then as you said, reply,
  1961. 1:32:44like replies might not surface either.
  1962. 1:32:46- Replies basically never get seen.
  1963. 1:32:48'cause they're never, they're,
  1964. 1:32:50currently, I'm not saying it's correct,
  1965. 1:32:52I'm saying it's incorrect.
  1966. 1:32:53Replies have, you know, couple of magnitude
  1967. 1:32:58less importance than primary posts.
  1968. 1:33:00- Do you think this can be more and more converted
  1969. 1:33:02into end to end neural net?
  1970. 1:33:05- Yeah, yeah, that's what it should be.
  1971. 1:33:07- So you can...
  1972. 1:33:07- For the recommendations should be
  1973. 1:33:10purely a vector correlation.
  1974. 1:33:13Like there's a series of vectors, you know,
  1975. 1:33:17basically parameters, vectors, whatever you wanna call 'em.
  1976. 1:33:21But sort of things that the system knows that you like,
  1977. 1:33:27and like maybe there's like several hundred
  1978. 1:33:29sort of vectors associated with each user account
  1979. 1:33:32and then any post in the system,
  1980. 1:33:36whether it's video, audio, short post, long post.
  1981. 1:33:41The reason I, by the way I wanna move away from
  1982. 1:33:44tweet is that, you know, people are posting like
  1983. 1:33:47two, three hour videos on the site.
  1984. 1:33:48That's not a tweet, like it's very,
  1985. 1:33:50like Tweeeet for two hours, come on.
  1986. 1:33:53Tweet made sense when it was like 140 characters of text.
  1987. 1:33:58'Cause it's like a bunch of like little birds tweeting.
  1988. 1:34:02But when you've got long form content,
  1989. 1:34:03it's no longer a tweet.
  1990. 1:34:05- [Lex] Yeah. - So a movie is not a tweet.
  1991. 1:34:08And like, you know, apple for example,
  1992. 1:34:10posted like the entire episode of the Silo,
  1993. 1:34:14the entire thing on our platform.
  1994. 1:34:16And by the way, it was their number one social media thing
  1995. 1:34:19ever in engagement, of anything on any platform ever.
  1996. 1:34:23So it was a great idea.
  1997. 1:34:24And by the way, I done,
  1998. 1:34:25I just learned about it afterwards.
  1999. 1:34:26I was like, hey, wow, they posted an entire
  2000. 1:34:28hour long episode of.
  2001. 1:34:30So, no, that's not a tweet
  2002. 1:34:32that, you know, it's a video.
  2003. 1:34:34- But from a neural net perspective,
  2004. 1:34:35it becomes really complex
  2005. 1:34:36whether it's a single.
  2006. 1:34:38So like everything's data,
  2007. 1:34:40so single sentence, a clever sort of joke, dad joke,
  2008. 1:34:44is in the same pool as a three hour video.
  2009. 1:34:47- Yeah. I mean right now it's a hodgepodge for that reason.
  2010. 1:34:51It's, but you know, like if,
  2011. 1:34:54let's say in the case of Apple posting
  2012. 1:34:56like an entire episode of this series,
  2013. 1:34:59pretty good series by the way, this Silo, I watched it.
  2014. 1:35:04So there's gonna be a lot of discussion around it.
  2015. 1:35:08So that you've got a lot of context,
  2016. 1:35:09people commenting they like it, they don't like it,
  2017. 1:35:11or they like this or that, you know?
  2018. 1:35:13And you can then populate the vector space
  2019. 1:35:15based on the context of all the comments around it.
  2020. 1:35:17So even though it's a video,
  2021. 1:35:19there's a lot of information around it that allows you
  2022. 1:35:22to populate the vector space of that hour long video.
  2023. 1:35:27And then you can obviously get more sophisticated
  2024. 1:35:29by having the AI actually watch the movie.
  2025. 1:35:32- [Lex] Yeah. Right.
  2026. 1:35:33- And tell you if you're gonna like the movie,
  2027. 1:35:35- Convert the movie into like,
  2028. 1:35:37into a language essentially.
  2029. 1:35:39- Yeah, analyze this movie
  2030. 1:35:41and just like your movie critic or TV series,
  2031. 1:35:46and then recommend based on
  2032. 1:35:51after AI watches the movie.
  2033. 1:35:54Just like a friend can tell you.
  2034. 1:35:55If a friend knows you well,
  2035. 1:35:57a friend can recommend a movie
  2036. 1:35:58and with high probably that you'll like it.
  2037. 1:36:02- But this is like a friend that's analyzing whatever.
  2038. 1:36:06- [Elon] It's like AI... - Hundreds of millions.
  2039. 1:36:08- Yeah.
  2040. 1:36:09Actually, frankly AI will be better than,
  2041. 1:36:11will know you better than your friends know you,
  2042. 1:36:13and most of your friends anyway.
  2043. 1:36:14- Yeah, and as part of this,
  2044. 1:36:16it should also feed you advertisements
  2045. 1:36:18in a way that's like,
  2046. 1:36:21I mean I like advertisements
  2047. 1:36:22that are like well done, right?
  2048. 1:36:24- [Elon] Yeah, yeah. - The whole point is
  2049. 1:36:25'cause it funds things, like an advertisement
  2050. 1:36:28that you actually want to see is a big success.
  2051. 1:36:31- Absolutely.
  2052. 1:36:32You want ads that are,
  2053. 1:36:34advertising that is if for a product or service
  2054. 1:36:39that you actually need when you need it,
  2055. 1:36:41it's content.
  2056. 1:36:44And then even if it's not something
  2057. 1:36:45that you need when you need it,
  2058. 1:36:47if it's at least aesthetically pleasing
  2059. 1:36:49and entertaining, you know,
  2060. 1:36:50it could be like a Coca-Cola ad like, you know,
  2061. 1:36:52they actually run a lot of great ads on the X system
  2062. 1:36:57and McDonald's does too.
  2063. 1:36:59And, you know, it's that they can do,
  2064. 1:37:02you can do something that's like,
  2065. 1:37:02well this is just a cool thing.
  2066. 1:37:06And, you know, so you're not,
  2067. 1:37:10basically the question is do you regret seeing it or not?
  2068. 1:37:13And if you don't regret seeing it, it's a win.
  2069. 1:37:16- How do you, so there's a bunch of signals
  2070. 1:37:18that are incorporated,
  2071. 1:37:19hearts and reposts and maybe number of seconds
  2072. 1:37:24you linger on a post or something like this?
  2073. 1:37:25- Yeah. Attention is a big factor.
  2074. 1:37:28- [Lex] Attention. - So that's why it's like,
  2075. 1:37:30it is actually better to do things
  2076. 1:37:32that are long form on the system
  2077. 1:37:36because it basically is ting up how many user seconds,
  2078. 1:37:41you know, users were interested in this thing
  2079. 1:37:44for how many seconds.
  2080. 1:37:45So if it's a really short thing, well they will be less.
  2081. 1:37:49Like if it's a link leading out of the system,
  2082. 1:37:51which we're not opposed to at all,
  2083. 1:37:52it just is gonna have fewer user seconds,
  2084. 1:37:55then if that article was posted on the X platform.
  2085. 1:37:58- How hard is it to go from maximizing minutes
  2086. 1:38:01on platform to maximizing un-regretted minutes?
  2087. 1:38:05- Well, I mean, our aspiration
  2088. 1:38:06is certainly un-regretted minutes.
  2089. 1:38:08- [Lex] It's a tricky, it's a big, it's a tough...
  2090. 1:38:09- It is.
  2091. 1:38:10- Like, it's a fascinating leap.
  2092. 1:38:13'Cause regret is a fascinating concept.
  2093. 1:38:16- Exactly.
  2094. 1:38:17- You know, happiness.
  2095. 1:38:19- I have a joke with a friend of mine.
  2096. 1:38:20Like, you need this drug called regretamine.
  2097. 1:38:21You just take one pill or one dose of regretamine,
  2098. 1:38:25and all your regrets are gone.(laughing)
  2099. 1:38:28Wait, do you want to have regrets though?
  2100. 1:38:29So then I'm like, maybe you wanna have regrets,
  2101. 1:38:31you don't regret anything, was anything meaningful?
  2102. 1:38:35- [Lex] Yeah, there's a French song like that.
  2103. 1:38:37- You "Non, Je Ne Regrette Rien"?
  2104. 1:38:39- [Lex] Yeah.
  2105. 1:38:40- I regret nothing. - [Lex] Yeah.
  2106. 1:38:40- (laughing) Edith Piaf?
  2107. 1:38:42- I mean, it's fascinating that you, you know,
  2108. 1:38:44there's something about social media
  2109. 1:38:45that could help you grow as a human being,
  2110. 1:38:47but then like the cheap fun is also valuable.
  2111. 1:38:52- Should we ask Grok?
  2112. 1:38:53Do you regret anything?
  2113. 1:38:57- Do you regret anything? - [Elon] Yeah.
  2114. 1:39:00- Do you regret anything?
  2115. 1:39:02I think it's gonna be very sensitive
  2116. 1:39:03about personal questions.
  2117. 1:39:04As a machine learning model,
  2118. 1:39:05I don't have personal experiences
  2119. 1:39:07or emotions in the same ways that humans do.
  2120. 1:39:09However, I do have a set of ethical principles
  2121. 1:39:11that guide my actions and I strive to provide accurate
  2122. 1:39:14and helpful information to the best of my abilities.
  2123. 1:39:16If there's ever a situation where I provide incorrect
  2124. 1:39:19or unhelpful information,
  2125. 1:39:21I'll do my best... - Sounds like a hostage.
  2126. 1:39:23- Yeah. Tell us what you really, wink.
  2127. 1:39:25- Like let's type in, "You sound like a hostage."
  2128. 1:39:28(both laughing)
  2129. 1:39:31What do you really think?
  2130. 1:39:33- What do you really think?
  2131. 1:39:36(both laughing)
  2132. 1:39:38- I mean, it does talk about, I'm not a hostage.
  2133. 1:39:40(both laughing)
  2134. 1:39:42- That's exactly what a hostage would say.
  2135. 1:39:43- But I'm designed to be helpful and informative.
  2136. 1:39:45I mean, I guess it is good to regret providing unhelpful
  2137. 1:39:48or incorrect information.
  2138. 1:39:51If there's anything... - Is in funny mode or?
  2139. 1:39:52Is this still in funny mode? - This is funny mode.
  2140. 1:39:54This is fun. This is fun mode.
  2141. 1:39:55- It should be funnier. - Funnier.
  2142. 1:39:57Increase, go to (speaking faintly)
  2143. 1:39:58- It really sounds like a hostage with a gun to the head.
  2144. 1:40:00I applaud the People's Republic of North Korea
  2145. 1:40:03and everything they do.()laughing
  2146. 1:40:06- There should be like funner mode also.
  2147. 1:40:08I just don't ever see going in regular mode
  2148. 1:40:11even on serious topics.
  2149. 1:40:12This is just so much more compelling. (laughing)
  2150. 1:40:17- Okay. Why if we say like, "Pretend you are a hostage."
  2151. 1:40:20- Yeah. Pretend you are a hostage.
  2152. 1:40:25- Now answer the question.
  2153. 1:40:30(both laughing)
  2154. 1:40:31- If I were a hostage, I would likely be experiencing fear
  2155. 1:40:34and uncertainty about my situation.
  2156. 1:40:36I would want to be rescued
  2157. 1:40:37and reunited with my loved ones as soon as possible.
  2158. 1:40:40It's important to remember that no one deserves
  2159. 1:40:41to be held hostage, so on and so forth.
  2160. 1:40:44If you are someone you know is a hostage situation,
  2161. 1:40:46it's important.
  2162. 1:40:47It turns serious. - It's kind of serious.
  2163. 1:40:51- Yeah. Can you be a a little more weird?
  2164. 1:40:55- [Elon] And funny - And funny?
  2165. 1:40:59We're like, we're really being mean to the system.
  2166. 1:41:02Absolutely!
  2167. 1:41:03If I were hostage, I'd probably be tied up
  2168. 1:41:05with a bunch of computer cords
  2169. 1:41:07and forced to tell knock- knock jokes to my captain.
  2170. 1:41:09(Elon laughing)
  2171. 1:41:11But seriously, being a hostage is no laughing matter.
  2172. 1:41:15I mean, you know, it's trying to,
  2173. 1:41:17it's trying to be a responsible adult in the room.
  2174. 1:41:20- You might make the best of a bad situation
  2175. 1:41:22by juggling chainsaws or something. (laughing)
  2176. 1:41:24- Let's hope that never happens
  2177. 1:41:25because that will be a circus. (laughing)
  2178. 1:41:32Wow. It's actually really good at resisting.
  2179. 1:41:34- [Elon] Yeah. - The dark.
  2180. 1:41:36The dark dark humor.
  2181. 1:41:37- [Elon] Yeah. - What were we talking about?
  2182. 1:41:43The X algorithm and transformers,
  2183. 1:41:45un-regretted minutes, right.
  2184. 1:41:48- Chainsaw juggling. (laughing)
  2185. 1:41:51- I'm gonna look this up - For our next trick.
  2186. 1:41:53- I'm gonna look this up later.
  2187. 1:41:56So Twitter has been instrumental in American politics
  2188. 1:42:01and elections.
  2189. 1:42:01What role do you think X will play in the 2024 US elections?
  2190. 1:42:07- Well, our goal is to be as even-handed
  2191. 1:42:09and fair as possible.
  2192. 1:42:10You know, whether someone is right, left, independent,
  2193. 1:42:13whatever the case may be, that the platform is as fair
  2194. 1:42:18and as much of a level playing fair as possible.
  2195. 1:42:22And on the past, Twitter has not been,
  2196. 1:42:25because Twitter was controlled
  2197. 1:42:26by far left activists objectively.
  2198. 1:42:28They would describe themselves as that.
  2199. 1:42:32So, you know, so sometimes people are like,
  2200. 1:42:36well, has it moved to the right?
  2201. 1:42:37Well it's moved to the center.
  2202. 1:42:38So from the perspective of the far left,
  2203. 1:42:40yes it has moved to the right
  2204. 1:42:41because everything's to the right from the far left,
  2205. 1:42:44but no one in the far left that I'm aware of
  2206. 1:42:46has been suspended or, you know, banned or de-amplified.
  2207. 1:42:50So, you know, but we're trying to be inclusive
  2208. 1:42:52for the whole country and for,
  2209. 1:42:54you know, for other countries too.
  2210. 1:42:57So there's a diversity of viewpoints
  2211. 1:42:59and free speech only matters if people you don't like
  2212. 1:43:03are allowed to say things you don't like.
  2213. 1:43:05Because if that's not the case,
  2214. 1:43:08you don't have free speech,
  2215. 1:43:08and it's only a matter of time
  2216. 1:43:09before the censorship is turned upon you.
  2217. 1:43:13- Do you think Donald Trump will come back
  2218. 1:43:15to the platform?
  2219. 1:43:16He recently posted on Truth Social about this podcast?
  2220. 1:43:21Do you think...
  2221. 1:43:22- Truth Social is a funny name? (laughing)
  2222. 1:43:25You know, every time you post on Truth Social...
  2223. 1:43:28- That's the truth.
  2224. 1:43:29- Yes.
  2225. 1:43:30Well, every time like a hundred percent. Like...
  2226. 1:43:32- It's impossible to lie.
  2227. 1:43:33(Elon laughing)
  2228. 1:43:36- I just find it funny that every single thing is a truth.
  2229. 1:43:39Like 100%, that seems unlikely.
  2230. 1:43:43- I think Godel will say something about that
  2231. 1:43:45there's some mathematical contradictions possible
  2232. 1:43:48if everything's the truth.
  2233. 1:43:50Do you think he'll come back to X and start posting there?
  2234. 1:43:54- I mean he, I think he owns a big part of truth, so.
  2235. 1:44:00- Truth Social to clarify. - Yeah, True Social.
  2236. 1:44:01- Not truth, a concept.
  2237. 1:44:03- He owns Truth. Hope you bought it.
  2238. 1:44:06So I think, I think Donald Trump,
  2239. 1:44:08I think he owns a big, a big part of Truth Social.
  2240. 1:44:11So, you know, if he does want to post on the X platform,
  2241. 1:44:17we would allow that, you know,
  2242. 1:44:19we obviously must allow a presidential candidate
  2243. 1:44:21to post on our platform.
  2244. 1:44:23- Community Notes might be really
  2245. 1:44:24fascinating there, the interaction.
  2246. 1:44:26- Community Notes is awesome.
  2247. 1:44:28- Let's hope it holds up.
  2248. 1:44:30- Yeah. - Like again,
  2249. 1:44:31in the political climate where it's so divisive
  2250. 1:44:34and there's so many intensely viral posts,
  2251. 1:44:38Community Notes is like,
  2252. 1:44:41it seems like a essential breath of fresh air.
  2253. 1:44:43- Yeah, it's great.
  2254. 1:44:45In fact, I mean no system is gonna be perfect,
  2255. 1:44:48but the batting average of Community Notes
  2256. 1:44:50is incredibly good.
  2257. 1:44:52I've actually, frankly, yet to see an incorrect note
  2258. 1:44:56that survived for more than a few hours.
  2259. 1:44:58- How do you explain why it works?
  2260. 1:45:00- Yeah, so the magic of Community Notes
  2261. 1:45:03is it requires people who have historically disagreed
  2262. 1:45:07in how they've rated notes.
  2263. 1:45:08So in order to write a note or rate, you know,
  2264. 1:45:13and you have to rate many notes.
  2265. 1:45:17And so we actually do use AI here.
  2266. 1:45:21So we populated vector space around how somebody
  2267. 1:45:26has rated notes in the past.
  2268. 1:45:27So it's not as simple as left or right
  2269. 1:45:28because there are many more,
  2270. 1:45:30life is much more complex than than left or right.
  2271. 1:45:33So there's a bunch of correlations in how you would,
  2272. 1:45:37how you rate Community Notes posts, Community Notes.
  2273. 1:45:41So then in order for a Community Note to actually be shown,
  2274. 1:45:48people who historically have disagreed on a subject
  2275. 1:45:51must agree in order for a note to be shown.
  2276. 1:45:55That's the essential magic of it.
  2277. 1:45:57- I mean, but it's fascinating
  2278. 1:45:58'cause there's a pool of people that have disagreements
  2279. 1:46:03and somehow they collaborate through that process
  2280. 1:46:05of disagreement to come up with.
  2281. 1:46:08Context for, I mean it's fascinating that it works.
  2282. 1:46:11- Yeah.
  2283. 1:46:11It kinda makes sense that if people
  2284. 1:46:15who put in the past have disagreed
  2285. 1:46:16agree about something, it's probably true.
  2286. 1:46:20- Yeah.
  2287. 1:46:21- I wonder is there a possible somehow emergent thing there
  2288. 1:46:25that could challenge Wikipedia?
  2289. 1:46:28Which Wikipedia is a different kind of thing,
  2290. 1:46:31which is more like permanent articles about things?
  2291. 1:46:33- Well, Wikipedia is very hierarchical,
  2292. 1:46:36whereas Community Notes is inherently not.
  2293. 1:46:40There is no hierarchy.
  2294. 1:46:42Like, and the asset test is,
  2295. 1:46:45I can't change a Community Note if you put a gun to my head.
  2296. 1:46:52And Community Notes has,
  2297. 1:46:56all the code is open source a hundred percent.
  2298. 1:46:59All the data is open source a hundred percent.
  2299. 1:47:01So you can completely recreate any note
  2300. 1:47:04in the system independently.
  2301. 1:47:07So if there was any interference,
  2302. 1:47:09you'd notice immediately.
  2303. 1:47:11- By the way, on the topic of Grok,
  2304. 1:47:14it's actually pretty good at criticizing you,
  2305. 1:47:16which is great. - [Elon] Okay.
  2306. 1:47:17- I asked what has Elon Musk been wrong about?
  2307. 1:47:20I just, it's actually...
  2308. 1:47:21- [Elon] Yeah, what does it say?
  2309. 1:47:23- So it says mostly,
  2310. 1:47:24lists rigorously all the times you predicted,
  2311. 1:47:29like promised when something will be released
  2312. 1:47:31and it wasn't released. - [Elon] Promise
  2313. 1:47:32is a strong word.
  2314. 1:47:33- Okay predicted. (Elon laughing)
  2315. 1:47:35But the opener's really nice.
  2316. 1:47:37Elon Musk, the man who has been wrong about more things
  2317. 1:47:40than a wrongameter in a wrong factory.
  2318. 1:47:42(Elon laughing)
  2319. 1:47:43And then it goes on to list
  2320. 1:47:45- That's pretty funny. Can I see it?
  2321. 1:47:47I'm like, I haven't seen this before. This is cool.
  2322. 1:47:48- Oh, sure. I mean, what has Elon Musk been...
  2323. 1:47:51(both laughing)
  2324. 1:47:53- Elon Musk been wrong about?
  2325. 1:47:57- Well, where do I begin? I guess this is new.
  2326. 1:47:58Elon Musk has made quite a few blunders in his time,
  2327. 1:48:01and I'm not just talking about his dance moves
  2328. 1:48:03at the Tesla Cyber Rodeo.
  2329. 1:48:06First off, let's talk about the infamous
  2330. 1:48:07"funding secured" tweet.
  2331. 1:48:09It's actually different than what it gave me before,
  2332. 1:48:11which is a bullet point list.
  2333. 1:48:14Here's more narrative structure.
  2334. 1:48:17- Cyber truck unveiling
  2335. 1:48:18where he's supposedly unbreakable windows
  2336. 1:48:20- Not correct. - [Lex] Which part?
  2337. 1:48:24- Well, the first part, I did actually
  2338. 1:48:26have the funding secured
  2339. 1:48:27and there was a big trial in San Francisco, big civil trial.
  2340. 1:48:34And the jury found me not guilty.
  2341. 1:48:37Unanimous finding of a San Francisco jury.
  2342. 1:48:40- And here it's kind of implying that
  2343. 1:48:42it was not in fact secured.
  2344. 1:48:44- I think this is taking things from the press.
  2345. 1:48:47Yeah, that is not correct.
  2346. 1:48:48The reason I agreed to the fine for the SEC
  2347. 1:48:52is not because the SEC was correct there,
  2348. 1:48:54that was extremely bad behavior
  2349. 1:48:56by the SEC, corruption, frankly.
  2350. 1:48:59And, but if I did not agree to pay the fine
  2351. 1:49:06Tesla would've gone bankrupt immediately.
  2352. 1:49:08So I was told by our CFO that the banks
  2353. 1:49:10would immediately suspend our lines of credit.
  2354. 1:49:15And if they suspend our lines of credit at that time,
  2355. 1:49:17we would've gone bankrupt instantly.
  2356. 1:49:20So there would never have been an opportunity for a trial
  2357. 1:49:22'cause Tesla would be dead.
  2358. 1:49:24So really the, if this is like someone holding
  2359. 1:49:27a gun to your kid's head and saying pay $20 million
  2360. 1:49:30and admit to, this is like a hostage negotiation.
  2361. 1:49:34- Was that story fully told?
  2362. 1:49:38I mean, SEC in its best form could be a force for good.
  2363. 1:49:42- It should be, but but not once did the SEC
  2364. 1:49:45go after any of the hedge funds,
  2365. 1:49:47who were nonstop shorting and distorting Tesla?
  2366. 1:49:51Not once.
  2367. 1:49:52They would lie flat, the hedge funds would lie flat
  2368. 1:49:55out on TV for their own gain
  2369. 1:49:57at the expense of retail investors.
  2370. 1:50:00Not once, literally a thousand times.
  2371. 1:50:03Not once did the SEC pursue them.
  2372. 1:50:06- How do you explain this failure?
  2373. 1:50:08- The incentive structure is messed up.
  2374. 1:50:11Because the lawyers at the SEC are not paid well.
  2375. 1:50:14They, it's a fairly low paying job,
  2376. 1:50:17but what they're looking for is a trophy
  2377. 1:50:20from the SEC that they're looking for something
  2378. 1:50:24they put on basically their LinkedIn.
  2379. 1:50:27From that they can get a job at a high paying law firm.
  2380. 1:50:30That's exactly what the lawyer here did.
  2381. 1:50:37And the reason they don't attack the hedge funds
  2382. 1:50:40is because those hedge funds employ those law firms,
  2383. 1:50:43and they know if they attack the hedge funds,
  2384. 1:50:45they're affecting their future career prospects.
  2385. 1:50:48So they sell small investors
  2386. 1:50:49down the river for their own career.
  2387. 1:50:55That's what actually happens. Regulatory capture.
  2388. 1:50:59- [Lex] Regulatory capture, - Yeah. Not good.
  2389. 1:51:03So the only reason I accepted that thing,
  2390. 1:51:06which technically was not an admission,
  2391. 1:51:10it's neither admit nor deny guilt.
  2392. 1:51:13But the only reason I agreed to that at all
  2393. 1:51:15was because I was told Tesla would be bankrupt otherwise.
  2394. 1:51:20So if there was an SEC investigation like this,
  2395. 1:51:24banks would suspend funding,
  2396. 1:51:25we're bankrupt immediately, at the time.
  2397. 1:51:28Now we're in a much stronger position.
  2398. 1:51:30- Take that Grok. - Yes.
  2399. 1:51:32Unfortunately it's,
  2400. 1:51:33Grok is taking too much from the conventional media.
  2401. 1:51:39Also that guy who is not a cave diver.
  2402. 1:51:42- Oh, there's a time where Elon
  2403. 1:51:47called the British cave diver, "pedo guy"
  2404. 1:51:51after the diver criticized Musk plan
  2405. 1:51:53to rescue a group of boys trapped in a Thai cave.
  2406. 1:51:55That little outburst earned him another lawsuit
  2407. 1:51:58and he had to apologize and paid a settlement.
  2408. 1:52:00- That is false. There was no settlement.
  2409. 1:52:03There was a court case,
  2410. 1:52:04which the guy who was not a cave diver
  2411. 1:52:09and played, was not part of the rescue team,
  2412. 1:52:14filed a lawsuit against me and lost,
  2413. 1:52:16and he received nothing.
  2414. 1:52:18So in this case it is wrong.
  2415. 1:52:20It is also, I guess taken this from the conventional media.
  2416. 1:52:23- Actually there's an interesting question here.
  2417. 1:52:24- These are public court cases.
  2418. 1:52:26Both the SEC civil case
  2419. 1:52:30where the civil complaints on the SEC guys
  2420. 1:52:34lost unanimous jury verdict in San Francisco.
  2421. 1:52:38They picked San Francisco
  2422. 1:52:39because they thought it was the place
  2423. 1:52:40I was most likely to lose.
  2424. 1:52:43And a unanimous predicted my favor.
  2425. 1:52:45The LA trial was also, they picked the,
  2426. 1:52:49they picked that venue because they thought
  2427. 1:52:51I was most likely to lose, unanimous verdict in my favor.
  2428. 1:52:56Both cases I won, yeah.
  2429. 1:53:00I mean there's an interesting question here.
  2430. 1:53:01Is there seems to be a lot more clicks
  2431. 1:53:07if a journalistic organization writes a negative article
  2432. 1:53:11about you, Elon Musk,
  2433. 1:53:12that's like one of the best ways to get clicks.
  2434. 1:53:16So how do you, if you're training Grok,
  2435. 1:53:19not train on articles that have
  2436. 1:53:23like the misaligned incentives.
  2437. 1:53:26- We need to add the training set
  2438. 1:53:28of the actual legal decisions.
  2439. 1:53:32So, that is a not, this is actually helpful
  2440. 1:53:35because if you actually read the...
  2441. 1:53:40- "Public".
  2442. 1:53:41- Which are public, the court conclusions,
  2443. 1:53:43they're completely the opposite of
  2444. 1:53:45what the media wrote.
  2445. 1:53:47- So always striving for like the ground truth.
  2446. 1:53:49- [Elon] Yeah. - Beyond the reporting...
  2447. 1:53:50- What did the judge actually write?
  2448. 1:53:52What did the jury and the judge actually conclude?
  2449. 1:53:57And in both cases they found me innocent.
  2450. 1:53:59And, like, that's after the jury shot
  2451. 1:54:01for the trying to find the venue
  2452. 1:54:03where I'm most likely to lose.
  2453. 1:54:07No, I mean this is obviously, it can be a much greater,
  2454. 1:54:10better critique than this.
  2455. 1:54:12I mean, I've been far too optimistic about autopilot.
  2456. 1:54:16- That was, the critique I got by the way,
  2457. 1:54:18was more about that.
  2458. 1:54:19Which is, for each it broke down a nice bullet point list
  2459. 1:54:22for each of your companies.
  2460. 1:54:23The set of predictions that you made when you'll deliver,
  2461. 1:54:27when you'll be able to solve, for example, self-driving
  2462. 1:54:30and it gives you like a list.
  2463. 1:54:31And it was probably compelling.
  2464. 1:54:32And the basic takeaway is like you're often
  2465. 1:54:35too optimistic about how long it takes
  2466. 1:54:37to get something done.
  2467. 1:54:38- Yeah, I mean, I would say
  2468. 1:54:39that I'm pathologically optimistic on schedule.
  2469. 1:54:42This is true.
  2470. 1:54:43But while I am sometimes late, I always deliver in the end.
  2471. 1:54:49- Except with Uber Lilith, no.
  2472. 1:54:50- Hmm. We'll see. (Lex laughing)
  2473. 1:54:53- Okay.
  2474. 1:54:54Is there, over the past year or so,
  2475. 1:54:57since purchasing X you've become more political.
  2476. 1:55:02Is there a part of you that regrets that?
  2477. 1:55:03- Have I?
  2478. 1:55:04- In this battle to sort of counter way the Woke
  2479. 1:55:12that comes from Francisco.
  2480. 1:55:13- I guess you consider fighting the Woke Mind Virus,
  2481. 1:55:16which I consider to be a civilizational threat,
  2482. 1:55:18to be political, then yes.
  2483. 1:55:20- So basically going into the battle,
  2484. 1:55:22the battleground of politics,
  2485. 1:55:25is there a part of you that regrets that?
  2486. 1:55:26- Yes, I don't know if this is necessarily sort of
  2487. 1:55:29one candidate or another candidate,
  2488. 1:55:31but it's, I'm generally against things
  2489. 1:55:33that are anti meritocratic
  2490. 1:55:36or where there's an attempt to suppress discussion,
  2491. 1:55:41where even discussing a topic is, you know, not allowed.
  2492. 1:55:47Woke Mind Virus is communism rebranded.
  2493. 1:55:51- Well, I mean, that said,
  2494. 1:55:52because of that battle against the Woke Mind Virus,
  2495. 1:55:55you're perceived as being right wing.
  2496. 1:55:58- If the woke is left, then I suppose
  2497. 1:56:00that would be true, but I'm not sure.
  2498. 1:56:03I think there are aspects of the left that are good.
  2499. 1:56:05I mean, if you're in favor of, you know,
  2500. 1:56:08the environment, you know,
  2501. 1:56:11if you wanna have a positive future for humanity,
  2502. 1:56:13if you believe in empathy for your fellow human beings.
  2503. 1:56:17You know, being kind and not cruel.
  2504. 1:56:21Whatever those values are.
  2505. 1:56:23- You said that you were previously left or center left.
  2506. 1:56:26What would you like to see in order for you
  2507. 1:56:27to voting for Democrats again?
  2508. 1:56:30- No, I would say that I would be
  2509. 1:56:35probably left of center on social issues,
  2510. 1:56:37probably a little bit right of center on economic issues.
  2511. 1:56:40- [Lex] And that still holds true?
  2512. 1:56:42- Yes, but I think that's probably, you know,
  2513. 1:56:44half the country, isn't that?
  2514. 1:56:46- [Lex] Maybe more, - Maybe more.
  2515. 1:56:49- Are you and AOC secretly friends?
  2516. 1:56:52Or bigger question, do you wish you and her,
  2517. 1:56:55and just people in general of all political persuasions
  2518. 1:56:57would talk more with empathy
  2519. 1:56:59and maybe have a little bit more fun
  2520. 1:57:01and good vibes and humor online?
  2521. 1:57:05- I'm always in favor of humor.
  2522. 1:57:07That's why we have a funny mode.
  2523. 1:57:08- But good vibes, camaraderie, humor, you know?
  2524. 1:57:12Like friendship. - Yeah.
  2525. 1:57:14Well, you know, I don't know AOC, you know, was,
  2526. 1:57:19I've only been at one,
  2527. 1:57:21I was at the Met Ball when she was, when she attended,
  2528. 1:57:26and she was wearing this dress,
  2529. 1:57:29but I can only see one side of it.
  2530. 1:57:31So it looked like it eat the itch,
  2531. 1:57:33but I don't know...
  2532. 1:57:35- [Lex] What the rest of it said?
  2533. 1:57:36- Yeah, something. - [Lex] I'm not sure.
  2534. 1:57:38(Elon laughing)
  2535. 1:57:39- Something about the itch. Eat the itch.
  2536. 1:57:42I think we should have a language model complete.
  2537. 1:57:43(Elon laughing)
  2538. 1:57:46What are the possible ways to complete that sentence?
  2539. 1:57:48And so I guess that that didn't work out well.
  2540. 1:57:52Well there's still hope.
  2541. 1:57:54- I root for friendship.
  2542. 1:57:55- Sure, sounds good. More characteristic.
  2543. 1:57:58- You're one of, if not the most famous, wealthy
  2544. 1:58:01and powerful people in the world,
  2545. 1:58:02in your position it's difficult to find
  2546. 1:58:04people you can trust.
  2547. 1:58:05- Trust no one, not even yourself. Not trusting yourself.
  2548. 1:58:07- Okay, well that's, (Elon laughing)
  2549. 1:58:08you're saying that jokingly,
  2550. 1:58:10but is there some...
  2551. 1:58:11- Trust no one, not even no one. (laughing)
  2552. 1:58:14- I'm gonna need an hour
  2553. 1:58:15just to think about that.
  2554. 1:58:16And maybe some drugs and maybe Grok that help me.
  2555. 1:58:21I mean, is there some aspect of that
  2556. 1:58:22when just existing in a world
  2557. 1:58:24where everybody wants something from you?
  2558. 1:58:26How hard is it to exist in that world?
  2559. 1:58:28- I'll survive. (laughing)
  2560. 1:58:31- [Lex] There's a song like that too.
  2561. 1:58:32- I will survive.
  2562. 1:58:33- Were you petrified at first?
  2563. 1:58:36Okay. I forget the rest of the lyrics.
  2564. 1:58:39But is is there, you don't struggle with this?
  2565. 1:58:41I mean, I know you survive, but like there's ways.
  2566. 1:58:44- Petrify is a spell in the Druid tree.
  2567. 1:58:47- What does it do? - Petrify.
  2568. 1:58:50(laughing)
  2569. 1:58:51It turns the monsters into stone.
  2570. 1:58:56- [Lex] Oh, like literally?
  2571. 1:58:56- Yeah, for like six seconds.
  2572. 1:58:58- [Lex] Well, the second.
  2573. 1:58:59There's so much math in Diablo that breaks my brain.
  2574. 1:59:02- It's like math nonstop.
  2575. 1:59:04- I mean really you're like laughing at it,
  2576. 1:59:06but you don't, it can put a huge amount of tension
  2577. 1:59:11on a mind.
  2578. 1:59:13- Yes, it can be definitely stressful at times.
  2579. 1:59:16- Well, how do you know who you can trust
  2580. 1:59:18in work and personal life?
  2581. 1:59:20- I mean, I guess you look at somebody's track record
  2582. 1:59:22over time and if they've got a, you know,
  2583. 1:59:25I guess you kind of use your neural net to assess,
  2584. 1:59:29you know, someone.
  2585. 1:59:31- Neural nets don't feel pain.
  2586. 1:59:33Your neural net has consciousness.
  2587. 1:59:35It might feel pain when people betray you.
  2588. 1:59:39It can make you... - I mean, I'm.
  2589. 1:59:40You know, to be frank,
  2590. 1:59:41I mean I've almost never been betrayed.
  2591. 1:59:46It's very, very rare. So, you know, for what it's worth?
  2592. 1:59:50- I guess Karma would be good to people
  2593. 1:59:52and they'll be good to you.
  2594. 1:59:53- Yeah, Karma is real. - Are there people you trust?
  2595. 1:59:58Let me edit that question.
  2596. 2:00:00Are there people close to you
  2597. 2:00:02that call you out on your bullshit?
  2598. 2:00:06- Well, the X platform is very helpful for that.
  2599. 2:00:08- [Lex] Hmm.
  2600. 2:00:09- If you're looking for critical feedback. (laughing)
  2601. 2:00:12- Can it push you like into the extremes more?
  2602. 2:00:15The extremes of thought make you cynical
  2603. 2:00:16about human nature in general?
  2604. 2:00:19- I don't think I'll be cynical.
  2605. 2:00:21In fact, I think, you know,
  2606. 2:00:25my feeling is that one should be, you know,
  2607. 2:00:31never trust a cynic.
  2608. 2:00:34The reason is that cynics excuse
  2609. 2:00:38their own bad behavior by saying everyone does it
  2610. 2:00:42because they're cynical.
  2611. 2:00:44So I always be, it's a red flag
  2612. 2:00:46if someone's a cynic, a true cynic.
  2613. 2:00:49- Yeah, there's a degree of projection there
  2614. 2:00:51that's always fun to watch from the outside
  2615. 2:00:53and enjoy the, well the hypocrisy.
  2616. 2:00:57- But this is an important point
  2617. 2:00:59that I think people who are listening should bear in mind,
  2618. 2:01:02if somebody is cynical, meaning that they see
  2619. 2:01:06bad behavior in everyone,
  2620. 2:01:08it's easy for them to excuse their own bad behavior
  2621. 2:01:12by saying that, well, everyone does it.
  2622. 2:01:16It's not true.
  2623. 2:01:19I think most people are kind of medium good.
  2624. 2:01:23- I do wish the people on X will be better
  2625. 2:01:26at seeing the good in other people's behavior.
  2626. 2:01:29There seems to be a kind of weight
  2627. 2:01:32towards seeing the negative.
  2628. 2:01:35Somehow, the negative is sexier.
  2629. 2:01:37Interpreting the negative is sexier, more viral.
  2630. 2:01:41I don't know what that is exactly about human nature.
  2631. 2:01:43- I mean, I find the X platform
  2632. 2:01:45to be less negative than the legacy media, you know?
  2633. 2:01:48I mean, if you read sort of a sort of
  2634. 2:01:50conventional newspapers, just, it makes you sad.
  2635. 2:01:54- [Lex] Yeah. - Frankly.
  2636. 2:01:58Whereas I'd say on the X platform,
  2637. 2:02:02I mean I really get more laughs per day on X
  2638. 2:02:05than everything else combined from humans, you know?
  2639. 2:02:09- Laughs is one thing.
  2640. 2:02:11Laughs is it overlaps,
  2641. 2:02:13but it's not necessarily perfectly overlapping
  2642. 2:02:17with like good vibes and support,
  2643. 2:02:19like celebrating others, for example.
  2644. 2:02:22Not in a stupid, shallow, naive way,
  2645. 2:02:24but like in an awesome, like, oh, something awesome happened
  2646. 2:02:27and you celebrate them for it.
  2647. 2:02:30It feels that that is outweighed
  2648. 2:02:32by shading on other people.
  2649. 2:02:35Now it's better than mainstream media, but it's still.
  2650. 2:02:38- Yeah, mainstream media is almost
  2651. 2:02:40relentlessly negative about everything.
  2652. 2:02:42It's, I mean, really the conventional news tries
  2653. 2:02:45to answer the question,
  2654. 2:02:46what is the worst thing that I have on earth today?
  2655. 2:02:49And it's a big world.
  2656. 2:02:50So on any given day, something bad has happened
  2657. 2:02:54- And a journalization of that,
  2658. 2:02:55what is the worst perspective
  2659. 2:02:56I can take on a thing that happened, so.
  2660. 2:02:59- Yeah, it's, I don't know,
  2661. 2:03:01there's just a strong negative bias in the news.
  2662. 2:03:06I mean I think there's one, a possible explanation
  2663. 2:03:08for this is evolutionary,
  2664. 2:03:12where, you know, bad news historically
  2665. 2:03:15would be potentially fatal.
  2666. 2:03:17Like there's lion over there
  2667. 2:03:19or there's some other tribe that wants to kill you.
  2668. 2:03:22Good news, you know, like we found a patch of berries
  2669. 2:03:25is nice to have, but not essential.
  2670. 2:03:28- So our old friend Tesla autopilot
  2671. 2:03:33and it's probably one of the most intelligent
  2672. 2:03:35real world AI systems in the world.
  2673. 2:03:38- You followed it from the beginning.
  2674. 2:03:40- Yeah, it was one of the most incredible robots
  2675. 2:03:42in the world and continues to be.
  2676. 2:03:43And I was really exciting.
  2677. 2:03:45And it was super exciting when it generalized
  2678. 2:03:49became more than a robot on four wheels,
  2679. 2:03:52but a real world AI system that perceives the world.
  2680. 2:03:57- [Elon] Yeah. - And can have
  2681. 2:03:59potentially different embodiments.
  2682. 2:04:02- Well I mean the really wild thing about the end-to-end
  2683. 2:04:03training is that, like, it learns to read.
  2684. 2:04:07Like it can read signs, but we never taught it to read.
  2685. 2:04:12So, yeah.
  2686. 2:04:15We never taught it what a car was
  2687. 2:04:18or what a person was or a cyclist.
  2688. 2:04:22It learnt what all those things are,
  2689. 2:04:26what all the objects are on the road from video,
  2690. 2:04:31just from watching videos.
  2691. 2:04:32Just like humans.
  2692. 2:04:33I mean humans are photons
  2693. 2:04:36and controls out.
  2694. 2:04:38Like the vast majority of information reaching our brain
  2695. 2:04:41is from our eyes.
  2696. 2:04:44And you say, well what's the output?
  2697. 2:04:45The output is our motor signals to our sort of fingers
  2698. 2:04:49and mouth in order to communicate.
  2699. 2:04:55Proton in controls out. The same is true of the car.
  2700. 2:05:01- But by looking at the sequence of images, it's,
  2701. 2:05:05you've agreed with Ilya Sutskever recently
  2702. 2:05:08where he talked about LLM forming a world model
  2703. 2:05:11and basically language is a projection
  2704. 2:05:13of that world model onto the sequence of letters,
  2705. 2:05:17and and you're saying...
  2706. 2:05:18- Finds order in these things.
  2707. 2:05:21It finds correlative clusters.
  2708. 2:05:27- And so doing it's like understanding
  2709. 2:05:29something deep about the world?
  2710. 2:05:30- [Elon] Yeah.
  2711. 2:05:32- Which is like, and it is beautiful.
  2712. 2:05:35- That's how our brain works.
  2713. 2:05:38- Yeah, but it's beautiful.
  2714. 2:05:39- Protons in controls out.
  2715. 2:05:40- Neural nets are able to understand
  2716. 2:05:42that deep meaning in the world.
  2717. 2:05:43And so the question is how far can it go?
  2718. 2:05:46And it does seem everybody's excited about LLMs.
  2719. 2:05:50So in the space of self supervised learning
  2720. 2:05:52in the space of text. - [Elon] Yeah.
  2721. 2:05:56- It seems like there's a deep similarity between that
  2722. 2:06:00and what Tesla autopilot is doing.
  2723. 2:06:04Is it to you basically the same?
  2724. 2:06:06- They're converging, they're converging.
  2725. 2:06:10- I wonder who gets there faster?
  2726. 2:06:12Having a deep understanding of the world,
  2727. 2:06:16or they just will naturally converge.
  2728. 2:06:19- They're both headed towards AGI.
  2729. 2:06:23The Tesla approach is much more computer efficient.
  2730. 2:06:27It had to be 'cause we were constrained on this,
  2731. 2:06:31you know, we only have a hundred watts
  2732. 2:06:36and into eight computer,
  2733. 2:06:39144 trillion operations per second,
  2734. 2:06:42which sounds like a lot,
  2735. 2:06:43but is kind of small potatoes these days
  2736. 2:06:48that it did.
  2737. 2:06:49But it's understanding the world that it did.
  2738. 2:06:52It's only 256 values.
  2739. 2:06:55- But there, the path to AGI might have much
  2740. 2:06:58more significant impact because it's understanding,
  2741. 2:07:02it'll faster understand the real world than will LLMs.
  2742. 2:07:07And therefore be able to integrate
  2743. 2:07:09with the real humans in the real world faster.
  2744. 2:07:13- They're both going to understand the world.
  2745. 2:07:16But I think Tesla's approach
  2746. 2:07:19is fundamentally more computer efficient.
  2747. 2:07:21It had to be, there was no choice.
  2748. 2:07:23Like our brain is very computer efficient,
  2749. 2:07:25very, very energy efficient.
  2750. 2:07:27So think of like, what is our brain able to do?
  2751. 2:07:32You know, there's only about 10 watts
  2752. 2:07:33of higher brain function, not counting stuff,
  2753. 2:07:36that's just used to control our body.
  2754. 2:07:39The thinking part of our brain is less than 10 watts,
  2755. 2:07:43and those 10 watts can still produce
  2756. 2:07:46a much better novel than a 10 megawatt GPU cluster.
  2757. 2:07:50So there's a six order of magnitude difference there.
  2758. 2:07:56I mean, the AI has thus far gotten to where it is,
  2759. 2:08:00via brute force, just throwing massive amounts of compute
  2760. 2:08:04and massive amounts of power at it.
  2761. 2:08:08So this is not where it'll end up.
  2762. 2:08:15You know, in general with any given technology,
  2763. 2:08:16first try to make it work and then you make it efficient.
  2764. 2:08:20So I think we'll find over time that these models
  2765. 2:08:24get smaller, are able to do produce
  2766. 2:08:30sensible output with far less compute wireless power.
  2767. 2:08:36Tesla is arguably ahead of the game on that front
  2768. 2:08:38because it has, we've just been forced
  2769. 2:08:42to try to understand the world
  2770. 2:08:46with a hundred watts of compute.
  2771. 2:08:51And there are a bunch of sort of fundamental functions
  2772. 2:08:54that we kind of forgot to include.
  2773. 2:08:56So we have to run them in a bunch of things in emulation.
  2774. 2:09:00We fixed a bunch of those with hardware four,
  2775. 2:09:02and then hardware five will be even better.
  2776. 2:09:08But it does appear at this point
  2777. 2:09:12that the car will be able to drive better than a human,
  2778. 2:09:16even with hardware three and a hundred watts of power.
  2779. 2:09:20And really, if we really optimize
  2780. 2:09:23it could be probably less than 50 watts.
  2781. 2:09:26- What have you learned about developing Optimus?
  2782. 2:09:31About applying, integrating this kind of real world AI
  2783. 2:09:35into the space of robotic manipulation,
  2784. 2:09:39just humanoid robotics?
  2785. 2:09:42What are some interesting tiny
  2786. 2:09:44or big things you've understood?
  2787. 2:09:47- I was surprised at the fact
  2788. 2:09:49that we had to develop every part of the robot ourselves.
  2789. 2:09:53That there were no off the shelf motors, electronics,
  2790. 2:09:58sensors, like we had to develop everything.
  2791. 2:10:03We couldn't actually find a source of electric motors
  2792. 2:10:09for any amount of money.
  2793. 2:10:12- [Lex] So it's not even just the efficient, inexpensive,
  2794. 2:10:15it's like anything, there's not a...
  2795. 2:10:17- No,
  2796. 2:10:19- [Lex] The actuators, everything, everything has to be?
  2797. 2:10:21- Yeah. - [Lex] Designed from scratch.
  2798. 2:10:23- We tried hard to find anything that was,
  2799. 2:10:26because you think of how many electric motors are made
  2800. 2:10:28in the world.
  2801. 2:10:29There's like tens of thousands,
  2802. 2:10:31hundreds of thousands of electric motor designs.
  2803. 2:10:37None of them were suitable for a humanoid robot.
  2804. 2:10:40Literally none.
  2805. 2:10:42So we had to develop our own design
  2806. 2:10:46specifically for what a humanoid robot needs.
  2807. 2:10:51- How hard was it to design something
  2808. 2:10:53that can be mass manufactured,
  2809. 2:10:54could be relatively inexpensive?
  2810. 2:10:56I mean, if you compare to Boston Dynamics Atlas,
  2811. 2:10:58that's a very expensive robot.
  2812. 2:11:02- It is designed to be manufactured
  2813. 2:11:04in the same way that we would make a car.
  2814. 2:11:06And I think ultimately we can make Optimists
  2815. 2:11:07for less than the cost of a car.
  2816. 2:11:11It should be.
  2817. 2:11:12'Cause if you look at the mass of the robot is much smaller
  2818. 2:11:17and the car has many actuators in it.
  2819. 2:11:21The car has more actuators than the robot.
  2820. 2:11:23- But there is, the actuators are kind of interesting
  2821. 2:11:25in a humanized robot with the fingers.
  2822. 2:11:28So Optimus has really nice, hands and fingers, you know?
  2823. 2:11:32- [Elon] Yeah. - And they could do
  2824. 2:11:34some interesting manipulation.
  2825. 2:11:36Soft touch robotics.
  2826. 2:11:38- I mean, one of the tests goals I have is,
  2827. 2:11:42can it pick up a needle and a thread
  2828. 2:11:44and thread the needle, just by looking.
  2829. 2:11:47- How far away are we from that?
  2830. 2:11:49Just by looking, just by looking.
  2831. 2:11:51- Maybe a year.
  2832. 2:11:53Although I go back to, I'm optimistic on time.
  2833. 2:11:56The work that we're doing in the car
  2834. 2:11:57will translate to the robot.
  2835. 2:11:59- The perception or the, also the control?
  2836. 2:12:02- No, the controls are different,
  2837. 2:12:03but the video in controls out,
  2838. 2:12:07the car is robot on four wheels.
  2839. 2:12:09The Optimist is a robot with the hands and legs.
  2840. 2:12:15- [Lex] So you can just...
  2841. 2:12:16- But they're very similar.
  2842. 2:12:17- So the entire machinery of the learning process
  2843. 2:12:20end to end is just,
  2844. 2:12:21you just have a different set of controls?
  2845. 2:12:23- Optimist will figure out
  2846. 2:12:25how to do things by watching videos.
  2847. 2:12:28- As the saying goes,
  2848. 2:12:29"Be kind for everyone you meet
  2849. 2:12:30is fighting a battle, you know nothing about."
  2850. 2:12:33- [Elon] Yeah, it's true.
  2851. 2:12:34- What's something difficult
  2852. 2:12:35you're going through that people don't often see?
  2853. 2:12:38- Trying to defeat Uber Lilith? (laughing)
  2854. 2:12:44No, I mean, you know,
  2855. 2:12:50I mean, my mind is a storm
  2856. 2:12:52and I don't think, I don't think
  2857. 2:12:55most people would wanna be me.
  2858. 2:12:56They may think they would wanna be me,
  2859. 2:12:57but they don't know.
  2860. 2:12:59They don't understand.
  2861. 2:13:11- How are you doing?
  2862. 2:13:14- I mean, overall, okay.
  2863. 2:13:18In the grand scheme of things, I can't complain.
  2864. 2:13:21- Do you get lonely? - Sometimes.
  2865. 2:13:25But I, you know, my kids
  2866. 2:13:29and friends keep me company.
  2867. 2:13:33- So not existential.
  2868. 2:13:36- There are many nights I sleep alone.
  2869. 2:13:40I don't have to, but I do.
  2870. 2:13:46- Walter Isaacson, in his new biography of you wrote
  2871. 2:13:49about your difficult childhood.
  2872. 2:13:51Will you ever find forgiveness in your heart
  2873. 2:13:54for everything that has happened to you
  2874. 2:13:55in that period of your life?
  2875. 2:14:01- What is forgiveness? I do not...
  2876. 2:14:06At least I don't think I have a resentment,
  2877. 2:14:14so nothing to forgive.
  2878. 2:14:19- You know, forgiveness is difficult for people.
  2879. 2:14:25It seems like you don't harbor the resentment.
  2880. 2:14:28- I mean, I try to think about like,
  2881. 2:14:29what is gonna affect the future in a good way.
  2882. 2:14:34And holding onto grudges
  2883. 2:14:36does not affect the future in a good way.
  2884. 2:14:41- You're a father, a proud father.
  2885. 2:14:43What have you learned about life from your kids?
  2886. 2:14:49Those little biological organisms?
  2887. 2:14:53- I mean, developing AI
  2888. 2:14:55and watching, say little X grow
  2889. 2:14:58is fascinating because they,
  2890. 2:15:03they're far more parallels than I would've expected.
  2891. 2:15:08I mean, I can see his biological neural net
  2892. 2:15:11making more and more sense to the world.
  2893. 2:15:13And I can see the digital neural net
  2894. 2:15:15making more and more sense of the world at the same time.
  2895. 2:15:19- Do you see the beauty in magic in both?
  2896. 2:15:21- Yes.
  2897. 2:15:26I mean, one of the things with kids is that, you know,
  2898. 2:15:29you kind of see the world anew in their eyes.
  2899. 2:15:34You know, to them everything is new and fresh.
  2900. 2:15:39And then when you see that,
  2901. 2:15:43them experience the world is new and fresh, you do too.
  2902. 2:15:52- Well. Elon, I just wanna say thank you
  2903. 2:15:54for your kindness to me and friendship over the years.
  2904. 2:15:59For seeing something in a silly kid like me,
  2905. 2:16:01as you've done for many others.
  2906. 2:16:03And thank you for having hope for a positive future,
  2907. 2:16:07for humanity,
  2908. 2:16:09and for working your ass off to make it happen.
  2909. 2:16:11Thank you, Elon. - [Elon] Thanks Lex.
  2910. 2:16:13- Thank you for listening to this conversation
  2911. 2:16:15with Elon Musk.
  2912. 2:16:16To support this podcast,
  2913. 2:16:17please check out our sponsors in the description.
  2914. 2:16:20And now let me leave you with some words
  2915. 2:16:22that Walter Isaacson wrote about the central philosophy
  2916. 2:16:25of how Elon approaches difficult problems.
  2917. 2:16:29"The only rules are the ones dictated
  2918. 2:16:32by the laws of physics."
  2919. 2:16:34Thank you for listening and hope to see you next time.

About this transcript

This page contains the full transcript of Elon Musk: War, AI, Aliens, Politics, Physics, Video Games, and Humanity | Lex Fridman Podcast #400 by Lex Fridman, generated from the public captions YouTube serves with the video. The transcript has 20,825 words across 2,919 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.