YouTube transcript (sRvrXL83N-c) — Transcript

Full transcript

1. 0:09You know I thought where we'd start Ali
2. 0:11is you know a lot of talk right before
3. 0:14you joined about there's world's moving
4. 0:17fast
5. 0:18>> XAI cursor open AI fighting anthropic
6. 0:21you know you guys have done such a great
7. 0:23job of stacking you know going from a
8. 0:25data business to a lakehouse business to
9. 0:26now an AI business just state of the
10. 0:28union. Mhm.
11. 0:29>> View from the top. What are you what are
12. 0:31you seeing? Like frame the landscape for
13. 0:33us. Um what are the biggest things that
14. 0:35you were thinking about? And you know,
15. 0:37I've got a bunch of questions that we
16. 0:38can talk, but I thought we'd just open
17. 0:39it up to like what is the biggest thing
18. 0:41on your mind as you as you think about
19. 0:42AI?
20. 0:43>> Yeah, I think that uh you know, I think
21. 0:44you guys can chill out. Don't be
22. 0:46stressed. You know, I think times are
23. 0:48crazy and uh I think it's not warranted
24. 0:52basically. And I think the stress uh
25. 0:54makes people do stupid things and chase
26. 0:56just uh you know whatever happens to be
27. 0:58the crazy thing that everybody's talking
28. 1:00about on Twitter. Uh I think it makes
29. 1:02people have tunnel vision and not work
30. 1:04on the right stuff.
31. 1:05>> Yeah.
32. 1:06>> Uh and I think that's what I see with
33. 1:07like the current generation. Like every
34. 1:09year we have interns coming to data
35. 1:10bricks and the interns I do always like
36. 1:12a session with them an hour or 90
37. 1:14minutes or something they can ask
38. 1:15questions and last two years have been
39. 1:17just insane. Before they would ask for
40. 1:19like good career advice and you would
41. 1:20give them good career advice. Now
42. 1:22they're like 22 year olds who are like
43. 1:24oh my god should I like start start my
44. 1:26own company and be a CEO or if I like
45. 1:28delay that by six months working on
46. 1:29something have I ruined my career and
47. 1:31life is over and you know AGI is going
48. 1:33to happen and I'm going to miss the boat
49. 1:35and like what am I going to do? I'm like
50. 1:36so I'm like just trying to tell people
51. 1:38like calm down take a deep breath.
52. 1:40Things take time.
53. 1:41>> Yeah.
54. 1:41>> You know so that's what I would say. I
55. 1:43would say actually I think also in
56. 1:45Silicon Valley if you think about it
57. 1:46right now what's happening is I think uh
58. 1:49and you might disagree with some of this
59. 1:50so feel free to push back you guys can
60. 1:51might disagree too you can push back as
61. 1:53well uh but there's this quest for super
62. 1:55intelligence which I think is
63. 1:56unwarranted
64. 1:58>> because first of all they're not even
65. 2:00defining what super intelligence is but
66. 2:02it's this kind of like godlike you know
67. 2:03it's like I think people reading kurs
68. 2:05well and you know it's take off
69. 2:07singularity you know this thing that
70. 2:09comes and like you know recursive
71. 2:11self-improvement
72. 2:12and you know cures all the diseases and
73. 2:14GDP jumps by like 10% and unemployment
74. 2:17goes to 20% and there's no more jobs and
75. 2:21UBI to everyone and so on. No, I think
76. 2:22they believe it. I think it's not
77. 2:24needed. I think we already have AGI so
78. 2:27we already have artificial general
79. 2:28intelligence.
80. 2:29>> Uh you know um how okay this is always
81. 2:33equally fun. How many people think we
82. 2:34have AGI already?
83. 2:36>> Okay, it's always the same. It's always
84. 2:38like 10%. [laughter] Okay. Uh, how many
85. 2:40of you think that a lot of people that
86. 2:42you interact with are not as smart as
87. 2:44the smartest models that you use?
88. 2:48>> Okay, [laughter] now let's start all
89. 2:49over. How many of you think we don't
90. 2:51have AGI yet? [laughter]
91. 2:55>> By the way, it always works. It's like,
92. 2:57you know, see, it's like the hypnosis is
93. 2:59working for some reason. They've gone
94. 3:02the whole world to believe we don't have
95. 3:03AGI, but it's like you just answered it
96. 3:05that you have it, you know,
97. 3:07>> but yet nah, no. uh you want to move the
98. 3:09goalpost. By the way, I was at the
99. 3:11research lab in 2009 at UC Berkeley
100. 3:13called AMPLab
101. 3:14>> was probably the biggest uh most active
102. 3:17kind of important AI lab of its time in
103. 3:192009.
104. 3:20>> And um you know the you know the god of
105. 3:25AI was working in that lab which is
106. 3:26Michael Jordan. His name is actually
107. 3:27that. So he's like the Michael Jordan of
108. 3:29AI. Uh [laughter]
109. 3:31and um back then our definition of AGI,
110. 3:34artificial general intelligence
111. 3:36um you know we've hit that
112. 3:39>> like anything we imagined would be AGI
113. 3:40we already hit that and those are all
114. 3:42the leading AI researchers in United
115. 3:43States many of them were working in that
116. 3:45lab but I was just I wanted to see like
117. 3:47if I'm just full of it. So I went and
118. 3:49asked some of those people that were
119. 3:50there at the time and I asked them as
120. 3:51hey do you agree and they all said yeah
121. 3:53according to that definition in 2009 for
122. 3:55sure we've hit that but you know and
123. 3:57then there's always some you know stupid
124. 3:59butt we moved the goalpost or we want to
125. 4:01change it or we want to have some other
126. 4:02definition or it did or the AI at some
127. 4:04there's some example that you know it
128. 4:06couldn't count the number of RS in
129. 4:07strawberry or something so therefore we
130. 4:09don't have AGI um we already have AGI
131. 4:12okay it's already smarter than many of
132. 4:14the people that you interact with that
133. 4:15is general intelligence it is artificial
134. 4:17it's not exactly a human it's not the
135. 4:19way human brain works so we already have
136. 4:21that so in some sense uh you know
137. 4:24blowing a lot of money on GPUs and data
138. 4:26centers and all of that kind of stuff is
139. 4:28not really needed
140. 4:29>> okay then there is at the same time so
141. 4:31you ask for the state of the union on
142. 4:33the other hand you have like the MIT
143. 4:34tech report that says that 95% of the
144. 4:36PC's are failing right
145. 4:38>> uh it's kind of right directionally I
146. 4:39don't know if the 95% might be wrong
147. 4:41maybe it's just 75% who knows
148. 4:43>> u but if you go inside of an enterprise
149. 4:45and or inside of an organization you go
150. 4:46into any and you look at how they're
151. 4:48using stuff. The reality is that there's
152. 4:52no like lots of agentic co-workers
153. 4:54running around doing all the work, you
154. 4:56know, blending with humans. That's not
155. 4:58happening. Okay? It's just humans
156. 5:01shuffling TPS reports.
157. 5:03>> Okay? It's like office the QA the movie
158. 5:05is office the movie the office space the
159. 5:07movie is still like how the world runs.
160. 5:10>> Yeah.
161. 5:10>> This the reality. This is just the
162. 5:12truth. Like
163. 5:13>> even inside AI companies, that's how
164. 5:14they run them. It's like they they like
165. 5:15to think but they're hiring salespeople
166. 5:17from old school companies and they're
167. 5:18running things in old school ways and I
168. 5:20don't see like that futuristic thing.
169. 5:22>> So then what's going on? We have AGI but
170. 5:24on the other hand uh none of this is
171. 5:27working and no company is using it. What
172. 5:29the hell is going on?
173. 5:30>> Uh I think it's very simple. Um, if you
174. 5:33don't get all the context that exists
175. 5:34inside of these organizations and how
176. 5:36humans work and everything, all the
177. 5:37context we have in our heads, if you
178. 5:39don't get that to the models and the
179. 5:41agents, they're going to do lots of
180. 5:43stupid mistakes and they're useless. And
181. 5:44that's what's happening right now. The
182. 5:46models just don't have or the agents
183. 5:48don't have the context that humans have
184. 5:50inside of organizations. Therefore,
185. 5:52they're useless. They do stupid mistakes
186. 5:53because they don't know all the stuff
187. 5:54that we know. Mhm.
188. 5:55>> You know, inside of every company,
189. 5:57there's always like this one guy or this
190. 5:59one gal
191. 6:00>> who's like, "Oh, go ask John or Jane."
192. 6:02Like she knows everything, you know, and
193. 6:04everybody's like tapping on that
194. 6:05person's, you know, and that's the one
195. 6:06person you can't lose in the company. If
196. 6:07you lose that person, the whole company
197. 6:08collapses.
198. 6:09>> Yeah.
199. 6:10>> That one person has that one person
200. 6:11exists in every department, in every
201. 6:13company, in every organization. And that
202. 6:15one person has all the context in their
203. 6:16head.
204. 6:17>> And that person what they have in their
205. 6:19head is not inside of the model.
206. 6:20>> So therefore, the model can't operate.
207. 6:22it just doesn't know a lot of the stuff
208. 6:24that's sort of usually John or Jane in
209. 6:27that company have been there for 10
210. 6:28years 15 years 20 years sometimes 30 40
211. 6:30years um you need to get that
212. 6:32transferred to the AI if you don't the
213. 6:35AI doesn't matter if you get super
214. 6:36intelligence and you can solve really
215. 6:38difficult you know math questions um uh
216. 6:42you know and if you can get that context
217. 6:44into the AI we already have AGI and they
218. 6:46can already crack the problem so my uh
219. 6:49urge to you guys would be uh you If you
220. 6:52want to have impact in the world, figure
221. 6:54out how to get that context into the AIS
222. 6:57uh inside of an like take an
223. 6:59organization, how do you transform how
224. 7:01old school business is happening and how
225. 7:03do you get those processes into the
226. 7:05agents then you will have massive impact
227. 7:07because AGI is already here.
228. 7:08>> Yeah,
229. 7:08>> right. That's my state of the unit.
230. 7:10>> AGI is already here.
231. 7:11>> You got to download the brain
232. 7:13>> into the silicon,
233. 7:14>> get the carbon to talk to the silicon.
234. 7:16>> Yes.
235. 7:16>> You know what? Actually, we were just
236. 7:17talking about this.
237. 7:18>> And by the way, queue up your your like
238. 7:19push backs. I'm very curious to hear.
239. 7:22>> I'm sure a majority disagrees. How many
240. 7:23disagree with this?
241. 7:26>> Oh, not that many. I was hoping. Okay,
242. 7:28I'm going to be more provocative. Okay,
243. 7:30we need more push back. Okay, so you
244. 7:32know, before we before we go into AI,
245. 7:34you know, there's a shadow of AI.
246. 7:36>> Uhhuh.
247. 7:36>> Software is dead.
248. 7:37>> Mhm.
249. 7:38>> Software has been dead for a while.
250. 7:39We've had this
251. 7:40>> four times. It happens. Every time it
252. 7:42happens, it bounces back.
253. 7:43>> Yeah.
254. 7:44>> Some macro reason,
255. 7:46Brexit, taper tantrum, inflation. This
256. 7:48time it's AI and the question that class
257. 7:51is asking is should we be loading up on
258. 7:53on software stocks. So, so is software
259. 7:56dead? Is this a buy the dip situation?
260. 7:59>> Uhhuh.
261. 8:00>> And you know I can't think of a better
262. 8:01person to ask because depending on the
263. 8:03day you ask there's like software you
264. 8:05know we AI company software company
265. 8:07speak about that is software.
266. 8:08>> I think you know better you're an
267. 8:09investor I'm not an investor also I
268. 8:11don't give financial advice. Uh but
269. 8:14[laughter]
270. 8:16having said that, if all software is
271. 8:17dead, then isn't OpenAI entropic dead?
272. 8:20>> They're just software companies with a
273. 8:21bunch of researchers writing software.
274. 8:24>> So those companies would be dead, too,
275. 8:26right? So they shouldn't have trillion
276. 8:28dollar valuations. SpaceX might make
277. 8:29sense because they make rockets, but
278. 8:31everybody else should be dead. Um,
279. 8:33Nvidia should be dead because they just
280. 8:35have really smart people who create chip
281. 8:38designs, humans that use some software
282. 8:40to create chip designs and then they
283. 8:42ship them over the internet probably
284. 8:43over to TSMC which is a real company
285. 8:46creating actual chips. But then Nvidia
286. 8:48would be dead as well. So the world's
287. 8:49most valuable company should be dead as
288. 8:51well cuz software is dead, right? So
289. 8:53software obviously isn't dead and it's
290. 8:54not going to be dead and Nvidia and Open
291. 8:56AAI and Entropic are not going to be
292. 8:57dead companies uh, you know, because of
293. 9:00whatever SAS apocalypse or whatever we
294. 9:01want to call it.
295. 9:03Yeah.
296. 9:03>> Um but um I do think two things are
297. 9:06true. I think that um two uh big changes
298. 9:11have happened which is one is barriers
299. 9:13to entry
300. 9:14>> uh have significantly gone down and then
301. 9:16switching costs have significantly gone
302. 9:18down. So let's talk about those. Um
303. 9:21barriers to entry because it's easier
304. 9:22than ever to write software.
305. 9:24>> Mh.
306. 9:25>> Um so that's like a new weapon.
307. 9:27>> Yeah. anyone can produce software uh
308. 9:30very cheaply
309. 9:31>> almost at zero cost. It's not quite zero
310. 9:32cost and it will never be zero cost but
311. 9:34much much cheaper than before
312. 9:35>> but that weapon is available to
313. 9:37everyone.
314. 9:37>> Mhm.
315. 9:38>> So also the people that create software
316. 9:39now also have that weapon. It's not like
317. 9:41>> only some some new players have that
318. 9:44everyone now.
319. 9:45>> Mhm.
320. 9:45>> So including data bricks like we're a
321. 9:46software company but we also have that
322. 9:48weapon and it's an awesome weapon using
323. 9:49Have you used that weapon and
324. 9:51substituted any of your core software
325. 9:53expenses like your CRM, your IT help
326. 9:55desk, your office of the CFO software?
327. 9:58>> No, I think that's stupid. Uh, also I
328. 9:59think switching costs are lowered
329. 10:01because it's easier to switch between
330. 10:02UIs.
331. 10:03>> Mhm.
332. 10:03>> Like, you know, humans get locked into
333. 10:05software. They get, you know, I don't
334. 10:06know if you're Are you How many use
335. 10:08Android?
336. 10:09>> Oh, no one. Okay. Wow. How many use
337. 10:12>> you guys?
338. 10:12>> Okay. Wow. Okay. All right.
339. 10:16Um, you don't want to switch to Android.
340. 10:18Why? You don't know the UI. You don't
341. 10:20know how to use it, right? It's like a
342. 10:22different UI. You would have to also
343. 10:23transfer all your data, your phone
344. 10:25contacts, all that. It's too much
345. 10:27inertia switching cost. That's a
346. 10:28switching cost, right? But if in the
347. 10:30future you're just talking to an agent,
348. 10:32that switching cost gets eliminated
349. 10:33because you're just talking to an agent.
350. 10:34So, who cares if the agent is
351. 10:36instrumenting your Android or your
352. 10:38iPhone or your Gmail or Outlook or your
353. 10:41Salesforce or the competitor or whatever
354. 10:43it is. So, that's like the switching
355. 10:44costs coming down as well. M
356. 10:46>> um so yeah I think it's going to be more
357. 10:48competition
358. 10:49>> um so I think software companies will
359. 10:51have to run more efficiently
360. 10:53>> uh that I think is going to happen
361. 10:55>> um but software is not the only moat
362. 10:57there's a good book you should read it's
363. 10:59called the seven powers how many have
364. 11:00read the seven powers
365. 11:02>> okay bunch of people here okay yeah so
366. 11:03there's like there are moes that are not
367. 11:05just software right I mean like
368. 11:07economies of scale if you can do things
369. 11:09at scale better than anyone else uh so
370. 11:11that you can afford crazy fixed costs
371. 11:14cuz you're advertising them way because
372. 11:16of your scale you know Amazon AWS
373. 11:20um you know uh then that's a moat
374. 11:23>> uh if you have a brand like Ferrari or
375. 11:25Rolex you know that's a that's a moat
376. 11:28people writing cheap software can't just
377. 11:30come replace that brand people care
378. 11:31about that brand
379. 11:33>> uh trust you know like I'm the only one
380. 11:35providing you can trust my company we
381. 11:38don't get hacked we have like really
382. 11:39secure software we have special
383. 11:41certification maybe we have patents that
384. 11:43remains a moat that you cannot break
385. 11:45that that easily. So um all these remain
386. 11:49uh there's a bunch of other ones
387. 11:50switching costs on so um data is a big
388. 11:53moat. If you have special data that no
389. 11:55one else has
390. 11:56>> um that only you have
391. 11:58>> that's a moat doesn't matter if they can
392. 11:59write cheap software. So so I think it's
393. 12:02the answer is in between.
394. 12:04>> Yeah.
395. 12:04>> The way I say it is if if a company has
396. 12:07uh been around for 10 years and they
397. 12:09have not innovated if their software
398. 12:10looks the same as 10 years ago but the
399. 12:12revenue has been going up
400. 12:13>> Yeah. they should be worried. Yeah.
401. 12:15>> Because they have not been innovating
402. 12:16and it's probably easier for a company
403. 12:18that starts today to then with you know
404. 12:21barriers of entry being lower write
405. 12:23software quickly that's much better than
406. 12:25that company because that company hasn't
407. 12:26done anything for 10 years.
408. 12:27>> Yeah.
409. 12:28>> They should be really afraid.
410. 12:29>> Yeah.
411. 12:29>> Uh and probably they don't have the
412. 12:31innovation muscle anymore because they,
413. 12:33you know, they're not innovating. So
414. 12:35obviously they're not they don't have
415. 12:36innovators. Um those kind of companies
416. 12:38are going to be wiped out.
417. 12:39>> Yeah. But there's going to be other
418. 12:40companies that have been innovating the
419. 12:42last 10 years and they're software
420. 12:43companies or companies that now get
421. 12:45their together because they're nervous
422. 12:47>> and they'll be fine too.
423. 12:49>> Perfect.
424. 12:49>> So what do you think? You're an
425. 12:50investor.
426. 12:51>> I mean I think um there's a grade
427. 12:53exactly as you said like if I was to
428. 12:54give the grades to types of software
429. 12:57companies I would say if you have got a
430. 12:59lot of data like you said if you've got
431. 13:01you know some cyber like you're like in
432. 13:04some core loop you're probably most
433. 13:05robust and immune from it. Somewhere in
434. 13:08the middle is all the workflow software
435. 13:09like which has not innovated. The UX
436. 13:11still looks same and you're like
437. 13:12scrunching down on your shoulder and
438. 13:13typing I met Ali Goatsy today. These are
439. 13:15the notes like that stuff's probably
440. 13:17gone.
441. 13:17>> Yeah.
442. 13:18>> If you you were exactly you said no
443. 13:20innovation you were a part of the old
444. 13:23habit.
445. 13:23>> But any one of those they have customers
446. 13:25they have data. If they build great AI
447. 13:28and start innovating they can keep the
448. 13:30keep keep on going. They might have to
449. 13:32change their pricing structure and their
450. 13:33cost basis but they'll be fine. Uh in
451. 13:35fact they have a lot of advantages
452. 13:37against incumbents. They have data, they
453. 13:39have customers and they have some scale.
454. 13:41So they have some economies of scale
455. 13:42going
456. 13:43>> but they do have to get their together
457. 13:45and that's you know easier said than
458. 13:46done.
459. 13:47>> Yeah. Yeah. Yeah. I'll um flash a chart.
460. 13:50Have you guys seen this uh chart from
461. 13:52Ethan Malik? He talks about AI is very
462. 13:55good at some things. He calls it the
463. 13:57jagged frontier.
464. 13:59this is customer support. Software
465. 14:02engineering uh would be like this would
466. 14:04be the frontier of like maybe software
467. 14:05engineering, maybe this is customer
468. 14:06support or whatever. But then there's a
469. 14:08lot of stuff that's like terrible at
470. 14:10like like this scale or this scale or
471. 14:12and so on. And you know you Ali, you see
472. 14:16a lot of
473. 14:17>> we're here man
474. 14:19already.
475. 14:19>> We're over there.
476. 14:20>> We're here. We're here.
477. 14:22>> Um
478. 14:22>> they kind of admitted to it.
479. 14:23>> That's right. That's right. That's
480. 14:25right.
481. 14:25>> Reluctantly.
482. 14:26>> Yeah.
483. 14:26>> Yeah. So, you know, you you've got what,
484. 14:29like 6,000 7,000 customers? Those are
485. 14:3110,000 customers.
486. 14:32>> No, we have probably 20,000 customers.
487. 14:34>> 20,000 customers. Sorry.
488. 14:35>> Plus, yeah.
489. 14:36>> As you see this and you have that's a
490. 14:38very good sample of the entire universe
491. 14:39of what's happening. So, in that sample
492. 14:41of 20,000 customers,
493. 14:43>> what are what are areas where AI is like
494. 14:45hitting home runs and like working as
495. 14:48advertised?
496. 14:49>> Yep.
497. 14:49>> And what are areas where the frontier is
498. 14:52still uh rough and it's not working? The
499. 14:54PC's are failing.
500. 14:55>> Yeah. Look, it's not AI's fault. I mean,
501. 14:57most companies are somewhere here.
502. 14:59>> I think we have AGI, but I think most
503. 15:01companies, if you look at how much are
504. 15:02they, maybe they're having AI is helping
505. 15:04me in some tasks. That's what most
506. 15:06companies are doing. That's just how it
507. 15:08is.
508. 15:08>> Mhm.
509. 15:08>> And uh it's because that context isn't
510. 15:11there in the model.
511. 15:13>> So, the model can't do it. Take support,
512. 15:15which everybody said, okay, that's going
513. 15:16to be dead. Support is like gone. Yeah.
514. 15:18>> Right. Support is very hard.
515. 15:20>> Support are l literally the things that
516. 15:21humans don't know what to do. Like they
517. 15:23they get stuck. So, take data bricks.
518. 15:24Databix offers support.
519. 15:26>> Datab is a company that offers.
520. 15:28>> It's a platform, advanced platform where
521. 15:30you can do data science, machine
522. 15:31learning, you can do advanced things on
523. 15:33the platform. These are smart people who
524. 15:35make, you know, big salaries. They have
525. 15:37education, you know, they have data
526. 15:39science education. They're trying to use
527. 15:40data bricks and maybe they get stuck.
528. 15:42So, their machine learning models
529. 15:43doesn't have, you know, the right it's
530. 15:46not getting the [clears throat] right F1
531. 15:47score or, you know, something like that
532. 15:49>> and they're stuck and they tried
533. 15:50everything. They call our support.
534. 15:53So it's pretty hard to automate that.
535. 15:55You can't actually give it to none of
536. 15:56the current support automation. We tried
537. 15:58them all
538. 15:59>> companies all of them immediately even
539. 16:01actually when they start talking to us
540. 16:03as soon as they know who we are we're
541. 16:04like whoa whoa we can't help you like
542. 16:06you go do get out of here [laughter] you
543. 16:09know uh so uh so yeah most of the world
544. 16:12is over here
545. 16:13>> but it's because we don't have the
546. 16:14context. If the AI could have all the
547. 16:16context of how our support engineers at
548. 16:18data bricks operate
549. 16:19>> then the AI could do it.
550. 16:21>> Yeah. it just doesn't have it.
551. 16:22>> Yeah. You know, one of the things we
552. 16:23used to say at panel is your AI strategy
553. 16:25starts at your data strategy. Yes. You
554. 16:27got to get the roads paved and have the
555. 16:29data flowing and
556. 16:31>> is that you know if you were to bucket
557. 16:33the best enterprises who are like maybe
558. 16:35like starting to head towards the right
559. 16:36in your customer base of 20,000
560. 16:39>> what is common between the ones who are
561. 16:41making it work
562. 16:42>> and the Ferraris are flying
563. 16:44>> um and and and and ones where I'm I'm
564. 16:47guessing it's a context problem for the
565. 16:48ones that it's not working and what does
566. 16:50it take to get the context working?
567. 16:51>> Yeah, it's very hard. It's a human
568. 16:53problem like it's not an AI problem. we
569. 16:55already have AGI. It's a human problem.
570. 16:57I I don't see anyone really doing an
571. 16:58excellent job at this.
572. 17:00>> You have to kind of rewire all your
573. 17:01processes in the organization uh to to
574. 17:03be able to do it. This is like well
575. 17:05known. I mean my favorite is there's an
576. 17:07article actually that I recommend people
577. 17:08reading from 1990 uh produced by
578. 17:11actually Stanford professor or
579. 17:13researcher. Um it's called I think um
580. 17:17you know from the dynamo to the
581. 17:18computer.
582. 17:19>> Okay, check it out. So dynamo to
583. 17:21computer and it looks at different uh
584. 17:23sort of u technological revolutions and
585. 17:26how long it h how long it took for them
586. 17:28to have impact on productivity of econ
587. 17:30of the economy
588. 17:31>> and it's just you know takes just
589. 17:34forever like when the PCs came out the
590. 17:36joke was the Nobel laureate economist um
591. 17:40you know uh um uh Richard Solo said that
592. 17:44>> computers or PCs you can find them
593. 17:46everywhere except in uh the productivity
594. 17:49statistics
595. 17:51you know, like it just doesn't show up
596. 17:53in the statistics. Um,
597. 17:56>> why people were buying PCs and they were
598. 17:58using them as typewriters.
599. 18:00>> So, they would have people type on PCs
600. 18:02but then print out the sheets and then
601. 18:04put them in folders and then have
602. 18:06assistants that like index them and do
603. 18:08things. So, like you didn't see any
604. 18:10productivity gains from it.
605. 18:11>> And same thing with if you look at the
606. 18:13industrial revolution, same thing
607. 18:15happened. Uh, you know, we had these
608. 18:17steam engines and the steam factories
609. 18:20were sort of super dense and they were
610. 18:22running like with these, you know, they
611. 18:24were called the line uh shafts
612. 18:27>> which were like these things that
613. 18:28rotate.
614. 18:29>> Mhm.
615. 18:29>> When the electric engine came, that's a
616. 18:31dynamo.
617. 18:32>> Uh, it took 40 years before they saw any
618. 18:34productivity gains in the in the
619. 18:36economy.
620. 18:37>> Wow.
621. 18:37>> Yeah. Check it out. This is in that
622. 18:39article. It took from 1880.
623. 18:40>> The diffusion took 40 years. from 1880
624. 18:43to 1920 when the electric engine came
625. 18:46>> uh to see impact. So what they were
626. 18:47doing is they were going to these
627. 18:48factories that already were these line
628. 18:49shaft factories
629. 18:51>> that were these dense factories where
630. 18:52you have a steam engine that's rotating
631. 18:54this line shaft and it's rotating these
632. 18:56belts and then everything is working you
633. 18:58have these multiple stories um and all
634. 19:01they did is just like the PC they use
635. 19:03typewriter they would replace the steam
636. 19:04engine with an electric engine
637. 19:07>> and that doesn't just like replacing the
638. 19:08PC with you don't get any um
639. 19:11productivity gains it took till 1920
640. 19:14>> but maybe it was 1915 but I'm roughly
641. 19:17until they realize we have to change the
642. 19:19whole factory floor.
643. 19:20>> We have to move the factories out of the
644. 19:22cities.
645. 19:22>> We have to have like floor plans that
646. 19:24are much bigger cuz now we can
647. 19:27>> distribute the electricity. Electricity
648. 19:29is much more it doesn't you know it's
649. 19:31not like the um the torque that has you
650. 19:34know inefficiency. Uh we can spread it
651. 19:36out. We can have floor pans that are big
652. 19:38and we can run different parts of the
653. 19:39factory at different rates. Unit drive
654. 19:41versus group drive. Um took a very long
655. 19:44time.
656. 19:44>> Yeah. That's what's going to happen.
657. 19:45Same same thing now. Rewiring. I know it
658. 19:48because I have 20,000 customers and I
659. 19:49talked to them. I was late to this
660. 19:50meeting because I was meeting one of the
661. 19:52CEOs of one of the big banks and same
662. 19:55problem. He has the same problem. All
663. 19:57the organizations I work with have the
664. 19:58same problem. They're like I'm not
665. 20:00seeing any advant like I don't see.
666. 20:02They're all like AI is amazing. It's
667. 20:03coming. It's like I need it. I need to
668. 20:05do that. But they're like I don't see
669. 20:06any productivity gains in my
670. 20:07organization. You know what the hell am
671. 20:09I doing wrong?
672. 20:10>> And I tell them we have AGI and they're
673. 20:11like what?
674. 20:12>> Like that is not true. Like we don't see
675. 20:15anything.
676. 20:15>> It's a very tough problem because you're
677. 20:17like, "Hey, I got the brain, but I got
678. 20:18to rebuild the human body."
679. 20:20>> Yeah.
680. 20:20>> The hands, the legs.
681. 20:21>> Yeah. Let me give you the body.
682. 20:23>> Yeah. Let me give you an example from
683. 20:24data bricks.
684. 20:25>> So, databicks helps you get data from
685. 20:27all the different systems like
686. 20:28Salesforce, workday, and so on. Collect
687. 20:30them in one place,
688. 20:31>> secure it, and then do AI on it. Like
689. 20:34you can do predictions, you can build
690. 20:35predictive models. That's what database.
691. 20:37>> So, we built connectors to all these
692. 20:38systems.
693. 20:39>> These connectors are it would take us
694. 20:41three quarters to build a production
695. 20:43connector. We're good at this what we do
696. 20:44for a living. We build these connectors
697. 20:46like we can build the connector from
698. 20:47data bricks to Salesforce production
699. 20:48ready.
700. 20:49>> It would take us three quarters so 9
701. 20:51months to do that
702. 20:53>> shipped secure
703. 20:54>> nice with its own. That's like that's
704. 20:55what we did.
705. 20:56>> So you know as uh you know uh the LLMs
706. 21:00got faster and faster and faster I
707. 21:01started sort of experimenting with this
708. 21:02myself and I was like oh I could write a
709. 21:04connector in two days. So I went to the
710. 21:06team that builds this and um and I was
711. 21:09like hey I can do this in two days. How
712. 21:11come it takes you guys three quarters?
713. 21:12They're like, "Okay, great point. Let us
714. 21:14come back to you." So, they went and
715. 21:16they thought about it and they came back
716. 21:17in two weeks and they said, "Okay,
717. 21:19you're right. Uh, it's but you're also
718. 21:22not right." We looked at it and yeah,
719. 21:25this AI is useful. We can compress it
720. 21:27down from 3/4 by one and a half month.
721. 21:29So, we can get it from 9 months to 7 and
722. 21:311/2 months.
723. 21:32>> That's it.
724. 21:33>> That's it. I'm like, well, I can do it
725. 21:35in two days. And like no no no no
726. 21:36offense to you but you know this is
727. 21:38production code and it really actually
728. 21:40works and you know we have like customer
729. 21:43feedback and you know it's like secure
730. 21:46and you know you wrote some toy God
731. 21:48knows what that I mean no offense you're
732. 21:50great but you know let us let us
733. 21:52>> that's a missing link.
734. 21:53>> Uh so I was like a man this is kind of
735. 21:55depressing but yeah I'll take the one
736. 21:56and a half month improvement and you
737. 21:57know maybe it's something but maybe I'm
738. 21:58just stupid and I don't get it.
739. 22:00>> Yeah.
740. 22:00>> Then I found another guy in the company.
741. 22:02We went to him and we sort of said,
742. 22:03"Hey, can you look at this problem?" And
743. 22:05he's very first principal. He's a very
744. 22:07smart guy and he doesn't care about all
745. 22:09this like you know fluff. He's like he
746. 22:11cuts through the fluff and he cut
747. 22:12through the fluff and he worked with a
748. 22:15team and he came back and they said,
749. 22:17"Hey, after looking at the problem we
750. 22:19can do seven connectors in one quarter."
751. 22:21>> Boom.
752. 22:22>> Yeah.
753. 22:22>> Let's go. What is the difference?
754. 22:24>> So what's the difference? Okay. So what
755. 22:25he did is he he went from first
756. 22:26principles with some team members and
757. 22:28they looked at it and they said okay
758. 22:30first quarter they're just sending our
759. 22:32very expensive very smart Stanford
760. 22:34educated product managers out to the
761. 22:37customers to talk to the customers and
762. 22:39collect feedback what exactly is your
763. 22:40requirements how do you use Salesforce
764. 22:42and so on that takes a full quarter at
765. 22:44the end of that quarter our amazing
766. 22:46smart uh product managers come back with
767. 22:48like a 60 70 80 page super nice report
768. 22:51on exactly all the requirements
769. 22:52>> okay so you're blocked for a whole
770. 22:54quarter so for sure You can't doll's law
771. 22:56you can't compress it below below that
772. 22:59>> then codew writing starts but we have to
773. 23:01test this stuff so testing requires you
774. 23:02to set up Salesforce workday Netswuite
775. 23:04but those are not software by data
776. 23:06bricks so we're not very good at that
777. 23:07that takes a very long time and it's
778. 23:08hard to find people to do that data
779. 23:10bricks so that again is like a process
780. 23:12that takes a long time for us to stand
781. 23:14up and it's very errorprone so we
782. 23:15couldn't do that either um and then we
783. 23:18have one person for each connector
784. 23:21>> they go on vacation they get sick you
785. 23:23know so on so all of So what he did is
786. 23:25he just from first principles looked at
787. 23:26it and said we're going to just rewire
788. 23:28all of this. And lots of people didn't
789. 23:29like this. They were unhappy about it.
790. 23:30But he said that u you know the product
791. 23:34requirements instead of one quarter
792. 23:36we're just going to take one week and
793. 23:37quickly write down whatever we have. We
794. 23:39might get things wrong but because the
795. 23:40software is so fast to write we can
796. 23:42rewrite it again.
797. 23:43>> Right?
798. 23:43>> So let's iterate faster. Uh the standing
799. 23:46up the Salesforce instances let's
800. 23:48outsource that to firms that can do that
801. 23:50for us and we can just pay them a lot
802. 23:51and they do it in parallel. So we can
803. 23:53shrink that as well. And then one person
804. 23:55per connector. Let's change it. Let's
805. 23:56have seven people, seven connectors, and
806. 23:58then they all work on all the connectors
807. 23:59together. So we don't have what's
808. 24:01called, you know, bus factor one.
809. 24:03>> If someone is hit by a bus,
810. 24:04>> the whole project is not stopped. Right.
811. 24:06>> Right. Uh so um so yeah so got it all
812. 24:10done into one quarter and seven seven
813. 24:12connectors shipped and you know so but
814. 24:16this had nothing to do with uh like
815. 24:18really it didn't have anything to do
816. 24:19with AI or AGI or smarter models or
817. 24:22super intelligence or gi gigantic like
818. 24:25you can have the next like GPT7 or OPU 6
819. 24:29would not have helped us
820. 24:31>> u do this better we needed to do those
821. 24:33make those changes
822. 24:34>> and that's like a human refactoring
823. 24:36problem and process change and um so
824. 24:38this is what the whole world is going
825. 24:39through. So%
826. 24:41>> that's what you need to do well if you
827. 24:42want to if you want to succeed. Some are
828. 24:43doing it better, others are not.
829. 24:46>> Hamilton Helmer actually talks about
830. 24:47this quite a bit actually. So for all of
831. 24:49you who are picking assignment
832. 24:52option one and want to be investors,
833. 24:54Hamilton Helmer's uh book is a must
834. 24:56readad on process power. We were
835. 24:58debating this um before this if Ali you
836. 25:00had uh $100 to invest across what Jensen
837. 25:04calls the five layer stack energy chips
838. 25:07infra model and apps
839. 25:11where does value acrue if you were to
840. 25:13put a 100 bucks in the in in the index
841. 25:15of energy and chips and in so on uh with
842. 25:18let's say a long-term time frame where
843. 25:21does where would you put it how would
844. 25:23you allocate the $100 um and why
845. 25:26>> I'm a computer scientist I'm not an
846. 25:27investor. I don't give financial advice,
847. 25:30>> but
848. 25:30>> but you're allocating,
849. 25:32>> you know, you're you're
850. 25:33>> $500.
851. 25:33>> Yeah, you are allocating money data
852. 25:35bricks time, right? Data bricks is
853. 25:36across three of these.
854. 25:37>> Yeah. I would just say look, it's
855. 25:39obvious that the applications are going
856. 25:40to be the winners,
857. 25:41>> right?
858. 25:42>> So, I would put put it in the top. Uh
859. 25:44it's kind of like uh and I'll give you
860. 25:46some some guesses that you know, but who
861. 25:48knows actually it's very hard to
862. 25:50predict. So you would have to kind of
863. 25:52have a I would go early stage and I
864. 25:55would have a seed strategy and I would
865. 25:56invest in many many startups and I would
866. 25:58get most of them wrong but a few would
867. 26:00actually make it and they would be the
868. 26:01next Google or whatever. Um but you know
869. 26:05in 1990 like when I did my PhD
870. 26:09um in the early 2000
871. 26:11>> um I was in the networking field.
872. 26:14Networking was like the cool thing to
873. 26:15do. It was the advanced thing cuz the
874. 26:16internet was like you want to work on it
875. 26:18like the internet was the big thing at
876. 26:20the time and you want to the coolest
877. 26:22thing on the internet was
878. 26:24>> uh networking
879. 26:25>> and the hardest problem like the
880. 26:28smartest math brains were working on at
881. 26:30the time. We all knew what the future
882. 26:31would look like.
883. 26:32>> The future everybody knew what the most
884. 26:34important problem everyone's going to
885. 26:35work on is the what's called the
886. 26:36multiccast problem
887. 26:38>> which is yeah see [laughter]
888. 26:40>> it's problematic that no one knows what
889. 26:42that is today. We were we were clearly
890. 26:44wrong. So multiccast is, you know, you
891. 26:46want to broadcast from one source, let's
892. 26:49say a soccer game or a football game or
893. 26:51basketball game to the whole world
894. 26:52because everybody wants to watch it at
895. 26:53the same time.
896. 26:54>> We didn't know how to solve that
897. 26:55efficiently. So all of the smartest
898. 26:57brains in the world were trying to work
899. 26:58on this problem and bandwidth was scarce
900. 27:01>> while we were doing this. And by the
901. 27:02way, we actually, you know, had pretty
902. 27:04good problems and I started a company on
903. 27:05this
904. 27:06>> uh and we had great solution.
905. 27:09Unfortunately, the cost of bandwidth
906. 27:11just plummeted and they just deployed so
907. 27:13much fiber that no one this problem was
908. 27:15not a problem ever.
909. 27:16>> So, no one needed to buy this software.
910. 27:18So, it was complete waste of time. Uh
911. 27:20and at that time we thought the hardest
912. 27:22problems the most interesting things to
913. 27:23work on are Cisco routers, routing, BGP,
914. 27:26border gateway protocol, internet
915. 27:28protocol like you know queuing theory,
916. 27:30quality of service, these kind of
917. 27:31things. Those are like the most
918. 27:32interesting things you can because we
919. 27:33had tunnel vision on the internet
920. 27:36>> and the what is the internet? Well, at
921. 27:38the time it was the internet protocols
922. 27:39and those things. No apps really
923. 27:41existed, right?
924. 27:42>> So, we were all focused on that. And
925. 27:44today, everybody's focused, I would say,
926. 27:46on
927. 27:46>> I think like, you know, well, I think
928. 27:48chips and you know, I think
929. 27:49infrastructure,
930. 27:51>> yeah, I think people are really right
931. 27:52now the hot new thing is like Nvidia,
932. 27:54OpenAI, Anthropic, Deep Mind, these are
933. 27:56the things everybody's focused on. AGI,
934. 27:58super intelligence. That's
935. 28:00>> what I said at the beginning. But, uh,
936. 28:02on the internet, there were like really
937. 28:04weird things. Yeah.
938. 28:05>> That took off. The really weird things
939. 28:07that took off were like taxi business,
940. 28:10you know, which is Uber.
941. 28:12>> Uber. Yeah.
942. 28:12>> Uh or selling books, which is the lamest
943. 28:15thing ever, but that became Amazon,
944. 28:18which became AWS.
945. 28:19>> Yeah.
946. 28:20>> You know, uh or uh renting your bedroom
947. 28:23to people like, you know, that's Airbnb.
948. 28:26Uh and or um sending people short text,
949. 28:31>> right,
950. 28:31>> which became Twitter,
951. 28:32>> right?
952. 28:32>> You know, right? uh these are like and
953. 28:35if you said them in those words in 2000
954. 28:37to people people would say you're out of
955. 28:38your mind like you're insane you're full
956. 28:40of it uh but that's those were the great
957. 28:42ideas of the time those are the ones
958. 28:43that came so I think it's the same thing
959. 28:44here
960. 28:45>> right
961. 28:45>> uh to throw a few of them out there um I
962. 28:48think healthcare is like 17% of US GDP
963. 28:52>> um you know
964. 28:54>> we we we all still unfortunately will
965. 28:57die and we all care about our health and
966. 28:59the health of our loved ones I think
967. 29:02there's a huge we have, you know, uh the
968. 29:05propensity to pay for this. Like we'd
969. 29:07pay anything to be able to save lives or
970. 29:09of our loved ones or our own lives or
971. 29:12our own health issues. Uh and it's not
972. 29:14particularly well done today. Surprise
973. 29:16surprise, you know, healthcare is not
974. 29:18like awesome. Uh so imagine a company
975. 29:20that has seen a million patients
976. 29:22>> like I have seen 100 million patients
977. 29:25with your kind of genetic composition
978. 29:26and the kind of issues that you might
979. 29:28have in the future and I can help you
980. 29:30>> but what are you willing to pay for me
981. 29:32to help you with that? That could be a
982. 29:34company that's trillions of dollars
983. 29:35worth,
984. 29:35>> right?
985. 29:36>> Um to take something out of left field
986. 29:38that I think people think is really, you
987. 29:41know, not interesting and not but take
988. 29:43education.
989. 29:44>> Education actually in in VC space the
990. 29:47consensus has always been education is
991. 29:48like a terrible investment, right? Isn't
992. 29:50that like VC people say always like
993. 29:51never invest in education?
994. 29:53>> What's the last public market company
995. 29:55you know?
996. 29:56>> Yeah. What's the last trillion dollar
997. 29:57education company?
998. 29:58>> Not even 100 billion. Yeah.
999. 29:59>> Yeah. Yeah. Anything, right? Um so but
1000. 30:03most people have kids
1001. 30:04>> and you know more kids are produced and
1002. 30:07they they do need to go through uh get
1003. 30:10to get an education whether people
1004. 30:11believe it or not
1005. 30:13>> and um uh and people do care actually if
1006. 30:16the education for their kids are good or
1007. 30:17not. Elections are won and lost. There's
1008. 30:20cultural issues on these things like you
1009. 30:22know of what you you're allowed to teach
1010. 30:24my kids or not, right? Elections are won
1011. 30:25and lost on that. Not because it's a
1012. 30:27stupid topic, because it matters. Like
1013. 30:28what are you teaching my kids matters
1014. 30:30and are my
1015. 30:31>> kids being brainwashed to do the right
1016. 30:33thing or the wrong thing or are they
1017. 30:34gonna be do the are they you know well
1018. 30:36equipped to get the jobs of the future.
1019. 30:38>> Uh I think if if there's a company that
1020. 30:40can provide amazing education, right?
1021. 30:42>> Uh using AI um I think right
1022. 30:45>> people a lot of people will pay for that
1023. 30:47and if it's like proven that that does a
1024. 30:49better job than um than than you know
1025. 30:53whatever they're getting right now. Um
1026. 30:55just two flavors of like obvious
1027. 30:57companies that I think will exist and
1028. 30:58they could be trillion dollar companies
1029. 31:00>> if they do it well. They will have data
1030. 31:01mode.
1031. 31:02>> Yeah.
1032. 31:03>> Um they will have economies of scale
1033. 31:05mode.
1034. 31:05>> There's like winner takes it all kind of
1035. 31:08dynamics in those markets at least
1036. 31:10>> in countries in geos.
1037. 31:12>> Yeah.
1038. 31:12>> So uh so I think the value acrews to the
1039. 31:14top.
1040. 31:15>> Yeah.
1041. 31:15>> We can't wait.
1042. 31:16>> But I'm not an investor.
1043. 31:17>> Yeah.
1044. 31:17>> Can't wait for that to happen.
1045. 31:19>> Would you push back? No, I think I mean,
1046. 31:20look, I've I've written extensively
1047. 31:22about this, eagerly waiting for this
1048. 31:24what I call the blue triangle to uh to
1049. 31:27invert. Uh I don't know if you've seen
1050. 31:28this, but basically this is uh all of
1051. 31:31the money in AI
1052. 31:32>> Yeah.
1053. 31:32>> is with one guy.
1054. 31:34>> Yeah.
1055. 31:34>> That's why Jensen's so happy all the
1056. 31:36time as as you know.
1057. 31:37>> Yeah.
1058. 31:38>> Um
1059. 31:40>> these guys are fighting for
1060. 31:42>> dollars. There's like no money there.
1061. 31:44There's very little money here. I mean,
1062. 31:45people are making some money here and
1063. 31:47>> uh so we'll see. But but that's the bet.
1064. 31:49the bet is that this thing will look
1065. 31:50like a more sustainable
1066. 31:51>> Yeah, it will go that way. I mean% all
1067. 31:54value in Silicon Valley and in tech and
1068. 31:56in technology moves up the stack all the
1069. 31:58time.
1070. 31:58>> Yeah.
1071. 31:59>> Like you know you even look at the
1072. 32:00greatest companies like okay the company
1073. 32:02that created the PCs IBM was like the
1074. 32:04greatest market cap and all the value
1075. 32:05accured there. But then that became
1076. 32:06commoditized then it became the software
1077. 32:08on top of it which is like the operating
1078. 32:09systems and the Microsoft of the world
1079. 32:11and so on. Then you know here at
1080. 32:13Stanford actually a while back it was
1081. 32:15like 20 years ago VMware which is how do
1082. 32:17you virtualize that software and but
1083. 32:19that became commoditized and then like
1084. 32:20you know so it keeps moving up the stack
1085. 32:22all the time.
1086. 32:22>> Yeah
1087. 32:22>> that's how that's that's how it's going
1088. 32:24to be here too
1089. 32:25>> 100%. And you know the big one of the
1090. 32:26forces that is commoditizing this you've
1091. 32:28spoken about this is open source.
1092. 32:30>> Open source is uh getting pretty good.
1093. 32:32>> Yep.
1094. 32:32>> This blue line is open source.
1095. 32:34>> The the gap is closing. This is like
1096. 32:36what three three four months.
1097. 32:38>> Yeah.
1098. 32:38>> This gap is now like a month.
1099. 32:40>> Yeah. And but still people are spending
1100. 32:43so much money on these frontier models.
1101. 32:46>> Yeah.
1102. 32:46>> People cannot wait to get their hands on
1103. 32:4847 from open from from cloud or 55 from
1104. 32:52GPT. But
1105. 32:54>> then there's this whole economy of of
1106. 32:55very good open source models.
1107. 32:58>> What what do you make of all this? Like
1108. 32:59you on one side you've got people
1109. 33:01earning what 30 billion now or maybe 40
1110. 33:03billion entropic or
1111. 33:05>> but on the other side this open source
1112. 33:06stuff is like nearly free. Obviously
1113. 33:07you've got to pay the hosting.
1114. 33:09>> Yeah. How do you think this shakes out?
1115. 33:11Will will that model layer, the
1116. 33:13proprietary model layer acrew any value?
1117. 33:15>> No, I think it's going to be valuable.
1118. 33:17Yeah.
1119. 33:17>> And I think people will want it whether
1120. 33:18it's open source or not. Let's put that
1121. 33:20aside for a second. I think there will
1122. 33:21be token factories
1123. 33:22>> which serve this stuff up. It's just
1124. 33:24like the cloud,
1125. 33:25>> right? I don't think like I think we
1126. 33:27foolish to say you all will have your
1127. 33:28own little mini data center in your
1128. 33:30living rooms and you're going to run you
1129. 33:33know your own PCs and you're going to
1130. 33:34insert GPU cards that you buy at home
1131. 33:36and you're going to run this yourself
1132. 33:37>> or on your phone or MacBook or the edge
1133. 33:39>> some of them will exist it will come to
1134. 33:41the edges but I do think there'll be
1135. 33:42like big yeah
1136. 33:43>> centralized data centers where this
1137. 33:45happens
1138. 33:46>> but we haven't discussed are they
1139. 33:47running open source models or are they
1140. 33:48running proprietary models and um and
1141. 33:52here's a fun fact so moonshot the
1142. 33:54Chinese company released Kimmy. Yeah. Uh
1143. 33:562.6.
1144. 33:57>> Very good model.
1145. 33:58>> Two days two days ago or two days ago.
1146. 33:59>> Yeah, Tuesday. Uh yeah, Tuesday. Uh so
1147. 34:01two days ago they released three days
1148. 34:03two days ago they released 2.6. In
1149. 34:05January they released 2.5.
1150. 34:08>> And here's a fun fact. 2.6 that they
1151. 34:10released on Tuesday is the best model
1152. 34:12ever in the history of mankind ever
1153. 34:14produced. Frontier non-frontier if it
1154. 34:16just had been released in January.
1155. 34:18>> Yeah.
1156. 34:18>> But open source will be here.
1157. 34:20>> Yeah. And it will apply pricing
1158. 34:21pressure. And this business of frontier
1159. 34:24models,
1160. 34:25>> that core business of providing frontier
1161. 34:27models is going to be economies of scale
1162. 34:29game.
1163. 34:30>> And you'll have to uh do it at small
1164. 34:33margins. It's like an Amazon.com book
1165. 34:36selling business. Yeah.
1166. 34:37>> That's what it's going to look like in
1167. 34:38the future. Therefore, there not going
1168. 34:39to be that many people doing it.
1169. 34:40>> Yeah.
1170. 34:41>> It's just like an Amazon.com. And gross
1171. 34:44margins are going to be tiny and
1172. 34:46operating margins are going to be small.
1173. 34:47That's my
1174. 34:48>> Yeah.
1175. 34:48>> take.
1176. 34:49>> Yeah. Yeah. I think so too. Um, three
1177. 34:52rapidfire questions before we uh wrap.
1178. 34:55>> Your uh favorite AI product that you use
1179. 34:57every day.
1180. 34:58>> I don't know. That's a tough one. Uh um
1181. 35:01I mean I use all of these.
1182. 35:02>> Yeah.
1183. 35:02>> Uh you know I actually like cursor. I
1184. 35:05know that's like everybody loves cloud
1185. 35:06code.
1186. 35:06>> I like the diffs and how how how it
1187. 35:08works. So like on coding I use like a
1188. 35:10combo of those.
1189. 35:11>> I still kind of like it. Yeah.
1190. 35:12>> Are you still using it after uh Elon
1191. 35:14owns it?
1192. 35:15>> No, I stopped. No, of course. Yeah.
1193. 35:16[laughter]
1194. 35:18>> Yeah. because you're going to lose
1195. 35:19access to entropic and open tokens
1196. 35:21through cursor I presume.
1197. 35:23>> Yeah. Yeah. Awesome. It's great. Good,
1198. 35:26good, good, good supporter.
1199. 35:28>> Um,
1200. 35:28>> you've been
1201. 35:29>> the truth is I do use data bricks as
1202. 35:31gener products. This is the truth
1203. 35:32>> because you know it just most of my
1204. 35:35>> inside data bricks most of my decisions
1205. 35:36are like numerical and quantitive in
1206. 35:38nature like should we do this? What's
1207. 35:40the ROI on this? What's the cost on
1208. 35:41that? What it's going to cost us? What's
1209. 35:42the So, I need something that can
1210. 35:44understand numerical data and time
1211. 35:46series data. So Genie is like really
1212. 35:48good for that. So that's that's what I
1213. 35:50honestly go to quite a bit
1214. 35:51>> quite often.
1215. 35:52>> Right.
1216. 35:53>> Um future for data bricks. You've been
1217. 35:56at this for for 15 years or so. Um what
1218. 35:59is your vision for the next decade for
1219. 36:01data bricks?
1220. 36:02>> Well, I think the cost of software is
1221. 36:03going down.
1222. 36:04>> Yeah.
1223. 36:04>> And so barriers to entry and switching
1224. 36:07costs are going down. So there is an a
1225. 36:09SAS apocalypse of sorts, but not all
1226. 36:11software is going to be dead.
1227. 36:12>> Yeah. uh we would love to partake in
1228. 36:14that and
1229. 36:15>> right
1230. 36:15>> kill some software
1231. 36:16>> right
1232. 36:17>> right right any advice for uh students
1233. 36:20in the room who are about to uh make
1234. 36:23career decisions
1235. 36:24>> yeah I think don't don't be worried
1236. 36:26about the fear-mongering don't be
1237. 36:28stressed out take it easy um I think
1238. 36:32that uh that's those I was very stressed
1239. 36:34doing my PhD in the early 2000 I thought
1240. 36:36like the world is ending with the
1241. 36:37internet and everything and uh you know
1242. 36:39working on this most important problem
1243. 36:41that we all knew was the most important
1244. 36:42problem which was the multiccast problem
1245. 36:44which none [laughter] of you which none
1246. 36:46of you have heard of
1247. 36:48>> uh turn out not to be a problem. Uh but
1248. 36:51I think one interesting thing is that in
1249. 36:532000 we had the internet. In 2009 Airbnb
1250. 36:56was started,
1251. 36:56>> right?
1252. 36:57>> Okay. But there's no reason why Airbnb
1253. 36:59should start in 2009. Airbnb could have
1254. 37:01started. We I've made this argument to
1255. 37:02you. Airbnb could have started in 2001.
1256. 37:04>> Yeah.
1257. 37:04>> There's nothing like we needed something
1258. 37:06additional to happen in the world.
1259. 37:08>> You know, uh Airbnb could have happened
1260. 37:11and disrupted hotel businesses in 2001.
1261. 37:13Yet it took 9 years for someone to have
1262. 37:15that idea,
1263. 37:16>> right?
1264. 37:16>> And that that was Brian. And by the way,
1265. 37:18Brian is not like he sat there and he
1266. 37:19was taking a Stanford class thinking
1267. 37:21about like a case study project.
1268. 37:23>> Uh Brian needed like bed and breakfast,
1269. 37:25right?
1270. 37:26>> And like he was like conference or
1271. 37:27something. Yeah, it [clears throat] was
1272. 37:28at some conference and he's like why is
1273. 37:29this so hard? Like can't I just solve
1274. 37:30this myself? So it took nine years to
1275. 37:32come up with that good idea. So I think
1276. 37:34good ideas are very hard to come by
1277. 37:36actually. I think humans are very bad at
1278. 37:38coming up with great ideas,
1279. 37:40>> right?
1280. 37:40uh and we have like this tunnel vision
1281. 37:42and we focus on the wrong problems like
1282. 37:44we did with multiccast in my earlier you
1283. 37:46know my PhD was really stupid um so uh
1284. 37:52chill out and take a long-term
1285. 37:55perspective and uh you know work on the
1286. 37:58things that you think will have
1287. 37:59long-term good impact. I think Jeff
1288. 38:01Bessos did it pretty well uh when he was
1289. 38:04an investment banker in Wall Street and
1290. 38:07uh and he said, "Hey, zooming out,
1291. 38:10what's like the big thing that's
1292. 38:11happening? It's the internet."
1293. 38:12>> Yeah.
1294. 38:12>> And then he said, "Hey, let's just make
1295. 38:14a secular bet on internet's going
1296. 38:15there's going to be more and more
1297. 38:16internet. So, it's going to slowly over
1298. 38:18time disrupt things."
1299. 38:19>> So then he said said, "Okay, can we in
1300. 38:21the long run probably purchasing can
1301. 38:24move more to the net. Maybe not right
1302. 38:25now." And then he started with he was
1303. 38:27very modest and he started with kind of
1304. 38:29the dumbest thing you could possibly
1305. 38:30[laughter] no one like the unsexiest
1306. 38:32thing which was a complete commodity
1307. 38:33that looks identical and there's no
1308. 38:35differentiation which is books.
1309. 38:36>> Yeah.
1310. 38:37>> And he just started with that
1311. 38:38>> and he just bet on that secular trend
1312. 38:40and every year it was more and more
1313. 38:41right
1314. 38:42>> and you know and now it's like
1315. 38:44everything on the planet. It's the
1316. 38:45everything store. So, kind of think long
1317. 38:47term like that and don't be swayed by
1318. 38:49the coolest thing that everybody's like
1319. 38:51right now um sort of making lots of
1320. 38:54noise on Twitter on because chances are
1321. 38:56it's probably something like multiccast.
1322. 38:59[laughter]
1323. 38:59Yeah.
1324. 39:00>> Awesome. Well, thank you so much for
1325. 39:02staying longer, folks. Thank you all.