WEBVTT 00:00:00.480 --> 00:00:03.837 you know, I would separate two things there. There's I think inevitably as we, 00:00:03.920 --> 00:00:08.397 you know, talk about AI in the future, this kind of sci-fi sheen on on on on, 00:00:08.480 --> 00:00:11.677 you know, the on the whole thing. And that's, you know, in a way it's helpful 00:00:11.760 --> 00:00:15.437 as a metaphor, but also it can be distracting because, you know, it just 00:00:15.520 --> 00:00:19.677 it just kind of conjures everyone's mind into this into this kind of uh, you 00:00:19.760 --> 00:00:25.117 know, into this kind of uh, you know, future dreamy mode. Um, uh, but, you 00:00:25.200 --> 00:00:28.397 know, I think there's some things that are real. One thing that's real is that, 00:00:28.480 --> 00:00:31.357 you know, every three months we make a model that's smarter than the month than 00:00:31.440 --> 00:00:35.037 the model we made before. And indeed, the models are slowly getting to where 00:00:35.120 --> 00:00:38.637 they're, you know, better than most humans at most things. Maybe at some 00:00:38.720 --> 00:00:41.197 point better all humans than all things. 00:00:41.280 --> 00:00:45.757 The other thing I think is real is, you know, the model is not building itself 00:00:45.840 --> 00:00:50.157 in the sense that like I have a specific copy of the model that's like making a 00:00:50.240 --> 00:00:53.757 specific copy of the the, you know, the the next model. you know, we don't have 00:00:53.840 --> 00:00:56.317 our self-replicating, you know, nanobots 00:00:56.400 --> 00:00:58.797 or robots or or whatever, but but what 00:00:58.880 --> 00:01:01.117 absolutely is happening is that within 00:01:01.200 --> 00:01:03.837 Anthropic and elsewhere, we are using 00:01:03.920 --> 00:01:07.037 Claude to help build products on top of 00:01:07.120 --> 00:01:10.157 quad and to help train the next Quad to 00:01:10.240 --> 00:01:13.357 help serve cla um you know, six months 00:01:13.440 --> 00:01:15.837 ago I I made this prediction that, you 00:01:15.920 --> 00:01:20.797 know, in in six months 90% of code would be written by by by AI models. Some 00:01:20.880 --> 00:01:24.557 people think that prediction is wrong but within Enthropic and within a number 00:01:24.640 --> 00:01:29.037 of companies that we work with that is absolutely true. Um now 90 you know 00:01:29.120 --> 00:01:33.437 >> so you're saying that 90% of all code at anthropics being written by the by the 00:01:33.520 --> 00:01:37.757 model today >> on on many teams you know not uniformly every 00:01:37.840 --> 00:01:41.197 >> something that you've kind of provided something kind of 00:01:41.280 --> 00:01:45.117 >> something auditable around that or like shown how that's happened. Yeah, the 00:01:45.200 --> 00:01:50.397 teams play an editing and supervisory role, but you know Claude can both you 00:01:50.480 --> 00:01:54.157 can say write this feature and Claude will do it and the human kind of looks 00:01:54.240 --> 00:01:56.397 over the code or asks for help from 00:01:56.480 --> 00:02:01.597 another another model or another copy of Claude. Um uh the other thing is 00:02:01.680 --> 00:02:04.877 sometimes we have these long running bugs. you know, when we were training 00:02:04.960 --> 00:02:08.637 our our the most recently released Claude, it was like, you know, the 00:02:08.720 --> 00:02:12.877 cluster was broken. There was this bug, engineers were spending a few days on 00:02:12.960 --> 00:02:16.397 it, and you know, we just we just went to Claude and we said, I don't know, 00:02:16.480 --> 00:02:18.877 just kind of play around with the cluster and see if you can figure out 00:02:18.960 --> 00:02:23.037 what's wrong. And, you know, Claude found this just super obscure bug that 00:02:23.120 --> 00:02:26.477 the engineers had missed. So, this this was one area where it was like Claude is 00:02:26.560 --> 00:02:28.397 like having a very talented teammate. So 00:02:28.480 --> 00:02:31.837 if quad is now running writing 90% of 00:02:31.920 --> 00:02:34.317 all of the code in entropic do you need 00:02:34.400 --> 00:02:36.637 less engineers if 90% of all the code 00:02:36.720 --> 00:02:39.357 when will it be 95% or 99%. 00:02:39.440 --> 00:02:43.997 >> Yeah. So you know I you know I think people can misinterpret it as you know 00:02:44.080 --> 00:02:48.077 you know like claude writing 90% of the code means we're firing 90% of the 00:02:48.160 --> 00:02:51.517 software engineers. But you know if you're familiar with the principle of 00:02:51.600 --> 00:02:52.957 >> you have made some very aggressive 00:02:53.040 --> 00:02:59.917 comments around firing executives >> that that so 00:03:00.000 --> 00:03:01.757 um >> I just want to put two and two together 00:03:01.840 --> 00:03:07.357 to get four. Okay. >> Um uh so uh um you know if 00:03:07.440 --> 00:03:10.397 >> one thing I've learned in knowing you for so long is I have to come back and 00:03:10.480 --> 00:03:13.917 kind of like get that qualifying question again. Yes. because you're go, 00:03:14.000 --> 00:03:18.637 you know, you you you're you're going fast. So trying to I'm trying to keep up 00:03:18.720 --> 00:03:22.157 with you and it's amazing, but I want to just make sure I understand exactly what 00:03:22.240 --> 00:03:24.157 you're doing. >> Yes, absolutely. Let me let me let me 00:03:24.240 --> 00:03:27.517 take them one by one. So um uh you know, 00:03:27.600 --> 00:03:30.557 under comparative advantage, if Claude 00:03:30.640 --> 00:03:33.517 is writing 90% of the code, what that 00:03:33.600 --> 00:03:36.397 means usually is you need just as many 00:03:36.480 --> 00:03:41.117 software engineers, you might need more because they can they can then be more 00:03:41.200 --> 00:03:43.277 leveraged. they can focus on the 10% 00:03:43.360 --> 00:03:45.357 that's editing the code or writing the 00:03:45.440 --> 00:03:48.237 10% that's the hardest or supervising a 00:03:48.320 --> 00:03:52.637 group of a of AI models. And so what happens is, you know, you can just end 00:03:52.720 --> 00:03:54.957 up being 10 times more productive 00:03:55.040 --> 00:03:58.317 because the remaining 10% is is now a 00:03:58.400 --> 00:04:02.077 complement to the to the AI system that's that's getting much better at at 00:04:02.160 --> 00:04:04.557 getting much better at what it's producing. So So 00:04:04.640 --> 00:04:07.677 >> it's a rebalancing of the work is what you're really saying. 00:04:07.760 --> 00:04:08.957 >> It's it's not a replacement. 00:04:09.040 --> 00:04:10.077 >> It's not a replacement. It's a 00:04:10.160 --> 00:04:13.357 rebalancing. Okay. Now, as that 90% goes 00:04:13.440 --> 00:04:18.317 to 95 99 totally automated end to end, there may be some point at which all 00:04:18.400 --> 00:04:22.317 bets are off. >> What's your time frame for that? It's a pretty big jump there. 00:04:22.400 --> 00:04:24.717 >> It's very hard to say. Um, but you know, 00:04:24.800 --> 00:04:28.477 we're looking at it. We're looking at it closely. We're always looking for ways 00:04:28.560 --> 00:04:32.477 to encourage complimentarity. >> You see that in the short term or the long term? 00:04:32.560 --> 00:04:35.117 >> Um, I, you know, >> is that a 5year goal or a one year goal? 00:04:35.200 --> 00:04:37.117 >> I'm optimistic about 10ear goal. 00:04:37.200 --> 00:04:39.277 >> I'm optimistic about complimentarity in 00:04:39.360 --> 00:04:42.157 the short run. I you know I I I I I do 00:04:42.240 --> 00:04:45.197 worry as we get out to two years 5 years 00:04:45.280 --> 00:04:47.837 that across the economy not that there 00:04:47.920 --> 00:04:50.237 won't be anything for humans to do but 00:04:50.320 --> 00:04:52.477 that there will be enough disruption 00:04:52.560 --> 00:04:55.597 that that we'll have a you know a kind 00:04:55.680 --> 00:04:57.917 of labor disruption that is you know not 00:04:58.000 --> 00:05:00.397 fundamentally different from what we've 00:05:00.480 --> 00:05:02.797 seen with previous technologies but that 00:05:02.880 --> 00:05:08.157 operates faster that is broader across the economy because it relates to all 00:05:08.240 --> 00:05:09.997 >> you think that's in the next 2 to 5 years. 00:05:10.080 --> 00:05:11.437 >> Two to five years I would say. 00:05:11.520 --> 00:05:13.197 >> Okay. And at what point just coming back 00:05:13.280 --> 00:05:18.557 to my previous question, at what point is Enthropic the model writing 100% of 00:05:18.640 --> 00:05:22.477 the code? >> Yeah. Again, it's it's a tricky question 00:05:22.560 --> 00:05:26.317 because maybe even if the model's writing 100% of the code, the humans are 00:05:26.400 --> 00:05:31.880 still supervising the model or prompting it or clarif 00:05:34.560 --> 00:05:36.637 it's it's it's it's a surprisingly 00:05:36.720 --> 00:05:41.517 clever thing. I don't, you know, I I'm I I I am concerned particularly about 00:05:41.600 --> 00:05:45.597 people's ability to adapt as the jobs are changing fast. The jobs are changing 00:05:45.680 --> 00:05:49.117 so fast, but comparative comparative advantage is powerful. That's going to 00:05:49.200 --> 00:05:51.357 be one of our tools in this. >> Well, you've mentioned healthcare 00:05:51.440 --> 00:05:55.437 several times, but we've seen these kind of very conflicting studies recently in 00:05:55.520 --> 00:05:59.437 healthcare around accuracy, especially because these are word models. There 00:05:59.520 --> 00:06:01.437 seems to be some level of accuracy 00:06:01.520 --> 00:06:06.877 that's not 100%. and were you know you know and then when applied to some of 00:06:06.960 --> 00:06:10.397 these kind of other things like radiology is one where we saw these very 00:06:10.480 --> 00:06:12.717 interesting you know studies where we 00:06:12.800 --> 00:06:17.517 just didn't see the accuracy of a radiologist reflected in what the model 00:06:17.600 --> 00:06:21.597 was doing is that how you see it going forward or how did you reflect back on 00:06:21.680 --> 00:06:25.757 that >> yeah I I would say where we are now is I 00:06:25.840 --> 00:06:28.957 think of the models and I think a lot of people who use the models in practice 00:06:29.040 --> 00:06:30.397 >> you know where I'm going right where I'm 00:06:30.480 --> 00:06:34.477 trying to I'm trying to like understand Where are we really in the evolution of 00:06:34.560 --> 00:06:36.957 the world? >> Where's the arc? So, so I think a lot of 00:06:37.040 --> 00:06:41.037 you I I think and a lot of people who use the models in practice, I think 00:06:41.120 --> 00:06:43.597 think of the models as maybe a bit like 00:06:43.680 --> 00:06:46.237 a second opinion, right? Where you don't 00:06:46.320 --> 00:06:50.957 necessarily trust them alone. Just like if a doctor says something to me, I you 00:06:51.040 --> 00:06:54.637 know, I don't I don't necessarily take that as fact. They're an expert, but and 00:06:54.720 --> 00:06:58.797 they're often right, but they're not always right. Um uh so you know I'll 00:06:58.880 --> 00:07:00.477 tell you >> especially in this one category 00:07:00.560 --> 00:07:04.637 radiology it seems strange like it just can't get its 00:07:04.720 --> 00:07:07.437 >> I was going to say can't get its head around it but that's not really the 00:07:07.520 --> 00:07:11.357 right metaphor but doesn't have >> it does there is something about either 00:07:11.440 --> 00:07:15.997 it doesn't have enough data or it's not getting it's too nuanced. What is it? 00:07:16.080 --> 00:07:19.517 >> Yeah. So let me let me actually tell you a story about about one case where it 00:07:19.600 --> 00:07:23.357 where it where it did work. Um so you know we're talking about you know 00:07:23.440 --> 00:07:26.637 founding the company with with my sister. She recently had her second 00:07:26.720 --> 00:07:30.477 child while she was pregnant. Um, you know, pregnancy can suppress your immune 00:07:30.560 --> 00:07:34.637 system. Uh, was was viral. For some reason, it didn't occur to to any of 00:07:34.720 --> 00:07:37.997 them that it could be a bacterial infection. She took all her data. She 00:07:38.080 --> 00:07:42.717 uploaded it to Claude and, you know, Claude, you know, systematically listed 00:07:42.800 --> 00:07:45.917 the possibilities in the way that the doctors had somehow missed. 00:07:46.000 --> 00:07:49.517 >> That is amazing, right? How many of us have already used the models in some way 00:07:49.600 --> 00:07:51.757 to look at our healthcare? Raise your hand. 00:07:51.840 --> 00:07:53.837 >> Yeah, I think we're all having exactly that. 00:07:53.920 --> 00:07:56.797 >> That's that's a lot of people. It's amazing, right? Where you can kind of 00:07:56.880 --> 00:07:58.717 say especially how many of you are the 00:07:58.800 --> 00:08:04.237 hypochondrics that are doing that? Raise your hand. Yeah, I got a few. Okay, I 00:08:04.320 --> 00:08:07.197 know I'm >> But but but here's the point, which is 00:08:07.280 --> 00:08:10.477 Claude listed possibilities. If she had 00:08:10.560 --> 00:08:13.837 just blindly taken the advice of Claude, 00:08:13.920 --> 00:08:17.837 that would not have gone well. Um Claude was good for saying, "Hey, have you 00:08:17.920 --> 00:08:22.397 considered this? Have you considered that?" So, so it has this assistive 00:08:22.480 --> 00:08:25.277 augmentative ability >> especially in regards to words, right? 00:08:25.360 --> 00:08:28.557 It's very good on words. >> Words. Yeah. Yeah. I mean, you know this 00:08:28.640 --> 00:08:32.637 specialized stuff of reading images like I think that's more about specialized 00:08:32.720 --> 00:08:36.237 models like if you don't have the right data maybe it's not doing well again. 00:08:36.320 --> 00:08:39.597 It's less reflective of the >> is that the crux of the issue like in 00:08:39.680 --> 00:08:41.597 radiology because like when we're saying 00:08:41.680 --> 00:08:43.997 it's going to improve health care but we 00:08:44.080 --> 00:08:47.277 also saw this weird study where doctors 00:08:47.360 --> 00:08:52.237 who were using the models some in some cases were relying too much of the 00:08:52.320 --> 00:08:55.197 models and then ending up with bad outcomes. 00:08:55.280 --> 00:08:59.517 >> On the other side then we saw models that were like trying to read radiology 00:08:59.600 --> 00:09:03.357 but then were not able to achieve accuracy. There was like this kind of 00:09:03.440 --> 00:09:05.197 strange bifurcation between the two 00:09:05.280 --> 00:09:07.517 things. Yeah, you know, I think there's 00:09:07.600 --> 00:09:12.717 a lot still left to work out in what the role of the human is and what the role 00:09:12.800 --> 00:09:16.877 of the model is. And of course, that's changing as the models improve rapidly. 00:09:16.960 --> 00:09:20.237 But, you know, I think it's something you often see in new technologies where 00:09:20.320 --> 00:09:24.717 there's this great technology and then these practical issues of exactly how do 00:09:24.800 --> 00:09:29.037 you use it, exactly how do you interact with it? They take surprisingly long to 00:09:29.120 --> 00:09:31.117 work out. And you know if you look back 00:09:31.200 --> 00:09:33.917 at you know the early days when people 00:09:34.000 --> 00:09:38.157 you know kind of you know deployed deployed new technologies I don't know 00:09:38.240 --> 00:09:41.117 it's it's just it's often >> you think this is going to be about 00:09:41.200 --> 00:09:42.957 humans and agents working together. 00:09:43.040 --> 00:09:45.037 >> Yeah. No no no exactly just just the 00:09:45.120 --> 00:09:49.997 collaboration model needs to be worked out a bit more than it has in these 00:09:50.080 --> 00:09:51.997 practical areas. code is good to point 00:09:52.080 --> 00:09:54.637 to like code happens so fast right 00:09:54.720 --> 00:09:57.757 developers are early adopters and so 00:09:57.840 --> 00:09:59.997 that cycle we've gone through it in like 00:10:00.080 --> 00:10:05.037 months right you can see people talking in blog posts about oh you know what's 00:10:05.120 --> 00:10:08.957 the best way to use quad code do I want it to suggest and I make edits or I 00:10:09.040 --> 00:10:13.277 write code and it makes edits and these people riff off each other and you know 00:10:13.360 --> 00:10:16.557 over the course of three or four months they work out what the best way to use 00:10:16.640 --> 00:10:20.557 quad code is I think that same thing is happening in healthcare and radiology 00:10:20.640 --> 00:10:23.917 technology, but because there are a lot of barriers in these industries, because 00:10:24.000 --> 00:10:26.637 they're high stakes, that that process 00:10:26.720 --> 00:10:29.357 is working itself out much slower. 00:10:29.440 --> 00:10:33.117 >> So, that's really kind of getting down to it, right? That it's a collaborative 00:10:33.200 --> 00:10:35.677 thing. Like, here's the best way to use 00:10:35.760 --> 00:10:37.757 cloud code. Cloud code is writing the 00:10:37.840 --> 00:10:40.317 code, but here's really the best way as 00:10:40.400 --> 00:10:43.277 a developer now to work on cloud code. 00:10:43.360 --> 00:10:47.437 It's not writing 100% of the code. It's writing code now. I'm like checking 00:10:47.520 --> 00:10:51.357 here's here's and now now I'm going to have to maintain the code. I'm going to 00:10:51.440 --> 00:10:55.597 have to provide all the capabilities the whole infrastructure around the code. Is 00:10:55.680 --> 00:10:58.397 this the right way to think about it? I I think that's the right way to think 00:10:58.480 --> 00:11:01.357 about it right now. Um uh you know as as 00:11:01.440 --> 00:11:04.237 the models get stronger then the 00:11:04.320 --> 00:11:09.117 collaboration may change to more distant supervision by the human. Um so it's 00:11:09.200 --> 00:11:13.197 it's very hard to say what will happen even six months from now. But but I 00:11:13.280 --> 00:11:16.397 think what you've said is a very accurate description of where we are now 00:11:16.480 --> 00:11:21.437 as of you know our our you know as of Sonnet 4.5 which was the latest model 00:11:21.520 --> 00:11:24.877 released at >> which is probably one of the best coders that we've seen right 00:11:24.960 --> 00:11:29.277 >> it it you know every every you know we have these benchmarks that are like you 00:11:29.360 --> 00:11:31.597 know what fraction of you know a random 00:11:31.680 --> 00:11:36.557 sample of GitHub PRs can this thing write from a description and that's over 00:11:36.640 --> 00:11:40.797 the last year and a half that's gone from like 5% which was basically the 00:11:40.880 --> 00:11:45.437 thing can't code at all to something like you know 77%. 00:11:45.520 --> 00:11:49.517 Uh so it's it's it's really >> and it's also yielding also this whole 00:11:49.600 --> 00:11:54.157 other ecosystem of things on top of it as well like even we showed agent force 00:11:54.240 --> 00:11:58.397 fives yesterday that I think our customers will benefit from or other you 00:11:58.480 --> 00:12:01.597 know other tools lovable these things have been built on these things right 00:12:01.680 --> 00:12:05.277 has that been a surprise to you to see a whole ecosystem on top of the model 00:12:05.360 --> 00:12:08.557 >> I mean that's been that's been great to see there are there are many different 00:12:08.640 --> 00:12:12.317 models right if you look at the GitHub co-pilots of the world you know that's 00:12:12.400 --> 00:12:15.677 kind of complimentary to GitHub if you look at the cursors of the world. That's 00:12:15.760 --> 00:12:20.637 kind of an IDE model. If you look at the lovables and replets, that's more of a 00:12:20.720 --> 00:12:24.957 consumer model. It allows people who are not traditionally software developers to 00:12:25.040 --> 00:12:29.277 write code. Um, you know, there's like a, you know, versel is like a different 00:12:29.360 --> 00:12:30.957 a different way of looking at it. 00:12:31.040 --> 00:12:35.117 Augment code, wind surf, you know, there's a whole ecosystem. And then 00:12:35.200 --> 00:12:37.037 there's the command line tools like like 00:12:37.120 --> 00:12:39.917 claude code. Uh, so I think it's too 00:12:40.000 --> 00:12:44.077 early to say, you know, what's going to be the winner there or or even if there 00:12:44.160 --> 00:12:47.117 is one winner, it's just possible there's just different things that work 00:12:47.200 --> 00:12:50.077 for different people or or different >> there's a whole civilization. 00:12:50.160 --> 00:12:51.997 >> There's a whole eos capability, right? 00:12:52.080 --> 00:12:56.237 It's a whole civilization like it's an eco it's even it's broader like it's 00:12:56.320 --> 00:13:00.397 really coming across the whole span. It it's really impressive, 00:13:00.480 --> 00:13:05.197 >> you know? It's it's it's so fundamental that it's able to yield all this 00:13:05.280 --> 00:13:07.357 incredible innovation from others. 00:13:07.440 --> 00:13:10.157 >> Yeah. No, this I mean >> which one of those is the most like the 00:13:10.240 --> 00:13:11.917 one you did not expect? 00:13:12.000 --> 00:13:14.237 >> Which one is the one that I most did not 00:13:14.320 --> 00:13:19.117 expect? Um I think the model of kind of 00:13:19.200 --> 00:13:21.677 allowing people to write code end to end 00:13:21.760 --> 00:13:26.477 like allowing non-coders to write code it totally makes sense once I see it and 00:13:26.560 --> 00:13:30.397 it's doing really well as a business model. But but I I I just when I was 00:13:30.480 --> 00:13:33.197 conceptualizing all the >> which product do you love the best? 00:13:33.280 --> 00:13:37.597 >> Uh yeah this is this is one of those questions that you know you know it's C 00:13:37.680 --> 00:13:39.357 your customers are like your children. 00:13:39.440 --> 00:13:42.237 You you you you really shouldn't give an answer. So, I'm not going to give an 00:13:42.320 --> 00:13:45.320 answer. 00:13:56.400 --> 00:13:58.157 All right. So, are you getting a sense 00:13:58.240 --> 00:14:01.277 for Dario, by the way? Are you getting a 00:14:01.360 --> 00:14:05.997 sense for Dario? Okay. So that's I'm trying to like you know help them to 00:14:06.080 --> 00:14:10.237 understand you and how you're thinking about this and where we're going. So now 00:14:10.320 --> 00:14:12.077 here you are, you know, you're 00:14:12.160 --> 00:14:15.357 entrepreneur, you've left Open AI, this 00:14:15.440 --> 00:14:17.357 thing is taking off. It's you and 00:14:17.440 --> 00:14:21.037 Daniela, you know, um you're in the 00:14:21.120 --> 00:14:25.997 Slack building, you're get where is it going to be in 5 years. What 00:14:26.080 --> 00:14:30.317 does it look like? We've seen zero to five years now. You don't have to tell 00:14:30.400 --> 00:14:33.997 me as an investor and there's nobody hold gonna hold you accountable for it, 00:14:34.080 --> 00:14:36.877 but just if you just fantasized out and 00:14:36.960 --> 00:14:41.437 now it's not 2025, but it's 2030 and you 00:14:41.520 --> 00:14:45.837 know where's Enthropic? What does it look like? Forget about all the other 00:14:45.920 --> 00:14:49.757 companies and all the other competitors, whatever. Nobody cares. Where's 00:14:49.840 --> 00:14:51.597 Enthropic 5 years from now? What's your 00:14:51.680 --> 00:14:54.317 dream big what's your big dream? So, you 00:14:54.400 --> 00:14:58.637 know, I I think I think maybe maybe one of the best ways to think about this is, 00:14:58.720 --> 00:15:01.997 you know, I wrote this essay called Machines of Loving Grace where where 00:15:02.080 --> 00:15:04.877 it's really it was written from a societal perspective, but you know, of 00:15:04.960 --> 00:15:08.557 course, the company is trying to enact that and and then contribute to it and 00:15:08.640 --> 00:15:12.717 so you can back out from that where we where we hope the the company will be. 00:15:12.800 --> 00:15:15.997 So, you know, I I really thought of it as like the upside of AI, right? There's 00:15:16.080 --> 00:15:17.997 lots of concerns with AI. I I you know 00:15:18.080 --> 00:15:19.677 I'm one of the loudest in voicing them 00:15:19.760 --> 00:15:23.197 but but I am you know I am I am I am 00:15:23.280 --> 00:15:27.597 incredible believer in the upside of AI you know I think I think more so than 00:15:27.680 --> 00:15:29.597 people who code or describe themselves 00:15:29.680 --> 00:15:31.997 as more optimistic I have just a really 00:15:32.080 --> 00:15:35.277 expansive view I am perhaps most excited 00:15:35.360 --> 00:15:39.917 about what it can do for curing diseases um you know going back to my background 00:15:40.000 --> 00:15:42.157 in biology like you know we look at 00:15:42.240 --> 00:15:44.477 these tough diseases like cancer like 00:15:44.560 --> 00:15:47.517 Alzheimer's um you know like like you 00:15:47.600 --> 00:15:52.317 know various kinds of metabolic disease like I I I just think AI is the right 00:15:52.400 --> 00:15:55.517 thing to cure them and so I think of all the work we can do working with 00:15:55.600 --> 00:15:59.277 pharmaceutical companies working with scientific research institutions and 00:15:59.360 --> 00:16:03.037 working with the health care system to kind of enact it out and you know you 00:16:03.120 --> 00:16:06.157 again our partnership is about regulated industries and one of the regulated 00:16:06.240 --> 00:16:08.717 industries I think especially is is 00:16:08.800 --> 00:16:13.357 healthcare but you know that is one of the most positive applications but I 00:16:13.440 --> 00:16:17.037 think that we will see applications across the industry I'm very excited 00:16:17.120 --> 00:16:19.277 about financial services. I'm very 00:16:19.360 --> 00:16:21.757 excited about insurance. I'm very 00:16:21.840 --> 00:16:24.397 excited about taking basically every 00:16:24.480 --> 00:16:28.477 enterprise um and learning to serve 00:16:28.560 --> 00:16:32.077 every enterprise with smarter and smarter models. And the skill ceiling is 00:16:32.160 --> 00:16:35.997 very high. One way to say it is what is an enterprise? It's kind of a super 00:16:36.080 --> 00:16:38.077 intelligence, right? It it is an entity 00:16:38.160 --> 00:16:40.637 that acts in the world in in in in in in 00:16:40.720 --> 00:16:45.597 a way with a level of strategy and and knowledge and power in the world that 00:16:45.680 --> 00:16:49.597 goes far beyond what any individual human can do. So, you know, even as 00:16:49.680 --> 00:16:53.757 these models get smarter than than than any than any individual human, there's a 00:16:53.840 --> 00:16:57.757 ceiling substantially higher than that in terms of doing the things that a 00:16:57.840 --> 00:17:01.677 company tries to do in the world. Um, you know, most of which are good for the 00:17:01.760 --> 00:17:06.157 world, most of which create economic value. Um and and and so then my my 00:17:06.240 --> 00:17:09.917 vision of Anthropic is a company that serves the enterprise of the world that 00:17:10.000 --> 00:17:12.877 transforms every every enterprise puts 00:17:12.960 --> 00:17:15.437 AI at the center at the center of every 00:17:15.520 --> 00:17:17.917 enterprise. Um uh you know and and and 00:17:18.000 --> 00:17:20.157 and and you know I think I won't give 00:17:20.240 --> 00:17:22.637 specific numbers but like if you if you 00:17:22.720 --> 00:17:24.957 think about the actual market potential 00:17:25.040 --> 00:17:30.157 of that it's very very high right our revenue has been going up 10x per year. 00:17:30.240 --> 00:17:32.877 It's currently, I think it was reported this morning, which I'm happy to confirm 00:17:32.960 --> 00:17:36.157 is correct, is approaching approaching 7 00:17:36.240 --> 00:17:40.797 billion run rate. Um, that's gone up 10x a year for the for the last three years. 00:17:40.880 --> 00:17:45.037 It can't continue like that for much longer, of course, but uh you know, 00:17:45.120 --> 00:17:48.477 >> as an investor, I'm looking forward to the $70 billion year next year. 00:17:48.560 --> 00:17:51.037 >> You should not feel bad investing at 4 00:17:51.120 --> 00:17:52.957 bill. You should not feel bad. 00:17:53.040 --> 00:17:57.117 >> Um uh uh uh so, you know, but I I I 00:17:57.200 --> 00:17:59.197 think the potential is is is very large. 00:17:59.280 --> 00:18:02.957 But but I think even even more important than that or perhaps even necessary to 00:18:03.040 --> 00:18:07.277 that success, we want to be known for, you know, the company that is 00:18:07.360 --> 00:18:09.757 trustworthy, the company that behaves 00:18:09.840 --> 00:18:12.477 responsibly, the company that has a that 00:18:12.560 --> 00:18:15.517 has a reputation for, you know, doing 00:18:15.600 --> 00:18:19.997 what it says, saying what it does, and building products that are trustworthy 00:18:20.080 --> 00:18:23.757 that, you know, if you if you if you build on us, you won't get any 00:18:23.840 --> 00:18:26.317 surprises. You won't get anything crazy. 00:18:26.400 --> 00:18:29.277 You won't get the model, you know, saying it's Hitler. you won't you won't 00:18:29.360 --> 00:18:32.557 get, you know, all kinds of um you won't 00:18:32.640 --> 00:18:36.397 get all kinds of crazy stuff. Um and and I think that's very important to us 00:18:36.480 --> 00:18:39.517 being good stewards of this. >> Are there certain companies that you're 00:18:39.600 --> 00:18:42.077 trying to imply that that might be >> None whatsoever. 00:18:42.160 --> 00:18:44.397 >> There's an entrepreneur down in Palo Alto. 00:18:44.480 --> 00:18:46.717 >> None at all. I don't know what you're talking about. 00:18:46.800 --> 00:18:50.717 >> Okay, I see. That's a certain letter in the alphabet. 00:18:50.800 --> 00:18:53.437 >> I don't I don't I I have no idea what you're talking about. Um 00:18:53.520 --> 00:18:57.037 >> Okay, it's before the letter Y. I sorry 00:18:57.120 --> 00:18:59.197 I still don't know what you mean. 00:18:59.280 --> 00:19:03.197 >> Okay, one more thing. I'm going to take a couple I'm going to use moderator 00:19:03.280 --> 00:19:04.477 privilege here and take a couple extra 00:19:04.560 --> 00:19:08.237 minutes with you. So number one, we see 00:19:08.320 --> 00:19:11.197 right now a tremendous fever on building 00:19:11.280 --> 00:19:14.797 data centers and you know we've seen a 00:19:14.880 --> 00:19:17.997 tremendous fever around the chips, the 00:19:18.080 --> 00:19:22.077 financing, the size and scale of the data centers, the energy of the data 00:19:22.160 --> 00:19:23.837 centers. you haven't really addressed it 00:19:23.920 --> 00:19:26.157 in your comments per se besides the 00:19:26.240 --> 00:19:31.757 scaling laws but let's talk about it just in part and parcel to anthropic and 00:19:31.840 --> 00:19:35.117 when you think about obviously one of your core partners has been Amazon it's 00:19:35.200 --> 00:19:40.477 been very important for your growth very important company um even though you do 00:19:40.560 --> 00:19:43.677 have you know obviously relationship with others as well but that one has 00:19:43.760 --> 00:19:47.757 been especially important to you know just give us a vision of what's 00:19:47.840 --> 00:19:51.517 happening with these data centers right now what are we going to see in the next 00:19:51.600 --> 00:19:54.957 few years with these these data centers and how do you think about it as an 00:19:55.040 --> 00:19:59.117 entrepreneur who's going to need data centers to grow? Who will you go to? How 00:19:59.200 --> 00:20:00.877 will you do it? Will you build your own? 00:20:00.960 --> 00:20:04.157 Will you just continue to partner? What what is your plan? 00:20:04.240 --> 00:20:07.677 >> Yeah. So, you know, my my my view is is 00:20:07.760 --> 00:20:11.917 that there is going to be an enormous need for compute for for all the reasons 00:20:12.000 --> 00:20:15.597 that uh you know that that we've laid out, right? The the scaling laws suggest 00:20:15.680 --> 00:20:18.397 that we're going to need more and more compute to train smarter and smarter 00:20:18.480 --> 00:20:21.837 models. And then as the revenue goes up, we're gonna we're, you know, we're going 00:20:21.920 --> 00:20:25.197 to need to serve more. We're going to need to serve more of our models. So 00:20:25.280 --> 00:20:30.397 when that revenue is not seven billion, but a hundred billion or more. Um uh and 00:20:30.480 --> 00:20:33.917 you know, when when model training requires millions of chips, you know, 00:20:34.000 --> 00:20:37.917 we're we're going to need many many gigawatts of power to to train the 00:20:38.000 --> 00:20:39.997 models. So we're trying to secure that 00:20:40.080 --> 00:20:44.957 power basically wherever we can. Um you know, and I'm confident in the company's 00:20:45.040 --> 00:20:46.637 uh I'm confident in the company's 00:20:46.720 --> 00:20:49.677 ability to do so. I I I do get a bit 00:20:49.760 --> 00:20:52.397 worried when the press focuses too much 00:20:52.480 --> 00:20:55.357 on the building of the data centers and 00:20:55.440 --> 00:20:59.757 deals to build data centers. Some of these deals seem a little bit fishy. 00:20:59.840 --> 00:21:03.597 They seem like there may be double counting. And I I also worry again 00:21:03.680 --> 00:21:05.117 >> deals are you ask talking about? 00:21:05.200 --> 00:21:06.237 >> I don't know what you're talking about. 00:21:06.320 --> 00:21:11.517 I have no idea. Um uh uh uh >> who is double counting? These are all 00:21:11.600 --> 00:21:15.117 public companies. I I >> we're going to want to know who's double counting. 00:21:15.200 --> 00:21:17.677 >> I don't I don't I don't know. I don't know anything more than you. 00:21:17.760 --> 00:21:19.197 >> What about the ones who are doing triple counting? 00:21:19.280 --> 00:21:21.517 >> I I don't There might be some of that, too. 00:21:21.600 --> 00:21:26.077 >> Oh, wow. All right. >> I don't know. Um uh uh uh but but but 00:21:26.160 --> 00:21:27.917 look look look look at the end of the 00:21:28.000 --> 00:21:30.397 day all this data center stuff is 00:21:30.480 --> 00:21:33.197 spending money, right? Um I think the 00:21:33.280 --> 00:21:35.517 thing that's most important is making 00:21:35.600 --> 00:21:38.397 money. Um and and so this is why I focus 00:21:38.480 --> 00:21:41.357 on, you know, the revenue and the actual 00:21:41.440 --> 00:21:43.517 work that we're doing with businesses. 00:21:43.600 --> 00:21:46.557 ROIC is growing faster than any other 00:21:46.640 --> 00:21:50.557 company in the space. The revenue is growing faster and you know I think I 00:21:50.640 --> 00:21:54.557 think basically if if you know if the if the demand is there if customers want 00:21:54.640 --> 00:21:58.317 the money and they're willing to pay you then you are willing to pay your compute 00:21:58.400 --> 00:22:01.757 providers and you'll get the compute you need at the end of the day. 00:22:01.840 --> 00:22:04.157 >> Daria, we could not be more grateful for 00:22:04.240 --> 00:22:09.517 you to be here to be in San Francisco to be a native son and to Daniela as well 00:22:09.600 --> 00:22:10.957 and for everything that you're doing 00:22:11.040 --> 00:22:17.240 with anthropic congratulations. Please give Dario a round of applause.