1 00:00:07,520 --> 00:00:10,557 All right. Good afternoon, everybody. 2 00:00:10,640 --> 00:00:13,917 Are you all having a good time? 3 00:00:14,000 --> 00:00:17,837 >> Oh, fantastic. Amazing. Are you ready for Metallica? 4 00:00:17,920 --> 00:00:20,157 >> Are you ready for Benson Boone? 5 00:00:20,240 --> 00:00:23,597 >> Raise your hand if you're going to Metallica. 6 00:00:23,680 --> 00:00:26,957 Awesome. Who's looking forward to Benson Boon? 7 00:00:27,040 --> 00:00:28,477 >> All right. Well, so we got a It's kind 8 00:00:28,560 --> 00:00:30,557 of a mix. 9 00:00:30,640 --> 00:00:36,840 I don't know. Um, please welcome Daario. 10 00:00:39,520 --> 00:00:44,957 >> Hey, Mark. Thanks for having me. Great to have you, Dario. 11 00:00:45,040 --> 00:00:48,477 Welcome. >> Thank you. 12 00:00:48,560 --> 00:00:52,157 >> Uh, we're really excited to have you, Daario, and introduce you to the 13 00:00:52,240 --> 00:00:55,677 audience and everybody here. And, you know, we've had such a great 14 00:00:55,760 --> 00:00:57,677 partnership. We've had an opportunity to 15 00:00:57,760 --> 00:01:01,357 speak so many times and um you know work 16 00:01:01,440 --> 00:01:04,237 with some customers together and uh also 17 00:01:04,320 --> 00:01:05,997 build some technology together. It's 18 00:01:06,080 --> 00:01:08,717 been exciting to use your model inside 19 00:01:08,800 --> 00:01:11,597 Slack with our new Slackbot. But you 20 00:01:11,680 --> 00:01:15,757 know, as I've talked to so many of my friends about you, we've gotten to know 21 00:01:15,840 --> 00:01:20,477 each other, but a lot of people don't know you. So maybe you could tell a 22 00:01:20,560 --> 00:01:22,717 little folks about a little about 23 00:01:22,800 --> 00:01:25,037 yourself, kind of your beginnings, where 24 00:01:25,120 --> 00:01:28,477 you're from, you know, and also kind of 25 00:01:28,560 --> 00:01:33,277 uh I'll ask you a couple of the core questions along the way, but to take it 26 00:01:33,360 --> 00:01:37,197 from the start. Yeah, absolutely. So I grew up right here in uh in I grew up 27 00:01:37,280 --> 00:01:39,437 right here in San Francisco. I was uh 28 00:01:39,520 --> 00:01:42,637 originally interested in science, not in 29 00:01:42,720 --> 00:01:45,277 uh not in tech per se. So you know I did 30 00:01:45,360 --> 00:01:48,397 my did my uh uh graduate work in uh 31 00:01:48,480 --> 00:01:50,477 neuroscience as a posttock in 32 00:01:50,560 --> 00:01:55,597 computational biology at at Stanford and you know as I was working hard on 33 00:01:55,680 --> 00:01:59,517 solving the problems of biology you know trying to understand the diseases we we 34 00:01:59,600 --> 00:02:02,477 we face like uh like cancer and to 35 00:02:02,560 --> 00:02:06,877 understand biology one of the things I noticed is that the complexity of 36 00:02:06,960 --> 00:02:10,637 biology felt like it was beyond human scale you know and I was starting to 37 00:02:10,720 --> 00:02:14,717 despair a little bit that that we would be able to make progress at any speed 38 00:02:14,800 --> 00:02:18,477 just because of you know if you've ever seen one of these diagrams of human 39 00:02:18,560 --> 00:02:20,157 metabolism it's just it's just beyond 40 00:02:20,240 --> 00:02:23,277 human scale but almost exactly at the 41 00:02:23,360 --> 00:02:26,077 time that I was wondering about this out 42 00:02:26,160 --> 00:02:28,637 of Google out of University of Toronto 43 00:02:28,720 --> 00:02:33,517 came the first work on deep neural networks uh and I looked at it you know 44 00:02:33,600 --> 00:02:36,637 I I replicated some of the results myself I said my god this stuff really 45 00:02:36,720 --> 00:02:37,917 works this stuff's really going to go 46 00:02:38,000 --> 00:02:40,717 there so I went into AI ended up working 47 00:02:40,800 --> 00:02:43,997 at um uh Google I joined open AI I in 48 00:02:44,080 --> 00:02:49,597 the early days I led research there for for for several years um uh and and and 49 00:02:49,680 --> 00:02:53,837 you know was one of the people who kind of documented this trend of exponential 50 00:02:53,920 --> 00:02:58,077 scaling of the technology um you know myself and the folks who eventually 51 00:02:58,160 --> 00:02:59,757 became the anthropic co-founders were 52 00:02:59,840 --> 00:03:03,197 among the first to uh study the scaling 53 00:03:03,280 --> 00:03:05,277 of the technology and so you know around 54 00:03:05,360 --> 00:03:08,957 2019 or 2020 had really two realizations 55 00:03:09,040 --> 00:03:12,637 one was wow AI is is really going to you 56 00:03:12,720 --> 00:03:16,237 know it's going to It's going to scale to the moon and it's really going to be 57 00:03:16,320 --> 00:03:18,957 you know this whole thing is going to reshape the economy. It'll take a few 58 00:03:19,040 --> 00:03:20,557 years but it's going to reshape the 59 00:03:20,640 --> 00:03:22,317 economy and you know they are almost 60 00:03:22,400 --> 00:03:26,317 there now. Um uh and and the second was 61 00:03:26,400 --> 00:03:30,397 look there's you know these systems are different from ordinary software. You 62 00:03:30,480 --> 00:03:33,677 don't write every line of code right building ordinary software it's like 63 00:03:33,760 --> 00:03:36,797 building a skyscraper or something. You make the blueprint you make everything 64 00:03:36,880 --> 00:03:41,837 to design. This is ironically a bit more like biological. It's organic or 65 00:03:41,920 --> 00:03:43,517 something. You're growing these models. 66 00:03:43,600 --> 00:03:48,077 You set a high level recipe. It's like growing a plant or baking a cake or 67 00:03:48,160 --> 00:03:51,277 something like that. And that unpredictability combined with the 68 00:03:51,360 --> 00:03:55,517 incredible economic impact means that safety and responsibility are going to 69 00:03:55,600 --> 00:03:59,597 be very important to make sure that the models behave predictably as well as as 70 00:03:59,680 --> 00:04:03,117 we deploy them and they have enormous impact on the world. We you know we we 71 00:04:03,200 --> 00:04:06,477 need to make sure that we deploy them responsibly. And so those were kind of 72 00:04:06,560 --> 00:04:09,117 the two pillars on which myself and the 73 00:04:09,200 --> 00:04:11,837 seven co-founders um who left OpenAI to 74 00:04:11,920 --> 00:04:15,357 to found Enthropic um you know generally 75 00:04:15,440 --> 00:04:17,357 generally based our generally based our 76 00:04:17,440 --> 00:04:19,997 view uh and you know the the the rest is 77 00:04:20,080 --> 00:04:22,717 history. Enthropic has been um you know 78 00:04:22,800 --> 00:04:25,597 around for 4 and 1/2 years. Um we're 79 00:04:25,680 --> 00:04:30,397 growing very quickly. um you know I think we're we're at this point um by 80 00:04:30,480 --> 00:04:35,117 revenue have at least a plurality perhaps the majority of uh um you know 81 00:04:35,200 --> 00:04:39,197 enterprise enterprise customers and enterprise revenue while most of the 82 00:04:39,280 --> 00:04:42,957 other players are are focused on the consumer side and we find that you know 83 00:04:43,040 --> 00:04:45,997 as as you know I think you would know from running an enterprise focused 84 00:04:46,080 --> 00:04:48,797 company these these values of trust and 85 00:04:48,880 --> 00:04:53,757 responsibility you know are really are synergistic with serving the enterprises 86 00:04:53,840 --> 00:04:57,357 right these values help us to serve the enterprise and Having a business model 87 00:04:57,440 --> 00:05:01,277 that serves the enterprise helps to reinforce our values. 88 00:05:01,360 --> 00:05:03,917 >> One of the one of the things um and you 89 00:05:04,000 --> 00:05:05,517 went pretty fast into kind of starting 90 00:05:05,600 --> 00:05:08,557 the company and kind of shot out of uh 91 00:05:08,640 --> 00:05:11,997 uh uh shot through uh Google and API to 92 00:05:12,080 --> 00:05:15,837 Anthropic, but you didn't mention your sister and that was like one of the 93 00:05:15,920 --> 00:05:19,277 things that first really shocked me when we started to get to know you was you 94 00:05:19,360 --> 00:05:23,437 and your sister together are working on this company and can you kind of tell us 95 00:05:23,520 --> 00:05:26,637 the origin story? >> Yeah. Yeah. So, you know, there are 96 00:05:26,720 --> 00:05:30,397 seven co-founders and me and Daniela are are two of them. As I mentioned, we 97 00:05:30,480 --> 00:05:34,397 both, you know, we both grew up in San Francisco. We went to to LOL High School 98 00:05:34,480 --> 00:05:37,997 and, you know, even even very early on, we really figured that we wanted to do 99 00:05:38,080 --> 00:05:41,197 >> My mom went to LOL and she's sitting right there. So, she's going to be very 100 00:05:41,280 --> 00:05:45,917 happy about that. >> Amazing. Um, uh, no, no, it was it was 101 00:05:46,000 --> 00:05:49,357 it was a great place. I actually didn't over Daniela was four years younger than 102 00:05:49,440 --> 00:05:52,557 me, so we were never there at the same year, but I was there for four and then, 103 00:05:52,640 --> 00:05:56,797 you know, the next four years she was she was there. So um uh yeah you know 104 00:05:56,880 --> 00:06:01,037 very early on you know we we we had these you know these ambitions to you 105 00:06:01,120 --> 00:06:04,557 know to to do something to do something good for the world to have some kind of 106 00:06:04,640 --> 00:06:06,797 grand project and and to work together. 107 00:06:06,880 --> 00:06:10,957 I think I think many many people you know kind of or many siblings I guess 108 00:06:11,040 --> 00:06:15,117 have these have these ambitions but you know some somehow things found their way 109 00:06:15,200 --> 00:06:17,677 to us actually actually being able to do 110 00:06:17,760 --> 00:06:21,197 it. So um uh you know she joined me at 111 00:06:21,280 --> 00:06:25,517 at she joined me at OpenAI in uh uh 2018 112 00:06:25,600 --> 00:06:29,677 um and then the two of us along with some folks we worked with many of whom 113 00:06:29,760 --> 00:06:32,237 had been longtime uh uh friends decided 114 00:06:32,320 --> 00:06:36,717 to start Anthropic. So me and Daniela essentially essentially run the company 115 00:06:36,800 --> 00:06:41,517 together. You know she's responsible for the dayto-day running of the company. 116 00:06:41,600 --> 00:06:45,597 you know, kind of operate, you know, how how the company is put put together, how 117 00:06:45,680 --> 00:06:50,157 the org runs, what its processes are, does, you know, kind of kind of a fair 118 00:06:50,240 --> 00:06:54,397 amount of the the the, you know, kind of the business on the on the the business 119 00:06:54,480 --> 00:06:59,117 on the structural side. Um, uh, you know, I I I I just, you know, couldn't 120 00:06:59,200 --> 00:07:02,877 couldn't run the company without her and without her without her doing all those 121 00:07:02,960 --> 00:07:05,277 things. Um you know I tend to think 122 00:07:05,360 --> 00:07:07,997 about the strategic side where the where 123 00:07:08,080 --> 00:07:10,957 the research is going what our kind of 124 00:07:11,040 --> 00:07:15,197 you know plans are as a company strategically how we think about safety 125 00:07:15,280 --> 00:07:19,197 how we think about the company's uh uh role in the world and and the vision and 126 00:07:19,280 --> 00:07:21,997 and I find that's a very good um that's 127 00:07:22,080 --> 00:07:23,437 a very good split that's a very good 128 00:07:23,520 --> 00:07:26,477 separation um because both of us can do 129 00:07:26,560 --> 00:07:30,717 both pieces but it's it's really hard to switch between a kind of operational 130 00:07:30,800 --> 00:07:34,717 mode and a vision mode and I think both people can be more effective if you if 131 00:07:34,800 --> 00:07:37,037 you split. >> Thanks for allowing me to come back to 132 00:07:37,120 --> 00:07:40,957 that because I think it's such a great part of the company and the culture and 133 00:07:41,040 --> 00:07:45,277 how you're working and last year she was here and now this year you're here. Um 134 00:07:45,360 --> 00:07:48,237 and we've known you all from the beginning. I mean I think Salesforce 135 00:07:48,320 --> 00:07:52,477 owns like a some little point or something of anthropics. So we're very 136 00:07:52,560 --> 00:07:54,717 happy to be an investor. >> Very early investor. Very 137 00:07:54,800 --> 00:07:56,637 >> very early investor. Believe in you. 138 00:07:56,720 --> 00:07:59,517 >> I think you I think you invested when the valuation was 4 billion. Is that 139 00:07:59,600 --> 00:08:01,517 correct? >> I I don't know but it was too high. 140 00:08:01,600 --> 00:08:04,157 whatever it was. >> I haven't. 141 00:08:04,240 --> 00:08:05,997 >> Before you say it's too high, you should 142 00:08:06,080 --> 00:08:07,997 calculate how many times it's gone up since then. 143 00:08:08,080 --> 00:08:10,877 >> Okay. We made money so far. 144 00:08:10,960 --> 00:08:15,837 >> Well, um, no, I I'm thrilled obviously that we're part of your journey and, uh, 145 00:08:15,920 --> 00:08:20,077 you know, we love working with you. We love working with Daniela, um, so much. 146 00:08:20,160 --> 00:08:24,317 And you know, I think kind of getting back to this story of you and your 147 00:08:24,400 --> 00:08:26,877 sister and your friends, you know, starting this company here in San 148 00:08:26,960 --> 00:08:30,477 Francisco and then you came to us, we were starting to consolidate some of our 149 00:08:30,560 --> 00:08:33,997 real estate here and then you, you know, we had a conversation about you taking 150 00:08:34,080 --> 00:08:37,277 over the Slack building. We had just bought Slack, >> which we have done. 151 00:08:37,360 --> 00:08:38,797 >> And now you're in the Slack building. 152 00:08:38,880 --> 00:08:42,717 So, you're right on the park with us looking at the tower from your window. 153 00:08:42,800 --> 00:08:46,797 >> I I believe my office is what I I was told it was formerly Steuart's office. 154 00:08:46,880 --> 00:08:49,597 >> Formerly Steuart's office. Very good. 155 00:08:49,680 --> 00:08:51,917 And I would just, you know, it just just 156 00:08:52,000 --> 00:08:53,997 occurred to me like here you are. You're 157 00:08:54,080 --> 00:08:58,957 the San Franciscan. Um, you're right on you're right on the park with us. You're 158 00:08:59,040 --> 00:09:03,437 right on the tower with us. We're an investor in the company. Um, it's, you 159 00:09:03,520 --> 00:09:06,717 know, obviously a critical part of Slack today, right? In terms of how you 160 00:09:06,800 --> 00:09:10,317 couldn't use Slack without it, but you're at the very bleeding edge of 161 00:09:10,400 --> 00:09:12,237 these models. It's really incredible. 162 00:09:12,320 --> 00:09:15,997 And you've really been visionary uh, right from the very beginning. And 163 00:09:16,080 --> 00:09:19,757 when I didn't know you when you were at OpenAI, I only met you after you started 164 00:09:19,840 --> 00:09:22,797 the company. So just now kind of take us 165 00:09:22,880 --> 00:09:27,037 through your expectation. What has been your biggest surprise, 166 00:09:27,120 --> 00:09:30,637 you know, since it's only been what, four and a half, 5 years since you kind 167 00:09:30,720 --> 00:09:33,357 of walked out the door, said goodbye to 168 00:09:33,440 --> 00:09:37,997 Sam, right? And then you just said, "Okay, Daniela, we're going to we're 169 00:09:38,080 --> 00:09:41,757 going to get this going." What what was your biggest shock right there? Yeah, I 170 00:09:41,840 --> 00:09:44,317 mean, you know, there were there were so many moments because I think that, you 171 00:09:44,400 --> 00:09:46,077 know, the last four and a half years 172 00:09:46,160 --> 00:09:49,117 were were just so surprising overall. 173 00:09:49,200 --> 00:09:51,597 Um, you know, I think I think at a high 174 00:09:51,680 --> 00:09:54,397 level there's a dichotomy where on one 175 00:09:54,480 --> 00:09:59,357 hand, everything we've seen over the last 4 and 1/2 years was in some sense 176 00:09:59,440 --> 00:10:02,717 predicted by those straight lines on graphs that I made with the scaling 177 00:10:02,800 --> 00:10:06,717 laws, right? these enormous data center buildouts, the way that our revenue is 178 00:10:06,800 --> 00:10:11,757 going up, you know, 10x per year, the the use across the economy. You could 179 00:10:11,840 --> 00:10:15,117 you you know, you could see it all in these scribbles on the page. But 180 00:10:15,200 --> 00:10:18,477 actually seeing it play out is is wild. 181 00:10:18,560 --> 00:10:22,797 And and a lot of the individual things that that I'm seeing happening, I just I 182 00:10:22,880 --> 00:10:24,637 just never would have I never would have 183 00:10:24,720 --> 00:10:27,437 expected. I guess if I were to pick one 184 00:10:27,520 --> 00:10:30,797 most surprising thing, it is, you know, 185 00:10:30,880 --> 00:10:35,677 we left one company and started another company. And uh something you see over 186 00:10:35,760 --> 00:10:37,437 time is that is that as you build a 187 00:10:37,520 --> 00:10:40,477 company, you really do see through the 188 00:10:40,560 --> 00:10:43,437 people you hire, through the choices 189 00:10:43,520 --> 00:10:47,757 that the company makes, you kind of see the paths diverge and you see 190 00:10:47,840 --> 00:10:51,997 differentiation, right? OpenAI has gone very much in the consumer direction. 191 00:10:52,080 --> 00:10:56,157 We've gone very much in the enterprise direction. we've taken, you know, just 192 00:10:56,240 --> 00:11:00,077 just the way we approach things is is very different. And so you have this 193 00:11:00,160 --> 00:11:03,357 market structure where there are these two companies that are like it's sort of 194 00:11:03,440 --> 00:11:06,397 a natural experiment. They're they're just taking a very different approach. 195 00:11:06,480 --> 00:11:10,717 Both exist in the ecosystem, but but the approach they take is very different. 196 00:11:10,800 --> 00:11:14,077 >> And then you have to reflect back at your time at Google as well. They're 197 00:11:14,160 --> 00:11:18,877 also very much with Gemini and, you know, very much a top model as well. How 198 00:11:18,960 --> 00:11:23,037 do you look at them? Do you look at them as a blend then? Yeah, I mean, you know, 199 00:11:23,120 --> 00:11:26,877 Google's a partner to us. They're an investor in us. Obviously, their models 200 00:11:26,960 --> 00:11:31,277 compete with us, compete with us at the same time. But I think this co-optition 201 00:11:31,360 --> 00:11:34,397 dynamic is becoming more and more common, right? As we have all these 202 00:11:34,480 --> 00:11:36,397 layers, we have like the chip layer, the 203 00:11:36,480 --> 00:11:39,757 cloud layer, the the kind of model 204 00:11:39,840 --> 00:11:43,757 layer, the the like the like go to market and distribution layer, and 205 00:11:43,840 --> 00:11:47,597 various companies are involved in more than one of these layers. and and so I 206 00:11:47,680 --> 00:11:52,157 think it's not uncommon for companies to cooperate in one area and and compete in 207 00:11:52,240 --> 00:11:56,237 another area. It's fairly normal. I think it's also been helped by, you 208 00:11:56,320 --> 00:11:59,837 know, I I I know quite well some of the folks behind Google. I've been, you 209 00:11:59,920 --> 00:12:04,237 know, friends with with Demis for uh who, you know, who leads uh who leads 210 00:12:04,320 --> 00:12:07,997 research at Google for something like 15 years, right? I think he was maybe one 211 00:12:08,080 --> 00:12:11,997 of the only people who was earlier than me in in seeing the potential of the 212 00:12:12,080 --> 00:12:16,077 technology. He also has a neuroscience background like me. So I you know I 213 00:12:16,160 --> 00:12:19,517 think that helps us to get along even as there are some competitive aspects 214 00:12:19,600 --> 00:12:22,637 between the companies. >> What was your reaction when he won the 215 00:12:22,720 --> 00:12:27,037 Nobel Prize this year? >> Uh I I was I was very happy for uh 216 00:12:27,120 --> 00:12:31,037 Demis. Uh uh actually you know I was writing this essay about the potential 217 00:12:31,120 --> 00:12:35,517 of AI in biology called machines of loving grace. And two days before I was 218 00:12:35,600 --> 00:12:37,517 about to publish it of course Deis and 219 00:12:37,600 --> 00:12:40,077 his colleagues Alphafold won the won the 220 00:12:40,160 --> 00:12:44,557 Nobel Prize. So um uh you know I I I ended up dedicating the essay to the 221 00:12:44,640 --> 00:12:46,637 2024 you know w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w 222 00:12:46,720 --> 00:12:47,117 w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w w 223 00:12:47,200 --> 00:12:48,637 w w w w w w w w w w w w w w w w w w w w w w w w w w w w of the Nobel Prize in 224 00:12:48,720 --> 00:12:52,317 chemistry for for AlphaFold because it was an example of the future that I was 225 00:12:52,400 --> 00:12:56,077 talking about. >> Very exciting. So just getting back to 226 00:12:56,160 --> 00:13:00,317 you and Daniela and you're working on and growing the company and up it goes 227 00:13:00,400 --> 00:13:03,917 and you're doing it right here in San Francisco, you know, what are some of 228 00:13:04,000 --> 00:13:06,717 the characteristics and the models that you're looking at now that are most 229 00:13:06,800 --> 00:13:10,637 surprising? What what's really exciting about, you know, as you as you see the 230 00:13:10,720 --> 00:13:13,597 next next generations of these models unfold? 231 00:13:13,680 --> 00:13:18,557 >> Yeah. You know, I think broadly it's the jump to being able to do agentic 232 00:13:18,640 --> 00:13:20,477 abilities. Being able to do things. 233 00:13:20,560 --> 00:13:21,757 >> What was it? I didn't hear the word. 234 00:13:21,840 --> 00:13:24,797 >> Oh, sorry. Agentic abilities. Um, uh, 235 00:13:24,880 --> 00:13:29,837 uh, uh, turning the models into agents that Oh. Oh. Oh. Oh, I see. You've been 236 00:13:29,920 --> 00:13:35,117 I >> don't worry about it, Dario. Don't worry. I I wasn't. I just wanted I'm a 237 00:13:35,200 --> 00:13:37,837 little weak in my right ear. genuinely thought you didn't hear. 238 00:13:37,920 --> 00:13:43,640 >> I see. I understand. >> Um, 239 00:13:43,760 --> 00:13:46,477 >> uh, yeah, it was very funny, wasn't it? 240 00:13:46,560 --> 00:13:50,237 >> Definitely a buzz word and you've, you know, perhaps contributed more more to 241 00:13:50,320 --> 00:13:53,597 mo more than most to to making it to making it a buzz word. But I, you know, 242 00:13:53,680 --> 00:13:55,117 I think I think it's I think it's a real 243 00:13:55,200 --> 00:13:58,717 thing, which is being able to do tasks 244 00:13:58,800 --> 00:14:00,797 end to end in a way that was not 245 00:14:00,880 --> 00:14:03,277 possible before. And I think we see it 246 00:14:03,360 --> 00:14:07,277 more uh you know in code than uh you 247 00:14:07,360 --> 00:14:09,677 know we see it in in any other area. But 248 00:14:09,760 --> 00:14:11,917 I think code is just a leading indicator 249 00:14:12,000 --> 00:14:14,717 honestly in other tasks in tasks like 250 00:14:14,800 --> 00:14:17,277 financial services in tasks like you 251 00:14:17,360 --> 00:14:22,317 know health and medical in tasks like uh you know insurance or manufacturing many 252 00:14:22,400 --> 00:14:26,317 of the regulated industries that were working together on our on our on our 253 00:14:26,400 --> 00:14:31,117 partnership. Agents are agents are just as you know valuable. you just as much 254 00:14:31,200 --> 00:14:34,797 want to do things end to end, right? You think of the, you know, the sales and go 255 00:14:34,880 --> 00:14:38,957 to market and, you know, customer management workflows. You want to string 256 00:14:39,040 --> 00:14:41,837 a bunch of things together and you want the models to pull in a bunch of things 257 00:14:41,920 --> 00:14:47,277 in the context and learn this, do that, access this information. Um, so it's 258 00:14:47,360 --> 00:14:50,877 it's just as valuable. It's just the coders are early adopters. So, you know, 259 00:14:50,960 --> 00:14:54,077 they've they've kind of given us a glimpse of the future because they they 260 00:14:54,160 --> 00:14:58,397 adopt things so quickly. But I think we're going to see a similar spike in in 261 00:14:58,480 --> 00:15:02,157 other areas. So I want to ask you I have a few a few 262 00:15:02,240 --> 00:15:06,077 questions in that area. So the f you know when we first started getting in 263 00:15:06,160 --> 00:15:09,837 probably I think one of our very first conversations we were starting to talk 264 00:15:09,920 --> 00:15:12,077 about this theory and also AGI 265 00:15:12,160 --> 00:15:14,317 especially and what the potential in the 266 00:15:14,400 --> 00:15:17,117 future and AGI at that time and this is 267 00:15:17,200 --> 00:15:21,597 kind of now we go back several years was kind of being defined as yeah the model 268 00:15:21,680 --> 00:15:25,277 is almost kind of recursive it's building itself it's growing it's 269 00:15:25,360 --> 00:15:28,237 evolving it knows it is coding itself it 270 00:15:28,320 --> 00:15:32,317 can code and then it's it's enhancing and it's off and going off and you know 271 00:15:32,400 --> 00:15:36,637 we had Brett Adcock on the stage earlier today and he was talking about how kind 272 00:15:36,720 --> 00:15:40,717 of like had a scene out of you know we are Bob we are Legion that the robot had 273 00:15:40,800 --> 00:15:42,397 taken off and was now you know 274 00:15:42,480 --> 00:15:44,957 self-replicating into the future and you 275 00:15:45,040 --> 00:15:48,317 know an infinite number of uh you know 276 00:15:48,400 --> 00:15:52,477 uh robots it was it was incredibly 277 00:15:52,560 --> 00:15:57,917 compelling would go off and that was that you know to this to that point a 278 00:15:58,000 --> 00:16:01,517 few years ago when we were talking it was like well now the model is kind of 279 00:16:01,600 --> 00:16:03,357 building itself It's growing itself. 280 00:16:03,440 --> 00:16:06,877 It's develop. It didn't turn out exactly like that though, right? Or it's not 281 00:16:06,960 --> 00:16:09,357 we're not at that point yet, are we? 282 00:16:09,440 --> 00:16:10,957 >> Yeah. So, you know, I would separate two 283 00:16:11,040 --> 00:16:14,557 things. There there's I think inevitably as we, you know, talk about AI in the 284 00:16:14,640 --> 00:16:19,517 future, this kind of sci-fi sheen on on on, you know, the on the whole thing. 285 00:16:19,600 --> 00:16:22,557 And that's, you know, in a way it's helpful as a metaphor, but also it can 286 00:16:22,640 --> 00:16:26,557 be distracting because, you know, it just it just kind of conjures everyone's 287 00:16:26,640 --> 00:16:29,117 mind into this into this kind of uh, you 288 00:16:29,200 --> 00:16:34,477 know, into this kind of uh, you know, future dreamy mode. Um, uh, but, you 289 00:16:34,560 --> 00:16:37,757 know, I think there are some things that are real. One thing that's real is that, 290 00:16:37,840 --> 00:16:40,957 you know, every 3 months we make a model that's smarter than the month than the 291 00:16:41,040 --> 00:16:44,477 model we made before. And indeed, the models are slowly getting to where 292 00:16:44,560 --> 00:16:48,077 they're, you know, better than most humans at most things. maybe at some 293 00:16:48,160 --> 00:16:50,557 point better all humans than all things. 294 00:16:50,640 --> 00:16:52,797 The other thing I think is real is you 295 00:16:52,880 --> 00:16:57,597 know the model is not building itself in the sense that like I have a specific 296 00:16:57,680 --> 00:17:01,597 copy of the model that's like making a specific copy of the the you know the 297 00:17:01,680 --> 00:17:05,757 the next model you know we don't have our self-replicating you know nanobots 298 00:17:05,840 --> 00:17:08,237 or robots or or whatever but but what 299 00:17:08,320 --> 00:17:10,637 absolutely is happening is that within 300 00:17:10,720 --> 00:17:13,197 anthropic and elsewhere we are using 301 00:17:13,280 --> 00:17:16,397 claude to help build products on top of 302 00:17:16,480 --> 00:17:19,357 claude and to help train the next claude 303 00:17:19,440 --> 00:17:24,157 to help serve serve quad faster. Um, you know, 6 months ago I I made this 304 00:17:24,240 --> 00:17:28,237 prediction that, you know, in in 6 months 90% of code would be written by 305 00:17:28,320 --> 00:17:32,317 by by AI models. Some people think that prediction is wrong, but within 306 00:17:32,400 --> 00:17:35,997 Anthropic and within a number of companies that we work with, that is 307 00:17:36,080 --> 00:17:40,557 absolutely true. Um now 90 you know >> so you're saying that 90% of all code 308 00:17:40,640 --> 00:17:43,677 adenthropics being written by the by the model today 309 00:17:43,760 --> 00:17:46,877 >> on on many teams you know not uniformly every 310 00:17:46,960 --> 00:17:50,637 >> is that something that you've kind of provided something kind of 311 00:17:50,720 --> 00:17:54,477 >> something auditable around that or like shown how that's happened. Yeah, the 312 00:17:54,560 --> 00:17:59,837 teams play an editing and supervisory role, but you know, Claude can both you 313 00:17:59,920 --> 00:18:03,597 can say write this feature and Claude will do it and the human kind of looks 314 00:18:03,680 --> 00:18:05,837 over the code or asks for help from 315 00:18:05,920 --> 00:18:10,957 another another model, another copy of Claude. Um uh the other thing is 316 00:18:11,040 --> 00:18:14,317 sometimes we have these long running bugs. you know, when we were training 317 00:18:14,400 --> 00:18:17,997 our our the most recently released Claude, it was like, you know, the 318 00:18:18,080 --> 00:18:20,317 cluster was broken. There was this bug. 319 00:18:20,400 --> 00:18:24,397 Engineers were spending a few days on it. And, you know, we just we just went 320 00:18:24,480 --> 00:18:26,877 to Claude and we said, I don't know, just kind of play around with the 321 00:18:26,960 --> 00:18:30,157 cluster and see if you can figure out what's wrong. And, you know, Claude 322 00:18:30,240 --> 00:18:33,997 found this just super obscure bug that the engineers had missed. So, this this 323 00:18:34,080 --> 00:18:35,917 was one area where it was like Claude is 324 00:18:36,000 --> 00:18:37,757 like having a very talented teammate. So 325 00:18:37,840 --> 00:18:41,197 if claude is now running writing 90% of 326 00:18:41,280 --> 00:18:43,677 all of the code in enthropic do you need 327 00:18:43,760 --> 00:18:46,077 less engineers if 90% of all the code 328 00:18:46,160 --> 00:18:48,797 when will it be 95% or 99%. 329 00:18:48,880 --> 00:18:53,357 >> Yeah. So you know I you know I think people can misinterpret it as you know 330 00:18:53,440 --> 00:18:57,437 you know like claude writing 90% of the code means we're firing 90% of the 331 00:18:57,520 --> 00:19:00,717 software engineers. But you know if you're familiar with the principle of 332 00:19:00,800 --> 00:19:02,397 >> but you have made some very aggressive 333 00:19:02,480 --> 00:19:09,357 comments around firing executives >> that that so 334 00:19:09,440 --> 00:19:11,117 um >> I just want to put two and two together 335 00:19:11,200 --> 00:19:12,797 to get four. Okay. 336 00:19:12,880 --> 00:19:18,157 >> Um uh so uh um you know if >> one thing I've learned in knowing you 337 00:19:18,240 --> 00:19:21,197 for so long is I have to come back and kind of like get that qualifying 338 00:19:21,280 --> 00:19:25,597 question again. Yes. Because you're go you know you you're you're going fast. 339 00:19:25,680 --> 00:19:30,157 So trying to I'm trying to keep up with you and it's amazing but I want to just 340 00:19:30,240 --> 00:19:31,917 make sure I understand exactly what you're saying. 341 00:19:32,000 --> 00:19:33,517 >> Yes, absolutely. Let me let me let me 342 00:19:33,600 --> 00:19:36,877 take them one by one. So um uh you know 343 00:19:36,960 --> 00:19:40,237 under comparative advantage if Claude is 344 00:19:40,320 --> 00:19:43,597 writing 90% of the code what that means 345 00:19:43,680 --> 00:19:48,397 usually is you need just as many software engineers you might need more 346 00:19:48,480 --> 00:19:52,717 because they can they can then be more leveraged. they can focus on the 10% 347 00:19:52,800 --> 00:19:54,797 that's editing the code or writing the 348 00:19:54,880 --> 00:19:57,677 10% that's the hardest or supervising a 349 00:19:57,760 --> 00:20:01,997 group of a of AI models. And so what happens is, you know, you can just end 350 00:20:02,080 --> 00:20:04,397 up being 10 times more productive 351 00:20:04,480 --> 00:20:07,677 because the remaining 10% is is now a 352 00:20:07,760 --> 00:20:11,517 complement to the to the AI system that's that's getting much better at at 353 00:20:11,600 --> 00:20:13,997 getting much better at what it's producing. So So 354 00:20:14,080 --> 00:20:17,037 >> it's a rebalancing of the work is what you're really saying. 355 00:20:17,120 --> 00:20:18,317 >> It's it's not a replacement. 356 00:20:18,400 --> 00:20:21,117 >> It's not a replacement. It's a rebalancing. Oh, okay. 357 00:20:21,200 --> 00:20:24,717 >> As that 90% goes to 95 99 totally 358 00:20:24,800 --> 00:20:28,557 automated end to end, there may be some point at which all bets are off. 359 00:20:28,640 --> 00:20:31,677 >> What's your time frame for that? You It's a pretty big jump there. 360 00:20:31,760 --> 00:20:34,157 >> It's very hard to say. Um, but you know, 361 00:20:34,240 --> 00:20:37,837 we're looking at it. We're looking at it closely. We're always looking for ways 362 00:20:37,920 --> 00:20:41,917 to encourage complimentarity. >> You see that in the short term or the long term? 363 00:20:42,000 --> 00:20:44,557 >> Um, I you know, >> is that a 5year goal or a one-year goal? 364 00:20:44,640 --> 00:20:46,477 >> I'm optimistic about 10 year goal. 365 00:20:46,560 --> 00:20:48,717 >> I'm optimistic about complimentarity in 366 00:20:48,800 --> 00:20:51,517 the short run. I you know I I I I I do 367 00:20:51,600 --> 00:20:54,637 worry as we get out to two years 5 years 368 00:20:54,720 --> 00:20:57,197 that across the economy not that there 369 00:20:57,280 --> 00:20:59,597 won't be anything for humans to do but 370 00:20:59,680 --> 00:21:01,917 that there will be enough disruption 371 00:21:02,000 --> 00:21:04,957 that that we'll have a you know a kind 372 00:21:05,040 --> 00:21:07,357 of labor disruption that is you know not 373 00:21:07,440 --> 00:21:09,837 fundamentally different from what we've 374 00:21:09,920 --> 00:21:12,157 seen with previous technologies but that 375 00:21:12,240 --> 00:21:17,277 operates faster that is broader across the economy because it relates to all 376 00:21:17,360 --> 00:21:19,357 kinds. I think that's in the next two to five years. 377 00:21:19,440 --> 00:21:20,797 >> Two to five years I would say. 378 00:21:20,880 --> 00:21:22,637 >> Okay. And at what point just coming back 379 00:21:22,720 --> 00:21:27,997 to my previous question, at what point is Enthropic the model writing 100% of 380 00:21:28,080 --> 00:21:31,917 the code? >> Yeah. Again, it's it's a tricky question 381 00:21:32,000 --> 00:21:35,757 because maybe even if the model's writing 100% of the code, the humans are 382 00:21:35,840 --> 00:21:41,320 still supervising the model or prompting it or clarif 383 00:21:44,000 --> 00:21:48,317 it's it's it's it's a surprisingly clever thing. I don't you know I I'm I I 384 00:21:48,400 --> 00:21:52,877 I am concerned particularly about people's ability to adapt as the jobs 385 00:21:52,960 --> 00:21:57,117 are changing fast jobs are changing so fast but comparative comparative 386 00:21:57,200 --> 00:21:59,997 advantage is powerful that's going to be one of our tools in this well you've 387 00:22:00,080 --> 00:22:03,117 mentioned healthcare several times but we've seen these kind of very 388 00:22:03,200 --> 00:22:07,197 conflicting studies recently in healthcare around accuracy especially 389 00:22:07,280 --> 00:22:08,797 because these are word models there 390 00:22:08,880 --> 00:22:10,797 seems to be some level of accuracy 391 00:22:10,880 --> 00:22:16,237 that's not 100% and we're you know you know and then when applied to some of 392 00:22:16,320 --> 00:22:19,757 these kind of other things like radiology is one where we saw these very 393 00:22:19,840 --> 00:22:22,157 interesting, you know, studies where we 394 00:22:22,240 --> 00:22:26,957 just didn't see the accuracy of a radiologist reflected in what the model 395 00:22:27,040 --> 00:22:31,037 was doing. Is that how you see it going forward or how did you reflect back on 396 00:22:31,120 --> 00:22:34,877 that? >> Yeah, I I would say where we are now is 397 00:22:34,960 --> 00:22:38,397 I think of the models and I think a lot of people who use the models in practice, 398 00:22:38,480 --> 00:22:40,957 >> you know where I'm going, right, where I'm trying to I'm trying to like 399 00:22:41,040 --> 00:22:44,237 understand where are we really in the evolution of the world. 400 00:22:44,320 --> 00:22:48,157 >> Where's the arc? So, so I think a lot of I I think and a lot of the people who 401 00:22:48,240 --> 00:22:50,477 use the models in practice, I think 402 00:22:50,560 --> 00:22:53,037 think of the models as maybe a bit like 403 00:22:53,120 --> 00:22:55,677 a second opinion, right? Where you don't 404 00:22:55,760 --> 00:23:00,317 necessarily trust them alone. Just like if a doctor says something to me, I you 405 00:23:00,400 --> 00:23:03,997 know, I don't I don't necessarily take that as fact. They're an expert, but and 406 00:23:04,080 --> 00:23:08,157 they're often right, but they're not always right. Um uh so, you know, I'll 407 00:23:08,240 --> 00:23:12,397 tell >> especially in this one category, radiology. It seems strange like it just 408 00:23:12,480 --> 00:23:15,437 can't get its >> I was going to say can't get its head 409 00:23:15,520 --> 00:23:18,797 around it but it's not really the right metaphor but doesn't have 410 00:23:18,880 --> 00:23:22,637 >> it does there's something about either it doesn't have enough data or it's not 411 00:23:22,720 --> 00:23:25,437 getting it's too nuanced. What is it? 412 00:23:25,520 --> 00:23:28,877 >> Yeah. So let me let me actually tell you a story about about one case where it 413 00:23:28,960 --> 00:23:32,717 where it where it did work. Um so you know we're talking about you know 414 00:23:32,800 --> 00:23:35,997 founding the company with with my sister. She recently had her second 415 00:23:36,080 --> 00:23:39,917 child while she was pregnant. um you know pregnancy can suppress your immune 416 00:23:40,000 --> 00:23:44,477 system was was viral. For some reason it didn't occur to to any of them that it 417 00:23:44,560 --> 00:23:48,957 could be a bacterial infection. She took all her data. She uploaded it to Claude 418 00:23:49,040 --> 00:23:52,877 and you know Claude you know systematically listed the possibilities 419 00:23:52,960 --> 00:23:55,277 in the way that the doctors had somehow missed. 420 00:23:55,360 --> 00:23:58,957 >> That is amazing right? How many of us have already used the models in some way 421 00:23:59,040 --> 00:24:01,117 to look at our healthcare? Raise your hand. 422 00:24:01,200 --> 00:24:03,277 >> Yeah I think we're all having exactly that. 423 00:24:03,360 --> 00:24:04,237 >> That's that's a lot of people. 424 00:24:04,320 --> 00:24:07,757 >> Yeah it's amazing right? where you can kind of say especially how many of you 425 00:24:07,840 --> 00:24:13,117 are the hyperchondriacs that are doing that raise your hand. Yeah, I got a few. 426 00:24:13,200 --> 00:24:16,637 Okay, I know I'm >> but but but here's the point which is 427 00:24:16,720 --> 00:24:19,917 Claude listed possibilities. If she had 428 00:24:20,000 --> 00:24:23,197 just blindly taken the advice of Claude, 429 00:24:23,280 --> 00:24:27,197 that would not have gone well. Um Claude was good for saying, "Hey, have you 430 00:24:27,280 --> 00:24:31,757 considered this? Have you considered that?" So, so it has this assistive 431 00:24:31,840 --> 00:24:34,637 augmentative ability, >> especially in regards to words, right? 432 00:24:34,720 --> 00:24:37,757 is very good on words. >> Words. Yeah. Yeah. I mean, you know, 433 00:24:37,840 --> 00:24:41,437 this specialized stuff of reading images like I think that's more about 434 00:24:41,520 --> 00:24:44,957 specialized models like if you don't have the right data, maybe it's not 435 00:24:45,040 --> 00:24:49,037 doing well. It's less reflective of the >> is that the crux of the issue like in 436 00:24:49,120 --> 00:24:50,957 radiology cuz like when we're saying 437 00:24:51,040 --> 00:24:53,437 it's going to improve healthare but we 438 00:24:53,520 --> 00:24:56,637 also saw this weird study where doctors 439 00:24:56,720 --> 00:25:01,597 who were using the models some in some cases were relying too much of the 440 00:25:01,680 --> 00:25:05,997 models and then ending up with bad outcomes. On the other side then we saw 441 00:25:06,080 --> 00:25:10,237 models that were like trying to read radiology but then were not able to 442 00:25:10,320 --> 00:25:14,637 achieve accuracy. was like this kind of strange bifurcation between the two 443 00:25:14,720 --> 00:25:17,837 things. >> Yeah. You know, I think there's a lot 444 00:25:17,920 --> 00:25:22,397 still left to work out in what the role of the human is and what the role of the 445 00:25:22,480 --> 00:25:26,877 model is. And of course, that's changing as the models improve rapidly. But, you 446 00:25:26,960 --> 00:25:29,917 know, I think it's something you often see in new technologies where there's 447 00:25:30,000 --> 00:25:34,317 this great technology and then these practical issues of exactly how do you 448 00:25:34,400 --> 00:25:38,717 use it, exactly how do you interact with it, they take surprisingly long to work 449 00:25:38,800 --> 00:25:40,877 out. And you know if you look back at 450 00:25:40,960 --> 00:25:43,437 you know the early days when people you 451 00:25:43,520 --> 00:25:48,077 know kind of you know deployed deployed new technologies I don't know it's it's 452 00:25:48,160 --> 00:25:50,477 just it's often >> you think this is going to be about 453 00:25:50,560 --> 00:25:52,317 humans and agents working together. 454 00:25:52,400 --> 00:25:54,397 >> Yeah. No no no exactly just just the 455 00:25:54,480 --> 00:25:57,517 collaboration model needs to be worked 456 00:25:57,600 --> 00:26:01,437 out a bit more than it has in these practical areas. Code is good to point 457 00:26:01,520 --> 00:26:03,917 to like code happens so fast right? 458 00:26:04,000 --> 00:26:07,197 developers are early adopters and so 459 00:26:07,280 --> 00:26:09,357 that cycle we've gone through it in like 460 00:26:09,440 --> 00:26:14,397 months right you can see people talking in blog posts about oh you know what's 461 00:26:14,480 --> 00:26:18,397 the best way to use cloud code do I want it to suggest and I make edits or I 462 00:26:18,480 --> 00:26:22,717 write code and it makes edits and these people riff off each other and you know 463 00:26:22,800 --> 00:26:25,997 over the course of three or four months they work out what the best way to use 464 00:26:26,080 --> 00:26:30,397 quad code is I think that same thing is happening in healthcare and radiology 465 00:26:30,480 --> 00:26:33,917 but because there are a lot of barriers in these industries because they're high 466 00:26:34,000 --> 00:26:36,557 stakes that that process is working 467 00:26:36,640 --> 00:26:41,037 itself out much slower. >> So that's really kind of getting down to 468 00:26:41,120 --> 00:26:44,717 it, right? That it's a collaborative thing and like here's the best way to 469 00:26:44,800 --> 00:26:46,877 use cloud code. Cloud code is writing 470 00:26:46,960 --> 00:26:49,437 the code, but here's really the best way 471 00:26:49,520 --> 00:26:52,317 as a developer now to work on cloud 472 00:26:52,400 --> 00:26:54,877 code. It's not writing 100% of the code. 473 00:26:54,960 --> 00:26:58,877 It's writing code now I'm like checking here's here's and now now I'm going to 474 00:26:58,960 --> 00:27:02,717 have to maintain the code. I'm gonna have to provide all the capabilities, 475 00:27:02,800 --> 00:27:06,317 the whole infrastructure around the code. Is this the right way to think about it? 476 00:27:06,400 --> 00:27:07,837 >> I I think that's the right way to think 477 00:27:07,920 --> 00:27:10,397 about it right now. Um uh you know, as 478 00:27:10,480 --> 00:27:13,597 as the models get stronger, then the 479 00:27:13,680 --> 00:27:18,557 collaboration may change to more distant supervision by the human. Um so it's 480 00:27:18,640 --> 00:27:22,717 it's very hard to say what will happen even 6 months from now. But but I think 481 00:27:22,800 --> 00:27:26,397 what you've said is a very accurate description of where we are now as of 482 00:27:26,480 --> 00:27:31,597 you know our our you know as of Sonnet 4.5 which was the latest model released at 483 00:27:31,680 --> 00:27:34,237 >> which is probably one of the best coders that we've seen right 484 00:27:34,320 --> 00:27:38,557 >> it it you know every every we you know we have these benchmarks that are like 485 00:27:38,640 --> 00:27:43,357 you know what fraction of you know a random sample of GitHub PRs can this 486 00:27:43,440 --> 00:27:47,037 thing write from a description and that's over the last year and a half 487 00:27:47,120 --> 00:27:49,757 that's gone from like 5% which was 488 00:27:49,840 --> 00:27:54,797 basically the thing can't code at all to something like you know 77%. 489 00:27:54,880 --> 00:27:58,957 Uh so it's it's it's really >> and it's also yielding also this whole 490 00:27:59,040 --> 00:28:03,517 other ecosystem of things on top of it as well like even we showed agent force 491 00:28:03,600 --> 00:28:07,757 vibes yesterday that I think our customers will benefit from or other you 492 00:28:07,840 --> 00:28:11,037 know other tools lovable these things have been built on these things right 493 00:28:11,120 --> 00:28:14,637 has that been a surprise to you to see a whole ecosystem on top of the model 494 00:28:14,720 --> 00:28:17,917 >> I mean that's been that's been great to see there are there are many different 495 00:28:18,000 --> 00:28:21,757 models right if you look at the GitHub co-pilots of the world you know that's 496 00:28:21,840 --> 00:28:25,117 kind of complimentary to GitHub if you look at the cursors of the world. That's 497 00:28:25,200 --> 00:28:30,077 kind of an IDE model. If you look at the lovables and replets, that's more of a 498 00:28:30,160 --> 00:28:34,397 consumer model. It allows people who are not traditionally software developers to 499 00:28:34,480 --> 00:28:38,717 write code. Um, you know, there's like a, you know, Versell is like a different 500 00:28:38,800 --> 00:28:40,397 a different way of looking at it. 501 00:28:40,480 --> 00:28:44,557 Augment code, wind surf, you know, there's a whole ecosystem. And then 502 00:28:44,640 --> 00:28:46,477 there's the command line tools like like 503 00:28:46,560 --> 00:28:49,357 claude code. Uh, so I think it's too 504 00:28:49,440 --> 00:28:53,517 early to say, you know, what's going to be the winner there or or even if there 505 00:28:53,600 --> 00:28:56,477 is one winner, it's just possible there's just different things that work 506 00:28:56,560 --> 00:28:59,677 for different people or or different >> there's a whole civilization. There's a 507 00:28:59,760 --> 00:29:03,837 whole EOS capability, right? It's a whole civilization like it's an EO it's 508 00:29:03,920 --> 00:29:05,997 even it's broader like it's really 509 00:29:06,080 --> 00:29:08,557 coming across the whole span. It it's 510 00:29:08,640 --> 00:29:13,597 really impressive, you know? It's it's it's so fundamental that it's able to 511 00:29:13,680 --> 00:29:17,917 yield all this incredible innovation from others. Yeah. No, this I mean 512 00:29:18,000 --> 00:29:21,357 >> which one of those is the most like the one you did not expect? 513 00:29:21,440 --> 00:29:23,677 >> Which one is the one that I most did not 514 00:29:23,760 --> 00:29:28,477 expect? Um I think the model of kind of 515 00:29:28,560 --> 00:29:31,037 allowing people to write code end to end 516 00:29:31,120 --> 00:29:33,757 like allowing non-coders to write code. 517 00:29:33,840 --> 00:29:37,357 It totally makes sense once I see it and it's doing really well as a business 518 00:29:37,440 --> 00:29:40,957 model. But but I I I just when I was conceptualizing all the 519 00:29:41,040 --> 00:29:44,477 >> which product do you love the best? Uh yeah, this is this is one of those 520 00:29:44,560 --> 00:29:48,957 questions that you know you know c your customers are like your children. You 521 00:29:49,040 --> 00:29:51,677 you you you really shouldn't give an answer. So I'm not going to give an 522 00:29:51,760 --> 00:29:54,760 answer. 523 00:30:05,840 --> 00:30:07,597 All right. So are you getting a sense 524 00:30:07,680 --> 00:30:10,477 for Daario, by the way? Are you getting 525 00:30:10,560 --> 00:30:15,437 a sense for Daario? Okay. So that's I'm trying to like you know help them to 526 00:30:15,520 --> 00:30:19,597 understand you and how you're thinking about this and where we're going. So now 527 00:30:19,680 --> 00:30:21,437 here you are you know you're 528 00:30:21,520 --> 00:30:24,797 entrepreneur you've left open AI this 529 00:30:24,880 --> 00:30:26,717 thing is taking off. It's you and 530 00:30:26,800 --> 00:30:30,797 Daniela you know um you're in the Slack 531 00:30:30,880 --> 00:30:35,437 building you're get where is it going to be in 5 years? What 532 00:30:35,520 --> 00:30:39,677 does it look like? We've seen zero to five years now. You don't have to tell 533 00:30:39,760 --> 00:30:43,037 me as an investor and there's nobody hold going to hold you accountable for 534 00:30:43,120 --> 00:30:45,997 it. But just if you just fantasized out 535 00:30:46,080 --> 00:30:50,637 and now it's not 2025 but it's 2030 and 536 00:30:50,720 --> 00:30:55,277 you know where's Enthropic? What does it look like? Forget about all the other 537 00:30:55,360 --> 00:30:59,197 companies and all the other competitors whatever. Nobody cares. Where's 538 00:30:59,280 --> 00:31:00,957 Enthropic 5 years from now? What's your 539 00:31:01,040 --> 00:31:03,757 dream big what's your big dream? So, you 540 00:31:03,840 --> 00:31:07,997 know, I I think I think maybe maybe one of the best ways to think about this is, 541 00:31:08,080 --> 00:31:11,437 you know, I wrote this essay called Machines of Loving Grace where where 542 00:31:11,520 --> 00:31:14,317 it's really it was written from a societal perspective, but you know, of 543 00:31:14,400 --> 00:31:18,077 course, the company is trying to enact that and and and contribute to it and so 544 00:31:18,160 --> 00:31:22,157 you can back out from that where we where we hope the the company will be. 545 00:31:22,240 --> 00:31:25,437 So, you know, I I really thought of it as like the upside of AI, right? There's 546 00:31:25,520 --> 00:31:27,357 lots of concerns with AI. I I you know 547 00:31:27,440 --> 00:31:29,117 I'm one of the loudest in voicing them 548 00:31:29,200 --> 00:31:32,637 but but I am you know I am I am I am 549 00:31:32,720 --> 00:31:37,037 incredible believer in the upside of AI you know I think I think more so than 550 00:31:37,120 --> 00:31:38,957 people who code or describe themselves 551 00:31:39,040 --> 00:31:41,437 as more optimistic I have just a really 552 00:31:41,520 --> 00:31:44,637 expansive view I am perhaps most excited 553 00:31:44,720 --> 00:31:49,357 about what it can do for curing diseases um you know going back to my background 554 00:31:49,440 --> 00:31:51,517 in biology like you know we look at 555 00:31:51,600 --> 00:31:53,917 these tough diseases like cancer like 556 00:31:54,000 --> 00:31:56,957 Alzheimer's um you know like like you 557 00:31:57,040 --> 00:32:01,757 know various kinds of metabolic disease like I I I just think AI is the right 558 00:32:01,840 --> 00:32:04,957 thing to cure them and so I think of all the work we can do working with 559 00:32:05,040 --> 00:32:08,717 pharmaceutical companies working with scientific research institutions and 560 00:32:08,800 --> 00:32:12,397 working with the health care system to kind of enact it out and you know you 561 00:32:12,480 --> 00:32:15,517 again our partnership is about regulated industries and one of the regulated 562 00:32:15,600 --> 00:32:18,157 industries I think especially is is 563 00:32:18,240 --> 00:32:22,717 healthcare but you know that is one of the most positive applications but I 564 00:32:22,800 --> 00:32:26,397 think that we will see applications across the industry I'm very excited 565 00:32:26,480 --> 00:32:28,637 about financial services. I'm very 566 00:32:28,720 --> 00:32:31,197 excited about insurance. I'm very 567 00:32:31,280 --> 00:32:33,837 excited about taking basically every 568 00:32:33,920 --> 00:32:37,837 enterprise um and learning to serve 569 00:32:37,920 --> 00:32:41,517 every enterprise with smarter and smarter models. And the skill ceiling is 570 00:32:41,600 --> 00:32:45,437 very high. One way to say it is what is an enterprise? It's kind of a super 571 00:32:45,520 --> 00:32:47,517 intelligence, right? It it is an entity 572 00:32:47,600 --> 00:32:49,997 that acts in the world in in in in in in 573 00:32:50,080 --> 00:32:55,037 a way with a level of strategy and and knowledge and power in the world that 574 00:32:55,120 --> 00:32:58,957 goes far beyond what any individual human can do. So, you know, even as 575 00:32:59,040 --> 00:33:03,117 these models get smarter than than than any than any individual human, there's a 576 00:33:03,200 --> 00:33:07,197 ceiling substantially higher than that in terms of doing the things that a 577 00:33:07,280 --> 00:33:11,117 company tries to do in the world. Um, you know, most of which are good for the 578 00:33:11,200 --> 00:33:15,517 world, most of which create economic value. Um and and and so then my my 579 00:33:15,600 --> 00:33:19,277 vision of Enthropic is a company that serves the enterprise of the world that 580 00:33:19,360 --> 00:33:22,237 transforms every every enterprise puts 581 00:33:22,320 --> 00:33:24,797 AI at the center at the center of every 582 00:33:24,880 --> 00:33:27,277 enterprise. Um uh you know and and and 583 00:33:27,360 --> 00:33:29,517 and and you know I think I won't give 584 00:33:29,600 --> 00:33:31,997 specific numbers but like if you if you 585 00:33:32,080 --> 00:33:34,317 think about the actual market potential 586 00:33:34,400 --> 00:33:39,517 of that is very very high right our revenue has been going up 10x per year. 587 00:33:39,600 --> 00:33:42,317 It's currently, I think it was reported this morning, which I'm happy to confirm 588 00:33:42,400 --> 00:33:45,517 is correct, is approaching approaching 7 589 00:33:45,600 --> 00:33:50,557 billion run rate. Um, that's gone up 10x a year for the for the last 3 years. It 590 00:33:50,640 --> 00:33:54,397 can't continue like that for much longer, of course, but uh you know, 591 00:33:54,480 --> 00:33:57,917 >> as an investor, I'm looking forward to the $70 billion year next year. You 592 00:33:58,000 --> 00:34:00,717 should not feel bad investing at 4 bill. 593 00:34:00,800 --> 00:34:02,397 You should not feel bad. All right. 594 00:34:02,480 --> 00:34:06,637 >> Um uh uh uh so you know but I I I think 595 00:34:06,720 --> 00:34:10,397 the potential is is is very large. But I but I think even even more important 596 00:34:10,480 --> 00:34:15,277 than that or perhaps even necessary to that success, we want to be known for, 597 00:34:15,360 --> 00:34:19,197 you know, the company that is trustworthy, the company that behaves 598 00:34:19,280 --> 00:34:21,917 responsibly, the company that has a that 599 00:34:22,000 --> 00:34:24,957 has a reputation for, you know, doing 600 00:34:25,040 --> 00:34:29,357 what it says, saying what it does, and building products that are trustworthy 601 00:34:29,440 --> 00:34:33,117 that, you know, if you if you if you build on us, you won't get any 602 00:34:33,200 --> 00:34:35,677 surprises. You won't get anything crazy. 603 00:34:35,760 --> 00:34:38,717 You won't get the model, you know, saying it's Hitler. you won't you won't 604 00:34:38,800 --> 00:34:41,917 get, you know, all kinds of um you won't 605 00:34:42,000 --> 00:34:45,837 get all kinds of crazy stuff. Um and and I think that's very important to us 606 00:34:45,920 --> 00:34:48,877 being good stewards of this. >> Are there certain companies that you're 607 00:34:48,960 --> 00:34:50,557 trying to imply that that might be? 608 00:34:50,640 --> 00:34:53,757 >> None whatsoever. >> There's an entrepreneur down in Palo Alto. 609 00:34:53,840 --> 00:34:56,157 >> None at all. I don't know what you're talking about. 610 00:34:56,240 --> 00:35:00,077 >> Okay, I see. That's a certain letter in the alphabet. 611 00:35:00,160 --> 00:35:02,797 >> I don't I don't I I have no idea what you're talking about. Um 612 00:35:02,880 --> 00:35:06,077 >> Okay, that's before the letter Y. I 613 00:35:06,160 --> 00:35:08,557 sorry I still don't know what you mean. 614 00:35:08,640 --> 00:35:12,637 >> Okay, one more thing. I'm going to take a couple I'm going to use moderator 615 00:35:12,720 --> 00:35:13,917 privilege here and take a couple extra 616 00:35:14,000 --> 00:35:17,597 minutes with you. So number one, we see 617 00:35:17,680 --> 00:35:20,557 right now a tremendous fever on building 618 00:35:20,640 --> 00:35:24,237 data centers and you know we've seen a 619 00:35:24,320 --> 00:35:27,357 tremendous fever around the chips, the 620 00:35:27,440 --> 00:35:31,517 financing, the size and scale of the data centers, the energy of the data 621 00:35:31,600 --> 00:35:33,277 centers. you haven't really addressed it 622 00:35:33,360 --> 00:35:35,597 in your comments per se besides the 623 00:35:35,680 --> 00:35:41,117 scaling laws but let's talk about it just in part and parcel to enthropic and 624 00:35:41,200 --> 00:35:44,477 when you think about obviously one of your core partners has been Amazon it's 625 00:35:44,560 --> 00:35:49,837 been very important for your growth very important company um even though you do 626 00:35:49,920 --> 00:35:53,037 have you know obviously relationship with others as well but that one has 627 00:35:53,120 --> 00:35:57,197 been especially important to you know just give us a vision of what's 628 00:35:57,280 --> 00:36:00,877 happening with these data centers right now what are we going to see in the next 629 00:36:00,960 --> 00:36:04,317 few years with these data centers and how do you think about it as an 630 00:36:04,400 --> 00:36:08,477 entrepreneur who's going to need data centers to grow? Who will you go to? How 631 00:36:08,560 --> 00:36:10,237 will you do it? Will you build your own? 632 00:36:10,320 --> 00:36:13,517 Will you just continue to partner? What what what is your plan? 633 00:36:13,600 --> 00:36:17,037 >> Yeah. So, you know, my my my view is is 634 00:36:17,120 --> 00:36:21,277 that there is going to be an enormous need for compute for for all the reasons 635 00:36:21,360 --> 00:36:25,037 that uh you know that that we've laid out, right? The the scaling laws suggest 636 00:36:25,120 --> 00:36:27,837 that we're going to need more and more compute to train smarter and smarter 637 00:36:27,920 --> 00:36:31,197 models. And then as the revenue goes up, we're gonna we're, you know, we're going 638 00:36:31,280 --> 00:36:34,637 to need to serve more. We're going to need to serve more of our models. So 639 00:36:34,720 --> 00:36:39,757 when that revenue is not seven billion, but a hundred billion or more um uh and 640 00:36:39,840 --> 00:36:43,277 you know, when when model training requires millions of chips, you know, 641 00:36:43,360 --> 00:36:47,357 we're we're going to need many many gigawatts of power to to train the 642 00:36:47,440 --> 00:36:49,357 models. So we're trying to secure that 643 00:36:49,440 --> 00:36:54,397 power basically wherever we can. Um you know, and I'm confident in the company's 644 00:36:54,480 --> 00:36:55,997 uh uh I'm confident in the company's 645 00:36:56,080 --> 00:36:59,117 ability to do so. I I I do get a bit 646 00:36:59,200 --> 00:37:01,837 worried when the press focuses too much 647 00:37:01,920 --> 00:37:04,797 on the building of the data centers and 648 00:37:04,880 --> 00:37:09,117 deals to build data centers. Some of these deals seem a little bit fishy. 649 00:37:09,200 --> 00:37:13,037 They seem like they're maybe double counting. And I I also worry again 650 00:37:13,120 --> 00:37:14,557 >> deals are you ask talking about? 651 00:37:14,640 --> 00:37:15,677 >> I don't know what you're talking about. 652 00:37:15,760 --> 00:37:20,877 I have no idea. Um uh uh uh >> who is double counting? These are all 653 00:37:20,960 --> 00:37:24,557 public companies. I I >> We're going to want to know who's double counting. 654 00:37:24,640 --> 00:37:27,037 >> I don't I don't I don't know. I don't know anything more than you. 655 00:37:27,120 --> 00:37:28,557 >> What about the ones who are doing triple counting? 656 00:37:28,640 --> 00:37:32,157 >> I I don't There might be some of that, too. Wow. All right. 657 00:37:32,240 --> 00:37:35,517 >> I don't know. Um uh uh uh but but but 658 00:37:35,600 --> 00:37:38,077 look look look at the end of the day, 659 00:37:38,160 --> 00:37:40,317 all this data center stuff is spending 660 00:37:40,400 --> 00:37:42,717 money, right? Um I think the thing 661 00:37:42,800 --> 00:37:45,597 that's most important is making money. 662 00:37:45,680 --> 00:37:48,317 Um and and so this is why I focus on, 663 00:37:48,400 --> 00:37:50,717 you know, the revenue and the actual 664 00:37:50,800 --> 00:37:52,717 work that we're doing with businesses. 665 00:37:52,800 --> 00:37:57,277 Andropic is growing faster than any other company in the space. The revenue 666 00:37:57,360 --> 00:38:01,837 is growing faster. And you know, I think I think basically if if you know, if the 667 00:38:01,920 --> 00:38:05,117 if the demand is there, if customers want the money and they're willing to 668 00:38:05,200 --> 00:38:08,717 pay you, then you are willing to pay your comput providers and you'll get the 669 00:38:08,800 --> 00:38:11,117 compute you need at the end of the day. 670 00:38:11,200 --> 00:38:13,517 >> Daria, we could not be more grateful for 671 00:38:13,600 --> 00:38:18,557 you to be here, to be in San Francisco, to be a native son, and to Daniela as 672 00:38:18,640 --> 00:38:22,317 well, and for everything that you're doing with Enthropic. Congratulations. 673 00:38:22,400 --> 00:38:28,040 Please give Dario a round of applause.