1 00:00:00,000 --> 00:00:04,557 So, Dario, we're at Davos. There's a lot going on, 2 00:00:04,640 --> 00:00:11,080 but I wanted to start with a a big picture question, which I'll characterize like 3 00:00:12,360 --> 00:00:17,400 this. It feels to me that the 4 00:00:22,200 --> 00:00:25,437 Time last year, everybody was very excited about AI, and everyone was 5 00:00:25,520 --> 00:00:30,400 talking about what AI can do, its potential, 6 00:00:30,880 --> 00:00:33,477 its capabilities. It feels to me as 7 00:00:33,560 --> 00:00:35,157 though the debate has shifted somewhat 8 00:00:35,240 --> 00:00:37,917 this year to be more less what can AI do 9 00:00:38,000 --> 00:00:41,237 to what is AI doing to the world. 10 00:00:41,320 --> 00:00:44,997 Um and I I know that you think a lot about these things. So, my question is, 11 00:00:45,080 --> 00:00:48,000 do you think business 12 00:00:49,760 --> 00:00:54,840 is policy makers, governments, whatever, are doing enough 13 00:00:56,400 --> 00:00:58,877 to prepare for the impact? >> No. 14 00:00:58,960 --> 00:01:00,197 Um 15 00:01:00,280 --> 00:01:03,557 I'll explain the longer version now. Um 16 00:01:03,640 --> 00:01:06,037 you know, I've been watching this field 17 00:01:06,120 --> 00:01:08,397 for 15 years, and I've been in this 18 00:01:08,480 --> 00:01:11,117 field for 10 years. And and one of the 19 00:01:11,200 --> 00:01:15,597 things I've most noticed is that there's been 20 00:01:15,680 --> 00:01:17,797 the the actual trajectory of the field 21 00:01:17,880 --> 00:01:20,397 has been surprisingly 22 00:01:20,480 --> 00:01:24,397 on on a a you know, the same trajectory. 23 00:01:24,480 --> 00:01:28,037 Whereas the kind of public opinion and the reaction of the public has 24 00:01:28,120 --> 00:01:32,157 oscillated wildly. I would say that in in in two different ways. One is the 25 00:01:32,240 --> 00:01:34,917 capabilities of the technology. Every 3 26 00:01:35,000 --> 00:01:40,157 to 6 months, we have this reversal of of polarity, where the media is incredibly 27 00:01:40,240 --> 00:01:43,757 excited about what the technology can do, it's going to change everything, and 28 00:01:43,840 --> 00:01:47,117 then it's you know, it's all a bubble, it's all it's all going to fall apart. 29 00:01:47,200 --> 00:01:52,077 And and what I see is this smooth exponential line where similar to 30 00:01:52,160 --> 00:01:56,037 Moore's law for compute, we basically have a Moore's law for intelligence, 31 00:01:56,120 --> 00:02:00,157 where the model is getting more and more cognitively capable every every few 32 00:02:00,240 --> 00:02:04,637 months. And that that march has just been constant. The up and down, the we 33 00:02:04,720 --> 00:02:06,477 invented a new thing, it's all going to 34 00:02:06,560 --> 00:02:08,837 crash, it's hitting a wall, it's it's 35 00:02:08,920 --> 00:02:12,117 going to go crazy. That that is a public 36 00:02:12,200 --> 00:02:16,637 perception phenomenon. That's on the capability of the technology. I think 37 00:02:16,720 --> 00:02:20,477 there's a similar thing on the polarity of whether the technology is good or 38 00:02:20,560 --> 00:02:22,517 bad. Um 39 00:02:22,600 --> 00:02:27,437 uh you know, in 2023 and 2024, there was 40 00:02:27,520 --> 00:02:31,997 a lot of concern about AI, right? There was, you know, concern AI you know, AI's 41 00:02:32,080 --> 00:02:33,877 going to AI's are going to take over. 42 00:02:33,960 --> 00:02:36,397 There was a lot of talk about AI risk, 43 00:02:36,480 --> 00:02:40,037 AI misuse. Then in 2025, the political 44 00:02:40,120 --> 00:02:42,997 wind shifted, as you say, to AI 45 00:02:43,080 --> 00:02:47,517 opportunity. And now they're sort of shifting back. And and I think 46 00:02:47,600 --> 00:02:52,357 throughout throughout all of this, the the approach that I have tried to take 47 00:02:52,440 --> 00:02:56,557 and the approach that Anthropic has tried to take is is is one of constancy, 48 00:02:56,640 --> 00:02:59,077 of saying that there is balance here. 49 00:02:59,160 --> 00:03:01,677 Uh and and uh balance of a very strange 50 00:03:01,760 --> 00:03:04,197 form, because I think the technology is 51 00:03:04,280 --> 00:03:07,357 very extreme in in what it's capable of 52 00:03:07,440 --> 00:03:11,357 doing. But I think its positive impacts and its negative impacts that, you know, 53 00:03:11,440 --> 00:03:15,437 they they both exist, right? I wrote this essay Machines of Loving Grace 54 00:03:15,520 --> 00:03:19,077 about a year and a half ago. It had a very radical view of the upside of AI, 55 00:03:19,160 --> 00:03:23,717 that you know, it would it would help us to, you know, cure cancer, eradicate 56 00:03:23,800 --> 00:03:28,597 tropical diseases, you know, kind of bring bring economic development to, you 57 00:03:28,680 --> 00:03:32,357 know, parts of the world that haven't seen it. And I my view hasn't changed. I 58 00:03:32,440 --> 00:03:34,557 believe all of those things. Um 59 00:03:34,640 --> 00:03:39,917 but the the other side of it, which, you know, I'm I'm I'm now uh uh 60 00:03:40,000 --> 00:03:43,477 you know, kind of writing writing more about and and what you know, may may 61 00:03:43,560 --> 00:03:46,677 release something about soon, is yeah, 62 00:03:46,760 --> 00:03:50,917 there you know, bad things will happen as well. If we just take just for an 63 00:03:51,000 --> 00:03:56,000 example as as one of the risks, we take the economic 64 00:03:56,920 --> 00:03:58,797 side of it. Um 65 00:03:58,880 --> 00:04:01,517 my my view is the signature of this 66 00:04:01,600 --> 00:04:07,920 technology is it's going to take us to a world where we have very 67 00:04:21,760 --> 00:04:25,397 high GDP growth and potentially also 68 00:04:25,480 --> 00:04:28,717 very high unemployment and inequality. 69 00:04:28,800 --> 00:04:32,077 Now, that's not a combination we've we've almost ever seen before, right? 70 00:04:32,160 --> 00:04:36,477 You think of it as high GDP growth, that's lots of stuff to do, lots of jobs 71 00:04:36,560 --> 00:04:39,837 for everyone. It's always been like that in the past. We've never had a 72 00:04:39,920 --> 00:04:42,357 technology that's this disruptive. So, 73 00:04:42,440 --> 00:04:45,557 the idea that we could have 5 or 10% GDP 74 00:04:45,640 --> 00:04:50,437 growth, but also, you know, 10 10% unemployment, it's it's not logically 75 00:04:50,520 --> 00:04:52,437 inconsistent at all. It's just never 76 00:04:52,520 --> 00:04:55,037 happened that that that that way before. 77 00:04:55,120 --> 00:04:57,877 And I'm I'm really quite uh you know, 78 00:04:57,960 --> 00:04:59,877 for those both reasons, excited and 79 00:04:59,960 --> 00:05:02,237 worried. If I take an example, something 80 00:05:02,320 --> 00:05:05,157 like AI coding, um 81 00:05:05,240 --> 00:05:07,237 you know, the latest model we released, 82 00:05:07,320 --> 00:05:11,717 Claude Opus 4.5, um I have some 83 00:05:11,800 --> 00:05:14,117 uh uh engineers, some engineering leads 84 00:05:14,200 --> 00:05:17,557 within Anthropic who have basically said 85 00:05:17,640 --> 00:05:22,397 to me, I don't write any code anymore. I I you know, I just let Opus do the work 86 00:05:22,480 --> 00:05:25,997 and I edit it. We just released a new thing called Claude co-work. We can go 87 00:05:26,080 --> 00:05:28,117 into go you know, we can go into that 88 00:05:28,200 --> 00:05:30,877 later, but this was a a version of our 89 00:05:30,960 --> 00:05:35,317 tool, Claude code, for non-coding. This was built in a week and a half, almost 90 00:05:35,400 --> 00:05:39,197 entirely with Claude Opus. There's still things for the software engineers to do, 91 00:05:39,280 --> 00:05:43,357 right? It's like, even if the software engineers are only doing 10% of it, they 92 00:05:43,440 --> 00:05:46,917 can you know, they still have a job to do, or they can take a level up. That's 93 00:05:47,000 --> 00:05:51,317 not going to last forever. The models are going to do more and more. And so 94 00:05:51,400 --> 00:05:54,437 there's an incredible This is a microcosm. You can see there's an 95 00:05:54,520 --> 00:05:56,477 incredible amount of productivity here. 96 00:05:56,560 --> 00:05:58,797 Software is going to become cheap, maybe 97 00:05:58,880 --> 00:06:02,157 essentially free. The premise that you 98 00:06:02,240 --> 00:06:08,760 need to amortize a piece of software you build across millions of users 99 00:06:20,960 --> 00:06:25,640 that may start to be false. Like for the 100 00:06:55,280 --> 00:06:59,597 meaning it might cost a few cents to just say, I don't know, let's let's 101 00:06:59,680 --> 00:07:03,197 let's make some app so people can talk to each other or everything, you know, I 102 00:07:03,280 --> 00:07:05,157 it just it it it just may be very 103 00:07:05,240 --> 00:07:07,717 flexible and recyclable. But but at the 104 00:07:07,800 --> 00:07:10,357 same time there are whole jobs, whole 105 00:07:10,440 --> 00:07:14,997 careers that we built built for decades that that may not be be present. And you 106 00:07:15,080 --> 00:07:19,037 know, I I think we can deal with it. I think we can we can adjust to it, but I 107 00:07:19,120 --> 00:07:23,797 don't I don't think there's an an awareness at all of what of what is 108 00:07:23,880 --> 00:07:25,957 coming here and the magnitude of it. 109 00:07:26,040 --> 00:07:30,197 >> So how that's it's so interesting when you say that. So how do you think in a 110 00:07:30,280 --> 00:07:34,037 world of of high GDP growth but also high 111 00:07:34,120 --> 00:07:37,517 unemployment, you know, what does that do to society? And you say there are, 112 00:07:37,600 --> 00:07:40,637 you know, people aren't thinking about it now. Can you give concrete examples 113 00:07:40,720 --> 00:07:45,117 of how society might organize itself to adapt to such a world. 114 00:07:45,200 --> 00:07:49,317 >> Yeah. So, you know, I think I think there's a few things. The The first 115 00:07:49,400 --> 00:07:51,717 thing that that we've done that we've 116 00:07:51,800 --> 00:07:56,597 focused on, and this is not a solution so much as it is a first step, is we 117 00:07:56,680 --> 00:08:00,117 have this thing called the Anthropic Economic Index. We've had it for about a 118 00:08:00,200 --> 00:08:03,717 year. We've updated it I think four or five times now. 119 00:08:03,800 --> 00:08:05,837 Uh and what that does is is a real-time 120 00:08:05,920 --> 00:08:11,357 index that lets you track, um you know, what our model Claude is being 121 00:08:11,440 --> 00:08:14,597 used for. It goes across all the conversations and kind of uses Claude in 122 00:08:14,680 --> 00:08:17,557 a privacy-preserving way to, you know, 123 00:08:17,640 --> 00:08:21,637 kind of kind of statistically query how Claude is being used. What are the tasks 124 00:08:21,720 --> 00:08:23,557 it's being used for? To what extent it 125 00:08:23,640 --> 00:08:26,877 is Is it automating versus augmenting 126 00:08:26,960 --> 00:08:31,357 tasks? Um what industries is it being used in? How is it diffusing across 127 00:08:31,440 --> 00:08:35,197 states in the United States and countries countries in the world? We've 128 00:08:35,280 --> 00:08:39,277 We've just added kind of more and more detail here. And And And my view is 129 00:08:39,360 --> 00:08:44,237 until we can measure the shape of this economic transition, any policy is going 130 00:08:44,320 --> 00:08:46,077 to be blind and and misinformed, right? 131 00:08:46,160 --> 00:08:49,037 Many many policies have gone wrong because they're they're based on 132 00:08:49,120 --> 00:08:53,277 premises that are that are fundamentally incorrect. So, so that's that's step 133 00:08:53,360 --> 00:08:55,797 one. Step two is I think we need to 134 00:08:55,880 --> 00:09:00,477 think very carefully about how do we allow people to adapt, right? People can 135 00:09:00,560 --> 00:09:02,797 adapt more quickly or they can adapt 136 00:09:02,880 --> 00:09:05,397 more slowly. This can mean adapting to 137 00:09:05,480 --> 00:09:08,077 use the technology within existing jobs. 138 00:09:08,160 --> 00:09:11,317 This can mean adapting, you know, from 139 00:09:11,400 --> 00:09:15,437 one job to another job. For example, I think there are probably going to be 140 00:09:15,520 --> 00:09:20,477 more jobs in the physical world and less jobs in the knowledge work economy. Now, 141 00:09:20,560 --> 00:09:24,477 maybe eventually robotics, you know, makes progress, but I think that's on a 142 00:09:24,560 --> 00:09:29,317 that's on a slower trajectory. So So that's one. Are there jobs that have, 143 00:09:29,400 --> 00:09:33,717 you know, still kind of, you know, really value a human touch? Um some of 144 00:09:33,800 --> 00:09:37,557 them do, some of them don't. We may find out how important how important that is 145 00:09:37,640 --> 00:09:41,477 in the market and where it's most important. At the level of companies, 146 00:09:41,560 --> 00:09:45,397 what are the moats when software becomes cheap and then subsequently the rest of 147 00:09:45,480 --> 00:09:48,877 knowledge work becomes cheap. We don't know. We've never quite asked that 148 00:09:48,960 --> 00:09:52,037 question. We thought about moats in a certain way. So, there's there's going 149 00:09:52,120 --> 00:09:56,237 to be a huge scramble at the level of at the level of companies. So, you know, 150 00:09:56,320 --> 00:09:59,957 teaching people to adapt, teaching them what to expect, I think is the second 151 00:10:00,040 --> 00:10:03,117 step. And and the third step is I think 152 00:10:03,200 --> 00:10:05,637 there's there's going to need to be some 153 00:10:05,720 --> 00:10:08,397 government in in the displacement that's 154 00:10:08,480 --> 00:10:12,677 that's this macroeconomically large. I just I just don't see how I don't see 155 00:10:12,760 --> 00:10:16,997 how it doesn't happen. The pie is going to grow much larger, right? Like the 156 00:10:17,080 --> 00:10:20,397 money is going to be there. Like, you know, it may you know, the budget may 157 00:10:20,480 --> 00:10:24,797 balance without us doing anything because there's so much growth. Um uh 158 00:10:24,880 --> 00:10:28,757 the issue is distributing it to the right people. And so and so I think I 159 00:10:28,840 --> 00:10:32,637 think this is probably a time to worry less about disincentivizing growth and 160 00:10:32,720 --> 00:10:37,677 and worry more about making making sure that everyone gets gets gets a part of 161 00:10:37,760 --> 00:10:40,797 that growth. Um which which, you know, which I know is the opposite of the 162 00:10:40,880 --> 00:10:45,677 prevailing sentiment now, but I think technological reality is about to change 163 00:10:45,760 --> 00:10:47,997 in a way that forces our ideas to change. 164 00:10:48,080 --> 00:10:51,157 >> So, obviously, in your in your desire [snorts] to create this greater sense of 165 00:10:51,240 --> 00:10:54,957 urgency, are you speaking to people in the administration? I mean, Anthropic 166 00:10:55,040 --> 00:10:58,557 hasn't always been the sort of first on the guest list for this administration, 167 00:10:58,640 --> 00:11:00,917 but you do you have people there that you're talking to? 168 00:11:01,000 --> 00:11:03,357 >> I I've I I I I have said it to them 169 00:11:03,440 --> 00:11:06,997 myself. Um uh and, you know, to to be 170 00:11:07,080 --> 00:11:10,957 clear, there are plenty of things there are plenty of things we agree on, right? 171 00:11:11,040 --> 00:11:13,157 You know, I think the AI action plan 172 00:11:13,240 --> 00:11:15,877 that the administration put out in the 173 00:11:15,960 --> 00:11:20,477 middle of this year actually had some some, you know, some very good ideas 174 00:11:20,560 --> 00:11:24,037 here. You know, I think I think, you know, we probably probably agreed with 175 00:11:24,120 --> 00:11:28,877 the vast vast majority of it. But I think most of all, we just want to say 176 00:11:28,960 --> 00:11:32,557 these things in public and kind of ha and kind of have a public debate about 177 00:11:32,640 --> 00:11:34,077 them, right? We don't we don't control 178 00:11:34,160 --> 00:11:36,597 policy. I think the most useful thing we 179 00:11:36,680 --> 00:11:39,797 can do is describe to the world what 180 00:11:39,880 --> 00:11:44,637 we're seeing and and provide data to the world. And then and then, you know, it's 181 00:11:44,720 --> 00:11:46,797 it's left to the the public in in a 182 00:11:46,880 --> 00:11:49,437 democracy to to, you know, to take that 183 00:11:49,520 --> 00:11:54,037 data and and to use it and to use it to drive policy. We can't drive policy on 184 00:11:54,120 --> 00:11:55,997 our own. >> Are you going to be talking to officials 185 00:11:56,080 --> 00:11:59,357 while you're here? Have you Have you been along to USA house yet? 186 00:11:59,440 --> 00:12:01,877 >> Uh I've not I've not been to USA house. 187 00:12:01,960 --> 00:12:05,677 You know, I I will I will I I, you know, I will be talking to officials during my 188 00:12:05,760 --> 00:12:08,837 during my trip to Davos. >> [laughter] 189 00:12:08,920 --> 00:12:14,237 >> Good. So, um just to go back to our topic then. So, you founded Anthropic um 190 00:12:14,320 --> 00:12:17,277 specifically because you were worried about the that you didn't think that 191 00:12:17,360 --> 00:12:21,437 OpenAI was taking safety seriously enough. Now, some people say that, you 192 00:12:21,520 --> 00:12:25,437 know, the competitive pressures mean that you've gone more hawkish now. I 193 00:12:25,520 --> 00:12:28,477 mean, do Are those competitive pressures, you know, to keep up with 194 00:12:28,560 --> 00:12:31,997 China and keep ahead of China and all the rest of it? Have they Do you think 195 00:12:32,080 --> 00:12:34,317 they have compromised your safety principles? 196 00:12:34,400 --> 00:12:39,277 >> So, we've taken a very different route than some of the other players have. Uh 197 00:12:39,360 --> 00:12:41,277 I I think one of the good choices we 198 00:12:41,360 --> 00:12:45,557 made early was to be a company that is 199 00:12:45,640 --> 00:12:48,117 focused on enterprise rather than 200 00:12:48,200 --> 00:12:52,637 consumers. Um and I think, you know, it's it's very hard to fight your own 201 00:12:52,720 --> 00:12:56,717 business incentives. It's easier to choose a business model where there's 202 00:12:56,800 --> 00:13:01,237 less need to fight your own business incentives. So, you know, I have a lot 203 00:13:01,320 --> 00:13:03,477 of worries about consumer AI that it it 204 00:13:03,560 --> 00:13:06,037 kind of leads to needing to maximize 205 00:13:06,120 --> 00:13:08,997 engagement. It leads to sloth. You know, 206 00:13:09,080 --> 00:13:11,557 we've seen a lot of stuff around ads 207 00:13:11,640 --> 00:13:16,077 from from some of the other players, you know, Anthropic is not a player that 208 00:13:16,160 --> 00:13:21,037 works like that or needs to work like that. We just sell things to businesses 209 00:13:21,120 --> 00:13:25,077 and those things directly have value, right? We We don't need to monetize a 210 00:13:25,160 --> 00:13:27,237 billion free users. We don't need to 211 00:13:27,320 --> 00:13:32,117 maximize engagement for billion free users because we're in some, you know, 212 00:13:32,200 --> 00:13:35,837 death race with with some other, you know, some other some other large 213 00:13:35,920 --> 00:13:40,277 player. And and so I think that has led us think more carefully. But but even 214 00:13:40,360 --> 00:13:44,077 with that we have made sacrifices, you know, we do all these tests on our 215 00:13:44,160 --> 00:13:47,157 models that others have not done. 216 00:13:47,240 --> 00:13:50,597 Um you know, some other players have done them but but you know, I think we've 217 00:13:50,680 --> 00:13:53,797 been the most you know, aggressive and 218 00:13:53,880 --> 00:13:58,437 you know, when we when we run tests that you know, show up certain behaviors in 219 00:13:58,520 --> 00:14:03,157 our model, you know, these things around deception, black mail, sycophancy that 220 00:14:03,240 --> 00:14:07,717 we show in tests that are present in all of the models. But you know, 221 00:14:07,800 --> 00:14:12,117 we make sure to always talk to the public about about these things. 222 00:14:12,200 --> 00:14:16,237 Um and you know, we we've we you know, we pioneered the science of mechanistic 223 00:14:16,320 --> 00:14:20,157 interpretability for looking inside models. So, you know, have we been have 224 00:14:20,240 --> 00:14:23,837 we been perfect? Of course not. I think we've done a generally good job. I mean, 225 00:14:23,920 --> 00:14:28,157 you mentioned China. I you know, I think that's not about competition. That is 226 00:14:28,240 --> 00:14:30,437 that is about actually the public 227 00:14:30,520 --> 00:14:35,757 benefit mission is I'm I'm worried that if autocracies lead in this technology, 228 00:14:35,840 --> 00:14:38,197 it will be a bad outcome for every single person in this room. 229 00:14:38,280 --> 00:14:42,117 >> What are your specific concern there? Is it about the chips? About you know, 230 00:14:42,200 --> 00:14:44,757 sharing data around chips or 231 00:14:44,840 --> 00:14:46,797 >> Yeah, well, I think I mean, you know, 232 00:14:46,880 --> 00:14:51,917 the the the the kind of means of selling the chips, right? That's the that's the 233 00:14:52,000 --> 00:14:55,717 thing that I think will have the most impact on on who is ahead and and who is 234 00:14:55,800 --> 00:14:58,077 not. But you know, the the concern and 235 00:14:58,160 --> 00:15:02,757 you know, it's not about any particular country or certainly not the people in 236 00:15:02,840 --> 00:15:04,797 any country. It's about It's about a 237 00:15:04,880 --> 00:15:06,477 form of government. 238 00:15:06,560 --> 00:15:10,597 I am concerned that AI may be uniquely 239 00:15:10,680 --> 00:15:13,917 well suited to autocracy and to 240 00:15:14,000 --> 00:15:17,997 deepening the repression that we see in autocracies. We already see it in the 241 00:15:18,080 --> 00:15:21,717 kind of surveillance state that is possible with today's technology. But as 242 00:15:21,800 --> 00:15:24,117 you think of the extent to which AI can 243 00:15:24,200 --> 00:15:28,957 make individualized propaganda, can break into any computer system in the 244 00:15:29,040 --> 00:15:30,677 world, um 245 00:15:30,760 --> 00:15:34,437 uh can uh you know, surveil everyone 246 00:15:34,520 --> 00:15:36,757 everyone in a population, detect dissent 247 00:15:36,840 --> 00:15:41,797 everywhere and suppress it, you know, make, you know, a a huge army of drones 248 00:15:41,880 --> 00:15:46,157 that could go after each each each individual person, it's it's it's really 249 00:15:46,240 --> 00:15:49,517 scary. It's it's it's really scary and we have to stop it. 250 00:15:49,600 --> 00:15:53,757 >> And is that something that you feel governments aren't paying enough 251 00:15:53,840 --> 00:15:56,797 attention to? >> I mean, I you know, I think I think it's 252 00:15:56,880 --> 00:16:01,197 I think it's fair to say that, you know, obviously, you know, different countries 253 00:16:01,280 --> 00:16:02,917 think of themselves as having having 254 00:16:03,000 --> 00:16:04,957 geopolitical adversaries, but the 255 00:16:05,040 --> 00:16:08,557 specific focus on we we don't want 256 00:16:08,640 --> 00:16:13,157 autocracies to get this powerful technology and we should have targeted 257 00:16:13,240 --> 00:16:17,797 policies. Like, we don't need to fight them. Like, we just need to not sell 258 00:16:17,880 --> 00:16:20,517 these chips. Um uh 259 00:16:20,600 --> 00:16:25,157 that, you know, I I think there's not enough focus on that. 260 00:16:25,240 --> 00:16:28,957 >> I want to talk a bit more about Claude because I think it's fair to say it's 261 00:16:29,040 --> 00:16:31,477 having a real moment. I mean, we were >> It's it's having a moment. 262 00:16:31,560 --> 00:16:33,237 >> It is having a moment and we we recently 263 00:16:33,320 --> 00:16:37,197 reported on how engineers and regular users are like are getting 264 00:16:37,280 --> 00:16:40,917 Claude-pilled. Um and I just wondered how you feel about 265 00:16:41,000 --> 00:16:44,197 the state of the business today versus a year ago. 266 00:16:44,280 --> 00:16:47,277 >> Yeah, I mean, this is one of these things that's been the growth of the 267 00:16:47,360 --> 00:16:49,957 business has been fast, [snorts] but but 268 00:16:50,040 --> 00:16:54,477 but kind of on on the same smooth exponential curve as the technology. So, 269 00:16:54,560 --> 00:16:57,877 you know, we we you know, we we have this this this revenue curve that in 270 00:16:57,960 --> 00:17:03,237 2023 went from zero to roughly 100 100 million, [snorts] went in 2024 went from 271 00:17:03,320 --> 00:17:06,437 roughly 100 million to roughly a billion, [clears throat] 2025 went from 272 00:17:06,520 --> 00:17:10,197 roughly a a billion to roughly 10 billion. Not exactly, these are rounded 273 00:17:10,280 --> 00:17:14,197 numbers, but but that is that is roughly it. You know, through that, if you go on 274 00:17:14,280 --> 00:17:17,597 Twitter, every couple months, it's like, "Oh my god, Anthropic's changing the 275 00:17:17,680 --> 00:17:21,237 world. Oh my god, you know, Anthropic's totally destroying you know, just just 276 00:17:21,320 --> 00:17:25,397 the excitability of the of of the moment. But but we just watch it and we 277 00:17:25,480 --> 00:17:29,637 watch this curve. It's fast. It's constantly progressing. It's given us 278 00:17:29,720 --> 00:17:32,877 confidence. We never know for sure if it's going to continue. It might not, 279 00:17:32,960 --> 00:17:36,917 but but that has been empirically what we have observed the whole time. And 280 00:17:37,000 --> 00:17:40,077 then there are these moments where even though the curve is smooth, there's a 281 00:17:40,160 --> 00:17:42,317 breakout moment. And so right now, I 282 00:17:42,400 --> 00:17:44,877 think there's a breakout moment around 283 00:17:44,960 --> 00:17:49,677 Claude Code among developers. Um you know, this this thing about being 284 00:17:49,760 --> 00:17:54,397 able to make whole apps and doing things end end. Again, that advanced gradually, 285 00:17:54,480 --> 00:17:58,357 but with our root most recent model Opus [clears throat] 4.5, it just kind of 286 00:17:58,440 --> 00:18:00,197 reached an inflection point, right? 287 00:18:00,280 --> 00:18:03,997 Where you know, the the improvement was gradual, but you know, it's it's just 288 00:18:04,080 --> 00:18:06,917 like you know, you you you boiling the frog. You know, you see you see the 289 00:18:07,000 --> 00:18:10,277 gradual improvement and then there's a specific point at which 290 00:18:10,360 --> 00:18:12,797 suddenly that's the point people notice. 291 00:18:12,880 --> 00:18:17,197 I think the second thing that has that has maybe accelerated that further is we 292 00:18:17,280 --> 00:18:21,637 looked at Claude Code and one of the things we we noticed is there were a lot 293 00:18:21,720 --> 00:18:23,877 of people inside Anthropic and outside 294 00:18:23,960 --> 00:18:26,957 Anthropic who were not technical, but 295 00:18:27,040 --> 00:18:31,557 who realized that Claude Code could do these incredible agentic tasks for you. 296 00:18:31,640 --> 00:18:33,317 It couldn't just write code, it could 297 00:18:33,400 --> 00:18:35,757 also organize your to-do list or plan 298 00:18:35,840 --> 00:18:40,557 your projects or organize your folders or you know, process a bunch of 299 00:18:40,640 --> 00:18:42,437 information and kind of summarize it. So 300 00:18:42,520 --> 00:18:45,037 the the idea that not just a chat bot, 301 00:18:45,120 --> 00:18:49,037 but agentic tasks were were needed. Um 302 00:18:49,120 --> 00:18:52,677 non-technical people were realizing it and they wanted it so much that they 303 00:18:52,760 --> 00:18:55,597 were wrestling with the command line, right? Non-technical, you know, they 304 00:18:55,680 --> 00:18:58,637 have no reason if you're not a programmer, it's such a terrible 305 00:18:58,720 --> 00:19:01,477 interface to use if you're not a programmer. But people were going 306 00:19:01,560 --> 00:19:03,917 through and using it anyway. And so I looked at that and I said, that looks 307 00:19:04,000 --> 00:19:07,357 like unmet demand. And so we we used we 308 00:19:07,440 --> 00:19:10,077 used Claude Code again like two weeks, 309 00:19:10,160 --> 00:19:13,357 you know, to make basically a kind of a 310 00:19:13,440 --> 00:19:16,157 version with a better UI that's 311 00:19:16,240 --> 00:19:18,797 customized for tasks that 312 00:19:18,880 --> 00:19:23,797 tasks other than other than code. And you know, you know, we released it and 313 00:19:23,880 --> 00:19:27,757 you know, within like a day it you know, had like you know, 314 00:19:27,840 --> 00:19:32,557 you know, you know, most of the metrics on the tape were like four times as much as 315 00:19:32,640 --> 00:19:35,397 anything anything anything we'd ever released. 316 00:19:35,480 --> 00:19:38,597 Uh, so you know, those are the two moments. I don't know if these are new 317 00:19:38,680 --> 00:19:41,957 capabilities, but you know, there was just one of these kind of consensus 318 00:19:42,040 --> 00:19:46,197 moments where people got really excited and it's it's it's driving adoption 319 00:19:46,280 --> 00:19:50,157 really fast. I think people are catching up to what the technology is is capable 320 00:19:50,240 --> 00:19:52,997 of because it's reached a certain point and because we built interfaces that 321 00:19:53,080 --> 00:19:55,637 have made it accessible. >> Can you tell us a bit about how you 322 00:19:55,720 --> 00:19:59,597 personally in your life, your family life use a generative AI? 323 00:19:59,680 --> 00:20:03,817 >> Yeah, um, so you know, when I'm writing like, you know, an essay or something 324 00:20:03,900 --> 00:20:06,997 [clears throat] or like, you know, things I say in front of the company, 325 00:20:07,080 --> 00:20:10,557 um, I feel like a fair amount of my job is writing and so I kind of how I have 326 00:20:10,640 --> 00:20:14,757 to plot, you know, come up with sources, help me help me with my writing, that 327 00:20:14,840 --> 00:20:17,997 that kind of thing. >> And then obviously you're having this 328 00:20:18,080 --> 00:20:21,037 great moment and I think there are your it's widely expected that you're going 329 00:20:21,120 --> 00:20:24,317 to IPO this year. Can you tell us a bit about your plans for that? 330 00:20:24,400 --> 00:20:27,797 >> Yeah, I mean, you know, we we don't know for sure. We don't know for sure what 331 00:20:27,880 --> 00:20:31,877 we're going to do and you know, I would say we're we're more focused on just 332 00:20:31,960 --> 00:20:35,837 keeping the revenue curve going, making the models better, selling the models to 333 00:20:35,920 --> 00:20:39,477 people, you know, warning about the societal impacts and bringing bringing 334 00:20:39,560 --> 00:20:42,357 the good stuff. societal impacts. So you know, that's the that's that's kind 335 00:20:42,440 --> 00:20:46,077 that's kind of the highest that's kind of the highest priority right now, but 336 00:20:46,160 --> 00:20:49,477 you know, I'm not saying anything novel if I say that this is an industry with 337 00:20:49,560 --> 00:20:53,437 very high capital demands. Um, uh, and you know, that the there's 338 00:20:53,520 --> 00:20:57,157 only so much at some point that the private markets the private markets can 339 00:20:57,240 --> 00:21:00,197 provide. >> So another model that's absolutely 340 00:21:00,280 --> 00:21:03,917 having a moment is Gemini and it searched to the top of the app store 341 00:21:04,000 --> 00:21:06,837 recently and OpenAI declared code red. 342 00:21:06,920 --> 00:21:09,877 So everyone was got very excited about that. 343 00:21:09,960 --> 00:21:14,597 Do you worry about your ability to compete against Gemini given the sheer 344 00:21:14,680 --> 00:21:18,357 size of Google? >> So, I think I think this is another 345 00:21:18,440 --> 00:21:20,317 place where just just being different 346 00:21:20,400 --> 00:21:22,837 helps. So, you know, the enterprise 347 00:21:22,920 --> 00:21:25,837 strategy, Google and OpenAI are fighting 348 00:21:25,920 --> 00:21:28,397 it out in consumer, right? It They It is 349 00:21:28,480 --> 00:21:30,237 both is existential to both of them. 350 00:21:30,320 --> 00:21:34,757 Existential to OpenAI because that's their whole business. Existential to 351 00:21:34,840 --> 00:21:37,917 Google because they have the search business and that's that's what's being 352 00:21:38,000 --> 00:21:42,357 disrupted by this, so they they need to, you know, replay replace themselves and 353 00:21:42,440 --> 00:21:44,317 and fight the disruption. So, that's 354 00:21:44,400 --> 00:21:47,357 always their first priority. And, you 355 00:21:47,440 --> 00:21:51,037 know, they they seem more They seem much 356 00:21:51,120 --> 00:21:55,597 more focused on that than kind of operating in the enterprise. It's been 357 00:21:55,680 --> 00:21:59,557 great to see what what Gemini is is capable of you know, capable of in 358 00:21:59,640 --> 00:22:02,117 consumer. You know, I think I think they're, you know, I think they're going 359 00:22:02,200 --> 00:22:05,757 about it a different way. I was, you know, I was just on a panel with with 360 00:22:05,840 --> 00:22:08,717 Demis Hassabis who leads research at Google. You know, I think I think he's a 361 00:22:08,800 --> 00:22:12,317 great guy. I've known him for 15 years, so I'm rooting for him. 362 00:22:12,400 --> 00:22:15,837 >> Um while you talk about differences, one difference is, I believe, is that our 363 00:22:15,920 --> 00:22:19,877 topic doesn't have the ability to generate videos and photos. Do you see 364 00:22:19,960 --> 00:22:22,037 that as a a potential weakness? 365 00:22:22,120 --> 00:22:25,797 >> Um I You know, I think for enterprise business, you know, there's there's not 366 00:22:25,880 --> 00:22:28,117 really a demand for like, you know, you 367 00:22:28,200 --> 00:22:33,037 know, photos of you know, cats riding donkeys or, you know, what 368 00:22:33,120 --> 00:22:36,917 whatever. Whatever consumer video people want. 369 00:22:37,000 --> 00:22:40,677 There's maybe an edge case around like slides and presentations, but if we ever 370 00:22:40,760 --> 00:22:42,237 need it, we can just buy them, you know, 371 00:22:42,320 --> 00:22:48,360 we can just contract a model from from from from from a contract a model from 372 00:22:50,360 --> 00:22:53,797 someone else. So, you know, I don't I don't know what will happen. I don't 373 00:22:53,880 --> 00:22:57,317 know what will happen in the future, but I at least don't anticipate needing 374 00:22:57,400 --> 00:23:00,717 this. Um and I think there are problems associated with this. Like, you know, I 375 00:23:00,800 --> 00:23:04,117 think I think, you know, we look at the amount of short-form video out there. 376 00:23:04,200 --> 00:23:08,397 Like, a lot of it's fake. A lot of it's pretty pretty addictive. A lot of it's 377 00:23:08,480 --> 00:23:11,117 slop. Um not to say that all of it is 378 00:23:11,200 --> 00:23:13,637 bad or that necessarily doing it means 379 00:23:13,720 --> 00:23:18,157 you're bad, but but it's not it's not a part of the market that I'm I'm like, 380 00:23:18,240 --> 00:23:22,357 you know, tripping over myself to to get involved in. 381 00:23:22,440 --> 00:23:26,157 >> You mentioned that you were on a panel with Demis Hassabis. And when we were 382 00:23:26,240 --> 00:23:29,797 chatting earlier yesterday, you talked you said something that I thought was 383 00:23:29,880 --> 00:23:34,317 quite interesting, but scientists are approaching the AI AI era or scientists 384 00:23:34,400 --> 00:23:37,917 who are leading these big AI companies are approaching the era differently from 385 00:23:38,000 --> 00:23:42,397 sort of tech entrepreneurs. Can you say a bit more about what you mean by that? 386 00:23:42,480 --> 00:23:44,597 >> Yeah, well, so you know, when I when you 387 00:23:44,680 --> 00:23:47,197 think about this technology, it's really 388 00:23:47,280 --> 00:23:50,197 it's really the intersection of research 389 00:23:50,280 --> 00:23:54,517 that has going been going on for many decades, much of which was academic in 390 00:23:54,600 --> 00:23:57,997 nature until, you know, a decade decade and a half ago. 391 00:23:58,080 --> 00:23:58,597 Um 392 00:23:58,680 --> 00:24:03,037 and the kind of scale needed to um you 393 00:24:03,120 --> 00:24:07,437 know, needed to develop and deploy these technologies over the last decade and a 394 00:24:07,520 --> 00:24:11,797 half, which has only come from the large-scale kind of internet and social 395 00:24:11,880 --> 00:24:15,837 media companies, right? They have the infrastructure, they have the cash. So, 396 00:24:15,920 --> 00:24:20,757 we've seen a world in which some of the companies are essentially led by people 397 00:24:20,840 --> 00:24:22,277 who who have a scientific background. 398 00:24:22,360 --> 00:24:25,757 That's my background. That's Demis's background. Some of them are led by the 399 00:24:25,840 --> 00:24:30,357 generation of entrepreneurs that did social media. Um and I I I think that's 400 00:24:30,440 --> 00:24:32,557 very different. That um you know, 401 00:24:32,640 --> 00:24:34,997 scientists there's a long tradition of 402 00:24:35,080 --> 00:24:38,637 scientists thinking about the effects of 403 00:24:38,720 --> 00:24:40,797 the technology they build, of thinking 404 00:24:40,880 --> 00:24:45,797 of themselves as having responsibility for the technology they build, not 405 00:24:45,880 --> 00:24:49,117 ducking responsibility, right? They're they're they're they're motivated in the 406 00:24:49,200 --> 00:24:51,797 first place by creating something for 407 00:24:51,880 --> 00:24:54,477 the world. And so then they they worry 408 00:24:54,560 --> 00:24:59,877 in the cases where that something can go wrong. Um, I think the motivation of 409 00:24:59,960 --> 00:25:03,117 entrepreneurs, particularly the generation of 410 00:25:03,200 --> 00:25:06,997 uh uh social media entrepreneurs, is very different. The selection effects 411 00:25:07,080 --> 00:25:11,477 that operated on them, the way in which they interacted with, you might say 412 00:25:11,560 --> 00:25:15,517 manipulated consumers, is very different. Um, and and so I think that 413 00:25:15,600 --> 00:25:19,717 that leads to that leads to different attitudes. 414 00:25:19,800 --> 00:25:23,957 >> Um, now we've been taking some questions from readers who submitted their 415 00:25:24,040 --> 00:25:26,997 questions online. But before we do that, I just wanted to ask you one more thing, 416 00:25:27,080 --> 00:25:29,957 which is that you know, in the big picture, tensions 417 00:25:30,040 --> 00:25:32,637 are running very high at the moment between the US and 418 00:25:32,720 --> 00:25:36,437 China. Do you think or wonder about how that might impact how you operate your 419 00:25:36,520 --> 00:25:38,757 business should things escalate? 420 00:25:38,840 --> 00:25:42,277 >> Look, I mean, you know, we we have we have always, you know, we always speak 421 00:25:42,360 --> 00:25:43,957 for ourselves. We've always thought of 422 00:25:44,040 --> 00:25:46,077 ourselves as as as, you know, kind of 423 00:25:46,160 --> 00:25:49,357 kind of our own thing and and and and 424 00:25:49,440 --> 00:25:51,077 and independent, right? We don't go out 425 00:25:51,160 --> 00:25:54,877 of our way to be for or against anyone, 426 00:25:54,960 --> 00:25:57,917 but when we, you know, when we disagree 427 00:25:58,000 --> 00:26:00,117 on policy, we say so. When we agree on 428 00:26:00,200 --> 00:26:02,997 policy, we say so. And we we really keep 429 00:26:03,080 --> 00:26:05,477 it focused on AI. And and so, you know, 430 00:26:05,560 --> 00:26:10,037 I you know, I haven't seen any reluctance folks in other part of parts 431 00:26:10,120 --> 00:26:11,917 of the world to work with us, right? 432 00:26:12,000 --> 00:26:16,517 We're our own thing. We're providing AI models. We try to do that responsibly. 433 00:26:16,600 --> 00:26:20,317 >> And I see a lot of talk this week about AI sovereignty. I'm not entirely sure 434 00:26:20,400 --> 00:26:23,397 what but everybody seems to have their own >> don't know what it means either. 435 00:26:23,480 --> 00:26:25,687 >> Oh, you don't have your own definition? 436 00:26:25,770 --> 00:26:28,197 >> [laughter] >> Do you want to hold this for me for a 437 00:26:28,280 --> 00:26:29,797 minute so I can make it while you do this? 438 00:26:29,880 --> 00:26:35,800 >> readers online. Um, so I'm going to start now with one from Trevor Loomis. 439 00:26:37,040 --> 00:26:40,917 Um, and his question is, what is the single most important technical 440 00:26:41,000 --> 00:26:45,997 breakthrough still missing to make frontier AI reliably safe and controllable 441 00:26:46,080 --> 00:26:48,877 >> [cough] >> in real-world deployment? 442 00:26:48,960 --> 00:26:52,797 >> So, I think we need to make more progress on mechanistic 443 00:26:52,880 --> 00:26:56,597 interpretability, which is the science of looking inside the models. 444 00:26:56,680 --> 00:26:58,357 One of the problem is when we train 445 00:26:58,440 --> 00:27:03,797 these models is that we don't know, you can't be sure they're going to do what 446 00:27:03,880 --> 00:27:06,997 you think they're going to do. You can talk to the model in one context, it can 447 00:27:07,080 --> 00:27:11,437 say all kinds of things. Just as with a human, that may not be a faithful 448 00:27:11,520 --> 00:27:15,757 representation of what they're actually thinking. If they tell you, "I'm doing X 449 00:27:15,840 --> 00:27:19,477 because Y." they might be doing X for a completely different reason. They might 450 00:27:19,560 --> 00:27:21,757 be lying about doing X. Like, we're very 451 00:27:21,840 --> 00:27:23,997 used to these problems with humans, but 452 00:27:24,080 --> 00:27:26,757 but they exist with AI as well. And so, 453 00:27:26,840 --> 00:27:28,917 any kind of phenomenological testing or 454 00:27:29,000 --> 00:27:32,277 training, we can't be certain of. But, 455 00:27:32,360 --> 00:27:36,197 similar to how, you know, you can learn things about human brains by doing an 456 00:27:36,280 --> 00:27:38,997 MRI or an X-ray that you can't learn 457 00:27:39,080 --> 00:27:43,437 just by talking to a human, the science of looking inside the AI models, I am 458 00:27:43,520 --> 00:27:45,317 convinced that this ultimately holds the 459 00:27:45,400 --> 00:27:50,637 key to making the model safe and controllable because it's the only 460 00:27:50,720 --> 00:27:53,117 ground truth we have. >> Right. 461 00:27:53,200 --> 00:27:55,597 Okay. Um I have another question here 462 00:27:55,680 --> 00:28:01,237 from Jim O'Connell. How will AI affect current K-12 educational achievement 463 00:28:01,320 --> 00:28:04,357 gaps? A practical question there from no doubt a parent. 464 00:28:04,440 --> 00:28:08,197 >> Yeah. Yeah. So, you know, I I my you know, there there's kind of the 465 00:28:08,280 --> 00:28:12,317 short-term stuff about, you know, people using AI for cheating, which I think is, 466 00:28:12,400 --> 00:28:17,077 you know, I think is problematic. But, you know, in in in relative terms, okay, 467 00:28:17,160 --> 00:28:20,917 fine, you can have a kind of a different way of um 468 00:28:21,000 --> 00:28:24,037 uh you know, of teaching using AI and you know, we've we've thought about 469 00:28:24,120 --> 00:28:27,717 that. We've released versions of of Claude for for education that are kind 470 00:28:27,800 --> 00:28:29,517 of designed around that. But, I think 471 00:28:29,600 --> 00:28:33,277 the harder problem behind that is, okay, 472 00:28:33,360 --> 00:28:37,197 what skills are we actually teaching in in the world of AI? What does education 473 00:28:37,280 --> 00:28:41,357 look like it look like in the world of AI? And it's it's not so easy because 474 00:28:41,440 --> 00:28:45,837 the disruption is broad and we don't, you know, if someone asked me what 475 00:28:45,920 --> 00:28:48,277 exactly what career should I go into?" 476 00:28:48,360 --> 00:28:52,397 The The uncomfortable truth is I I'm I'm not sure. I can't tell the direction 477 00:28:52,480 --> 00:28:56,557 that it's going to I can't tell the direction that it's going to go yet. I I 478 00:28:56,640 --> 00:29:01,797 will say that I think we should go back to some concepts that we had earlier 479 00:29:01,880 --> 00:29:03,477 about education. Like we we've had a 480 00:29:03,560 --> 00:29:06,877 very kind of like um like economically 481 00:29:06,960 --> 00:29:12,317 inflected, almost mercenary notion notion of education. Um and and and I 482 00:29:12,400 --> 00:29:15,517 think one of the things that we should do is we should maybe move away move 483 00:29:15,600 --> 00:29:20,197 away from that notion back to the idea that like, you know, education is is is 484 00:29:20,280 --> 00:29:22,197 designed to shape you as a person, is 485 00:29:22,280 --> 00:29:24,677 designed to build character, is is, you 486 00:29:24,760 --> 00:29:27,757 know, is is is is is kind of designed to 487 00:29:27,840 --> 00:29:31,717 enrich you and and like make you a better person. I think that's actually a 488 00:29:31,800 --> 00:29:34,357 safer foundation for for education's future. 489 00:29:34,440 --> 00:29:37,837 >> That sounds I'm rather envious of the kids kids who are yet to be educated. 490 00:29:37,920 --> 00:29:40,997 It's kind of education that I think we'd all have liked to have. Um so, to be 491 00:29:41,080 --> 00:29:43,877 fair to everybody in the room then, I just I think we've got time for one 492 00:29:43,960 --> 00:29:48,677 question if anybody would like to ask a question. Yeah? 493 00:29:48,760 --> 00:29:50,517 This lady here. 494 00:29:50,600 --> 00:29:53,117 Yeah Yeah, no no, here's the mic. 495 00:29:53,200 --> 00:29:55,997 >> Um I wanted to ask from a point of view 496 00:29:56,080 --> 00:30:00,397 of the AI labs, uh what kind of responsibility do you hold when there 497 00:30:00,480 --> 00:30:04,397 are economies, countries, and people that are being left behind? Um would 498 00:30:04,480 --> 00:30:07,317 that expand into structurally involving 499 00:30:07,400 --> 00:30:12,397 them, um slowing down, or actually making sure that they're not being left 500 00:30:12,480 --> 00:30:15,477 out? >> Yeah, I worry about that on a whole 501 00:30:15,560 --> 00:30:19,397 bunch of scales. And it's you know, it's not just country versus country. 502 00:30:19,480 --> 00:30:23,557 Certainly I worry about the developing world versus the developed world. Um 503 00:30:23,640 --> 00:30:26,357 where, you know, sometimes the the developed world developing world will 504 00:30:26,440 --> 00:30:28,077 get passed by by by technological 505 00:30:28,160 --> 00:30:30,357 revolutions. But I But you know, I also 506 00:30:30,440 --> 00:30:33,557 worry about divisions within a country. 507 00:30:33,640 --> 00:30:35,397 It It has occurred to me as I've looked 508 00:30:35,480 --> 00:30:38,397 across our customers as the startups are 509 00:30:38,480 --> 00:30:42,997 so fast to adopt AI. And the traditional enterprises, because they're bigger, 510 00:30:43,080 --> 00:30:47,557 because they do a specific thing, they move much slower. Um and we can see it 511 00:30:47,640 --> 00:30:52,157 in our economic data. We can see the diffusion of the technology from states 512 00:30:52,240 --> 00:30:55,877 within the US that adopted quickly and states that move slowly. It is 513 00:30:55,960 --> 00:30:59,557 diffusing. It's getting out there. But there's there's no question that there's 514 00:30:59,640 --> 00:31:02,797 a differential here. If I were to describe the nightmare, and then then 515 00:31:02,880 --> 00:31:06,757 I'll try to describe some, you know, what I think of as a solutions. The 516 00:31:06,840 --> 00:31:08,797 nightmare would be that there's there 517 00:31:08,880 --> 00:31:11,877 there's like this emerging zeroth world 518 00:31:11,960 --> 00:31:16,837 country of like 10 million people that's like 7 million people in the bay, you 519 00:31:16,920 --> 00:31:21,037 know, in like Silicon Valley and you know, 3 million people kind of scattered 520 00:31:21,120 --> 00:31:24,517 scat- scattered scattered throughout that that you know, is is kind of 521 00:31:24,600 --> 00:31:28,037 forming its own economy and it's becoming decoupled or disconnected, 522 00:31:28,120 --> 00:31:32,797 right? Maybe the 10% GDP growth looks like 50% GDP growth in that part of like 523 00:31:32,880 --> 00:31:37,757 this technology is so crazy. It can pull things apart that way. Um I think that 524 00:31:37,840 --> 00:31:41,317 would be a really bad world. I would almost say that is a dystopian world, 525 00:31:41,400 --> 00:31:46,597 right? I And we should think about how to stop that. Um there are, you know, a 526 00:31:46,680 --> 00:31:49,997 number of things Anthropic is thinking about or doing. Um 527 00:31:50,080 --> 00:31:54,277 one is as regards the developing world, we're we are starting to do a lot of 528 00:31:54,360 --> 00:31:58,157 work around public health. Um we've announced stuff with, you know, run 529 00:31:58,240 --> 00:32:02,757 Ministry of Education. We're doing a a lot of work with the Gates Foundation. 530 00:32:02,840 --> 00:32:05,717 Um you know, it it would I wrote about this in Machines Evolving Grace. It 531 00:32:05,800 --> 00:32:08,957 would it would be really great to get these fast economic growth rates, which 532 00:32:09,040 --> 00:32:12,637 in theory should be even faster because it's catch-up growth, um in the 533 00:32:12,720 --> 00:32:16,477 developing world that I predict we're going to get in in the developed world. 534 00:32:16,560 --> 00:32:19,637 Um you know, within within countries, we 535 00:32:19,720 --> 00:32:23,837 need to we you know, we need to think about, you know, how how not to have a 536 00:32:23,920 --> 00:32:27,757 part of the world that just that just kind of decouples, right? How do we get 537 00:32:27,840 --> 00:32:31,477 the economic growth to Mississippi that that, you know, that is coming to this 538 00:32:31,560 --> 00:32:36,197 contained contained area of of of of Silicon Valley. And so, you know, there 539 00:32:36,280 --> 00:32:40,917 we've done work around kind of economic mobility and economic opportunity, but I 540 00:32:41,000 --> 00:32:44,197 think both of these again are going to need to have some some involvement of 541 00:32:44,280 --> 00:32:47,237 the government. We're going to find that that, you know, 542 00:32:47,320 --> 00:32:50,237 ideology will will not survive the 543 00:32:50,320 --> 00:32:51,957 nature of this technology. It won't 544 00:32:52,040 --> 00:32:57,437 survive reality. The things I'm talking about, you know, while you could today 545 00:32:57,520 --> 00:33:00,557 say, "Oh, yeah, they're they're like politically coded in some way." They're 546 00:33:00,640 --> 00:33:03,757 going to become bipartisan and universal because everyone will recognize the 547 00:33:03,840 --> 00:33:07,637 necessity of it. Just mark my words. If we come back, if not next year, the year 548 00:33:07,720 --> 00:33:11,960 after, everyone's going to think this.