WEBVTT - Anthropic President Daniela Amodei Talks the Future of Claude 0:00:02.520 --> 0:00:09.760 Bloomberg Audio Studios, podcasts, radio news. We were chatting about 0:00:10.039 --> 0:00:13.360 not so long ago when we first met and started 0:00:13.360 --> 0:00:18.159 covering Anthropic. But for those who aren't familiar, you and 0:00:18.200 --> 0:00:21.639 your brother Dario, and a handful of other people left 0:00:21.680 --> 0:00:24.480 Opening Eye to start Anthropic back in twenty twenty one. 0:00:25.040 --> 0:00:27.040 Now you employ thousands. 0:00:26.600 --> 0:00:29.520 Of employees, many of them in a reporting chain up 0:00:29.520 --> 0:00:35.280 to you and your role as president. I'm curious what 0:00:35.400 --> 0:00:38.560 is the difference in the leadership style between you and Dario. 0:00:39.200 --> 0:00:41.720 Well, first of all, thank you so much for having me. 0:00:41.760 --> 0:00:44.960 It's great to be here today. And you know, Dario 0:00:45.000 --> 0:00:47.199 and I I really think of how we sort of 0:00:47.240 --> 0:00:50.680 split duties. Dario is this incredible technical visionary. I think, 0:00:51.479 --> 0:00:56.120 you know, years before AI was this amazing idea that 0:00:56.160 --> 0:00:58.920 people were excited about around the world. We had this 0:00:59.040 --> 0:01:02.160 really strong fiction that artificial intelligence is going to be 0:01:02.200 --> 0:01:05.560 a big deal. This is like early twenty tens, and 0:01:05.800 --> 0:01:09.200 I think Daria is constantly reminding the company about the 0:01:09.280 --> 0:01:14.520 scale the ambition, sort of where the technology itself is going. 0:01:14.840 --> 0:01:17.640 I think of my role much more as actually running 0:01:17.680 --> 0:01:19.760 and managing the day to day of the organization. So 0:01:20.400 --> 0:01:23.560 I managed the executive team. We spend time making decisions 0:01:23.600 --> 0:01:26.959 about customers, about our products, about the research, and how 0:01:26.959 --> 0:01:30.360 that applies to the people that are using artificial intelligence. 0:01:30.720 --> 0:01:32.000 And I like to think that the two of us 0:01:32.040 --> 0:01:34.959 are a great balance. And I have to say, like, 0:01:35.319 --> 0:01:37.960 I don't know how one CEO just runs a company 0:01:38.000 --> 0:01:40.120 by themselves. It seems like it's actually much easier to 0:01:40.160 --> 0:01:42.240 do with two people, particularly when they know each other 0:01:42.280 --> 0:01:45.080 really well. 0:01:45.160 --> 0:01:48.280 I remember in one of our conversations earlier, are you 0:01:48.360 --> 0:01:52.800 telling me that you and Daria rarely fight? And I'm curious 0:01:52.840 --> 0:01:55.720 how you manage What are some examples of places where 0:01:55.720 --> 0:01:59.000 maybe you have disagreed and have been able to settle 0:01:59.040 --> 0:01:59.600 it in. 0:01:59.520 --> 0:02:02.280 A healthy way with regard to the business. 0:02:02.640 --> 0:02:04.440 Well As someone who also has a brother, I know, 0:02:04.560 --> 0:02:08.320 you know, you know. I think we've had, like you know, 0:02:08.680 --> 0:02:11.920 several decades to kind of practice fighting and getting and 0:02:11.960 --> 0:02:15.040 still loving each other right, so as early as like 0:02:15.600 --> 0:02:16.200 why did you. 0:02:16.160 --> 0:02:19.400 Take my toy? Can I have that back? Please? 0:02:19.480 --> 0:02:21.320 I think that has served us well in being able 0:02:21.360 --> 0:02:23.840 to operate well together. In anthropic I think what I 0:02:23.880 --> 0:02:26.080 would say is that you know, we do disagree sometimes, 0:02:26.280 --> 0:02:30.000 but I think how you disagree really matters. And I 0:02:30.040 --> 0:02:33.000 feel like there is always like an incredible mutual respect 0:02:33.000 --> 0:02:35.400 between us. So if we look at the same situation 0:02:35.639 --> 0:02:39.839 and we see something different, I like to think that 0:02:39.919 --> 0:02:42.360 what our default is is to be curious, right, So 0:02:42.440 --> 0:02:44.280 I usually view it as like, oh, this about like 0:02:44.320 --> 0:02:45.840 an obvious call to me, Like why do you see 0:02:45.880 --> 0:02:47.800 it that way? Because usually he'll have a piece of 0:02:47.800 --> 0:02:50.639 information that I don't have, or we'll have a different perspective. 0:02:50.840 --> 0:02:52.760 And I would say about eighty to ninety percent of 0:02:52.800 --> 0:02:54.519 the time what ends up happening is we meet in 0:02:54.560 --> 0:02:55.480 the middle, and so. 0:02:55.480 --> 0:02:57.000 We'll think like, oh, I think this is the right 0:02:57.000 --> 0:02:58.160 thing to do. You think this is the right thing 0:02:58.200 --> 0:02:58.440 to do. 0:02:58.680 --> 0:03:00.880 Actually, there's probably a third way to do this that's 0:03:00.919 --> 0:03:03.000 the combination of what we both think that's actually a 0:03:03.000 --> 0:03:05.280 little bit better. I think it's part of the strength 0:03:05.280 --> 0:03:09.560 of having both of us the ability to practice looking 0:03:09.560 --> 0:03:12.959 at situations and decisions just from different points of view 0:03:12.960 --> 0:03:15.960 and perspective, and I think it helps us make better decisions. 0:03:17.120 --> 0:03:19.720 So two years ago, when you were sitting on the 0:03:19.760 --> 0:03:24.000 Bloomberg Tech stage with Dario, Anthropic was still very much 0:03:24.040 --> 0:03:27.680 the underdog in the competition against open AI and just 0:03:27.720 --> 0:03:30.959 in the race to AI generally. Now it's in a 0:03:31.120 --> 0:03:34.160 very different position. No longer the underdog. Claude is a 0:03:34.240 --> 0:03:37.960 runaway success, particularly among coders. Your company is projecting forty 0:03:38.000 --> 0:03:41.520 seven billion dollars in annualized run rate revenue and the 0:03:41.560 --> 0:03:44.440 company has eclipsed Opening Eyes valuation for the first time. 0:03:45.920 --> 0:03:48.200 Do you feel like you're definitively the front runner right 0:03:48.240 --> 0:03:51.200 now and how does that change your approach it. 0:03:51.280 --> 0:03:53.480 It's something that I think both Dario and I, you know, 0:03:53.760 --> 0:03:57.320 really hammer at Anthropic is that that is just completely 0:03:57.400 --> 0:03:59.480 the wrong way to think. I think our view has 0:03:59.520 --> 0:04:01.720 always been and you know, this is really about how 0:04:01.720 --> 0:04:04.160 do we show up every day for our customers, how 0:04:04.160 --> 0:04:06.440 do we stay humble and make sure that we're really 0:04:06.480 --> 0:04:08.840 focused on the mission. Right The whole reason we started 0:04:08.840 --> 0:04:11.520 Anthropic is we want to be able to build and 0:04:11.600 --> 0:04:15.320 develop this technology in a way that is ethical, that's responsible, 0:04:15.640 --> 0:04:19.480 that's fair. And I think it's really incumbent upon everybody 0:04:19.520 --> 0:04:22.120 at the company, but especially leadership, to say, you know, 0:04:22.279 --> 0:04:24.360 all of these numbers, they're actually not the point, right. 0:04:24.400 --> 0:04:27.719 We are really here to do a job. We're here 0:04:27.760 --> 0:04:30.279 to support the businesses that rely on us every day. 0:04:30.440 --> 0:04:34.800 AI is increasingly a bigger part of everybody's workflow, their 0:04:34.839 --> 0:04:37.760 experience at the office, their experience in their personal lives. 0:04:37.760 --> 0:04:40.600 That's a huge privilege and responsibility. And so I think, 0:04:40.640 --> 0:04:43.320 if anything, you know, we work really hard every day 0:04:43.400 --> 0:04:46.280 to say, like our job is to be the best 0:04:46.360 --> 0:04:49.320 version of who we can possibly be as a company 0:04:49.440 --> 0:04:52.599 every day. That takes work, that takes humility, that takes time, 0:04:52.680 --> 0:04:54.279 that takes focus and energy. 0:04:55.400 --> 0:04:57.200 I want to talk about the big news earlier this 0:04:57.240 --> 0:05:01.280 week that Anthropic has filed confidentially for an IPO SpaceX 0:05:01.320 --> 0:05:05.920 soulso filed Opening Eyes expected to file anytime now. How 0:05:06.000 --> 0:05:08.560 far back had you been planning to file and do 0:05:08.600 --> 0:05:11.320 you think that there's a race here to be first 0:05:11.560 --> 0:05:12.799 among BAA companies. 0:05:13.640 --> 0:05:16.480 So, yes, you're correct, we confidentially you know, filed our 0:05:16.560 --> 0:05:20.839 S one and that gives us the option you know, 0:05:20.920 --> 0:05:24.680 to potentially go public after the SEC review. Unfortunately, that's 0:05:24.680 --> 0:05:27.279 all I can say about about anything IPO related. 0:05:27.320 --> 0:05:28.320 I'm sure you understand. 0:05:28.920 --> 0:05:30.520 I mean, if we can just. 0:05:30.480 --> 0:05:33.880 Maybe level up a little, what is the calculus for 0:05:33.960 --> 0:05:37.760 a kind of an AI company generally to go public 0:05:37.839 --> 0:05:39.279 or not in this moment? What are some of the 0:05:39.320 --> 0:05:41.279 pros and cons here? Because On the one hand, yes, 0:05:41.320 --> 0:05:44.760 you can access more capital if you're not a startup, 0:05:44.800 --> 0:05:47.440 but on the other then you're beholden to you know, 0:05:47.520 --> 0:05:50.680 quarterly results, shareholder calls, et cetera. So what are some 0:05:50.720 --> 0:05:53.440 of the pros and cons just generally for an AI startup? 0:05:53.480 --> 0:05:55.000 Do you think to go public in this moment? 0:05:55.640 --> 0:05:55.760 So? 0:05:56.320 --> 0:05:58.960 I think, you know, at least speaking for you know, ourselves, 0:05:59.040 --> 0:06:01.839 and I think ideas probably really for the AI industry 0:06:01.880 --> 0:06:02.480 more broadly. 0:06:03.000 --> 0:06:03.920 It's a very. 0:06:03.839 --> 0:06:07.160 Capital intensitive business to train AI models. I think that's 0:06:07.200 --> 0:06:09.600 something we've been you know, very open about from from 0:06:09.680 --> 0:06:11.680 day one. I don't think that's you know, a surprise 0:06:11.720 --> 0:06:14.400 in the industry. And I think this kind of concept 0:06:14.400 --> 0:06:17.000 of like how do you access the level of capital 0:06:17.040 --> 0:06:19.880 that you need to train these models? Right, it's a 0:06:20.000 --> 0:06:22.880 it's a really big kind of upfront cost to train 0:06:22.920 --> 0:06:25.320 the models and to sort of inference on them. And 0:06:26.480 --> 0:06:29.040 I think that is you know, my guess is that 0:06:29.200 --> 0:06:33.159 you know, over time, the sort of core set of 0:06:33.200 --> 0:06:38.120 companies that are working to advance the frontier are just 0:06:38.160 --> 0:06:41.200 going to need access to capital, and I think the 0:06:41.240 --> 0:06:44.080 public market is very well suited to that. So there's obviously, 0:06:44.160 --> 0:06:46.640 of course, trade offs. But my sense is there's just 0:06:46.680 --> 0:06:51.720 a fundamental structure of how training and kind of serving 0:06:51.800 --> 0:06:56.960 these models for customers works that it will require that 0:06:57.040 --> 0:06:58.240 level of access. 0:06:59.560 --> 0:07:02.640 It's clear that many businesses can't get enough of Claude. 0:07:03.279 --> 0:07:06.440 But at the same time, we are hearing about some 0:07:06.839 --> 0:07:09.600 customers who are concerned about the pitfalls of so called 0:07:09.640 --> 0:07:13.560 token maxing of you know, companies making leaderboards of who 0:07:13.640 --> 0:07:18.240 uses the most AI tokens for using AI's sake Ubers, 0:07:18.920 --> 0:07:20.920 one of ers executives talked about it getting harder to 0:07:21.000 --> 0:07:23.520 justify AI spend if there isn't a clear metrics backed 0:07:23.680 --> 0:07:26.880 ROI do you think there's truth to idea that some 0:07:27.040 --> 0:07:32.040 businesses may be overspending on AI tools in this early 0:07:32.120 --> 0:07:35.000 experimental phase and they may cut back as they move 0:07:35.040 --> 0:07:35.760 out of that phase. 0:07:36.560 --> 0:07:38.960 So I think kind of two things are are true 0:07:38.960 --> 0:07:39.480 at once here. 0:07:39.560 --> 0:07:42.400 I think, you know, I like that you said, you know, experimental, 0:07:42.520 --> 0:07:45.280 because I think the reality is it feels like these 0:07:45.280 --> 0:07:48.840 AI tools are so powerful and so capable, which is true, 0:07:49.200 --> 0:07:51.640 and comparing them, you know, particularly to you know, even 0:07:51.680 --> 0:07:53.280 two years ago when I was when I was on 0:07:53.320 --> 0:07:56.360 stage at Bloomberg with Dario, I think the models have 0:07:56.440 --> 0:07:58.480 just come so far in what they're able to do, 0:07:58.560 --> 0:08:01.400 in the economic value that they but I actually think 0:08:01.440 --> 0:08:04.160 there's a lot more distance to go still for what 0:08:04.200 --> 0:08:06.559 the models will be able to do, you know, two 0:08:06.880 --> 0:08:09.360 to four to six to eight years in the future, 0:08:09.840 --> 0:08:13.840 and I would suspect that we will see some you know, 0:08:14.040 --> 0:08:18.160 some experimentation in different fields to understand how do you 0:08:18.240 --> 0:08:21.120 kind of get the most value out of these models 0:08:21.120 --> 0:08:22.480 for what particular workflows. 0:08:23.040 --> 0:08:24.160 I think the second thing. 0:08:24.040 --> 0:08:28.800 That's true is, you know, how businesses are choosing to 0:08:28.960 --> 0:08:33.360 use AI will change, So the use cases today, some 0:08:33.400 --> 0:08:36.120 of them, I expect, will continue to be kind of 0:08:35.559 --> 0:08:41.599 the primary driver of you know, efficiency or creativity or 0:08:41.880 --> 0:08:44.120 you know, new approaches to doing things within businesses, whether 0:08:44.120 --> 0:08:48.160 that's coding or you know, financial services, legal, healthcare. But 0:08:48.240 --> 0:08:52.120 I also think as the industry as a whole, not 0:08:52.160 --> 0:08:54.360 just the a the AI industry, but really the business 0:08:54.400 --> 0:08:57.400 community gets more familiar with the tools. 0:08:57.400 --> 0:08:59.600 We're all going to learn together like what is the. 0:08:59.520 --> 0:09:02.679 Best way to apply these tools that also supports employees. 0:09:03.120 --> 0:09:05.320 And so I think today there's this there's this feeling 0:09:05.320 --> 0:09:07.840 that it's like, oh, like AI, you know, the leaderboards, 0:09:07.880 --> 0:09:09.120 and it's like I have to use it and what 0:09:09.160 --> 0:09:11.120 do I even use it for? And I think my 0:09:11.280 --> 0:09:14.080 hope is that over time it'll be more incorporated into 0:09:14.120 --> 0:09:16.200 the day to day of how humans do our work, 0:09:16.320 --> 0:09:19.160 how we communicate together, and that there will actually be 0:09:19.200 --> 0:09:21.200 a lot more value realized in a way that that 0:09:21.240 --> 0:09:22.319 feels really good to people. 0:09:22.520 --> 0:09:24.280 Does Xanthropic have a leaderboard itself. 0:09:24.600 --> 0:09:27.120 We don't have a leaderboard, no, or at least not 0:09:27.120 --> 0:09:30.440 not in the way you've described. We do track how 0:09:30.520 --> 0:09:33.360 much we're using Claude internally and for what use cases, 0:09:33.600 --> 0:09:36.840 because we really try to prototype all of the tools 0:09:36.840 --> 0:09:38.960 that we give to our customers, and so all of 0:09:38.960 --> 0:09:41.040 the products that you know you see, whether that's you know, 0:09:41.080 --> 0:09:44.520 Claud code, Claud design, cowork, like all of these are 0:09:44.559 --> 0:09:48.160 things that we actually built Adanthropic first because we said, hey, 0:09:48.160 --> 0:09:50.000 this is a real need we have, could Claud help 0:09:50.080 --> 0:09:50.520 us with this? 0:09:50.880 --> 0:09:52.040 And so I think it's important for. 0:09:52.040 --> 0:09:54.839 Us to just have a kind of a metric for 0:09:55.240 --> 0:09:57.120 checking like how are we using it and how much 0:09:57.160 --> 0:09:59.720 are we using it? But there's not like you must 0:09:59.840 --> 0:10:02.000 use and you must use Claud and it's better if 0:10:02.040 --> 0:10:02.640 you use it or not. 0:10:03.240 --> 0:10:07.560 What type of job function outside of coding or research 0:10:08.240 --> 0:10:10.480 are using Claude most internally. 0:10:11.200 --> 0:10:14.840 The entire company uses Claude for topics large and small. 0:10:15.120 --> 0:10:17.640 Probably the second biggest, I would actually say is finance 0:10:17.960 --> 0:10:19.439 and so a lot of the kind of back end 0:10:19.480 --> 0:10:24.200 financial planning analysis, you know, like just number analysis. I 0:10:24.200 --> 0:10:27.800 think claud is really really helpful at that. But honestly, 0:10:27.840 --> 0:10:31.240 like our people team built an incredible tool internally for 0:10:31.280 --> 0:10:33.480 how we do performance reviews that leverages Claude run in 0:10:33.480 --> 0:10:35.680 the middle performance review season. I have to say, like 0:10:35.720 --> 0:10:38.920 the feedback from our team has been so positive. Claude either, 0:10:39.000 --> 0:10:40.560 like this just feels so much more fun to do this. 0:10:40.640 --> 0:10:44.520 It's so much more interesting, interactive. It's pulling in information 0:10:44.679 --> 0:10:46.320 about what I did over the course of the past 0:10:46.360 --> 0:10:48.640 six months when I'm assessing myself, and I think that 0:10:48.720 --> 0:10:52.600 it really just speaks to the generality of these tools 0:10:53.160 --> 0:10:56.520 their ability to I think, reach across a business and 0:10:56.960 --> 0:11:00.120 just take a lot of information in and help you, 0:11:00.360 --> 0:11:02.480 as an employee be more successful in your role. 0:11:04.000 --> 0:11:07.120 Dario Anthropics leadership has been vocal about the need for 0:11:07.200 --> 0:11:11.200 more compute. Your company recently made a major deal with 0:11:11.440 --> 0:11:13.040 Xai to least compute. 0:11:13.040 --> 0:11:16.400 There are others. Opening Eye made a. 0:11:16.320 --> 0:11:21.360 Big splash sort of earlier on with securing huge data 0:11:21.360 --> 0:11:24.120 center deals and spending you know, up to a trillion 0:11:24.160 --> 0:11:27.439 on infrastructure. Why did Anthropic take a little bit of 0:11:27.480 --> 0:11:29.920 a different approach, And in hindsight, do you think you 0:11:29.960 --> 0:11:30.320 should have. 0:11:30.320 --> 0:11:32.440 Done more earlier on on the compute front. 0:11:33.520 --> 0:11:35.960 We've talked publicly about this concept of sort of the 0:11:36.040 --> 0:11:39.760 cone of uncertainty related to compute, and I think something 0:11:39.800 --> 0:11:43.760 that's important to know is the sort of structure of 0:11:43.800 --> 0:11:47.120 these deals is you have to commit to a certain 0:11:47.120 --> 0:11:50.120 amount of compute you know, reasonably far in advance. And 0:11:50.160 --> 0:11:52.640 so I think in thropics view has always been we 0:11:52.760 --> 0:11:55.360 are we are wanting to, you know, plan for the 0:11:55.360 --> 0:11:59.360 best outcome, but not over extend ourselves such that we're 0:11:59.400 --> 0:12:03.440 buying more compute than we could productively use. It's really 0:12:03.520 --> 0:12:07.120 hard to predict that, you know, perfectly, and I think 0:12:07.200 --> 0:12:09.040 we would much prefer to be on the side of 0:12:09.120 --> 0:12:12.880 having a little bit more demand for the product then 0:12:12.920 --> 0:12:16.160 we're able to serve than the inverse where you overshoot 0:12:16.160 --> 0:12:18.400 and then you're actually like, not in a great situation 0:12:18.600 --> 0:12:20.760 because you've bought something you can't pay for it down 0:12:20.800 --> 0:12:21.120 the road. 0:12:21.679 --> 0:12:23.120 It is really hard to get this right. 0:12:23.160 --> 0:12:26.200 I think the industry as a whole is still grappling 0:12:26.240 --> 0:12:29.880 with how to kind of structure these deals. But I 0:12:29.920 --> 0:12:34.040 think my sense is it's impossible to know what the 0:12:34.080 --> 0:12:37.360 future will look like. On balance, I continue to think 0:12:37.400 --> 0:12:40.520 it's better to be fiscally responsible and think you know 0:12:40.600 --> 0:12:42.760 carefully about how much you're going to buy, make sure 0:12:42.760 --> 0:12:44.720 that you have the ability to actually use all of 0:12:44.720 --> 0:12:48.240 that compute in the future, and we'll probably undershoot or 0:12:48.280 --> 0:12:49.839 overshoot at some point A little bit. 0:12:50.679 --> 0:12:54.559 In the announcement with SpaceX, Anthropic express preliminary interest in 0:12:54.679 --> 0:12:57.280 using potential data centers in space. 0:12:58.240 --> 0:12:59.880 What do you think about that? Is that a real 0:13:00.360 --> 0:13:01.520 and how far out are we for that. 0:13:03.080 --> 0:13:05.199 I don't think that data centers in space are something 0:13:05.200 --> 0:13:08.720 that will be on our RTD list in twenty twenty seven, 0:13:09.600 --> 0:13:14.160 but I will say AI is the field that has 0:13:14.200 --> 0:13:16.560 surprised me the most. It's probably surprised the world the 0:13:16.559 --> 0:13:19.559 most in terms of just what new things are possible 0:13:19.840 --> 0:13:22.240 are made possible by the advent of this technology. So 0:13:22.880 --> 0:13:26.319 no immediate immediate plans for working with the astronauts to 0:13:26.600 --> 0:13:29.480 get space center data centers going, but you never know. 0:13:31.320 --> 0:13:34.280 A few months ago Anthropic had a very public battle 0:13:34.320 --> 0:13:36.880 with the Pentagon over restrictions on. 0:13:36.880 --> 0:13:38.480 Its AI software. 0:13:38.520 --> 0:13:41.360 Are you more or less optimistic today about finding the 0:13:41.400 --> 0:13:42.319 solution to that? 0:13:43.679 --> 0:13:48.319 Anthropic has been really leaned in from day one and 0:13:48.520 --> 0:13:52.440 talked very publicly about our commitment to national security, and 0:13:52.679 --> 0:13:56.040 we were the first AI company available on the top 0:13:56.080 --> 0:13:59.760 secret cloud. I think our commitment to these values, in 0:13:59.800 --> 0:14:02.800 these principles are also very very old, right. 0:14:02.800 --> 0:14:04.600 It's sort of not not a new not a new 0:14:04.640 --> 0:14:05.160 thing for us. 0:14:05.160 --> 0:14:08.080 But I have to say I've been very impressed the 0:14:08.120 --> 0:14:10.880 degree to which we've been able to work productively, you know, 0:14:10.880 --> 0:14:14.000 with the administration around a wide variety of topics, and 0:14:14.040 --> 0:14:16.760 I think that's actually been the been the much bigger 0:14:16.800 --> 0:14:19.720 story over a long period of time is our ability 0:14:19.760 --> 0:14:23.840 to partner productively with government at all kinds of levels, 0:14:24.200 --> 0:14:27.360 because fundamentally, I think artificial intelligence is and will be 0:14:27.400 --> 0:14:31.200 a geopolitical issue, and in order to be an ethical 0:14:31.280 --> 0:14:35.320 and responsible lab we need to work with governments around 0:14:35.320 --> 0:14:38.440 the world, in the US but also in partner countries 0:14:38.520 --> 0:14:40.760 to say, how can we how can we roll this 0:14:40.840 --> 0:14:43.200 technology out in a way that's safe, in a way 0:14:43.240 --> 0:14:46.320 that's good for people, that's going to protect democracy, and 0:14:46.360 --> 0:14:48.320 so I do really feel optimistic about this in the 0:14:48.320 --> 0:14:48.760 long run. 0:14:49.360 --> 0:14:51.760 Do you think other companies like Open Eye, Google have 0:14:51.840 --> 0:14:54.720 been able to negotiate better deals with the government because 0:14:54.720 --> 0:14:56.800 of Anthropics sticking to its position. 0:14:58.000 --> 0:15:01.760 You know, I think every individual jewel company is going 0:15:01.800 --> 0:15:04.240 to have their own stance and sort of their own 0:15:04.360 --> 0:15:07.920 principles about like what their redlines and what their values are. 0:15:08.520 --> 0:15:11.040 And my sense is Anthropic. 0:15:10.640 --> 0:15:14.320 Will always be as open as we can be about 0:15:14.400 --> 0:15:17.800 what our principles and values are why we have those principles. 0:15:18.360 --> 0:15:19.640 But ultimately, I think it's. 0:15:19.520 --> 0:15:22.920 Important that whatever those values are for you as a company, 0:15:23.160 --> 0:15:25.520 that you are true to them, that you feel like 0:15:25.560 --> 0:15:28.040 you can explain them to employees and to the world 0:15:28.040 --> 0:15:30.880 more broadly and fundamentally. You know, we can't really control 0:15:30.920 --> 0:15:33.960 what other businesses are going to do. I think anthropics 0:15:34.080 --> 0:15:36.400 role is to be the sort of best version of 0:15:36.440 --> 0:15:39.800 itself again, to say, like, if we were the only 0:15:39.960 --> 0:15:43.440 organization in this, you know, in this situation, how would 0:15:43.480 --> 0:15:46.560 we want to behave I think that's that's really the 0:15:46.600 --> 0:15:49.520 best way We've been able to navigate many difficult decisions 0:15:49.520 --> 0:15:51.080 at the company over over the years. 0:15:52.640 --> 0:15:54.720 I think philanthropy it seems to be a big part 0:15:54.760 --> 0:15:57.120 of anthropics culture. I know you and many Anthropic co 0:15:57.160 --> 0:16:00.440 founders have pledged to give away half of your equity 0:16:00.000 --> 0:16:04.000 eighty percent, sorry, eighty percent of your equity. I'm curious 0:16:04.040 --> 0:16:07.040 would you be in support of something like the proposed 0:16:07.680 --> 0:16:10.560 billionaire tax ballot in California that would impose a one 0:16:10.600 --> 0:16:13.640 time tax on individuals whose networth exceeds one billion. 0:16:14.360 --> 0:16:17.480 So anthropic has I think, again, like pretty publicly talked 0:16:17.480 --> 0:16:19.840 about the fact that we think AI is going to 0:16:19.880 --> 0:16:22.480 create a huge amount of wealth. We're already seeing that, 0:16:22.960 --> 0:16:25.840 and I think we've been pretty open about this idea 0:16:25.920 --> 0:16:29.960 that that should be redistributed in some fashion. Both the 0:16:30.000 --> 0:16:32.520 pledge and then a lot of sort of views of 0:16:32.560 --> 0:16:36.280 anthropic when we're talking about, you know, potential for labor displacement, 0:16:36.520 --> 0:16:39.600 are really grounded on this belief that if we don't 0:16:39.800 --> 0:16:45.360 take intentional actions, AI, like many technologies, will probably widen 0:16:45.600 --> 0:16:48.280 the gap of inequality and disparity. 0:16:48.360 --> 0:16:49.840 It's not what we want to see. 0:16:50.000 --> 0:16:51.840 So I think there are a lot of interesting suggestions 0:16:51.880 --> 0:16:53.960 kind of floating around for how to approach that. 0:16:54.280 --> 0:16:55.720 Of course, the part we can control is. 0:16:55.680 --> 0:16:57.480 You know, what do we as a company and as 0:16:57.520 --> 0:17:00.280 individuals as co founders sort of choose to do with 0:17:00.360 --> 0:17:03.280 you know, any you know, economic benefits that we get 0:17:03.320 --> 0:17:05.920 from AI. But I think this is a really big 0:17:05.960 --> 0:17:08.560 and important question that actually goes beyond the walls of 0:17:08.560 --> 0:17:13.080 any individual company. 0:17:13.560 --> 0:17:16.359 There's a lot of potential upside with AI to cure 0:17:16.440 --> 0:17:19.080 disease or lower the cost of consumer goods, but also 0:17:19.200 --> 0:17:20.520 real fears that. 0:17:20.440 --> 0:17:23.280 You know, Dario is validated saying last year that AI. 0:17:23.160 --> 0:17:25.680 Could potentially wipe out half of all white collar work. 0:17:26.240 --> 0:17:29.320 Do you agree with that, and if so, what's the solution? 0:17:29.520 --> 0:17:33.960 Is it something like more taxes? We talked about basic income, 0:17:34.119 --> 0:17:36.679 job retraining, and how do you pay for all of that? 0:17:38.520 --> 0:17:39.000 Sorry? 0:17:39.640 --> 0:17:43.639 I think AI is We're already seeing the ways that 0:17:43.680 --> 0:17:47.080 it's disruptive. I don't necessarily think we know the future. 0:17:47.400 --> 0:17:51.320 So exactly what the type of disruption that might happen, 0:17:52.440 --> 0:17:54.560 you know, a year or three years or five years. 0:17:54.359 --> 0:17:55.680 From now is unknown. 0:17:56.440 --> 0:17:59.639 I do think it's important for companies to be open 0:17:59.680 --> 0:18:02.520 about and study what we're seeing today, which is part 0:18:02.520 --> 0:18:06.040 of why we publish our Societal Impacts research. We say, hey, 0:18:06.080 --> 0:18:09.720 here is how people are using AI. Is it displacing jobs? 0:18:09.760 --> 0:18:13.080 Is it supplementing jobs. What we found so far is 0:18:13.080 --> 0:18:16.399 that in twenty twenty five and twenty twenty six, replacement 0:18:16.440 --> 0:18:18.399 is a tiny, tiny, tiny. 0:18:18.119 --> 0:18:20.560 Fraction of what AI is doing. 0:18:20.920 --> 0:18:25.080 And really where we see that is mostly in jobs 0:18:25.119 --> 0:18:28.159 that are overseas and mostly in customer support, and so 0:18:28.200 --> 0:18:31.080 these are jobs that are sort of already being automated 0:18:31.080 --> 0:18:34.800 by more traditional mL non generative AI systems. Could that 0:18:34.880 --> 0:18:39.000 change in the future, absolutely, And I think we're a 0:18:39.040 --> 0:18:41.240 little bit unusual in the sense that we talk about 0:18:41.280 --> 0:18:43.679 it a lot. We say, hey, this might happen. But 0:18:43.760 --> 0:18:46.679 I think there is this just broader again sort of 0:18:46.720 --> 0:18:49.560 societal challenge. AI is going to be able to do 0:18:49.800 --> 0:18:53.240 so many productive things that humans can do today. 0:18:53.480 --> 0:18:54.679 What does that mean. 0:18:54.760 --> 0:18:57.840 For how we find meaning, how we earn income, how 0:18:57.840 --> 0:19:00.280 we relate to one another? And again, I think the 0:19:00.359 --> 0:19:04.160 default will be to treat it like past technologies and say, Okay, 0:19:04.400 --> 0:19:06.440 we're going to integrate this into existing workflows and that 0:19:06.520 --> 0:19:09.479 sort of the end of the story. I personally believe 0:19:09.600 --> 0:19:12.960 there's a real opportunity here for us to say, how 0:19:13.000 --> 0:19:18.199 do we accelerate and sort of accentuate the parts of 0:19:18.320 --> 0:19:21.000 doing work and finding meaning that only humans can do? 0:19:21.640 --> 0:19:25.040 And this very fundamental belief that humans like to spend 0:19:25.080 --> 0:19:27.520 time with other humans. We like to create things together, 0:19:27.600 --> 0:19:30.320 we like to relate to one another. Sometimes we like 0:19:30.359 --> 0:19:33.000 to disagree with each other. And I don't think AI 0:19:33.080 --> 0:19:35.879 will fundamentally take that away from us. But we have 0:19:35.960 --> 0:19:38.280 to figure out how we apply that within the existing 0:19:38.359 --> 0:19:42.080 economic infrastructure so that people can still find meaning in 0:19:42.280 --> 0:19:44.520 their work and so that people have a way to 0:19:44.640 --> 0:19:46.920 earn a livelihood. Again, I think there could be a 0:19:46.960 --> 0:19:48.840 lot of different ways to structure that, but I think 0:19:48.840 --> 0:19:51.639