WEBVTT 00:00:04.050 --> 00:00:08.527 I love this library. It is, it is, it is absolutely beautiful library. 00:00:08.610 --> 00:00:10.860 Are you a big reader? You know, I mean, I, I, 00:00:11.960 --> 00:00:15.787 I read a lot in general. I don't know that I've had that much time in 00:00:15.870 --> 00:00:17.160 over the last year or so. 00:00:18.780 --> 00:00:21.000 Dario Amodei is an unlikely AI celebrity. 00:00:22.080 --> 00:00:26.047 He's known for warning the world about the risks of artificial intelligence. 00:00:26.130 --> 00:00:31.267 Now, his company is an AI frontrunner valued at nearly a trillion dollars. 00:00:31.350 --> 00:00:34.207 You're shipping so much, so fast. How are you doing that? 00:00:34.290 --> 00:00:36.817 We use Claude, you know, across the product development cycle, 00:00:36.900 --> 00:00:39.510 and it allows us to release very fast. 00:00:40.710 --> 00:00:44.677 Founded by a team of OpenAI defectors in 2021, Anthropic 00:00:44.760 --> 00:00:47.227 started out as an underdog lab. 00:00:47.310 --> 00:00:50.857 Today it's the breakout star, wiping billions in value off 00:00:50.940 --> 00:00:53.977 software stocks, going head-to-head with the Pentagon, 00:00:54.060 --> 00:00:56.947 and creating models allegedly powerful enough to burst 00:00:57.030 --> 00:00:59.227 through the walls of modern cybersecurity. 00:00:59.310 --> 00:01:01.597 Some of the early companies that we gave this 00:01:01.680 --> 00:01:06.180 to said things like, this is a super weapon, please don't release this. 00:01:09.570 --> 00:01:13.147 Guiding Anthropic through this is a sibling duo, Dario, 00:01:13.230 --> 00:01:15.937 the brother and visionary, and Daniela, the sister 00:01:16.020 --> 00:01:20.077 and operator who puts Dario's swirling cosmic thoughts into action. 00:01:20.160 --> 00:01:22.507 Dario and I, we've always been really close since we were 00:01:22.590 --> 00:01:25.327 little, but I think we always really wanted to do something big together. 00:01:25.410 --> 00:01:27.150 Okay, but when you argue who wins? 00:01:28.680 --> 00:01:33.727 No one, No one. As the AI arms race escalates, the Amodeis are eager 00:01:33.810 --> 00:01:36.457 to establish themselves as the good guys. 00:01:36.540 --> 00:01:39.217 But Anthropic's technology could have profound implications 00:01:39.300 --> 00:01:43.507 for how humanity works, learns thinks, even fights wars. 00:01:43.590 --> 00:01:46.057 Anthropic is run by an ideological lunatic. 00:01:46.140 --> 00:01:51.697 This is a debate about what the proper use of AI by the government is. 00:01:51.780 --> 00:01:54.697 At some point, the models become very dangerous, you know, 00:01:54.780 --> 00:01:55.897 the bad guys will have it too. 00:01:55.980 --> 00:01:59.887 And at that point, we have to make sure that good guys have an even better model. 00:01:59.970 --> 00:02:02.647 The name Anthropic actually comes from the Greek word 00:02:02.730 --> 00:02:07.357 for human, which speaks to their mission to build responsible AI 00:02:07.440 --> 00:02:09.600 for the long-term benefit of humanity. 00:02:10.440 --> 00:02:12.517 Can you actually do that when you're building the most 00:02:12.600 --> 00:02:14.430 powerful technology in the world? 00:02:20.310 --> 00:02:23.137 You are at the center of the AI universe right now. 00:02:23.220 --> 00:02:27.877 What does that feel like? The experience I've had for my whole career, 00:02:27.960 --> 00:02:31.117 and certainly the whole time at Anthropic, is 00:02:31.200 --> 00:02:33.607 that there's this kind of smooth exponential, 00:02:33.690 --> 00:02:35.977 and the experience of the smooth exponential is, 00:02:36.060 --> 00:02:38.407 nothing's happening, nothing's happening, nothing's happening. 00:02:38.490 --> 00:02:41.677 Little things happen, and then zoom, it goes crazy. 00:02:41.760 --> 00:02:43.657 So, you know, I was watching this graph for a while 00:02:43.740 --> 00:02:46.177 and I said, oh yeah, we'll probably become the, you know, 00:02:46.260 --> 00:02:49.837 the, the AI company with, you know, the, the most revenue 00:02:49.920 --> 00:02:52.867 and the most valuation some sometime around this time. 00:02:52.950 --> 00:02:55.687 And, and indeed, indeed, it has happened. 00:02:55.770 --> 00:02:59.027 We're just keeping in mind all the things we usually keep in 00:02:59.110 --> 00:03:00.887 mind, you know, how do we train good models? 00:03:00.970 --> 00:03:02.417 How do we put them in good products? 00:03:02.500 --> 00:03:04.457 How do we make sure that everything's safe? 00:03:04.540 --> 00:03:07.067 What were you like as a kid growing up in San Francisco? 00:03:07.150 --> 00:03:08.837 I know your, your dad was a leather craftsman. 00:03:08.920 --> 00:03:12.527 Your mom worked in libraries. How did that shape you? 00:03:12.610 --> 00:03:14.987 You know, the whole, you know, like first, you know, 00:03:15.070 --> 00:03:19.240 internet revolution was happening around me and I had absolutely no interest in it. 00:03:20.620 --> 00:03:24.497 I was just interested in like, doing math and like scra, you know, scrawling things. 00:03:24.580 --> 00:03:27.857 I was interested in, like, you know, understanding the universe. 00:03:27.940 --> 00:03:29.860 I was interested in science fiction. 00:03:30.820 --> 00:03:33.580 I think I just felt a lot of curiosity about the world. 00:03:35.590 --> 00:03:37.427 What was it like growing up with Dario? 00:03:37.510 --> 00:03:43.037 He was so smart. He was taking, you know, calculus when he was, you know, 00:03:43.120 --> 00:03:45.137 in I think like middle school. 00:03:45.220 --> 00:03:46.997 He took math classes at UC Berkeley, 00:03:47.080 --> 00:03:52.577 when he was in high school. I was actually more into to reading and arts. 00:03:52.660 --> 00:03:55.460 So we were almost like complete compliments in that way. 00:03:56.260 --> 00:04:00.947 Dario studied neuroscience before turning to AI at Baidu and later Google. 00:04:01.030 --> 00:04:04.427 While Daniela started out as an early employee at Stripe. 00:04:04.510 --> 00:04:06.407 They lived together in San Francisco, along 00:04:06.490 --> 00:04:09.347 with Daniela's husband, Holden Karnofsky. 00:04:09.430 --> 00:04:14.477 Then in 2016, Dario joined the newly formed OpenAI followed by Daniela. 00:04:14.560 --> 00:04:17.717 In 2018, the company began as a nonprofit 00:04:17.800 --> 00:04:21.677 that promised a safe, more open path to super intelligence. 00:04:21.760 --> 00:04:23.837 I think we need to think right now about 00:04:23.920 --> 00:04:26.357 how we want this deployed, how everyone gets 00:04:26.440 --> 00:04:28.907 to benefit from it, how we're gonna govern it, how we're 00:04:28.990 --> 00:04:31.440 gonna make it safe and sort of good for humanity. 00:04:32.020 --> 00:04:35.657 At OpenAI, Dario developed the concept of scaling laws, 00:04:35.740 --> 00:04:39.707 predicting that large language models would improve simply by adding more data 00:04:39.790 --> 00:04:43.937 and computing power, even if the underlying algorithm stayed the same 00:04:44.020 --> 00:04:47.657 At that point in time, I know it sounds crazy now, looking back, not a lot 00:04:47.740 --> 00:04:49.757 of people believed, Hey, scale up is the way 00:04:49.840 --> 00:04:51.767 that these models are gonna get smarter and better. 00:04:51.850 --> 00:04:55.457 That was sort of an unusual counter-cultural scientific 00:04:55.540 --> 00:04:59.897 perspective that I think was really held by our, you know, founding research team. 00:04:59.980 --> 00:05:05.537 That approach helped supercharge OpenAI's models, paving the way for chatGPT, 00:05:05.620 --> 00:05:07.127 but the Amodeis reportedly clashed 00:05:07.210 --> 00:05:10.397 with Altman over the company's direction and values. 00:05:10.480 --> 00:05:12.677 Altman has said, for all the differences I have 00:05:12.760 --> 00:05:15.310 with Anthropic, I mostly trust them as a company. 00:05:16.780 --> 00:05:20.777 Your decision to leave OpenAI has become Silicon Valley lore. 00:05:20.860 --> 00:05:25.037 What did you disagree on? Look, I'll, I'm gonna say it, 00:05:25.120 --> 00:05:26.357 I'm gonna say it very simply. 00:05:26.440 --> 00:05:29.537 There are many valid disagreements to be had on safety. 00:05:29.620 --> 00:05:31.967 We certainly had some of those disagreements with them, 00:05:32.050 --> 00:05:36.527 but you know, people that, that, that, that alone is not sufficient to leave. 00:05:36.610 --> 00:05:39.587 When you feel that you can't trust someone, when you feel 00:05:39.670 --> 00:05:43.007 that their values are not what they say they are, 00:05:43.090 --> 00:05:45.497 when you feel that they're not honest, 00:05:45.580 --> 00:05:49.517 that makes it very hard to, you know, to continue to work 00:05:49.600 --> 00:05:51.767 with a company, to continue to trust the company. 00:05:51.850 --> 00:05:57.207 And look, at the end of the day, why argue with someone when you don't have the same 00:05:57.290 --> 00:05:58.890 vision and you don't trust them? 00:05:59.690 --> 00:06:02.667 The way to resolve it is you go off and do your thing. 00:06:02.750 --> 00:06:04.760 They go off and do their thing. 00:06:08.570 --> 00:06:13.857 So this is Precita Park. When Dario and Daniela decided to leave OpenAI 00:06:13.940 --> 00:06:18.477 to start Anthropic, this is where the early team would get together. 00:06:18.560 --> 00:06:21.357 It was during the pandemic. So the story goes, 00:06:21.440 --> 00:06:24.087 and a bunch of early employees would come here, 00:06:24.170 --> 00:06:27.027 they'd pull up a chair on the grass, they'd have lunch, 00:06:27.110 --> 00:06:29.410 and they'd talk about what they were building. 00:06:30.560 --> 00:06:32.967 When we started this company, there were seven co-founders 00:06:33.050 --> 00:06:34.677 of which me and Daniela are two. 00:06:34.760 --> 00:06:37.047 And now, you know, we we're, we're the, 00:06:37.130 --> 00:06:39.177 we're the basically the only company in the space 00:06:39.260 --> 00:06:41.337 that has all of its co-founders still here. 00:06:41.420 --> 00:06:45.327 You don't get to, you know, be a company of the size and scale that we are. 00:06:45.410 --> 00:06:47.960 With that happening like that almost never happens. 00:06:49.220 --> 00:06:51.087 From the start, Anthropic pitched itself 00:06:51.170 --> 00:06:53.690 as the ultimate safety conscious AI company. 00:06:54.560 --> 00:06:57.537 Dario has published long essays with names like Machines 00:06:57.620 --> 00:06:59.517 of Loving Grace and The Adolescence 00:06:59.600 --> 00:07:03.627 of Technology, musing on the miraculous potential of AI, 00:07:03.710 --> 00:07:05.510 as well as the worst case scenarios. 00:07:06.560 --> 00:07:09.417 Anthropic's chatbot, Claude has been trained to follow a set 00:07:09.500 --> 00:07:12.327 of principles called a constitution, intended 00:07:12.410 --> 00:07:14.517 to keep it on the straight and narrow. 00:07:14.600 --> 00:07:19.077 Claude has a very distinct style and feel a human name. 00:07:19.160 --> 00:07:20.637 What are you trying to convey? 00:07:20.720 --> 00:07:22.977 I think when people interact with Claude versus, you know, 00:07:23.060 --> 00:07:26.067 other systems, there is more of a feeling of, I like 00:07:26.150 --> 00:07:28.257 to describe it as professional warmth. 00:07:28.340 --> 00:07:31.077 So the goal is not for it to be your best friend, 00:07:31.160 --> 00:07:34.557 but it's not for it to be sort of cold, rote, calculating. 00:07:34.640 --> 00:07:37.647 It should feel approachable, but distant, right? Professional. 00:07:37.730 --> 00:07:41.547 Anthropic has talked about teaching Claude to be good. 00:07:41.630 --> 00:07:44.607 What is a good model? What is a bad model? 00:07:44.690 --> 00:07:48.087 You don't want a model that lies accidentally or intentionally, right? 00:07:48.170 --> 00:07:50.547 Lying, we call hallucinations. It makes something up. 00:07:50.630 --> 00:07:52.587 The models are just trained to predict the next word. 00:07:52.670 --> 00:07:55.977 So sometimes they don't know and they just invent something. 00:07:56.060 --> 00:07:58.197 Models sometimes, as we've shown in our research, 00:07:58.280 --> 00:08:00.177 can purposely try to deceive you. 00:08:00.260 --> 00:08:01.887 We have to make sure that doesn't happen in production 00:08:01.970 --> 00:08:03.837 models that we expose to customers. 00:08:03.920 --> 00:08:05.547 And then there's a lot of work around harmlessness, 00:08:05.630 --> 00:08:08.007 just making sure that the model is not accidentally 00:08:08.090 --> 00:08:11.397 producing, you know, information that is wrong or harmful 00:08:11.480 --> 00:08:13.347 or, or could cause someone to do something bad 00:08:13.430 --> 00:08:17.067 Whose values are being put into Claude, like is there a universal good? 00:08:17.150 --> 00:08:19.737 There's so many religions, so many different beliefs, like 00:08:19.820 --> 00:08:23.097 how do you even decide that? So of course there's no universal standard for 00:08:23.180 --> 00:08:25.167 what makes something helpful or harmless, 00:08:25.250 --> 00:08:27.867 but there are founding documents in human history, 00:08:27.950 --> 00:08:30.927 like the UN Declaration of Human Rights that we can use 00:08:31.010 --> 00:08:33.627 to train Claude's character. 00:08:33.710 --> 00:08:35.847 And I think, you know, interestingly on the religious front, 00:08:35.930 --> 00:08:38.247 we've actually started to have a lot of conversations 00:08:38.330 --> 00:08:41.187 with religious leaders around how should we think about, 00:08:41.270 --> 00:08:46.047 you know, Claude, the entity, and how can we bake in some of the kind of core values 00:08:46.130 --> 00:08:50.187 that are consistent across religions types of belief 00:08:50.270 --> 00:08:53.517 that really just transcend, you know, a specific worldview, 00:08:53.600 --> 00:08:55.447 but that are really things that human beings been grappling 00:08:55.530 --> 00:09:00.157 with for, for millennia. Have you tried to give Claude a trait that didn't take, 00:09:00.240 --> 00:09:03.697 or was there any trait that emerged that surprised you? 00:09:03.780 --> 00:09:07.477 I think in the early days, it's funny to look back on them now, the kind 00:09:07.560 --> 00:09:09.727 of early Claudes, Claude two era, 00:09:09.810 --> 00:09:11.947 sometimes Claude would be almost a little bit nannyish. 00:09:12.030 --> 00:09:14.647 Claude was like, I'm really concerned about you, and you're like, Claude, I was 00:09:14.730 --> 00:09:16.117 asking for the weather, right? 00:09:16.200 --> 00:09:17.737 It was like, it was something really benign. 00:09:17.820 --> 00:09:19.657 Thankfully we didn't, you know, release most of, most 00:09:19.740 --> 00:09:21.577 of the most egregious versions of those. 00:09:21.660 --> 00:09:23.767 But I think it is, you know, it is, when you think about it, 00:09:23.850 --> 00:09:27.457 it's like tuning a dial. And so our researchers have this very fine needle 00:09:27.540 --> 00:09:30.150 to thread in order to land at the center of that. 00:09:31.680 --> 00:09:35.917 Whatever Anthropic is doing with Claude, it seems to be working. 00:09:36.000 --> 00:09:38.707 Anthropic's revenue has skyrocketed over the past year, 00:09:38.790 --> 00:09:41.857 making the company profitable for the first time. 00:09:41.940 --> 00:09:46.267 That's largely thanks to the company's focus on more lucrative business tools. 00:09:46.350 --> 00:09:50.857 Claude Code, a major leap that automated large chunks of software engineering 00:09:50.940 --> 00:09:54.240 and Claude Cowork, which gave that power to everyone else. 00:09:55.140 --> 00:09:59.017 Now, early on, others focused on fun splashy consumer apps. 00:09:59.100 --> 00:10:02.077 You made a bet on coding and enterprise. 00:10:02.160 --> 00:10:03.607 Why did you make that bet? 00:10:03.690 --> 00:10:07.687 Was it a, was it a values decision or a business decision? 00:10:07.770 --> 00:10:11.887 Look, if you pick a business model that fundamentally conflicts with your values, 00:10:11.970 --> 00:10:13.837 you're gonna have a hard time, right? 00:10:13.920 --> 00:10:17.947 Either you betray your own values or you become irrelevant. 00:10:18.030 --> 00:10:20.437 And so when we thought about it, we said, look, you know, 00:10:20.520 --> 00:10:23.797 we've seen the world of social media, the consumer world, 00:10:23.880 --> 00:10:27.637 it, it really seems to, you know, encourage engagement, 00:10:27.720 --> 00:10:30.637 even addiction, you know, the slop we've seen 00:10:30.720 --> 00:10:32.947 with AI video models, it's like, what's going on? 00:10:33.030 --> 00:10:35.587 Is it wants to maximize the number of minutes that you're, 00:10:35.670 --> 00:10:39.727 you're paying attention to because that's the advertising revenue-driven incentive. 00:10:39.810 --> 00:10:43.357 Whereas if we look at enterprise, look, I mean, you know, 00:10:43.440 --> 00:10:46.027 we want to make these models useful to people. 00:10:46.110 --> 00:10:50.257 We want to use AI to, you know, cure diseases 00:10:50.340 --> 00:10:51.637 that we couldn't cure before, right? 00:10:51.720 --> 00:10:54.157 Well, that's working with biotech, it's working with pharma, 00:10:54.240 --> 00:10:56.467 it's working with academic research groups. 00:10:56.550 --> 00:10:58.567 All of those are enterprises, right? 00:10:58.650 --> 00:11:03.247 We want to use AI to like, you know, to make energy cheaper and more efficient. 00:11:03.330 --> 00:11:04.957 That's, that's all enterprise. 00:11:05.040 --> 00:11:08.017 And so I think it served us well to have this business model 00:11:08.100 --> 00:11:10.537 that largely aligns with our values. 00:11:10.620 --> 00:11:12.577 Soon after Claude Cowork was released, 00:11:12.660 --> 00:11:16.897 $285 billion in market value vanished overnight. 00:11:16.980 --> 00:11:18.907 Traders called it the SaaSpocalypse. 00:11:18.990 --> 00:11:23.377 This kind of white collar wipeout story in the software sector. Terrifying. 00:11:23.460 --> 00:11:27.127 Some of those are down for nine days in a row, so clearly the tension is building. 00:11:27.210 --> 00:11:30.517 If AI continues improving at this pace, how much 00:11:30.600 --> 00:11:34.417 of traditional software gets replaced and how fast. 00:11:34.500 --> 00:11:38.107 I think with AI like the pie is getting bigger, right? 00:11:38.190 --> 00:11:40.987 So the existing incumbents may be smaller 00:11:41.070 --> 00:11:44.587 and relative terms, some of them may, may go down in value, 00:11:44.670 --> 00:11:47.587 some of them may even may even go out of business if they don't, 00:11:47.670 --> 00:11:49.237 if they don't adapt in the right way. 00:11:49.320 --> 00:11:53.357 But like, I would guess that the software industry gets larger, not 00:11:53.440 --> 00:11:56.627 although there will be some big losers, those who don't kind 00:11:56.710 --> 00:11:58.727 of see what's coming, who don't identify the moats they 00:11:58.810 --> 00:12:00.947 have, they're gonna have a really hard time. 00:12:01.030 --> 00:12:02.327 It's sort of the quiet cafe. 00:12:02.410 --> 00:12:06.677 Okay. Anthropic's recent growth spurt might not have happened without this man. 00:12:06.760 --> 00:12:09.497 Boris Cherny, the engineer behind Claude Code 00:12:09.580 --> 00:12:13.547 and Claude Cowork. When Anthropic hired him in 2024, 00:12:13.630 --> 00:12:17.147 Cherny was living a very different life in rural Japan. 00:12:17.230 --> 00:12:19.367 There was a lot of farmer's markets, like very slow. 00:12:19.450 --> 00:12:21.497 We were making miso. That was sort of the big hobby. 00:12:21.580 --> 00:12:24.047 And I remember using the first AI chatbot 00:12:24.130 --> 00:12:27.587 that I'd ever used, and it just took my breath away 00:12:27.670 --> 00:12:31.247 and I was just like, oh my God, I have to, I have to just be a part of this. 00:12:31.330 --> 00:12:33.257 And I, I'm also just such a big scifi reader 00:12:33.340 --> 00:12:35.387 and so I just know how bad this thing can go. 00:12:35.470 --> 00:12:38.897 Like this technology is incredibly powerful and so we move, move back. 00:12:38.980 --> 00:12:44.357 You created Claude Code, you led the development of Claude Cowork, 00:12:44.440 --> 00:12:47.607 what problem were you trying to solve? 00:12:47.690 --> 00:12:49.877 If you look at the coding products, they were all pretty simple. 00:12:49.960 --> 00:12:51.377 It was like, it was sort of like, you know, 00:12:51.460 --> 00:12:53.837 like complete the word, complete the sentence. 00:12:53.920 --> 00:12:59.357 That was the extent of AI in coding and we just wanted to make a way bigger bet. 00:12:59.440 --> 00:13:03.107 And our bet was, we think actually a coding agent can do all of it. 00:13:03.190 --> 00:13:07.187 A year and a half ago you wrote the code by hand and sometimes you press tab 00:13:07.270 --> 00:13:08.867 and it would auto complete a line. 00:13:08.950 --> 00:13:13.427 Now I talk to my Claude and it writes the code and then while it does that, I talk 00:13:13.510 --> 00:13:15.197 to the next Claude and it writes some code. 00:13:15.280 --> 00:13:18.467 And at any point I have either, you know, like a few Claudes running and up 00:13:18.550 --> 00:13:21.347 to a few thousand Claudes running, doing things. 00:13:21.430 --> 00:13:25.597 How much code is Claude writing internally for engineers here? 00:13:25.680 --> 00:13:27.730 It's writing almost all of it on my team. 00:13:28.930 --> 00:13:31.697 So for me personally, it's been writing a hundred percent 00:13:31.780 --> 00:13:33.167 of my code for at least six months. 00:13:33.250 --> 00:13:36.107 The work of engineering has just completely changed. 00:13:36.190 --> 00:13:38.117 I feel like I suddenly have superpowers. 00:13:38.200 --> 00:13:41.447 I have like a jet pack and the engineering has never been this fun. 00:13:41.530 --> 00:13:43.997 Can we give Claude a spin? Yeah, let's do this. 00:13:44.080 --> 00:13:46.787 All right. Alright, so let's, let's make like a little recipe app. 00:13:46.870 --> 00:13:49.690 So make a recipe app. What do you want it to do? 00:13:51.130 --> 00:13:56.127 I would love it to suggest meals for the week. 00:13:56.210 --> 00:13:58.607 So, you know, this is gonna take maybe a few minutes 00:13:58.690 --> 00:14:01.547 and let's, let's just see if Claude builds this. 00:14:01.630 --> 00:14:07.727 Oh, it gave me some options. Hmm. Let's do These look pretty tasty actually. Yeah, 00:14:07.810 --> 00:14:09.797 I could do a Greek chicken power bowl. 00:14:09.880 --> 00:14:13.637 When I click on a recipe, I want it 00:14:13.720 --> 00:14:18.720 to show me how to do it. Thank you. 00:14:19.330 --> 00:14:22.590 Aw, I'm sure Claude appreciates that. 00:14:24.460 --> 00:14:25.637 I always try to say thank you. 00:14:25.720 --> 00:14:27.107 Yeah, I always try to be nice too. I 00:14:27.190 --> 00:14:30.317 don't know. A recipe app might not be the most mind 00:14:30.400 --> 00:14:32.717 blowing demo of Anthropic's technology, 00:14:32.800 --> 00:14:35.807 but what Claude Code just did in minutes, used to take hours 00:14:35.890 --> 00:14:39.827 or days for some. That looks like a big opportunity. 00:14:39.910 --> 00:14:41.927 Are you like nervous about next week or is it like.. 00:14:42.010 --> 00:14:43.937 What's next week, Your developer conference? 00:14:44.020 --> 00:14:48.580 Oh! Yes, yes. There's so many things happening every week. 00:14:52.190 --> 00:14:55.907 Year over year. API volume is up nearly 17 x on the cloud. 00:14:55.990 --> 00:15:01.827 In the last 12 months, we shipped eight frontier models to developers and users. 00:15:01.910 --> 00:15:04.887 Welcome to the second annual Code with Claude conference, 00:15:04.970 --> 00:15:08.337 where Claude superfans come, hoping to get a glimpse of the future. 00:15:08.420 --> 00:15:11.097 We're having a lot of fun. There's a ton of adrenaline. 00:15:11.180 --> 00:15:14.817 This is the first year we've grown faster than the exponential. 00:15:14.900 --> 00:15:18.507 In the first quarter of this year, we saw, if you were 00:15:18.590 --> 00:15:22.047 to annualize it, 80 x growth per year. 00:15:22.130 --> 00:15:25.647 Ooh. I 00:15:25.730 --> 00:15:30.927 use Claude every day in the form of Claude Code to do all sorts of stuff. Not 00:15:31.010 --> 00:15:34.467 just write code. I write way more code than I did five years ago. 00:15:34.550 --> 00:15:40.467 I've got the confidence of, I don't know, a 22-year-old with VC money. I'm not 22. 00:15:40.550 --> 00:15:43.377 My friend invited me and it's been, it's amazing. 00:15:43.460 --> 00:15:45.087 As someone that's not really technical, 00:15:45.170 --> 00:15:47.187 which I think it's really interesting just to see 00:15:47.270 --> 00:15:49.137 what everyone's working on, what's out there. 00:15:49.220 --> 00:15:51.267 I think the, the concerning part or the, 00:15:51.350 --> 00:15:54.987 and the exciting part is hey, they can do things that are, 00:15:55.070 --> 00:15:58.857 we couldn't dream of and timelines we couldn't ever expect. 00:15:58.940 --> 00:16:03.447 But the kind of alternate side of that reality is you really have to be prepared 00:16:03.530 --> 00:16:06.580 because they will really be a thousand times more productive. 00:16:08.090 --> 00:16:10.587 All of this raises an obvious question. 00:16:10.670 --> 00:16:14.397 Will engineers be the first casualties of the AI they're building? 00:16:14.480 --> 00:16:17.750 It's been revenge of the nerds for a while? Is that over? 00:16:19.700 --> 00:16:23.000 I think we all become nerds, But what happens to the actual nerds? 00:16:24.170 --> 00:16:28.827 I think the actual nerds will, they're they have to figure it out. 00:16:28.910 --> 00:16:32.187 Yeah. I I think for a lot of them, the skill that they had 00:16:32.270 --> 00:16:34.647 before is gonna help them in the future 00:16:34.730 --> 00:16:37.377 because they sort of have a pretty big head start. 00:16:37.460 --> 00:16:39.177 They do a lot of things that are not coding 00:16:39.260 --> 00:16:40.917 and general also have to talk to users. 00:16:41.000 --> 00:16:44.097 They have to plan, they have to think about what's next. 00:16:44.180 --> 00:16:48.130 So I, I think these are still parts of engineering that are gonna stick around. 00:16:48.260 --> 00:16:53.577 Silicon Valley may have AI fever, but elsewhere the mood is less upbeat. 00:16:53.660 --> 00:16:56.877 70% of Americans think AI will kill jobs 00:16:56.960 --> 00:16:59.610 and nearly a third worry, theirs will be one of them. 00:17:00.770 --> 00:17:03.567 Dario Amodei has been outspoken about this issue 00:17:03.650 --> 00:17:06.897 and some of his predictions don't exactly sound reassuring. 00:17:06.980 --> 00:17:10.077 I think we could have this very unusual combination 00:17:10.160 --> 00:17:14.397 of very fast GDP growth and high unemployment, 00:17:14.480 --> 00:17:19.197 or at least underemployment or, you know, low wage job. 00:17:19.280 --> 00:17:21.597 A lot of low wage jobs, high inequality. 00:17:21.680 --> 00:17:24.387 You've been really direct about job loss. 00:17:24.470 --> 00:17:27.387 AI could eliminate half of all entry level white collar jobs in 00:17:27.470 --> 00:17:32.487 the next one to five years. That was a year ago. AI has moved incredibly fast. 00:17:32.570 --> 00:17:35.127 Is it still 50% or is it higher? 00:17:35.210 --> 00:17:38.607 I don't know exactly, but I'm, I'm still, I'm still pretty concerned. 00:17:38.690 --> 00:17:40.617 I'm still the same order of concerns. 00:17:40.700 --> 00:17:45.207 You know, we are seeing right now that AI is making people more productive. 00:17:45.290 --> 00:17:48.507 But that's the usual hump. You automate 90% of the job, 00:17:48.590 --> 00:17:51.247 great people are 10 times more productive in the other 10% 00:17:51.330 --> 00:17:52.987 'cause they're 10 times more leveraged. 00:17:53.070 --> 00:17:56.047 But eventually it gets close to a hundred percent. 00:17:56.130 --> 00:17:58.147 Now the sequel to that is, well then you have 00:17:58.230 --> 00:17:59.797 to find something else for them to do. 00:17:59.880 --> 00:18:02.167 Right now AI makes the software engineers more productive, 00:18:02.250 --> 00:18:05.467 even though AI writes all the code or almost all the code. 00:18:05.550 --> 00:18:08.647 But we're already starting to see the beginning of like, 00:18:08.730 --> 00:18:10.597 you know, there may be some people that it's, 00:18:10.680 --> 00:18:13.237 it's not making more productive, that it's better for the AI 00:18:13.320 --> 00:18:14.977 to just, to just do the thing. 00:18:15.060 --> 00:18:16.410 How does that sit with you? 00:18:17.610 --> 00:18:20.250 It's, it's very uncomfortable. Yeah. 00:18:21.090 --> 00:18:23.977 I I think this is, this is the reason why I chose to go 00:18:24.060 --> 00:18:26.647 to Anthropic and I think this is the reason that a lot 00:18:26.730 --> 00:18:30.457 of people here chose to go here is artificial intelligence 00:18:30.540 --> 00:18:33.397 is this force that is far bigger than we are. 00:18:33.480 --> 00:18:36.667 But here we can hopefully make it go a little bit better. 00:18:36.750 --> 00:18:39.757 Do you feel like it's your job to do something about that? 00:18:39.840 --> 00:18:42.157 Or is that someone else's problem? 00:18:42.240 --> 00:18:43.687 I think it's a thing that we have to talk about. 00:18:43.770 --> 00:18:47.287 We have to advocate for it. Ultimately, it's up to society to solve it. 00:18:47.370 --> 00:18:48.997 This is bigger than one company. 00:18:49.080 --> 00:18:51.907 There has been a lot of pushback on, you know, your, 00:18:51.990 --> 00:18:54.667 and I know you've said you're trying to warn people, 00:18:54.750 --> 00:18:56.707 but that, you know, you're, you know, Jensen Huang said, 00:18:56.790 --> 00:18:58.627 you're conflating tasks with jobs. 00:18:58.710 --> 00:19:03.037 AI is creating jobs. Anybody who is saying that AI is wiping out jobs 00:19:03.120 --> 00:19:06.337 is scaring people. Other folks have said this, you know, it's sort 00:19:06.420 --> 00:19:08.867 of doom marketing that benefits anthropic. 00:19:08.950 --> 00:19:13.117 So, so I wanna be really clear and push back hard against this. 00:19:13.200 --> 00:19:16.597 In every interview I talk about the, 00:19:16.680 --> 00:19:21.787 the possible ways to address these risks from tax and macroeconomic policy to 00:19:21.870 --> 00:19:25.117 what the new jobs are in The Adolescence of Technology. 00:19:25.200 --> 00:19:29.707 I have like five pages where I lay out the difference between tasks and jobs. 00:19:29.790 --> 00:19:34.807 Why this time is different than other times, but social media, which I detest, 00:19:34.890 --> 00:19:40.447 which I detest as a category, people have these three second clips from, you know, 00:19:40.530 --> 00:19:45.697 from a year ago. I've, I've written much more carefully about these things 00:19:45.780 --> 00:19:47.107 where, where I talk about the risks. 00:19:47.190 --> 00:19:49.357 So these are this, this, the idea 00:19:49.440 --> 00:19:52.327 that this is cheap marketing is itself cheap marketing. 00:19:52.410 --> 00:19:55.327 I think it's, it's part of the disease of Silicon Valley. 00:19:55.410 --> 00:20:00.517 It's been caught up in this social media world of, of, of three seconds. 00:20:00.600 --> 00:20:03.037 And so my message is just definitely not doom is coming. 00:20:03.120 --> 00:20:05.827 My message is like, this is something, you know, 00:20:05.910 --> 00:20:07.807 that we should see coming, that we're worried about 00:20:07.890 --> 00:20:10.590 and that we need to actually respond to positively 00:20:13.770 --> 00:20:15.217 Beyond the software industry, 00:20:15.300 --> 00:20:19.927 the potential impact of AI on jobs seems harder to predict. 00:20:20.010 --> 00:20:22.627 Anthropic has published a paper estimating which fields 00:20:22.710 --> 00:20:25.160 could make the most use of AI in the near future. 00:20:25.980 --> 00:20:28.897 If its predictions are right, management, finance, 00:20:28.980 --> 00:20:31.280 and legal jobs could soon look very different. 00:20:32.460 --> 00:20:36.847 Which jobs go away, who gets replaced and what new jobs are created. 00:20:36.930 --> 00:20:38.430 So no one knows for sure 00:20:39.420 --> 00:20:41.587 because you know, the economy's unpredictable, right? 00:20:41.670 --> 00:20:46.447 The thing that that we have going for us here is the pie is gonna expand a lot. 00:20:46.530 --> 00:20:49.307 And so because the pie is gonna expand a lot, 00:20:49.390 --> 00:20:51.947 there are probably going to be places where people can go. 00:20:52.030 --> 00:20:54.287 It's just a matter of finding them fast enough. 00:20:54.370 --> 00:20:56.087 So play this out for me a little bit. 00:20:56.170 --> 00:21:02.087 You know, you wake up in five years, you know, what does this country look like? 00:21:02.170 --> 00:21:03.797 What are those people doing? 00:21:03.880 --> 00:21:05.957 Yeah, so 'cause if there's that much unemployment, is 00:21:06.040 --> 00:21:08.657 that not how revolutions start? 00:21:08.740 --> 00:21:10.607 Yeah, no, this is the outcome we wanna prevent. 00:21:10.690 --> 00:21:12.707 This is absolutely the outcome we want to prevent. 00:21:12.790 --> 00:21:15.527 I think there's a few places, none of them are guaranteed. 00:21:15.610 --> 00:21:18.377 We're not sure, but there's the physical world, 00:21:18.460 --> 00:21:21.527 we need a lot of more people to make, build, 00:21:21.610 --> 00:21:24.017 manufacture things in the physical world. 00:21:24.100 --> 00:21:27.857 Anything that's human centered, I think that's gonna be a big deal, right? 00:21:27.940 --> 00:21:30.617 People, or at least some people wanna talk to humans. 00:21:30.700 --> 00:21:33.707 So these kind of human relationship driven jobs, like, 00:21:33.790 --> 00:21:35.267 I think those are gonna be important 00:21:35.350 --> 00:21:39.077 and I think there'll be some effort by the humans to kind 00:21:39.160 --> 00:21:42.347 of direct AIs right at, at some level it has to be in line 00:21:42.430 --> 00:21:45.407 with someone's values and someone's intentions and, 00:21:45.490 --> 00:21:48.407 and so I, I think there's gonna be some role there, 00:21:48.490 --> 00:21:51.407 although I don't know how thin versus how thick it will be. 00:21:51.490 --> 00:21:53.177 I think I feel a little bit more hopeful 00:21:53.260 --> 00:21:56.507 that humans will continue to find ways to leverage AI 00:21:56.590 --> 00:21:59.627 to be productive, to do the parts of the work 00:21:59.710 --> 00:22:01.907 that are meaningful to us that only humans can do. 00:22:01.990 --> 00:22:05.687 I think the human to human interaction will never fully go away. 00:22:05.770 --> 00:22:10.187 The example I often reach for is, is what will happen in medicine today. 00:22:10.270 --> 00:22:13.720 We hire doctors who are expert diagnosticians. 00:22:14.620 --> 00:22:17.357 I think AI is gonna soon be pretty good at telling you 00:22:17.440 --> 00:22:19.697 what the suite of options of things that are wrong with you and what test to run. 00:22:19.780 --> 00:22:21.827 And you won't need a doctor to do that. 00:22:21.910 --> 00:22:24.437 But an AI can't physically examine you 00:22:24.520 --> 00:22:27.317 and say, Hey, does it hurt when I press, you know, here. 00:22:27.400 --> 00:22:30.407 They can't have a bedside manner with you that says like, 00:22:30.490 --> 00:22:31.607 tell me how you're feeling about this. 00:22:31.690 --> 00:22:34.377 How are you, how are you coping with going through this process? 00:22:34.460 --> 00:22:36.467 And I think we're going to pivot something like medicine 00:22:36.550 --> 00:22:38.597 to be much more focused on the interpersonal 00:22:38.680 --> 00:22:41.987 because the diagnostic tools are going to become much better. 00:22:42.070 --> 00:22:45.860 But the interpersonal human part, that's, that's not gonna change. 00:22:49.360 --> 00:22:52.187 Explain this to me. Daniela runs day-to-day operations. 00:22:52.270 --> 00:22:54.647 All the leadership team reports to you. Yes. 00:22:54.730 --> 00:22:57.407 No one reports to you. That sounds like a pretty sweet job. 00:22:57.490 --> 00:23:00.737 It's incredibly freeing. It lets me do all the things 00:23:00.820 --> 00:23:05.267 that I do much more easily than I would otherwise 00:23:05.350 --> 00:23:06.527 Because it's and she does all the work. 00:23:06.610 --> 00:23:08.797 Is that what you're saying? It's, it's, it's a, I'm just, 00:23:08.880 --> 00:23:13.727 I if if you, if you had to go through the things I had 00:23:13.810 --> 00:23:18.340 to go through during Department of War or no, 00:23:19.420 --> 00:23:22.780 The Amodeis hope AI can have a utopian influence on society, 00:23:23.710 --> 00:23:28.337 but there's no denying the technology's destructive and dystopian potential. 00:23:28.420 --> 00:23:31.697 Anthropic is at the center of a technological arms race 00:23:31.780 --> 00:23:33.980 with powerful governments vying for control. 00:23:34.990 --> 00:23:37.727 Dario has charged head on into this arena. 00:23:37.810 --> 00:23:39.827 He's not afraid to blast his competitors. 00:23:39.910 --> 00:23:42.587 And then I think there are some players who, you know, 00:23:42.670 --> 00:23:46.437 who, who are yoloing who, who pull, pull the risk dial too far. 00:23:46.520 --> 00:23:48.207 And I'm very concerned. Who is yoloing? 00:23:48.290 --> 00:23:53.097 So I, that's a question I'm not gonna answer Or critique the US government 00:23:53.180 --> 00:23:56.847 and Anthropic's own partners for selling AI chips to China. 00:23:56.930 --> 00:23:59.937 It's a bit like, you know, I I I don't know like selling, 00:24:00.020 --> 00:24:03.507 selling, you know, nuclear weapons to, to North Korea. 00:24:03.590 --> 00:24:08.877 I've been very outspoken about the need for export controls on chips to China. 00:24:08.960 --> 00:24:11.187 I say this because I think it would be really bad 00:24:11.270 --> 00:24:13.647 for America, for, you know, the state 00:24:13.730 --> 00:24:17.967 of democracy in the world for, you know, China to be ahead in AI capabilities. 00:24:18.050 --> 00:24:19.917 And you know, it's, it's like some 00:24:20.000 --> 00:24:23.367 of the chipmakers obviously don't agree with that view, 00:24:23.450 --> 00:24:25.137 but it hasn't stopped me from saying it. 00:24:25.220 --> 00:24:27.567 Even after we've signed more partnerships, 00:24:27.650 --> 00:24:30.117 I'm sure they wish we didn't say these things, 00:24:30.200 --> 00:24:32.277 but these things are what are what I believe. 00:24:32.360 --> 00:24:34.257 And so, you know, look, we're all adults here. 00:24:34.340 --> 00:24:37.940 We can work together on one thing while disagreeing about another thing. 00:24:39.050 --> 00:24:41.037 Amodei's statements about geopolitics date back 00:24:41.120 --> 00:24:44.157 to his days at Caltech. As a sophomore physics student, 00:24:44.240 --> 00:24:47.127 he developed a reputation as an anti-war advocate 00:24:47.210 --> 00:24:50.337 who believed scientists shouldn't sit in ivory towers. 00:24:50.420 --> 00:24:55.437 It's a worldview that would later define Anthropic's position on the future of war. 00:24:55.520 --> 00:24:57.717 You've had a longstanding anti-war stance 00:24:57.800 --> 00:25:00.567 and yet you were one of the first AI companies 00:25:00.650 --> 00:25:04.707 to sign a contract with the Department of Defense to operate on classified networks. 00:25:04.790 --> 00:25:09.177 These are the, are the networks that the US uses to fight wars. Explain that. 00:25:09.260 --> 00:25:10.887 Look, I mean, the world changes. 00:25:10.970 --> 00:25:13.797 Like, you know, my view of this technology, you know, 00:25:13.880 --> 00:25:17.667 when I see Russia invading Ukraine, when I see the risk 00:25:17.750 --> 00:25:22.017 of China invading Taiwan, you know, it worries me 00:25:22.100 --> 00:25:25.370 that we have a kind of resurgent authoritarian block 00:25:26.270 --> 00:25:29.967 that they're very aggressive and that we need to defend ourselves. 00:25:30.050 --> 00:25:34.107 You know, I may not agree with every policy of I of either administration, 00:25:34.190 --> 00:25:37.137 but you know, that's why we've generally been supportive of this. 00:25:37.220 --> 00:25:39.387 You've been working with Palantir since 2024. 00:25:39.470 --> 00:25:41.127 That's right. You know, their technology is used 00:25:41.210 --> 00:25:43.557 by ICE, police departments, in Gaza. 00:25:43.640 --> 00:25:46.287 Is Claude being used for surveillance in other ways? 00:25:46.370 --> 00:25:50.727 We, we don't work with ICE either through Palantir or anyone else. 00:25:50.810 --> 00:25:54.477 We don't work with CBP. I don't believe we work in Gaza. 00:25:54.560 --> 00:25:57.777 We're very careful about, you know, scoping our engagements 00:25:57.860 --> 00:25:59.310 to things that we believe in. 00:26:00.080 --> 00:26:03.927 In 2025, Anthropic along with OpenAI xAI 00:26:04.010 --> 00:26:07.647 and Google won a $200 million contract with the Pentagon. 00:26:07.730 --> 00:26:09.567 Anthropic framed it as an opportunity 00:26:09.650 --> 00:26:12.100 to become a leading AI vendor for the government. 00:26:14.210 --> 00:26:17.577 Claude was reportedly used by the US military in the operation 00:26:17.660 --> 00:26:20.547 to seize Venezuelan president Nicolas Maduro. 00:26:20.630 --> 00:26:23.277 Weeks later, everything started to unravel. 00:26:23.360 --> 00:26:26.187 We bring you breaking news from Anthropic, the Department 00:26:26.270 --> 00:26:28.437 of Defense demand that Anthropic allow full use 00:26:28.520 --> 00:26:31.167 of its AI technology without guardrails. 00:26:31.250 --> 00:26:35.697 Anthropic drew red lines refusing to let Claude be used for mass surveillance 00:26:35.780 --> 00:26:39.867 and autonomous weapons putting the company on a collision course with the Pentagon. 00:26:39.950 --> 00:26:44.047 The tech giant facing a deadline today to accept conditions or be blacklisted. 00:26:44.130 --> 00:26:47.137 Does a commercial entity have the right to determine 00:26:47.220 --> 00:26:51.247 how their products are used by the military? 00:26:51.330 --> 00:26:55.297 Lawsuits were filed and Anthropic was banned from the Pentagon, underscored 00:26:55.380 --> 00:26:59.677 by strong words from President Trump and defense secretary Pete Hegseth 00:26:59.760 --> 00:27:02.197 Anthropic is run by an ideological lunatic who 00:27:02.280 --> 00:27:04.477 shouldn't have a, so that's, but that's not my question. 00:27:04.560 --> 00:27:07.687 My question is AI decision making about what we do. 00:27:07.770 --> 00:27:11.947 Do you mind being called an ideological lunatic or a bunch of left wing nut jobs? 00:27:12.030 --> 00:27:14.677 You know, I've been called worse things than that all the time. 00:27:14.760 --> 00:27:17.317 What does winning this fight actually look like? 00:27:17.400 --> 00:27:23.257 I won't even call it a fight. This is more a debate about what the proper use of AI 00:27:23.340 --> 00:27:27.397 by the government is. And AI is an emerging new technology. 00:27:27.480 --> 00:27:31.987 We don't understand the ways in which it's reliable or unreliable. 00:27:32.070 --> 00:27:35.017 We don't understand the ways in which it promotes our values 00:27:35.100 --> 00:27:36.547 or undermines our values. 00:27:36.630 --> 00:27:39.427 And so one of the things that I thought was important was 00:27:39.510 --> 00:27:44.797 to establish a precedent on some of the, some of the use cases we think are good, 00:27:44.880 --> 00:27:46.597 which frankly is most of them. 00:27:46.680 --> 00:27:50.287 And, and some of the use cases that we're concerned about. 00:27:50.370 --> 00:27:52.837 A US official has said, with the help of LLMs, 00:27:52.920 --> 00:27:54.727 the US military has gone from being able 00:27:54.810 --> 00:27:58.957 to hit a thousand targets a day to 5,000 targets a day. 00:27:59.040 --> 00:28:03.097 That means Claude can help kill more people more quickly. 00:28:03.180 --> 00:28:04.597 Are you comfortable with that? 00:28:04.680 --> 00:28:06.757 Basically, you're, you're asking like, you know, 00:28:06.840 --> 00:28:08.047 do you believe in this country, right? 00:28:08.130 --> 00:28:12.037 Do you want this country to be a more powerful actor rather than a less powerful 00:28:12.120 --> 00:28:15.697 actor on the world stage? I do. I'm a patriot. It's not up to me. 00:28:15.780 --> 00:28:18.637 If we provide a technology, it's not up to us to say, 00:28:18.720 --> 00:28:22.417 you can do this military operation and you can't do that military operation. 00:28:22.500 --> 00:28:27.397 Now, I might privately believe that this military operation makes sense 00:28:27.480 --> 00:28:31.717 and that military operation is a bad idea, but we're not gonna deny the technology. 00:28:31.800 --> 00:28:36.157 You, you basically have to, you know, you have to leave policy in the hands 00:28:36.240 --> 00:28:38.827 of the military decision makers. 00:28:38.910 --> 00:28:41.437 Bloomberg has reported that Claude is being used 00:28:41.520 --> 00:28:43.687 by the US military and the war in Iran 00:28:43.770 --> 00:28:47.167 to do AI assisted targeting via a platform made 00:28:47.250 --> 00:28:50.317 by Palantir Maven Smart system. In February, 00:28:50.400 --> 00:28:54.307 US missile reportedly hit a girl school in Iran killing more 00:28:54.390 --> 00:28:56.850 than 150 people, most of them children. 00:28:57.930 --> 00:29:00.037 Did Claude play a role in that strike? 00:29:00.120 --> 00:29:02.977 We don't know exactly how, you know, 00:29:03.060 --> 00:29:06.630 these models were used, you know, obviously like, you know, 00:29:07.980 --> 00:29:09.817 these things that, that, you know, mistakes 00:29:09.900 --> 00:29:12.427 that happen in warfare are really, really terrible. 00:29:12.510 --> 00:29:15.307 Like this is a really terrible thing to happen. 00:29:15.390 --> 00:29:17.467 We were willing to risk the future of our company 00:29:17.550 --> 00:29:21.667 to like limit how, you know, these models are used 00:29:21.750 --> 00:29:24.727 and you know, what, what you're talking about is a use case 00:29:24.810 --> 00:29:26.317 that doesn't even violate our red lines. 00:29:26.400 --> 00:29:29.257 We're worried that there will be a hundred times as much, 00:29:29.340 --> 00:29:33.217 you know, with, with use cases that, that, that do violate our red lines. 00:29:33.300 --> 00:29:35.347 Now I, you know, I I I, you know, again, again, 00:29:35.430 --> 00:29:39.337 I would say I think overall the use of these, the, the use 00:29:39.420 --> 00:29:41.237 of these models is appropriate. 00:29:41.320 --> 00:29:43.577 I think it's good on net, you know, 00:29:43.660 --> 00:29:47.957 but military decision makers make terrible mistakes even 00:29:48.040 --> 00:29:49.787 when, even, even at the best of times, 00:29:49.870 --> 00:29:51.827 and I, I don't know if we're in the best of times. 00:29:51.910 --> 00:29:56.327 What we've seen here is Claude assists, but a human makes the final call. 00:29:56.410 --> 00:29:59.117 So a human made that final call, not Claude. 00:29:59.200 --> 00:30:02.447 Imagine if you had a world in which, not Claude, 00:30:02.530 --> 00:30:05.207 because we haven't allowed it, but someone else's AI model, 00:30:05.290 --> 00:30:07.937 the AI model just makes the decision and the human never sees it. 00:30:08.020 --> 00:30:09.437 That's what we were standing up for. 00:30:09.520 --> 00:30:11.087 That's what we were fighting against. 00:30:11.170 --> 00:30:14.477 This school had a website. You could have found it in a Google Search, 00:30:14.560 --> 00:30:16.517 like shouldn't Claude have spotted that? 00:30:16.600 --> 00:30:20.117 And it doesn't speak to a scarier issue about using 00:30:20.200 --> 00:30:22.337 technology as a shortcut in war. 00:30:22.420 --> 00:30:26.537 The principle that was obeyed here is a human makes the 00:30:26.620 --> 00:30:28.337 human makes the final decision. 00:30:28.420 --> 00:30:31.517 I don't know what role Claude or any other AI had, 00:30:31.600 --> 00:30:34.637 but like, if, if this isn't an illustration why 00:30:34.720 --> 00:30:40.817 that principle is so important, I don't know what is, Is AI warfare more likely 00:30:40.900 --> 00:30:45.587 to stop World War III, a war between the US and China? 00:30:45.670 --> 00:30:48.107 Or is it more likely to make it happen? 00:30:48.190 --> 00:30:50.567 I would say on balance it is more likely to stop it. 00:30:50.650 --> 00:30:55.097 But if we have no limits on how it's used, then I think, 00:30:55.180 --> 00:30:56.897 you know, it could be more likely to cause it. 00:30:56.980 --> 00:30:58.547 You've seen Dr. StrangeLove, right? 00:30:58.630 --> 00:31:01.697 The premise of it was like, you know, you have a doomsday device 00:31:01.780 --> 00:31:04.007 that automatically fires nuclear weapons when it thinks 00:31:04.090 --> 00:31:05.507 nuclear weapons are being fired at it. 00:31:05.590 --> 00:31:07.067 What could go wrong? Right? 00:31:07.150 --> 00:31:10.367 I think the way conflicts happen is that, you know, the, 00:31:10.450 --> 00:31:11.747 the two sides jump at each other. 00:31:11.830 --> 00:31:13.337 They misunderstand each other. 00:31:13.420 --> 00:31:16.037 And when we don't have proper oversight of this technology, 00:31:16.120 --> 00:31:19.337 I think those kinds of accidents are more likely to happen. 00:31:19.420 --> 00:31:22.667 Now, I think if AI is used in an appropriate way, in, in, 00:31:22.750 --> 00:31:28.637 in, not even warfare, but think of just, just intelligence collection, you know, 00:31:28.720 --> 00:31:31.727 let, let's say we're able to, you know, predict an invasion 00:31:31.810 --> 00:31:34.247 of Taiwan or a new movement in Ukraine. 00:31:34.330 --> 00:31:38.777 Like, you know, our adversaries will think twice about, 00:31:38.860 --> 00:31:42.827 you know, about conducting some kind of invasion or military operation if we know 00:31:42.910 --> 00:31:44.410 everything that they're doing. 00:31:46.510 --> 00:31:49.277 AI's role on the battlefield raises difficult questions, 00:31:49.360 --> 00:31:53.807 even for a company willing to debate the ethical dilemmas of its own technology. 00:31:53.890 --> 00:31:57.160 It's a conversation Amodei seems eager to have in public. 00:31:58.090 --> 00:32:00.107 How do you handle the pressure? 00:32:00.190 --> 00:32:04.757 I try very hard to communicate and always be straightforward and honest. 00:32:04.840 --> 00:32:08.057 I get up in front of the company every two weeks 00:32:08.140 --> 00:32:11.177 and just talk for an hour about just what's on my mind, 00:32:11.260 --> 00:32:14.147 what's going on in the industry, what's going on in the outside world. 00:32:14.230 --> 00:32:19.037 It's totally uncensored. And not only does it build trust, it's very freeing for me 00:32:19.120 --> 00:32:23.777 where I feel that I have 3,000 people who are on the same page as me. 00:32:23.860 --> 00:32:28.367 That is an incredible amplifier and is one of the most useful things in 00:32:28.450 --> 00:32:29.897 handing, handling the pressure. 00:32:29.980 --> 00:32:33.377 Then when we have to confront an external challenge, 00:32:33.460 --> 00:32:37.157 stand up to, you know, some very difficult situations 00:32:37.240 --> 00:32:39.567 that we've seen in the last few months, that is 00:32:39.650 --> 00:32:42.867 what allows us to have a consistent and coherent position. 00:32:42.950 --> 00:32:45.597 And, and I think, I think it's an incredible advantage 00:32:45.680 --> 00:32:47.480 because I never feel like I'm alone. 00:32:54.020 --> 00:32:56.427 Lurking in the background at Anthropic's headquarters was 00:32:56.510 --> 00:32:59.660 the surprise development of a new AI model called Mythos. 00:33:01.160 --> 00:33:03.470 A model so powerful, it spooked everyone. 00:33:04.550 --> 00:33:06.567 Anthropic is making headlines almost on a weekly basis. 00:33:06.650 --> 00:33:08.367 Yes, Most notably now around mythos. 00:33:08.450 --> 00:33:11.547 The company believes it's this enormous threat. 00:33:11.630 --> 00:33:15.387 Imagine a world where everyone had a nuclear bazooka basically 00:33:15.470 --> 00:33:18.777 Mythos identified thousands of cybersecurity vulnerabilities, 00:33:18.860 --> 00:33:22.887 exposing potential flaws in every major operating system. 00:33:22.970 --> 00:33:26.127 Anthropic signaled that if fully released mythos could hack 00:33:26.210 --> 00:33:30.260 banks, pry open state secrets and cripple critical infrastructure. 00:33:31.310 --> 00:33:35.637 I think the thing that surprised me most about it was the 00:33:35.720 --> 00:33:39.477 models had been climbing in their ability to find vulnerabilities. 00:33:39.560 --> 00:33:41.247 It was a particularly large jump. 00:33:41.330 --> 00:33:43.557 Some of the early companies that we gave this 00:33:43.640 --> 00:33:46.947 to said things like, this is a super weapon, you should have 00:33:47.030 --> 00:33:49.227 to own a gun license to use it. 00:33:49.310 --> 00:33:51.230 Please don't release this. In 00:33:52.370 --> 00:33:54.147 an initiative called Project Glasswing, 00:33:54.230 --> 00:33:57.560 Anthropic gave select organizations access to Mythos. 00:33:58.490 --> 00:34:02.007 Even federal agencies like the National Security Administration clamored 00:34:02.090 --> 00:34:05.180 to use it despite Anthropic's blacklisting by the Pentagon. 00:34:06.050 --> 00:34:11.127 I think the future is this kind of cat and mouse game where we need to make sure 00:34:11.210 --> 00:34:15.147 that the good guys have the tools that they need to defend, 00:34:15.230 --> 00:34:17.937 and then at some point the, you know, the bad guys will have it too. 00:34:18.020 --> 00:34:19.437 And, and at that point we have 00:34:19.520 --> 00:34:21.687 to make sure the good guys have an even better model 00:34:21.770 --> 00:34:23.187 so they can be ready for this. 00:34:23.270 --> 00:34:26.697 Is it possible to stay ahead of the bad guys really though? 00:34:26.780 --> 00:34:29.367 That's what we hope the criticism is. 00:34:29.450 --> 00:34:32.487 You're effectively deciding who gets access and who doesn't. 00:34:32.570 --> 00:34:36.777 Why should anyone be comfortable with that kind of concentration of power? 00:34:36.860 --> 00:34:40.497 It wasn't like, oh, it's so powerful and let's decide who gets the power. 00:34:40.580 --> 00:34:44.367 It was a very specific concern around cybersecurity 00:34:44.450 --> 00:34:47.397 and so the way that we decided who to give the model 00:34:47.480 --> 00:34:49.677 to was grounded in that specific fear. 00:34:49.760 --> 00:34:53.007 There's obviously nuance to decide, like where do you draw that circle? 00:34:53.090 --> 00:34:54.417 I think that's really complicated. 00:34:54.500 --> 00:34:58.137 I think we've tried to be as publicly open as possible 00:34:58.220 --> 00:35:00.687 to say we're trying our best to make this decision well, 00:35:00.770 --> 00:35:03.297 but like we might not do it perfectly. 00:35:03.380 --> 00:35:06.280 What about the folks who say this was just good marketing. 00:35:06.710 --> 00:35:10.017 You know, we, we have suffered enormously commercially 00:35:10.100 --> 00:35:11.547 from not releasing this model. 00:35:11.630 --> 00:35:15.057 This model has incredibly accelerated research within 00:35:15.140 --> 00:35:17.877 Anthropic and production and next models. 00:35:17.960 --> 00:35:21.447 It would do the same in the outside world if, if we were to release it. 00:35:21.530 --> 00:35:24.357 This has hurt us enormously commercially. 00:35:24.440 --> 00:35:27.837 Have you had to make trade-offs already that you're not entirely comfortable with? 00:35:27.920 --> 00:35:32.097 Throughout the entire history of Anthropic has been trade-offs, right? 00:35:32.180 --> 00:35:34.287 In some ideal world, you would prefer to, 00:35:34.370 --> 00:35:36.597 before you release the first chat bot, you know, 00:35:36.680 --> 00:35:39.067 you could spend years studying, you know, every, 00:35:39.150 --> 00:35:41.197 every possible thing that could go wrong with it. 00:35:41.280 --> 00:35:44.167 Now, we did delay, we did delay the initial release 00:35:44.250 --> 00:35:46.537 of Claude, but you know, we did it for a few months. 00:35:46.620 --> 00:35:51.037 So everything is a trade off now that we're in, you know, 00:35:51.120 --> 00:35:54.097 what I would describe as a commercially leading position, 00:35:54.180 --> 00:35:57.817 we can afford to move the dial even further towards being careful, right? 00:35:57.900 --> 00:35:59.947 That's what the Mythos release was about, right? 00:36:00.030 --> 00:36:03.997 It's very hard to do something like that if you're not the leading player. 00:36:04.080 --> 00:36:07.807 There's this argument, why wouldn't the government take you over? 00:36:07.890 --> 00:36:12.427 Why would they let a private company control technology? That's so powerful? 00:36:12.510 --> 00:36:14.917 So I think, I actually think that's a very, 00:36:15.000 --> 00:36:18.277 that's a very serious question and I share those concerns. 00:36:18.360 --> 00:36:21.260 I don't think the government should outright take us over. 00:36:22.260 --> 00:36:26.737 Every previous powerful technology we've seen in history was 00:36:26.820 --> 00:36:29.587 either built by the government or originated with the government. 00:36:29.670 --> 00:36:33.097 So nuclear weapons, obviously, you know, initially built 00:36:33.180 --> 00:36:38.887 by the government, the internet, GPS, cell phones, AI is the first technology 00:36:38.970 --> 00:36:40.920 that's been built in the private sector 00:36:41.700 --> 00:36:44.467 and where government has not really had a serious role 00:36:44.550 --> 00:36:46.567 and is coming in late to the game. 00:36:46.650 --> 00:36:49.747 I think that's actually a dangerous and unstable situation. 00:36:49.830 --> 00:36:53.287 It is not the situation I would've chosen this technology. 00:36:53.370 --> 00:36:56.827 I'm scared of companies having it, but I'm also scared of government having it. 00:36:56.910 --> 00:37:00.607 And then, you know, we need basic regulation of the technology. 00:37:00.690 --> 00:37:04.627 You know, more and more as I've seen what we've seen with Mythos, you know, 00:37:04.710 --> 00:37:07.447 I think we need to start doing pre-release testing, 00:37:07.530 --> 00:37:11.647 required pre-release testing, testing and auditing of the models. 00:37:11.730 --> 00:37:13.597 And this was an approach the White House rejected 00:37:13.680 --> 00:37:18.667 initially on his first day back in office, President Trump with the help of former AI 00:37:18.750 --> 00:37:22.477 and crypto czar, David Sachs dismantled President Biden's AI 00:37:22.560 --> 00:37:26.557 executive order seeking guardrails, instead favoring a hands-off, 00:37:26.640 --> 00:37:29.377 let Silicon Valley do its thing approach. 00:37:29.460 --> 00:37:31.717 We believe that excessive regulation 00:37:31.800 --> 00:37:36.247 of the AI sector could kill a transformative industry just 00:37:36.330 --> 00:37:40.657 as it's taking off. But with Mythos and its national security implications, 00:37:40.740 --> 00:37:44.077 proving hard to ignore the White House now seems to want 00:37:44.160 --> 00:37:46.620 to gatekeep the world's most powerful ai 00:37:47.910 --> 00:37:51.037 It's very funny to me how there's a particular group 00:37:51.120 --> 00:37:54.037 of people in the tech world in Silicon Valley, they started 00:37:54.120 --> 00:37:57.847 with a position of like, you know, even having transparency 00:37:57.930 --> 00:38:00.367 around this technology, even export control, you know, 00:38:00.450 --> 00:38:05.437 this is all, you know, just totally, it'll apocalyptically destroy our potential 00:38:05.520 --> 00:38:08.437 to create the technology. It'll kill innovation. 00:38:08.520 --> 00:38:10.567 And then as soon as they see the first real danger, 00:38:10.650 --> 00:38:13.057 which I've been expecting all along, there's all this talk 00:38:13.140 --> 00:38:15.907 of like nationalization and the government should just seize it. 00:38:15.990 --> 00:38:17.257 Come on folks here, you're yo, 00:38:17.340 --> 00:38:21.997 you're yo-yoing from like the most extreme anti-regulatory. 00:38:22.080 --> 00:38:25.267 If you look at us the wrong way, you're destroying the industry to, you know, 00:38:25.350 --> 00:38:28.477 this completely communist, the government should grab it all. 00:38:28.560 --> 00:38:32.977 We, we need, we, we need a more, we need a more sensible, moderate approach. 00:38:33.060 --> 00:38:36.797 That's the one we've been favoring all along because we've, we've understood the 00:38:36.880 --> 00:38:41.087 power of this technology. We're not panicking, we're not denying it. 00:38:41.170 --> 00:38:44.720 We see the smooth exponential and we're responding to it appropriately. 00:38:51.820 --> 00:38:56.087 The reaction to AI right now, it's intense. 00:38:56.170 --> 00:38:58.937 Anthropic has built this really loyal following. 00:38:59.020 --> 00:39:01.637 There are some people who just love what they stand for, 00:39:01.720 --> 00:39:04.967 but there've also been protests right outside their office. 00:39:05.050 --> 00:39:07.817 There's a lot of anxiety, a lot of confusion, 00:39:07.900 --> 00:39:12.617 there's some real anger right now about what's happening 00:39:12.700 --> 00:39:15.047 and it actually feels like it's escalating. 00:39:15.130 --> 00:39:20.530 Artificial intelligence is the next industrial revolution. Whoa. 00:39:25.960 --> 00:39:29.327 If you look at the data, people are more concerned than excited 00:39:29.410 --> 00:39:33.347 about what's going on. They think the risks outweigh the benefits 00:39:33.430 --> 00:39:37.097 and the truth is, if you talk to the people who are building this, 00:39:37.180 --> 00:39:41.320 even they will tell you they don't fully know how it's all going to play out. 00:39:45.160 --> 00:39:47.867 How do you think about the weight of this moment? 00:39:47.950 --> 00:39:50.477 I worry that something will go wrong. 00:39:50.560 --> 00:39:52.427 You know, are we doing literally everything we can? 00:39:52.510 --> 00:39:55.907 We're certainly trying our best, we're certainly trying very hard. 00:39:55.990 --> 00:39:58.217 What I want is to create a situation where if this set 00:39:58.300 --> 00:40:00.197 of people can't do it, it couldn't be done. 00:40:00.280 --> 00:40:03.330 You can't guarantee success, but maybe you can guarantee that 00:40:04.030 --> 00:40:08.417 It's getting personal. Sam Altman's home has been attacked. 00:40:08.500 --> 00:40:11.717 How is that affecting you? It was really scary to read that. 00:40:11.800 --> 00:40:15.497 I mean, it was, we were obviously extremely relieved that he 00:40:15.580 --> 00:40:17.477 and his family were okay in general. 00:40:17.560 --> 00:40:19.997 I think this is a time, you know, technologically, 00:40:20.080 --> 00:40:23.747 politically where unfortunately there's just a lot more 00:40:23.830 --> 00:40:26.747 rhetoric and, and words that I think can lead to, 00:40:26.830 --> 00:40:29.597 to bad outcomes and bad things happening. 00:40:29.680 --> 00:40:31.577 I hope that this is a topic we can all just debate, 00:40:31.660 --> 00:40:34.007 you know, as peacefully as possible. 00:40:34.090 --> 00:40:38.237 It was scary to me too. I mean, you know, this is like a, a less savory aspect 00:40:38.320 --> 00:40:41.417 of the exponential, right? That as AI gets more 00:40:41.500 --> 00:40:43.457 and more of, of a big thing like, you know, 00:40:43.540 --> 00:40:46.787 it becomes just such a big deal to society. 00:40:46.870 --> 00:40:51.347 There's more attention on it. There's been massive backlash against social media. 00:40:51.430 --> 00:40:54.550 Countries are starting to ban it. Could that happen to AI? 00:40:55.960 --> 00:40:58.097 I think it's, it's absolutely possible. 00:40:58.180 --> 00:41:00.617 If the social media companies could go back in time 00:41:00.700 --> 00:41:04.577 and see the world that they see today, would they do anything differently? 00:41:04.660 --> 00:41:05.987 I like to think the answer to that is yes. 00:41:06.070 --> 00:41:09.677 I don't know if we sort of project some of the challenges 00:41:09.760 --> 00:41:11.597 that the social media companies have faced 00:41:11.680 --> 00:41:15.527 around child welfare, mental health, election integrity, all 00:41:15.610 --> 00:41:19.097 of these topics we're really lucky that we're we're second. 00:41:19.180 --> 00:41:22.307 We view it as our job to try and proactively think about all 00:41:22.390 --> 00:41:24.137 of the things that could go wrong. 00:41:24.220 --> 00:41:25.847 Because if we don't, who's going to, 00:41:25.930 --> 00:41:28.277 You know, I don't know that they actually set out 00:41:28.360 --> 00:41:30.707 to do the right thing or make the world a better place. 00:41:30.790 --> 00:41:32.937 And so I don't think if they were going back, 00:41:33.020 --> 00:41:34.497 they would even knowing what they do, 00:41:34.580 --> 00:41:36.687 and they certainly should do things differently. 00:41:36.770 --> 00:41:39.207 I don't know if they actually will, but we can't. 00:41:39.290 --> 00:41:43.047 This is why we're trying to get this right the first time instead of waiting 00:41:43.130 --> 00:41:48.207 for things to go wrong, then scrambling to justify why it's all okay. 00:41:48.290 --> 00:41:52.227 The main way I could see a AI being, you know, banned 00:41:52.310 --> 00:41:55.467 or blocked is if something really went wrong and if something really went wrong, 00:41:55.550 --> 00:42:00.147 then maybe it deserves to be, You have technologists saying it's gonna be amazing 00:42:00.230 --> 00:42:02.007 and others saying it could be awful. 00:42:02.090 --> 00:42:04.617 Where are you on that spectrum? 00:42:04.700 --> 00:42:07.317 I hope for the best, but plan for the worst. 00:42:07.400 --> 00:42:10.167 For me, this is the most important work I've ever done. 00:42:10.250 --> 00:42:12.987 For some people it feels like this is the last job 00:42:13.070 --> 00:42:16.377 because this is sort of the, it could mean the end of work 00:42:16.460 --> 00:42:20.547 and I think for other people it's, you know, like maybe it's not the last job, 00:42:20.630 --> 00:42:23.457 but it's existential to get this right. 00:42:23.540 --> 00:42:25.677 And so there's just this, this kind of burden. 00:42:25.760 --> 00:42:30.657 If the impacts could be as significant as we're talking about, as significant 00:42:30.740 --> 00:42:35.427 as Anthropic has warned about, what responsibility do you think Anthropic has 00:42:35.510 --> 00:42:41.810 to cushion the blow? What do you owe the people whose lives you've upended? 00:42:43.130 --> 00:42:45.747 I think our view has always been, 00:42:45.830 --> 00:42:48.567 we ultimately are responsible as an industry, the industry 00:42:48.650 --> 00:42:52.857 that's developing this technology for thinking through what 00:42:52.940 --> 00:42:55.857 are the risks, what are the bad things that could happen? 00:42:55.940 --> 00:42:59.817 And if some of those things come to pass, what is our role in helping to fix them? 00:42:59.900 --> 00:43:03.777 That is our job, right? We should not just say, well, we were just trying 00:43:03.860 --> 00:43:05.667 to grow the product and suddenly there's an entire 00:43:05.750 --> 00:43:09.267 generation of young women who have eating disorders 00:43:09.350 --> 00:43:10.977 or who have mental health problems 00:43:11.060 --> 00:43:13.167 because whoops, we were just trying to grow the product. 00:43:13.250 --> 00:43:16.137 That's not the stance that I think any technology company should take. 00:43:16.220 --> 00:43:19.070 I don't think that's the stance that we're trying to take 00:43:19.820 --> 00:43:23.487 For a company whose identity is so wrapped up in, We want to do this right, 00:43:23.570 --> 00:43:25.467 We exist as a AI safety company. 00:43:25.550 --> 00:43:27.837 How can we just help all of this go? Well, 00:43:27.920 --> 00:43:30.357 It can be hard to understand why Anthropic is pushing 00:43:30.440 --> 00:43:36.267 so hard to advance AI while being so upfront about the dangers. In his essays, 00:43:36.350 --> 00:43:38.967 Dario lays out what the end game looks like if everything 00:43:39.050 --> 00:43:45.417 goes right with ai, a utopian future where machines and humans work side by side AI, 00:43:45.500 --> 00:43:48.657 an inevitable force steered toward prosperity rather than 00:43:48.740 --> 00:43:53.097 catastrophe to mitigate the devastation of job loss. 00:43:53.180 --> 00:43:56.217 He proposes solutions like universal basic income 00:43:56.300 --> 00:43:58.820 and progressive taxation of AI companies. 00:44:00.920 --> 00:44:03.897 But as the Amodeis confront the messy realities of power, 00:44:03.980 --> 00:44:06.507 politics, and profit, the real test is whether 00:44:06.590 --> 00:44:10.040 that founding mission can survive the scale of what they're building. 00:44:11.390 --> 00:44:14.007 Google started with the motto, don't be evil. 00:44:14.090 --> 00:44:17.240 A founding promise the company quietly retired as it grew. 00:44:19.610 --> 00:44:22.167 You are building something incredibly powerful 00:44:22.250 --> 00:44:25.197 and stand to gain enormously from it. 00:44:25.280 --> 00:44:29.877 Why should we trust you? I think starting from a position of distrust, you know, 00:44:29.960 --> 00:44:31.917 if you don't know anything about, if you don't know anything about 00:44:32.000 --> 00:44:33.427 philanthropic, it's pretty rational. 00:44:33.510 --> 00:44:36.487 I think Silicon Valley has lost a lot of the world's trust 00:44:36.570 --> 00:44:38.377 and kind of has to re-earn it, 00:44:38.460 --> 00:44:42.817 and the message, you know, we're trying to send is, we're actually different and, 00:44:42.900 --> 00:44:45.937 and that has to be earned in things that we actually do. 00:44:46.020 --> 00:44:49.387 I understand one of your favorite books is The Making of the Atomic Bomb. 00:44:49.470 --> 00:44:52.897 That is correct. Do you see parallels between yourself and Oppenheimer? 00:44:52.980 --> 00:44:55.927 The figure I most identified with was Leo Szilard 00:44:56.010 --> 00:45:00.390 who was the one who first had the idea that there could be a chain reaction. 00:45:01.380 --> 00:45:03.997 Look, my view is we're, we're not gonna get through this 00:45:04.080 --> 00:45:06.157 with like larger than life personalities 00:45:06.240 --> 00:45:09.907 or like figures who try and be at the center of everything. 00:45:09.990 --> 00:45:12.937 In some ways, I actually see Oppenheimer as a failure case, as 00:45:13.020 --> 00:45:17.767 what should not happen. There's a lot of powerful actors who have interests here, 00:45:17.850 --> 00:45:19.537 and the only way it's gonna end well 00:45:19.620 --> 00:45:24.727 for everyone is if there is some, there's basically checks and balances everywhere. 00:45:24.810 --> 00:45:28.297 You've said there's roughly a 10 to 25% chance 00:45:28.380 --> 00:45:32.917 of civilization collapse, that is not insignificant. 00:45:33.000 --> 00:45:38.827 Is there a scenario where it's something that Anthropic built, that caused that? 00:45:38.910 --> 00:45:44.707 I mean, I certainly hope not. My view is that that probability comes from the very 00:45:44.790 --> 00:45:47.287 straightforward recipe of the technology, the existence 00:45:47.370 --> 00:45:49.087 of many countries in the world. 00:45:49.170 --> 00:45:52.987 The existence of many companies within an economy and new ones created. 00:45:53.070 --> 00:45:56.137 If the void is isn't filled, like that's a dilemma 00:45:56.220 --> 00:45:59.227 that we're in half of what we do within the company is try 00:45:59.310 --> 00:46:03.547 and, you know, reduce the risk as much as we can, but it's never gonna be zero. 00:46:03.630 --> 00:46:06.217 Suppose there are a bunch of, you know, airline companies out there 00:46:06.300 --> 00:46:08.737 and you're like, well, I'm gonna make an airline company that's safer. 00:46:08.820 --> 00:46:11.047 It can both be the case that, you know, 00:46:11.130 --> 00:46:14.977 your airline company is 10 times safer than all the other airline companies. 00:46:15.060 --> 00:46:18.187 But if someone comes and asks you, like, can you guarantee 00:46:18.270 --> 00:46:20.347 that your airplane will never crash? 00:46:20.430 --> 00:46:22.687 I mean, how could you, how could you possibly, 00:46:22.770 --> 00:46:25.327 But if there was a 25% chance of an airplane crashing, 00:46:25.410 --> 00:46:26.737 you wouldn't get on that plane. 00:46:26.820 --> 00:46:28.417 That's right. 25% is too high. 00:46:28.500 --> 00:46:30.937 We're trying to make that probability much, much lower. 00:46:31.020 --> 00:46:33.210 That's, that is, that is the goal. 00:47:01.770 --> 00:47:04.597 How do you find your zen? How do you relax? 00:47:04.680 --> 00:47:07.927 You know, it honestly, a lot of it's just exposure to, to it. 00:47:08.010 --> 00:47:11.527 You know, I sometimes I'll just like, you know, I'll just take a weekend 00:47:11.610 --> 00:47:13.477 and I'll like play some video games. 00:47:13.560 --> 00:47:15.637 Sometimes with Daniela Me and my wife go to Italy. 00:47:15.720 --> 00:47:20.197 Sometimes we have a horse there, so I'll just, I'll just sit there next. 00:47:20.280 --> 00:47:21.487 I'll just, I'll just look at the horse 00:47:21.570 --> 00:47:25.147 and I'll be like, you know, Calypso, who's her horse like, 00:47:25.230 --> 00:47:29.800 you know, she doesn't know about any of this, like she's just a happy horse.