WEBVTT 00:00:00.060 --> 00:00:03.720 Can we start with the kind of rough status of how the industry is going? 00:00:04.410 --> 00:00:09.020 You and your rivals, how close are we now to artificial general intelligence? 00:00:09.540 --> 00:00:13.500 Set the stage for us first and then we can talk about the morality of it all. 00:00:13.650 --> 00:00:15.547 Yeah. Yeah. Thanks for. Thanks for having me. 00:00:15.630 --> 00:00:19.957 I mean, you know, I've never liked the artificial general intelligence or 00:00:20.040 --> 00:00:24.227 superintelligence. Not because I don't think is very powerful. Right. 00:00:24.310 --> 00:00:27.187 I'm not I'm not a skeptic. I'm actually extreme in terms of my 00:00:27.270 --> 00:00:29.557 views of how powerful the technology is going to be. 00:00:29.640 --> 00:00:32.977 But it's the wrong model for thinking about it, that there will be some one 00:00:33.060 --> 00:00:35.820 point where we build something completely different. 00:00:36.240 --> 00:00:41.017 What we see here, actually, what we've seen for the last ten years, maybe even 00:00:41.100 --> 00:00:45.157 15, and you know, me and therapist co-founders were among the first to 00:00:45.240 --> 00:00:50.250 document it is this very smooth, exponential, exponential process. 00:00:50.550 --> 00:00:54.870 Just like in the nineties, you saw Moore's Law that, you know, the the 00:00:54.990 --> 00:00:58.740 amount of computing power would double every 12 months or every 18 months. 00:00:58.980 --> 00:01:04.057 We have a moore's Law like law, except it's for intelligence itself, for for 00:01:04.140 --> 00:01:07.350 the cognitive ability of the model across many tasks. 00:01:07.680 --> 00:01:12.090 And so what we are seeing is that that cognitive ability is, you know, 00:01:12.750 --> 00:01:16.457 depending on how you measure it, doubling every, you know, four to 4 to 00:01:16.540 --> 00:01:19.477 12 months. And so, you know, we're just climbing 00:01:19.560 --> 00:01:22.897 the ladder of cognitive ability. And where we are now is, you know, to 00:01:22.980 --> 00:01:27.637 take coding as one example that's been exploding really fast over the last year 00:01:27.720 --> 00:01:29.857 or two. I have engineers. 00:01:29.940 --> 00:01:34.747 In fact, the team that leads one of our lead products called Code, which is the 00:01:34.830 --> 00:01:41.377 way you use our models for for coding. He says he hasn't written any code in 00:01:41.460 --> 00:01:45.787 the last two months. He's it's all caught he's he's edited it 00:01:45.870 --> 00:01:48.487 he's looked at it but it's all been written by Claude. 00:01:48.570 --> 00:01:50.870 Indeed. We recently released this thing called 00:01:50.970 --> 00:01:53.520 Coachwork, which was bringing code code to, 00:01:54.870 --> 00:01:58.830 you know, non-coding tasks. It really seems to have taken off. 00:01:58.980 --> 00:02:02.760 We wrote it in like a week and a half, almost entirely with code code. 00:02:03.060 --> 00:02:06.720 So I think we're entering the world where, you know, the junior level 00:02:06.900 --> 00:02:11.137 software engineers, maybe many of the tasks of the more senior level software 00:02:11.220 --> 00:02:15.997 engineers are starting to be done, you know, most of the way by buying AI 00:02:16.080 --> 00:02:20.017 systems now that's going to go further can be more end to end the project to me 00:02:20.100 --> 00:02:22.650 more. But I think we're going to be surprised 00:02:22.890 --> 00:02:25.117 at how the exponential turns upward. Right? 00:02:25.200 --> 00:02:28.777 The whole the whole thing about exponentials is, you know, it looks like 00:02:28.860 --> 00:02:32.227 it's going very, very slowly. It speeds up a little bit and then it 00:02:32.310 --> 00:02:35.617 just zooms past you. And I think we're I think we're on the 00:02:35.700 --> 00:02:37.837 precipice. I think we're a year or two away from 00:02:37.920 --> 00:02:40.477 it, really zooming past us. So it's very like going bankrupt. 00:02:40.560 --> 00:02:45.787 But when when this happens in terms of the kind of running order does yourself, 00:02:45.870 --> 00:02:50.347 this Gemini, this open air, where do you put you all together at the moment 00:02:50.430 --> 00:02:51.307 roughly? Yeah. 00:02:51.390 --> 00:02:58.087 I mean, you know, so I think I think as the field has developed, you know, I 00:02:58.170 --> 00:03:01.177 think it's no longer appropriate to think of it as like, you know, like a 00:03:01.260 --> 00:03:04.567 scalar, you know, like runners in a race who are at one position. 00:03:04.650 --> 00:03:06.937 The players have gone in different directions. 00:03:07.020 --> 00:03:11.730 So some of the other players have gone into very consumer oriented direction. 00:03:11.940 --> 00:03:16.597 And that leads you to make your models, you know, super humanly engaging or 00:03:16.680 --> 00:03:21.307 super humanly good at recommending shopping or ads or various things like 00:03:21.390 --> 00:03:24.457 that. Anthropic has focused, I think, first 00:03:24.540 --> 00:03:29.617 and foremost on enterprises, developers. And to the extent we do consumer, we're 00:03:29.700 --> 00:03:35.557 very focused on productivity and the kind of high value end of the of the 00:03:35.640 --> 00:03:38.617 consumer work. And so that that gives very that gives 00:03:38.700 --> 00:03:39.937 very different incentives. Right? 00:03:40.020 --> 00:03:43.507 If you if you want to get things done. You know, we talked about AGI in 00:03:43.590 --> 00:03:47.227 superintelligence before. One of the things I would say is that, 00:03:47.310 --> 00:03:50.017 you know, there are superintelligence is today and they're basically large 00:03:50.100 --> 00:03:53.050 corporations, right? They're smarter than any human can be, 00:03:53.730 --> 00:03:57.607 you know, solving the problem of, you know, shipping, shipping commerce at the 00:03:57.690 --> 00:04:01.410 lowest possible cost or making solar panels the lowest possible cost or 00:04:01.590 --> 00:04:04.447 launching rockets at the lowest possible possible cost. 00:04:04.530 --> 00:04:09.187 And so there's really high returns to intelligence in this area, which I think 00:04:09.270 --> 00:04:13.380 incentivizes us to to kind of to kind of build the right thing. 00:04:13.680 --> 00:04:18.007 It's it's also, you know, I think it's I think it's a business that is more 00:04:18.090 --> 00:04:22.040 stable than consumer. We don't need you know, we don't need ads. We don't need, 00:04:23.820 --> 00:04:29.047 you know, large numbers of free users. We can we can just very directly create 00:04:29.130 --> 00:04:33.457 value. Right? There aren't all these weird externalities of, like your prioritizing 00:04:33.540 --> 00:04:35.677 engagement. You're prioritizing these other things, 00:04:35.760 --> 00:04:39.487 you're generating all this slop. We just make stuff that that people can 00:04:39.570 --> 00:04:44.647 use. Year ago here, you what? You said the Chinese were catching up a bit. 00:04:44.730 --> 00:04:47.130 Do you think now the Chinese have fallen behind? 00:04:47.430 --> 00:04:50.947 You know, I think they never really caught up that much. 00:04:51.030 --> 00:04:54.430 So, you know, of course, there was this huge excitement around deep. 00:04:54.720 --> 00:04:59.580 Yeah, right. But, you know, the truth was a couple of things. One. 00:05:00.300 --> 00:05:02.867 You know, those models are very optimized for the benchmarks. 00:05:02.950 --> 00:05:07.680 It's actually very easy to optimize the model for a finite list of benchmarks. 00:05:07.950 --> 00:05:11.580 When we go out into the world, right, when we're when we're when we're, 00:05:12.630 --> 00:05:16.417 you know, competing against other companies for enterprise contracts, we 00:05:16.500 --> 00:05:20.040 see just honestly and candidly, we see Google and we see Openai. 00:05:20.340 --> 00:05:22.987 Every once in a while, we see a couple other U.S. 00:05:23.070 --> 00:05:28.507 players. I have almost never lost a deal, lost a 00:05:28.590 --> 00:05:32.467 contract to to a Chinese model. But now you have the Trump 00:05:32.550 --> 00:05:36.150 administration, and I think you've already protested about this, giving 00:05:36.300 --> 00:05:39.120 high speed chips and video chips to the Chinese. 00:05:39.210 --> 00:05:39.977 That's right. That's right. 00:05:40.060 --> 00:05:43.327 The thing that is holding them back and they've said it themselves. 00:05:43.410 --> 00:05:47.737 The CEOs of these companies say it's the embargo on chips that's holding us back. 00:05:47.820 --> 00:05:52.147 They explicitly say this. And now, indeed, you know, there are 00:05:52.230 --> 00:05:57.187 some policies and I hope they change their mind to, you know, to explicitly 00:05:57.270 --> 00:06:01.117 send not quite our latest generation of chips, although it was reported that 00:06:01.200 --> 00:06:04.950 even that was being considered. But, you know, the generation of chips, 00:06:05.190 --> 00:06:08.580 that's that's just one back that's still extremely powerful. 00:06:09.270 --> 00:06:14.857 And we are many years ahead of China in terms of our in terms of our ability to 00:06:14.940 --> 00:06:17.587 make chips. So I think it would be a big mistake to 00:06:17.670 --> 00:06:21.847 ship these chips. You know, the the analogy I thought of, 00:06:21.930 --> 00:06:27.360 if you think about the incredible national security implications of 00:06:27.630 --> 00:06:31.117 building model, building models that are essentially cognition, that are 00:06:31.200 --> 00:06:32.497 essentially intelligence. Right. 00:06:32.580 --> 00:06:36.247 I've called where we're going with this, a country of geniuses in the data 00:06:36.330 --> 00:06:40.170 center, right. So imagine 100,000, 100 million people 00:06:40.410 --> 00:06:44.317 smarter than any Nobel Prize winner. And it's going to be under the control 00:06:44.400 --> 00:06:48.967 of one, one country or another. So so I think this is crazy. 00:06:49.050 --> 00:06:52.777 I think it's you know, it's a bit like, you know, I don't know, like selling, 00:06:52.860 --> 00:06:56.617 selling, you know, nuclear weapons to North Korea and, you know, bragging. 00:06:56.700 --> 00:07:00.337 Oh, yeah, But we made that. You know, your friend your friend David 00:07:00.420 --> 00:07:05.377 Sachs is basically arming the Chinese. You know, I wouldn't refer to any 00:07:05.460 --> 00:07:07.957 particular people, but, you know, I would I would just say that this 00:07:08.040 --> 00:07:11.190 particular policy, I think, is is akin to. 00:07:11.580 --> 00:07:13.890 Yeah, not not, not, not, I think well-advised. 00:07:14.220 --> 00:07:17.467 Another issue in this, you know, could be you could be right all this 00:07:17.550 --> 00:07:21.397 technology be going in the right direction but economically we could be 00:07:21.480 --> 00:07:24.420 heading towards a bubble. Do you think that's true? 00:07:24.600 --> 00:07:25.237 Yeah. Yeah. 00:07:25.320 --> 00:07:31.717 So I would separate out the two things. One is the basic exponential, and that's 00:07:31.800 --> 00:07:34.777 what we've been talking about. That's been kind of the technological 00:07:34.860 --> 00:07:38.587 direction. And there, you know, just like with 00:07:38.670 --> 00:07:43.967 Moore's Law, you can never be sure that the technology is sort of fully going to 00:07:44.050 --> 00:07:45.247 advanced. It's going to keep going. 00:07:45.330 --> 00:07:47.630 Right? It's a fundamentally inductive problem. 00:07:47.970 --> 00:07:53.017 But now that I've seen well over a decade of it, I would say I am pretty 00:07:53.100 --> 00:07:57.067 confident that it's going to continue, or at least I'm more confident than I've 00:07:57.150 --> 00:08:01.297 ever been at any time in the past that it's going to continue to where the 00:08:01.380 --> 00:08:05.760 models are basically smarter than humans at almost everything. 00:08:06.030 --> 00:08:09.427 And I think there's a good chance that even happens in the next year or two. 00:08:09.510 --> 00:08:13.817 Again, it seems far away. But but this property of exponential 00:08:13.900 --> 00:08:16.537 that they like, they they they catch you by surprise. 00:08:16.620 --> 00:08:20.287 And if it's not a year or two, I think it's pretty likely that it's, you know, 00:08:20.370 --> 00:08:22.740 at least less than five years within the 2020s. 00:08:22.980 --> 00:08:24.787 I think this moment will come in the 2020. 00:08:24.870 --> 00:08:29.277 So so, you know, kind of I'll get into that economic angle. 00:08:29.360 --> 00:08:33.780 Yeah, I'll get to the economic diffusion, which I think is complicated, 00:08:33.990 --> 00:08:37.087 but I don't want anyone to forget like that. 00:08:37.170 --> 00:08:42.360 You know that as long as we have that fundamental technological gain, like 00:08:42.630 --> 00:08:46.207 there's going to be many trillions of revenue, maybe many trillions per 00:08:46.290 --> 00:08:50.490 company in this in this area, because the economic potential is so great. 00:08:50.910 --> 00:08:57.487 Now there's a separate thing, which is that we don't know how fast enterprises 00:08:57.570 --> 00:09:02.040 companies are actually going to be able to use this technology. 00:09:02.190 --> 00:09:05.077 Right. The the the uses of this, what the 00:09:05.160 --> 00:09:09.517 technology is capable of today is probably ten times what the enterprise 00:09:09.600 --> 00:09:11.940 of the world are able to are able to deploy. 00:09:12.210 --> 00:09:17.227 And I see it with our customers every day where I'll talk to a CEO and they'll 00:09:17.310 --> 00:09:21.360 understand and their executive team will understand the power of this technology, 00:09:21.630 --> 00:09:25.237 you know, for for automating customer service, for coding, for many other 00:09:25.320 --> 00:09:27.637 things. But they have a company of tens of 00:09:27.720 --> 00:09:31.897 thousands of people who are perfectly brilliant people but expert in something 00:09:31.980 --> 00:09:35.737 that is not A.I. and have to learn to use A.I. 00:09:35.820 --> 00:09:39.247 and that you know, different names for it change management, enterprise 00:09:39.330 --> 00:09:43.147 transformation, whatever it's called the need to do. 00:09:43.230 --> 00:09:45.930 This is very slow. It can take years. 00:09:46.290 --> 00:09:50.587 And so we have this powerful technology that I have an incredible amount of 00:09:50.670 --> 00:09:53.340 confidence will generate trillions in revenue. 00:09:53.580 --> 00:09:58.147 But we don't know exactly when, you know, plus, plus or minus a few years. 00:09:58.230 --> 00:10:02.587 We don't know exactly when. And and and in the meantime, companies 00:10:02.670 --> 00:10:05.280 have to buy compute to serve all that revenue. 00:10:05.550 --> 00:10:09.487 And you don't want to buy too much because, you know, you could financially 00:10:09.570 --> 00:10:13.717 overextend yourself and you don't want to buy too little because then you can't 00:10:13.800 --> 00:10:18.097 serve all the revenue. And so that's where the kind of economic 00:10:18.180 --> 00:10:21.907 and financial problems. So you are involved in building data 00:10:21.990 --> 00:10:25.440 centers, spending vast fortunes. You've got the deals with Belvedere. 00:10:25.620 --> 00:10:27.757 So you worry that that's going to come back and bite you. 00:10:27.840 --> 00:10:32.977 So so we have some advantages in this. There is a stability to enterprise 00:10:33.060 --> 00:10:34.987 business that there is not to consumer, right. 00:10:35.070 --> 00:10:40.057 Consumers, very fickle enterprise purchasing enterprise predictions are 00:10:40.140 --> 00:10:43.207 predictable. Also, enterprise tends to have better 00:10:43.290 --> 00:10:46.237 margins. The margin is basically how much buffer 00:10:46.320 --> 00:10:50.270 you have between buying too little and buying too much between what you have to 00:10:50.370 --> 00:10:53.130 pay for and what you have to support in revenue. 00:10:53.340 --> 00:10:57.487 So we have a number of advantages here. But I you know, I won't deny that there 00:10:57.570 --> 00:11:02.940 is some inherent risk in any in all of the companies to this process. 00:11:03.120 --> 00:11:08.377 Whenever you have an upside whose amount is uncertain or even whose timing is 00:11:08.460 --> 00:11:13.417 uncertain, and you have to make financial decisions somewhat blind to 00:11:13.500 --> 00:11:18.187 those things, you have to you have to decide early to, you know, to to build 00:11:18.270 --> 00:11:21.900 out capital, to build out resources a few years in advance. 00:11:22.170 --> 00:11:25.537 There's going to be some inherent risk that's driven by that uncertainty. 00:11:25.620 --> 00:11:29.130 And I do think that some companies have probably overbought. 00:11:29.430 --> 00:11:31.590 I'm not able to look at their financials. 00:11:31.860 --> 00:11:35.557 I just know what they announce. And I'm like, Whoa, I wouldn't 00:11:35.640 --> 00:11:39.067 necessarily have done that. I'm pretty happy with the decisions that 00:11:39.150 --> 00:11:42.607 we've made. But, you know, and I can't be I can't be 00:11:42.690 --> 00:11:47.767 sure about others, but it is entirely consistent that this could be the most 00:11:47.850 --> 00:11:51.847 transformative technology in the history of the world, that some companies will 00:11:51.930 --> 00:11:55.147 do really well. But but not every company may very 00:11:55.230 --> 00:11:57.067 quickly on that. As you said, you own the enterprise 00:11:57.150 --> 00:12:00.217 side. On the consumer side, Openai with a big 00:12:00.300 --> 00:12:01.987 lead. You know, there seems to be look to 00:12:02.070 --> 00:12:05.170 Gemini as an eye catching up. Is that your impression as well? 00:12:05.370 --> 00:12:09.180 I mean, you know, we're not in that market, so I so I don't know about that. 00:12:09.870 --> 00:12:13.057 Yeah. You know, the statistics would seem to 00:12:13.140 --> 00:12:16.897 suggest that it would seem to suggest that, you know, with their kind of 00:12:16.980 --> 00:12:21.097 greater distribution, with their greater existing reach, they've been able to 00:12:21.180 --> 00:12:23.637 make a lot of progress in. Kind of recent times. 00:12:23.720 --> 00:12:26.457 And again, this is why, you know, consumers fickle. 00:12:26.540 --> 00:12:31.707 This is why I would be careful if, you know, if I were a consumer company in in 00:12:31.790 --> 00:12:36.177 in you know, in kind of you know, what you know in what was on my balance 00:12:36.260 --> 00:12:37.357 sheet. Let's jump then. 00:12:37.440 --> 00:12:39.927 It's that big question which we originally asked Claude at the 00:12:40.010 --> 00:12:45.177 beginning, will I do do do good. Obviously, as you said, it's going to 00:12:45.260 --> 00:12:48.800 have a positive effect on GDP, perhaps on your wealth, too. 00:12:49.190 --> 00:12:52.227 But what about employment? Last year you predicted the kind of 00:12:52.310 --> 00:12:56.637 white collar bloodbath. You said it would wipe out 50% of entry 00:12:56.720 --> 00:13:00.410 level jobs. That's now four and a half years away. 00:13:00.860 --> 00:13:02.877 Do you still stand by that? Yeah. 00:13:02.960 --> 00:13:09.207 So, you know, I you know, my view of of of I there's maybe two axes in terms of 00:13:09.290 --> 00:13:13.250 which two think there's good things happening versus bad things happening. 00:13:13.550 --> 00:13:16.197 And there's like A.I. is a small deal versus A.I. 00:13:16.280 --> 00:13:19.797 is a big deal. You know, my view is on the extreme side 00:13:19.880 --> 00:13:25.167 of I is a big deal, but I'm kind of in both side, both both top quadrants where 00:13:25.250 --> 00:13:27.297 I think some really good things will happen. 00:13:27.380 --> 00:13:33.200 And if we don't act to prevent them, some really bad things will happen. 00:13:34.010 --> 00:13:37.077 You know, we you know, I wouldn't be building this technology if I didn't 00:13:37.160 --> 00:13:41.697 believe that the good outweighed the bad and that we could mitigate the bad. 00:13:41.780 --> 00:13:44.247 That's why I warn about the bad, so that we could We can. 00:13:44.330 --> 00:13:45.717 How do you how do you think we mitigate it? 00:13:45.800 --> 00:13:48.320 Does that mean things like taxation? Yeah. 00:13:48.500 --> 00:13:52.340 Well, let me let me just get to what I think the kind of the situation is. 00:13:52.640 --> 00:13:57.087 You know, I think exactly what you said. I think we could have this very unusual 00:13:57.170 --> 00:14:03.590 combination of very fast GDP growth and high unemployment or at least 00:14:03.890 --> 00:14:09.027 underemployment or, you know, low wage job, a lot of low wage jobs, high 00:14:09.110 --> 00:14:11.637 inequality. I don't think that's a macro correct me 00:14:11.720 --> 00:14:13.587 if I'm wrong. I don't think that's a macroeconomic 00:14:13.670 --> 00:14:17.547 combination we've ever seen before. You think of fast growth, you're like, 00:14:17.630 --> 00:14:21.237 well, okay, maybe there's inflation, but you know, you're not going to have high 00:14:21.320 --> 00:14:26.007 unemployment when there's fast growth. I think this technology is a bit 00:14:26.090 --> 00:14:29.787 different because it's extreme in the way in the way it's going to it's going 00:14:29.870 --> 00:14:34.137 to it's going to generate value. But also because it's moving up the 00:14:34.220 --> 00:14:37.647 cognitive waterline, there's going to be, unfortunately, a whole class of 00:14:37.730 --> 00:14:41.907 people who are who are, I think, across a lot of industries going to have a hard 00:14:41.990 --> 00:14:44.967 time coping. And that's really a problem we need to 00:14:45.050 --> 00:14:47.997 solve. So to get to the question you just 00:14:48.080 --> 00:14:50.430 asked, there's a few things Anthropic is doing. 00:14:51.020 --> 00:14:56.337 One is we maintain for almost a year now what we call the anthropic economic 00:14:56.420 --> 00:15:00.980 index, which is tracking how our models are used in real time. 00:15:01.250 --> 00:15:05.397 And because we see all these conversations, we can use quite itself 00:15:05.480 --> 00:15:10.427 to in a privacy preserving manner. Look across all the conversations and 00:15:10.510 --> 00:15:15.437 ask questions like, Is someone using this to augment a task to work together 00:15:15.520 --> 00:15:18.580 with the model or to delegate or fully automated task? 00:15:18.790 --> 00:15:22.667 What are the industries people are using called for statistically by 00:15:22.750 --> 00:15:26.957 distribution? What are the subtasks within those industries? Right. 00:15:27.040 --> 00:15:30.167 We can get into, you know, a lot of these like lead, you know, these very 00:15:30.250 --> 00:15:33.287 fine grained Labor Department statistics and ask all about it. 00:15:33.370 --> 00:15:37.037 Which states are using cloud more? What is we can watch the economic 00:15:37.120 --> 00:15:41.327 diffusion of cloud in real time. And and so the reason to do all this is 00:15:41.410 --> 00:15:45.260 that I don't think you can make good policy if you don't have the right data. 00:15:45.610 --> 00:15:49.097 And I worry that the data that the government produces as comprehensive as 00:15:49.180 --> 00:15:53.177 it is, just doesn't move fast enough and isn't detailed enough for this. 00:15:53.260 --> 00:15:57.130 So so, you know, that is that is kind of our first contribution. 00:15:58.000 --> 00:16:02.387 We are thinking increasingly about this kind of I don't know if you want to call 00:16:02.470 --> 00:16:05.980 it retraining or ability to help people adapt. 00:16:06.310 --> 00:16:10.577 It's it's partly in line with kind of what we do in terms of go to market, 00:16:10.660 --> 00:16:12.857 Right. Part of our go to market activity is how 00:16:12.940 --> 00:16:15.167 to help you with that, but with great risk, great respect. 00:16:15.250 --> 00:16:18.407 I mean, you can do these things, but fundamentally it's a societal problem. 00:16:18.490 --> 00:16:21.737 You know, government, if you think about it, what you're describing is a perfect 00:16:21.820 --> 00:16:25.060 storm. You have GDP going up, the wealth of 00:16:25.240 --> 00:16:27.647 certain people going massively. Absolutely. 00:16:27.730 --> 00:16:32.380 And you are saying 50% of entry level jobs go people in real distressed. 00:16:32.680 --> 00:16:36.727 You are surely going to see a political. Change coming from this. 00:16:36.810 --> 00:16:39.627 And my personal bet would be, yes, you're going to get people trying to 00:16:39.710 --> 00:16:44.337 restrain the tech barons, but you would also get people saying, let's have 00:16:44.420 --> 00:16:46.400 higher taxes. We need to pay for this. 00:16:46.640 --> 00:16:50.997 Do you think higher taxes are coming? So, you know, my view again, you know, I 00:16:51.080 --> 00:16:54.987 was going to the kind of voluntary stuff we could do first, but but actually, I 00:16:55.070 --> 00:16:58.107 actually do believe that. I think this one is going to be big 00:16:58.190 --> 00:17:02.487 enough that, you know, at some point I think everyone's going to come to the 00:17:02.570 --> 00:17:08.607 realization that there needs to be some kind of macroeconomic intervention 00:17:08.690 --> 00:17:11.217 there. You know, if we if we look at just the 00:17:11.300 --> 00:17:18.207 disparities in wealth that we have now, if we look at it as a fraction of GDP, I 00:17:18.290 --> 00:17:21.020 believe we've kind of exceeded the Gilded Age already. 00:17:21.170 --> 00:17:25.107 And this is mostly without without I. So I think things are going to go even 00:17:25.190 --> 00:17:27.717 further. And so, yeah, you know, my my guess is 00:17:27.800 --> 00:17:30.087 this is this is not even going to be a partisan thing. 00:17:30.170 --> 00:17:32.517 There's a wealth there's a wealth tax now in California. 00:17:32.600 --> 00:17:35.760 Are you supportive of that? No, I think it's a great start. 00:17:36.170 --> 00:17:40.467 I think that's poorly designed. And, you know, my my my worry would be 00:17:40.550 --> 00:17:45.057 if we don't think about these things in a sober way, then we'll get things that 00:17:45.140 --> 00:17:48.027 are that are you know, that are that are kind of poorly designed. 00:17:48.110 --> 00:17:52.347 So, you know, that would be that would be my message to, you know, the others 00:17:52.430 --> 00:17:57.687 in the world who are doing who are doing well in this boom is that, you know, if 00:17:57.770 --> 00:18:03.447 we if we don't think proactively about how to make this revolution work for 00:18:03.530 --> 00:18:07.500 everyone, we will get these proposals that don't make sense. 00:18:08.330 --> 00:18:10.797 Just taking it a bit further, there's you know, people worry about the 00:18:10.880 --> 00:18:13.197 economic consequences, but there are other ones. 00:18:13.280 --> 00:18:16.767 Not that long ago I went to a panel with, I think a couple of your rivals, a 00:18:16.850 --> 00:18:20.877 couple of your peers, where they again, they put forward all the good things I 00:18:20.960 --> 00:18:22.377 would do. And then someone said, what about the 00:18:22.460 --> 00:18:24.567 risks? And one of them rather casually said, 00:18:24.650 --> 00:18:27.327 well, of course there's extinction risk and then moved on to another. 00:18:27.410 --> 00:18:31.490 And collectively around the room, people took a deep breath. 00:18:31.820 --> 00:18:35.247 And what he was talking about was the possibility of a bad actor getting hold 00:18:35.330 --> 00:18:39.107 of it, or the possibility ultimately of things like Claud running things. 00:18:39.190 --> 00:18:41.307 What does that fit into your. Yeah, yeah. 00:18:41.390 --> 00:18:44.217 You know, I've long been been worried about that as well. 00:18:44.300 --> 00:18:47.937 You know, just, just as I said, you know, I'm on the direction of like, this 00:18:48.020 --> 00:18:52.130 is a very powerful technology. Therefore the benefits are extreme. 00:18:52.280 --> 00:18:55.737 We're going to be able to, you know, do things like really seriously, really 00:18:55.820 --> 00:18:59.090 cure cancer or maybe eradicate tropical diseases. 00:18:59.450 --> 00:19:03.567 But on the other side, look, we're building cognitive systems that have 00:19:03.650 --> 00:19:06.957 their own autonomy, Right. And we really need to think about that. 00:19:07.040 --> 00:19:12.807 And lycanthropy has been founded since the beginning with an intention to think 00:19:12.890 --> 00:19:13.537 about that. Right. 00:19:13.620 --> 00:19:16.617 You know, we publish probably almost every, you know, probably three or four 00:19:16.700 --> 00:19:23.187 times a month research on how to control these models, how to make sure that, you 00:19:23.270 --> 00:19:26.727 know, they they do what we want them to do and they don't run out of control. 00:19:26.810 --> 00:19:31.287 We've pioneered one of my co-founders, Chris Ola, is, you know, kind of 00:19:31.370 --> 00:19:34.977 arguably, you know, the inventor of a field called mechanistic 00:19:35.060 --> 00:19:38.517 interpretability, which is basically when you look inside the brain, the 00:19:38.600 --> 00:19:42.867 artificial brain of of Claude or another A.I. 00:19:42.950 --> 00:19:48.590 model and try to trace mechanistically why it's doing exactly what it's doing. 00:19:48.800 --> 00:19:54.260 And we've seen things inside the model like, you know, in in lab environments, 00:19:54.560 --> 00:19:58.647 you know, sometimes the models will develop the intent to blackmail, the 00:19:58.730 --> 00:20:01.317 intent to deceive. And this isn't unique to court. 00:20:01.400 --> 00:20:03.550 If anything, this is worse in other models. 00:20:03.710 --> 00:20:08.150 These are things that if we don't train the models in the right way can emerge. 00:20:08.450 --> 00:20:13.017 And but we've pioneered the science of looking inside them so we can diagnose 00:20:13.100 --> 00:20:17.457 them and and, you know, prevent the models from happening, intervene and 00:20:17.540 --> 00:20:20.127 retrain the models so they don't behave in this way. 00:20:20.210 --> 00:20:24.260 That said, you know, we this is something we have to be careful about. 00:20:24.530 --> 00:20:29.390 We've we've, you know, supported various what, you know, light touch, 00:20:29.540 --> 00:20:36.327 transparency focused measures for making sure that, you know, all the companies 00:20:36.410 --> 00:20:38.817 have to talk about the tests that they run. 00:20:38.900 --> 00:20:44.067 We always disclose our tests. We always, you know, basically we try 00:20:44.150 --> 00:20:48.417 and test our models to try and really stream them and get them to do the worst 00:20:48.500 --> 00:20:52.167 things possible in a test environment so that they never do those things in the 00:20:52.250 --> 00:20:56.127 real world. And we generally think every company 00:20:56.210 --> 00:20:59.997 should have to disclose or should have to run those tests and should have to 00:21:00.080 --> 00:21:03.590 disclose those tests. So this is you spend quite a lot of time 00:21:03.680 --> 00:21:05.637 thinking about this. You write pamphlets about it, about the 00:21:05.720 --> 00:21:08.787 consequences. There is an impression that most of your 00:21:08.870 --> 00:21:12.447 peers are going to head down and they're thinking purely about whether they can 00:21:12.530 --> 00:21:15.747 just keep ahead of the other. Do you think your industry is kind of 00:21:15.830 --> 00:21:19.497 immature in that way? You know, I can't speak to what the 00:21:19.580 --> 00:21:22.640 other players are are doing or why they're doing it. 00:21:23.450 --> 00:21:28.310 You know, I think there are at least some other players in in the ecosystem 00:21:28.520 --> 00:21:31.587 who I, you know, I think of as responsible who are at least thinking of 00:21:31.670 --> 00:21:34.527 things in the right way. I agree there are others who are not. 00:21:34.610 --> 00:21:38.407 But. I would say what I have always tried to 00:21:38.490 --> 00:21:44.107 do and what Anthropic has always tried to do is, is to set an example and to 00:21:44.190 --> 00:21:47.040 try to inspire others to follow the example. 00:21:47.220 --> 00:21:50.237 We always publish our work on mechanistic interpretability. 00:21:50.320 --> 00:21:54.697 We always publish the tests we run. And honestly, there are a lot of other 00:21:54.780 --> 00:21:58.837 companies where, you know, their researchers basically say, Hey, why 00:21:58.920 --> 00:22:01.807 can't why can't we do that too? That seems like a responsible thing to 00:22:01.890 --> 00:22:03.937 do. That seems like the right way to develop 00:22:04.020 --> 00:22:07.387 the technology. And so you can kind of have an effect on 00:22:07.470 --> 00:22:11.947 the other players in the ecosystem. 111 very clear zone where you are 00:22:12.030 --> 00:22:15.967 different from your peers that you have not very obviously queued up to kiss the 00:22:16.050 --> 00:22:18.210 ring of Donald J. Trump. 00:22:19.110 --> 00:22:21.217 Are you still? What do you think of the current US 00:22:21.300 --> 00:22:24.007 President? You showed no signs of wanting to see 00:22:24.090 --> 00:22:26.827 him. Look, so, I mean, you know, I wish I 00:22:26.910 --> 00:22:28.327 would. I would really put it in a different 00:22:28.410 --> 00:22:33.150 way. I don't think being for or against 00:22:33.480 --> 00:22:39.997 administrations or for or against politicians is is the right approach or 00:22:40.080 --> 00:22:44.670 that anthropic has anything to say one way or another on those topics. 00:22:44.880 --> 00:22:50.617 What I would say is that what Anthropic knows is I Anthropic knows very well the 00:22:50.700 --> 00:22:56.137 policy issues around A.I. and our approach, which, you know, I 00:22:56.220 --> 00:22:59.347 can't speak for others, but you know, indeed, indeed our approach may have 00:22:59.430 --> 00:23:05.077 been different is we think through the issues, we try and get a substantive 00:23:05.160 --> 00:23:09.570 view based on those issues, and then we say what we think. 00:23:10.020 --> 00:23:15.277 And sometimes we will disagree with the current administration, just as we 00:23:15.360 --> 00:23:20.317 sometimes just like with China, with the last administration, and sometimes we 00:23:20.400 --> 00:23:22.987 agree. And it's worth highlighting the areas of 00:23:23.070 --> 00:23:26.340 agreement like the energy data center buildouts. 00:23:26.610 --> 00:23:30.270 We did this health pledge at the White House. 00:23:31.170 --> 00:23:35.587 You know, there are a number of areas, you know, when we saw the you know, the 00:23:35.670 --> 00:23:40.050 White House's air action plan last summer, actually, we liked most of that. 00:23:40.350 --> 00:23:42.700 That was actually a very well-written document. 00:23:42.960 --> 00:23:45.697 So this is complex. There's not against and for. 00:23:45.780 --> 00:23:52.020 But again, when it comes to things like the chips in China, yeah, we disagree 00:23:52.320 --> 00:23:56.670 when it comes to putting a moratorium on state deregulation. 00:23:56.910 --> 00:23:59.850 Yeah, we disagree because we disagree on the merits. 00:24:00.030 --> 00:24:03.390 It's not about liking or disliking a particular person. 00:24:03.660 --> 00:24:07.657 I don't I don't think that that that kind of thinking is going to get us out 00:24:07.740 --> 00:24:11.707 of this situation that we're in. We have to think based on substance. 00:24:11.790 --> 00:24:14.227 Are you going to see him when he comes to Davos tomorrow? 00:24:14.310 --> 00:24:17.047 I may. I think that would be an interesting 00:24:17.130 --> 00:24:19.987 change. One final last question. 00:24:20.070 --> 00:24:22.387 What happens to entropic? What happens? 00:24:22.470 --> 00:24:27.367 Do you have a I think what since we have supposedly a 350 billion implied 00:24:27.450 --> 00:24:30.090 valuation, would you consider an IPO this year? 00:24:30.420 --> 00:24:35.437 Look, I mean, you know, I we are we are most focused on, you know, just trying 00:24:35.520 --> 00:24:39.367 to make the best models and just trying to, you know, build products on top of 00:24:39.450 --> 00:24:43.297 those models and just trying to, you know, sell those models to enterprises 00:24:43.380 --> 00:24:48.067 in ways that are useful, you know, their their large capital needs for that for 00:24:48.150 --> 00:24:50.587 the for the field. So that that should be considered. 00:24:50.670 --> 00:24:52.720 But, you know, that's where our focus is. 00:24:53.040 --> 00:24:55.640 That implies that you do need a lot of money though. 00:24:56.130 --> 00:24:59.330 I mean, you don't need to ask me to know that this is a capital. 00:25:00.840 --> 00:25:04.837 So an IPO isn't completely out of the question and it's never completely out 00:25:04.920 --> 00:25:07.207 of the question. Those the IPO and Donald Trump with both 00:25:07.290 --> 00:25:12.757 possibles in your life. Derek Derek, thank you very much for 00:25:12.840 --> 00:25:15.720 talking to us. We now know all about mechanistic 00:25:16.320 --> 00:25:20.137 interoperability and we've had a view inside your brain and it's not a bad 00:25:20.220 --> 00:25:22.520 one. Thank you. Thank you for having me, John.