WEBVTT 00:00:00.080 --> 00:00:03.197 It seems like your worldview is compatible with somebody who says uh 00:00:03.280 --> 00:00:06.877 we're like 10 years away from a world in which like we're generating trillions. 00:00:06.960 --> 00:00:08.157 >> That's just that's just not my view. 00:00:08.240 --> 00:00:12.477 Yeah, that is that is not my view. Like I I so so I'll like I'll like make 00:00:12.560 --> 00:00:14.957 another prediction. It is hard for me to 00:00:15.040 --> 00:00:17.757 see that that there won't be trillions 00:00:17.840 --> 00:00:20.957 of dollars in revenue before 2030. Um 00:00:21.040 --> 00:00:26.077 like uh I can I can construct a plausible world. It takes maybe three 00:00:26.160 --> 00:00:28.717 years. So that that you know that would be the end of what I think it's 00:00:28.800 --> 00:00:33.517 plausible like in 2028 we get the the real country of geniuses in a data 00:00:33.600 --> 00:00:37.197 center. You know the revenue's been been go you know the revenue has been going 00:00:37.280 --> 00:00:39.757 into the maybe is is in the low hundreds 00:00:39.840 --> 00:00:43.677 of billions by by by by 2028 and and and 00:00:43.760 --> 00:00:47.757 then the country of geniuses accelerates it to trillions you know and and we're 00:00:47.840 --> 00:00:51.117 basically we're basically on the slow end of diffusion. It takes two years to 00:00:51.200 --> 00:00:54.157 get to the trillions. that that that would that that that would be the world 00:00:54.240 --> 00:00:55.677 where it takes until that would be the 00:00:55.760 --> 00:00:58.717 world where it takes until 2030. I I I 00:00:58.800 --> 00:01:01.517 suspect even composing the technical 00:01:01.600 --> 00:01:03.117 exponential and diffusion exponential 00:01:03.200 --> 00:01:05.837 will get there before 2030. So you laid 00:01:05.920 --> 00:01:08.877 out a model where anthropic makes profit 00:01:08.960 --> 00:01:12.957 because it seems like fundamentally we're in a computed world and so it's 00:01:13.040 --> 00:01:14.717 like eventually we keep growing comput. 00:01:14.800 --> 00:01:18.877 No, I think I think the way the profit comes is again and and you know, let's 00:01:18.960 --> 00:01:22.077 let's just abstract the whole industry here. Like we have a you know, let's 00:01:22.160 --> 00:01:25.677 just imagine we're we're in like an economics textbook. We have a small 00:01:25.760 --> 00:01:28.957 number of firms each can invest a 00:01:29.040 --> 00:01:31.757 limited amount in you or or or like each 00:01:31.840 --> 00:01:34.237 can invest some fra fraction in R&D. 00:01:34.320 --> 00:01:36.397 They have some marginal cost to serve. 00:01:36.480 --> 00:01:40.077 the margins on that the profit margin the gross profit margins on that 00:01:40.160 --> 00:01:44.877 marginal cost are like very high because because because inference is efficient 00:01:44.960 --> 00:01:48.557 there's some competition but the models are also differentiated there's some 00:01:48.640 --> 00:01:53.037 there's some um you know companies will compete to push their research budgets 00:01:53.120 --> 00:01:57.437 up but like because there's a small number of players you know we have the 00:01:57.520 --> 00:02:01.037 what is it called in the corno equilibrium I think is what the what the 00:02:01.120 --> 00:02:05.997 small number of firm equilibrium is it the point is it it doesn't equilibrate 00:02:06.080 --> 00:02:10.077 to perfect competition with with with with with with 00:02:10.160 --> 00:02:14.717 zero margins. If there's like three firms, if there's three firms in the 00:02:14.800 --> 00:02:19.197 economy, all are kind of independently behaving behaving rationally, it doesn't 00:02:19.280 --> 00:02:22.717 equilibrate to zero. >> Um, help me understand that because 00:02:22.800 --> 00:02:26.637 right now we do have three leading firms and they're not making profit. Um, and 00:02:26.720 --> 00:02:29.037 so what what uh yeah, what what is 00:02:29.120 --> 00:02:32.557 changing? Yeah. So the the again the 00:02:32.640 --> 00:02:37.757 gross margins right now are very positive. What's happen what what's 00:02:37.840 --> 00:02:41.037 happening is a combination of two things. One is we're still in the 00:02:41.120 --> 00:02:43.677 exponential scale up phase of compute. 00:02:43.760 --> 00:02:46.557 Um so what basically what that means is 00:02:46.640 --> 00:02:49.597 we're training like a model gets trained 00:02:49.680 --> 00:02:51.357 it costs you know let's say a model got 00:02:51.440 --> 00:02:54.637 trained that costs uh a billion last 00:02:54.720 --> 00:02:58.157 year um and then uh this year it 00:02:58.240 --> 00:03:02.237 produced uh $4 billion of revenue and 00:03:02.320 --> 00:03:06.237 cost $1 billion to to uh to to to 00:03:06.320 --> 00:03:10.077 inference from. Um so you know again I'm using stylized number here but you know 00:03:10.160 --> 00:03:13.037 that would be 75% you know gross gross 00:03:13.120 --> 00:03:15.677 gross margins and you know this this 25% 00:03:15.760 --> 00:03:19.517 tax. So that model as a whole makes $2 00:03:19.600 --> 00:03:24.797 billion. Um but at the same time we're spending $10 billion to train the next 00:03:24.880 --> 00:03:28.397 model because there's an exponential scale up and so the company loses money. 00:03:28.480 --> 00:03:32.717 Each model makes money but the company loses money. The equilibrium I'm talking 00:03:32.800 --> 00:03:36.397 about is an equilibrium where we have the country of geniuses. We have the 00:03:36.480 --> 00:03:38.157 country of geniuses in a data center, 00:03:38.240 --> 00:03:42.797 but that that um model training scale up 00:03:42.880 --> 00:03:46.877 has equilibrated more. Maybe maybe it's still it's still going up. We're still 00:03:46.960 --> 00:03:51.437 trying to predict the demand, but it's more it's more um leveled out. If you 00:03:51.520 --> 00:03:55.117 enjoyed this clip, you can watch the full episode here and subscribe for more 00:03:55.200 --> 00:03:58.480 clips. Thanks.