WEBVTT 00:00:16.560 --> 00:00:19.960 Hello everyone. 00:00:21.360 --> 00:00:23.517 Welcome to lunch. Not for you and me, 00:00:23.600 --> 00:00:27.677 but welcome to lunch. Um, 00:00:27.760 --> 00:00:29.917 uh, I think we've had a, uh, a great day 00:00:30.000 --> 00:00:33.357 so far and, uh, and I'm honored to have 00:00:33.440 --> 00:00:36.957 with me, uh, Sam Alman, uh, joining us 00:00:37.040 --> 00:00:40.237 for our discussion about, 00:00:40.320 --> 00:00:43.197 uh, AI and and where we're going from 00:00:43.280 --> 00:00:45.837 here. So, um, 00:00:45.920 --> 00:00:47.917 so thanks for thanks for joining us. You 00:00:48.000 --> 00:00:50.637 have really the greatest countries in 00:00:50.720 --> 00:00:52.957 the world all around you. So would you 00:00:53.040 --> 00:00:55.997 mind giving us a little history like how 00:00:56.080 --> 00:00:58.557 do how do you go from being Sam Alman to 00:00:58.640 --> 00:01:03.117 being the head of AI and really at the frontier like sort of walk us through 00:01:03.200 --> 00:01:05.277 how you got from there to here? 00:01:05.360 --> 00:01:08.317 >> First of all, thank you very much for having me. Um as you said, this is 00:01:08.400 --> 00:01:11.597 greatest countries in the world and it's it's a great honor to get to address all 00:01:11.680 --> 00:01:14.957 of you at the same time. Um 00:01:15.040 --> 00:01:19.837 I I was always an AI nerd. I was like a real sci-fi nerd as a kid. I thought AI 00:01:19.920 --> 00:01:23.757 was the coolest thing. I always wanted to work on it. Um, I never like really 00:01:23.840 --> 00:01:27.677 thought I'd actually get to do it. I kind of can't believe this happened to 00:01:27.760 --> 00:01:31.517 me, but I I always I always thought it was awesome. I went 00:01:31.600 --> 00:01:33.277 to school to study it at the time. 00:01:33.360 --> 00:01:37.117 Nothing was working and I kind of gave up on it and I went on this side quest 00:01:37.200 --> 00:01:38.637 of entrepreneurship, which was also 00:01:38.720 --> 00:01:43.597 great. Um, and then in 2012, 00:01:43.680 --> 00:01:46.637 um, everything changed. None of us paid 00:01:46.720 --> 00:01:50.957 enough attention, but there was a a real watershed moment where all of a sudden 00:01:51.040 --> 00:01:55.277 deep learning started to work and it got better with more compute. And I paid a 00:01:55.360 --> 00:01:59.117 lot of attention to that for the next couple of years. And by the time 2015 00:01:59.200 --> 00:02:03.357 rolled around, I was like, hey, this is really this is going to happen. Like we 00:02:03.440 --> 00:02:06.797 got to take we got to try this could be like the most important thing ever. Uh 00:02:06.880 --> 00:02:08.237 and so we thought for a while about what 00:02:08.320 --> 00:02:11.837 to do and started OpenAI. Uh and 00:02:11.920 --> 00:02:17.357 >> so what year did you start OpenAI? uh the very end of 2015 um we announced it 00:02:17.440 --> 00:02:18.797 and then the first day of work was like 00:02:18.880 --> 00:02:21.997 January of 2016. 00:02:22.080 --> 00:02:25.277 >> So all right so you start in uh in in 00:02:25.360 --> 00:02:27.997 2016 and and what are your like what's 00:02:28.080 --> 00:02:32.717 the big goal that you start with and then how does that evolve? 00:02:32.800 --> 00:02:34.317 So we knew that there was this magic 00:02:34.400 --> 00:02:36.397 piece of technology. We knew that we 00:02:36.480 --> 00:02:40.237 could apply compute 00:02:40.320 --> 00:02:42.557 to a problem and to learn it. And we 00:02:42.640 --> 00:02:47.517 knew that the more compute we used, the smarter and more capable the system got. 00:02:47.600 --> 00:02:53.277 We didn't really know anything else. We didn't know how we were going to um 00:02:53.360 --> 00:02:55.597 what kind of system we were going to build. This was way before the idea of 00:02:55.680 --> 00:02:57.517 language models. We were trying to train 00:02:57.600 --> 00:03:02.797 a robot hand to do a Rubik's cube. Um we were trying to, you know, show that we 00:03:02.880 --> 00:03:06.237 could play video games. And we were really developing the algorithms and 00:03:06.320 --> 00:03:08.397 trying to understand how this amazing 00:03:08.480 --> 00:03:11.357 property of the universe worked. And and 00:03:11.440 --> 00:03:14.557 then not until about 2018 00:03:14.640 --> 00:03:17.117 did we start making progress on 00:03:17.200 --> 00:03:19.037 unsupervised learning. And then the 00:03:19.120 --> 00:03:23.677 first GPT1 model and not until 2020 did 00:03:23.760 --> 00:03:26.157 we train GPT3 which was the first model 00:03:26.240 --> 00:03:30.877 that I think actually was over some threshold of of usability. even that 00:03:30.960 --> 00:03:34.717 most people in the world looked at this and said, you know, okay, the model can 00:03:34.800 --> 00:03:37.597 like generate a co coherent sentence. 00:03:37.680 --> 00:03:39.277 That was really the limit at the time. 00:03:39.360 --> 00:03:42.077 Kind of impressive, but like what is this going to what is this going to 00:03:42.160 --> 00:03:44.797 mean? And why is why is this going to 00:03:44.880 --> 00:03:50.237 like matter to the economy or society or anything else? Um, but we had discovered 00:03:50.320 --> 00:03:53.837 by this point these scaling laws which said, well, if we could, you know, apply 00:03:53.920 --> 00:03:55.997 like a trillion dollars of compute to 00:03:56.080 --> 00:04:01.037 this idea, something quite amazing would happen. And people said, "Well, you 00:04:01.120 --> 00:04:03.277 can't apply a trillion dollars to compute, so it doesn't matter." And we 00:04:03.360 --> 00:04:07.517 sort of said, "Why not?" Like, let's try. 00:04:07.600 --> 00:04:09.277 So, 00:04:09.360 --> 00:04:12.477 you've got the idea and and it's moving 00:04:12.560 --> 00:04:16.157 forward, but when when does it come 00:04:16.240 --> 00:04:19.197 together that it's really bottling 00:04:19.280 --> 00:04:23.917 knowledge? Like, h how does it 00:04:24.000 --> 00:04:27.997 different people at OpenAI would give you different answers to this, but I 00:04:28.080 --> 00:04:30.797 would say Mo the kind of consensus to 00:04:30.880 --> 00:04:37.160 the degree there is one was the time of GPT4. This was like 00:04:37.920 --> 00:04:40.317 I believe this was like March of 2023 00:04:40.400 --> 00:04:44.957 but I may be wrong on the timeline there. Um we had that model for maybe 00:04:45.040 --> 00:04:48.317 eight months before it launched. that we had had time internally to really get to 00:04:48.400 --> 00:04:52.397 know it. And that was the time where people even some of the people 00:04:52.480 --> 00:04:53.917 internally had been skeptics of this 00:04:54.000 --> 00:04:57.597 approach would looked at it and said 00:04:57.680 --> 00:05:02.077 this is going to go very far. Um and it 00:05:02.160 --> 00:05:06.157 hasn't been that long since that moment you know three and a half years now. uh 00:05:06.240 --> 00:05:07.757 and 00:05:07.840 --> 00:05:12.957 and the fact that we now have a system 00:05:13.040 --> 00:05:15.757 uh and we're launching our new model soon which is another step forward and 00:05:15.840 --> 00:05:20.397 there be many more step forwards coming soon after that. Um but that it is so 00:05:20.480 --> 00:05:22.077 capable that it is discovering new 00:05:22.160 --> 00:05:25.117 knowledge doing science able to add huge 00:05:25.200 --> 00:05:27.917 economic value create entire comp pieces 00:05:28.000 --> 00:05:32.637 of complex software. Um this is 00:05:32.720 --> 00:05:34.717 this is like a very crazy moment. I 00:05:34.800 --> 00:05:37.917 think it hit us. Yeah. Over a period of 00:05:38.000 --> 00:05:40.317 time, but centered around this GPT4 00:05:40.400 --> 00:05:42.877 period. 00:05:42.960 --> 00:05:46.397 >> So you understand that 00:05:46.480 --> 00:05:48.957 you can sort of capture education, 00:05:49.040 --> 00:05:53.837 right? You capture knowledge. Um how how 00:05:53.920 --> 00:06:00.680 should the the G20 economies think about this? 00:06:02.320 --> 00:06:06.077 I I don't want to overstate things, but I'm like very excited about our new 00:06:06.160 --> 00:06:08.157 model, so forgive me if I get a little 00:06:08.240 --> 00:06:13.640 bit too enthusiastic here. Um, I 00:06:14.480 --> 00:06:18.717 I think this, for example, will be the greatest boom in entrepreneurship and 00:06:18.800 --> 00:06:20.237 the creation and development of small 00:06:20.320 --> 00:06:23.277 businesses um the world has ever seen. 00:06:23.360 --> 00:06:26.077 If you have a great idea, but you didn't 00:06:26.160 --> 00:06:28.957 have the resources or the ability to get 00:06:29.040 --> 00:06:33.517 enough capital together or hire enough experts to work on it or to figure out 00:06:33.600 --> 00:06:37.277 what it's going to take to invent a new piece of science or manage a complex 00:06:37.360 --> 00:06:40.317 supply chain, you can do that with AI 00:06:40.400 --> 00:06:44.317 now. Um, and I remember after chatbt 00:06:44.400 --> 00:06:48.637 launched, one of the stories that has stuck with me the most from that moment 00:06:48.720 --> 00:06:50.957 was a guy that managed a laundromat. 00:06:51.040 --> 00:06:53.997 Apparently, laundromats are usually good businesses. Wouldn't have been obvious 00:06:54.080 --> 00:06:57.597 to me. Apparently, they are. Um, but this one was not going well. And he was 00:06:57.680 --> 00:07:00.717 using chatbt for everything. Um, to make 00:07:00.800 --> 00:07:02.877 his marketing materials, to, you know, 00:07:02.960 --> 00:07:08.157 manage give him advice on legal contracts, to manage like relationships 00:07:08.240 --> 00:07:09.837 with vendors, this long list of stuff. 00:07:09.920 --> 00:07:11.677 He he was really the first person I met 00:07:11.760 --> 00:07:15.277 that had figured out how to use AI to 00:07:15.360 --> 00:07:19.757 effectively run a business. And this was amazing to me at the time. Now, I know 00:07:19.840 --> 00:07:23.197 you all have heard stories about this a bunch and it's no longer surprising, but 00:07:23.280 --> 00:07:26.077 in 2023, the fact that someone was using 00:07:26.160 --> 00:07:30.717 Chat GBT, broken and embarrassing though it was at the time, to let them manage a 00:07:30.800 --> 00:07:31.997 business that otherwise they couldn't 00:07:32.080 --> 00:07:35.277 have done was was like quite remarkable. 00:07:35.360 --> 00:07:39.117 Um, in 2025, we saw a real acceleration 00:07:39.200 --> 00:07:44.637 of this with coding agents and and white collar work agents. Um, I think we'll 00:07:44.720 --> 00:07:47.677 see a third thing soon where we'll have these long-term persistent agents that 00:07:47.760 --> 00:07:51.997 can work like a real virtual collaborator constantly helping. But, 00:07:52.080 --> 00:07:54.317 um, 00:07:54.400 --> 00:07:57.437 it it's it's it's amazing to me what is 00:07:57.520 --> 00:08:01.037 happening with the boom in 00:08:01.120 --> 00:08:02.877 entrepreneurship and business creation 00:08:02.960 --> 00:08:06.797 because of AI. If I were a country, um, 00:08:06.880 --> 00:08:10.317 you know, many of you are going to take very different approaches to how you 00:08:10.400 --> 00:08:14.477 regulate AI, how you welcome it into your country, how you use it. But for 00:08:14.560 --> 00:08:18.077 sure, I think it is non-negotiable that you have to use it. The economic growth 00:08:18.160 --> 00:08:23.357 and benefit to people that can come from this. Um, the value to a country is too 00:08:23.440 --> 00:08:26.237 high to ignore. uh there will be different ways people try to do this but 00:08:26.320 --> 00:08:28.717 this is like you know the smartest 00:08:28.800 --> 00:08:32.317 people in the world uh decided to move 00:08:32.400 --> 00:08:35.677 to your country to learn your language 00:08:35.760 --> 00:08:40.557 um to work very hard on whatever problem you'd like um to go discover new things 00:08:40.640 --> 00:08:44.477 that you need and you can put as much of this to work as you want some of you 00:08:44.560 --> 00:08:47.357 will choose to build data centers some of you will choose to rent data centers 00:08:47.440 --> 00:08:49.677 from other people because you don't want them in your country I think all that's 00:08:49.760 --> 00:08:52.717 fine um but we want to help we want to 00:08:52.800 --> 00:08:55.437 support not just our company, our whole 00:08:55.520 --> 00:08:58.317 industry like we want to bring this 00:08:58.400 --> 00:09:01.837 incredible magic of, 00:09:01.920 --> 00:09:06.237 you know, intelligence in a bottle, education in a bottle um to all of you. 00:09:06.320 --> 00:09:08.477 I think this is like this is an 00:09:08.560 --> 00:09:11.197 incredible moment for the world. What 00:09:11.280 --> 00:09:17.357 people can do with this and the economic 00:09:17.440 --> 00:09:19.837 boom that will come with it, the sort of 00:09:19.920 --> 00:09:22.317 individual increased quality of life. 00:09:22.400 --> 00:09:26.957 People talk about, you know, medical care and what that means, education, 00:09:27.040 --> 00:09:29.037 what that means, services that only very 00:09:29.120 --> 00:09:34.237 rich people used to have access to, now anybody can have with AI. Um, and the 00:09:34.320 --> 00:09:37.997 ability for people to just come up with ideas and get them to fruition so 00:09:38.080 --> 00:09:40.877 quickly, uh, is going to be incredible. 00:09:40.960 --> 00:09:45.357 Do you do you think that um 00:09:45.440 --> 00:09:49.917 your company can create the context 00:09:50.000 --> 00:09:53.437 um that people can use to create uh 00:09:53.520 --> 00:09:55.357 their outcomes or do you think there'll 00:09:55.440 --> 00:09:58.797 be an industry sort of surrounding you 00:09:58.880 --> 00:10:02.477 that will contextualize it? Because if you think about it, you say the the 00:10:02.560 --> 00:10:05.677 smartest people in the world move in to my neighborhood and they're willing to 00:10:05.760 --> 00:10:09.677 work for me. But if I don't know what to ask them to do, that's not really a fun 00:10:09.760 --> 00:10:12.797 dinner party having the smartest people in the world sitting with you and they 00:10:12.880 --> 00:10:14.637 can't really have a conversation, right? 00:10:14.720 --> 00:10:18.157 So, how do we how do we think about context that that's not going to come 00:10:18.240 --> 00:10:19.597 from us and I don't think that should. 00:10:19.680 --> 00:10:22.797 Um we are we will provide this this 00:10:22.880 --> 00:10:25.837 engine you know we will provide this 00:10:25.920 --> 00:10:28.637 incredible 00:10:28.720 --> 00:10:33.757 commodity of AI that can you can use for whatever you want but the people of the 00:10:33.840 --> 00:10:36.797 world and I have complete trust in the people of the world's creativity and 00:10:36.880 --> 00:10:39.437 ambition um will have to figure out what 00:10:39.520 --> 00:10:42.317 to ask it to do what ideas to to to 00:10:42.400 --> 00:10:46.797 throw this at what what problems and what context in a particular society or 00:10:46.880 --> 00:10:51.437 company is important to understand what to Um so that won't be our role. I don't 00:10:51.520 --> 00:10:55.517 think we'd be good at it even if it were. Um but we will be able to enable 00:10:55.600 --> 00:10:58.477 everyone around the world to bring that context and those questions and those 00:10:58.560 --> 00:11:03.357 ideas. So should 00:11:03.440 --> 00:11:05.117 should the governments of the world 00:11:05.200 --> 00:11:08.637 think of AI 00:11:08.720 --> 00:11:12.637 and companies like your company differently 00:11:12.720 --> 00:11:14.877 than they thought about prior 00:11:14.960 --> 00:11:18.077 technology? Meaning the if if what 00:11:18.160 --> 00:11:22.237 you're bringing them is, you know, the smartest people in the world, an 00:11:22.320 --> 00:11:24.157 educated 00:11:24.240 --> 00:11:26.877 capacity, then if they look at that as 00:11:26.960 --> 00:11:30.317 technology and they uh keep it out of 00:11:30.400 --> 00:11:32.797 their country or they constrain it, are 00:11:32.880 --> 00:11:35.197 they are they constraining their 00:11:35.280 --> 00:11:40.120 capacity to think and grow and build? 00:11:40.880 --> 00:11:44.957 I am always tempted to say this time is different. This is really special. You 00:11:45.040 --> 00:11:49.837 know, this is and I suspect 00:11:49.920 --> 00:11:51.757 well to zoom out even further than that 00:11:51.840 --> 00:11:56.237 as a kid my understanding of 00:11:56.320 --> 00:12:00.157 the world was that there were these few punctuated moments in history each of 00:12:00.240 --> 00:12:04.237 which was a very different thing. You had um the agricultural revolution and 00:12:04.320 --> 00:12:07.757 then you had the industrial revolution and then you had the computer revolution 00:12:07.840 --> 00:12:11.837 and these were all completely different unrelated things and there were not that 00:12:11.920 --> 00:12:16.077 many lessons to learn from them and they were best understood in isolation and 00:12:16.160 --> 00:12:17.837 even if you zoomed into one pick the 00:12:17.920 --> 00:12:20.157 industrial revolution um the right way 00:12:20.240 --> 00:12:24.317 to understand that was um 20 or 30 00:12:24.400 --> 00:12:27.597 important pieces of technology that were all developed over a period of a couple 00:12:27.680 --> 00:12:29.037 hundred years that came together in this 00:12:29.120 --> 00:12:32.317 important way. uh 00:12:32.400 --> 00:12:35.517 as an adult I look at it as you know it 00:12:35.600 --> 00:12:39.517 has been a single revolution a single exponential of technology. Each 00:12:39.600 --> 00:12:43.597 generation has built on the scaffolding of the generations before put a new 00:12:43.680 --> 00:12:47.197 brick in place. People have been able to see or layer of scaffolding to stick 00:12:47.280 --> 00:12:50.157 with the analogy. People have been able to climb and see a little higher and do 00:12:50.240 --> 00:12:54.557 the new thing. Um, if you think about what it took to get us to this point 00:12:54.640 --> 00:12:58.797 where we can build AI, um, it was all of those previous revolutions uh, stacked 00:12:58.880 --> 00:13:03.597 together. And it was the collective discovery of science throughout society 00:13:03.680 --> 00:13:05.357 and the global economy that we all 00:13:05.440 --> 00:13:08.397 contribute to that made this incredible 00:13:08.480 --> 00:13:13.517 thing possible and now happen. And it is tempting whenever you're in the 00:13:13.600 --> 00:13:15.917 moment to say this is the last revolution. This is the biggest 00:13:16.000 --> 00:13:19.437 revolution. There will be more after this. There will be uh bigger things 00:13:19.520 --> 00:13:22.797 after this. people will use AI to invent whatever thing that comes next. It's 00:13:22.880 --> 00:13:24.317 probably very difficult for us to 00:13:24.400 --> 00:13:28.797 imagine right now. Um and 00:13:28.880 --> 00:13:33.437 and I think the right answer is to look at the history of technology as one 00:13:33.520 --> 00:13:36.717 continuous exponential. The history of the economy and society as one 00:13:36.800 --> 00:13:40.957 continuous exponential. Um and I really mean that. It's not just the technology. 00:13:41.040 --> 00:13:44.877 It's this social fabric and the kind of complex structure 00:13:44.960 --> 00:13:47.757 of the world that allows this technology to flourish and people to figure out 00:13:47.840 --> 00:13:52.317 what to do with it. And you have to embrace it continuously all the way 00:13:52.400 --> 00:13:55.517 through. If you miss out on any part of it, um you're not ready for the next 00:13:55.600 --> 00:13:58.797 thing that will come. If you if you 00:13:58.880 --> 00:14:02.877 don't enable your people and your businesses with AI, uh they will not 00:14:02.960 --> 00:14:06.317 invent the next level of the scaffolding. But there will be more 00:14:06.400 --> 00:14:10.397 things and I think it is tempting to say this one is different and it clearly is 00:14:10.480 --> 00:14:14.877 really important. Um but I would view it as a continuum. 00:14:14.960 --> 00:14:16.717 So 00:14:16.800 --> 00:14:19.837 we all read about the the various 00:14:19.920 --> 00:14:22.877 companies at the frontier. Can you can 00:14:22.960 --> 00:14:24.877 you sort of differentiate them for us? 00:14:24.960 --> 00:14:27.437 What what do you what do you view Open 00:14:27.520 --> 00:14:30.797 AI as best at? What what's the you know 00:14:30.880 --> 00:14:33.757 your DNA if you will that separates you 00:14:33.840 --> 00:14:38.157 from the others and what are you great at? And just give us that sort of from 00:14:38.240 --> 00:14:43.517 your perspective. Um 00:14:43.600 --> 00:14:46.157 I think from a from what matters to this 00:14:46.240 --> 00:14:49.197 room um we are the pragmatic centrists. 00:14:49.280 --> 00:14:55.037 We we realize like we want to be we want to be accurate in how we talk 00:14:55.120 --> 00:14:57.197 about what we see in the future. We don't want to fall into the trap of 00:14:57.280 --> 00:15:00.637 blind optimism. We don't want to fall into the trap of dumerism. This is 00:15:00.720 --> 00:15:04.397 clearly complex. There's clearly a combination of both excitement and 00:15:04.480 --> 00:15:07.037 anxiety of transformation and risks that 00:15:07.120 --> 00:15:11.997 the world has to navigate. Um but 00:15:12.080 --> 00:15:14.717 by getting pulled into either extreme we 00:15:14.800 --> 00:15:16.237 are not going to successfully figure out 00:15:16.320 --> 00:15:19.037 the challenges in front of us. Um we 00:15:19.120 --> 00:15:23.917 want to be a reliable partner. Uh we want like we really do view what we are 00:15:24.000 --> 00:15:27.677 doing as a platform as a service to the world. We realize that different people 00:15:27.760 --> 00:15:32.317 around the world are going to use it very differently and we want to put 00:15:32.400 --> 00:15:37.197 sufficient constraints on our system to avoid catastrophic risks. But we really 00:15:37.280 --> 00:15:39.837 do want to respect all of your sovereignity and that you're going to 00:15:39.920 --> 00:15:44.237 use it in very different ways. And we personally may not always like it. Um, 00:15:44.320 --> 00:15:48.637 but I don't like everything that people do with electricity either. And I still 00:15:48.720 --> 00:15:50.157 think everybody should get to have it. 00:15:50.240 --> 00:15:54.200 Um, so 00:15:54.560 --> 00:15:58.957 oh and then also I would say like we want the whole global economy to 00:15:59.040 --> 00:16:02.797 succeed. We do not want to be the only company. Um, we do not want to suck up 00:16:02.880 --> 00:16:06.077 all the value. We do not think that is good for global stability. nor do we 00:16:06.160 --> 00:16:09.277 think that's possible for the way that people need to flourish and people need 00:16:09.360 --> 00:16:12.397 to stay at the center of this. Oh, I'll get to that one more thing after this. 00:16:12.480 --> 00:16:16.717 Um, and we want to like enable every 00:16:16.800 --> 00:16:20.317 entrepreneur, every government, every big company in the world to thrive with 00:16:20.400 --> 00:16:24.397 this technology. Uh, the last thing I'll say that I think differentiates us from 00:16:24.480 --> 00:16:27.997 some of the other companies is we are very firmly and very proudly on team 00:16:28.080 --> 00:16:30.557 humanity. Um, we would like to build 00:16:30.640 --> 00:16:32.477 tools for people. We would like people's 00:16:32.560 --> 00:16:35.597 lives to be better. Um we think that 00:16:35.680 --> 00:16:38.957 people need to be at the center of the 00:16:39.040 --> 00:16:41.197 economy and society and global 00:16:41.280 --> 00:16:44.317 decision-m um and that people should run 00:16:44.400 --> 00:16:46.557 the future in AI. 00:16:46.640 --> 00:16:51.517 >> So it it seems that entrepreneurship 00:16:51.600 --> 00:16:53.917 should be really the fire starter of all 00:16:54.000 --> 00:16:56.797 this, right? You're are you the you know 00:16:56.880 --> 00:16:59.197 the the kindling of it? I mean maybe 00:16:59.280 --> 00:17:02.397 sort of tell me how entrepreneurship and 00:17:02.480 --> 00:17:06.957 and uh open AI sort of work together. 00:17:07.040 --> 00:17:09.197 Before OpenAI, I used to run a startup 00:17:09.280 --> 00:17:14.877 accelerator and I had always been interested, not always, after I came to 00:17:14.960 --> 00:17:18.317 college, I'd always been interested in entrepreneurship. And then having had 00:17:18.400 --> 00:17:20.477 this incredible front row seat, I really 00:17:20.560 --> 00:17:24.237 became convinced that entrepreneurship 00:17:24.320 --> 00:17:28.317 was maybe the most important 00:17:28.400 --> 00:17:31.117 novel force in all of capitalism. Um, 00:17:31.200 --> 00:17:35.037 and I fell in love with it. and 00:17:35.120 --> 00:17:37.757 watching what new companies were able to 00:17:37.840 --> 00:17:41.917 do and watching how 00:17:42.000 --> 00:17:46.237 large of a percentage of the new ideas that mattered in the tech industry over 00:17:46.320 --> 00:17:49.837 the last you know 20 years or whatever came from startups and not big 00:17:49.920 --> 00:17:52.877 companies. Um now having seen us in 00:17:52.960 --> 00:17:55.197 other industries I am a huge believer in 00:17:55.280 --> 00:18:00.120 entrepreneurship. I also think that, 00:18:00.320 --> 00:18:02.957 you know, like I talked to a lot of people around the world and they would 00:18:03.040 --> 00:18:05.997 much rather run a great lifestyle business than go work as like a 00:18:06.080 --> 00:18:09.517 corporate cog and I think that's great too. Like I think that's just we should 00:18:09.600 --> 00:18:14.557 enable that. Um, one of the things I got excited about early with AI, of course 00:18:14.640 --> 00:18:16.477 AI is going to transform big companies 00:18:16.560 --> 00:18:17.997 and of course AI is going to transform 00:18:18.080 --> 00:18:21.997 governments, but what AI can do 00:18:22.080 --> 00:18:25.037 for entrepreneurs and for people that 00:18:25.120 --> 00:18:27.517 want to take an idea and bring it into 00:18:27.600 --> 00:18:32.237 the world quickly um is really quite amazing. When I was running the startup 00:18:32.320 --> 00:18:35.597 accelerator, it lasted for three months and the work that 00:18:35.680 --> 00:18:39.517 >> it lasted for three months. You were in it for three months and the work that we 00:18:39.600 --> 00:18:41.437 expected a startup to accomplish in 00:18:41.520 --> 00:18:43.917 three months is now probably doable in 00:18:44.000 --> 00:18:47.837 like 17 minutes with codeex. This is 00:18:47.920 --> 00:18:51.837 just amazing. The world is not caught up with what this is going to mean. Not 00:18:51.920 --> 00:18:56.477 just for the economic value that gets created from that, but how quickly 00:18:56.560 --> 00:18:59.917 people can test ideas, build things, get feedback from real customers. We're all 00:19:00.000 --> 00:19:02.877 just going to get much better stuff. 00:19:02.960 --> 00:19:06.477 So, you use the term uh I want to sort 00:19:06.560 --> 00:19:10.637 of flush it out a bit a bit non-negotiable. 00:19:10.720 --> 00:19:13.357 Can you maybe uh help everybody 00:19:13.440 --> 00:19:18.317 understand what you mean about that and and what that means for each of uh our 00:19:18.400 --> 00:19:23.400 partner ministers who are here today with them? 00:19:24.720 --> 00:19:26.797 >> Again, I'm biased. I'm trying to talk my 00:19:26.880 --> 00:19:29.437 book here. Forgive me for this analogy. 00:19:29.520 --> 00:19:32.877 Um, I think it would be approximately as 00:19:32.960 --> 00:19:37.437 bad of an idea to say we're not gonna have AI in our country as it was to say 00:19:37.520 --> 00:19:40.637 we're not going to have electricity in our country, you know, back 100 plus 00:19:40.720 --> 00:19:44.797 years ago. I think this is something at that level. Um, this is something that 00:19:44.880 --> 00:19:47.437 every person and every business and 00:19:47.520 --> 00:19:49.517 really the entire 00:19:49.600 --> 00:19:53.117 machine of society is going to need to 00:19:53.200 --> 00:19:56.317 deliver the quality of life that your 00:19:56.400 --> 00:20:00.957 citizens should demand. Now, electricity also came with some 00:20:01.040 --> 00:20:04.317 complexities, and it took a little while for us to figure out how to make it 00:20:04.400 --> 00:20:08.157 safe, but we made fairly rapid progress as a society to make it generally pretty 00:20:08.240 --> 00:20:12.877 safe, even though it can do a lot of bad things. Um, and at this point, we barely 00:20:12.960 --> 00:20:16.157 think about electricity. You know, we turn these lights on, we expect them to 00:20:16.240 --> 00:20:19.917 work. You know, if this the power shut off and I had to scream because this 00:20:20.000 --> 00:20:24.077 microphone was not working, like I would be annoyed about that. You 00:20:24.160 --> 00:20:26.477 I won't go through all the examples. I see all the phones on the table. Like 00:20:26.560 --> 00:20:30.237 you just it would be hard for society to function well without electricity at 00:20:30.320 --> 00:20:33.677 this point. But we don't think about it much. We think about you know there's 00:20:33.760 --> 00:20:37.277 light at night. That's great. We can hear these people speak in the room 00:20:37.360 --> 00:20:39.517 without them having to yell. That's nice. Our phones can do these amazing 00:20:39.600 --> 00:20:41.117 things for us. We care about our phones. 00:20:41.200 --> 00:20:44.077 We don't think too much about the electricity that gets generated to power 00:20:44.160 --> 00:20:46.637 them. Um and I think the same thing is 00:20:46.720 --> 00:20:49.517 going to happen with AI. We won't, my 00:20:49.600 --> 00:20:51.277 prediction is we won't be talking about 00:20:51.360 --> 00:20:54.557 AI as much as we do in a decade from now 00:20:54.640 --> 00:20:56.877 because we'll just expect it everywhere. 00:20:56.960 --> 00:20:59.517 You know, like a kid growing up today 00:20:59.600 --> 00:21:03.277 will never be smarter than AI, but he or 00:21:03.360 --> 00:21:05.597 she will also never 00:21:05.680 --> 00:21:10.477 have understood a world where every product and service that they interact 00:21:10.560 --> 00:21:14.957 with is not really smart and really capable and really helpful. And because 00:21:15.040 --> 00:21:17.597 we'll just be so used to that, we won't think about the AI. will just be like, 00:21:17.680 --> 00:21:19.597 "Yeah, I want my products to be good." 00:21:19.680 --> 00:21:22.797 You know, if I have an idea, I want to be able to like make it right away. I 00:21:22.880 --> 00:21:26.397 want this software to be created instantly for me. Um, if I have a 00:21:26.480 --> 00:21:28.157 disease, like I expect a cure to get 00:21:28.240 --> 00:21:33.597 discovered very quickly. Uh, so I think countries need to need to embrace this. 00:21:33.680 --> 00:21:34.797 Again, there's many ways to do it. 00:21:34.880 --> 00:21:38.397 People will set different rules. People will decide on what they want to build 00:21:38.480 --> 00:21:40.717 versus rent, but I think everyone will 00:21:40.800 --> 00:21:43.917 need the this new commodity. 00:21:44.000 --> 00:21:48.557 If um if we look forward five years and 00:21:48.640 --> 00:21:53.357 and we come back to North Carolina alto together um 00:21:53.440 --> 00:21:59.160 what what will be holding us back do you think in your opinion? 00:22:00.480 --> 00:22:03.837 Well, there are clearly challenges that 00:22:03.920 --> 00:22:06.157 we have to get right along the way. 00:22:06.240 --> 00:22:10.637 There will be specifics. For example, I'm sure cyber security is a top-of- 00:22:10.720 --> 00:22:13.757 mind issue for many of you. Uh we work with many of the governments in the 00:22:13.840 --> 00:22:16.157 room. Uh we work with the state of North 00:22:16.240 --> 00:22:18.477 Carolina. Uh we work with companies big 00:22:18.560 --> 00:22:22.957 and small and we are 00:22:23.040 --> 00:22:27.277 I think we have a big challenge in front of us like to not to not try to sidestep 00:22:27.360 --> 00:22:31.197 this one. I think some things are going to go very wrong with cyber security 00:22:31.280 --> 00:22:35.437 unless people act quite urgently. There will be other challenges in the next 00:22:35.520 --> 00:22:38.637 five years. People talk about uh biocurity and the the things we're going 00:22:38.720 --> 00:22:42.237 to face there. um 00:22:42.320 --> 00:22:44.157 there will be bigger ones yet to come 00:22:44.240 --> 00:22:46.077 and I think if we don't successfully 00:22:46.160 --> 00:22:49.277 navigate those um 00:22:49.360 --> 00:22:51.357 society will not be able to adopt AI to 00:22:51.440 --> 00:22:54.717 the degree or to the speed with which 00:22:54.800 --> 00:22:57.997 people deserve. So getting 00:22:58.080 --> 00:23:02.157 successfully navigating the risks and downsides in front of us is something 00:23:02.240 --> 00:23:05.837 that could go very wrong and I think will set this technology back could set 00:23:05.920 --> 00:23:09.037 this technology back a great deal. Um, 00:23:09.120 --> 00:23:10.477 another thing that could go wrong is too 00:23:10.560 --> 00:23:15.357 much concentration of power. Um, 00:23:15.440 --> 00:23:20.637 AI should be in many ways AI should be an equalizing force in the world. You 00:23:20.720 --> 00:23:23.757 talked about what it means to have like a reset on the educational capacity 00:23:23.840 --> 00:23:27.197 available to everybody. Um, you can also 00:23:27.280 --> 00:23:31.757 see worlds where AI concentrates because 00:23:31.840 --> 00:23:35.277 some people make early investments and others don't and there's, you know, bad 00:23:35.360 --> 00:23:39.837 inequities with comput. uh that'd be bad. So I think if we fail to make AI 00:23:39.920 --> 00:23:42.237 abundant, it ends up being something 00:23:42.320 --> 00:23:47.117 that is very hotly competed over, rich people spend a huge amount of money on 00:23:47.200 --> 00:23:50.957 and um 00:23:51.040 --> 00:23:55.197 that would be a bad outcome. So I think we need to make it abundant so that we 00:23:55.280 --> 00:23:59.197 get the the benefits of lifting everybody up and not much worse than 00:23:59.280 --> 00:24:02.397 inequality. >> Right. A as you said, if if the smartest 00:24:02.480 --> 00:24:04.717 people in the world learned your 00:24:04.800 --> 00:24:07.037 language and all agreed to move to your 00:24:07.120 --> 00:24:09.677 country and work on whatever you wanted 00:24:09.760 --> 00:24:12.477 them to work on, that that has to be the 00:24:12.560 --> 00:24:17.597 most incredible thing. Do you think that 00:24:17.680 --> 00:24:21.277 since you'll have all the smartest people speaking all those languages, 00:24:21.360 --> 00:24:23.037 can't they figure out a way do you think 00:24:23.120 --> 00:24:25.917 over time to deliver the same amount of 00:24:26.000 --> 00:24:29.517 compute more inexpensively without the 00:24:29.600 --> 00:24:34.637 same use of power without the same I mean aren't these the kind of things 00:24:34.720 --> 00:24:39.277 >> that are going to happen that will democratize it by sort of by itself in a 00:24:39.360 --> 00:24:42.957 way >> it it will still require huge investment 00:24:43.040 --> 00:24:47.357 but yes we will continue to make things vastly more efficient and we'll figure 00:24:47.440 --> 00:24:50.877 out how to make more and better chips and we'll figure out how to make smarter 00:24:50.960 --> 00:24:55.037 algorithms. All of that will happen. But to put some numbers on the magnitude of 00:24:55.120 --> 00:24:57.117 how much we we need this to happen. 00:24:57.200 --> 00:24:59.677 Again, I know these numbers in front of me. I'm not going to get this exactly 00:24:59.760 --> 00:25:03.357 right. Um but forgive me for the approximation. 00:25:03.440 --> 00:25:08.157 At the beginning of 2020, um the token leader in the world was an OpenAI 00:25:08.240 --> 00:25:10.237 employee using roughly a 100,000 tokens 00:25:10.320 --> 00:25:12.477 a month. 00:25:12.560 --> 00:25:15.757 Um, in the middle of 2026, 00:25:15.840 --> 00:25:20.157 the token leader that I know of, there may be somebody using more than this at 00:25:20.240 --> 00:25:22.637 outside of OpenAI, um, was using 00:25:22.720 --> 00:25:25.517 hundreds of billions of tokens a month. 00:25:25.600 --> 00:25:30.397 So, you know, a multi-millionx increase in six and a half years. And the average 00:25:30.480 --> 00:25:34.237 person on Earth had gone from using effectively zero tokens to 100,000 00:25:34.320 --> 00:25:37.597 tokens a month. That was the global per capita thing. 00:25:37.680 --> 00:25:40.797 If we just naively project this forward again, which seems like a reasonable 00:25:40.880 --> 00:25:46.317 thing to do to me. Um, and I hate tokens as a unit here. It's a dumb one. No one 00:25:46.400 --> 00:25:49.677 should care, but like let's just take that as shorthand for amount of 00:25:49.760 --> 00:25:52.877 intelligence. Then 00:25:52.960 --> 00:25:57.037 in another six and a half years, the average per capita use of tokens will go 00:25:57.120 --> 00:26:01.757 up to, let's round it down to 100 billion tokens a month. And the person 00:26:01.840 --> 00:26:03.437 using the most in the world will be 00:26:03.520 --> 00:26:06.157 using 00:26:06.240 --> 00:26:08.957 you know 100 quadrillion tokens a month 00:26:09.040 --> 00:26:12.877 uh or one quadrillion 100 trillion like 00:26:12.960 --> 00:26:14.397 >> no I was right 100 trillion tokens a 00:26:14.480 --> 00:26:17.997 month 100 quadrillion tokens a month um 00:26:18.080 --> 00:26:23.277 even with great gains on efficiency we are going to need to build much more 00:26:23.360 --> 00:26:27.597 infrastructure unless we want this to be a highly 00:26:27.680 --> 00:26:30.717 priced commodity which we don't we want this to be abundant We want this to be 00:26:30.800 --> 00:26:31.917 inexpensive. We want people to just be 00:26:32.000 --> 00:26:34.317 able to use this for whatever they need. 00:26:34.400 --> 00:26:37.757 >> So that's the driving force. If we if we 00:26:37.840 --> 00:26:42.397 globally can build enough then it will 00:26:42.480 --> 00:26:45.757 intelligence will be broadly distributed 00:26:45.840 --> 00:26:48.797 uh more democratized across the world. 00:26:48.880 --> 00:26:52.557 Uh the world will be educated or they'll 00:26:52.640 --> 00:26:56.477 have the access to education. Is that sort of the 00:26:56.560 --> 00:26:58.557 >> to to go back to the analogy of 00:26:58.640 --> 00:27:01.357 electricity? Um, there's this amazing 00:27:01.440 --> 00:27:04.877 chart of what it cost to have an hour of 00:27:04.960 --> 00:27:09.917 light at night. And there was a time, one of the 00:27:10.000 --> 00:27:11.837 earliest data points you can find is 00:27:11.920 --> 00:27:14.797 that it it took an average of 5 hours of 00:27:14.880 --> 00:27:16.957 labor at the wage at the time to be able 00:27:17.040 --> 00:27:20.477 to pay for one hour of light at night. 00:27:20.560 --> 00:27:24.557 Electricity came along. It fell. We figured out how to make more 00:27:24.640 --> 00:27:27.677 electricity. So, there was like, you know, the supply and demand change 00:27:27.760 --> 00:27:32.637 curve. It fell again. We figured out how to make vastly more efficient light 00:27:32.720 --> 00:27:36.557 bulbs. We went from incandescent lights to LEDs. A whole bunch of other things 00:27:36.640 --> 00:27:39.437 happened. And now an hour of light at 00:27:39.520 --> 00:27:43.597 night is not something that people think of at all about the cost. I don't even 00:27:43.680 --> 00:27:46.397 know what the number is, but someone in here can do it. It's it's like very 00:27:46.480 --> 00:27:48.397 cheap to run, you know, a five watt bulb 00:27:48.480 --> 00:27:52.797 for an hour. Um the 00:27:52.880 --> 00:27:56.317 if that hadn't happened if we and all of those things together had to happen lots 00:27:56.400 --> 00:27:59.197 of electricity generation huge innovation on the efficiency all of 00:27:59.280 --> 00:28:04.237 these things if that hadn't happened then rich people get light at night and 00:28:04.320 --> 00:28:06.797 they can study more they have a compounding effect they can learn more 00:28:06.880 --> 00:28:09.517 whatever and and you know poor people 00:28:09.600 --> 00:28:13.917 don't so I think it's been very good that energy has become abundant and that 00:28:14.000 --> 00:28:16.157 light at night has become relatively 00:28:16.240 --> 00:28:19.757 like shockingly inexpensive Um, if that 00:28:19.840 --> 00:28:22.317 same thing happens with AI and anybody 00:28:22.400 --> 00:28:24.237 on earth, rich, poor, whatever, can use 00:28:24.320 --> 00:28:29.517 AI to get the abundant services that they need and to express their ideas and 00:28:29.600 --> 00:28:33.677 to start their businesses and whatever else. And we fulfill our mission of 00:28:33.760 --> 00:28:38.477 making AI so inexpensive and so smart that people are just like they think 00:28:38.560 --> 00:28:41.917 about it like they think of the cost of an hour at night of light at night. I 00:28:42.000 --> 00:28:45.437 think that'd be really wonderful. And then this huge reset and democratization 00:28:45.520 --> 00:28:47.917 of education you're talking about can happen. 00:28:48.000 --> 00:28:52.317 So are you you know there are look there 00:28:52.400 --> 00:28:54.877 are doomers out there who worry about 00:28:54.960 --> 00:28:57.597 everything but we know that if you look 00:28:57.680 --> 00:28:59.437 back through the industrial re as you 00:28:59.520 --> 00:29:01.917 said it it always seems to work out 00:29:02.000 --> 00:29:04.797 fine. Are you uh fundamentally 00:29:04.880 --> 00:29:07.517 optimistic to what your products will do 00:29:07.600 --> 00:29:11.117 to the world and will serve the world? 00:29:11.200 --> 00:29:12.877 I I think part of the reason it tends to 00:29:12.960 --> 00:29:16.797 work out fine is people worry and they 00:29:16.880 --> 00:29:19.917 stress and they look at the downsides and they try to address those and they 00:29:20.000 --> 00:29:22.797 try to learn how we avoid accidents. So 00:29:22.880 --> 00:29:25.437 if we go back to an early example even 00:29:25.520 --> 00:29:27.917 before electricity of a technology fire 00:29:28.000 --> 00:29:32.797 was incredible incredible for human evolution incredible for letting us cook 00:29:32.880 --> 00:29:36.077 our food and heat our homes at night and whatever else. It also burnt down whole 00:29:36.160 --> 00:29:40.637 cities a few times and people really worried about that because we don't want 00:29:40.720 --> 00:29:43.037 cities to burn down and we figured out 00:29:43.120 --> 00:29:47.597 fire safety and through a lot of human 00:29:47.680 --> 00:29:49.597 ingenuity um through a lot of good 00:29:49.680 --> 00:29:51.917 policy definitely through learning from 00:29:52.000 --> 00:29:56.637 some very bad accidents we figured out how to make it safer. Now had we just 00:29:56.720 --> 00:30:00.077 said like yeah you know some cities are going to burn down whatever we got other 00:30:00.160 --> 00:30:02.957 problems to work on probably wouldn't have figured that out and that wouldn't 00:30:03.040 --> 00:30:06.957 have been as good. Um, so I think worry 00:30:07.040 --> 00:30:10.397 is warranted. Worry is how we anticipate 00:30:10.480 --> 00:30:13.437 problems and how we from a matter of 00:30:13.520 --> 00:30:17.437 policy and a matter of technology figure out to address them. But I am optimistic 00:30:17.520 --> 00:30:20.317 that we can mitigate the downsides of this technology. And if you look at 00:30:20.400 --> 00:30:24.877 what's happened, you know, if you look at what the doomers were saying 5 years 00:30:24.960 --> 00:30:29.517 ago about the chances of us being able to make AI of this level, of this sort 00:30:29.600 --> 00:30:33.917 of near human, sometimes superhuman capacity level, you know, broadly 00:30:34.000 --> 00:30:37.597 considered safe and reliable. Not perfect. There have been accidents. 00:30:37.680 --> 00:30:40.877 There will be accidents, of course. I don't think most people would have 00:30:40.960 --> 00:30:42.717 thought it was going to go this well as 00:30:42.800 --> 00:30:46.477 it has so far. And I I believe in human 00:30:46.560 --> 00:30:51.117 ingenuity and the ability for to solve very difficult, very complex 00:30:51.200 --> 00:30:54.237 problems. And I think I think we'll keep doing that. 00:30:54.320 --> 00:30:57.437 And I think the economic value and the 00:30:57.520 --> 00:31:02.557 personal value is so huge. Um I met with someone just this morning and told and 00:31:02.640 --> 00:31:05.757 again I'm sure you've all heard stories like this that they wouldn't be alive 00:31:05.840 --> 00:31:08.717 without CHPT. Um and that they had a 00:31:08.800 --> 00:31:12.877 very difficult disease and you know got a diagnosis that doctors weren't able to 00:31:12.960 --> 00:31:14.557 >> give them. We shouldn't take that away. 00:31:14.640 --> 00:31:17.757 >> No. But we need to figure out how to make the technology safe so that people 00:31:17.840 --> 00:31:22.157 can have more of that. >> Great. Sam, thank you for joining us. It 00:31:22.240 --> 00:31:24.717 was really an incredible conversation. 00:31:24.800 --> 00:31:29.277 Um 2:00, so 2 p.m. in the ministerial 00:31:29.360 --> 00:31:32.797 room where we will have uh Alex Karp 00:31:32.880 --> 00:31:36.317 will be our next guest uh speaking to us. 00:31:36.400 --> 00:31:39.517 >> He's a good one. >> He's a good one. But uh 00:31:39.600 --> 00:31:42.717 >> thank you Sam for joining us. We we really appreciate it. Thank you. 00:31:42.800 --> 00:31:49.400 [applause] Thanks. [applause]