1 00:00:00,160 --> 00:00:03,597 where we're going from here. So, um, 2 00:00:03,680 --> 00:00:05,597 so thanks for thanks for joining us. You 3 00:00:05,680 --> 00:00:08,397 have really the greatest countries in 4 00:00:08,480 --> 00:00:10,717 the world all around you. So, would you 5 00:00:10,800 --> 00:00:13,757 mind giving us a little history like how 6 00:00:13,840 --> 00:00:16,317 do how do you go from being Sam Alman to 7 00:00:16,400 --> 00:00:20,877 being the head of AI and really at the frontier like sort of walk us through 8 00:00:20,960 --> 00:00:23,037 how you got from there to here. 9 00:00:23,120 --> 00:00:26,077 >> First of all, thank you very much for having me. Um, as you said, this is 10 00:00:26,160 --> 00:00:29,037 greatest countries in the world, and it's it's a great honor to get to 11 00:00:29,120 --> 00:00:32,717 address all of you at the same time. Um, 12 00:00:32,800 --> 00:00:37,517 I I was always an AI nerd. I was like a real sci-fi nerd as a kid. I thought AI 13 00:00:37,600 --> 00:00:41,517 was the coolest thing. I always wanted to work on it. Um, I never like really 14 00:00:41,600 --> 00:00:45,357 thought I'd actually get to do it. I kind of can't believe this happened to 15 00:00:45,440 --> 00:00:49,197 me, but I I always I always thought it was awesome. I went 16 00:00:49,280 --> 00:00:52,877 to school to study it. At the time nothing was working and I kind of gave 17 00:00:52,960 --> 00:00:54,717 up on it and I went on this side quest 18 00:00:54,800 --> 00:00:56,397 of entrepreneurship which was also 19 00:00:56,480 --> 00:01:01,357 great. Um and then in 2012 20 00:01:01,440 --> 00:01:04,317 um everything changed. None of us paid 21 00:01:04,400 --> 00:01:08,717 enough attention but there was a a real watershed moment where all of a sudden 22 00:01:08,800 --> 00:01:12,957 deep learning started to work and it got better with more compute. And I paid a 23 00:01:13,040 --> 00:01:16,797 lot of attention to that for the next couple of years. And by the time 2015 24 00:01:16,880 --> 00:01:21,117 rolled around, I was like, "Hey, this is really this is going to happen. Like we 25 00:01:21,200 --> 00:01:24,557 got to take we got to try. This could be like the most important thing ever." Uh, 26 00:01:24,640 --> 00:01:25,997 and so we thought for a while about what 27 00:01:26,080 --> 00:01:29,597 to do and started OpenAI. Uh, and 28 00:01:29,680 --> 00:01:31,837 >> so what year did you start OpenAI? 29 00:01:31,920 --> 00:01:34,717 >> Uh, the very end of 2015. Um, we 30 00:01:34,800 --> 00:01:35,997 announced it and then the first day of 31 00:01:36,080 --> 00:01:39,757 work was like January of 2016. 32 00:01:39,840 --> 00:01:42,957 >> So all right. So you start in uh in in 33 00:01:43,040 --> 00:01:45,757 2016 and and what are your like what's 34 00:01:45,840 --> 00:01:50,477 the big goal that you start with and then how does that evolve? 35 00:01:50,560 --> 00:01:52,077 So we knew that there was this magic 36 00:01:52,160 --> 00:01:54,077 piece of technology. We knew that we 37 00:01:54,160 --> 00:01:57,997 could apply compute 38 00:01:58,080 --> 00:02:00,477 to a problem and to learn it and we knew 39 00:02:00,560 --> 00:02:05,277 that the more compute we used the smarter and more capable the system got. 40 00:02:05,360 --> 00:02:11,037 We didn't really know anything else. We didn't know how we were going to um 41 00:02:11,120 --> 00:02:13,357 what kind of system we were going to build. This was way before the idea of 42 00:02:13,440 --> 00:02:15,197 language models. We were trying to train 43 00:02:15,280 --> 00:02:18,717 a robot hand to do a Rubik's cube. Um we 44 00:02:18,800 --> 00:02:22,237 were trying to you know show that we could play video games. And we were 45 00:02:22,320 --> 00:02:26,077 really developing the algorithms and trying to understand how this amazing 46 00:02:26,160 --> 00:02:29,117 property of the universe worked. And and 47 00:02:29,200 --> 00:02:32,317 then not until about 2018 48 00:02:32,400 --> 00:02:35,037 did we start making progress on on 49 00:02:35,120 --> 00:02:36,957 supervised learning. and then the first 50 00:02:37,040 --> 00:02:41,597 GPT1 model and not until 2020 did we 51 00:02:41,680 --> 00:02:45,517 train GPT3 which was the first model that I think actually was over some 52 00:02:45,600 --> 00:02:49,997 threshold of of usability. Even then most people in the world looked at this 53 00:02:50,080 --> 00:02:52,397 and said you know okay the model can 54 00:02:52,480 --> 00:02:57,037 like generate a co coherent sentence that was really the limit at the time. 55 00:02:57,120 --> 00:03:00,717 Kind of impressive but like what is this going to what is this going to mean and 56 00:03:00,800 --> 00:03:02,957 why is why is this going to like matter 57 00:03:03,040 --> 00:03:04,877 to the economy or society or anything 58 00:03:04,960 --> 00:03:08,317 else? Um but we had discovered by this 59 00:03:08,400 --> 00:03:11,917 point these scaling laws which said well if we could you know apply like a 60 00:03:12,000 --> 00:03:15,197 trillion dollars of compute to this idea 61 00:03:15,280 --> 00:03:19,277 something quite amazing would happen and people said well you can't apply a 62 00:03:19,360 --> 00:03:21,677 trillion dollars of compute so it doesn't matter and we sort of said why 63 00:03:21,760 --> 00:03:25,277 not like let's try 64 00:03:25,360 --> 00:03:27,037 so 65 00:03:27,120 --> 00:03:30,237 you've got the idea and and it's moving 66 00:03:30,320 --> 00:03:33,917 forward but when when does it come 67 00:03:34,000 --> 00:03:36,957 together that It's really bottling 68 00:03:37,040 --> 00:03:41,597 knowledge like h how does it 69 00:03:41,680 --> 00:03:43,677 >> different people at openi would give you 70 00:03:43,760 --> 00:03:45,917 different answers to this but I would 71 00:03:46,000 --> 00:03:48,717 say mo the kind of consensus to the 72 00:03:48,800 --> 00:03:54,920 degree there is one was the time of gpt4 this was like 73 00:03:55,680 --> 00:03:58,077 I believe this was like March of 2023 74 00:03:58,160 --> 00:04:02,957 but I may be wrong on the timeline there um we had that model for maybe eight 75 00:04:03,040 --> 00:04:07,037 months before it launched so we had time internally to really get to know it. And 76 00:04:07,120 --> 00:04:10,877 that was the time where people even some of the people internally had been 77 00:04:10,960 --> 00:04:13,757 skeptics of this approach would looked 78 00:04:13,840 --> 00:04:17,037 at it and said this is going to go very 79 00:04:17,120 --> 00:04:20,637 far. Um and it hasn't been that long 80 00:04:20,720 --> 00:04:22,077 since that moment you know three and a 81 00:04:22,160 --> 00:04:25,517 half years now. uh and 82 00:04:25,600 --> 00:04:30,717 and the fact that we now have a system 83 00:04:30,800 --> 00:04:33,517 uh and we're launching our new model soon which is another step forward and 84 00:04:33,600 --> 00:04:38,157 there be many more step forwards coming soon after that. Um but that it is so 85 00:04:38,240 --> 00:04:39,837 capable that it is discovering new 86 00:04:39,920 --> 00:04:42,877 knowledge doing science able to add huge 87 00:04:42,960 --> 00:04:45,677 economic value create entire comp pieces 88 00:04:45,760 --> 00:04:50,317 of complex software. Um this is 89 00:04:50,400 --> 00:04:54,477 this was like a very crazy moment. I think it hit us. 90 00:04:54,560 --> 00:04:56,157 Yeah. Over a period of time, but 91 00:04:56,240 --> 00:05:00,557 centered around this GPT4 period. 92 00:05:00,640 --> 00:05:04,157 >> So you understand that 93 00:05:04,240 --> 00:05:06,717 you can sort of capture education, 94 00:05:06,800 --> 00:05:11,597 right? You capture knowledge. Um how how 95 00:05:11,680 --> 00:05:18,440 should the the G20 economies think about this? 96 00:05:20,080 --> 00:05:23,757 I I don't want to overstate things, but I'm like very excited about our new 97 00:05:23,840 --> 00:05:25,837 model. So, forgive me if I get a little 98 00:05:25,920 --> 00:05:31,400 bit too enthusiastic here. Um, I 99 00:05:32,160 --> 00:05:36,477 I think this, for example, will be the greatest boom in entrepreneurship and 100 00:05:36,560 --> 00:05:37,997 the creation and development of small 101 00:05:38,080 --> 00:05:41,117 businesses um the world has ever seen. 102 00:05:41,200 --> 00:05:43,757 If you have a great idea, but you didn't 103 00:05:43,840 --> 00:05:46,717 have the resources or the ability to get 104 00:05:46,800 --> 00:05:51,277 enough capital together or hire enough experts to work on it or to figure out 105 00:05:51,360 --> 00:05:54,957 what it's going to take to invent a new piece of science or manage a complex 106 00:05:55,040 --> 00:05:58,077 supply chain, you can do that with AI 107 00:05:58,160 --> 00:06:02,077 now. Um, and I remember after ChatBT 108 00:06:02,160 --> 00:06:06,397 launched, one of the stories that has stuck with me the most from that moment 109 00:06:06,480 --> 00:06:08,637 was a guy that managed a laundromat. 110 00:06:08,720 --> 00:06:11,757 Apparently laundromats are usually good businesses. Wouldn't have been obvious 111 00:06:11,840 --> 00:06:15,597 to me. Apparently they are. Um but this one was not going well. And he was using 112 00:06:15,680 --> 00:06:18,717 chatbt for everything. Um to make his 113 00:06:18,800 --> 00:06:21,677 marketing materials to, you know, manage 114 00:06:21,760 --> 00:06:26,717 give him advice on legal contracts, to manage like relationships with vendors, 115 00:06:26,800 --> 00:06:29,997 this long list of stuff. He he was really the first person I met that had 116 00:06:30,080 --> 00:06:33,037 figured out how to use AI to 117 00:06:33,120 --> 00:06:37,517 effectively run a business. And this was amazing to me at the time. Now, I know 118 00:06:37,600 --> 00:06:40,957 you all have heard stories about this a bunch and it's no longer surprising, but 119 00:06:41,040 --> 00:06:43,837 in 2023, the fact that someone was using 120 00:06:43,920 --> 00:06:48,477 Chat GBT, broken and embarrassing though it was at the time, to let them manage a 121 00:06:48,560 --> 00:06:49,757 business that otherwise they couldn't 122 00:06:49,840 --> 00:06:52,957 have done was was like quite remarkable. 123 00:06:53,040 --> 00:06:56,877 Um, in 2025, we saw a real acceleration 124 00:06:56,960 --> 00:07:02,317 of this with coding agents and and white collar work agents. Um, I think we'll 125 00:07:02,400 --> 00:07:05,437 see a third thing soon where we'll have these long-term persistent agents that 126 00:07:05,520 --> 00:07:09,757 can work like a real virtual collaborator constantly helping. But, 127 00:07:09,840 --> 00:07:12,077 um, 128 00:07:12,160 --> 00:07:15,117 it it's it's it's amazing to me what is 129 00:07:15,200 --> 00:07:18,797 happening with the boom in 130 00:07:18,880 --> 00:07:20,637 entrepreneurship and business creation 131 00:07:20,720 --> 00:07:24,477 because of AI. If I were a country, um, 132 00:07:24,560 --> 00:07:27,997 you know, many of you are going to take very different approaches to how you 133 00:07:28,080 --> 00:07:32,157 regulate AI, how you welcome it into your country, how you use it. But for 134 00:07:32,240 --> 00:07:35,757 sure, I think it is non-negotiable that you have to use it. The economic growth 135 00:07:35,840 --> 00:07:41,037 and benefit to people that can come from this. Um, the value to a country is too 136 00:07:41,120 --> 00:07:43,997 high to ignore. uh there will be different ways people try to do this but 137 00:07:44,080 --> 00:07:46,477 this is like you know the smartest 138 00:07:46,560 --> 00:07:49,997 people in the world uh decided to move 139 00:07:50,080 --> 00:07:53,357 to your country to learn your language 140 00:07:53,440 --> 00:07:58,237 um to work very hard on whatever problem you'd like um to go discover new things 141 00:07:58,320 --> 00:08:02,237 that you need and you can put as much of this to work as you want some of you 142 00:08:02,320 --> 00:08:05,037 will choose to build data centers some of you will choose to rent data centers 143 00:08:05,120 --> 00:08:07,437 from other people because you don't want them in your country I think all that's 144 00:08:07,520 --> 00:08:12,237 fine um but we want to help we want to support not 145 00:08:12,320 --> 00:08:14,157 our company, our whole industry like we 146 00:08:14,240 --> 00:08:19,597 want to bring this incredible magic of, 147 00:08:19,680 --> 00:08:23,997 you know, intelligence in a bottle, education in a bottle um to all of you. 148 00:08:24,080 --> 00:08:26,237 I think this is like this is an 149 00:08:26,320 --> 00:08:28,957 incredible moment for the world. What 150 00:08:29,040 --> 00:08:35,117 people can do with this and the economic 151 00:08:35,200 --> 00:08:37,597 boom that will come with it, the sort of 152 00:08:37,680 --> 00:08:40,077 individual increased quality of life. 153 00:08:40,160 --> 00:08:44,637 People talk about, you know, medical care and what that means, education, 154 00:08:44,720 --> 00:08:46,797 what that means, services that only very 155 00:08:46,880 --> 00:08:51,917 rich people used to have access to, now anybody can have with AI. Um, and the 156 00:08:52,000 --> 00:08:55,677 ability for people to just come up with ideas and get them to fruition so 157 00:08:55,760 --> 00:08:58,637 quickly, uh, is going to be incredible. 158 00:08:58,720 --> 00:09:03,117 Do you do you think that um 159 00:09:03,200 --> 00:09:07,597 your company can create the context 160 00:09:07,680 --> 00:09:11,197 um that people can use to create uh 161 00:09:11,280 --> 00:09:13,117 their outcomes or do you think there'll 162 00:09:13,200 --> 00:09:16,557 be an industry sort of surrounding you 163 00:09:16,640 --> 00:09:20,237 that will contextualize it? Because if you think about it, you say the the 164 00:09:20,320 --> 00:09:23,437 smartest people in the world move in to my neighborhood and they're willing to 165 00:09:23,520 --> 00:09:27,437 work for me. But if I don't know what to ask them to do, that's not really a fun 166 00:09:27,520 --> 00:09:30,557 dinner party having the smartest people in the world sitting with you and they 167 00:09:30,640 --> 00:09:32,397 can't really have a conversation, right? 168 00:09:32,480 --> 00:09:35,917 So, how do we how do we think about context? That that's not going to come 169 00:09:36,000 --> 00:09:37,357 from us and I don't think that should. 170 00:09:37,440 --> 00:09:40,557 Um we are we will provide this this 171 00:09:40,640 --> 00:09:43,597 engine you know we will provide this 172 00:09:43,680 --> 00:09:46,397 incredible 173 00:09:46,480 --> 00:09:51,437 commodity of AI that can you can use for whatever you want but the people of the 174 00:09:51,520 --> 00:09:54,557 world and I have complete trust in the people of the world's creativity and 175 00:09:54,640 --> 00:09:57,197 ambition um will have to figure out what 176 00:09:57,280 --> 00:10:00,077 to ask it to do what ideas to to to 177 00:10:00,160 --> 00:10:04,557 throw this at what what problems and what context in a particular society or 178 00:10:04,640 --> 00:10:09,197 company is important to understand what to Um so that won't be our role. I don't 179 00:10:09,280 --> 00:10:13,277 think we'd be good at it even if it were. Um but we will be able to enable 180 00:10:13,360 --> 00:10:16,157 everyone around the world to bring that context and those questions and those 181 00:10:16,240 --> 00:10:21,037 ideas. So should 182 00:10:21,120 --> 00:10:22,877 should the governments of the world 183 00:10:22,960 --> 00:10:26,397 think of AI 184 00:10:26,480 --> 00:10:30,397 and companies like your company differently 185 00:10:30,480 --> 00:10:32,557 than they thought about prior 186 00:10:32,640 --> 00:10:35,757 technology? Meaning the if if what 187 00:10:35,840 --> 00:10:39,997 you're bringing them is, you know, the smartest people in the world, an 188 00:10:40,080 --> 00:10:41,917 educated 189 00:10:42,000 --> 00:10:44,557 capacity, then if they look at that as 190 00:10:44,640 --> 00:10:48,077 technology and they uh keep it out of 191 00:10:48,160 --> 00:10:50,557 their country or they constrain it, are 192 00:10:50,640 --> 00:10:52,957 they are they constraining their 193 00:10:53,040 --> 00:10:57,880 capacity to think and grow and build? 194 00:10:58,640 --> 00:11:02,717 >> I am always tempted to say this time is different. This is really special. You 195 00:11:02,800 --> 00:11:07,597 know, this is and I suspect 196 00:11:07,680 --> 00:11:09,517 well to zoom out even further than that 197 00:11:09,600 --> 00:11:13,917 as a kid my understanding of 198 00:11:14,000 --> 00:11:17,917 the world was that there were these few punctuated moments in history each of 199 00:11:18,000 --> 00:11:21,917 which was a very different thing. You had um the agricultural revolution and 200 00:11:22,000 --> 00:11:25,437 then you had the industrial revolution and then you had the computer revolution 201 00:11:25,520 --> 00:11:29,597 and these were all completely different unrelated things and there were not that 202 00:11:29,680 --> 00:11:33,757 many lessons to learn from them and they were best understood in isolation and 203 00:11:33,840 --> 00:11:35,597 even if you zoomed into one pick the 204 00:11:35,680 --> 00:11:37,917 industrial revolution um the right way 205 00:11:38,000 --> 00:11:42,077 to understand that was um 20 or 30 206 00:11:42,160 --> 00:11:45,357 important pieces of technology that were all developed over a period of a couple 207 00:11:45,440 --> 00:11:46,797 hundred years that came together in this 208 00:11:46,880 --> 00:11:50,077 important way. uh 209 00:11:50,160 --> 00:11:53,197 as an adult I look at it as you know it 210 00:11:53,280 --> 00:11:57,277 has been a single revolution a single exponential of technology each 211 00:11:57,360 --> 00:12:01,357 generation has built on the scaffolding of the generations before put a new 212 00:12:01,440 --> 00:12:04,957 brick in place people have been able to see or layer of scaffolding to stick 213 00:12:05,040 --> 00:12:07,837 with the analogy people have been able to climb and see a little higher and do 214 00:12:07,920 --> 00:12:12,637 the new thing um if you think about what it took to get us to this point where we 215 00:12:12,720 --> 00:12:16,797 can build AI um it was all of those previous revolutions uh stacked together 216 00:12:16,880 --> 00:12:21,357 and it was the collective discovery of science throughout society 217 00:12:21,440 --> 00:12:23,117 and the global economy that we all 218 00:12:23,200 --> 00:12:26,077 contribute to that made this incredible 219 00:12:26,160 --> 00:12:31,277 thing possible and now happen. And it is tempting whenever you're in the 220 00:12:31,360 --> 00:12:33,597 moment to say this is the last revolution. This is the biggest 221 00:12:33,680 --> 00:12:37,197 revolution. There will be more after this. There will be uh bigger things 222 00:12:37,280 --> 00:12:40,557 after this. People will use AI to invent whatever thing that comes next. It's 223 00:12:40,640 --> 00:12:42,077 probably very difficult for us to 224 00:12:42,160 --> 00:12:46,557 imagine right now. Um and 225 00:12:46,640 --> 00:12:51,117 and I think the right answer is to look at the history of technology as one 226 00:12:51,200 --> 00:12:54,397 continuous exponential. The history of the economy and society as one 227 00:12:54,480 --> 00:12:58,717 continuous exponential. Um and I really mean that. It's not just the technology. 228 00:12:58,800 --> 00:13:02,557 It's this social fabric and the kind of complex structure 229 00:13:02,640 --> 00:13:05,517 of the world that allows this technology to flourish and people to figure out 230 00:13:05,600 --> 00:13:10,077 what to do with it. And you have to embrace it continuously all the way 231 00:13:10,160 --> 00:13:13,277 through. if you miss out on any part of it, um you're not ready for the next 232 00:13:13,360 --> 00:13:16,557 thing that will come if you if you 233 00:13:16,640 --> 00:13:20,637 don't enable your people and your businesses with AI, uh they will not 234 00:13:20,720 --> 00:13:24,077 invent the next level of the scaffolding. But there will be more 235 00:13:24,160 --> 00:13:28,157 things and I think it is tempting to say this one is different and it clearly is 236 00:13:28,240 --> 00:13:32,637 really important. Um but I would view it as a continuum. 237 00:13:32,720 --> 00:13:34,477 >> So 238 00:13:34,560 --> 00:13:37,597 we all read about the the various 239 00:13:37,680 --> 00:13:40,637 companies at the frontier. Can you can 240 00:13:40,720 --> 00:13:42,637 you sort of differentiate them for us? 241 00:13:42,720 --> 00:13:45,197 What what do you what do you view Open 242 00:13:45,280 --> 00:13:48,557 AI as best at? What what's the you know 243 00:13:48,640 --> 00:13:51,517 your DNA if you will that separates you 244 00:13:51,600 --> 00:13:55,597 from the others and and what are you great at? And just give us that sort of 245 00:13:55,680 --> 00:14:01,277 from your perspective. Um 246 00:14:01,360 --> 00:14:03,917 I think from a from what matters to this 247 00:14:04,000 --> 00:14:06,957 room um we are the pragmatic centrists. 248 00:14:07,040 --> 00:14:12,797 We we realize like we want to be We want to be accurate in how we talk 249 00:14:12,880 --> 00:14:14,957 about what we see in the future. We don't want to fall into the trap of 250 00:14:15,040 --> 00:14:18,397 blind optimism. We don't want to fall into the trap of doomerism. This is 251 00:14:18,480 --> 00:14:22,077 clearly complex. There's clearly a combination of both excitement and 252 00:14:22,160 --> 00:14:24,797 anxiety of transformation and risks that 253 00:14:24,880 --> 00:14:29,757 the world has to navigate. Um but 254 00:14:29,840 --> 00:14:32,237 by getting pulled into either extreme, 255 00:14:32,320 --> 00:14:33,837 we are not going to successfully figure 256 00:14:33,920 --> 00:14:36,797 out the challenges in front of us. Um we 257 00:14:36,880 --> 00:14:41,677 want to be a reliable partner. uh we want like we really do view what we are 258 00:14:41,760 --> 00:14:45,437 doing as a platform as a service to the world. We realize that different people 259 00:14:45,520 --> 00:14:49,997 around the world are going to use it very differently and we want to put 260 00:14:50,080 --> 00:14:54,877 sufficient constraints on our system to avoid catastrophic risks but we really 261 00:14:54,960 --> 00:14:57,757 do want to respect all of your sovereignty and that you're going to use 262 00:14:57,840 --> 00:15:01,997 it in very different ways and we personally may not always like it. Um 263 00:15:02,080 --> 00:15:06,397 but I don't like everything that people do with electricity either and I still 264 00:15:06,480 --> 00:15:07,837 think everybody should get to have it. 265 00:15:07,920 --> 00:15:11,880 Um, so, 266 00:15:12,240 --> 00:15:16,717 oh, and then also I would say like we want the whole global economy to 267 00:15:16,800 --> 00:15:20,557 succeed. We do not want to be the only company. Um, we do not want to suck up 268 00:15:20,640 --> 00:15:23,837 all the value. We do not think that is good for global stability, nor do we 269 00:15:23,920 --> 00:15:27,037 think that's possible for the way that people need to flourish and people need 270 00:15:27,120 --> 00:15:30,157 to stay at the center of this. Oh, I'll get to that one more thing after this. 271 00:15:30,240 --> 00:15:34,477 Um and we want to like enable every 272 00:15:34,560 --> 00:15:38,077 entrepreneur, every government, every big company in the world to thrive with 273 00:15:38,160 --> 00:15:42,157 this technology. Uh the last thing I'll say that I think differentiates us from 274 00:15:42,240 --> 00:15:45,757 some of the other companies is we are very firmly and very proudly on team 275 00:15:45,840 --> 00:15:48,317 humanity. Um we would like to build 276 00:15:48,400 --> 00:15:50,237 tools for people. We would like people's 277 00:15:50,320 --> 00:15:53,357 lives to be better. Um we think that 278 00:15:53,440 --> 00:15:56,717 people need to be at the center of the 279 00:15:56,800 --> 00:15:58,957 economy and society and global 280 00:15:59,040 --> 00:16:02,077 decision-m um and that people should run 281 00:16:02,160 --> 00:16:04,237 the future in AI. 282 00:16:04,320 --> 00:16:09,277 >> So it it seems that entrepreneurship 283 00:16:09,360 --> 00:16:11,677 should be really the fire starter of all 284 00:16:11,760 --> 00:16:14,477 this, right? You're are you the you know 285 00:16:14,560 --> 00:16:16,957 the the kindling of it? I mean maybe 286 00:16:17,040 --> 00:16:20,157 sort of tell me how entrepreneurship and 287 00:16:20,240 --> 00:16:24,637 and uh open AI sort of work together. 288 00:16:24,720 --> 00:16:26,957 >> Before open AI I used to run a startup 289 00:16:27,040 --> 00:16:32,637 accelerator and I had always been interested not always after I came to 290 00:16:32,720 --> 00:16:36,077 college I'd always been interested in entrepreneurship and then having had 291 00:16:36,160 --> 00:16:38,157 this incredible front row seat I really 292 00:16:38,240 --> 00:16:40,957 became convinced that 293 00:16:41,040 --> 00:16:43,437 entrepreneurship was 294 00:16:43,520 --> 00:16:45,997 maybe the most important 295 00:16:46,080 --> 00:16:48,797 novel force in all of capitalism. Um, 296 00:16:48,880 --> 00:16:52,717 and I fell in love with it and 297 00:16:52,800 --> 00:16:55,517 watching what new companies were able to 298 00:16:55,600 --> 00:16:59,677 do and watching how 299 00:16:59,760 --> 00:17:03,997 large of a percentage of the new ideas that mattered in the tech industry over 300 00:17:04,080 --> 00:17:07,517 the last, you know, 20 years or whatever came from startups and not big 301 00:17:07,600 --> 00:17:10,557 companies. Um, now having seen this in 302 00:17:10,640 --> 00:17:12,477 other industries, I am a huge believer 303 00:17:12,560 --> 00:17:17,880 in entrepreneurship. I also think that, 304 00:17:18,000 --> 00:17:20,637 you know, like I talked to a lot of people around the world and they would 305 00:17:20,720 --> 00:17:23,677 much rather run a great lifestyle business than go work as like a 306 00:17:23,760 --> 00:17:27,197 corporate cog. And I think that's great, too. Like I think that's just we should 307 00:17:27,280 --> 00:17:30,317 enable that. Um, one of the things I got 308 00:17:30,400 --> 00:17:34,237 excited about early with AI, of course AI is going to transform big companies 309 00:17:34,320 --> 00:17:35,677 and of course AI is going to transform 310 00:17:35,760 --> 00:17:39,757 governments, but what AI can do 311 00:17:39,840 --> 00:17:42,717 for entrepreneurs and for people that 312 00:17:42,800 --> 00:17:45,277 want to take an idea and bring it into 313 00:17:45,360 --> 00:17:49,997 the world quickly um is really quite amazing. When I was running the startup 314 00:17:50,080 --> 00:17:53,357 accelerator, it lasted for three months and the work that 315 00:17:53,440 --> 00:17:57,277 >> it lasted for three months. You were in it for three months and the work that we 316 00:17:57,360 --> 00:17:59,117 expected a startup to accomplish in 317 00:17:59,200 --> 00:18:01,597 three months is now probably doable in 318 00:18:01,680 --> 00:18:05,197 like 17 minutes with codeex. 319 00:18:05,280 --> 00:18:08,877 This is just amazing. The world is not caught up with what this is going to 320 00:18:08,960 --> 00:18:13,757 mean. Not just for the economic value that gets created from that, but how 321 00:18:13,840 --> 00:18:17,037 quickly people can test ideas, build things, get feedback from real 322 00:18:17,120 --> 00:18:18,077 customers. We're all just going to get 323 00:18:18,160 --> 00:18:20,637 much better stuff. 324 00:18:20,720 --> 00:18:24,237 So, you use the term uh I want to sort 325 00:18:24,320 --> 00:18:28,317 of flush it out a bit a bit, non-negotiable. 326 00:18:28,400 --> 00:18:31,037 Can you maybe uh help everybody 327 00:18:31,120 --> 00:18:33,357 understand what you mean about that and 328 00:18:33,440 --> 00:18:35,997 and what that means for each of uh our 329 00:18:36,080 --> 00:18:40,520 partner ministers who are here today? 330 00:18:42,400 --> 00:18:44,557 >> Again, I'm biased. I'm trying to talk my 331 00:18:44,640 --> 00:18:47,117 book here. Forgive me for this analogy. 332 00:18:47,200 --> 00:18:50,637 Um, I think it would be approximately as 333 00:18:50,720 --> 00:18:55,117 bad of an idea to say we're not going to have AI in our country as it was to say 334 00:18:55,200 --> 00:18:58,397 we're not going to have electricity in our country, you know, back 100 plus 335 00:18:58,480 --> 00:19:02,477 years ago. I think this is something at that level. Um, this is something that 336 00:19:02,560 --> 00:19:05,117 every person and every business and 337 00:19:05,200 --> 00:19:07,277 really the entire 338 00:19:07,360 --> 00:19:10,797 machine of society is going to need to 339 00:19:10,880 --> 00:19:13,997 deliver the quality of life that your 340 00:19:14,080 --> 00:19:18,717 citizens should demand. Now, electricity also came with some 341 00:19:18,800 --> 00:19:22,077 complexities and it took a little while for us to figure out how to make it 342 00:19:22,160 --> 00:19:25,837 safe, but we made fairly rapid progress as a society to make it generally pretty 343 00:19:25,920 --> 00:19:30,557 safe even though it can do a lot of bad things. Um, and at this point, we barely 344 00:19:30,640 --> 00:19:33,837 think about electricity. You know, we turn these lights on, we expect them to 345 00:19:33,920 --> 00:19:37,677 work. You know, if this the power shut off and I had to scream because this 346 00:19:37,760 --> 00:19:42,237 microphone was not working, like I would be annoyed about that. You I won't go 347 00:19:42,320 --> 00:19:44,717 through all the examples. I see all the phones on the table. like you just it 348 00:19:44,800 --> 00:19:48,397 would be hard for society to function well without electricity at this point. 349 00:19:48,480 --> 00:19:51,757 But we don't think about it much. We think about, you know, there's light at 350 00:19:51,840 --> 00:19:55,357 night. That's great. We can hear these people speak in the room without them 351 00:19:55,440 --> 00:19:57,837 having to yell. That's nice. Our phones can do these amazing things for us. We 352 00:19:57,920 --> 00:20:00,957 care about our phones. We don't think too much about the electricity that gets 353 00:20:01,040 --> 00:20:03,677 generated to power them. Um, and I think 354 00:20:03,760 --> 00:20:04,877 the same thing is going to happen with 355 00:20:04,960 --> 00:20:08,397 AI. We won't, my prediction is we won't 356 00:20:08,480 --> 00:20:11,197 be talking about AI as much as we do in 357 00:20:11,280 --> 00:20:16,397 a decade from now because we'll just expect it everywhere. You know, like a 358 00:20:16,480 --> 00:20:21,757 kid growing up today will never be smarter than AI, but he or she will also 359 00:20:21,840 --> 00:20:25,997 never have understood a world where every 360 00:20:26,080 --> 00:20:29,757 product and service that they interact with is not really smart and really 361 00:20:29,840 --> 00:20:33,997 capable and really helpful. And because we'll just be so used to that, we won't 362 00:20:34,080 --> 00:20:37,277 think about the AI. will just be like, "Yeah, I want my products to be good." 363 00:20:37,360 --> 00:20:40,557 You know, if I have an idea, I want to be able to like make it right away. I 364 00:20:40,640 --> 00:20:44,157 want this software to be created instantly for me. Um, if I have a 365 00:20:44,240 --> 00:20:45,837 disease, like I expect a cure to get 366 00:20:45,920 --> 00:20:51,277 discovered very quickly. Uh, so I think countries need to need to embrace this. 367 00:20:51,360 --> 00:20:52,557 Again, there's many ways to do it. 368 00:20:52,640 --> 00:20:56,077 People will set different rules. People will decide on what they want to build 369 00:20:56,160 --> 00:20:58,397 versus rent, but I think everyone will 370 00:20:58,480 --> 00:21:01,677 need the this new commodity. 371 00:21:01,760 --> 00:21:06,237 If um if we look forward five years and 372 00:21:06,320 --> 00:21:11,037 and we come back to North Carolina alto together um 373 00:21:11,120 --> 00:21:16,840 what what will be holding us back do you think in your opinion? 374 00:21:18,160 --> 00:21:21,597 >> Well there are clearly challenges that 375 00:21:21,680 --> 00:21:23,837 we have to get right along the way. 376 00:21:23,920 --> 00:21:28,637 There will be specifics. For example I'm sure cyber security is a top-of- mind 377 00:21:28,720 --> 00:21:32,397 issue for many of you. Uh we work with many of the governments in the room. Uh 378 00:21:32,480 --> 00:21:33,837 we work with the state of North 379 00:21:33,920 --> 00:21:36,237 Carolina. Uh we work with companies big 380 00:21:36,320 --> 00:21:40,637 and small and we are 381 00:21:40,720 --> 00:21:45,037 I think we have a big challenge in front of us like to not to not try to sidestep 382 00:21:45,120 --> 00:21:48,877 this one. I think some things are going to go very wrong with cyber security 383 00:21:48,960 --> 00:21:53,197 unless people act quite urgently. There will be other challenges in the next 384 00:21:53,280 --> 00:21:56,397 five years. People talk about uh biocurity and the the things we're going 385 00:21:56,480 --> 00:21:59,997 to face there. um 386 00:22:00,080 --> 00:22:01,917 there will be bigger ones yet to come 387 00:22:02,000 --> 00:22:03,837 and I think if we don't successfully 388 00:22:03,920 --> 00:22:07,037 navigate those um 389 00:22:07,120 --> 00:22:09,037 society will not be able to adopt AI to 390 00:22:09,120 --> 00:22:12,477 the degree or to the speed with which 391 00:22:12,560 --> 00:22:15,677 people deserve. So getting 392 00:22:15,760 --> 00:22:19,837 successfully navigating the risks and downsides in front of us is something 393 00:22:19,920 --> 00:22:23,517 that could go very wrong and I think will set this technology back could set 394 00:22:23,600 --> 00:22:26,797 this technology back a great deal. Um, 395 00:22:26,880 --> 00:22:28,237 another thing that could go wrong is too 396 00:22:28,320 --> 00:22:33,037 much concentration of power. Um, 397 00:22:33,120 --> 00:22:38,397 AI should be in many ways AI should be an equalizing force in the world. You 398 00:22:38,480 --> 00:22:41,437 talked about what it means to have like a reset on the educational capacity 399 00:22:41,520 --> 00:22:44,957 available to everybody. Um, you can also 400 00:22:45,040 --> 00:22:49,437 see worlds where AI concentrates because 401 00:22:49,520 --> 00:22:53,037 some people make early investments and others don't and there's, you know, bad 402 00:22:53,120 --> 00:22:56,957 inequities with comput or whatever. uh that'd be bad. So I think if we fail to 403 00:22:57,040 --> 00:23:01,757 make AI abundant, it ends up being something that is very hotly competed 404 00:23:01,840 --> 00:23:03,917 over. Rich people spend a huge amount of 405 00:23:04,000 --> 00:23:08,717 money on and um 406 00:23:08,800 --> 00:23:12,957 that would be a bad outcome. So I think we need to make it abundant so that we 407 00:23:13,040 --> 00:23:16,957 get the the benefits of lifting everybody up and not much worsening 408 00:23:17,040 --> 00:23:20,157 inequality. >> Right. As you said, if if the smartest 409 00:23:20,240 --> 00:23:22,397 people in the world learned your 410 00:23:22,480 --> 00:23:24,717 language and all agreed to move to your 411 00:23:24,800 --> 00:23:27,357 country and work on whatever you wanted 412 00:23:27,440 --> 00:23:30,157 them to work on, that that has to be the 413 00:23:30,240 --> 00:23:35,277 most incredible thing. Do you think that 414 00:23:35,360 --> 00:23:38,957 since you'll have all the smartest people speaking all those languages, 415 00:23:39,040 --> 00:23:40,717 can't they figure out a way do you think 416 00:23:40,800 --> 00:23:43,677 over time to deliver the same amount of 417 00:23:43,760 --> 00:23:47,197 compute more inexpensively without the 418 00:23:47,280 --> 00:23:51,997 same use of power without the same I mean aren't these the kind of things 419 00:23:52,080 --> 00:23:53,997 >> absolutely >> that are going to happen that will 420 00:23:54,080 --> 00:23:57,597 democratize it by sort of by itself in a way 421 00:23:57,680 --> 00:24:00,717 >> it it will still require huge investment 422 00:24:00,800 --> 00:24:05,117 but yes we will continue to make things vastly more efficient and we'll figure 423 00:24:05,200 --> 00:24:08,637 out how to make more and better chips and we'll figure out how to make smarter 424 00:24:08,720 --> 00:24:12,797 algorithms. All of that will happen. But to put some numbers on the magnitude of 425 00:24:12,880 --> 00:24:15,997 how much we we need this to happen, again, I know these numbers in front of 426 00:24:16,080 --> 00:24:19,517 me. I'm not going to get this exactly right. Um but forgive me for the 427 00:24:19,600 --> 00:24:23,917 approximation. At the beginning of 2020, um the token 428 00:24:24,000 --> 00:24:27,997 leader in the world was an OpenAI employee using roughly a 100,000 tokens 429 00:24:28,080 --> 00:24:30,157 a month. 430 00:24:30,240 --> 00:24:33,517 Um, in the middle of 2026, 431 00:24:33,600 --> 00:24:37,917 the token leader that I know of, there may be somebody using more than this at 432 00:24:38,000 --> 00:24:40,397 outside of OpenAI, um, was using 433 00:24:40,480 --> 00:24:43,197 hundreds of billions of tokens a month. 434 00:24:43,280 --> 00:24:48,077 So, you know, a multi-millionx increase in six and a half years. And the average 435 00:24:48,160 --> 00:24:51,917 person on Earth had gone from using effectively zero tokens to 100,000 436 00:24:52,000 --> 00:24:55,357 tokens a month. That was the global per capita thing. 437 00:24:55,440 --> 00:24:58,477 If we just naively project this forward again, which seems like a reasonable 438 00:24:58,560 --> 00:25:03,997 thing to do to me. Um, and I hate tokens as a unit here. It's a dumb one. No one 439 00:25:04,080 --> 00:25:07,357 should care, but like let's just take that as shorthand for amount of 440 00:25:07,440 --> 00:25:10,557 intelligence. Then 441 00:25:10,640 --> 00:25:14,797 in another six and a half years, the average per capita use of tokens will go 442 00:25:14,880 --> 00:25:19,437 up to, let's round it down to 100 billion tokens a month. And the person 443 00:25:19,520 --> 00:25:23,837 using the most in the world will be using 444 00:25:23,920 --> 00:25:29,517 you know a 100 quadrillion tokens a month uh or one quadrillion 100 trillion 445 00:25:29,600 --> 00:25:32,157 like >> no I was right 100 trillion tokens a 446 00:25:32,240 --> 00:25:35,677 month 100 quadrillion tokens a month um 447 00:25:35,760 --> 00:25:41,037 even with great gains on efficiency we are going to need to build much more 448 00:25:41,120 --> 00:25:45,357 infrastructure unless we want this to be a highly 449 00:25:45,440 --> 00:25:48,397 priced commodity which we don't we want this to be abund on it. We want this to 450 00:25:48,480 --> 00:25:51,997 be inexpensive. We want people to just be able to use this for whatever they need. 451 00:25:52,080 --> 00:25:54,957 >> So that's the driving force. If we 452 00:25:55,040 --> 00:25:58,717 if we globally can build enough 453 00:25:58,800 --> 00:26:01,357 then it will intelligence will be 454 00:26:01,440 --> 00:26:03,437 broadly distributed 455 00:26:03,520 --> 00:26:06,557 uh more democratized across the world. 456 00:26:06,640 --> 00:26:10,237 Uh the world will be educated or they'll 457 00:26:10,320 --> 00:26:14,237 have the access to education. Is that 458 00:26:14,320 --> 00:26:16,237 >> to to go back to the analogy of 459 00:26:16,320 --> 00:26:19,117 electricity? Um, there's this amazing 460 00:26:19,200 --> 00:26:22,557 chart of what it cost to have an hour of 461 00:26:22,640 --> 00:26:27,677 light at night. And there was a time, one of the 462 00:26:27,760 --> 00:26:32,557 earliest data points you can find is that it it took an average of 5 hours of 463 00:26:32,640 --> 00:26:34,637 labor at the wage at the time to be able 464 00:26:34,720 --> 00:26:38,237 to pay for one hour of light at night. 465 00:26:38,320 --> 00:26:42,237 Electricity came along, it fell. We figured out how to make more 466 00:26:42,320 --> 00:26:45,357 electricity. So, there was like, you know, the supply and demand change 467 00:26:45,440 --> 00:26:50,397 curve. It fell again. We figured out how to make vastly more efficient light 468 00:26:50,480 --> 00:26:54,237 bulbs. We went from incandescent lights to LEDs. A whole bunch of other things 469 00:26:54,320 --> 00:26:57,117 happened and now an hour of light at 470 00:26:57,200 --> 00:27:01,277 night is not something that people think of at all about the cost. I don't even 471 00:27:01,360 --> 00:27:04,157 know what the number is, but someone in here can do it. It's it's like very 472 00:27:04,240 --> 00:27:06,077 cheap to run, you know, a five watt bulb 473 00:27:06,160 --> 00:27:10,477 for an hour. Um the 474 00:27:10,560 --> 00:27:13,997 if that hadn't happened if we and all of those things together had to happen lots 475 00:27:14,080 --> 00:27:16,877 of electricity generation huge innovation on the efficiency all of 476 00:27:16,960 --> 00:27:21,917 these things. If that hadn't happened then rich people get light at night and 477 00:27:22,000 --> 00:27:24,557 they can study more they have a compounding effect they can learn more 478 00:27:24,640 --> 00:27:27,197 whatever and and you know poor people 479 00:27:27,280 --> 00:27:31,677 don't. So I think it's been very good that energy has become abundant and that 480 00:27:31,760 --> 00:27:33,917 light at night has become relatively 481 00:27:34,000 --> 00:27:37,517 like shockingly inexpensive. Um, if that 482 00:27:37,600 --> 00:27:39,997 same thing happens with AI and anybody 483 00:27:40,080 --> 00:27:41,917 on earth, rich, poor, whatever, can use 484 00:27:42,000 --> 00:27:47,197 AI to get the abundant services that they need and to express their ideas and 485 00:27:47,280 --> 00:27:51,357 to start their businesses and whatever else. And we fulfill our mission of 486 00:27:51,440 --> 00:27:56,237 making AI so inexpensive and so smart that people are just like they think 487 00:27:56,320 --> 00:27:59,597 about it like they think of the cost of an hour at night of light at night. I 488 00:27:59,680 --> 00:28:03,197 think that'd be really wonderful. And then this huge reset and democratization 489 00:28:03,280 --> 00:28:05,677 of education you're talking about can happen. 490 00:28:05,760 --> 00:28:09,997 So are you you know there are look there 491 00:28:10,080 --> 00:28:12,557 are doomers out there who worry about 492 00:28:12,640 --> 00:28:15,277 everything but we know that if you look 493 00:28:15,360 --> 00:28:17,117 back through the industrial re as you 494 00:28:17,200 --> 00:28:19,677 said it it always seems to work out 495 00:28:19,760 --> 00:28:22,557 fine. Are you uh fundamentally 496 00:28:22,640 --> 00:28:25,277 optimistic to what your products will do 497 00:28:25,360 --> 00:28:28,797 to the world and will serve the world? 498 00:28:28,880 --> 00:28:30,557 I I think part of the reason it tends to 499 00:28:30,640 --> 00:28:34,477 work out fine is people worry and they 500 00:28:34,560 --> 00:28:37,597 stress and they look at the downsides and they try to address those and they 501 00:28:37,680 --> 00:28:40,477 try to learn how we avoid accidents. So 502 00:28:40,560 --> 00:28:43,197 if we go back to an early example even 503 00:28:43,280 --> 00:28:45,597 before electricity of a technology fire 504 00:28:45,680 --> 00:28:50,477 was incredible incredible for human evolution incredible for letting us cook 505 00:28:50,560 --> 00:28:53,757 our food and heat our homes at night and whatever else. It also burnt down whole 506 00:28:53,840 --> 00:28:57,037 cities a few times and people really 507 00:28:57,120 --> 00:28:58,317 worried about that because we don't want 508 00:28:58,400 --> 00:29:00,797 cities to burn down and we figured out 509 00:29:00,880 --> 00:29:05,277 fire safety and through a lot of human 510 00:29:05,360 --> 00:29:07,277 ingenuity um through a lot of good 511 00:29:07,360 --> 00:29:09,597 policy definitely through learning from 512 00:29:09,680 --> 00:29:14,397 some very bad accidents we figured out how to make it safer. Now had we just 513 00:29:14,480 --> 00:29:17,837 said like yeah you know some cities are going to burn down whatever we got other 514 00:29:17,920 --> 00:29:20,717 problems to work on probably wouldn't have figured that out and that wouldn't 515 00:29:20,800 --> 00:29:24,717 have been as good. Um, so I think worry 516 00:29:24,800 --> 00:29:28,077 is warranted. Worry is how we anticipate 517 00:29:28,160 --> 00:29:32,637 problems and how we from a matter of policy and a matter of technology figure 518 00:29:32,720 --> 00:29:36,157 out how to address them. But I am optimistic that we can mitigate the 519 00:29:36,240 --> 00:29:40,797 downsides of this technology. And if you look at what's happened, you know, if 520 00:29:40,880 --> 00:29:44,397 you look at what the doomers were saying 5 years ago about the chances of us 521 00:29:44,480 --> 00:29:48,877 being able to make AI of this level, of this sort of near human, sometimes 522 00:29:48,960 --> 00:29:53,917 superhuman capacity level, you know, broadly considered safe and reliable. 523 00:29:54,000 --> 00:29:55,277 Not perfect. There have been accidents. 524 00:29:55,360 --> 00:29:58,557 There will be accidents, of course. I don't think most people would have 525 00:29:58,640 --> 00:30:00,477 thought it was going to go this well as 526 00:30:00,560 --> 00:30:04,157 it has so far. And I I believe in human 527 00:30:04,240 --> 00:30:08,797 ingenuity and the ability for to solve very difficult, very complex 528 00:30:08,880 --> 00:30:11,997 problems. And I think I think we'll keep doing that. 529 00:30:12,080 --> 00:30:15,117 And I think the economic value and the 530 00:30:15,200 --> 00:30:20,317 personal value is so huge. Um I met with someone just this morning and told and 531 00:30:20,400 --> 00:30:23,437 again I'm sure you've all heard stories like this that they wouldn't be alive 532 00:30:23,520 --> 00:30:26,477 without Chachi PT um and that they had a 533 00:30:26,560 --> 00:30:30,557 very difficult disease and you know got a diagnosis that doctors weren't able to 534 00:30:30,640 --> 00:30:32,237 >> give them. We shouldn't take that away. 535 00:30:32,320 --> 00:30:35,437 >> Yeah. But we need to figure out how to make the technology safe so that people 536 00:30:35,520 --> 00:30:39,757 can have more of that. >> Great. So Sam, thank you for joining us. 537 00:30:39,840 --> 00:30:41,197 It was really an incredible 538 00:30:41,280 --> 00:30:46,317 conversation. Um 2:00, so 2 p.m. in the 539 00:30:46,400 --> 00:30:49,837 ministerial room where we will have uh 540 00:30:49,920 --> 00:30:54,077 Alex Karp will be our next guest uh speaking to us. 541 00:30:54,160 --> 00:30:57,197 >> He's a good one. >> He's a good one. But uh 542 00:30:57,280 --> 00:31:00,797 >> thank you Sam for joining us. We we really appreciate it. Thank you. 543 00:31:00,880 --> 00:31:05,160 [applause] Thanks.