1 00:00:00,120 --> 00:00:03,600 Speaker 1: You guys made some news today announcing some new guidelines 2 00:00:03,600 --> 00:00:06,240 Speaker 1: around the use of AI in elections. I'm sure it's 3 00:00:06,280 --> 00:00:11,080 Speaker 1: all uh stuff that the Davos set love to hear. 4 00:00:12,039 --> 00:00:15,320 Speaker 1: You banned the use of GPT and political campaigns. You 5 00:00:15,480 --> 00:00:20,360 Speaker 1: introduced cryptographic watermarks or images created by Dali to create 6 00:00:20,440 --> 00:00:23,920 Speaker 1: kind of providence and transparency around the use of AI 7 00:00:23,960 --> 00:00:24,880 Speaker 1: generated images. 8 00:00:25,200 --> 00:00:27,200 Speaker 2: I read it and I thought, you know, this is great. 9 00:00:27,880 --> 00:00:28,480 Speaker 2: Some of these. 10 00:00:28,320 --> 00:00:31,560 Speaker 1: Principles are shared by much larger platforms like Facebook and 11 00:00:31,640 --> 00:00:34,839 Speaker 1: TikTok and YouTube, and they have struggled to enforce it. 12 00:00:35,040 --> 00:00:37,800 Speaker 2: How do you make it real? 13 00:00:39,840 --> 00:00:41,440 Speaker 3: These A lot of these are things that we've been 14 00:00:41,440 --> 00:00:43,840 Speaker 3: doing for a long time, and we have a really 15 00:00:44,040 --> 00:00:48,680 Speaker 3: strong safety systems team that not only sort of has monitoring, 16 00:00:48,760 --> 00:00:51,120 Speaker 3: but we're actually able to leverage our own tools in 17 00:00:51,240 --> 00:00:53,520 Speaker 3: order to scale our enforcement, which gives us I think 18 00:00:53,560 --> 00:00:59,960 Speaker 3: a significant advantage. But so there are this they're also 19 00:01:00,080 --> 00:01:02,480 Speaker 3: really boarding partnership like with the National Association of the 20 00:01:02,480 --> 00:01:06,000 Speaker 3: Secretaries of States, so we can surface authoritative voting information. 21 00:01:06,480 --> 00:01:08,600 Speaker 4: So we have quite a few ways that we are 22 00:01:08,640 --> 00:01:10,000 Speaker 4: able to enforce this, Ran. 23 00:01:09,959 --> 00:01:12,319 Speaker 1: Sam, are you does this put your mind at ease 24 00:01:12,440 --> 00:01:16,000 Speaker 1: that we don't that open ai doesn't move the needle 25 00:01:16,360 --> 00:01:20,720 Speaker 1: in some seventy seven upcoming critical democratic elections in twenty 26 00:01:20,760 --> 00:01:21,200 Speaker 1: twenty four. 27 00:01:22,000 --> 00:01:24,200 Speaker 5: No, we're quite focused on it, and I think it's 28 00:01:24,200 --> 00:01:25,640 Speaker 5: good that ourline is not of these. I think it's 29 00:01:25,640 --> 00:01:27,360 Speaker 5: good that we have a lot of anxiety and are 30 00:01:27,360 --> 00:01:28,759 Speaker 5: going to do everything we can to get it as 31 00:01:28,800 --> 00:01:32,560 Speaker 5: right as we can. I think our role is very 32 00:01:32,600 --> 00:01:35,520 Speaker 5: different than the role of a distribution platform, but still important. 33 00:01:35,600 --> 00:01:37,880 Speaker 2: We'll have to work with them too, you know. 34 00:01:37,920 --> 00:01:41,679 Speaker 5: It's like you generate here and distribute here, and there 35 00:01:41,680 --> 00:01:44,880 Speaker 5: needs to be a good conversation between them. But we 36 00:01:45,000 --> 00:01:48,320 Speaker 5: also have the benefit of having watched what's happened in 37 00:01:48,400 --> 00:01:53,840 Speaker 5: previous cycles with previous technologies, and I don't think this 38 00:01:53,880 --> 00:01:55,680 Speaker 5: will be the same as before. I think it's always 39 00:01:55,680 --> 00:01:57,760 Speaker 5: a mistake to try to fight the last war, but 40 00:01:57,840 --> 00:01:59,960 Speaker 5: we do get to take away some learnings from that, 41 00:02:00,480 --> 00:02:03,680 Speaker 5: and so I wouldn't you know, I think it'd be 42 00:02:03,720 --> 00:02:05,080 Speaker 5: terrible if I said, oh, yeah, I'm not worried. I 43 00:02:05,120 --> 00:02:07,480 Speaker 5: feel great, like we're gonna have to watch this incredibly 44 00:02:07,520 --> 00:02:10,320 Speaker 5: close to this year, super tight monitoring, super tight feedback group. 45 00:02:11,280 --> 00:02:15,080 Speaker 1: You were at Facebook for open AI, so I almost 46 00:02:15,120 --> 00:02:19,440 Speaker 1: apologize for asking this in this way, probably a trigger phrase, 47 00:02:19,480 --> 00:02:23,320 Speaker 1: But do you worry about another Cambridge analytical analytical moment? 48 00:02:23,919 --> 00:02:26,079 Speaker 4: I think, as Sam alluded to, there are a lot 49 00:02:26,120 --> 00:02:28,520 Speaker 4: of learnings that we can leverage but also open. 50 00:02:28,520 --> 00:02:31,760 Speaker 3: AYE from its inception has been a company that thinks 51 00:02:31,800 --> 00:02:33,680 Speaker 3: about these issues. That it was one of the reasons 52 00:02:33,680 --> 00:02:35,880 Speaker 3: that it was founded. So I think I'm a lot 53 00:02:35,960 --> 00:02:38,680 Speaker 3: less concerned because these are issues that our teams have 54 00:02:38,720 --> 00:02:42,480 Speaker 3: been thinking about from the beginning of our building. 55 00:02:42,120 --> 00:02:42,760 Speaker 4: Of these tools. 56 00:02:43,760 --> 00:02:48,400 Speaker 1: Sam, Donald Trump just won the Iowa coccus yesterday. We 57 00:02:48,440 --> 00:02:50,520 Speaker 1: are now sort of confronted with the reality of this 58 00:02:50,600 --> 00:02:51,360 Speaker 1: upcoming election. 59 00:02:52,240 --> 00:02:53,839 Speaker 2: What do you think is at stake. 60 00:02:55,040 --> 00:02:58,760 Speaker 1: In the US election for tech and for the safe 61 00:02:58,800 --> 00:02:59,840 Speaker 1: stewardship of AID. 62 00:03:00,080 --> 00:03:02,280 Speaker 2: Feel like that's a critical. 63 00:03:01,880 --> 00:03:04,840 Speaker 1: Issue that voters should and we'll have to consider in 64 00:03:04,840 --> 00:03:05,280 Speaker 1: this election. 65 00:03:05,320 --> 00:03:08,080 Speaker 5: I think they're now confronted as part of the problem. Actually, 66 00:03:08,080 --> 00:03:10,040 Speaker 5: I think most people who come to dot say that. Again, 67 00:03:10,080 --> 00:03:11,440 Speaker 5: I didn't quite get that. I think part of the 68 00:03:11,440 --> 00:03:13,799 Speaker 5: problems we're saying, we're now confronted. You know, it never 69 00:03:13,840 --> 00:03:16,079 Speaker 5: occurred to us that what Trump is saying might be 70 00:03:16,120 --> 00:03:17,880 Speaker 5: resonating with a lot of people and now all of 71 00:03:17,919 --> 00:03:22,359 Speaker 5: a sudden after this performance in Iowa. Oh man, that's 72 00:03:22,400 --> 00:03:26,040 Speaker 5: a very like Davos Centriet. You know, I've been here 73 00:03:26,040 --> 00:03:26,600 Speaker 5: for two days. 74 00:03:26,639 --> 00:03:30,760 Speaker 2: I guess I just so. 75 00:03:31,200 --> 00:03:33,680 Speaker 5: I would love if we had a lot more reflection 76 00:03:33,760 --> 00:03:36,880 Speaker 5: and if we started it a lot sooner about and 77 00:03:37,040 --> 00:03:40,000 Speaker 5: we didn't feel now confronted. But I think there's a 78 00:03:40,040 --> 00:03:42,000 Speaker 5: lot at stake at this election. I think elections are, 79 00:03:42,440 --> 00:03:46,560 Speaker 5: you know, huge deals. I believe that America is going 80 00:03:46,600 --> 00:03:48,920 Speaker 5: to be fine no matter what happens in this election. 81 00:03:49,400 --> 00:03:51,120 Speaker 5: I believe that AI is going to be fine no 82 00:03:51,120 --> 00:03:53,360 Speaker 5: matter what happens in this election, and we will have 83 00:03:53,400 --> 00:03:54,360 Speaker 5: to work very hard to. 84 00:03:54,320 --> 00:03:54,800 Speaker 2: Make it so. 85 00:03:56,360 --> 00:03:59,440 Speaker 5: But this is not you know, no one wants to 86 00:03:59,440 --> 00:04:02,080 Speaker 5: say set up here and like hear me rant about politics. 87 00:04:02,080 --> 00:04:02,920 Speaker 2: I'm going to stop after this. 88 00:04:04,680 --> 00:04:08,840 Speaker 5: But I think there has been a real failure to 89 00:04:09,040 --> 00:04:14,000 Speaker 5: sort of learn lessons about what's kind of like working 90 00:04:14,040 --> 00:04:15,400 Speaker 5: for the citizens of America and what's not. 91 00:04:16,320 --> 00:04:19,000 Speaker 1: And I want to ask you the same question, you know, 92 00:04:19,080 --> 00:04:21,960 Speaker 1: taking your political background into account, what do you feel 93 00:04:22,000 --> 00:04:25,880 Speaker 1: like for Silicon Valley, for AI is at stake in 94 00:04:25,920 --> 00:04:26,640 Speaker 1: the US election. 95 00:04:27,560 --> 00:04:29,400 Speaker 4: I think what has struck. 96 00:04:29,120 --> 00:04:31,400 Speaker 3: Me and has been really remarkable is that the conversation 97 00:04:31,520 --> 00:04:37,080 Speaker 3: around AI has remained very bipartisan. And so you know, 98 00:04:37,360 --> 00:04:39,080 Speaker 3: I think that the one concern I have is that 99 00:04:39,160 --> 00:04:40,720 Speaker 3: somehow both parties hate it. 100 00:04:41,000 --> 00:04:43,480 Speaker 2: Yeah. 101 00:04:43,720 --> 00:04:45,920 Speaker 4: No, but you know, this is like an area where. 102 00:04:48,000 --> 00:04:51,600 Speaker 3: Republicans tend to, of course have an approach where they 103 00:04:51,600 --> 00:04:53,640 Speaker 3: are not as in favor of regulation. But on this 104 00:04:53,920 --> 00:04:57,120 Speaker 3: I think there's agreement on both parties that they are considered. 105 00:04:57,240 --> 00:04:59,960 Speaker 3: Do they believe that something is needed on this second 106 00:05:00,000 --> 00:05:02,279 Speaker 3: Cologyina Sener Schumer has this big part is an effort 107 00:05:02,440 --> 00:05:04,560 Speaker 3: that he's running with his Republican counterparts. 108 00:05:04,839 --> 00:05:06,080 Speaker 2: Again, when we. 109 00:05:06,000 --> 00:05:09,040 Speaker 3: Speak to people in DC on both sides of the aisle, 110 00:05:10,000 --> 00:05:12,960 Speaker 3: for now, it seems like they're on the same page. 111 00:05:13,040 --> 00:05:15,760 Speaker 1: And do you feel like all the existing campaigns are 112 00:05:15,760 --> 00:05:21,240 Speaker 1: equally articulate about the issues relating to aias. 113 00:05:20,279 --> 00:05:22,560 Speaker 3: That AI has really been a campaign issue to date. 114 00:05:22,920 --> 00:05:24,440 Speaker 3: So it will be interesting to see. 115 00:05:24,279 --> 00:05:26,800 Speaker 5: How worry right about what's going to happen here. This 116 00:05:26,960 --> 00:05:30,559 Speaker 5: is like bigger than just a technological revolution in some sense, 117 00:05:30,800 --> 00:05:34,240 Speaker 5: I mean sort of like all technological revolutions or societal revolutions, 118 00:05:34,600 --> 00:05:36,760 Speaker 5: but this one feels like it can be much more 119 00:05:36,800 --> 00:05:40,880 Speaker 5: of that than usual, and so it is going to 120 00:05:40,920 --> 00:05:45,600 Speaker 5: become a social issue, a political issue. It already has 121 00:05:45,640 --> 00:05:49,440 Speaker 5: in some ways, but I think it is strange to 122 00:05:49,480 --> 00:05:51,279 Speaker 5: both of us that it's not more of that already, 123 00:05:51,320 --> 00:05:53,680 Speaker 5: but with what we expect to happen this year, not 124 00:05:53,800 --> 00:05:55,720 Speaker 5: with the election, but just with the increase in the 125 00:05:55,760 --> 00:06:00,680 Speaker 5: capabilities of the products and as people really catch up 126 00:06:00,720 --> 00:06:04,480 Speaker 5: with what's going to happen, what is happening, what's already happened. 127 00:06:05,000 --> 00:06:06,960 Speaker 5: There's like a lot of inertia always and society well, 128 00:06:07,040 --> 00:06:07,400 Speaker 5: I mean, there. 129 00:06:07,320 --> 00:06:09,560 Speaker 1: Are political figures in the UF and around the world, 130 00:06:10,120 --> 00:06:13,400 Speaker 1: like Donald Trump who have successfully tapped into a feeling 131 00:06:13,520 --> 00:06:18,760 Speaker 1: of dislocation, anger of the working class, the feeling of it. 132 00:06:18,920 --> 00:06:24,000 Speaker 1: You know, after baiting inequality or a technology leaving people behind, 133 00:06:24,160 --> 00:06:28,040 Speaker 1: is there the danger that you know, AI furthers those trends. 134 00:06:28,160 --> 00:06:30,559 Speaker 5: Yes, for sure, I think that's something to think about. 135 00:06:30,640 --> 00:06:34,839 Speaker 5: But one of the things that surprised us very pleasantly 136 00:06:34,880 --> 00:06:37,000 Speaker 5: on the upside, because you know, when you start building 137 00:06:37,040 --> 00:06:39,400 Speaker 5: a technology, you start doing to research, you kind of say, well, 138 00:06:39,440 --> 00:06:41,240 Speaker 5: we'll follow where the science leads us, and when you 139 00:06:41,240 --> 00:06:43,000 Speaker 5: put a product, you'll say this is going to coabob 140 00:06:43,040 --> 00:06:46,000 Speaker 5: with society and we'll follow where users lead us. But 141 00:06:46,520 --> 00:06:49,799 Speaker 5: it's not you get to steer it, but only somewhat. 142 00:06:49,839 --> 00:06:52,640 Speaker 5: There's some which is just like, this is what the 143 00:06:52,680 --> 00:06:55,200 Speaker 5: technology can do, this is how people want to use it, 144 00:06:55,720 --> 00:06:57,680 Speaker 5: and this is what it's capable of. And this has 145 00:06:57,760 --> 00:07:00,799 Speaker 5: been much more of a tool than I think we expected. 146 00:07:00,839 --> 00:07:04,200 Speaker 5: It is not yet and again in the future it'll 147 00:07:04,240 --> 00:07:07,400 Speaker 5: get better, but it's not yet like replacing jobs in 148 00:07:07,440 --> 00:07:09,480 Speaker 5: the way to the degree that people thought it was 149 00:07:09,520 --> 00:07:13,240 Speaker 5: going to. It is this incredible tool for productivity and 150 00:07:13,280 --> 00:07:16,680 Speaker 5: you can see people magnifying what they can do by 151 00:07:16,680 --> 00:07:18,880 Speaker 5: a factor of two or five or in some way 152 00:07:18,920 --> 00:07:20,560 Speaker 5: that doesn't even talk to it makes sense to talk 153 00:07:20,560 --> 00:07:22,120 Speaker 5: about a number because they just couldn't do the things 154 00:07:22,160 --> 00:07:26,760 Speaker 5: at all before. And that is I think quite exciting, 155 00:07:27,160 --> 00:07:30,000 Speaker 5: this new vision in the future that we didn't really 156 00:07:30,040 --> 00:07:31,880 Speaker 5: see when we started. We kind of didn't know it 157 00:07:31,920 --> 00:07:34,200 Speaker 5: was going to go, and very thankful the technology did 158 00:07:34,240 --> 00:07:36,600 Speaker 5: go in this direction. But where this is a tool 159 00:07:36,640 --> 00:07:39,920 Speaker 5: that magnifies what humans do, lets people do their jobs better, 160 00:07:40,240 --> 00:07:42,840 Speaker 5: lets the AI do parts of jobs, and of course 161 00:07:42,920 --> 00:07:45,000 Speaker 5: jobs will change and of course some jobs will totally 162 00:07:45,000 --> 00:07:48,160 Speaker 5: go away, but the human drives are so strong in 163 00:07:48,160 --> 00:07:51,080 Speaker 5: the sort of way that society works, so strong that 164 00:07:51,160 --> 00:07:53,480 Speaker 5: I think, and I can't believe I'm saying this, because 165 00:07:53,640 --> 00:07:56,480 Speaker 5: it would have sounded like an ungrammatical sentence to me 166 00:07:56,520 --> 00:07:58,880 Speaker 5: at some point, But I think AGI will get developed 167 00:07:59,200 --> 00:08:02,640 Speaker 5: and reasonably close ish future, and it'll change the world 168 00:08:02,760 --> 00:08:04,640 Speaker 5: much less than we all think it'll change jobs and 169 00:08:04,720 --> 00:08:07,040 Speaker 5: much less than we all think. And again that sounds 170 00:08:08,000 --> 00:08:10,240 Speaker 5: I may be wrong again now, but that wouldn't even 171 00:08:10,280 --> 00:08:12,559 Speaker 5: compiled for me as a sentence at some point, given 172 00:08:12,560 --> 00:08:14,440 Speaker 5: my conception then of how EGI was going to go. 173 00:08:14,680 --> 00:08:17,880 Speaker 1: You've watched the technology develop have you both changed your 174 00:08:17,960 --> 00:08:22,440 Speaker 1: views on how significant the job dislocation and disruption. 175 00:08:22,080 --> 00:08:24,080 Speaker 2: Will be as AGI comes into focus. 176 00:08:24,560 --> 00:08:26,440 Speaker 3: So this is actually an area that we have a 177 00:08:26,440 --> 00:08:29,320 Speaker 3: policy research team that studies this, and they've seen pretty 178 00:08:29,360 --> 00:08:32,440 Speaker 3: significant impact in terms of changing the way people do 179 00:08:32,520 --> 00:08:35,200 Speaker 3: jobs rather than job dislocation. And I think that's actually 180 00:08:35,200 --> 00:08:37,480 Speaker 3: going to accelerate and that it's going to change. 181 00:08:37,320 --> 00:08:40,480 Speaker 4: More people's jobs that a Sam said. 182 00:08:40,559 --> 00:08:44,160 Speaker 3: So far, it hasn't been the significant replacement of jobs. 183 00:08:44,960 --> 00:08:46,840 Speaker 5: You know, you hear a coder say, okay, I'm like 184 00:08:46,840 --> 00:08:49,040 Speaker 5: two times well productive, three times won't productive whatever than 185 00:08:49,040 --> 00:08:50,680 Speaker 5: they used to be, and I like, can never code 186 00:08:50,679 --> 00:08:52,360 Speaker 5: again without this tool. You mostly hear that from the 187 00:08:52,360 --> 00:08:56,800 Speaker 5: younger ones, but it turns out and I think this 188 00:08:56,800 --> 00:08:58,800 Speaker 5: would be true for a lot of industries. The world 189 00:08:58,880 --> 00:09:00,720 Speaker 5: just needs a lot more code than we have people 190 00:09:00,720 --> 00:09:03,280 Speaker 5: to write right now. And so it's not like we 191 00:09:03,360 --> 00:09:05,560 Speaker 5: run out of demand. It's that people can just do more. 192 00:09:06,040 --> 00:09:07,800 Speaker 5: Expectations go up, but ability goes up. 193 00:09:07,760 --> 00:09:11,600 Speaker 1: To I want to ask you about another news report 194 00:09:11,640 --> 00:09:15,920 Speaker 1: today that suggested that open AI was relaxing its restrictions 195 00:09:16,480 --> 00:09:21,479 Speaker 1: around the use of AI in military projects and developing weapons. 196 00:09:21,600 --> 00:09:23,200 Speaker 2: Can you say more about that? 197 00:09:23,800 --> 00:09:25,400 Speaker 1: And you know what work are you doing with the 198 00:09:25,480 --> 00:09:28,000 Speaker 1: US Department of Defense and other military agencies. 199 00:09:28,480 --> 00:09:31,640 Speaker 3: So a lot of these policies were written before we 200 00:09:31,720 --> 00:09:34,200 Speaker 3: even knew what these people use our tools for. So 201 00:09:34,800 --> 00:09:37,800 Speaker 3: what this was not actually just the adjustment of the 202 00:09:37,840 --> 00:09:40,079 Speaker 3: military use case policies, but across the board to make 203 00:09:40,080 --> 00:09:42,520 Speaker 3: it more clear so people understand what. 204 00:09:42,559 --> 00:09:43,880 Speaker 4: Is possible when it's not possible. 205 00:09:44,000 --> 00:09:49,160 Speaker 3: But specifically on this area, we actually still prohibit the 206 00:09:49,200 --> 00:09:53,319 Speaker 3: development of weapons, the destruction of property, harn to individuals. 207 00:09:53,520 --> 00:09:55,480 Speaker 3: But for example, we've been doing work with the Department 208 00:09:55,520 --> 00:10:00,199 Speaker 3: of Defense on cybersecurity tools for open source soft where 209 00:10:00,240 --> 00:10:03,320 Speaker 3: that secure's critical infrastructure. We've been exploring whether it can 210 00:10:03,360 --> 00:10:06,640 Speaker 3: assist with veteran suicide. And because we previously have a 211 00:10:06,960 --> 00:10:10,120 Speaker 3: what essentially was a blank of provision on military many 212 00:10:10,120 --> 00:10:12,240 Speaker 3: people felt like that would have prohibited any of these 213 00:10:12,320 --> 00:10:14,440 Speaker 3: use cases, which we think are very much aligned with 214 00:10:14,480 --> 00:10:15,880 Speaker 3: what we want to see in the world. 215 00:10:16,120 --> 00:10:19,320 Speaker 1: Has the US government asked you to restrict the level 216 00:10:19,320 --> 00:10:23,280 Speaker 1: of cooperation with militaries in other countries. 217 00:10:24,480 --> 00:10:27,640 Speaker 3: They haven't asked this, but we certainly are not, you know, 218 00:10:27,720 --> 00:10:31,640 Speaker 3: right for now, Actually our discussions are focused on United States. 219 00:10:31,520 --> 00:10:33,080 Speaker 4: National security agencies, and. 220 00:10:34,360 --> 00:10:36,719 Speaker 3: You know, I think we have always believed that democracies 221 00:10:37,120 --> 00:10:38,800 Speaker 3: need to be in the lead on this technology. 222 00:10:39,200 --> 00:10:41,840 Speaker 1: Sam changing topics, give us an update on the GPT 223 00:10:42,000 --> 00:10:45,439 Speaker 1: store and are you seeing maybe probably explain it briefly, 224 00:10:45,640 --> 00:10:47,760 Speaker 1: and are you seeing the same kind of explosion of 225 00:10:47,840 --> 00:10:50,640 Speaker 1: creativity we saw in the early days of the mobile 226 00:10:50,640 --> 00:10:51,200 Speaker 1: app stores. 227 00:10:51,920 --> 00:10:54,319 Speaker 2: Yeah, the same level creativity and the same level of crap. 228 00:10:54,400 --> 00:10:56,960 Speaker 5: But I mean that happens in the early days as 229 00:10:57,000 --> 00:10:59,560 Speaker 5: people like feel out technology. There's some incredible stuff in 230 00:10:59,600 --> 00:11:03,080 Speaker 5: there too. Give us an example the GPTs. Did I 231 00:11:03,080 --> 00:11:04,040 Speaker 5: say what GPT's are the first? 232 00:11:04,120 --> 00:11:04,280 Speaker 2: Yeah? 233 00:11:04,280 --> 00:11:06,800 Speaker 5: Sure, So GPTs are a way to do a very 234 00:11:06,840 --> 00:11:10,560 Speaker 5: lightweight customization of chat GPT and if you want it 235 00:11:10,600 --> 00:11:13,200 Speaker 5: to behave in a particular way, to use particular data, 236 00:11:13,240 --> 00:11:15,080 Speaker 5: to be able to call out to an external service, 237 00:11:15,880 --> 00:11:17,720 Speaker 5: you can make this thing and you can do all 238 00:11:17,760 --> 00:11:21,240 Speaker 5: sorts of like great stuff with it. And then we 239 00:11:21,400 --> 00:11:23,400 Speaker 5: just recently launched a store where you can see what 240 00:11:23,440 --> 00:11:25,839 Speaker 5: other people have built and you can share it. And 241 00:11:26,800 --> 00:11:29,439 Speaker 5: I mean personally, one that I have loved is All Trails. 242 00:11:29,920 --> 00:11:31,959 Speaker 5: I have this like every other weekend I would like 243 00:11:32,040 --> 00:11:34,200 Speaker 5: to go for a long hike, and there's always like 244 00:11:34,240 --> 00:11:36,079 Speaker 5: the version of Netflix that other people have where it's 245 00:11:36,080 --> 00:11:37,600 Speaker 5: like takes an hour to figure out what to watch. 246 00:11:37,640 --> 00:11:39,000 Speaker 5: It takes me like two hours to figure out what 247 00:11:39,080 --> 00:11:41,679 Speaker 5: hike to do. And the All Trails thing to like 248 00:11:42,040 --> 00:11:43,800 Speaker 5: say I want this, I want that. You know I've 249 00:11:43,800 --> 00:11:45,600 Speaker 5: already done this one, and like here's a great hike. 250 00:11:46,320 --> 00:11:49,040 Speaker 2: It's been. It sounds silly, but I love that one. 251 00:11:49,160 --> 00:11:51,400 Speaker 2: Have yet added any GPTs of your own? Have I 252 00:11:51,400 --> 00:11:51,960 Speaker 2: never read any? 253 00:11:52,040 --> 00:11:52,240 Speaker 1: Yeah? 254 00:11:52,920 --> 00:11:56,640 Speaker 2: I have not put any in this store. Maybe I will, great. 255 00:11:57,240 --> 00:11:59,760 Speaker 1: Can you give us an update on the volume or 256 00:11:59,800 --> 00:12:01,520 Speaker 1: the so what you're seeing new GPTs. 257 00:12:02,120 --> 00:12:03,760 Speaker 5: The number I know is that there have been three 258 00:12:03,800 --> 00:12:05,960 Speaker 5: million created before we launched the store. I have been 259 00:12:06,000 --> 00:12:07,480 Speaker 5: in the middle of this trip around the world that 260 00:12:07,520 --> 00:12:09,679 Speaker 5: has been quite hectic, and I have not been doing 261 00:12:09,679 --> 00:12:12,160 Speaker 5: my normal daily metrics tracking, So I don't know how 262 00:12:12,160 --> 00:12:14,120 Speaker 5: it's gone since launch, but I'll tell you by the 263 00:12:14,160 --> 00:12:19,239 Speaker 5: slowness of chat gpt it's probably doing really well. 264 00:12:19,320 --> 00:12:21,920 Speaker 1: I want to ask you about open AI's copyright issues. 265 00:12:22,280 --> 00:12:26,480 Speaker 1: How important are publisher relations to open AI's business, considering, 266 00:12:26,559 --> 00:12:29,960 Speaker 1: for example, the lawsuit last month filed against open ai 267 00:12:30,080 --> 00:12:31,120 Speaker 1: by the New York Times. 268 00:12:31,520 --> 00:12:33,800 Speaker 5: They are important, but not for the reason people think. 269 00:12:34,080 --> 00:12:37,240 Speaker 5: There is this belief held by some people that man, 270 00:12:37,320 --> 00:12:39,480 Speaker 5: you need all of my training data, and my training 271 00:12:39,520 --> 00:12:43,280 Speaker 5: data is so valuable, and actually that is generally not 272 00:12:43,320 --> 00:12:45,280 Speaker 5: the case. We do not want to train on the 273 00:12:45,320 --> 00:12:49,240 Speaker 5: New York Times data, for example, and a more generally, 274 00:12:49,280 --> 00:12:52,319 Speaker 5: we're getting to a world where it's been like data, data, data, 275 00:12:52,480 --> 00:12:54,040 Speaker 5: you just need more, you need more, you need more. 276 00:12:54,240 --> 00:12:55,680 Speaker 5: You're going to run out of that at some point anyway. 277 00:12:55,720 --> 00:12:57,079 Speaker 5: So a lot of our research has been how can 278 00:12:57,120 --> 00:13:00,360 Speaker 5: we learn more from smaller amounts of very high quality data. 279 00:13:00,400 --> 00:13:02,080 Speaker 5: And I think the world is going to figure that out. 280 00:13:02,440 --> 00:13:04,840 Speaker 5: What we want to do with publishers if they want 281 00:13:05,240 --> 00:13:08,120 Speaker 5: is when one of our users says, what happened to 282 00:13:08,160 --> 00:13:11,200 Speaker 5: Davos today, be able to say, here's an article from Bloomberg, 283 00:13:11,200 --> 00:13:13,040 Speaker 5: here's an from New York Times, and here's like a 284 00:13:13,080 --> 00:13:15,320 Speaker 5: little snippet, or probably want to snippet. There's probably some 285 00:13:15,360 --> 00:13:17,839 Speaker 5: cooler thing that we can do with the technology. And 286 00:13:18,040 --> 00:13:19,960 Speaker 5: you know, some people want to partner with us, some 287 00:13:20,000 --> 00:13:22,559 Speaker 5: people don't. We've been striking a lot of great partnerships 288 00:13:22,559 --> 00:13:25,360 Speaker 5: and we have a lot more coming. And then you know, 289 00:13:25,559 --> 00:13:29,160 Speaker 5: some people don't want to. We'd rather they just say 290 00:13:29,160 --> 00:13:30,880 Speaker 5: we don't want to do that rather than sue us. 291 00:13:30,920 --> 00:13:33,679 Speaker 5: But like we'll we'll end ourselves. But it's fine too. 292 00:13:33,720 --> 00:13:35,560 Speaker 2: I just heard you say you don't want to train 293 00:13:36,040 --> 00:13:36,840 Speaker 2: the New York Times. 294 00:13:37,000 --> 00:13:40,160 Speaker 1: Does that mean, given the legal exposure, you would have 295 00:13:40,200 --> 00:13:41,760 Speaker 1: done things differently if you train your mind? 296 00:13:41,960 --> 00:13:45,599 Speaker 5: Is a triggy thing about that people. The web is 297 00:13:45,640 --> 00:13:47,520 Speaker 5: a big thing, and there are people who like copy 298 00:13:47,559 --> 00:13:49,960 Speaker 5: from the New York Times and put an article without 299 00:13:49,960 --> 00:13:51,880 Speaker 5: attribution up on some website and you don't know that's 300 00:13:51,880 --> 00:13:52,839 Speaker 5: a New York Times article. 301 00:13:53,360 --> 00:13:54,559 Speaker 2: If the New York Times wants to. 302 00:13:54,520 --> 00:13:56,760 Speaker 5: Give us a database of all their articles or someone 303 00:13:56,800 --> 00:13:58,719 Speaker 5: else does and say, hey, don't put anything out, that's 304 00:13:58,760 --> 00:14:00,880 Speaker 5: like a match for this. We can probably do a 305 00:14:00,880 --> 00:14:04,680 Speaker 5: pretty good job and solve we don't want to regurgitate 306 00:14:04,720 --> 00:14:08,200 Speaker 5: someone else's content. But the problem is not as easy 307 00:14:08,200 --> 00:14:10,480 Speaker 5: as it sounds. In a vacuum, I think we can 308 00:14:10,520 --> 00:14:12,160 Speaker 5: get that number down and down and down and have 309 00:14:12,200 --> 00:14:14,520 Speaker 5: to be quite low. And that seems like a super 310 00:14:14,559 --> 00:14:17,040 Speaker 5: reasonable thing to evaluate us on. You know, if you 311 00:14:17,160 --> 00:14:21,080 Speaker 5: have copyrighted content, whether or not it got put into 312 00:14:21,080 --> 00:14:24,000 Speaker 5: someone else's thing without our knowledge, and you're willing to 313 00:14:24,000 --> 00:14:26,040 Speaker 5: show us what it is and say, don't put this 314 00:14:26,080 --> 00:14:27,360 Speaker 5: stuff is a direct response. 315 00:14:27,640 --> 00:14:30,560 Speaker 2: We should be able to do that. Again. 316 00:14:30,600 --> 00:14:33,800 Speaker 5: It won't like thousand you know, monkeys, thousand typewriters, whatever 317 00:14:33,800 --> 00:14:35,200 Speaker 5: it is. Once in a while the model will just 318 00:14:35,240 --> 00:14:37,720 Speaker 5: generate something very close. But on the whole we should 319 00:14:37,720 --> 00:14:40,760 Speaker 5: be able to do a great job with this. So 320 00:14:40,800 --> 00:14:43,040 Speaker 5: there's like there's all the negatives of these people like, ah, 321 00:14:43,080 --> 00:14:46,160 Speaker 5: you know, don't do this, but the positives are I 322 00:14:46,200 --> 00:14:49,120 Speaker 5: think there's going to be great new ways to consume 323 00:14:49,240 --> 00:14:53,240 Speaker 5: and monetize news and other published content. And for every 324 00:14:53,320 --> 00:14:55,680 Speaker 5: one New York time situation we have, we have many 325 00:14:55,680 --> 00:14:58,800 Speaker 5: more super productive things about people that are excited to 326 00:14:58,800 --> 00:15:02,280 Speaker 5: build the future and not do it theatrics and. 327 00:15:02,320 --> 00:15:04,560 Speaker 2: And what about DOLLI. 328 00:15:04,640 --> 00:15:07,120 Speaker 1: I mean, there have been artists who have been upset 329 00:15:07,160 --> 00:15:08,600 Speaker 1: with DOLLI two, DOLLI three. 330 00:15:08,640 --> 00:15:10,680 Speaker 2: What what has that taught you? And how will you 331 00:15:10,720 --> 00:15:11,560 Speaker 2: do things differently? 332 00:15:11,680 --> 00:15:14,640 Speaker 5: We engage with the artist's community a lot, and you know, 333 00:15:14,760 --> 00:15:16,920 Speaker 5: we try to like do their requests. So one is, 334 00:15:17,000 --> 00:15:21,360 Speaker 5: don't generate in my style even if you're not training 335 00:15:21,400 --> 00:15:23,920 Speaker 5: on my data. Super reasonable, so you know, implement things 336 00:15:23,960 --> 00:15:27,760 Speaker 5: like that, you know, let me opt out of training. 337 00:15:27,840 --> 00:15:29,440 Speaker 5: Even if my images are all over the internet, you 338 00:15:29,440 --> 00:15:31,320 Speaker 5: don't know what they are, what I'm And so there's 339 00:15:31,320 --> 00:15:32,960 Speaker 5: a lot of other things to where I'm really excited 340 00:15:33,000 --> 00:15:35,280 Speaker 5: to do and the technology isn't here yet. But get 341 00:15:35,320 --> 00:15:37,360 Speaker 5: to a point where rather than the artists that I 342 00:15:37,360 --> 00:15:39,640 Speaker 5: don't want this thing for these reasons, be able to 343 00:15:39,680 --> 00:15:42,720 Speaker 5: deliver something where an artist can make a great version 344 00:15:42,880 --> 00:15:45,520 Speaker 5: of DOLLI in their style, sell access to that if 345 00:15:45,520 --> 00:15:47,040 Speaker 5: they want, don't if they don't want, Just use it 346 00:15:47,080 --> 00:15:51,520 Speaker 5: for themselves or get some sort of economic benefit or otherwise. 347 00:15:51,880 --> 00:15:54,920 Speaker 5: When someone does use their stuff and it's not just 348 00:15:54,920 --> 00:15:58,720 Speaker 5: training the images, it really is like you know, it 349 00:15:58,760 --> 00:16:02,520 Speaker 5: really is about style. And that's the thing that at 350 00:16:02,600 --> 00:16:05,160 Speaker 5: least in the artist conversations I've had that people are 351 00:16:05,160 --> 00:16:07,480 Speaker 5: super interested in. So for now, it's like, all right, 352 00:16:07,520 --> 00:16:09,640 Speaker 5: let'sen know what people don't want, make sure that we 353 00:16:09,720 --> 00:16:13,000 Speaker 5: respect that. Of course, you can't make everybody happy, but 354 00:16:13,000 --> 00:16:15,440 Speaker 5: try to make the community feel like we're being a 355 00:16:15,440 --> 00:16:18,920 Speaker 5: good partner. But what I think will be better and 356 00:16:18,960 --> 00:16:21,400 Speaker 5: more exciting is when we can do things that artists 357 00:16:21,400 --> 00:16:22,480 Speaker 5: are like, that's awesome. 358 00:16:24,720 --> 00:16:28,840 Speaker 1: Anna, you are open AI's ambassador to Washington to other 359 00:16:29,480 --> 00:16:32,840 Speaker 1: capitals around the world. I'm curious what you've taken from 360 00:16:32,840 --> 00:16:37,280 Speaker 1: your experience a Facebook, what you've taken from the tense 361 00:16:37,400 --> 00:16:40,480 Speaker 1: relations between a lot of tech companies and governments and 362 00:16:40,520 --> 00:16:43,400 Speaker 1: regulators over the past few decades, and how you're. 363 00:16:43,240 --> 00:16:45,680 Speaker 2: Putting that to use now and Open AI. 364 00:16:46,240 --> 00:16:48,120 Speaker 3: I mean, so, I think one thing that I really 365 00:16:48,160 --> 00:16:50,520 Speaker 3: learned working in government, and of course I worked in 366 00:16:50,560 --> 00:16:53,760 Speaker 3: the White House during the twenty sixteen Russia election interference, and. 367 00:16:53,640 --> 00:16:55,360 Speaker 4: People think that that was the first time we'd ever 368 00:16:55,400 --> 00:16:55,720 Speaker 4: heard of. 369 00:16:55,680 --> 00:16:57,800 Speaker 3: It, but it was something that we had actually been 370 00:16:57,840 --> 00:17:00,000 Speaker 3: working on for years and thinking, you know, we know that. 371 00:17:00,040 --> 00:17:02,320 Speaker 4: This happens, what do we do about it? And one 372 00:17:02,360 --> 00:17:03,080 Speaker 4: thing I never. 373 00:17:02,960 --> 00:17:05,360 Speaker 2: Did during that period is go out and talk. 374 00:17:05,200 --> 00:17:07,840 Speaker 3: To the companies because it's not actually a typical thing 375 00:17:07,880 --> 00:17:10,840 Speaker 3: you do in government and was much more rare back then, 376 00:17:10,920 --> 00:17:14,240 Speaker 3: especially with these emerging tools. And I thought about that 377 00:17:14,280 --> 00:17:15,760 Speaker 3: a lot as I entered the tech space that I 378 00:17:15,840 --> 00:17:18,240 Speaker 3: regretted that, and that I wanted governments to be able 379 00:17:18,280 --> 00:17:20,639 Speaker 3: to really understand the technology and how the decisions are 380 00:17:20,680 --> 00:17:23,280 Speaker 3: made by these companies. And also just honestly, when I 381 00:17:23,359 --> 00:17:26,119 Speaker 3: first joined OPENEI, no one of course had heard of 382 00:17:26,160 --> 00:17:30,119 Speaker 3: OpenAI in government for the most part, and I thought. 383 00:17:30,560 --> 00:17:32,480 Speaker 4: Every time I used it, I thought, oh my god, 384 00:17:32,520 --> 00:17:33,040 Speaker 4: if I'd. 385 00:17:32,880 --> 00:17:35,000 Speaker 3: Had this for eight years I was in the administration, 386 00:17:35,200 --> 00:17:37,000 Speaker 3: I could have gotten ten times more done. 387 00:17:37,119 --> 00:17:38,520 Speaker 4: So for me, it was really how do I get 388 00:17:38,520 --> 00:17:39,639 Speaker 4: my colleagues to use. 389 00:17:39,520 --> 00:17:42,560 Speaker 3: It, especially with Opene's mission to make sure these tools 390 00:17:42,600 --> 00:17:45,560 Speaker 3: benefit everyone. I don't think that'll ever happen unless governments 391 00:17:45,600 --> 00:17:48,879 Speaker 3: are incorporating it to serve citizens more efficiently and faster. 392 00:17:49,280 --> 00:17:51,159 Speaker 3: And so this is actually One of the things I've 393 00:17:51,200 --> 00:17:54,280 Speaker 3: been most excited about is to just really get governments 394 00:17:54,280 --> 00:17:55,560 Speaker 3: to use it for everyone's benefit. 395 00:17:55,600 --> 00:17:57,600 Speaker 1: I mean, I'm hearing like a lot of sincerity in 396 00:17:57,680 --> 00:18:00,960 Speaker 1: that pitch. Are regulators receptive to it? 397 00:18:00,960 --> 00:18:01,680 Speaker 2: It feels like a. 398 00:18:01,640 --> 00:18:04,560 Speaker 1: Lot are coming to the conversation, probably a good deal 399 00:18:04,560 --> 00:18:08,119 Speaker 1: of skepticism because of past interactions with Silicon Valley. 400 00:18:08,200 --> 00:18:10,080 Speaker 3: I think I mostly don't even really get to talk 401 00:18:10,080 --> 00:18:12,199 Speaker 3: about it because for the most part, people are interested 402 00:18:12,200 --> 00:18:15,080 Speaker 3: in governance and regulation, and I think that they know 403 00:18:16,160 --> 00:18:18,960 Speaker 3: theoretically that there's a lot of benefits the government. Many 404 00:18:18,960 --> 00:18:21,040 Speaker 3: governments are not quite ready to incorporate. I mean, there 405 00:18:21,080 --> 00:18:24,400 Speaker 3: are exceptions, obviously, people who are really at the forefront, 406 00:18:24,480 --> 00:18:26,439 Speaker 3: So it's not you know, I think often I just 407 00:18:26,440 --> 00:18:28,000 Speaker 3: don't even really get to that conversation. 408 00:18:29,119 --> 00:18:32,080 Speaker 1: So I want to ask you both about the dramatic 409 00:18:32,119 --> 00:18:36,040 Speaker 1: kurn of events in November. Sam, One day, the window 410 00:18:36,200 --> 00:18:37,560 Speaker 1: on these questions will close. 411 00:18:38,480 --> 00:18:39,840 Speaker 2: That is, not have they will? 412 00:18:41,440 --> 00:18:44,080 Speaker 1: I think at some point they probably will, but it 413 00:18:44,080 --> 00:18:46,200 Speaker 1: hasn't happened yet, so it doesn't doesn't matter. 414 00:18:47,320 --> 00:18:47,840 Speaker 2: I guess my. 415 00:18:47,920 --> 00:18:51,879 Speaker 1: Question is is you know, have you addressed the governance 416 00:18:52,160 --> 00:18:56,879 Speaker 1: the governance issues the very unique uh corporate structure at 417 00:18:56,920 --> 00:19:00,200 Speaker 1: open AI with the nonprofit board and the have profit. 418 00:19:01,200 --> 00:19:03,360 Speaker 2: That led to your ulster. 419 00:19:03,640 --> 00:19:06,520 Speaker 5: We're going to focus first on putting a great full 420 00:19:06,520 --> 00:19:09,239 Speaker 5: board in place. I expect us to make a lot 421 00:19:09,240 --> 00:19:11,560 Speaker 5: of progress on that in the coming months, and then 422 00:19:11,680 --> 00:19:14,320 Speaker 5: after that the new board will take a look at 423 00:19:14,320 --> 00:19:15,240 Speaker 5: the government structure. 424 00:19:15,440 --> 00:19:17,240 Speaker 2: But I think we debated. 425 00:19:16,840 --> 00:19:20,359 Speaker 1: What does that mean is that should open AIBF traditional 426 00:19:20,400 --> 00:19:22,200 Speaker 1: Silicon Valley for profit company. 427 00:19:22,320 --> 00:19:25,080 Speaker 5: We will never be a traditional company, but the structure. 428 00:19:25,600 --> 00:19:27,320 Speaker 5: I think we should take a look at the structure. 429 00:19:27,320 --> 00:19:28,960 Speaker 5: Maybe the answer we have now is right, but I 430 00:19:29,000 --> 00:19:31,760 Speaker 5: think we should be willing to consider other things. But 431 00:19:31,760 --> 00:19:33,080 Speaker 5: I think this is not the time for it, and 432 00:19:33,119 --> 00:19:35,239 Speaker 5: the focus on the board first, and then we'll go 433 00:19:35,240 --> 00:19:36,480 Speaker 5: look at it from all angles. 434 00:19:36,880 --> 00:19:41,199 Speaker 1: I mean, presumably you have investors, including Microsoft, including your 435 00:19:41,320 --> 00:19:47,080 Speaker 1: venture capital supporters, your employees, who over the long term, 436 00:19:47,119 --> 00:19:49,560 Speaker 1: are seeking a return on their investment. 437 00:19:51,200 --> 00:19:53,879 Speaker 5: I think one of the things that's difficult to express 438 00:19:53,960 --> 00:19:56,960 Speaker 5: about open AI is the degree to which our team 439 00:19:57,240 --> 00:20:00,560 Speaker 5: and the people around us, investors, Microsoft, whatever, A committed 440 00:20:00,600 --> 00:20:05,000 Speaker 5: to this mission. In the middle of that crazy few days. 441 00:20:06,320 --> 00:20:09,919 Speaker 5: At one point I think like ninety seven something like 442 00:20:09,960 --> 00:20:14,080 Speaker 5: that ninety eight percent of the company signed a letter saying, 443 00:20:14,240 --> 00:20:16,280 Speaker 5: you know, we're all going to resign and go to 444 00:20:16,320 --> 00:20:19,840 Speaker 5: something else, and that would have torched everyone's equity. And 445 00:20:19,880 --> 00:20:21,640 Speaker 5: for a lot of our employees, like this is all 446 00:20:21,880 --> 00:20:24,720 Speaker 5: the great majority of their wealth, and people being willing 447 00:20:24,760 --> 00:20:28,560 Speaker 5: to go do that, I think is quite unusual. Our 448 00:20:28,600 --> 00:20:31,840 Speaker 5: investors who also were about to watch their stakes go 449 00:20:31,920 --> 00:20:34,000 Speaker 5: to zero, which is like, how can we support you 450 00:20:34,000 --> 00:20:36,400 Speaker 5: and whatever is best for the mission Microsoft too. 451 00:20:37,040 --> 00:20:38,800 Speaker 2: I feel very very fortunate about that. 452 00:20:40,480 --> 00:20:41,879 Speaker 5: Of course, I also would like to make all of 453 00:20:41,880 --> 00:20:44,440 Speaker 5: our shareholders a bunch of money, but it was very 454 00:20:44,480 --> 00:20:46,800 Speaker 5: clear to me what people's priorities were, and that meant 455 00:20:46,800 --> 00:20:47,040 Speaker 5: a lot. 456 00:20:47,920 --> 00:20:49,800 Speaker 1: I sort of smiled because you came to the Bloomberg 457 00:20:49,840 --> 00:20:54,719 Speaker 1: Tech conference in last June and Emily Chang asked it 458 00:20:54,760 --> 00:20:57,639 Speaker 1: was something along the lines of why should we trust you, 459 00:20:57,720 --> 00:21:00,400 Speaker 1: and you very candidly says you shouldn't, and you said 460 00:21:00,400 --> 00:21:02,840 Speaker 1: the board should be able to fire me if they want, 461 00:21:03,119 --> 00:21:07,479 Speaker 1: and of course then they did, and you quite definitely 462 00:21:07,680 --> 00:21:08,680 Speaker 1: orchestrated your return. 463 00:21:08,760 --> 00:21:11,720 Speaker 5: Actually, let me tell you something I the board did 464 00:21:11,760 --> 00:21:16,080 Speaker 5: that I was like, I think this is wild, super confused, 465 00:21:16,080 --> 00:21:18,239 Speaker 5: super caught off guard, but this is the structure, and 466 00:21:18,280 --> 00:21:20,359 Speaker 5: I immediately just went to go thinking about what I 467 00:21:20,440 --> 00:21:22,439 Speaker 5: was going to do next. It was not until some 468 00:21:22,480 --> 00:21:24,840 Speaker 5: board members called me the next morning that I even 469 00:21:24,920 --> 00:21:26,160 Speaker 5: thought about really coming back. 470 00:21:27,280 --> 00:21:29,360 Speaker 2: When I asked, you know, you want to come back? 471 00:21:29,760 --> 00:21:33,000 Speaker 5: I want to talk about that, but like the board 472 00:21:33,040 --> 00:21:36,360 Speaker 5: did have all of the power there. Now you know what, 473 00:21:36,480 --> 00:21:38,760 Speaker 5: I'm not going to say that next thing, but I 474 00:21:39,280 --> 00:21:41,240 Speaker 5: think you should continue. 475 00:21:40,960 --> 00:21:44,320 Speaker 3: I know I would also just say that I think 476 00:21:44,359 --> 00:21:46,359 Speaker 3: that there's a lot of narratives out there. It's like, oh, well, 477 00:21:46,400 --> 00:21:48,600 Speaker 3: this was orchestrated by all these other forces. 478 00:21:48,760 --> 00:21:49,720 Speaker 4: It's not accurate. 479 00:21:50,000 --> 00:21:52,320 Speaker 3: I mean, it was the employees of Opening Eye that 480 00:21:52,480 --> 00:21:55,639 Speaker 3: wanted this and that that was the right thing for 481 00:21:55,680 --> 00:21:56,440 Speaker 3: Sam to be back. 482 00:21:56,640 --> 00:21:59,000 Speaker 5: But you know, like, yeah, I think I'll say is 483 00:21:59,680 --> 00:22:02,080 Speaker 5: I think it's important that I have an entity that 484 00:22:03,040 --> 00:22:05,040 Speaker 5: we confide this, but that entity has got to have 485 00:22:05,080 --> 00:22:09,520 Speaker 5: some accountability too, and that is a clear issue. 486 00:22:09,160 --> 00:22:10,320 Speaker 2: With what happened. 487 00:22:10,480 --> 00:22:13,680 Speaker 1: Right Anna, you wrote a remarkable letter to employees during 488 00:22:13,720 --> 00:22:16,159 Speaker 1: the saga, and one of the many reasons I was 489 00:22:16,160 --> 00:22:19,679 Speaker 1: excited to have you on stage today was to just 490 00:22:19,760 --> 00:22:22,720 Speaker 1: ask you what were those five days like for you? 491 00:22:22,760 --> 00:22:25,560 Speaker 2: And why did you step up and write that. Honor 492 00:22:25,600 --> 00:22:27,439 Speaker 2: can clearly answer this if she wants to. 493 00:22:27,480 --> 00:22:29,600 Speaker 5: But like, is really what you want to spend our 494 00:22:29,680 --> 00:22:31,920 Speaker 5: time on, like the soap opera rather than like what 495 00:22:32,119 --> 00:22:32,840 Speaker 5: HI is going to do? 496 00:22:32,880 --> 00:22:36,280 Speaker 2: I mean, I'm wrapping it up, but I mean I 497 00:22:36,320 --> 00:22:38,359 Speaker 2: think people are interested. Oh well, we can leave it 498 00:22:38,359 --> 00:22:41,640 Speaker 2: here if you want to. YEA, let's question the time 499 00:22:41,960 --> 00:22:42,879 Speaker 2: whe we can move on. 500 00:22:43,720 --> 00:22:46,480 Speaker 3: I would just say, for color that it happened the 501 00:22:46,560 --> 00:22:48,560 Speaker 3: day before the entire company was supposed to take a 502 00:22:48,560 --> 00:22:51,280 Speaker 3: week off, so we were all on Friday preparing to, 503 00:22:51,440 --> 00:22:54,200 Speaker 3: you know, have a RESTful week after an insane year. 504 00:22:54,320 --> 00:22:56,040 Speaker 4: So then you know many of us slept on the 505 00:22:56,040 --> 00:22:57,159 Speaker 4: floor of the office for a week. 506 00:22:57,760 --> 00:23:00,639 Speaker 2: Right, There's a question here that I think is a 507 00:23:00,680 --> 00:23:02,160 Speaker 2: really good one. We are at Davos. 508 00:23:02,600 --> 00:23:07,480 Speaker 1: Climate change is on the agenda. The question is does well, 509 00:23:07,520 --> 00:23:10,040 Speaker 1: I'm going to give it a different spin. Considering the 510 00:23:10,200 --> 00:23:15,120 Speaker 1: compute costs and the need for chips, does the development 511 00:23:15,160 --> 00:23:17,399 Speaker 1: of AI and the path to ag I threatened to 512 00:23:17,440 --> 00:23:21,520 Speaker 1: take us in the opposite direction on the climate. 513 00:23:24,359 --> 00:23:28,200 Speaker 5: We do need way more energy in the world than 514 00:23:28,240 --> 00:23:30,960 Speaker 5: I think we thought we needed before. My whole model 515 00:23:31,000 --> 00:23:34,159 Speaker 5: of the world is that the two important currencies of 516 00:23:34,200 --> 00:23:39,480 Speaker 5: the future are compute slash intelligence and energy. You know, 517 00:23:39,560 --> 00:23:42,119 Speaker 5: the ideas that we want and the ability to make 518 00:23:42,160 --> 00:23:45,960 Speaker 5: stuff happen and the ability to run the compute. And 519 00:23:46,040 --> 00:23:49,800 Speaker 5: I think we still don't appreciate the energy needs of 520 00:23:49,840 --> 00:23:54,760 Speaker 5: this technology. The good news, to the degree there's good news, 521 00:23:54,800 --> 00:23:57,320 Speaker 5: is there's no way to get there without a breakthrough. 522 00:23:57,640 --> 00:24:00,440 Speaker 5: We need fusion, or we need like radically cheaper solar 523 00:24:00,480 --> 00:24:03,879 Speaker 5: plus storage or something at massive scale, like a scale 524 00:24:03,880 --> 00:24:05,439 Speaker 5: that no one is really planning for. 525 00:24:06,680 --> 00:24:08,080 Speaker 2: So we. 526 00:24:09,920 --> 00:24:11,960 Speaker 5: It's totally fair to say that AI is going to 527 00:24:12,000 --> 00:24:14,000 Speaker 5: need a lot of energy, but it will force us, 528 00:24:14,040 --> 00:24:17,239 Speaker 5: I think, to invest more in the technologies that can 529 00:24:17,240 --> 00:24:19,400 Speaker 5: deliver this, none of which are the ones that are 530 00:24:19,880 --> 00:24:22,560 Speaker 5: burning the carbon, Like that'll be those all those unbelievable 531 00:24:22,600 --> 00:24:23,480 Speaker 5: number of fuel trucks. 532 00:24:23,480 --> 00:24:25,919 Speaker 2: And by the way, you back all pro one or 533 00:24:25,920 --> 00:24:26,640 Speaker 2: more nuclear. 534 00:24:27,480 --> 00:24:33,240 Speaker 5: Yeah, I personally think that is either the most likely 535 00:24:33,320 --> 00:24:35,000 Speaker 5: or the second most likely approach. 536 00:24:35,119 --> 00:24:37,439 Speaker 1: I feel like the world is more receptive to that 537 00:24:37,520 --> 00:24:40,480 Speaker 1: technology now, certainly historically not in the US. 538 00:24:40,760 --> 00:24:45,520 Speaker 5: I think the world is still unfortunately pretty negative on 539 00:24:45,600 --> 00:24:49,240 Speaker 5: fission super positive on fusion. It's a much easier story, 540 00:24:50,240 --> 00:24:53,080 Speaker 5: but I wish the world would embrace fish in much more. 541 00:24:54,200 --> 00:24:56,760 Speaker 5: I look, I may be too optimistic about this, but 542 00:24:56,840 --> 00:25:02,120 Speaker 5: I think I think we have paths now to massive, 543 00:25:03,760 --> 00:25:06,240 Speaker 5: a massive energy transition away from burning carbon. 544 00:25:06,520 --> 00:25:07,200 Speaker 2: It'll take a while. 545 00:25:07,200 --> 00:25:09,480 Speaker 5: Those cars are going to keep driving, you know, there's 546 00:25:09,560 --> 00:25:10,760 Speaker 5: all of the transport stuff. 547 00:25:11,160 --> 00:25:11,760 Speaker 2: It'll be a while. 548 00:25:11,840 --> 00:25:13,920 Speaker 5: So there's like a fusion reactor in every cargo ship. 549 00:25:14,680 --> 00:25:17,080 Speaker 5: But if we can drop the cost of energy as 550 00:25:17,119 --> 00:25:20,359 Speaker 5: dramatically as I hope we can, then the math on 551 00:25:20,480 --> 00:25:22,760 Speaker 5: carbon capture just so changes. 552 00:25:23,960 --> 00:25:25,080 Speaker 2: I still expect. 553 00:25:25,720 --> 00:25:27,480 Speaker 5: Unfortunately, the world is on a path we're going to 554 00:25:27,520 --> 00:25:31,600 Speaker 5: have to do something dramatic with climate like geoengineering as 555 00:25:31,680 --> 00:25:34,560 Speaker 5: a as a band aid, as a stop gap. But 556 00:25:34,640 --> 00:25:36,199 Speaker 5: I think we do not see a path to the 557 00:25:36,240 --> 00:25:36,960 Speaker 5: long term solution. 558 00:25:37,280 --> 00:25:39,600 Speaker 1: So I want to just go back to my question 559 00:25:40,359 --> 00:25:42,879 Speaker 1: in terms of moving in the opposite direction. It sounds 560 00:25:42,880 --> 00:25:46,440 Speaker 1: like the answer is potentially yes on the demand side 561 00:25:46,880 --> 00:25:50,960 Speaker 1: unless we take graphic action on the supply. 562 00:25:51,080 --> 00:25:54,119 Speaker 5: But there there is no I see no way to 563 00:25:54,119 --> 00:25:59,640 Speaker 5: supply this with to manage the supply side without a really. 564 00:25:59,440 --> 00:26:02,760 Speaker 2: Big breakthrough, right, which is does this frighten you guys? 565 00:26:02,920 --> 00:26:07,600 Speaker 1: Because you know, the world hasn't been that versatile when 566 00:26:07,640 --> 00:26:10,719 Speaker 1: it comes to supply, but AI, as you know you 567 00:26:10,760 --> 00:26:13,120 Speaker 1: have pointed out, it's not going to take its time 568 00:26:13,200 --> 00:26:14,800 Speaker 1: until we start generating enough power. 569 00:26:14,880 --> 00:26:17,679 Speaker 5: It motivates us to go invest more in fusion and 570 00:26:17,720 --> 00:26:21,520 Speaker 5: invest more in norse new storage, and not only the technology, 571 00:26:21,800 --> 00:26:23,960 Speaker 5: but what it's going to take to deliver this at 572 00:26:23,960 --> 00:26:27,040 Speaker 5: the scale that AI needs and that the whole globe needs. 573 00:26:27,480 --> 00:26:29,760 Speaker 5: So I think it would be not helpful for us 574 00:26:29,760 --> 00:26:32,520 Speaker 5: to just sit there and be nervous. We're just like, hey, 575 00:26:32,600 --> 00:26:35,280 Speaker 5: we see what's coming with very high conviction it's coming. 576 00:26:35,760 --> 00:26:40,879 Speaker 5: How can we use our abilities, our capital or whatever 577 00:26:40,920 --> 00:26:42,960 Speaker 5: else to do this and in the process of that 578 00:26:43,000 --> 00:26:45,080 Speaker 5: hopefully deliver a solution for the rest of the world, 579 00:26:45,119 --> 00:26:48,240 Speaker 5: not just AI training workloads or inference workloads. 580 00:26:48,600 --> 00:26:50,560 Speaker 1: And I felt like in twenty twenty three we had 581 00:26:50,560 --> 00:26:55,800 Speaker 1: the beginning of an almost hypothetical conversation about regulating AI. 582 00:26:56,440 --> 00:27:00,560 Speaker 1: What should we expect in twenty twenty four and you know, 583 00:27:00,640 --> 00:27:04,680 Speaker 1: does it do the Governments actes? Does it become real? 584 00:27:04,800 --> 00:27:06,800 Speaker 1: And what does AI safety look like? 585 00:27:07,560 --> 00:27:09,720 Speaker 4: So I think it is becoming real. 586 00:27:09,920 --> 00:27:12,960 Speaker 3: You know that you is on the cusp of actually 587 00:27:13,000 --> 00:27:16,160 Speaker 3: finalizing this regulation, which is going to be quite extensive, 588 00:27:16,600 --> 00:27:19,880 Speaker 3: and the bid. The Administration wrote the longest executive order 589 00:27:19,920 --> 00:27:21,680 Speaker 3: I think in the history of executive. 590 00:27:21,359 --> 00:27:23,560 Speaker 4: Orders covering this technology, and. 591 00:27:23,800 --> 00:27:26,959 Speaker 3: It's being implemented in twenty twenty four because they gave agencies, 592 00:27:27,000 --> 00:27:29,720 Speaker 3: you know, a bunch of homework for how to implement 593 00:27:29,760 --> 00:27:31,000 Speaker 3: this and governing this technology. 594 00:27:31,040 --> 00:27:34,359 Speaker 4: And it's happening. So I think it is really moving forward. 595 00:27:35,280 --> 00:27:38,080 Speaker 3: But what exactly safety looks like, of what it even is, 596 00:27:38,160 --> 00:27:39,760 Speaker 3: I think this is still a conversation we. 597 00:27:39,720 --> 00:27:41,199 Speaker 4: Haven't bottomed out on. 598 00:27:41,359 --> 00:27:43,680 Speaker 3: You know, we've founded this frenchier model form in part 599 00:27:44,040 --> 00:27:46,880 Speaker 3: maybe explain what yeah, so this is for now, this 600 00:27:46,920 --> 00:27:52,040 Speaker 3: is Microsoft opening Eyeanthropic and Google, but it will I think, 601 00:27:52,080 --> 00:27:55,480 Speaker 3: expand to other frontier labs. But really, right now, all 602 00:27:55,480 --> 00:27:57,800 Speaker 3: of us are working on safety. We all read some 603 00:27:57,840 --> 00:28:00,320 Speaker 3: of our models, we all do a lot of this work, 604 00:28:00,400 --> 00:28:03,800 Speaker 3: but we really don't have even a common vocabulary or 605 00:28:03,800 --> 00:28:06,840 Speaker 3: a standardized approach. And to the extent that people think like, well, 606 00:28:06,880 --> 00:28:09,960 Speaker 3: this is just industry, but this is in part and 607 00:28:10,000 --> 00:28:12,919 Speaker 3: response to many governments that have asked us for this 608 00:28:13,040 --> 00:28:15,480 Speaker 3: very thing. So like, what is it across industry that 609 00:28:15,560 --> 00:28:18,399 Speaker 3: you think are viable best practices. 610 00:28:19,119 --> 00:28:24,200 Speaker 1: Is there a risk that regulation starts to discourage entrepreneurial 611 00:28:24,200 --> 00:28:25,520 Speaker 1: activity in AI? 612 00:28:25,800 --> 00:28:29,080 Speaker 3: I mean, I think people are terrified of this. This 613 00:28:29,240 --> 00:28:32,920 Speaker 3: is why I think Germany and France and Italy interjected 614 00:28:33,000 --> 00:28:36,720 Speaker 3: into the eu AI Act discussion because they are really 615 00:28:36,760 --> 00:28:39,960 Speaker 3: concerned about their own domestic industries being sort of undercut 616 00:28:39,960 --> 00:28:42,000 Speaker 3: before they've even had a chance to develop. 617 00:28:42,600 --> 00:28:46,160 Speaker 1: Were you satisfied with your old boss's executive order and 618 00:28:46,280 --> 00:28:50,000 Speaker 1: was there anything in there that you had lobbied against? 619 00:28:50,720 --> 00:28:54,000 Speaker 3: No, And in fact, you know, I think it was 620 00:28:54,080 --> 00:28:56,960 Speaker 3: really good and that it wasn't just these are the restrictions, 621 00:28:56,960 --> 00:28:59,760 Speaker 3: it's like, and then also please go and think about 622 00:28:59,760 --> 00:29:03,360 Speaker 3: how agency will actually leverage this to do your work better. 623 00:29:03,680 --> 00:29:06,880 Speaker 3: So I was really encouraged that they actually did have 624 00:29:07,040 --> 00:29:07,960 Speaker 3: a balanced approach. 625 00:29:10,040 --> 00:29:12,320 Speaker 2: Sam, first time a dabos fursan. Okay, is. 626 00:29:14,480 --> 00:29:17,000 Speaker 1: You mentioned that you'd prefer to spend more of our 627 00:29:17,040 --> 00:29:19,560 Speaker 1: time here on stage talking about AGI. What is the 628 00:29:19,680 --> 00:29:23,320 Speaker 1: message you're bringing to political leaders and other business leaders 629 00:29:23,360 --> 00:29:24,560 Speaker 1: here if you could distill it? 630 00:29:24,600 --> 00:29:24,960 Speaker 2: Thank you. 631 00:29:28,840 --> 00:29:28,959 Speaker 1: So. 632 00:29:29,680 --> 00:29:31,680 Speaker 5: I think twenty twenty three was a year where the 633 00:29:31,720 --> 00:29:35,600 Speaker 5: world woke up to the possibility of these systems becoming 634 00:29:35,680 --> 00:29:40,520 Speaker 5: increasingly capable and increasingly general. But GPT four I think 635 00:29:40,640 --> 00:29:45,320 Speaker 5: is best understood as a preview, and it was more 636 00:29:45,440 --> 00:29:48,160 Speaker 5: over the bar than we expected of utility for more 637 00:29:48,160 --> 00:29:51,680 Speaker 5: people in more ways. But you know, it's easy to 638 00:29:51,720 --> 00:29:55,080 Speaker 5: point out the limitations. And again we're thrilled that people 639 00:29:55,200 --> 00:29:56,760 Speaker 5: love it and use it as much as they do. 640 00:29:57,560 --> 00:30:02,000 Speaker 5: But this is progress here is not linear. And this 641 00:30:02,120 --> 00:30:04,880 Speaker 5: is the thing that I think is really tricky. Humans 642 00:30:04,880 --> 00:30:08,160 Speaker 5: have horrible intuition for exponentials, at least speaking for myself, 643 00:30:08,160 --> 00:30:10,560 Speaker 5: but it seems like a common part of the human condition. 644 00:30:11,640 --> 00:30:13,920 Speaker 5: What does it mean if GPT five is as much 645 00:30:13,960 --> 00:30:16,240 Speaker 5: better than GPT four is four to us to three 646 00:30:16,320 --> 00:30:18,440 Speaker 5: and six is to five, And what does it mean 647 00:30:18,440 --> 00:30:20,240 Speaker 5: if we're just on this trajectory now? 648 00:30:22,000 --> 00:30:23,000 Speaker 2: What you know? 649 00:30:23,000 --> 00:30:24,760 Speaker 5: On the question of regulation, I think it's great that 650 00:30:24,760 --> 00:30:27,200 Speaker 5: different countries are going to try different things. Some countries 651 00:30:27,200 --> 00:30:30,080 Speaker 5: will probably banning AI, some countries will probably say no 652 00:30:30,160 --> 00:30:32,240 Speaker 5: guardrails at all. Both of those I think will turn 653 00:30:32,240 --> 00:30:35,000 Speaker 5: out to be suboptimal, and we'll get to see different 654 00:30:35,000 --> 00:30:40,719 Speaker 5: things work. But as these systems become more powerful, as 655 00:30:40,840 --> 00:30:43,400 Speaker 5: they as they become more deeply integrated to the economy 656 00:30:43,400 --> 00:30:45,400 Speaker 5: as they become something we all used to do our work, 657 00:30:45,920 --> 00:30:48,040 Speaker 5: and then as things beyond that happen, as they become 658 00:30:48,520 --> 00:30:53,720 Speaker 5: capable of discovering new scientific knowledge for humanity, even as 659 00:30:53,760 --> 00:30:55,480 Speaker 5: they become capable of doing AI research. 660 00:30:55,480 --> 00:30:56,240 Speaker 2: At some point. 661 00:30:57,800 --> 00:31:01,800 Speaker 5: The world is gonna change more slowly and then more 662 00:31:01,880 --> 00:31:03,840 Speaker 5: quickly than we might imagine. But the world is going 663 00:31:03,880 --> 00:31:07,920 Speaker 5: to change. This is you know, A thing I always 664 00:31:07,920 --> 00:31:09,560 Speaker 5: say to people is no one knows what happens next. 665 00:31:09,600 --> 00:31:11,360 Speaker 5: And I really believe that, and I think keeping the 666 00:31:11,440 --> 00:31:13,840 Speaker 5: humility about that is really important. You can see a 667 00:31:13,880 --> 00:31:18,000 Speaker 5: few steps in front of you, but not too many. 668 00:31:18,040 --> 00:31:22,480 Speaker 5: But when cognition, when the cost of cognition falls by 669 00:31:22,520 --> 00:31:24,920 Speaker 5: a factor of a thousand or a million, when the 670 00:31:24,960 --> 00:31:29,040 Speaker 5: capability of it becomes, it augments us in ways we 671 00:31:29,080 --> 00:31:33,320 Speaker 5: can't even imagine, you know. Like one example I try 672 00:31:33,320 --> 00:31:35,200 Speaker 5: to give to people is, what if everybody in the 673 00:31:35,200 --> 00:31:39,440 Speaker 5: world had a really competent company of ten thousand great 674 00:31:39,600 --> 00:31:42,880 Speaker 5: virtual employees, experts in every area. They never follow with 675 00:31:42,920 --> 00:31:45,880 Speaker 5: each other, they didn't need to rest, they got really smart, 676 00:31:45,880 --> 00:31:48,160 Speaker 5: they got smarter at this rapid pace. What would we 677 00:31:48,200 --> 00:31:50,240 Speaker 5: be able to create for each other? What would that 678 00:31:50,240 --> 00:31:53,240 Speaker 5: do to the world that we experience. And the answer is, 679 00:31:53,280 --> 00:31:55,680 Speaker 5: none of us know. Of course, none of us have 680 00:31:55,760 --> 00:31:58,560 Speaker 5: these strong intuitions for that. I can imagine it sort of, 681 00:31:59,280 --> 00:32:03,840 Speaker 5: but it's not like a clear picture. And this is 682 00:32:03,880 --> 00:32:06,680 Speaker 5: going to happen. It doesn't mean we don't get to 683 00:32:06,680 --> 00:32:08,360 Speaker 5: steer it. It doesn't mean we don't get to work 684 00:32:08,400 --> 00:32:10,040 Speaker 5: really hard to make it safe and to do it 685 00:32:10,040 --> 00:32:12,360 Speaker 5: in a responsible way. But we are going to go 686 00:32:12,400 --> 00:32:14,880 Speaker 5: to the future, and I think the best way to 687 00:32:14,960 --> 00:32:18,800 Speaker 5: get there in a way that works is the level 688 00:32:18,800 --> 00:32:19,880 Speaker 5: of engagement. 689 00:32:19,400 --> 00:32:19,880 Speaker 2: We now have. 690 00:32:20,240 --> 00:32:21,840 Speaker 5: Part of the reason, a big part of the reason 691 00:32:21,840 --> 00:32:25,560 Speaker 5: we believe in iterative deployment of our technology is that 692 00:32:25,680 --> 00:32:28,800 Speaker 5: people need time to gradually get used to it, to 693 00:32:28,880 --> 00:32:31,440 Speaker 5: understand it. We need time to make mistakes while the 694 00:32:31,560 --> 00:32:34,640 Speaker 5: stakes are low. Governments need time to make some policy mistakes. 695 00:32:34,960 --> 00:32:38,440 Speaker 5: And also technology and society have to coevolve in a 696 00:32:38,440 --> 00:32:42,040 Speaker 5: case like this. So technology is going to change with 697 00:32:42,120 --> 00:32:44,479 Speaker 5: each iteration, but so is the way society works. And 698 00:32:44,520 --> 00:32:48,920 Speaker 5: that's got to be this interactive iterative process, and we 699 00:32:48,960 --> 00:32:51,880 Speaker 5: need to embrace it, but have caution without fear. 700 00:32:51,960 --> 00:32:54,880 Speaker 2: And how long do we have for this iterative process 701 00:32:54,880 --> 00:32:55,360 Speaker 2: to play. 702 00:32:55,720 --> 00:32:59,360 Speaker 5: I think it's surprisingly continuous. I don't like if I 703 00:32:59,400 --> 00:33:01,760 Speaker 5: try to think about discontinuities. I can sort of see 704 00:33:01,760 --> 00:33:05,960 Speaker 5: one when AI can do really good AI research, and 705 00:33:06,000 --> 00:33:07,600 Speaker 5: I can see a few others too, But that's like 706 00:33:07,640 --> 00:33:11,480 Speaker 5: an evocative example. But on the whole, I don't think 707 00:33:11,520 --> 00:33:14,680 Speaker 5: it's about like crossing this one line. I think it's 708 00:33:14,720 --> 00:33:19,120 Speaker 5: about this continuous, exponential curve we'd climb together. And so 709 00:33:19,280 --> 00:33:22,239 Speaker 5: how long do we have? Like no time at all 710 00:33:22,240 --> 00:33:27,120 Speaker 5: in infinite. 711 00:33:25,480 --> 00:33:29,360 Speaker 1: I saw GPT five trending on X earlier this week 712 00:33:29,400 --> 00:33:32,120 Speaker 1: and I clicked, and I, you know, couldn't I It sounded, 713 00:33:32,960 --> 00:33:36,480 Speaker 1: you know, probably misinformed. But what what can you tell 714 00:33:36,560 --> 00:33:40,760 Speaker 1: us about GPT five And is it an exponential. 715 00:33:41,640 --> 00:33:43,960 Speaker 2: You know, improvement over what we see? Look, I don't 716 00:33:44,000 --> 00:33:45,840 Speaker 2: know what we're going to call our next model. I 717 00:33:45,880 --> 00:33:47,360 Speaker 2: don't know when you're going to get creative with the 718 00:33:47,920 --> 00:33:48,720 Speaker 2: naming process. 719 00:33:49,680 --> 00:33:53,640 Speaker 5: I don't want to be like shipping iPhone twenty seven, 720 00:33:55,840 --> 00:34:00,320 Speaker 5: so you know it's not my style quite But I 721 00:34:00,360 --> 00:34:04,280 Speaker 5: think the next model we release, I expect it to 722 00:34:04,320 --> 00:34:06,880 Speaker 5: be very impressive, to do new things that were not 723 00:34:06,920 --> 00:34:09,279 Speaker 5: possible with GPT four, to do a lot of things better, 724 00:34:09,600 --> 00:34:11,400 Speaker 5: And I expect us to like take our time and 725 00:34:11,440 --> 00:34:14,040 Speaker 5: make sure we can launch something that we feel good about, 726 00:34:14,239 --> 00:34:15,200 Speaker 5: I'm responsible about. 727 00:34:16,040 --> 00:34:20,680 Speaker 1: Within open AI, some employees consider themselves to be quote 728 00:34:20,920 --> 00:34:21,840 Speaker 1: building God. 729 00:34:22,480 --> 00:34:24,680 Speaker 2: Is that Is that okay? 730 00:34:26,280 --> 00:34:29,560 Speaker 5: I mean I've heard like people say that facetiously, but 731 00:34:31,040 --> 00:34:35,279 Speaker 5: I think almost all employees would say they're building a 732 00:34:35,360 --> 00:34:37,920 Speaker 5: tool more so than they thought they were going to be, 733 00:34:37,960 --> 00:34:40,480 Speaker 5: which they're thrilled about. You know this confusion in the 734 00:34:40,480 --> 00:34:42,880 Speaker 5: industry of are we building a creature? Are we building 735 00:34:42,920 --> 00:34:45,799 Speaker 5: a tool? I think we're much more building a tool. 736 00:34:45,800 --> 00:34:50,799 Speaker 5: And that's much better to try addition to something. You know, No, no, no, no, 737 00:34:50,800 --> 00:34:52,480 Speaker 5: no no, I finished your thought. Oh, I was just 738 00:34:52,480 --> 00:34:56,920 Speaker 5: going to say, like the we think of ourselves as 739 00:34:56,960 --> 00:35:01,920 Speaker 5: tool builders. AI is much more of a tool than 740 00:35:01,960 --> 00:35:04,279 Speaker 5: a product, and much much more of a tool than 741 00:35:04,480 --> 00:35:09,920 Speaker 5: this like entity. And one of the most wonderful things 742 00:35:09,920 --> 00:35:13,160 Speaker 5: about last year was seen just how much people around 743 00:35:13,160 --> 00:35:15,640 Speaker 5: the world could do with that tool, and they astonished us. 744 00:35:15,719 --> 00:35:17,880 Speaker 5: And I think we'll just see more and more human 745 00:35:17,880 --> 00:35:22,719 Speaker 5: creativity and ability to do more with better tools is remarkable. 746 00:35:23,040 --> 00:35:25,640 Speaker 1: And before we have to start wrapping up now, there 747 00:35:25,680 --> 00:35:27,680 Speaker 1: was a report that you were working with Johnny ive 748 00:35:28,440 --> 00:35:33,040 Speaker 1: on an AI powered device, either within Open Ai, perhaps 749 00:35:33,120 --> 00:35:34,200 Speaker 1: as a separate company. 750 00:35:34,520 --> 00:35:37,359 Speaker 2: And I bring it up because CEES was earlier this. 751 00:35:37,320 --> 00:35:41,480 Speaker 1: Month and AI powered devices were the talk of the conference. 752 00:35:41,680 --> 00:35:43,839 Speaker 2: You know, can you give us an update on that 753 00:35:44,239 --> 00:35:46,520 Speaker 2: and are we approach Does AI bring us to the 754 00:35:46,560 --> 00:35:47,520 Speaker 2: beginning of the end. 755 00:35:47,400 --> 00:35:52,480 Speaker 5: Of the smartphone era? Smartphones are fantastic. I don't think smartphones. 756 00:35:51,960 --> 00:35:52,640 Speaker 2: Are going anywhere. 757 00:35:53,800 --> 00:35:55,359 Speaker 5: I think what they do they do really really well, 758 00:35:55,360 --> 00:35:57,960 Speaker 5: and they're very general. If there is a new thing 759 00:35:58,040 --> 00:36:01,440 Speaker 5: to make, I don't think it replaces a smartphone in 760 00:36:01,440 --> 00:36:03,759 Speaker 5: the way I don't think smartphones replace computers. But if 761 00:36:03,760 --> 00:36:05,640 Speaker 5: there's a new thing to make that helps us do 762 00:36:05,840 --> 00:36:09,040 Speaker 5: more better, you know, in a new way. Given that 763 00:36:09,080 --> 00:36:12,480 Speaker 5: we have this unbelievable change, like I don't think we 764 00:36:12,560 --> 00:36:15,080 Speaker 5: quite I don't spend enough time. I think like marveling 765 00:36:15,080 --> 00:36:17,040 Speaker 5: at the fact that we can now talk to computers 766 00:36:17,080 --> 00:36:18,880 Speaker 5: and they understand us and do stuff for us, Like 767 00:36:19,080 --> 00:36:21,400 Speaker 5: it is a new affordance, a new way to use 768 00:36:21,400 --> 00:36:24,120 Speaker 5: a computer. And if we can do something great there 769 00:36:25,080 --> 00:36:27,239 Speaker 5: a new kind of computer, we should do that. And 770 00:36:27,280 --> 00:36:29,000 Speaker 5: if it turns out that the smartphone is really good 771 00:36:29,040 --> 00:36:31,919 Speaker 5: and this is all software, then fine, but I bet 772 00:36:31,960 --> 00:36:33,520 Speaker 5: there is something great to be done. 773 00:36:33,280 --> 00:36:36,200 Speaker 2: And the partnership with Johnny. 774 00:36:36,360 --> 00:36:39,720 Speaker 1: Is at an open AI effort is then another company. 775 00:36:39,400 --> 00:36:41,279 Speaker 2: I have not heard anything official about a partnership with 776 00:36:41,360 --> 00:36:47,000 Speaker 2: done that Auto. I'm going to give you the last word. 777 00:36:47,320 --> 00:36:50,640 Speaker 1: As you and Sam meet with business and world leaders 778 00:36:50,640 --> 00:36:53,000 Speaker 1: here at Davos, what's the message you want to lead 779 00:36:53,040 --> 00:36:53,399 Speaker 1: them with. 780 00:36:56,120 --> 00:37:00,440 Speaker 3: I think that there is a trend where people feel 781 00:37:00,680 --> 00:37:03,480 Speaker 3: more fear than excitement about this technology, and I understand 782 00:37:03,480 --> 00:37:05,680 Speaker 3: that we have to work very hard to make sure 783 00:37:05,760 --> 00:37:08,480 Speaker 3: that the best version of this technology is realized. But 784 00:37:08,560 --> 00:37:11,279 Speaker 3: I do think that many people are engaging with this 785 00:37:11,960 --> 00:37:15,160 Speaker 3: via the leaders here, and that they really have responsibility 786 00:37:15,160 --> 00:37:18,240 Speaker 3: to make sure that they are setting a balanced message 787 00:37:18,239 --> 00:37:21,799 Speaker 3: so that people can really actually engage with it and 788 00:37:21,960 --> 00:37:23,799 Speaker 3: realize the benefit of this technology. 789 00:37:23,840 --> 00:37:25,640 Speaker 2: Can I have twenty seconds? Absolutely. 790 00:37:25,920 --> 00:37:27,799 Speaker 5: One of the things that I think Openey has not 791 00:37:27,840 --> 00:37:30,240 Speaker 5: always done right in the field. Hasn't either is find 792 00:37:30,280 --> 00:37:33,600 Speaker 5: a way to build these tools in a way and 793 00:37:33,640 --> 00:37:37,640 Speaker 5: also talk about them that don't get that kind of response. 794 00:37:37,920 --> 00:37:39,960 Speaker 5: I think CHATGBT one of the best things it did 795 00:37:40,040 --> 00:37:42,520 Speaker 5: is it shifted the conversation of the positive not because 796 00:37:42,520 --> 00:37:44,640 Speaker 5: we said trust us it'll be great. But because people 797 00:37:44,800 --> 00:37:46,640 Speaker 5: used it and I are like, oh, I get this, 798 00:37:46,719 --> 00:37:48,880 Speaker 5: I use this in a very natural way. The smartphone 799 00:37:48,880 --> 00:37:50,480 Speaker 5: was cool because I didn't have to use a keyboard 800 00:37:50,480 --> 00:37:52,520 Speaker 5: and phone. I could use it more naturally. Talking is 801 00:37:52,560 --> 00:37:55,960 Speaker 5: even more natural. Speaking of Johnny, Johnny's a genius, and 802 00:37:56,000 --> 00:37:57,600 Speaker 5: one of the things that I think he has done 803 00:37:57,960 --> 00:38:01,200 Speaker 5: again and again about computers is figuring out a way 804 00:38:01,280 --> 00:38:05,279 Speaker 5: to make them very human compatible. And I think that's 805 00:38:05,320 --> 00:38:09,800 Speaker 5: super important with this technology, making this feel like, you know, 806 00:38:09,960 --> 00:38:12,280 Speaker 5: not this mystical thing from sci fi, not this scary 807 00:38:12,280 --> 00:38:15,160 Speaker 5: thing from sci fi, but this new way to use 808 00:38:15,200 --> 00:38:18,319 Speaker 5: a computer that you love, and it really feels like 809 00:38:18,360 --> 00:38:21,719 Speaker 5: I still remember the first iMac I got and what 810 00:38:21,760 --> 00:38:22,440 Speaker 5: that felt. 811 00:38:22,200 --> 00:38:23,520 Speaker 2: Like to me. Relative and it's heavy. 812 00:38:24,160 --> 00:38:26,000 Speaker 5: It was heavy, but the fact that it had that handle, 813 00:38:26,000 --> 00:38:27,360 Speaker 5: even though it was like a kid, it was very 814 00:38:27,400 --> 00:38:30,759 Speaker 5: heavy to carry it did mean that I was like 815 00:38:31,160 --> 00:38:33,000 Speaker 5: I had a different relationship with it because of that 816 00:38:33,080 --> 00:38:35,719 Speaker 5: handle and because of the way it looked. I was like, oh, 817 00:38:35,760 --> 00:38:38,040 Speaker 5: I can move this thing around. I could unplug it 818 00:38:38,080 --> 00:38:39,279 Speaker 5: and throw it out the window if it tried to 819 00:38:39,320 --> 00:38:42,840 Speaker 5: like wake up and take over. That's nice, and I 820 00:38:42,880 --> 00:38:45,640 Speaker 5: think the way we design, our technology and our products 821 00:38:45,640 --> 00:38:46,440 Speaker 5: really does matter.