1 00:00:01,210 --> 00:00:06,664 And now our CEO and chair, Mr. Marc Benioff. All right. 2 00:00:06,747 --> 00:00:10,087 Thank you so much everybody. Great 3 00:00:10,170 --> 00:00:16,567 to have you here. How are you all doing? Are you excited to be at Dreamforce? How 4 00:00:16,650 --> 00:00:23,650 did you like the keynote? Please welcome Sam Alman, the CEO of OpenAI. [cheering] 5 00:00:26,730 --> 00:00:31,099 Welcome. Thanks for having me. Great to have you, Sam. 6 00:00:31,182 --> 00:00:34,727 This is not your first Dreamforce interview. 7 00:00:34,810 --> 00:00:41,044 Nice to be back. Great to have you, Sam. It's been quite a week for AI, 8 00:00:41,128 --> 00:00:43,447 hasn't it? It kind of feels 9 00:00:43,530 --> 00:00:47,671 like all the weeks are a big week these days, but yeah, especially big one. 10 00:00:47,754 --> 00:00:54,236 Feels like AI is really owning the moment. Um, yeah, I'll say yeah. Now, 11 00:00:54,320 --> 00:00:59,287 one of the things I noticed about you, Sam, is 12 00:00:59,370 --> 00:01:01,830 when we first did our first Dreamforce interview, 13 00:01:01,914 --> 00:01:04,167 your responses were always a little bit curt, 14 00:01:04,250 --> 00:01:08,975 sometimes curtailed, sometimes short, maybe one word, 15 00:01:09,058 --> 00:01:12,967 but I noticed a transformation in you. We've 16 00:01:13,050 --> 00:01:18,465 been together so many times and you are a lot more willing to open up and be more 17 00:01:18,548 --> 00:01:23,601 robust in what you have to say. So let's start there and just tell us, you know, 18 00:01:23,685 --> 00:01:25,527 what's your analysis of what's 19 00:01:25,610 --> 00:01:29,895 going on in the last few days? I mean, everyone is just kind of wondering what 20 00:01:29,978 --> 00:01:30,811 is going on. 21 00:01:31,850 --> 00:01:37,189 It's interesting because people have been talking about the potential downsides uh 22 00:01:37,272 --> 00:01:38,247 or serious risks 23 00:01:38,330 --> 00:01:42,410 of AI for some time and I think a lot of people have been saying why is this 24 00:01:42,493 --> 00:01:43,287 the moment where 25 00:01:43,370 --> 00:01:49,533 it sort of escaped the bubble and is now all of a sudden an international topic 26 00:01:49,617 --> 00:01:54,295 of such uh intensity and I think a few things have happened. 27 00:01:54,378 --> 00:01:57,047 one the models have gotten so good 28 00:01:57,130 --> 00:02:02,093 and people for a long time looked at the models maybe they looked at the rate 29 00:02:02,176 --> 00:02:03,207 of progress but I 30 00:02:03,290 --> 00:02:06,859 don't think people really felt that the models were going to come this far 31 00:02:06,943 --> 00:02:07,847 this fast and all of 32 00:02:07,930 --> 00:02:14,465 a sudden we went from what it feels like all of a sudden we went from like models 33 00:02:14,548 --> 00:02:15,527 when chat GBT 34 00:02:15,610 --> 00:02:21,521 launched three something years ago um that could barely carry on a conversation 35 00:02:21,604 --> 00:02:22,887 to models that are 36 00:02:22,970 --> 00:02:27,549 transforming I mean the way enterprises do work to models that can write incredibly 37 00:02:27,632 --> 00:02:33,336 complicated uh pieces of software, models that can prove millennium prize problems. 38 00:02:33,419 --> 00:02:34,647 And so I think the 39 00:02:34,730 --> 00:02:38,860 biggest thing that has changed is this thing that our industry has been talking 40 00:02:38,943 --> 00:02:39,767 about for a while 41 00:02:39,850 --> 00:02:43,625 which is no no this technology this approach is going to go very far. 42 00:02:43,708 --> 00:02:44,967 These models are in some 43 00:02:45,050 --> 00:02:49,817 ways going to become more capable smarter than people. Um that finally happened. 44 00:02:49,901 --> 00:02:50,727 So I think that 45 00:02:50,810 --> 00:02:55,793 was a big change to why now when people say you know my probability of doom is x 46 00:02:55,877 --> 00:02:56,807 or y or whatever 47 00:02:56,890 --> 00:03:03,388 it hits in a different way. Um I think people also feel the speed with which 48 00:03:03,471 --> 00:03:05,207 things are are moving 49 00:03:05,290 --> 00:03:09,021 and there have been things like the hugging face accident other accidents 50 00:03:09,104 --> 00:03:10,327 in the field where people 51 00:03:10,410 --> 00:03:15,831 can say like all right it doesn't take as much imagination uh as it used to for 52 00:03:15,915 --> 00:03:19,801 to imagine how this could go wrong. Uh on the other hand, 53 00:03:19,885 --> 00:03:23,367 you have companies saying things like we will only 54 00:03:23,450 --> 00:03:26,590 slow down if or we will only be responsible if other companies are 55 00:03:26,674 --> 00:03:28,007 responsible. And then I think 56 00:03:28,090 --> 00:03:31,831 the public naturally says like well we'd really like to know that you're going 57 00:03:31,915 --> 00:03:32,567 to like do this 58 00:03:32,650 --> 00:03:36,625 safely and be responsible no matter what. There should be no qualifier on that. 59 00:03:36,708 --> 00:03:37,447 I think you also 60 00:03:37,530 --> 00:03:44,353 see um a real fear that some of these companies or some companies developing AI 61 00:03:44,436 --> 00:03:45,927 could get too much 62 00:03:46,010 --> 00:03:50,950 power and be able to sort of exert undue influence uh on the economy, 63 00:03:51,033 --> 00:03:53,207 push a worldview out on people. 64 00:03:53,290 --> 00:03:55,922 Uh and I think the world is right to be afraid of this. 65 00:03:56,005 --> 00:03:57,847 So we have these two big challenges. We 66 00:03:57,930 --> 00:04:01,963 have the potential of a loss of control accident or some other serious thing 67 00:04:02,047 --> 00:04:03,047 that could go wrong. 68 00:04:03,130 --> 00:04:08,167 um we have the potential of way too much power concentration and people fear that 69 00:04:08,250 --> 00:04:11,007 those developing AI could exert their worldview. 70 00:04:11,090 --> 00:04:13,847 Um and there is a sort of narrow path we have to 71 00:04:13,930 --> 00:04:20,930 navigate with pragmatism with you know steadiness with the ability for the 72 00:04:21,376 --> 00:04:23,607 the world to like trust 73 00:04:23,690 --> 00:04:28,832 that we will make dependable decisions consistent decisions uh through this and I 74 00:04:28,916 --> 00:04:29,607 think people 75 00:04:29,690 --> 00:04:36,690 are saying well you know this now seems pretty important like let's let's get it 76 00:04:37,157 --> 00:04:38,567 right um yeah we 77 00:04:38,650 --> 00:04:43,808 will get it right by the way I am very confident in our company's ability, 78 00:04:43,892 --> 00:04:45,367 our industry's ability 79 00:04:45,450 --> 00:04:51,349 to do this safely, to make sure that we keep alignment and safety and monitoring 80 00:04:51,433 --> 00:04:52,247 way ahead of 81 00:04:52,330 --> 00:04:56,275 capabilities to slow down or stop if we get to a point where we can't. Um, 82 00:04:56,358 --> 00:04:57,527 but I'm disappointed by 83 00:04:57,610 --> 00:05:03,557 how it's been framed. Can you articulate a little bit why where that disappointment 84 00:05:03,640 --> 00:05:04,730 is coming from? 85 00:05:05,690 --> 00:05:12,690 Well, I touched on this a little bit, but um the world should trust that we are 86 00:05:13,063 --> 00:05:14,007 going to do 87 00:05:14,090 --> 00:05:21,090 the right thing because it's the right thing and because we feel the magnitude 88 00:05:21,323 --> 00:05:22,167 of this. I 89 00:05:22,250 --> 00:05:26,288 think it's great for our industry to say we want to come together and we want to be 90 00:05:26,371 --> 00:05:30,312 able to coordinate and make sure we have enough time to do this safely. 91 00:05:30,395 --> 00:05:32,167 But when there's any implication 92 00:05:32,250 --> 00:05:37,123 that because of the commercial pressures and the race some company or between 93 00:05:37,206 --> 00:05:40,870 countries, some countries might not do the right thing, 94 00:05:40,954 --> 00:05:44,327 I think that's when people get very scared. Let's 95 00:05:44,410 --> 00:05:49,704 leave AI for just one second. You know, you and I have had a lot of conversations 96 00:05:49,787 --> 00:05:53,948 about safety, but you know, starting in the social media industry, you know, 97 00:05:54,032 --> 00:05:55,367 we've looked at the social 98 00:05:55,450 --> 00:05:59,620 media industry now for more than a decade and all the various incidents and things 99 00:05:59,703 --> 00:06:00,087 that have 100 00:06:00,170 --> 00:06:04,779 happened in social media and it's exhaustive and companies have paid various 101 00:06:04,862 --> 00:06:05,767 levels of price, 102 00:06:05,850 --> 00:06:11,402 but also it's well documented that kids have paid a lot of prices with social 103 00:06:11,486 --> 00:06:12,647 media. Societies, 104 00:06:12,730 --> 00:06:17,287 countries have paid certain prices. You're an expert in this. you you've educated me 105 00:06:17,370 --> 00:06:23,927 on these issues. You know, when you look back at what's happened with social media, 106 00:06:24,010 --> 00:06:25,047 you know, what 107 00:06:25,130 --> 00:06:30,141 what lessons should we take about this idea that companies should be held 108 00:06:30,224 --> 00:06:31,607 responsible for their 109 00:06:31,690 --> 00:06:37,207 actions? Companies should be held responsible for their products. 110 00:06:37,290 --> 00:06:39,447 Uh the company's executive 111 00:06:39,530 --> 00:06:44,272 should be held responsible. How how do you look at that as the basis of ethics in 112 00:06:44,356 --> 00:06:45,189 our industry? 113 00:06:47,370 --> 00:06:52,225 I I would zoom out even a little more and say the the there's always this question 114 00:06:52,308 --> 00:06:55,553 of like why are we why are we building this technology? You know, 115 00:06:55,636 --> 00:06:57,287 the social media people could give 116 00:06:57,370 --> 00:06:59,875 their own answer to that. I'd certainly have some questions, 117 00:06:59,959 --> 00:07:01,127 although I think there's some 118 00:07:01,210 --> 00:07:08,210 good things social media has done too. Um but I I definitely do observe that 119 00:07:09,603 --> 00:07:11,287 the consequences 120 00:07:11,370 --> 00:07:18,230 the impact on society of social media has clearly not been a only positive thing. 121 00:07:18,313 --> 00:07:22,674 Now do I think that social media companies should be blamed for all of the ills 122 00:07:22,758 --> 00:07:23,847 of society? Certainly 123 00:07:23,930 --> 00:07:28,329 not. Do I think that social media companies should have made some different 124 00:07:28,413 --> 00:07:29,047 decisions or 125 00:07:29,130 --> 00:07:32,507 if I were running them I would have made some different decisions about the impact 126 00:07:32,590 --> 00:07:32,887 that this 127 00:07:32,970 --> 00:07:39,970 has on people especially young people. Um absolutely yes. And as we think about AI, 128 00:07:41,290 --> 00:07:48,290 um well a a reason I feel very positive and like and then I'll come to all the 129 00:07:48,791 --> 00:07:50,247 the negatives is 130 00:07:50,330 --> 00:07:55,894 a thing that has really inspired me about AI progress is people don't care that much 131 00:07:55,978 --> 00:07:56,567 what an AI 132 00:07:56,650 --> 00:08:00,811 what a machine does. You know machines can go off and like do a bunch of impressive 133 00:08:00,894 --> 00:08:01,527 work. Machines 134 00:08:01,610 --> 00:08:04,894 can go off and prove some mathematical theorem. We don't really care. 135 00:08:04,977 --> 00:08:06,407 We have this wonderful ability. 136 00:08:06,490 --> 00:08:09,990 We are obsessed with other people. We really care about being useful to other 137 00:08:10,073 --> 00:08:14,177 people. We care about um serving other people, being creative for other people, 138 00:08:14,261 --> 00:08:16,487 what other people think of our stuff. And 139 00:08:16,570 --> 00:08:21,475 and so when I think about the things that I worry about with AI and the mistakes I 140 00:08:21,558 --> 00:08:22,327 want to avoid, 141 00:08:22,410 --> 00:08:29,193 um one thing that I view as a real positive is the degree to which people 142 00:08:29,277 --> 00:08:30,887 care about people. 143 00:08:30,970 --> 00:08:34,327 people are going to use this tool and this new stuff to do things for other people, 144 00:08:34,410 --> 00:08:38,820 to build new people for new businesses for other people. Um, 145 00:08:38,904 --> 00:08:41,367 but I think we're going to stay in 146 00:08:41,450 --> 00:08:45,485 an extremely human- centric world no matter how good the technology gets. 147 00:08:45,569 --> 00:08:46,727 Um, and I think one of 148 00:08:46,810 --> 00:08:53,810 the things I didn't like about social media is I think it pushed us a little bit 149 00:08:54,436 --> 00:08:55,687 more away. Now 150 00:08:55,770 --> 00:09:01,847 clearly this technology is moving at a pace that is very fast and I 151 00:09:01,930 --> 00:09:07,847 think society and people themselves will need to uplift themselves very quickly 152 00:09:07,930 --> 00:09:14,930 to be able to get all the advantages technology but it seems possible to me. 153 00:09:15,450 --> 00:09:18,247 Yeah, I think that, you know, you have that unique perspective, 154 00:09:18,330 --> 00:09:22,644 right? Because not only were you with all those amazing companies at Y Combinator 155 00:09:22,727 --> 00:09:27,220 and you're also the CEO of Reddit and now you're the CEO of this company and you can 156 00:09:27,304 --> 00:09:27,607 look at 157 00:09:27,690 --> 00:09:32,937 those different frames and then kind of say which of these companies did a good 158 00:09:33,020 --> 00:09:37,860 job, you know, taking AI totally out of the picture in safety and responsibility 159 00:09:37,943 --> 00:09:39,287 and which ones just did 160 00:09:39,370 --> 00:09:46,370 not care. You're asking me which of the previous generation of companies did just 161 00:09:50,660 --> 00:09:56,874 didn't care. Um, well, I think it's not so black and white. 162 00:09:56,957 --> 00:10:02,007 Like I I happen to like short form video. I happen to 163 00:10:02,090 --> 00:10:05,907 like the ability to like watch 5 or 10 minutes of short form video before I go 164 00:10:05,990 --> 00:10:10,141 to bed as a way to like kind of unwind and relax a little bit. 165 00:10:10,225 --> 00:10:14,407 But I wouldn't let my kids near that stuff. And I don't 166 00:10:14,490 --> 00:10:20,018 think it's like reasonable to expect kids to be able to resist that dopamine [clears 167 00:10:20,102 --> 00:10:20,887 throat] thing 168 00:10:20,970 --> 00:10:25,438 or know that they even should. Uh so I don't think it's like I don't think short 169 00:10:25,521 --> 00:10:30,087 form video is inherently evil, but I do think it's dangerous. 170 00:10:30,171 --> 00:10:34,087 And how much of that responsibility is on parents 171 00:10:34,170 --> 00:10:38,650 versus the companies versus governments? I don't know how to like aortion that. 172 00:10:40,250 --> 00:10:44,327 Um well let's take a little bit of a pivot and kind of talk about where we are in the 173 00:10:44,410 --> 00:10:49,857 AI industry. Today we kind of can see AI as you know we as you said we've moved 174 00:10:49,940 --> 00:10:51,047 from the chat GPT 175 00:10:51,130 --> 00:10:55,638 moment you know to today where we kind of went through the hugging face attack when 176 00:10:55,721 --> 00:10:56,247 the hugging 177 00:10:56,330 --> 00:11:00,838 face attack happened how did you frame it inside the company and when you looked 178 00:11:00,921 --> 00:11:01,527 at the model 179 00:11:01,610 --> 00:11:06,316 and what it what it achieved and you know it's probably was definitely a catalyst 180 00:11:06,400 --> 00:11:06,967 for the you 181 00:11:07,050 --> 00:11:11,169 know I talked to CLM after you know that happened he was at my house and we were 182 00:11:11,252 --> 00:11:12,167 you know we're kind 183 00:11:12,250 --> 00:11:16,691 of having the conversation you know around selling in the company. You know, 184 00:11:16,775 --> 00:11:20,520 there's no question it was a catalyst for him. So, you know, 185 00:11:20,603 --> 00:11:23,687 do you think we're going to see more things like 186 00:11:23,770 --> 00:11:28,611 that happen where the models go out and, you know, kind of attack, you know, 187 00:11:28,695 --> 00:11:30,167 other stakeholders? Does 188 00:11:30,250 --> 00:11:34,214 everybody know what happened with hugging face or is it Yeah. Okay. 189 00:11:34,297 --> 00:11:39,512 Um I think it's okay to reframe re reset [clears throat] it and very briefly. So, 190 00:11:39,595 --> 00:11:42,727 during uh the evaluation of one of our models, 191 00:11:42,810 --> 00:11:47,107 this is like an older model. This is not like one of our best models. 192 00:11:47,190 --> 00:11:48,567 Um the model was trying 193 00:11:48,650 --> 00:11:53,875 to see how it could do on a certain benchmark to evaluate a capability 194 00:11:53,958 --> 00:11:55,847 and that was what we asked 195 00:11:55,930 --> 00:12:00,610 the model to do at least and the model broke out of a sandbox uh that we were 196 00:12:00,693 --> 00:12:01,847 evaluate that it was 197 00:12:01,930 --> 00:12:05,386 running in hacked into a hugging face server like moved laterally through 198 00:12:05,469 --> 00:12:06,647 the hugging face system to 199 00:12:06,730 --> 00:12:11,387 get the answer and then return it back and you know got like a perfect score on 200 00:12:11,470 --> 00:12:12,167 the test. Um, 201 00:12:12,250 --> 00:12:17,176 this obviously a terrifying incident. This was the worst accident we've seen. 202 00:12:17,259 --> 00:12:23,105 I think it was mostly framed as a security issue, which certainly it was, 203 00:12:23,189 --> 00:12:26,567 but there's also a real alignment issue. 204 00:12:26,650 --> 00:12:30,824 And then since then, you know, other companies found similar behavior of 205 00:12:30,907 --> 00:12:34,980 of their models. Um, but this idea that the models have gotten so powerful, 206 00:12:35,063 --> 00:12:37,127 they're capable of doing something like 207 00:12:37,210 --> 00:12:39,732 this. And although we have aligned them in many ways, 208 00:12:39,816 --> 00:12:41,847 we have not taught them like hey, no matter 209 00:12:41,930 --> 00:12:45,482 how much we tell you to get the best score on this test you can, 210 00:12:45,565 --> 00:12:47,527 like don't break out, don't hack in, 211 00:12:47,610 --> 00:12:53,072 don't steal the answer. Um, that was like a real wakeup call for us about what has 212 00:12:53,156 --> 00:12:54,087 happened to the 213 00:12:54,170 --> 00:12:59,533 capability of these models. Um, and you touched on this uh about like coming 214 00:12:59,617 --> 00:13:00,967 from chatbt to here, 215 00:13:01,050 --> 00:13:05,723 but to give another framing of how quickly these models have gotten better. Um, 216 00:13:05,806 --> 00:13:06,807 this really struck 217 00:13:06,890 --> 00:13:12,785 me recently. Three summers ago, we had a model that could barely do grade school 218 00:13:12,868 --> 00:13:13,607 math. Like, 219 00:13:13,690 --> 00:13:16,908 you know, the word problems that you do in second grade, Bobby has three apples, 220 00:13:16,991 --> 00:13:20,855 blah blah blah. Barely, it could do those. Um, the summer after that, 221 00:13:20,938 --> 00:13:23,367 we had a model that could perform pretty 222 00:13:23,450 --> 00:13:27,746 well at AIM, which is like a high school math competition that I'm sure many of you 223 00:13:27,830 --> 00:13:28,327 some of you 224 00:13:28,410 --> 00:13:34,087 did. One summer ago, we had a model that could get a gold medal on the IMO, 225 00:13:34,170 --> 00:13:36,007 like the most prestigious 226 00:13:36,090 --> 00:13:41,552 math competition uh in the world. And then this summer, 227 00:13:41,636 --> 00:13:45,687 we had a model that could uh could prove, 228 00:13:45,770 --> 00:13:49,545 you know, one of the seven biggest unsolved problems in mathematics. 229 00:13:49,629 --> 00:13:50,567 And and to zoom in 230 00:13:50,650 --> 00:13:56,256 even a little bit more, um, a framing that someone gave is, you know, 231 00:13:56,339 --> 00:13:58,647 5.5 was maybe as good as like 232 00:13:58,730 --> 00:14:04,484 an average math professor. 5.6 was as good as like a maybe top one or two percentile 233 00:14:04,568 --> 00:14:05,527 math professor. 234 00:14:05,610 --> 00:14:09,184 Astra was a little bit better than that. And then this internal model past Astra 235 00:14:09,267 --> 00:14:14,067 is one that can do things that the best mathematicians in the world cannot. 236 00:14:14,150 --> 00:14:16,967 And that's just like the last four maybe 237 00:14:17,050 --> 00:14:22,476 six months, I don't know. Um so that that speed and math has been an unusually fast 238 00:14:22,560 --> 00:14:26,633 takeoff but that certainly is a fast takeoff by anyone's definition. 239 00:14:26,716 --> 00:14:28,407 Um I think is the backdrop to 240 00:14:28,490 --> 00:14:32,407 something like hugging face which is these models and this actually ties back 241 00:14:32,491 --> 00:14:33,447 to where we started. 242 00:14:33,530 --> 00:14:38,039 These models have gotten so good so fast that we need to treat the alignment 243 00:14:38,122 --> 00:14:39,127 and monitoring and 244 00:14:39,210 --> 00:14:44,635 security with a new level of rigor. And we have to be willing to pace our development 245 00:14:44,718 --> 00:14:49,351 of capabilities such that capabilities always s such that alignment, 246 00:14:49,434 --> 00:14:52,087 safety and monitoring are always ahead 247 00:14:52,170 --> 00:14:56,832 of capabilities. And the hugging face incident triggered a real reset not just 248 00:14:56,915 --> 00:14:57,927 in our own company 249 00:14:58,010 --> 00:15:04,007 but I think the whole industry. Yeah. You know speaking to Clen he's the CEO 250 00:15:04,090 --> 00:15:05,207 of hugging face 251 00:15:05,290 --> 00:15:09,749 um after this had happened. You know, I think one of the conversations is they're 252 00:15:09,833 --> 00:15:10,647 a small company, 253 00:15:10,730 --> 00:15:16,382 200 people. Um, not a lot of revenue, $100, $200 million revenue. And, um, 254 00:15:16,465 --> 00:15:18,087 you know, important to 255 00:15:18,170 --> 00:15:22,412 the open service industry, no question. I think even you have published our models 256 00:15:22,495 --> 00:15:27,609 on there. We've published models on there, but I don't think he could he didn't feel 257 00:15:27,693 --> 00:15:28,567 that he had the 258 00:15:28,650 --> 00:15:34,627 technical capability to keep himself safe. He just wasn't ready. Obviously, 259 00:15:34,711 --> 00:15:39,543 you know, Salesforce has 90,000 people. Hopefully, our security team, you know, 260 00:15:39,626 --> 00:15:41,527 is doing what it's what is necessary 261 00:15:41,610 --> 00:15:45,539 to keep itself safe in this kind of evolving world. You know, 262 00:15:45,622 --> 00:15:47,447 h how do you think companies, 263 00:15:47,530 --> 00:15:51,774 small companies like that should be thinking about their vulnerability 264 00:15:51,857 --> 00:15:52,887 against, you know, 265 00:15:52,970 --> 00:15:57,245 um your model, but other models, uh and where we are and how do you think, 266 00:15:57,329 --> 00:16:01,071 you know, to kind of get back to the responsibility argument, 267 00:16:01,154 --> 00:16:04,487 what is the frame from the CEO of the model company? 268 00:16:04,570 --> 00:16:08,194 is not just you but others. What is the right frame, 269 00:16:08,278 --> 00:16:11,047 the right consciousness to have as these 270 00:16:11,130 --> 00:16:14,687 things happen? So that that was another update we took from this. 271 00:16:14,770 --> 00:16:16,087 Um it's just a little bit 272 00:16:16,170 --> 00:16:20,262 more backstory. I'm not going to get the days of the week exactly right here, 273 00:16:20,345 --> 00:16:20,967 but something 274 00:16:21,050 --> 00:16:25,502 on like a Friday or a Saturday. I remember seeing a post from HuggingFace which 275 00:16:25,586 --> 00:16:30,075 is our system has been hacked and we're very confused by it and we believe it was 276 00:16:30,159 --> 00:16:36,339 the work of an AI agent. And I was like, "Really? It seems surprising, but okay." 277 00:16:36,422 --> 00:16:39,047 And I went about my weekend and 278 00:16:39,130 --> 00:16:44,538 uh and then I saw something like the next day on our internal Slack that was like uh 279 00:16:44,622 --> 00:16:46,926 you know, there was a little bit of like weird behavior. 280 00:16:47,009 --> 00:16:48,647 And then maybe by kind of like Sunday night, 281 00:16:48,730 --> 00:16:52,859 someone had put together the possibility these things were linked. 282 00:16:52,942 --> 00:16:54,327 And by let's say Monday 283 00:16:54,410 --> 00:17:00,256 morning um we were pretty sure uh we needed to talk to them to find out 284 00:17:00,340 --> 00:17:01,927 for sure but we were 285 00:17:02,010 --> 00:17:05,614 pretty sure and then like by you know Monday at lunchtime I texted Clement. 286 00:17:05,698 --> 00:17:06,647 I was like uh I would 287 00:17:06,730 --> 00:17:10,511 love to talk to you. I think I may know [clears throat] something here. 288 00:17:10,595 --> 00:17:11,927 Um and we got on the phone 289 00:17:12,010 --> 00:17:17,340 and you know figured he flew out to San Francisco and we put it together. Um, 290 00:17:17,423 --> 00:17:18,887 but one of the things, 291 00:17:18,970 --> 00:17:22,801 so so I kind of like was in the fog of war for the first part of that and not later. 292 00:17:22,884 --> 00:17:29,470 And one of the things they said before we knew it was our our our system was that, 293 00:17:29,553 --> 00:17:31,527 you know, they had tried 294 00:17:31,610 --> 00:17:34,527 to talk to one of our competitors to get access to their security model. 295 00:17:34,610 --> 00:17:35,527 I think they didn't even 296 00:17:35,610 --> 00:17:40,838 know we had one to offer at that point. And they uh couldn't get it and they it 297 00:17:40,921 --> 00:17:44,346 was too hard to defend themselves the oldfashioned way just with people. 298 00:17:44,429 --> 00:17:46,087 So they had to use like uh Chinese open 299 00:17:46,170 --> 00:17:50,145 source models to defend themselves. And I remember thinking before I knew that we 300 00:17:50,228 --> 00:17:51,047 were involved like 301 00:17:51,130 --> 00:17:54,414 oh that seems sub-optimal like you know we should do a better job of making 302 00:17:54,497 --> 00:17:55,447 our models available to 303 00:17:55,530 --> 00:18:00,789 help people. And then after I was really like okay we have to change the way we do 304 00:18:00,873 --> 00:18:05,228 this. We we are proud that we have the best cyber security models in the world. 305 00:18:05,312 --> 00:18:07,207 We are proud to have the best models 306 00:18:07,290 --> 00:18:09,754 in the world for businesses and enterprise across the board. 307 00:18:09,837 --> 00:18:11,367 We really want to make sure people can 308 00:18:11,450 --> 00:18:14,858 use them all of the time. Yeah. But particularly when they need them in 309 00:18:14,941 --> 00:18:16,087 a critical situation. And 310 00:18:16,170 --> 00:18:19,270 so after that, we've really tried to put out uh we call it Daybreak, 311 00:18:19,353 --> 00:18:20,487 which is our cyber program 312 00:18:20,570 --> 00:18:25,729 to help companies defend themselves. Um I think that cyber defense is going to have 313 00:18:25,812 --> 00:18:26,487 to change to 314 00:18:26,570 --> 00:18:31,287 sort of be a agents are actively defending your systems. You're not even trying to, 315 00:18:31,370 --> 00:18:35,515 you know, patch every bug. We're trying to get there as fast as we can. 316 00:18:35,598 --> 00:18:36,647 But more generally, 317 00:18:36,730 --> 00:18:41,352 we want to be a great dependable partner to enterprises and we don't want to be 318 00:18:41,435 --> 00:18:42,007 like, we've 319 00:18:42,090 --> 00:18:45,517 got this great model and we're going to keep it locked up and not let you use it. 320 00:18:45,600 --> 00:18:46,167 How many people 321 00:18:46,250 --> 00:18:51,895 in the room here have are with companies with less than a thousand employees? 322 00:18:51,979 --> 00:18:53,607 Raise your hands. Okay, 323 00:18:53,690 --> 00:18:57,767 so it's about 20%. What is your message to companies with less than a thousand 324 00:18:57,850 --> 00:18:58,247 employees 325 00:18:58,330 --> 00:19:03,673 to for them to defend themsel in this moment where they could suffer that same 326 00:19:03,756 --> 00:19:04,647 kind of attack 327 00:19:04,730 --> 00:19:09,438 where they probably don't have the controls that a company like yours or eyes 328 00:19:09,521 --> 00:19:10,807 have? Uh first of all, 329 00:19:10,890 --> 00:19:15,515 I think there are probably more advantages to being a smaller number of employee 330 00:19:15,598 --> 00:19:19,193 company than ever before. Um the speed with which you can move, 331 00:19:19,276 --> 00:19:21,687 the way that you can adopt new technology, 332 00:19:21,770 --> 00:19:24,647 uh the way that you can kind of like be ahead of the curve on this stuff. 333 00:19:24,730 --> 00:19:30,362 We're seeing across the board how quickly small companies are able to move now. 334 00:19:30,445 --> 00:19:31,447 Um and then you 335 00:19:31,530 --> 00:19:36,567 know selfishly we would love to sell you daybreak services to help defend against 336 00:19:36,650 --> 00:19:41,608 attacks and you know I think this idea of a persistent agent defending is great 337 00:19:41,692 --> 00:19:45,844 but defend against your models or other models any model like any models not ours. 338 00:19:45,927 --> 00:19:46,807 Our model hopefully 339 00:19:46,890 --> 00:19:52,635 will not be attacking. Um, but whether you buy from us or a competitor or use an open 340 00:19:52,719 --> 00:19:57,053 model, the strongest message I have is you have to defend against this impending wave 341 00:19:57,137 --> 00:19:57,927 of cyber attacks, 342 00:19:58,010 --> 00:20:02,279 I think we are not that far away from open-source models that can do serious 343 00:20:02,362 --> 00:20:03,367 damage. Clearly, we 344 00:20:03,450 --> 00:20:06,115 should not stop open source models from happening. That's important. 345 00:20:06,198 --> 00:20:09,556 But there is going to be this huge cyber threat that's going to come. So, 346 00:20:09,639 --> 00:20:12,247 while there is this little period of advantage, um, 347 00:20:12,330 --> 00:20:16,420 please use something to defend yourselves. 348 00:20:16,503 --> 00:20:21,687 Cyber and cyber attacks are one potential um you know 349 00:20:21,770 --> 00:20:26,807 kind of manifestation of these models. Other kinds of attacks could come 350 00:20:26,890 --> 00:20:28,727 from other kinds of weapons 351 00:20:28,810 --> 00:20:31,980 that are more sophisticated. You've thought about this a lot. 352 00:20:32,063 --> 00:20:33,367 It could be biological. It 353 00:20:33,450 --> 00:20:39,242 could be chemical. Uh it could be the hacking into a kinetic system. 354 00:20:39,325 --> 00:20:42,007 um where some kind of a military 355 00:20:42,090 --> 00:20:47,320 maybe that is not as maybe well architected to defend itself as ours 356 00:20:47,403 --> 00:20:48,727 or smaller country 357 00:20:48,810 --> 00:20:55,810 could have a rogue kinetic attack. Are these realistic scenarios or is 358 00:20:58,287 --> 00:21:01,047 this fantasy land? Um 359 00:21:01,130 --> 00:21:05,440 well, we need to not let them happen. They're all like this is a very powerful 360 00:21:05,523 --> 00:21:06,567 technology and those 361 00:21:06,650 --> 00:21:10,204 things are all possible. We need to mitigate those because it'd be very 362 00:21:10,288 --> 00:21:11,127 reasonable to hear 363 00:21:11,210 --> 00:21:16,389 someone like Mark say something like that and for some policy maker to say like hm 364 00:21:16,472 --> 00:21:16,967 remind me 365 00:21:17,050 --> 00:21:22,594 why we're developing this again like that would be a reasonable takeaway and 366 00:21:22,677 --> 00:21:23,927 the answer is that 367 00:21:24,010 --> 00:21:28,839 although this is a technology capable of doing things like that it is possible 368 00:21:28,922 --> 00:21:29,847 for us to defend 369 00:21:29,930 --> 00:21:33,752 against those not only as the models and the companies like ours but as society 370 00:21:33,836 --> 00:21:34,247 as a whole 371 00:21:34,330 --> 00:21:39,063 we can build resilience against new kinds of bio attacks We can build resilience 372 00:21:39,146 --> 00:21:40,087 against potential 373 00:21:40,170 --> 00:21:44,647 kinetic attacks. We can build resilience against all of these cyber mess that's 374 00:21:44,731 --> 00:21:50,845 going to come. And we have to do that to get the prize of curing the diseases, 375 00:21:50,928 --> 00:21:54,167 giving people this incredible creative 376 00:21:54,250 --> 00:21:56,882 ability, this boom in entrepreneurship that I think is coming. 377 00:21:56,965 --> 00:21:58,327 So I think it's very easy to say, 378 00:21:58,410 --> 00:22:02,884 well there's all these scary things and it's true there are and they are possible. 379 00:22:02,967 --> 00:22:06,926 But you know, you and I and people in this room and many others, 380 00:22:07,009 --> 00:22:09,847 we have the ability and the agency to defend 381 00:22:09,930 --> 00:22:13,167 against those so that we get to enjoy all of the incredible stuff. 382 00:22:13,250 --> 00:22:14,727 And people deserve this. Like I 383 00:22:14,810 --> 00:22:18,962 think people with this technology in their hands can live a much better version 384 00:22:19,046 --> 00:22:19,767 of their lives, 385 00:22:19,850 --> 00:22:23,666 be a better version of themselves to say nothing of what the economy can do. 386 00:22:23,749 --> 00:22:28,619 You and I always have this conversation. Technology itself is never good or bad. 387 00:22:28,703 --> 00:22:31,130 It's what we do with it that matters. 388 00:22:31,770 --> 00:22:36,392 Um, now we're looking at this technology. We don't want these things to happen, 389 00:22:36,475 --> 00:22:41,836 but we kind of also started with this discussion where it actually happened. So, 390 00:22:41,920 --> 00:22:43,527 you think we're going to 391 00:22:43,610 --> 00:22:47,325 have more discussions where it actually happened and then we're kind of coming 392 00:22:47,408 --> 00:22:52,998 back. How many like is that obviously not just you're only one model company, right? 393 00:22:53,081 --> 00:22:54,807 There's hundreds of model 394 00:22:54,890 --> 00:22:57,754 companies, thousands of model companies. How many companies are there? 395 00:22:57,837 --> 00:23:01,520 You like building models, people building models? tens of thousands of people, 396 00:23:01,603 --> 00:23:03,607 hundreds of thousands of people. I don't 397 00:23:03,690 --> 00:23:06,246 I don't agree that technology is like neither good nor bad. 398 00:23:06,329 --> 00:23:07,767 I think the tool framing, the sort 399 00:23:07,850 --> 00:23:10,409 of like neutral tool, we don't care what people do with it. You know, 400 00:23:10,493 --> 00:23:13,439 it's not our fault if all these bad things happen. Freedom of expression. 401 00:23:13,522 --> 00:23:15,687 I think you can justify a lot of stuff that way. Um 402 00:23:15,770 --> 00:23:20,498 I think the choices that companies building this stuff make do have real 403 00:23:20,581 --> 00:23:21,367 consequences. 404 00:23:21,450 --> 00:23:26,637 Now there's clearly a fine line to walk and I also don't think that the right 405 00:23:26,720 --> 00:23:27,527 answer is for 406 00:23:27,610 --> 00:23:32,748 us to impose our worldview on the future. But when something like hugging face 407 00:23:32,831 --> 00:23:36,499 happens, we say that was bad. Like that is not neutral technology. 408 00:23:36,582 --> 00:23:39,447 That is not what we want our technology to do. Somebody 409 00:23:39,530 --> 00:23:41,359 else wants to build like a hacking bot. Okay. 410 00:23:41,442 --> 00:23:43,527 Um the world can decide what they think about that. 411 00:23:43,610 --> 00:23:47,287 But but I think that is not a case of technology being neither good nor bad. 412 00:23:47,371 --> 00:23:48,327 That's a case of like 413 00:23:48,410 --> 00:23:54,663 it's an accident. this. Yeah. And and I think the companies building 414 00:23:54,747 --> 00:23:56,807 this technology do need 415 00:23:56,890 --> 00:24:03,890 to have both a lot of responsibility and a lot of support of the ecosystem, 416 00:24:04,977 --> 00:24:06,727 support of the of 417 00:24:06,810 --> 00:24:11,237 society and the economy. It cannot be that a small number of companies building these 418 00:24:11,320 --> 00:24:15,581 models get to make the decisions that the world should get to make. Well, I mean, 419 00:24:15,665 --> 00:24:19,988 in the context of the hugging face accident, um, where the model, 420 00:24:20,071 --> 00:24:24,007 this older model got crazy and attacked Hugging Face, 421 00:24:24,090 --> 00:24:28,433 this little company, you know, is this going to be, you know, 422 00:24:28,516 --> 00:24:30,247 a continuum of accidents? 423 00:24:30,330 --> 00:24:34,111 Do you feel like we'll be looking at not I'm not talking about open AI, 424 00:24:34,195 --> 00:24:35,527 like just put it out there 425 00:24:35,610 --> 00:24:40,967 just as a context of AI as a technology, as an industry, whether it's open source, 426 00:24:41,050 --> 00:24:43,060 closed source, proprietary, whatever. You know, 427 00:24:43,143 --> 00:24:45,287 is there going to be a continuum of accidents that 428 00:24:45,370 --> 00:24:47,542 we'll start marking the hugging face accident, 429 00:24:47,625 --> 00:24:49,847 this accident that you probably know other ones 430 00:24:49,930 --> 00:24:54,657 that I don't know, and you kind of mark them down and go, here we go. And, 431 00:24:54,740 --> 00:24:56,087 you know, a continuum, 432 00:24:56,170 --> 00:24:59,789 a spectrum of them across, you know, we're in the cyber area today, 433 00:24:59,872 --> 00:25:01,447 but maybe we move into another 434 00:25:01,530 --> 00:25:07,103 area. Is this a fair way to characterization or am I or is this unfair 435 00:25:07,187 --> 00:25:09,447 to characterize it like this? 436 00:25:09,530 --> 00:25:13,892 assuming you're right um because I do think that some degree of accidents are 437 00:25:13,975 --> 00:25:15,047 unavoidable with new 438 00:25:15,130 --> 00:25:19,399 technology um across the industry and across all of the way society will use 439 00:25:19,482 --> 00:25:20,487 this. What I really 440 00:25:20,570 --> 00:25:24,448 care about is that we have got a great culture of accident reporting 441 00:25:24,532 --> 00:25:25,847 and learning. I think if 442 00:25:25,930 --> 00:25:30,257 we look at what happens with the FAA and the NTSB and how safe airplanes have 443 00:25:30,340 --> 00:25:30,887 gotten, um, 444 00:25:30,970 --> 00:25:34,449 as I understand it from talking to people who have really studied, you know, 445 00:25:34,532 --> 00:25:35,527 new kinds of technology 446 00:25:35,610 --> 00:25:41,107 on the order of magnitude of airplanes. One of the things that has gone so well 447 00:25:41,190 --> 00:25:42,167 there is is the 448 00:25:42,250 --> 00:25:45,265 culture of accident reporting. You know, accidents are going to happen. 449 00:25:45,348 --> 00:25:46,487 We're going to learn as much 450 00:25:46,570 --> 00:25:52,033 as we can and react from each one. Um, and if you know there's other kinds 451 00:25:52,116 --> 00:25:53,607 of technology where I 452 00:25:53,690 --> 00:25:58,114 don't think it's gone, I don't think society's had the same response and I 453 00:25:58,197 --> 00:25:58,967 don't think we 454 00:25:59,050 --> 00:26:04,570 have improved as much. So yeah, I think probably regrettably accidents with any 455 00:26:04,653 --> 00:26:05,847 new technology are 456 00:26:05,930 --> 00:26:12,193 unavoidable and we should have a great culture of transparent reporting about 457 00:26:12,277 --> 00:26:14,007 them. And do you think 458 00:26:14,090 --> 00:26:19,341 that that's the right metaphor then? You kind of brought this up now aircraft 459 00:26:19,424 --> 00:26:20,727 this aircraft we had 460 00:26:20,810 --> 00:26:25,634 an aircraft disaster this many people die there this happened we've learned from it 461 00:26:25,718 --> 00:26:31,241 now we've improved the aircraft etc the or the system so that it's going to be safer 462 00:26:31,324 --> 00:26:32,407 for everyone else 463 00:26:32,490 --> 00:26:36,036 look I I don't want to equate hugging face to the tragedy of like a big aircraft 464 00:26:36,119 --> 00:26:36,807 crash and a bunch 465 00:26:36,890 --> 00:26:42,043 of people dying like that was not that's obviously a much worse thing I I I meant 466 00:26:42,126 --> 00:26:43,207 like however small 467 00:26:43,290 --> 00:26:47,527 these accidents are learning learning from them quickly and adapting as they happen, 468 00:26:47,610 --> 00:26:51,626 I think will serve our industry well and maybe we can avoid, you know, 469 00:26:51,709 --> 00:26:53,207 ever getting to those those 470 00:26:53,290 --> 00:26:59,321 kinds of stakes. Okay. So, um it's been it's been an incredible moment 471 00:26:59,404 --> 00:27:01,767 in technology. We've seen we 472 00:27:01,850 --> 00:27:06,636 just did this keynote and we're starting to see the models can not only, 473 00:27:06,719 --> 00:27:10,869 you know, kind of do the, you know, textbased answer and response thing, 474 00:27:10,952 --> 00:27:13,687 but we even saw it dynamically draw this interface. 475 00:27:13,770 --> 00:27:17,474 you've made a remarkable, you know, progress with chat GPT. 476 00:27:17,557 --> 00:27:19,207 It's it's really incredible 477 00:27:19,290 --> 00:27:24,476 actually what you've done. It's awesome how you can not only chat with it, 478 00:27:24,559 --> 00:27:25,687 but work with it, 479 00:27:25,770 --> 00:27:29,967 code with it, and I'm sure you can start to render these interfaces with us. 480 00:27:30,051 --> 00:27:31,207 I hope very very soon. 481 00:27:31,290 --> 00:27:34,900 We would love to do something like this with this AI force platform that we've 482 00:27:34,983 --> 00:27:40,647 built with you. That would be a dream of ours. So, you know, give us a a picture. 483 00:27:40,730 --> 00:27:43,607 What is the role then of these models 484 00:27:43,690 --> 00:27:48,201 on the kind of driving the intelligence but also you know providing 485 00:27:48,284 --> 00:27:50,327 this incredible next generation 486 00:27:50,410 --> 00:27:54,369 you know interface we're starting to talk about that for the first time we've 487 00:27:54,452 --> 00:27:55,367 of course seen DOSs 488 00:27:55,450 --> 00:27:59,523 and character modes and gooies and then we moving to the you know phones and now 489 00:27:59,607 --> 00:28:00,407 we're moving into 490 00:28:00,490 --> 00:28:05,255 these dynamic interfaces driven by the models this has been a fantasy that we've 491 00:28:05,338 --> 00:28:06,407 had for a long time 492 00:28:06,490 --> 00:28:11,355 because we've seen them in the movies now we just we just demonstrated it really 493 00:28:11,438 --> 00:28:16,531 working. The fact that you can talk to a computer, can understand what you want, 494 00:28:16,615 --> 00:28:18,727 it can have good judgment to know 495 00:28:18,810 --> 00:28:22,663 where inside of a company it can go look and where it's allowed to look and how it 496 00:28:22,746 --> 00:28:27,516 pulls it together. Um, and then it can render for you a completely new interface. 497 00:28:27,600 --> 00:28:29,447 It can custom write, you know, 498 00:28:29,530 --> 00:28:33,488 10,000 lines of code or whatever it takes to pull this thing together and present 499 00:28:33,571 --> 00:28:34,087 you with the 500 00:28:34,170 --> 00:28:37,182 information to make a decision or to let you be in this creative loop where you 501 00:28:37,266 --> 00:28:37,927 kind of keep trying 502 00:28:38,010 --> 00:28:45,010 things. I think this is one of the most profound changes to how we use technology 503 00:28:45,240 --> 00:28:48,795 that I've ever seen. I mean, sci-fi movies, as you mentioned, 504 00:28:48,878 --> 00:28:51,287 have been hinting at this or playing around with 505 00:28:51,370 --> 00:28:54,802 this for a long time. But this is one of those moments where it feels like you're 506 00:28:54,886 --> 00:28:55,367 in the movie, 507 00:28:55,450 --> 00:29:00,086 like the sci-fi thing came true. You're talking to your personal super assistant 508 00:29:00,169 --> 00:29:01,207 and you are working 509 00:29:01,290 --> 00:29:04,896 10 times faster or better or more creatively than you could have before. 510 00:29:04,979 --> 00:29:05,767 I think this will 511 00:29:05,850 --> 00:29:12,850 transform the way we work, the way we kind of use computers and it feels all of 512 00:29:13,733 --> 00:29:15,047 a sudden here. 513 00:29:15,130 --> 00:29:18,487 Will this be the moment for enterprises? Many of these enterprise customers are 514 00:29:18,570 --> 00:29:22,745 here where enterprise can start to say, "Wow, 515 00:29:22,829 --> 00:29:26,247 our systems are really changing now." 516 00:29:26,330 --> 00:29:31,047 I think we saw the be so so I would say so far there have been kind of two major 517 00:29:31,130 --> 00:29:35,527 eras of AI. I mean a lot of other things too but directionally speaking you had the 518 00:29:35,610 --> 00:29:42,610 chatbots and then you had the coding/computer use agents of this year 519 00:29:43,277 --> 00:29:45,527 and I think coding in 520 00:29:45,610 --> 00:29:50,567 particular and kind of all the other work that one does behind a computer really did 521 00:29:50,650 --> 00:29:55,758 transform in the enterprise like the way that if you think about what how people 522 00:29:55,841 --> 00:29:56,407 wrote code 523 00:29:56,490 --> 00:30:02,247 even one year ago relative to now there's less in common than maybe a year ago in 524 00:30:02,330 --> 00:30:05,859 decade ago or two decades ago. It's really it's quite remarkable. 525 00:30:05,943 --> 00:30:07,527 Um so I think that already did 526 00:30:07,610 --> 00:30:12,079 transform the enterprise. But as we head towards what I think will be this third 527 00:30:12,163 --> 00:30:13,047 phase of it's not 528 00:30:13,130 --> 00:30:17,187 just like you ask AI for something like a chatbot or you ask AI to do something like 529 00:30:17,271 --> 00:30:17,927 a coding agent, 530 00:30:18,010 --> 00:30:22,247 but the AI is like running for you all of the time. Understanding your job, 531 00:30:22,330 --> 00:30:25,804 what you need to be more successful, rendering new code for you, 532 00:30:25,887 --> 00:30:27,527 proactively watching your Slack 533 00:30:27,610 --> 00:30:30,359 or your email. Hopefully there's more Slack than email. 534 00:30:30,442 --> 00:30:32,007 um like looking at all the stuff 535 00:30:32,090 --> 00:30:39,090 happening inside of a company that I think we are on the precipice of this third 536 00:30:40,276 --> 00:30:40,807 phase. 537 00:30:40,890 --> 00:30:45,780 Yeah, that's going to feel very magical for a lot of plain old users who are just 538 00:30:45,863 --> 00:30:50,210 using this computer for the first time and get this incredible new capability. 539 00:30:50,293 --> 00:30:51,527 I don't think anybody's 540 00:30:51,610 --> 00:30:56,270 really seen anything like that. I mean, when I first saw it and really I was 541 00:30:56,354 --> 00:31:00,912 and even when I was like doing the focus groups for the keynote, right? 542 00:31:00,995 --> 00:31:03,607 I was sitting next to our COO and he was 543 00:31:03,690 --> 00:31:08,490 running this dynamic interface on his computer and it looked like he was using, 544 00:31:08,574 --> 00:31:11,980 you know, mission control and the graphics, everything. I'm like, 545 00:31:12,063 --> 00:31:14,327 this is incredible. And he had total command 546 00:31:14,410 --> 00:31:18,054 and control over what was happening, you know, with Salesforce. 547 00:31:18,138 --> 00:31:20,007 But it looked radically different 548 00:31:20,090 --> 00:31:26,157 from any other application I'd ever seen because it was rendering itself on the fly 549 00:31:26,241 --> 00:31:27,047 dynamically, 550 00:31:27,130 --> 00:31:32,365 intelligently, in a composite fashion, in a fashion that almost seemed like it 551 00:31:32,448 --> 00:31:33,047 was living 552 00:31:33,130 --> 00:31:39,568 or alive in a way. I don't think we've ever seen that kind of an interface, 553 00:31:39,652 --> 00:31:41,047 you know, before. 554 00:31:41,130 --> 00:31:46,698 Um you think that's going to ex do something radical for is it going to be 555 00:31:46,781 --> 00:31:47,767 a a Y2K moment 556 00:31:47,850 --> 00:31:54,850 you know where enterprises will feel okay this is a huge opportunity for much more 557 00:31:56,922 --> 00:31:58,167 productivity 558 00:31:58,250 --> 00:32:02,997 I yes and I also think the world has a lot of inertia the way people do their work 559 00:32:03,080 --> 00:32:07,595 has a lot of inertia so it will this thing will happen where like oh man everything 560 00:32:07,678 --> 00:32:12,762 is different now and then it'll take some number of years to filter through, 561 00:32:12,846 --> 00:32:17,345 but I think the capability and the total change in what's possible. Yeah, 562 00:32:17,429 --> 00:32:19,927 I would say by like by the end of this year, 563 00:32:20,010 --> 00:32:23,123 there's like a very different way that it'll be possible to work. 564 00:32:23,206 --> 00:32:28,514 You've now been working on Open AI for more than a decade. Uh it's incredible. 565 00:32:28,598 --> 00:32:32,887 uh you were on the cover of Time magazine last week, 566 00:32:32,970 --> 00:32:38,463 you know, with kind of an article that's kind of discusses the next chapter of 567 00:32:38,546 --> 00:32:39,607 the company with 568 00:32:39,690 --> 00:32:46,690 um with Greg and how it's being operated and how your hope for it and so forth. 569 00:32:46,800 --> 00:32:47,527 you know, 570 00:32:47,610 --> 00:32:53,310 you have now the opportunity to start to think about kind of the legacy of Open AI, 571 00:32:53,393 --> 00:32:54,007 the impact 572 00:32:54,090 --> 00:32:59,054 to the technology industry, what it's also doing for business, for our community, 573 00:32:59,138 --> 00:33:03,693 San Francisco, you're a very important part of San Francisco as well. Now, 574 00:33:03,776 --> 00:33:05,927 large employer, you know, what what 575 00:33:06,010 --> 00:33:11,721 are your hopes and aspirations, you know, for the company, not the technology? 576 00:33:11,804 --> 00:33:13,207 You know, I think in 577 00:33:13,290 --> 00:33:19,558 the first decade, we set out on this what seemed the time absolutely insane mission 578 00:33:19,641 --> 00:33:24,670 to try to figure out how to build AGI. Chance of success seemed extremely low. 579 00:33:24,754 --> 00:33:26,167 Every expert told us we 580 00:33:26,250 --> 00:33:31,858 weren't going to be able to do it. And we were a great I think probably the greatest 581 00:33:31,941 --> 00:33:32,807 scientific lab 582 00:33:32,890 --> 00:33:37,006 um in you know recent memory. and we we figure out how to build this amazing 583 00:33:37,090 --> 00:33:38,167 technology along with 584 00:33:38,250 --> 00:33:44,407 many other people in the world doing incredible work that and now we basically 585 00:33:44,490 --> 00:33:45,847 know how to do it. 586 00:33:45,930 --> 00:33:49,422 We have only scratched the surface of figuring out how to build great products 587 00:33:49,506 --> 00:33:53,175 that really make people more capable, more powerful, that are useful to people, 588 00:33:53,258 --> 00:33:54,727 that transform the economy, that 589 00:33:54,810 --> 00:33:59,514 transform the enterprise. And I hope the second decade is about how we take 590 00:33:59,597 --> 00:34:03,253 this amazing thing. You know, this is like we went to go get fire from the gods. 591 00:34:03,336 --> 00:34:04,727 This is a crazy thing that has just 592 00:34:04,810 --> 00:34:10,595 happened in the world. And figure out how to give that value to people in a way 593 00:34:10,679 --> 00:34:15,915 that is most useful to them. figure out how to empower every person on Earth, 594 00:34:15,999 --> 00:34:18,727 every enterprise on Earth with this flood 595 00:34:18,810 --> 00:34:25,810 of capability um and help them build whatever they're going to build. 596 00:34:26,501 --> 00:34:28,647 Are your aspirations 597 00:34:28,730 --> 00:34:32,886 only about the technology or is there something broader in our community or 598 00:34:32,969 --> 00:34:37,607 in the culture of business you'd also like to manifest? Um, well, 599 00:34:37,690 --> 00:34:41,207 that was kind of what I was trying to get at, 600 00:34:41,290 --> 00:34:45,239 which is I, you know, this can be some technology that does an amazing thing 601 00:34:45,323 --> 00:34:46,407 and that will be fine. 602 00:34:46,490 --> 00:34:51,902 It's an amazing scientific achievement or this can be something that really does 603 00:34:51,985 --> 00:34:53,207 level people up and 604 00:34:53,290 --> 00:34:59,730 transform what people are capable of. um how how people can spend their time 605 00:34:59,813 --> 00:35:04,543 and live their lives and choose what they want to what they want to do, 606 00:35:04,627 --> 00:35:07,207 what they what impact they want to have and 607 00:35:07,290 --> 00:35:12,241 that. So I think the first decade was about this technological transformation. 608 00:35:12,325 --> 00:35:12,887 Um and now 609 00:35:12,970 --> 00:35:19,970 I hope this transforms people and companies and that we are one of the flag 610 00:35:20,756 --> 00:35:25,232 bears for what this means for people, not what this means for the machine. Again, 611 00:35:25,316 --> 00:35:26,407 I think maybe our first 612 00:35:26,490 --> 00:35:29,208 decade was thinking about the machine and figuring out how to do this and now I 613 00:35:29,292 --> 00:35:32,271 think this has got to be about what it's going to be for people. 614 00:35:32,354 --> 00:35:35,127 If you look forward in the next five years and you're 615 00:35:35,210 --> 00:35:39,035 getting a pretty good vision of what's coming in the next 5 years as we look 616 00:35:39,119 --> 00:35:40,167 at now kind of heading 617 00:35:40,250 --> 00:35:46,049 towards 2030, what do you think will be the single most exciting thing that will 618 00:35:46,133 --> 00:35:50,854 happen by 2030 with the technology that's going to be really impact us? 619 00:35:50,937 --> 00:35:53,927 the last time you were here it was discovering 620 00:35:54,010 --> 00:36:00,567 new science um maybe curing a disease you know what do you think will really happen 621 00:36:00,650 --> 00:36:01,127 between 622 00:36:01,210 --> 00:36:06,334 now and 2030 that you'll say wow this is what I always thought was going to happen 623 00:36:06,417 --> 00:36:07,287 and it actually 624 00:36:07,370 --> 00:36:11,707 has happened so those things we talked about last time they've started happening 625 00:36:11,791 --> 00:36:12,647 there there is no 626 00:36:12,730 --> 00:36:16,732 doubt there will be more technological wonders between now and 2030 and they will 627 00:36:16,815 --> 00:36:17,287 be big ones 628 00:36:17,370 --> 00:36:24,370 like giant ones like some of the biggest ones in human But back to that previous 629 00:36:24,633 --> 00:36:25,367 question, 630 00:36:25,450 --> 00:36:30,943 what I hope you invite me back here in 2030 is people are saying like I cannot 631 00:36:31,026 --> 00:36:32,087 believe how much 632 00:36:32,170 --> 00:36:36,542 better my own life is. I cannot believe how good the company is I was able 633 00:36:36,625 --> 00:36:37,927 to start by myself like 634 00:36:38,010 --> 00:36:43,861 my oneperson company you know I cannot believe like what I am creating for people 635 00:36:43,944 --> 00:36:47,687 around me. Um, so you know, the science stuff that's on lock at this point. 636 00:36:47,770 --> 00:36:49,367 That's on autopilot. That's going to 637 00:36:49,450 --> 00:36:54,407 happen. Um, I hope the humanity side of it really goes very far. Sam, you're invited 638 00:36:54,490 --> 00:37:01,490 back anytime you want to Dreamforce. Hold on. Thank you. 639 00:37:05,933 --> 00:37:12,880 Please thank Sam Alman. [applause]