1 00:00:01,120 --> 00:00:02,960 Speaker 1: We're going to now head over to the AI Action 2 00:00:03,040 --> 00:00:06,160 Speaker 1: Summit in Paris, where Bloomberg's Tom mackenzie is sitting down 3 00:00:06,160 --> 00:00:09,560 Speaker 1: with open AI's chief Global affairs officer, Chris Lahane. 4 00:00:12,960 --> 00:00:14,960 Speaker 2: Guys, thank you very much doing yes at Le Grand 5 00:00:15,000 --> 00:00:17,400 Speaker 2: Ballet in Paris. I'll start with that question about deep 6 00:00:17,440 --> 00:00:20,040 Speaker 2: sea because it is a key question for people attending 7 00:00:20,200 --> 00:00:23,040 Speaker 2: this event. Chris Lhane, head of Global policy of course 8 00:00:23,120 --> 00:00:26,279 Speaker 2: at open AI deep Sea, how do you characterise the 9 00:00:26,400 --> 00:00:29,120 Speaker 2: impact of deep sea. We're just hearing from Demisisarvice, who 10 00:00:29,120 --> 00:00:32,000 Speaker 2: I sat down with earlier. He said their claims around 11 00:00:32,040 --> 00:00:34,880 Speaker 2: the way they built and trained this model, the cost 12 00:00:35,360 --> 00:00:36,920 Speaker 2: and the amount of chips they were using. He says, 13 00:00:36,920 --> 00:00:37,800 Speaker 2: that's exaggerated. 14 00:00:38,440 --> 00:00:40,159 Speaker 1: So I think there can be two things that they 15 00:00:40,200 --> 00:00:44,160 Speaker 1: are true here. One that they have built a really 16 00:00:44,600 --> 00:00:48,720 Speaker 1: impressive model. It basically competes with what open AI had 17 00:00:48,760 --> 00:00:50,879 Speaker 1: put out back in September. Now we've since put out 18 00:00:50,920 --> 00:00:55,680 Speaker 1: other more advanced models, but clearly a very capable model. 19 00:00:56,120 --> 00:00:58,880 Speaker 1: I think the second thing with Demis said can also 20 00:00:58,920 --> 00:01:02,240 Speaker 1: be true, which is perhaps and we've seen news reports 21 00:01:02,280 --> 00:01:05,919 Speaker 1: on this, that the costs, that how the technology was derived, 22 00:01:06,760 --> 00:01:09,600 Speaker 1: that you know, whether they access certain types of chips, 23 00:01:10,880 --> 00:01:14,000 Speaker 1: you know, whether what was initially said was not actually 24 00:01:14,360 --> 00:01:17,399 Speaker 1: the case. I do think the big takeaway though, even 25 00:01:17,400 --> 00:01:19,480 Speaker 1: if everything I just said turns out to be accurate 26 00:01:19,520 --> 00:01:23,800 Speaker 1: and crewe, it is still a very impressive and competitive model. 27 00:01:23,840 --> 00:01:26,080 Speaker 1: And so to me, the big, big, big takeaway in 28 00:01:26,120 --> 00:01:29,560 Speaker 1: all of this is that this really reaffirmed something that 29 00:01:30,080 --> 00:01:32,479 Speaker 1: open ai has been saying since the summer of twenty 30 00:01:32,600 --> 00:01:36,080 Speaker 1: twenty four that there are two countries in the world 31 00:01:36,360 --> 00:01:40,399 Speaker 1: that can build AI at scale, the US, the CCP 32 00:01:40,560 --> 00:01:44,160 Speaker 1: led China. And what that really means is that there 33 00:01:44,200 --> 00:01:47,520 Speaker 1: is a global competition right now between whether the world's 34 00:01:47,560 --> 00:01:51,400 Speaker 1: going to be built on small, d democratic AI rails 35 00:01:52,080 --> 00:01:56,520 Speaker 1: or authoritarian, autocratic AI rails. And that's the big takeaway 36 00:01:56,520 --> 00:01:56,800 Speaker 1: for this. 37 00:01:56,920 --> 00:01:58,080 Speaker 2: I want to get to that, but I want to 38 00:01:58,080 --> 00:01:59,680 Speaker 2: push you on what you'll see with discovery because I 39 00:01:59,720 --> 00:02:01,600 Speaker 2: know this is investigation that has been a problem within 40 00:02:01,680 --> 00:02:04,920 Speaker 2: open ai as to whether or not deepseek inappropriately use 41 00:02:04,960 --> 00:02:07,920 Speaker 2: some of the inferencing data distilling from your own models. 42 00:02:08,000 --> 00:02:09,600 Speaker 2: Have you come to a conclusional We're. 43 00:02:09,440 --> 00:02:11,840 Speaker 1: Still looking at it. Obviously, there's we and we've already 44 00:02:11,880 --> 00:02:14,799 Speaker 1: made public that we had seen some evidence of that 45 00:02:14,919 --> 00:02:18,320 Speaker 1: taking place, And just so folks understand, because distillation is 46 00:02:18,360 --> 00:02:20,560 Speaker 1: not a normal word, at least certainly that I don't 47 00:02:20,600 --> 00:02:23,680 Speaker 1: use at the breakfast table all the time. There is 48 00:02:23,800 --> 00:02:27,720 Speaker 1: different kinds of distillation. And I'll use an analogy. If 49 00:02:27,760 --> 00:02:29,519 Speaker 1: you go to the library and you take out a 50 00:02:29,560 --> 00:02:31,720 Speaker 1: book and you learn from that book, and that ultimately 51 00:02:31,720 --> 00:02:34,239 Speaker 1: informs some of your work. That's fine. That takes place 52 00:02:34,240 --> 00:02:36,360 Speaker 1: a lot of that. That's part of what happens in 53 00:02:36,400 --> 00:02:38,920 Speaker 1: the AI space. There's another version of it where you 54 00:02:38,919 --> 00:02:41,720 Speaker 1: go in to the library, take the book, keep the book, 55 00:02:41,960 --> 00:02:43,760 Speaker 1: put your name on the book, slap a cover on 56 00:02:43,800 --> 00:02:45,720 Speaker 1: the book, and hand it out as if it's your book. 57 00:02:46,040 --> 00:02:48,519 Speaker 1: And that's the replication. And I think that's what we're 58 00:02:48,520 --> 00:02:51,560 Speaker 1: concerned about, and again what we've seen some evidence of 59 00:02:51,600 --> 00:02:54,280 Speaker 1: and are continuing to review to have a better understand 60 00:02:54,320 --> 00:02:56,800 Speaker 1: We've talked with government officials about it, and we'll share 61 00:02:56,840 --> 00:02:57,720 Speaker 1: more as we learn more. 62 00:02:58,040 --> 00:03:00,960 Speaker 2: Well open AIS critics have said, then trend your models 63 00:03:01,000 --> 00:03:03,320 Speaker 2: on data, and you haven't been fully transparent in terms 64 00:03:03,320 --> 00:03:04,880 Speaker 2: of the use of that data. I don't know you 65 00:03:04,919 --> 00:03:05,560 Speaker 2: would push back. 66 00:03:05,560 --> 00:03:07,680 Speaker 1: That's when there's good calories and bad calories there's a 67 00:03:07,680 --> 00:03:10,240 Speaker 1: good distallation and problematic distillation. 68 00:03:10,400 --> 00:03:12,040 Speaker 2: Have any of your customers pushed back and said, look, 69 00:03:12,040 --> 00:03:14,640 Speaker 2: we're uncomfortable with the pricing of your models at this 70 00:03:14,800 --> 00:03:16,919 Speaker 2: at this junct shot. 71 00:03:17,160 --> 00:03:20,200 Speaker 1: Well, let me put it this way. You know, open 72 00:03:20,280 --> 00:03:23,839 Speaker 1: ai came out in November of two years ago, twenty 73 00:03:23,919 --> 00:03:28,200 Speaker 1: twenty two, right, and within two months was it a 74 00:03:28,320 --> 00:03:32,239 Speaker 1: one hundred million users? Right? Were well over three hundred 75 00:03:32,360 --> 00:03:37,760 Speaker 1: million today continuing to see you know, that really strong growth. 76 00:03:38,640 --> 00:03:41,720 Speaker 1: And what we try to do is have different models 77 00:03:41,880 --> 00:03:45,920 Speaker 1: priced at different levels depending on how you ultimately want 78 00:03:46,120 --> 00:03:48,840 Speaker 1: to use it. And that's something just given the nature 79 00:03:48,840 --> 00:03:51,280 Speaker 1: of how fast the technology is moving and the pace 80 00:03:51,680 --> 00:03:55,000 Speaker 1: this moving at that you're always constantly trying to evaluate. 81 00:03:54,600 --> 00:03:55,520 Speaker 2: Prices go down from it. 82 00:03:56,000 --> 00:03:58,720 Speaker 1: Oh yeah, So I think amongst the most interesting things 83 00:03:58,760 --> 00:04:01,400 Speaker 1: that we have seen this is a an interesting aspect, 84 00:04:01,440 --> 00:04:06,240 Speaker 1: which is the efficiencies of systems, particularly using the reasoning 85 00:04:06,320 --> 00:04:09,560 Speaker 1: technology is coming down. Sam, When our CEO put a 86 00:04:09,640 --> 00:04:12,560 Speaker 1: blog out last night, maybe technically this morning, I don't know, 87 00:04:12,600 --> 00:04:14,240 Speaker 1: but at some point in the last twenty four hours 88 00:04:14,720 --> 00:04:18,000 Speaker 1: that sort of looked at three observations of AI. The 89 00:04:18,120 --> 00:04:21,240 Speaker 1: first was that the more you spend on compute to 90 00:04:21,279 --> 00:04:24,720 Speaker 1: build frontier models. It's pretty logrhithmic. The more powerful of 91 00:04:24,720 --> 00:04:26,960 Speaker 1: those gets so you're going to require more and more compute, 92 00:04:27,160 --> 00:04:31,039 Speaker 1: more and more investment in infrastructure. Secondly, over the last 93 00:04:31,160 --> 00:04:34,000 Speaker 1: year or so, we've seen a big increase in efficiencies 94 00:04:34,320 --> 00:04:36,880 Speaker 1: which are bringing down the cost of a token. Of 95 00:04:37,000 --> 00:04:38,600 Speaker 1: that I think of a token as a price of 96 00:04:38,640 --> 00:04:40,880 Speaker 1: computer as a unit of computing about one hundred and 97 00:04:40,960 --> 00:04:44,760 Speaker 1: fifty percent. But even as the costs come down, the 98 00:04:44,800 --> 00:04:47,440 Speaker 1: amount of people using it go up, so that then 99 00:04:47,480 --> 00:04:50,040 Speaker 1: puts pressures back on computing. Like an analogy here is 100 00:04:50,400 --> 00:04:52,840 Speaker 1: car prices come down, more people drive cars, but you 101 00:04:52,960 --> 00:04:55,560 Speaker 1: then need more energy, more roads. Third, piecet that has 102 00:04:55,600 --> 00:04:57,440 Speaker 1: come out and his observation, sorry, just bear with me 103 00:04:57,520 --> 00:05:01,640 Speaker 1: for a second, is that is that the economic productivity 104 00:05:01,680 --> 00:05:03,280 Speaker 1: that you're getting in is super exponential. 105 00:05:03,680 --> 00:05:05,400 Speaker 2: Where are you getting the money to spend one hundred 106 00:05:05,440 --> 00:05:08,400 Speaker 2: billion initially on star gates, elon must says, you probably 107 00:05:08,400 --> 00:05:10,760 Speaker 2: don't have the money, even something Adela says, we're good 108 00:05:10,760 --> 00:05:12,880 Speaker 2: for our eight two billion. That's Onion. 109 00:05:13,040 --> 00:05:15,800 Speaker 1: So first of all, we have incredible partners. We have 110 00:05:16,000 --> 00:05:19,400 Speaker 1: soft Bank, which is a proven track record of raising 111 00:05:19,520 --> 00:05:22,200 Speaker 1: enormous syndicated money from sovereigns and pensions. 112 00:05:23,720 --> 00:05:24,839 Speaker 2: Thought is coming from nails. 113 00:05:24,920 --> 00:05:27,479 Speaker 1: And then and then we have Oracle, right, which actually 114 00:05:27,600 --> 00:05:30,599 Speaker 1: builds these. Then you have open aies. What are piece 115 00:05:30,680 --> 00:05:33,080 Speaker 1: of this? First of all, in terms of the media question, 116 00:05:33,400 --> 00:05:34,920 Speaker 1: there's one hundred billion that's going to be going out 117 00:05:34,960 --> 00:05:37,280 Speaker 1: the door in the immediate future. We already have a 118 00:05:37,360 --> 00:05:40,520 Speaker 1: facility in Abilene. You guys need to come down. You 119 00:05:40,600 --> 00:05:43,159 Speaker 1: can abline Texas. Sorry, we'd love to give you, guys. 120 00:05:43,200 --> 00:05:45,720 Speaker 1: You can hang out the Oracle. Guys have done an 121 00:05:45,720 --> 00:05:48,880 Speaker 1: awesome job down there. No, no, no, no, why don't 122 00:05:48,880 --> 00:05:53,320 Speaker 1: you come see it? Show? We like to show, not tell. 123 00:05:53,800 --> 00:05:56,280 Speaker 1: And then and then the open AI piece comes in 124 00:05:56,320 --> 00:05:58,160 Speaker 1: two different pieces here, and I think this is to 125 00:05:58,320 --> 00:06:01,080 Speaker 1: understand the economic model here. First of all, what we 126 00:06:01,440 --> 00:06:06,039 Speaker 1: bring to this is the IP and you can think 127 00:06:06,160 --> 00:06:08,960 Speaker 1: of compute the same way. Maybe you can think of 128 00:06:09,240 --> 00:06:13,560 Speaker 1: gas regular gas, medium gas, and then premium gas. The 129 00:06:13,640 --> 00:06:15,760 Speaker 1: premium gas is what people are gonna pay for. The 130 00:06:15,880 --> 00:06:19,240 Speaker 1: premium compute is going to be the most expensive compute 131 00:06:19,279 --> 00:06:22,360 Speaker 1: in the world because it's gonna be the highest level compute. 132 00:06:22,600 --> 00:06:26,360 Speaker 1: You only get that premium compute with our IP. IP 133 00:06:26,600 --> 00:06:29,680 Speaker 1: going too, the chip design IP going into the data centers, 134 00:06:29,960 --> 00:06:32,560 Speaker 1: IP going how the clusters are sort of structured. And 135 00:06:32,680 --> 00:06:36,600 Speaker 1: then we also are a buyer of the compute, right, 136 00:06:36,680 --> 00:06:38,960 Speaker 1: so we commit to buying a certain amount of the 137 00:06:39,000 --> 00:06:41,880 Speaker 1: compute that's coming out, which helps the whole economic model work. 138 00:06:42,040 --> 00:06:44,640 Speaker 2: Chris, you worked in the Clinton White House. Is the 139 00:06:44,680 --> 00:06:49,560 Speaker 2: Trump administration there go go to great headline is to 140 00:06:49,600 --> 00:06:52,239 Speaker 2: the Trump administration right to slash AI regulation. 141 00:06:52,920 --> 00:06:55,320 Speaker 1: I think what the Trump administration is focused on is 142 00:06:55,360 --> 00:06:57,920 Speaker 1: one thing, and one thing very clearly, who is going 143 00:06:58,000 --> 00:07:01,640 Speaker 1: to prevail in the competition between democratic That's what they 144 00:07:01,680 --> 00:07:03,600 Speaker 1: get up and think about every day, at least from 145 00:07:03,600 --> 00:07:06,279 Speaker 1: what I have seen and from what I have heard, 146 00:07:06,680 --> 00:07:08,560 Speaker 1: And so I think they understand that you do have 147 00:07:08,680 --> 00:07:11,680 Speaker 1: to really be leading in leaning into the innovation. If 148 00:07:11,680 --> 00:07:15,320 Speaker 1: you think about about where the sort of comparative advantages are. 149 00:07:15,720 --> 00:07:17,520 Speaker 1: Right at the end of the day, this is actually 150 00:07:17,600 --> 00:07:21,520 Speaker 1: pretty simple. Whoever has access to compute is going to 151 00:07:21,560 --> 00:07:26,760 Speaker 1: be in a strong position. What makes up compute, it's data, 152 00:07:27,200 --> 00:07:30,760 Speaker 1: it's energy, it's chips, and it's talent. Right, and if 153 00:07:30,800 --> 00:07:32,760 Speaker 1: you think of what the PRC has, they have an 154 00:07:32,840 --> 00:07:37,040 Speaker 1: enormous amount of data. Authoritarian State Energy ten nuclear facilities 155 00:07:37,120 --> 00:07:39,560 Speaker 1: last year, another ten coming on this year. Our chips 156 00:07:39,600 --> 00:07:43,080 Speaker 1: are better. They're throwing a ton of money at it talent, 157 00:07:43,440 --> 00:07:46,000 Speaker 1: and I think the talent piece is really interesting because 158 00:07:46,320 --> 00:07:51,440 Speaker 1: they do have great talent in China. But in capitalist systems, 159 00:07:52,080 --> 00:07:58,440 Speaker 1: it is capitalism that unleashes the developer, the builder, the entrepreneur, 160 00:07:58,840 --> 00:08:00,960 Speaker 1: the people who are actually making this stuff in the 161 00:08:01,040 --> 00:08:03,400 Speaker 1: tools that starve hauled problems. And that's where our advantage is. 162 00:08:03,640 --> 00:08:06,040 Speaker 2: Is that why you've joined up and partnered with Aurae'm 163 00:08:06,040 --> 00:08:08,200 Speaker 2: to thinking of the AI principles a Google. They've adjusted 164 00:08:08,240 --> 00:08:11,280 Speaker 2: them and Google is no longer ruling out building tech 165 00:08:11,640 --> 00:08:14,520 Speaker 2: for defense for the military. Yeah, you're signing up, you 166 00:08:14,600 --> 00:08:17,000 Speaker 2: partner with Angurill. How far does that relationship go? How 167 00:08:17,080 --> 00:08:19,640 Speaker 2: much are you prepared to embed tech into weapons system? 168 00:08:19,680 --> 00:08:21,840 Speaker 1: Yeah? And we also just announced a partnership with the 169 00:08:21,960 --> 00:08:25,400 Speaker 1: National Labs, which are the you know, which play an 170 00:08:25,400 --> 00:08:29,080 Speaker 1: incredibly important role in how the US government thinks about 171 00:08:29,320 --> 00:08:31,720 Speaker 1: national security. Is on the FILS, Yeah, yes, and the 172 00:08:31,840 --> 00:08:34,520 Speaker 1: labs you know obviously like Los Alamos and others, are 173 00:08:35,040 --> 00:08:39,319 Speaker 1: very big players in the broader national security ecosystem. So 174 00:08:39,720 --> 00:08:41,719 Speaker 1: you know, for US, right, we do want to be 175 00:08:41,760 --> 00:08:45,439 Speaker 1: a partner on innovation with the government. We do believe 176 00:08:45,480 --> 00:08:49,200 Speaker 1: it's incredibly important that democratic AI prevails, and that means 177 00:08:49,280 --> 00:08:51,280 Speaker 1: making sure that the government is getting access to the 178 00:08:51,360 --> 00:08:54,120 Speaker 1: highest capabilities. You know, we'll certainly do it consistent with 179 00:08:54,240 --> 00:08:56,640 Speaker 1: our values and our principles, but at the end of 180 00:08:56,679 --> 00:08:59,120 Speaker 1: the day, like we do believe it is very consistent 181 00:08:59,440 --> 00:09:02,040 Speaker 1: with their mission that is to make sure AI benefits everyone. 182 00:09:02,520 --> 00:09:04,280 Speaker 1: That you're ensuring that AI is going to be built 183 00:09:04,320 --> 00:09:06,439 Speaker 1: in a democratic way with democratic values. 184 00:09:06,760 --> 00:09:09,040 Speaker 2: Chris la Haye, thank you very much. Indeed, Global head 185 00:09:09,080 --> 00:09:10,360 Speaker 2: of Policy at Open AI.