1 00:00:02,560 --> 00:00:07,040 Speaker 1: Bloomberg Audio Studios, podcasts, radio news. 2 00:00:09,240 --> 00:00:11,840 Speaker 2: I'm Stephen Carol and this is here is Why, where 3 00:00:11,840 --> 00:00:14,200 Speaker 2: we take one news story and explain it in just 4 00:00:14,240 --> 00:00:16,760 Speaker 2: a few minutes with our experts here at Bloomberg. 5 00:00:19,200 --> 00:00:23,000 Speaker 1: It was done very cheaply. So why the crazy amounts 6 00:00:23,040 --> 00:00:25,600 Speaker 1: of spending that are happening from the US big tech companies. 7 00:00:25,720 --> 00:00:30,000 Speaker 2: What's happening here is the collapse in the cost of innovation. 8 00:00:30,400 --> 00:00:34,200 Speaker 1: What this development has shown is the hardware that these 9 00:00:34,200 --> 00:00:37,959 Speaker 1: companies have procured, they weren't using it efficiently. A lot 10 00:00:38,000 --> 00:00:41,120 Speaker 1: of question marks around deep Seek, in particular, how much 11 00:00:41,320 --> 00:00:43,920 Speaker 1: did it really cost for them to develop this new 12 00:00:44,080 --> 00:00:46,400 Speaker 1: quote unquote cheap model that they've self reported. 13 00:00:46,560 --> 00:00:49,159 Speaker 2: The implication is, of course that things got cheap, that 14 00:00:49,200 --> 00:00:51,360 Speaker 2: things got more simple, but we'll have to see that 15 00:00:51,440 --> 00:00:54,840 Speaker 2: manifest itself in the real world. Shocking, a game changer, 16 00:00:54,960 --> 00:00:57,480 Speaker 2: a Spotnik moment, just a few of the things that 17 00:00:57,480 --> 00:01:00,680 Speaker 2: have been said about the Chinese AI startup deep Seek. 18 00:01:00,920 --> 00:01:04,600 Speaker 2: The company shot to global fame over a weekend after 19 00:01:04,640 --> 00:01:07,679 Speaker 2: the launch of it's R one chatbart, seen as arrival 20 00:01:07,720 --> 00:01:10,840 Speaker 2: to the likes of Chat GPT, but crucially, the company 21 00:01:10,840 --> 00:01:15,080 Speaker 2: says it used less expensive chips. So here's why Deep 22 00:01:15,080 --> 00:01:17,880 Speaker 2: Seek is a wake up call for the AI Titans. 23 00:01:20,240 --> 00:01:22,840 Speaker 2: Joining me Iwo to discuss Boomberg's TV anchor. Tom mackenzie, 24 00:01:22,880 --> 00:01:25,880 Speaker 2: a former China correspondent and our resident tech watcher. Thanks 25 00:01:25,880 --> 00:01:27,920 Speaker 2: for being with us. Tom, First of all, tell me 26 00:01:27,920 --> 00:01:30,680 Speaker 2: about R one, the product that Deep Seek makes. Is 27 00:01:30,720 --> 00:01:32,440 Speaker 2: it something that's really comparable to. 28 00:01:32,600 --> 00:01:35,960 Speaker 1: Chat GPT or to Gemini so on a number of 29 00:01:36,120 --> 00:01:40,280 Speaker 1: metrics measured by experts, and there are platforms online where 30 00:01:40,280 --> 00:01:43,039 Speaker 1: you can see how these measurements are displayed and how 31 00:01:43,040 --> 00:01:47,240 Speaker 1: they categorize these models. Then yes, R one competes with 32 00:01:47,400 --> 00:01:50,880 Speaker 1: chat gbt's most sophisticated model, the latest model to zero one, 33 00:01:50,880 --> 00:01:53,600 Speaker 1: but also the models from Gemini and Anthropic. It is 34 00:01:53,800 --> 00:01:57,240 Speaker 1: up there top of the list alongside those kind of players. 35 00:01:57,520 --> 00:02:01,160 Speaker 1: It is sophisticated. It is a text based model and 36 00:02:01,280 --> 00:02:03,760 Speaker 1: chat bop you can download it onto your phone. It 37 00:02:03,840 --> 00:02:06,360 Speaker 1: is not multimodal, so it doesn't produce video, it doesn't 38 00:02:06,400 --> 00:02:10,680 Speaker 1: produce pictures, but what it does do is reasoning. Like 39 00:02:11,080 --> 00:02:14,320 Speaker 1: one from chat GBT, it goes through how it comes 40 00:02:14,320 --> 00:02:17,400 Speaker 1: to its responses, and unlike O one, it actually displays 41 00:02:17,440 --> 00:02:19,640 Speaker 1: those for you, so you can see the chat bot 42 00:02:19,760 --> 00:02:23,120 Speaker 1: processing your question, going through how it's going to get 43 00:02:23,160 --> 00:02:26,679 Speaker 1: to the answer, and that level of transparency has led 44 00:02:26,720 --> 00:02:29,960 Speaker 1: to a lot of tech enthusiasts out there who are 45 00:02:29,960 --> 00:02:32,640 Speaker 1: getting their hands on this thing responding very favorably, and 46 00:02:32,639 --> 00:02:35,079 Speaker 1: that's led to a lot of optimism and some very 47 00:02:35,080 --> 00:02:38,280 Speaker 1: positive feedback about what this chatbot based on deep seeks 48 00:02:38,320 --> 00:02:40,440 Speaker 1: model can actually do. Now, one thing it can't do 49 00:02:40,960 --> 00:02:44,080 Speaker 1: is answer an honest question about Hi Jinping and his 50 00:02:44,200 --> 00:02:47,200 Speaker 1: leadership or what happened in Chaneman Square in nineteen eighty nine, 51 00:02:47,240 --> 00:02:50,520 Speaker 1: because it is not allowed to full foul of Chinese 52 00:02:50,520 --> 00:02:53,400 Speaker 1: government censorship. So there is that important caveat. 53 00:02:53,160 --> 00:02:55,639 Speaker 2: Okay, all very interesting. What do we know about how 54 00:02:55,680 --> 00:02:56,399 Speaker 2: it was developed? 55 00:02:56,800 --> 00:02:59,359 Speaker 1: What we know is what is claimed by deep Seek. 56 00:02:59,680 --> 00:03:02,240 Speaker 1: So they claim that it was developed at a fraction 57 00:03:02,480 --> 00:03:05,160 Speaker 1: of the cost of some of their competitors, around six 58 00:03:05,440 --> 00:03:09,040 Speaker 1: million US dollars. They claim the model was trained on 59 00:03:09,560 --> 00:03:13,080 Speaker 1: much older chips, not the most cutting edge in video chips, 60 00:03:13,120 --> 00:03:17,160 Speaker 1: because those are restricted from the Chinese markets. We know 61 00:03:17,360 --> 00:03:21,280 Speaker 1: that they have very innovative and sophisticated engineers. They've been 62 00:03:21,280 --> 00:03:25,440 Speaker 1: recruiting talent for the top universities domestically in China systems engineers. 63 00:03:25,480 --> 00:03:28,239 Speaker 1: We also know that they put in place an infrastructure 64 00:03:28,320 --> 00:03:30,360 Speaker 1: and a method of building models called a mixture of 65 00:03:30,360 --> 00:03:33,160 Speaker 1: experts procedure, which basically has lots of mini models much 66 00:03:33,160 --> 00:03:35,560 Speaker 1: easier to put together, and if you align them, you 67 00:03:35,560 --> 00:03:39,120 Speaker 1: can create these efficiencies. So the engineers, the mixture of 68 00:03:39,200 --> 00:03:42,640 Speaker 1: experts method that they've used, they say has led them 69 00:03:42,640 --> 00:03:45,800 Speaker 1: to creating these efficiencies, building a model much more cheaply, 70 00:03:45,920 --> 00:03:48,640 Speaker 1: using less efistigated chips, and much more quickly. They say 71 00:03:48,640 --> 00:03:52,080 Speaker 1: they built this model, design this model within about two months. 72 00:03:52,120 --> 00:03:55,040 Speaker 1: Now there are question marks. Microsoft and open ai are 73 00:03:55,040 --> 00:03:58,200 Speaker 1: scrutinizing whether or not deep Seak actually lent on open 74 00:03:58,240 --> 00:04:00,760 Speaker 1: AI's own model to learn from the outputs from that 75 00:04:00,760 --> 00:04:02,560 Speaker 1: model that they then fed into the training of the 76 00:04:02,560 --> 00:04:06,480 Speaker 1: deep Seat model that potentially went over and above what 77 00:04:06,760 --> 00:04:10,080 Speaker 1: was allowed. That is being scrutinized, And we have to 78 00:04:10,080 --> 00:04:11,600 Speaker 1: take them on face value when they talk about the 79 00:04:11,680 --> 00:04:15,000 Speaker 1: chips they're using, because we don't know exactly how they 80 00:04:15,000 --> 00:04:16,520 Speaker 1: built this. But if we take them on face value, 81 00:04:16,560 --> 00:04:19,880 Speaker 1: the more cheaply, more cost effectively, in a shorter time 82 00:04:20,320 --> 00:04:23,360 Speaker 1: and with a slightly different method, that created a very 83 00:04:23,400 --> 00:04:25,640 Speaker 1: efficient and very capable model. 84 00:04:25,880 --> 00:04:29,360 Speaker 2: Okay, got questions being asked, as you know, more broadly, 85 00:04:29,800 --> 00:04:32,119 Speaker 2: when we talk about AI open on now we've mainly 86 00:04:32,120 --> 00:04:35,160 Speaker 2: been talking about the models developed by American companies. What 87 00:04:36,040 --> 00:04:38,599 Speaker 2: is this revealing something that we didn't already know about 88 00:04:38,680 --> 00:04:41,960 Speaker 2: China's AI industry? How does it compare to what we 89 00:04:42,000 --> 00:04:42,839 Speaker 2: know out of the US. 90 00:04:43,160 --> 00:04:46,040 Speaker 1: There is an argument that we've had a blind spot 91 00:04:46,080 --> 00:04:48,520 Speaker 1: when it comes to the innovation that's coming out of China. 92 00:04:48,760 --> 00:04:50,680 Speaker 1: There's been a lot of talk about the slowdown in 93 00:04:50,720 --> 00:04:52,920 Speaker 1: the economy rightly, so, there's been a lot of concern 94 00:04:52,960 --> 00:04:54,680 Speaker 1: about the real estate market. There's been a lot of 95 00:04:54,680 --> 00:04:57,640 Speaker 1: concern about a crackdown on technology in recent years out 96 00:04:57,680 --> 00:05:01,040 Speaker 1: of China. That has allowed some to overlook the real 97 00:05:01,080 --> 00:05:03,919 Speaker 1: innovation that is happening in that Chinese market. One of 98 00:05:03,920 --> 00:05:07,240 Speaker 1: the most competitive places to build and test technology is 99 00:05:07,279 --> 00:05:09,440 Speaker 1: in the Chinese markets because you have so many people 100 00:05:09,440 --> 00:05:11,880 Speaker 1: who pile in, they test their products, they fight to 101 00:05:11,960 --> 00:05:14,000 Speaker 1: the death, and then the survivors come out on top. 102 00:05:14,000 --> 00:05:16,080 Speaker 1: And if they can compete in the Chinese market, my goodness, 103 00:05:16,080 --> 00:05:17,920 Speaker 1: they can compete globally. And we've seen that, whether it 104 00:05:18,000 --> 00:05:20,839 Speaker 1: is with drone makers like DGI, whether it's like TikTok, 105 00:05:20,920 --> 00:05:23,839 Speaker 1: social media companies like TikTok, or whether it's indeed the 106 00:05:23,880 --> 00:05:26,520 Speaker 1: solar panel makers of China. Across all those different areas, 107 00:05:26,560 --> 00:05:28,480 Speaker 1: they compete in the domestic market, they win, and then 108 00:05:28,480 --> 00:05:30,480 Speaker 1: they go to compete internationally. I haven't even mentioned the 109 00:05:30,520 --> 00:05:32,919 Speaker 1: electric vehicle makers that are posing a challenge to Europe's 110 00:05:32,960 --> 00:05:35,920 Speaker 1: EV model now as well. They have the engineers, They 111 00:05:35,920 --> 00:05:38,200 Speaker 1: have almost double the number of engineers of the US 112 00:05:38,279 --> 00:05:41,560 Speaker 1: in terms of AI engineers. They have the data in 113 00:05:41,839 --> 00:05:44,880 Speaker 1: vast quantities what they don't have, and the most sophisticated chips. 114 00:05:45,200 --> 00:05:47,799 Speaker 1: If Deep Sea really is an example of how to 115 00:05:47,839 --> 00:05:52,320 Speaker 1: circumvent that by using older chips, then China has all 116 00:05:52,360 --> 00:05:56,400 Speaker 1: the ingredients it needs to compete on the global level. Interestingly, 117 00:05:56,680 --> 00:05:59,720 Speaker 1: we also saw another Chinese company, Ali Baba, coming out 118 00:05:59,720 --> 00:06:03,000 Speaker 1: with model this week that is also as capable as 119 00:06:03,000 --> 00:06:06,080 Speaker 1: the most sophisticated models coming out of the US. So 120 00:06:06,400 --> 00:06:09,520 Speaker 1: every indication suggests that China is a serious player, and 121 00:06:09,560 --> 00:06:11,920 Speaker 1: if it hasn't caught up with the US yet, then 122 00:06:11,960 --> 00:06:13,560 Speaker 1: it's very very close to doing that. 123 00:06:13,720 --> 00:06:15,960 Speaker 2: It does seem like all of a sudden everyone has 124 00:06:16,000 --> 00:06:19,839 Speaker 2: an opinion on deep Seek and its breakthrough. Even the 125 00:06:19,960 --> 00:06:21,239 Speaker 2: US President. 126 00:06:21,160 --> 00:06:25,120 Speaker 1: The release of deep Seek AI from a Chinese company 127 00:06:25,160 --> 00:06:27,760 Speaker 1: should be a wake up call for our industries that 128 00:06:27,800 --> 00:06:30,760 Speaker 1: we need to be laser focused on competing to win. 129 00:06:31,160 --> 00:06:34,400 Speaker 2: So, Tom, what do the AIA companies in the US 130 00:06:34,600 --> 00:06:37,960 Speaker 2: need to be woken up to with this deep Seek story? 131 00:06:38,080 --> 00:06:40,800 Speaker 1: Well, interestingly, it's earnings week as well, so we've been 132 00:06:40,800 --> 00:06:42,800 Speaker 1: hearing from the CEOs of some of those major players. 133 00:06:42,839 --> 00:06:44,960 Speaker 1: We've been hearing from Mark Zuckerberg from Meta, We've been 134 00:06:45,000 --> 00:06:47,680 Speaker 1: hearing from sati In adell At, CEO of Microsoft. What 135 00:06:47,760 --> 00:06:50,400 Speaker 1: you're not hearing from them is any walk back on 136 00:06:50,440 --> 00:06:53,320 Speaker 1: the spend around CAPEX, the spend on the chips, the 137 00:06:53,360 --> 00:06:55,320 Speaker 1: spend on the data centers, the spend on the servers, 138 00:06:55,600 --> 00:06:57,960 Speaker 1: and the spend on the energy that's needed to power 139 00:06:58,000 --> 00:07:00,599 Speaker 1: this AI revolution. Because one of the major questions that 140 00:07:00,600 --> 00:07:02,960 Speaker 1: Deep Seek has posed this week is is all of 141 00:07:03,000 --> 00:07:05,839 Speaker 1: that spending worth it? Do we need to spend tens 142 00:07:05,839 --> 00:07:08,360 Speaker 1: of billions of dollars and all of that AI infrastructure? 143 00:07:08,800 --> 00:07:12,160 Speaker 1: If a bootstrap company in China with six million dollars 144 00:07:12,240 --> 00:07:14,800 Speaker 1: can produce a model that competes with Chat GBT at 145 00:07:14,840 --> 00:07:18,000 Speaker 1: a fraction of the price. But the CEOs are not 146 00:07:18,160 --> 00:07:22,520 Speaker 1: walking back from these investment commitments. In fact, Mark Zuckerberg, 147 00:07:22,520 --> 00:07:25,040 Speaker 1: who praised deep Seat and said there was real innovation there. 148 00:07:25,240 --> 00:07:28,280 Speaker 1: He's committed to sixty five billion dollars of spending this year. 149 00:07:28,680 --> 00:07:31,520 Speaker 1: He says they're going to be building their own AI agents, 150 00:07:31,520 --> 00:07:33,320 Speaker 1: and he hopes the Meta is going to be getting 151 00:07:33,360 --> 00:07:35,360 Speaker 1: those agents to a billion people by the end of 152 00:07:35,360 --> 00:07:37,200 Speaker 1: this year, and that they'll be leading in that space. 153 00:07:37,280 --> 00:07:40,320 Speaker 1: Microsoft Satiy and Nadella also praising deep Seek, but his 154 00:07:40,480 --> 00:07:42,800 Speaker 1: take was this is going to lead to greater adoption 155 00:07:42,840 --> 00:07:46,120 Speaker 1: of artificial intelligence. It will drive down prices, and longer 156 00:07:46,200 --> 00:07:48,960 Speaker 1: term that will be good news for Microsoft. Microsoft committing 157 00:07:48,960 --> 00:07:52,680 Speaker 1: to spend eighty billion dollars this fiscal year on AI infrastructure. 158 00:07:53,080 --> 00:07:55,560 Speaker 2: Okay, a rude awakening for some, perhaps not for everyone. 159 00:07:55,600 --> 00:07:58,040 Speaker 2: Tom McKenzie, our Blimberg TV anchor, thank you very much 160 00:07:58,080 --> 00:08:00,880 Speaker 2: for joining us for more xp like this from our 161 00:08:00,920 --> 00:08:03,679 Speaker 2: team of twenty nine hundred journalists and analysts around the world. 162 00:08:03,720 --> 00:08:06,560 Speaker 2: Search for Quick Take on the Bloomberg website or Bloomberg 163 00:08:06,600 --> 00:08:11,320 Speaker 2: Business app. I'm Stephen Carroll. This is Here's why I'll 164 00:08:11,360 --> 00:08:13,600 Speaker 2: be back next week with more thanks for listening