1 00:00:02,720 --> 00:00:07,200 Speaker 1: Bloomberg Audio Studios, podcasts, radio news. 2 00:00:08,600 --> 00:00:11,640 Speaker 2: Earlier this year, a new product from the Chinese AI 3 00:00:11,720 --> 00:00:15,400 Speaker 2: startup Deep Seek, shocked the world and rattled Wall Street. 4 00:00:16,079 --> 00:00:20,040 Speaker 2: China's deep Seek is freaking out the AI world right now. 5 00:00:20,079 --> 00:00:23,279 Speaker 2: Techtoks tumbled as It's absurd to the top of the 6 00:00:23,360 --> 00:00:27,920 Speaker 2: download charts. But despite the global attention, very little is 7 00:00:28,000 --> 00:00:32,800 Speaker 2: known about the man behind deep Seek. Chinese entrepreneur Liang Wu. 8 00:00:33,120 --> 00:00:35,839 Speaker 1: Liang Wenfeng is certainly a mystery figure. 9 00:00:36,400 --> 00:00:40,080 Speaker 2: Bloomberg crit the rye covers artificial intelligence in Asia. 10 00:00:40,680 --> 00:00:44,240 Speaker 1: He's certainly one of the most inaccessible and low key 11 00:00:44,800 --> 00:00:48,319 Speaker 1: tech entrepreneurs that I've come across. Just to illustrate how 12 00:00:48,360 --> 00:00:52,840 Speaker 1: private he is, we weren't able to find any pictures 13 00:00:52,960 --> 00:00:56,600 Speaker 1: of him on the internet when we scoured through his 14 00:00:56,760 --> 00:01:00,520 Speaker 1: website and all of that. But finally appear in a 15 00:01:00,600 --> 00:01:05,520 Speaker 1: really high profile meeting with President Sheijeng Peing, and that 16 00:01:05,560 --> 00:01:08,440 Speaker 1: picture got out into the world and he was everywhere. 17 00:01:09,240 --> 00:01:11,120 Speaker 2: What does this man of mystery look like? 18 00:01:11,920 --> 00:01:18,440 Speaker 1: He is slim, wears glasses, but doesn't talk much. Baby faced, Yes, 19 00:01:18,680 --> 00:01:20,319 Speaker 1: I think we could describe him like that. 20 00:01:25,240 --> 00:01:29,119 Speaker 2: Deep Seek rarely answers questions about Leang citing his privacy. 21 00:01:29,640 --> 00:01:32,399 Speaker 2: But Sarita and her colleagues were curious about this man 22 00:01:32,560 --> 00:01:35,480 Speaker 2: whose AI systems turned the tech world on its head. 23 00:01:36,040 --> 00:01:38,839 Speaker 2: So they spoke with dozens of people familiar with his work, 24 00:01:39,040 --> 00:01:43,040 Speaker 2: from former employees and fellow researchers to investors and insiders 25 00:01:43,040 --> 00:01:43,760 Speaker 2: in the industry. 26 00:01:44,360 --> 00:01:47,760 Speaker 1: And what we found is that, yes, he is extraordinarily 27 00:01:47,840 --> 00:01:54,880 Speaker 1: low key, very shy, but extraordinarily driven and talented and passionate. 28 00:01:55,240 --> 00:01:58,640 Speaker 1: And I think he has somewhat taken on Deep Seek 29 00:01:58,680 --> 00:02:03,040 Speaker 1: as a sort of a mission to establish China in AI, 30 00:02:03,560 --> 00:02:07,480 Speaker 1: trying to make sure that China is the force to 31 00:02:07,560 --> 00:02:09,320 Speaker 1: reckon with in AI. 32 00:02:12,680 --> 00:02:15,920 Speaker 2: Welcome to the Big Take Asia from Bloomberg News. I'm wanha. 33 00:02:16,600 --> 00:02:18,960 Speaker 2: Every week we take you inside some of the world's 34 00:02:18,960 --> 00:02:22,920 Speaker 2: biggest and most powerful economies and the markets, tycoons and 35 00:02:23,040 --> 00:02:27,520 Speaker 2: businesses that drive this ever shifting region. Today, on the show, 36 00:02:27,680 --> 00:02:31,079 Speaker 2: who is LNG One Fun, we learn about the mysterious 37 00:02:31,120 --> 00:02:34,040 Speaker 2: tech founder who led Deep Seek to the frontline of 38 00:02:34,120 --> 00:02:38,680 Speaker 2: AI advances. Plus what is the company's rapid rise? Tell 39 00:02:38,760 --> 00:02:45,840 Speaker 2: us about the US China artificial intelligence race. Sriita, thanks 40 00:02:45,840 --> 00:02:49,240 Speaker 2: for joining us. I'm fascinated by AI. You guys did 41 00:02:49,280 --> 00:02:52,040 Speaker 2: such an interesting job with a story. I wonder if 42 00:02:52,040 --> 00:02:54,800 Speaker 2: we can start with who is Lang one fung? What 43 00:02:54,840 --> 00:02:56,320 Speaker 2: do we know about his roots? 44 00:02:56,800 --> 00:03:01,040 Speaker 1: So? Liang is about forty years old in a small 45 00:03:01,120 --> 00:03:06,760 Speaker 1: village called Middling in the Guangdong province. His parents were 46 00:03:07,000 --> 00:03:12,960 Speaker 1: school teachers, mainly taught primary school kids. He was extremely 47 00:03:13,160 --> 00:03:18,280 Speaker 1: bright and went on to study at Xijiang University and 48 00:03:18,360 --> 00:03:20,519 Speaker 1: then also did his master's there. 49 00:03:21,040 --> 00:03:24,800 Speaker 2: At Jujiung University. Liang and his friends immersed themselves in 50 00:03:24,919 --> 00:03:30,600 Speaker 2: all things tech, machine learning, signal processing, electronic engineering. They 51 00:03:30,639 --> 00:03:34,680 Speaker 2: even developed programs to trade stocks during the financial crisis. 52 00:03:35,280 --> 00:03:39,160 Speaker 2: After graduating, Liang joined forces with two of his classmates 53 00:03:39,360 --> 00:03:43,080 Speaker 2: and set up a quantitative hedge fund called High Fire Management. 54 00:03:44,160 --> 00:03:48,360 Speaker 1: So quant funds basically work with mathematical models and statistical 55 00:03:48,400 --> 00:03:53,120 Speaker 1: analysis to do stock trading. Humans are not involved in 56 00:03:53,160 --> 00:03:57,680 Speaker 1: taking the decisions. At its peak, High Flyer Management was 57 00:03:57,760 --> 00:04:02,240 Speaker 1: managing something like fourteen billion in assets, so it was 58 00:04:02,480 --> 00:04:07,000 Speaker 1: quite a sizeable fund, and in its most successful runs, 59 00:04:07,080 --> 00:04:12,480 Speaker 1: it was providing annualized returns averaging thirty five percent to 60 00:04:12,560 --> 00:04:15,040 Speaker 1: its investors, so I would say that it was doing 61 00:04:15,120 --> 00:04:15,520 Speaker 1: very well. 62 00:04:15,560 --> 00:04:20,360 Speaker 2: Indeed, according to former employees. High Flyer had a geeky 63 00:04:20,400 --> 00:04:24,480 Speaker 2: startup culture. Its early job postings moosted of attracting top 64 00:04:24,560 --> 00:04:27,440 Speaker 2: talent from Google and Facebook and said they were looking 65 00:04:27,480 --> 00:04:30,760 Speaker 2: for math and coding geeks with quirky brilliants. 66 00:04:32,120 --> 00:04:37,359 Speaker 1: The early job postings also referred to Sheldon, who is 67 00:04:37,560 --> 00:04:43,919 Speaker 1: this very awkward main character in the prominent American sitcom 68 00:04:44,000 --> 00:04:45,520 Speaker 1: called The Big Bang Pory. 69 00:04:46,240 --> 00:04:48,960 Speaker 2: For example, I cry because others are stupid and it 70 00:04:48,960 --> 00:04:49,599 Speaker 2: makes me sad. 71 00:04:51,480 --> 00:04:56,680 Speaker 1: Sheldon has a legend of fans and it's extraordinarily funny 72 00:04:56,680 --> 00:05:00,520 Speaker 1: without meaning to be. So, you know, the whole culture 73 00:05:00,600 --> 00:05:04,360 Speaker 1: of deep Seek in the early days revolved around recreating 74 00:05:04,400 --> 00:05:08,520 Speaker 1: some of that geeky nerdy culture. They were free snacks, 75 00:05:08,760 --> 00:05:12,880 Speaker 1: poker game nights. Everybody was dressed in T shirts and slippers. 76 00:05:13,120 --> 00:05:14,520 Speaker 2: Sounds like a great place to work. 77 00:05:14,600 --> 00:05:19,320 Speaker 1: Yeah. It was really a very unorthodox startup culture, unlike 78 00:05:19,600 --> 00:05:22,440 Speaker 1: what you'll probably see in the big tech companies in 79 00:05:22,520 --> 00:05:24,560 Speaker 1: China such as the Aliba Bay and the Tencent. 80 00:05:25,080 --> 00:05:28,479 Speaker 2: And how did Liang transition from doing quant financials into 81 00:05:28,600 --> 00:05:30,120 Speaker 2: AI and building deep seak. 82 00:05:30,800 --> 00:05:37,720 Speaker 1: Liang was always extraordinarily interested in machine learning and artificial intelligence, 83 00:05:38,360 --> 00:05:41,640 Speaker 1: and then a few months after Open Ai you know 84 00:05:41,760 --> 00:05:47,640 Speaker 1: launched chat GPT, the chatbot that became an overnight global success. 85 00:05:47,960 --> 00:05:51,000 Speaker 1: It was then the spring of twenty twenty three, a 86 00:05:51,040 --> 00:05:54,120 Speaker 1: few months had passed after the launch of chat GPT, 87 00:05:55,080 --> 00:06:00,839 Speaker 1: and Liang then announced that deep Seek would be set up. 88 00:06:01,480 --> 00:06:06,400 Speaker 1: In its early manifesto, deep Seek talked about shunning mediocrity 89 00:06:07,080 --> 00:06:12,000 Speaker 1: and tackling the big challenges in AI, and of course, 90 00:06:12,120 --> 00:06:15,640 Speaker 1: ultimately cracking artificial general intelligence. 91 00:06:16,400 --> 00:06:20,440 Speaker 2: The manifesto also laid out deep Seek's ambition to position 92 00:06:20,640 --> 00:06:23,440 Speaker 2: China as a leader in cutting edge technologies. 93 00:06:24,360 --> 00:06:27,960 Speaker 1: You Know, Liang has given two interviews, rare as they 94 00:06:28,000 --> 00:06:32,080 Speaker 1: may be. In both those interviews, he's talked about bringing 95 00:06:32,200 --> 00:06:36,800 Speaker 1: China's AI ecosystem to the forefront of where the world is. 96 00:06:37,160 --> 00:06:40,279 Speaker 1: You know, China has been accused constantly of being a copycat. 97 00:06:40,560 --> 00:06:43,760 Speaker 1: He wanted in AI China to chart a different path. 98 00:06:44,720 --> 00:06:48,719 Speaker 2: Deep Seek worked fast since twenty twenty three, It's released 99 00:06:48,720 --> 00:06:51,919 Speaker 2: over half a dozen AI models and helped pioneer a 100 00:06:51,960 --> 00:06:55,880 Speaker 2: technique called sparsity, which enabled those models to train and 101 00:06:55,960 --> 00:07:01,400 Speaker 2: run more efficiently. Developers started to take note then earlier 102 00:07:01,480 --> 00:07:03,040 Speaker 2: this year. Back to that top. 103 00:07:02,839 --> 00:07:06,480 Speaker 1: Story, now deep Seek shaking up global tech lower technology 104 00:07:06,480 --> 00:07:09,680 Speaker 1: co host when they released their reasoning model R one 105 00:07:10,280 --> 00:07:14,560 Speaker 1: that cost such an upheaval in the industry and coused 106 00:07:15,000 --> 00:07:18,880 Speaker 1: a trillion dollar stock market meltdown. That's when the world 107 00:07:19,200 --> 00:07:25,440 Speaker 1: really started paying attention to this secretive AI entrepreneur in China. 108 00:07:29,640 --> 00:07:33,240 Speaker 2: And Trita. What is show groundbreaking about Deep seeks are 109 00:07:33,320 --> 00:07:33,880 Speaker 2: one model? 110 00:07:34,520 --> 00:07:39,240 Speaker 1: The AI industry until recently was always about billions of 111 00:07:39,280 --> 00:07:43,440 Speaker 1: dollars spent in building the infrastructure, the data centers, the 112 00:07:43,920 --> 00:07:47,640 Speaker 1: graphics processing units in the data centers that would train 113 00:07:47,720 --> 00:07:52,480 Speaker 1: these models. But what Deepsek did was show that its 114 00:07:52,600 --> 00:07:56,880 Speaker 1: models could match or even outdoor on some benchmark measures 115 00:07:57,480 --> 00:08:02,400 Speaker 1: what the latest open AI or anthropic models were doing, 116 00:08:03,000 --> 00:08:08,440 Speaker 1: and with far less computational power, with far less resources, 117 00:08:08,480 --> 00:08:12,520 Speaker 1: and as Deep Sea claimed, with far less capital as well. 118 00:08:14,360 --> 00:08:16,960 Speaker 2: So how did Liang and his team manage to achieve 119 00:08:17,120 --> 00:08:20,080 Speaker 2: true innovation at what it says is a fraction of 120 00:08:20,120 --> 00:08:23,320 Speaker 2: the cost? And what does deep seek success say about 121 00:08:23,320 --> 00:08:27,480 Speaker 2: the AI race between China and the US. That's after 122 00:08:27,520 --> 00:08:41,040 Speaker 2: the break. For much of the last decade, the US 123 00:08:41,080 --> 00:08:45,920 Speaker 2: has tried to restrict China's access to semiconductors. Tensions reached 124 00:08:45,920 --> 00:08:48,600 Speaker 2: a fever pitch in twenty twenty two and the following 125 00:08:48,679 --> 00:08:52,520 Speaker 2: year when Washington targeted Beijing with two rounds of chip 126 00:08:52,720 --> 00:08:55,640 Speaker 2: export controls in Vidia and shares. 127 00:08:55,679 --> 00:08:59,520 Speaker 1: Some semiconductor companies have slumped today after the Biden administration 128 00:08:59,679 --> 00:09:03,400 Speaker 1: said it would tighten restrictions on exports of AI chips 129 00:09:03,400 --> 00:09:05,960 Speaker 1: to China. 130 00:09:05,360 --> 00:09:08,920 Speaker 2: That limited sales from American firms like Nvidia, whose cutting 131 00:09:09,000 --> 00:09:11,800 Speaker 2: edge chips are used by tech companies to help train 132 00:09:11,960 --> 00:09:15,960 Speaker 2: their AI models. The move presented a significant challenge for 133 00:09:16,040 --> 00:09:19,680 Speaker 2: developers in China, but as Bloomberg's read the rises, it 134 00:09:19,840 --> 00:09:22,559 Speaker 2: also forced them to develop workarounds. 135 00:09:23,120 --> 00:09:27,160 Speaker 1: Necessity is always the mother of innovation. This has been 136 00:09:27,320 --> 00:09:33,680 Speaker 1: proven by AI developers in China. Never mind the export cubs. 137 00:09:34,120 --> 00:09:37,200 Speaker 1: They've still gone on to build good models that have 138 00:09:37,400 --> 00:09:39,840 Speaker 1: benchmarked with the best around the world. 139 00:09:40,559 --> 00:09:43,360 Speaker 2: And one of the most innovative approaches from Deepseek is 140 00:09:43,360 --> 00:09:45,439 Speaker 2: a sparsity technique we mentioned earlier. 141 00:09:47,120 --> 00:09:51,120 Speaker 1: Now. Sparsity is something to do with building a model 142 00:09:51,160 --> 00:09:55,400 Speaker 1: without having the high end computational power. It's when a 143 00:09:55,559 --> 00:09:59,120 Speaker 1: large language model doesn't have to be entirely harnessed to 144 00:09:59,160 --> 00:10:02,960 Speaker 1: give a nan to a query. Instead, Liang and his 145 00:10:03,400 --> 00:10:10,000 Speaker 1: fellow developers tried to apportion the expertise of the model 146 00:10:10,080 --> 00:10:14,120 Speaker 1: into smaller expert groups and then only harness those groups 147 00:10:14,160 --> 00:10:17,840 Speaker 1: that required to be used. So in doing that, they 148 00:10:18,240 --> 00:10:21,920 Speaker 1: made it much more computation efficient and also much more 149 00:10:21,960 --> 00:10:22,760 Speaker 1: cost efficient. 150 00:10:23,679 --> 00:10:26,679 Speaker 2: Is it basically, instead of using your whole brain, are 151 00:10:26,720 --> 00:10:30,240 Speaker 2: you using just certain parts of your brain to do 152 00:10:30,400 --> 00:10:31,679 Speaker 2: that computation. 153 00:10:31,880 --> 00:10:35,560 Speaker 1: That's exactly right on? You know, instead of entirely using 154 00:10:35,640 --> 00:10:39,400 Speaker 1: every little gray cell in your brain, it only fires 155 00:10:39,520 --> 00:10:42,880 Speaker 1: up those neurons or little portions in your brain that 156 00:10:42,960 --> 00:10:47,560 Speaker 1: contained that particular field of expertise and then bring that 157 00:10:47,880 --> 00:10:50,280 Speaker 1: to you know, respond to a query or give an 158 00:10:50,280 --> 00:10:53,599 Speaker 1: answer to a particular question, whether it's a command or 159 00:10:53,640 --> 00:10:54,520 Speaker 1: a coding question. 160 00:10:56,720 --> 00:11:00,560 Speaker 2: The sparsity breaks through impressed Deep seeks competitors, but its 161 00:11:00,640 --> 00:11:05,040 Speaker 2: price point is what ultimately made headlines. Deep Seek said 162 00:11:05,040 --> 00:11:07,760 Speaker 2: it cost them just five point six million dollars to 163 00:11:07,840 --> 00:11:11,360 Speaker 2: train its V three model. That's far less than they 164 00:11:11,440 --> 00:11:14,800 Speaker 2: estimated one hundred million dollars Open Ai spent on its 165 00:11:14,800 --> 00:11:17,120 Speaker 2: most advanced version of chat GPT. 166 00:11:18,000 --> 00:11:21,439 Speaker 1: Now, there is definitely a whole lot of skepticism around 167 00:11:21,880 --> 00:11:27,400 Speaker 1: that number because just the intrastructure, the training of the model, 168 00:11:27,520 --> 00:11:31,280 Speaker 1: the talent, and the time it takes all of it 169 00:11:31,280 --> 00:11:34,040 Speaker 1: adds to quite a sizable sum of money, so the 170 00:11:34,160 --> 00:11:39,240 Speaker 1: skepticism is warranted. People have estimated that there was no 171 00:11:39,360 --> 00:11:41,679 Speaker 1: way Deep Seak could have pulled that off without at 172 00:11:41,760 --> 00:11:43,280 Speaker 1: least a billion dollars or more. 173 00:11:44,679 --> 00:11:47,760 Speaker 2: Also in Deep Seek's favor is that AI startups like 174 00:11:47,800 --> 00:11:51,160 Speaker 2: it have a staunch ally in China's government and present. 175 00:11:51,200 --> 00:11:56,160 Speaker 2: She Jinkin Sriita says she sees generative AI, robotics and 176 00:11:56,200 --> 00:11:59,640 Speaker 2: other high tech ambitions as beneficial to the state's agenda, 177 00:12:00,120 --> 00:12:03,760 Speaker 2: part of a larger push for self reliance in key technologies, 178 00:12:04,360 --> 00:12:07,480 Speaker 2: and Deep seax success has spurred much bigger rivals such 179 00:12:07,480 --> 00:12:10,720 Speaker 2: as Ali, Baba, ten Cent, and byte Edance to release 180 00:12:10,760 --> 00:12:15,800 Speaker 2: their own AI models. Srida deep six model is entirely 181 00:12:15,840 --> 00:12:18,880 Speaker 2: open sourced at this point. That means any individual or 182 00:12:18,960 --> 00:12:23,319 Speaker 2: company could incorporate deep Six's algorithms into their own programs. 183 00:12:24,040 --> 00:12:26,760 Speaker 2: Why did the company choose this approach and why is 184 00:12:26,760 --> 00:12:29,040 Speaker 2: that important open source? 185 00:12:29,440 --> 00:12:34,240 Speaker 1: On one level, you could say that it is democratizing 186 00:12:34,400 --> 00:12:37,960 Speaker 1: AI and taking it out into the world, But let's 187 00:12:38,040 --> 00:12:43,560 Speaker 1: not forget that China's AI models would have otherwise found 188 00:12:43,640 --> 00:12:47,800 Speaker 1: fewer takers around the world if they were proprietary models 189 00:12:47,840 --> 00:12:51,439 Speaker 1: and were on par in terms of what they cost 190 00:12:51,600 --> 00:12:55,400 Speaker 1: with Western companies such as open ai. By making it 191 00:12:55,480 --> 00:12:59,400 Speaker 1: cheap and by making it open source, China allowed people 192 00:12:59,440 --> 00:13:01,680 Speaker 1: around the world to quickly take a look at the 193 00:13:01,720 --> 00:13:07,400 Speaker 1: models and begin using them, allowing them to be adopted 194 00:13:07,480 --> 00:13:13,199 Speaker 1: much faster in the business and AI ecosystem, thereby outdoing 195 00:13:13,400 --> 00:13:17,080 Speaker 1: the likes of OpenAI. Now that's huge. It's not only 196 00:13:17,160 --> 00:13:22,080 Speaker 1: about democratizing models, it's strategically about making sure that you 197 00:13:22,200 --> 00:13:25,440 Speaker 1: cut out your competitor by making things so cheap that 198 00:13:25,720 --> 00:13:33,199 Speaker 1: the world adopts it quickly and then it becomes mainstream. 199 00:13:33,360 --> 00:13:36,720 Speaker 2: As a result, Microsoft and Amazon both offer deep seek 200 00:13:36,760 --> 00:13:39,760 Speaker 2: on their cloud services, and deep seeks models have been 201 00:13:39,760 --> 00:13:43,600 Speaker 2: incorporated into Perplexity, an AI powered search engine that also 202 00:13:43,640 --> 00:13:46,160 Speaker 2: offers models from open Ai and Anthropic. 203 00:13:47,559 --> 00:13:50,920 Speaker 1: There is definitely a question about, you know, how fast 204 00:13:51,000 --> 00:13:54,400 Speaker 1: AI is advancing, and there's a fear around the world 205 00:13:54,640 --> 00:13:58,920 Speaker 1: about having all of the controls rest only with one 206 00:13:59,000 --> 00:14:01,120 Speaker 1: or two companies in the world. I think that was 207 00:14:01,160 --> 00:14:04,240 Speaker 1: what jep Seek and others were trying to put out 208 00:14:04,240 --> 00:14:06,800 Speaker 1: a message in the world, saying that all of the 209 00:14:06,800 --> 00:14:09,880 Speaker 1: controls cannot be left to one or two companies and 210 00:14:09,960 --> 00:14:13,400 Speaker 1: the proprietary models that they are building, it should be 211 00:14:13,480 --> 00:14:17,240 Speaker 1: much more democratic. Therefore, I think the open source philosophy 212 00:14:17,480 --> 00:14:23,480 Speaker 1: is about de risking concentration and allowing more people to 213 00:14:23,640 --> 00:14:27,960 Speaker 1: build with technologies that are much more available. 214 00:14:29,120 --> 00:14:31,880 Speaker 2: Is there potentially a kind of clash of cultures or 215 00:14:31,920 --> 00:14:34,040 Speaker 2: a clash of values as well when it comes to 216 00:14:34,080 --> 00:14:38,560 Speaker 2: building AI between the Wesk's approach and the Chinese approach. 217 00:14:38,800 --> 00:14:42,840 Speaker 1: Very clearly, because if you look at early models of 218 00:14:42,840 --> 00:14:47,360 Speaker 1: deep Seek or even the you know, not tweaked or 219 00:14:47,400 --> 00:14:50,280 Speaker 1: fine tune models of deep Seek, they are very much 220 00:14:50,440 --> 00:14:54,560 Speaker 1: working within the boundaries of China's censorship rules. For instance, 221 00:14:54,640 --> 00:14:59,240 Speaker 1: you cannot ask questions about Taiwan or Sheishingping without it 222 00:14:59,320 --> 00:15:04,280 Speaker 1: giving a very bland official answer. Whereas if you take 223 00:15:04,320 --> 00:15:07,960 Speaker 1: that same model and you can train it with other 224 00:15:08,040 --> 00:15:14,080 Speaker 1: data and make it culturally suitable to different geographies. That's 225 00:15:14,160 --> 00:15:17,080 Speaker 1: one of the things that deep Seek learned early on 226 00:15:17,400 --> 00:15:23,520 Speaker 1: that by open sourcing the model and by giving developers 227 00:15:23,560 --> 00:15:28,400 Speaker 1: and users a chance to customize to their own cultural context, 228 00:15:29,080 --> 00:15:33,400 Speaker 1: deep Seak could find much more quicker and faster adoption 229 00:15:33,520 --> 00:15:36,280 Speaker 1: around the world than by controlling a lot of it 230 00:15:36,640 --> 00:15:39,840 Speaker 1: and controlling it in a way that it could only 231 00:15:39,920 --> 00:15:42,920 Speaker 1: give China friendly answers around the world. 232 00:15:45,320 --> 00:15:49,240 Speaker 2: And while some applied China's innovations in AI, many of 233 00:15:49,280 --> 00:15:53,600 Speaker 2: the US suspect darker reasons for the success. An April 234 00:15:53,680 --> 00:15:57,840 Speaker 2: report from a US House of Representatives committee alleged significant 235 00:15:57,880 --> 00:16:01,640 Speaker 2: ties between Deep Seek and the Chinese got It concluded 236 00:16:01,680 --> 00:16:05,680 Speaker 2: that the company unlawfully stole data from Open Ai. The 237 00:16:05,800 --> 00:16:10,520 Speaker 2: Chinese embassy rejects those claims as groundless. Meanwhile, Deep Seek 238 00:16:10,560 --> 00:16:15,000 Speaker 2: and Liang haven't commented on the House report. Srita, it 239 00:16:15,080 --> 00:16:17,360 Speaker 2: seems like there is very much of an arms race 240 00:16:17,360 --> 00:16:19,800 Speaker 2: of sorts when it comes to AI, certainly between the 241 00:16:19,920 --> 00:16:20,600 Speaker 2: US and China. 242 00:16:20,600 --> 00:16:25,360 Speaker 1: At the moment, it's very much a race, and I 243 00:16:25,400 --> 00:16:29,000 Speaker 1: think it would be too early to call a winner. 244 00:16:29,880 --> 00:16:33,040 Speaker 1: All I can say is that a year ago I 245 00:16:33,080 --> 00:16:36,720 Speaker 1: would not have called it as a close race. It's 246 00:16:36,760 --> 00:16:39,120 Speaker 1: a marathon, but you have to go at a sprinting pace. 247 00:16:39,640 --> 00:16:42,000 Speaker 1: We are really at the very beginning of it, and 248 00:16:42,360 --> 00:16:47,120 Speaker 1: for whichever country cracks the race, there is a lot 249 00:16:47,200 --> 00:16:52,320 Speaker 1: of economic gains to be had. So every country, particularly 250 00:16:52,320 --> 00:16:54,560 Speaker 1: the US and China, do not want to let up 251 00:16:54,600 --> 00:16:55,080 Speaker 1: in AI. 252 00:16:55,800 --> 00:16:58,240 Speaker 2: And what are the challenges ahead right now for Deep 253 00:16:58,280 --> 00:16:59,400 Speaker 2: Seek that you see. 254 00:16:59,680 --> 00:17:03,200 Speaker 1: I think one of the main challenges is what next, 255 00:17:03,520 --> 00:17:06,879 Speaker 1: What can they do that outdes what they've already done. 256 00:17:07,040 --> 00:17:10,600 Speaker 1: But there's also, I think for Deep Seak competition within 257 00:17:10,640 --> 00:17:15,200 Speaker 1: its own home ground, a bunch of China companies such 258 00:17:15,240 --> 00:17:19,479 Speaker 1: as Ali Baba and Paita and Tencent building models that 259 00:17:19,600 --> 00:17:25,080 Speaker 1: are outdoing Deep Seek's last flagship model. So there's this 260 00:17:25,720 --> 00:17:30,119 Speaker 1: pressure on Deep Seek to do better. But also I 261 00:17:30,160 --> 00:17:35,240 Speaker 1: think there is also the question around commercializing these models. 262 00:17:35,240 --> 00:17:38,520 Speaker 1: How are companies like deep Seak going to make money? 263 00:17:39,119 --> 00:17:42,359 Speaker 1: There is no clear answer yet whether Deep Seak wants 264 00:17:42,359 --> 00:17:44,399 Speaker 1: to make money and if it does, how will it 265 00:17:44,480 --> 00:17:47,720 Speaker 1: make money. 266 00:17:49,320 --> 00:17:51,960 Speaker 2: This is the Big Take Asia from Bloomberg News. I'm 267 00:17:51,960 --> 00:17:55,680 Speaker 2: wan Ha. This episode was produced by Young Young and Naomi. 268 00:17:56,480 --> 00:17:59,280 Speaker 2: It was edited by Patty Hirsh and Joshua Brustin. It 269 00:17:59,320 --> 00:18:02,159 Speaker 2: was fact check by Bloomberg's editorial team and mixed and 270 00:18:02,200 --> 00:18:06,320 Speaker 2: sound designed by Taka Yasuzawa. Our senior producer is named Lushaven. 271 00:18:06,720 --> 00:18:10,800 Speaker 2: Our senior editor is Elizabeth Ponso. Our deputy executive producer 272 00:18:10,920 --> 00:18:14,359 Speaker 2: is Julia Weaver. Our executive producer is Nicole Beemster Bower. 273 00:18:14,520 --> 00:18:17,560 Speaker 2: Sage Bauman is Bloomberg's head of podcasts. If you liked 274 00:18:17,600 --> 00:18:20,280 Speaker 2: this episode, make sure to subscribe and review The Big 275 00:18:20,359 --> 00:18:23,480 Speaker 2: Tack Asia wherever you listen to podcasts. It really helps 276 00:18:23,480 --> 00:18:26,160 Speaker 2: people find the show. Thanks for listening, See you next time.