중국의 신규 AI 모델인 DeepSeek이 미국의 지배력을 위협한다는 주제의 동영상

How China’s New AI Model DeepSeek Is Threatening U.S. Dominance - YouTube

중국의 신규 AI 모델인 DeepSeek이 미국의 지배력을 위협한다는 주제의 동영상입니다. 아래는 관련된 YouTube 링크들입니다:

1. How China's New AI Model DeepSeek Is Threatening U.S. Dominance - 중국에서 개발한 새로운 AI 모델이 실리콘밸리를 긴장시키고 있다는 내용을 다룬 영상입니다.
2. China's 'Deepseek' threatening US dominance in AI market - Geoff Harris가 중국의 AI 기술을 분석한 내용입니다.
3. Chinese start-up DeepSeek threatens American AI dominance - CNBC에서 진행한 AI 경합에 관한 보고입니다.

원하시는 주제에 대해 자세히 시청하실 수 있습니다.

Transcript:
(00:00) China's latest AI breakthrough has leapfrogged the world. I think we should take the development out of China very, very seriously. A game changing move that does not come from OpenAI, Google or Meta. There is a new model that has all of the valley buzzing. But from a Chinese lab called Deepseek. It's opened a lot of eyes of like what is actually happening in AI in China.
(00:25) What took Google and OpenAI years and hundreds of millions of dollars to build... Deepseek says took it just two months and less than $6 million dollars. They have the best open source model, and all the American developers are building on that. I'm Deirdre Bosa with the tech check take... China's AI breakthrough.
(00:54) It was a technological leap that shocked Silicon Valley. A newly unveiled free, open-source AI model that beats some of the most powerful ones on the market. But it wasn't a new launch from OpenAI or model announcement from Anthropic. This one was built in the East by a Chinese research lab called Deepseek. And the details behind its development stunned top AI researchers here in the U.S.
(01:18) First-the cost. The AI lab reportedly spent just $5.6 million dollars to build Deepseek version 3. Compare that to OpenAI, which is spending $5 billion a year, and Google, which expects capital expenditures in 2024 to soar to over $50 billion. And then there's Microsoft that shelled out more than $13 billion just to invest in OpenAI.
(01:40) But even more stunning how Deepseek's scrap pier model was able to outperform the lavishly-funded American ones. To see the Deepseek, new model. It's super impressive in terms of both how they have really effectively done an open-source model that does what is this inference time compute. And it's super compute efficient.
(02:01) It beat Meta's Llama, OpenAI's GPT 4-O and Anthropic's Claude Sonnet 3.5 on accuracy on wide-ranging tests. A subset of 500 math problems, an AI math evaluation, coding competitions, and a test of spotting and fixing bugs in code. Quickly following that up with a new reasoning model called R1, which just as easily outperformed OpenAI's cutting-edge o1 in some of those third-party tests.
(02:27) Today we released Humanity's Last Exam, which is a new evaluation or benchmark of AI models that we produced by getting math, physics, biology, chemistry professors to provide the hardest questions they could possibly imagine. Deepseek, which is the leading Chinese AI lab, their model is actually the top performing, or roughly on par with the best American models.
(02:52) They accomplished all that despite the strict semiconductor restrictions that the U.S . government has imposed on China, which has essentially shackled the amount of computing power. Washington has drawn a hard line against China in the AI race. Cutting the country off from receiving America's most powerful chips like...
(03:08) Nvidia's H-100 GPUs. Those were once thought to be essential to building a competitive AI model. With startups and big tech firms alike scrambling to get their hands on any available. But Deepseek turned that on its head. Side-stepping the rules by using Nvidia's less performant H-800s to build the latest model and showing that the chip export controls were not the chokehold D.C. intended.
(03:33) They were able to take whatever hardware they were trained on, but use it way more efficiently. But just who's behind Deep seek anyway? Despite its breakthrough, very, very little is known about its lab and its founder, Liang Wenfeng. According to Chinese media reports, Deepseek was born out of a Chinese hedge fund called High Flyer Quant.
(03:54) That manages about $8 billion in assets. The mission, on its developer site, it reads simply: "unravel the mystery of AGI with curiosity. Answer the essential question with long-termism." The leading American AI startups, meanwhile – OpenAI and Anthropic – they have detailed charters and constitutions that lay out their principles and their founding missions, like these sections on AI safety and responsibility.
(04:20) Despite several attempts to reach someone at Deepeseek, we never got a response. How did they actually assemble this talent? How did they assemble all the hardware? How did they assemble the data to do all this? We don't know, and it's never been publicized, and hopefully we can learn that. But the mystery brings into sharp relief just how urgent and complex the AI face off against China has become.
(04:42) Because it's not just Deepseek. Other, more well -known Chinese AI models have carved out positions in the race with limited resources as well. Kai Fu Lee, he's one of the leading AI researchers in China, formerly leading Google's operations there. Now, his startup, "Zero One Dot AI," it's attracting attention, becoming a unicorn just eight months after founding and bringing in almost $14 million in revenue in 2024.
(05:07) The thing that shocks my friends in the Silicon Valley is not just our performance, but that we trained the model with only $3 million, and GPT-4 was trained by $80 to $100 million. Trained with just three million dollars. Alibaba's Qwen, meanwhile, cut costs by as much as 85% on its large language models in a bid to attract more developers and signaling that the race is on.
(05:38) China's breakthrough undermines the lead that our AI labs were once thought to have. In early 2024, former Google CEO Eric Schmidt. He predicted China was 2 to 3 years behind the U.S . in AI. But now , Schmidt is singing a different tune. Here he is on ABC's "This Week." I used to think we were a couple of years ahead of China, but China has caught up in the last six months in a way that is remarkable.
(06:03) The fact of the matter is that a couple of the Chinese programs, one, for example, is called Deep seek, looks like they've caught up. It raises major questions about just how wide open AI's moat really is. Back when OpenAI released ChatGPT to the world in November of 2022, it was unprecedented and uncontested. Now, the company faces not only the international competition from Chinese models, but fierce domestic competition from Google's Gemini, Anthropic's Claude, and Meta's open source Llama Model. And now the game has
(06:36) changed. The widespread availability of powerful open-source models allows developers to skip the demanding, capital-intensive steps of building and training models themselves. Now they can build on top of existing models, making it significantly easier to jump to the frontier, that is the front of the race, with a smaller budget and a smaller team.
(06:58) In the last two weeks, AI research teams have really opened their eyes and have become way more ambitious on what's possible with a lot less capital. So previously, to get to the frontier, you would have to think about hundreds of millions of dollars of investment and perhaps a billion dollars of investment.
(07:18) What Deepseek has now done here in Silicon Valley is it's opened our eyes to what you can actually accomplish with 10, 15, 20, or 30 million dollars. It also means any company, like OpenAI, that claims the frontier today ...could lose it tomorrow. That's how Deepseek was able to catch up so quickly. It started building on the existing frontier of AI, its approach focusing on iterating on existing technology rather than reinventing the wheel.
(07:45) They can take a really good big model and use a process called distillation. And what distillation is, basically you use a very large model to help your small model get smart at the thing that you want it to get smart at. And that's actually a very cost efficient. It closed the gap by using available data sets, applying innovative tweaks, and leveraging existing models.
(08:09) So much so, that Deepseek's model has run into an identity crisis. It's convinced that its ChatGPT, when you ask it directly, "what model are you?" Deepseek responds... I'm an AI language model created by OpenAI, specifically based on the GPT -4 architecture. Leading OpenAI CEO Sam Altman to post in a thinly veiled shot at Deepseek just days after the model was released.
(08:33) "It's relatively easy to copy something that you know works. It's extremely hard to do something new, risky, and difficult when you don't know if it will work." But that's not exactly what Deepseek did. It emulated GPT by leveraging OpenAI's existing outputs and architecture principles, while quietly introducing its own enhancements, really blurring the line between itself and ChatGPT.
(08:55) It all puts pressure on a closed source leader like OpenAI to justify its costlier model as more potentially nimbler competitors emerge. Everybody copies everybody in this field. You can say Google did the transformer first. It's not OpenAI and OpenAI just copied it. Google built the first large language models.
(09:14) They didn't productise it, but OpenAI did it into a productized way. So you can say all this in many ways. It doesn't matter. So if everyone is copying one another, it raises the question, is massive spend on individual L-L-Ms even a good investment anymore? Now, no one has as much at stake as OpenAI. The startup raised over $6 billion in its last funding round alone.
(09:40) But, the company has yet to turn a profit. And with its core business centered on building the models - it's much more exposed than companies like Google and Amazon, who have cloud and ad businesses bankrolling their spend. For OpenAI, reasoning will be key. A model that thinks before it generates a response, going beyond pattern recognition to analyze, draw logical conclusions, and solve really complex problems.
(10:04) For now, the startup's o1 reasoning model is still cutting edge. But for how long? Researchers at Berkeley showed that they could build a reasoning model for $450 just last week. So you can actually create these models that do thinking for much, much less. You don't need those huge amounts of to pre-train the models.
(10:23) So I think the game is shifting. It means that staying on top may require as much creativity as capital. Deepseek's breakthrough also comes at a very tricky time for the AI darling. Just as OpenAI is moving to a for-profit model and facing unprecedented brain drain. Can it raise more money at ever higher valuations if the game is changing? As Chamath Palihapitiya puts it...
(10:46) let me say the quiet part out loud: AI model building is a money trap. Those trip restrictions from the U.S . government, they were intended to slow down the race. To keep American tech on American ground, to stay ahead in the race. What we want to do is we want to keep it in this country. China is a competitor and others are competitors.
(11:13) So instead, the restrictions might have been just what China needed. Necessity is the mother of invention. B ecause they had to go figure out workarounds, they actually ended up building something a lot more efficient. It's really remarkable the amount of progress they've made with as little capital as it's taken them to make that progress.
(11:34) It drove them to get creative. With huge implications. Deepseek is an open-source model, meaning that developers have full access and they can customize its weights or fine -tune it to their liking. It's known that once open -source is caught up or improved over closed source software, all developers migrate to that.
(11:54) But key is that it's also inexpensive. The lower the cost, the more attractive it is for developers to adopt. The bottom line is our inference cost is 10 cents per million tokens, and that's 1/30th of what the typical comparable model charges. Where's it going? It's well, the 10 cents would lead to building apps for much lower costs.
(12:16) So if you wanted to build a u.com or Perplexity or some other app, you can either pay OpenAI $4.40 per million tokens, or if you have our model, it costs you just 10 cents. It could mean that the prevailing model in global AI may be open -source, as organizations and nations come around to the idea that collaboration and decentralization, those things can drive innovation faster and more efficiently than proprietary, closed ecosystems.
(12:44) A cheaper, more efficient, widely adopted open -source model from China that could lead to a major shift in dynamics. That's more dangerous, because then they get to own the mindshare, the ecosystem. In other words, the adoption of a Chinese open-source model at scale that could undermine U.S . leadership while embedding China more deeply into the fabric of global tech infrastructure.
(13:10) There's always a point where open source can stop being open -source, too, right? So, the licenses are very favorable today, but-it could close it. Exactly, over time, they can always change the license. So, it's important that we actually have people here in America building, and that's why Meta is so important.
(13:28) Another consequence of China's AI breakthrough is giving its Communist Party control of the narrative. AI models built in China t hey're forced to adhere to a certain set of rules set by the state. They must embody "core socialist values." Studies have shown that models created by Tencent and Alibaba, they will censor historical events like Tiananmen Square, deny human rights abuse, and filter criticism of Chinese political leaders.
(13:54) That contest is about whether we're going to have democratic AI informed by democratic values, built to serve democratic purposes, or we're going to end up with with autocratic AI. If developers really begin to adopt these models en masse because they're more efficient, that could have a serious ripple effect. Trickle down to even consumer-facing AI applications, and influence how trustworthy those AI-generated responses from chatbots really are.
(14:19) And there's really only two countries right now in the world that can build this at scale, you know, and that is the U.S . and China, and so, you know, the consequences of the stakes in and around this are just enormous. Enormous stakes, enormous consequences, and hanging in the balance: A merica's lead. For a topic so complex and new, we turn to an expert who's actually building in the space, and model-agnostic.
(14:50) Perplexity co-founder and CEO Arvind Srinivas – who you heard from throughout our piece. He sat down with me for more than 30 minutes to discuss Deepseek and its implications, as well as Perplexity's roadmap. We think it's worth listening to that whole conversation, so here it is. So first I want to know what the stakes are.
(15:08) What, like describe the AI race between China and the U.S . and what's at stake. Okay, so first of all, China has a lot of disadvantages in competing with the U.S. Number one is, the fact that they don't get access to all the hardware that we have access to here. So they're kind of working with lower end GPUs than us.
(15:32) I t's almost like working with the previous generation GPUs, scrappily. S o and the fact that the bigger models tend to be more smarter, naturally puts them at a disadvantage. But the flip side of this is that necessity is the mother of invention, because they had to go figure out workarounds. They actually ended up building something a lot more efficient.
(15:58) It's like saying, "hey look, you guys really got to get a top notch model, and I'm not going to give you resources and figure out something," right? Unless it's impossible, unless it's mathematically possible to prove that it's impossible to do so, you can always try to like come up with something more efficient.
(16:17) But that is likely to make them come up with a more efficient solution than America. And of course, they have open -sourced it, so we can still adopt something like that here. But that kind of talent they're building to do that will become an edge for them over time right? T he leading open-source model in America's Meta's Llama family.
(16:40) It's really good. It's kind of like a model that you can run on your computer. B ut even though it got pretty close to GBT-4, and at the time of its release, the model that was closest in quality was the giant 405B, not the 70B that you could run on your computer. And so there was still a not a small, cheap, fast, efficient, open-source model that rivaled the most powerful closed models from OpenAI, Anthropic.
(17:10) Nothing from America, nothing from Mistral AI either. And then these guys come out, with like a crazy model that's like 10x cheaper and API pricing than GPT -4 and 15x cheaper than Sonnet, I believe. Really fast, 16 tokens per second–60 tokens per second, and pretty much equal or better in some benchmarks and worse in some others.
(17:31) But like roughly in that ballpark of 4-O's quality. And they did it all with like approximately just 20, 48, 800 GPUs, which is actually equivalent to like somewhere around 1,500 or 1,000 to 1,500 H100 GPUs. That's like 20 to 30x lower than the amount of GPUs that GPT -4s is usually trained on, and roughly $5 million in total compute budget.
(17:59) They did it with so little money and such an amazing model, gave it away for free, wrote a technical paper, and definitely it makes us all question like, "okay, like if we have the equivalent of Doge for like model training, this is an example of that, right?" Right. Yeah. Efficiency, is what you're getting at. So, fraction of the price, fraction of the time.
(18:22) Yeah. Dumb down GPUs essentially. What was your surprise when you understood what they had done. So my surprise was that when I actually went through the technical paper, the amount of clever solutions they came up with, first of all, they train a mixture of experts model. It's not that easy to train, there's a lot of like, the main reason people find it difficult to catch up with OpenAI, especially on the MoE architecture, is that there's a lot of, irregular loss spikes.
(18:54) The numerics are not stable, so often, like, you've got to restart the training checkpoint again, and a lot of infrastructure needs to be built for that. And they came up with very clever solutions to balance that without adding additional hacks. T hey also figured out floating point-8 bit training, at least for some of the numerics.
(19:17) And they cleverly figured out which has to be in higher precision, which has to be in lower precision. T o my knowledge, I think floating point-8 training is not that well understood. Most of the training in America is still running in FP16. Maybe OpenAI and some of the people are trying to explore that, but it's pretty difficult to get it right.
(19:35) So because necessity is the mother of invention, because they don't have that much memory, that many GPUs. T hey figured out a lot of numerical stability stuff that makes their training work. And they claimed in the paper that for majority of the training was stable. Which means what? They can always rerun those training runs again and on more data or better data.
(20:00) And then, it only trained for 60 days. So that's pretty amazing. Safe to say you were surprised. So I was definitely surprised. Usually the wisdom or, like I wouldn't say, wisdom, the myth, is that Chinese are just good at copying. So if we start stop writing research papers in America, if we stop describing the details of our infrastructure or architecture, and stop open sourcing, they're not going to be able to catch up.
(20:29) But the reality is, some of the details in Deep seek v3 are so good that I wouldn't be surprised if Meta took a look at it and incorporated some of that –tried to copy them . Right. I wouldn't necessarily say copy. It's all like, you know, sharing science, engineering, but the point is like, it's changing. Like, it's not like China is just copycat.
(20:52) They're also innovating. We don't know exactly the data that it was trained on right? Even though it's open -source, we know some of the ways and things that was trained up, but not everything. And there's this idea that it was trained on public ChatGPT outputs, which would mean it just was copied. But you're saying it goes beyond that? There's real innovation in there? Yeah, look, I mean, they've trained it on 14.
(21:13) 8 trillion tokens. T he internet has so much ChatGPT. If you actually go to any LinkedIn post or X post. Now, most of the comments are written by AI. You can just see it, like people are just trying to write. In fact, even with an X, there's like a Grok tweet enhancer, or in LinkedIn there's an AI enhancer, or in Google Docs and Word.
(21:37) There are AI tools to like rewrite your stuff. So if you do something there and copy paste somewhere on the internet, it's naturally going to have some elements of a ChatGPT like training, right? And there's a lot of people who don't even bother to strip away that I'm a language model, right? –part.
(21:57) So, they just paste it somewhere and it's very difficult to control for this. I think xAI has spoken about this too, so I wouldn't like disregard their technical accomplishment just because like for some prompts like who are you, or like which model are you at response like that? It doesn't even matter in my opinion. For a long time we thought, I don't know if you agreed with us, China was behind in AI, what does this do to that race? Can we say that China is catching up or has it caught up? I mean, like if we say the matter is catching up to
(22:27) OpenAI and Anthropic, if you make that claim, then the same claim can be made for China catching up to America. A lot of papers from China that have tried to replicate o1, in fact, I saw more papers from China after o1 announcement that tried to replicate it than from America. Like, and the amount of compute Deepseek has access to is roughly similar to what PhD students in the U.S .
(22:54) have access to. By the way, this is not meant to criticize others like even for ourselves, like, you know, I for Perplexity, we decided not to train models because we thought it's like a very expensive thing. A nd we thought like, there's no way to catch up with the rest. But will you incorporate Deepseek into Perplexity? Oh, we already are beginning to use it.
(23:15) I think they have an API, and we're also they have open source weights, so we can host it ourselves, too. And it's good to, like, try to start using that because it's actually, allows us to do a lot of the things at lower cost. But what I'm kind of thinking is beyond that, which is like, okay, if these guys actually could train such a great model with, good team like, and there's no excuse anymore for companies in the U.S.
(23:41) , including ourselves, to like, not try to do something like that. You hear a lot in public from a lot of, you know, thought leaders in generative AI, both on the research side, on the entrepreneurial side, like Elon Musk and others say that China can't catch up. Like it's the stakes are too big. The geopolitical stakes, whoever dominates AI is going to kind of dominate the economy, dominate the world.
(24:03) You know, it's been talked about in those massive terms. Are you worried about what China proved it was able to do? Firstly, I don't know if Elon ever said China can't catch up. I'm not – just the threat of China. He's only identified the threat of letting China, and you know, Sam Altman has said similar things, we can't let China win the race.
(24:21) You know, it's all I think you got to decouple what someone like Sam says to like what is in his self-interest. Right? Look, I think the my point is, like, whatever you did to not let them catch up didn't even matter. They ended up catching up anyway. Necessity is the mother of invention like you said.
(24:47) And you it's actually, you know what's more dangerous than trying to do all the things to not let them catch up and, you know, all this stuff is what's more dangerous is they have the best open-source model. And all the American developers are building on that. Right. That's more dangerous because then they get to own the mindshare, the ecosystem.
(25:08) If the entire American AI ecosystem look, in general, it's known that once open-source is caught up or improved over closed source software, all developers migrate to that. It's historically known, right? When Llama was being built and becoming more widely used, there was this question should we trust Zuckerberg? But now the question is should we trust China? That's a very–You should trust open-source, that's the like it's not about who, is it Zuckerberg, or is it.
(25:36) Does it matter then if it's Chinese, if it's open-source? Look, it doesn't matter in the sense that you still have full control. Y ou run it as your own, like set of weights on your own computer, you are in charge of the model. But, it's not a great look for our own, like, talent to rely on software built by others.
(26:00) E ven if it's open-source, there's always, like, a point where open-source can stop being open-source, too, right? So the licenses are very favorable today, but if – you can close it – exactly, over time, they can always change the license. So, it's important that we actually have people here in America building, and that's why Meta is so important.
(26:24) Like I look I still think Meta will build a better model than Deep seek v3 and open-source it, and they'll call it Llama 4 or 3 point something, doesn't matter, but I think what is more key is that we don't try to focus all our energy on banning them, stopping them, and just try to outcompete and win them. That's just that's just the American way of doing things just be better.
(26:47) And it feels like there's, you know, we hear a lot more about these Chinese companies who are developing in a similar way, a lot more efficiently, a lot more cost effectively right? –Yeah, again, like, look, it's hard to fake scarcity, right? If you raise $10 billion and you decide to spend 80% of it on a compute cluster, it's hard for you to come up with the exact same solution that someone with $5 million would do.
(27:11) And there's no point, no need to, like, sort of berate those who are putting more money. They're trying to do it as fast as they can. When we say open -source, there's so many different versions. Some people criticize Meta for not publishing everything, and even Deepseek itself isn't totally transparent. Yeah, you can go to the limits of open-source and say, I should exactly be able to replicate your training run.
(27:33) But first of all, how many people even have the resources to do that. And I think the amount of detail they've shared in the technical report, actually Meta did that too, by the way, Meta's Llama 3.3 technical report is incredibly detailed, and very great for science. So the amount of details they get these people are sharing is already a lot more than what the other companies are doing right now.
(27:58) When you think about how much it costs Deepseek to do this, less than $6 million, I think about what OpenAI has spent to develop GPT models. What does that mean for the closed source model, ecosystem trajectory, momentum? What does it mean for OpenAI? I mean, it's very clear that we'll have an open-source version 4-O, or even better than that, and much cheaper than that open-source, like completely this year.
(28:25) Made by OpenAI? Probably not. Most likely not. And I don't think they care if it's not made by them. I think they've already moved to a new paradigm called the o1 family of models. I looked at I can't like Ilya Sutskever came and said, pre-training is a wall, right? So, I mean, he didn't exactly use the word, but he clearly said–yeah–the age of pre-training is over.
(28:52) –many people have said that . Right? So, that doesn't mean scaling has hit a wall. I think we're scaling on different dimensions now. The amount of time model spends thinking at test time. Reinforcement learning, like trying to, like, make the model, okay, if it doesn't know what to do for a new prompt, it'll go and reason and collect data and interact with the world, use a bunch of tools.
(29:14) I think that's where things are headed, and I feel like OpenAI is more focused on that right now. Yeah. –I nstead of just the bigger, better model? Correct. –Reasoning capacities. But didn't you say that deep seek is likely to turn their attention to reasoning? 100%, I think they will. A nd that's why I'm pretty excited about what they'll produce next.
(29:34) I guess that's then my question is sort of what's OpenAI's moat now? Well, I still think that, no one else has produced a system similar to the o1 yet, exactly. I know that there's debates about whether o1 is actually worth it. Y ou know, on maybe a few prompts, it's really better. But like most of the times, it's not producing any differentiated output from Sonnet.
(30:00) But, at least the results they showed in o3 where, they had like, competitive coding performance and almost like an AI software engineer level. Isn't it just a matter of time, though, before the internet is filled with reasoning data that. –yeah– Deepseek. Again, it's possible. Nobody knows yet. Yeah.
(30:24) So until it's done, it's still uncertain right? Right. So maybe that uncertainty is their moat. T hat, like, no one else has the same, reasoning capability yet, but will by end of this year, will there be multiple players even in the reasoning arena? I absolutely think so. So are we seeing the commoditization of large language models? I think we will see a similar trajectory, just like how in pre-training and post-training that that sort of system for getting commoditized this year will be a lot more commoditization there.
(30:59) I think the reasoning kind of models will go through a similar trajectory where in the beginning, 1 or 2 players really know how to do it, but over time –That's. and who knows right? Because OpenAI could make another advancement to focus on. But right now reasoning is their mode. By the way, if advancements keep happening again and again and again, like, I think the meaning of the word advancement also loses some of its value, right? Totally.
(31:26) Even now it's very difficult, right. Because there's pre-training advancements. Yeah. And then we've moved into a different phase. Yeah, so what is guaranteed to happen is whatever models exist today, that level of reasoning, that level of multimodal capability in like 5 or 10x cheaper models, open source, all that's going to happen.
(31:45) It's just a matter of time. What is unclear is if something like a model that reasons at test time will be extremely cheap enough that we can just run it on our phones. I think that's not clear to me yet. It feels like so much of the landscape has changed with what Deepseek was able to prove. Could you call it China's ChatGPT moment? Possible, I mean, I think it certainly probably gave them a lot of confidence that, like, you know, we're not really behind no matter what you do to restrict our compute.
(32:20) Like, we can always figure out some workarounds. And, yeah, I'm sure the team feels pumped about the results. How does this change, like the investment landscape, the hyperscalers that are spending tens of billions of dollars a year on CapEx have just ramped it up huge. And OpenAI and Anthropic that are raising billions of dollars for GPUs, essentially.
(32:40) But what Deepseek told us is you don't need, you don't necessarily need that. Yeah. I mean, look, I think it's very clear that they're going to go even harder on reasoning because they understand that, like, whatever they were building in the previous two years is getting extremely cheap, that it doesn't make sense to go justify raising that– Is the spending.
(33:02) proposition the same? Do they need the same amount of, you know, high end GPUs, or can you reason using the lower end ones that Deepseek– Again, it's hard to say no until proven it's not. But I guess, like in the spirit of moving fast, you would want to use the high end chips, and you would want to, like, move faster than your competitors.
(33:26) I think, like the best talent still wants to work in the team that made it happen first. You know, there's always some glory to like, who did this, actually? Like, who's the real pioneer? Versus who's the fast follow right? That was like kind of like Sam Altman's tweet kind of veiled response to what Deepseek has been able to, he kind of implied that they just copied, and anyone can copy.
(33:47) Right? Yeah, but then you can always say that, like, everybody copies everybody in this field. You can say Google did the transformer first. It's not OpenAI and OpenAI just copied it. Google built the first large language models. They didn't productise it, but OpenAI did it in a productized way. So you can say all this in many ways, it doesn't matter.
(34:10) I remember asking you being like, you know, why don't you want to build the model? Yeah, that's that's, you know, the glory. And a year later, just one year later, you look very, very smart. To not engage in that extremely expensive race that has become so competitive. And you kind of have this lead now in what everyone wants to see now, which is like real world applications, killer applications of generative AI.
(34:36) Talk a little bit about like that decision and how that's sort of guided you where you see Perplexity going from here. Look, one year ago, I don't even think we had something like, this is what, like 2024 beginning, right? I feel like we didn't even have something like Sonnet 3.5, right? W e had GPT -4, I believe, and it was kind of nobody else was able to catch up to it. Yeah.
(35:03) B ut there was no multimodal nothing, and my sense was like, okay, if people with way more resources and way more talent cannot catch up, it's very difficult to play that game. So let's play a different game. Anyway, people want to use these models. And there's one use case of asking questions and getting accurate answers with sources, with real time information, accurate information.
(35:29) There's still a lot of work there to do outside the model, and making sure the product works reliably, keep scaling it up to usage. Keep building custom UIs, there's just a lot of work to do, and we would focus on that, and we would benefit from all the tailwinds of models getting better and better.
(35:47) That's essentially what happened, in fact, I would say, Sonnet 3.5 made our product so good, in the sense that if you use Sonnet 3.5 as the model choice within Perplexity, it's very difficult to find a hallucination. I'm not saying it's impossible, but it dramatically reduced the rate of hallucinations, which meant, the problem of question-answering, asking a question, getting an answer, doing fact checks, research, going and asking anything out there because almost all the information is on the web,was such a big unlock.
(36:22) And that helped us grow 10x over the course of the year in terms of usage. And you've made huge strides in terms of users, and you know, we hear on CNBC a lot, like big investors who are huge fans. Yeah. Jensen Huang himself right? He mentioned it the other, in his keynote. Yeah. The other night. He's a pretty regular user, actually, he's not just saying it.
(36:40) He's actually a pretty regular user. So, a year ago we weren't even talking about monetization because you guys were just so new and you wanted to, you know, get yourselves out there and build some scale, but now you are looking at things like that, increasingly an ad model, right? Yeah, we're experimenting with it. I know there's some controversy on like, why should we do ads? Whether you can have a truthful answer engine despite having ads.
(37:06) And in my opinion, we've been pretty proactively thoughtful about it where we said, okay, as long as the answer is always accurate, unbiased and not corrupted by someone's advertising budget, only you get to see some sponsored questions, and even the answers to those sponsored questions are not influenced by them, and questions are also not picked in a way where it's manipulative.
(37:34) Sure, there are some things that the advertiser also wants, which is they want you to know about their brand, and they want you to know the best parts of their brand, just like how you go, and if you're introducing yourself to someone you want to, you want them to see the best parts of you, right? So that's all there.
(37:49) But you still don't have to click on a sponsored question. You can ignore it. And we're only charging them CPM right now. So we're not we ourselves are not even incentivized to make you click yet. So I think considering all this, we're actually trying to get it right long term. Instead of going the Google way of forcing you to click on links.
(38:09) I remember when people were talking about the commoditization of models a year ago and you thought, oh, it was controversial, but now it's not controversial. It's kind of like that's happening and you're keeping your eye on that is smart. By the way, we benefit a lot from model commoditization, except we also need to figure out something to offer to the paid users, like a more sophisticated research agent that can do like multi-step reasoning, go and like do like 15 minutes worth of searching and give you like an analysis, an analyst type of
(38:36) answer. All that's going to come, all that's going to stay in the product. Nothing changes there. But there's a ton of questions every free user asks day-to-day basis that that needs to be quick, fast answers, like it shouldn't be slow, and all that will be free, whether you like it or not, it has to be free.
(38:53) That's what people are used to. And that means like figuring out a way to make that free traffic also monetizable. So you're not trying to change user habits. But it's interesting because you are kind of trying to teach new habits to advertisers. They can't have everything that they have in a Google ten blue links search.
(39:10) What's the response been from them so far? Are they willing to accept some of the trade offs? Yeah, I mean that's why they are trying stuff like Intuit is working with us. And then there's many other brands. Dell, like all these people are working with us to test, right? They're also excited about, look, everyone knows that, like, whether you like it or not, 5 or 10 years from now, most people are going to be asking AIs most of the things, and not on the traditional search engine, everybody understands that.
(39:40) So everybody wants to be early adopters of the new platforms, new UX, and learn from it, and build things together. Not like they're not viewing it as like, okay, you guys go figure out everything else and then we'll come later. I'm smiling because it goes back perfectly to the point you made when you first sat down today, which is necessity is the mother of all invention, right? And that's what advertisers are essentially looking at.
(40:05) They're saying this field is changing. We have to learn to adapt with it. Okay, Arvind, I took up so much of your time. Thank you so much for taking the time.