Everyone's talking about deep-seak, deep-seak. Satya Nadella at Microsoft. I think deep-seak has, you know, had some real innovation. Mark Zuckerberg at Meta. You know, I think that there's a, there's a number of novel things that, that they did that I think were still digesting and
There are a number of things that they have advances that we will hope to implement in our systems. The president of America? The release of DeepSeek AI from a Chinese company should be a wake-up call for our industries that we need to be laser focused on competing to win. It's a chatbot. It's a white paper. It could help break code for a Tetris game. Tetris.
It could solve a meth theorem. It's chips, you guys. And today explained the week the world wilded out over deep-seek. And also, what is deep-seek coming up?
On Explain It To Me, we treat every single question you ask us with the utmost professionalism. What was your initial reaction when you read that question? Honestly, like, my gut, initial reaction was like, aw, honey. Like, yeah, I'm kind of like, okay, all right, I'm good, I said that.
There are no bad questions. But there are some that are really hard to answer. This week on Explain It To Me, Seagal Samuel tells us why those are the ones she gravitates towards. New episodes every Wednesday, wherever you get your podcasts.
Hey, it's Andy Roddick, and I'm not just a former tennis player. I am a tennis fan, a tennis nerd. I just can't stop watching it. I can't stop analyzing it. I can't stop talking about it to anyone that will listen, which is why I started my podcast served with Andy Roddick, now a part of the Vox Media Podcast Network. On the show, we talk about everything from new up-and-coming players to the champions dominating the narrative to whatever's on my mind.
This January is the Australian Open, and you know I've got some thoughts. So tune in for our Australian Open coverage by and serve wherever you get your podcast or on our YouTube channel.
You are listening to today explained. Eleanor Olcott is the Financial Times China technology correspondent. We reached her in Beijing where she has been following DeepSeek from its very beginnings. I first heard about this mysterious AI company in early 2023 because one of my contacts said that this hedge fund silently built one of the largest clusters of NVIDIA GPUs in China.
So Nvidia GPUs are graphic processing units. These are basically the AI chips that you need to power AI model training and inference. They're really, really important for this whole AI race and they're in short supply in China. So somehow
this quant fund that's a hedge fund in China had kind of silently built one of the biggest clusters in the country. We took notice and they started publishing more and more advanced models over the past year and their work finally pierced through the Western consciousness when we were all on Christmas holiday right at the end of 2024 with this new model released V3
There is a new model that has all the valley buzzing and it does not come from OpenAI or Meta or Google or any of those names. DeepSeek V3 is the first open source model in all of AI history that is better than the closed source models. DeepSeek version 3 is free and it's absolutely insane.
Later in January, it then published another model, which again shocked the world with its sophistication. And the key thing here is that the reason that they've prompted somewhat of an existential crisis, especially amongst the US players, is that they're claimed to have been doing this on such a bootstrap budget.
All right, when I learned about DeepSeek, it was because the stock market had absolutely collapsed amid the news that this Chinese company had made this thing. What exactly went on earlier this week? I mean, the stock market is an incredibly mysterious beast, right? I mean, we at the FT have been writing about how DeepSeek and other Chinese companies are building really competitive models.
for months now, but I think what happened over the past week was we saw all of this frenzied activity on Twitter.
It's not that people want deep-seek to win. It's that they want open AI to lose. Deep-seek R1 is one of the most amazing and impressive breakthroughs I've ever seen. Deep-seek this. Deep-seek that. A profound gift to the world. How about you seek a deep connection with a woman? Personally, I'm staying away from deep-seek. I don't want the Chinese spying on me and seeing what kind of videos I'm watching on TikTok. Wait. Wait. Wait.
what came out on Monday was a moment, right? It was really, really important because deep-seek, this kind of little-known Chinese lab, for the first time released a paper with a very, very detailed explanation, a kind of technical recipe, as it were, for building a reasoning model. Now, reasoning models are important. It's a fairly new area of AI.
But it basically means models that can teach themselves and improve themselves without human supervision. And this is really important because if we can kind of use this in practical applications, it means that AI will be capable of critical thinking and will be useful in tasks that are vastly more complex than what we currently have on the market.
The dream right is to have an AI, for example, running in the background of your computer and kind of preempting your needs, like booking travel, doing things that you haven't even thought of. Maybe it's kind of acting as your actual personal assistant. They don't just respond to demands. They preempt things. Hello, Noel. What can I get for you today?
They make decisions on their own. They might, for example, figure out that you have not got enough groceries in your fridge and think, OK, well, we'll preemptively order that so you don't even have to do it yourself, right? I ordered extra cheetahs this week. You deserve them. It's still very much an open question as to whether or not we're going to get there. It's important to know as well that this is just like
a big marketing strategy on the part of a lot of AI companies also to justify continuing to raise billions of dollars. But what I think Deepseak proved over the past week is actually China is a viable and competitive player in this field. So let's talk about where Deepseak comes from. Who's behind this?
So unlike other AI companies, AI startups in China, it hasn't raised any external financing. So you think, OK, how the hell has a company managed to build what we know is a very expensive endeavor of buying all of these GPUs and also hiring the best talent? They're known a long white dance for paying the top dollar for the best AI researchers.
in China. And that's basically a story about the founder Liang Wenfeng who has a background as a quant hedge fund manager. So he basically made a whole bunch of money trading stocks and decided to plow some of those resources into this new pet project. And he started in 2021 building this large Nvidia cluster because he recognized the potential for this technology
And the timing of that is important for two reasons. The first is that it was really before the world woke up to the potential of generative AI. It was before the release of chat TBT. And we, you know, the rest of the Chinese players had kind of neglected generative AI as a field research. They were much more focused on surveillance technology, surveillance AI, because it was clear.
you could make money with that form of AI. The other reason it's significant is because it was really before the first tranche of kind of blanket export controls would live in place on China. The restrictions will limit Chinese companies' access to advanced computer chips and slow their progress in artificial intelligence.
U.S. chip makers NVIDIA and AMD tumbling after the U.S. ramped up its chip export rules. Washington says the aim is to prevent Beijing using the most advanced semiconductors for its military modernization. So, when the race in China in early 2023 started to replicate or seeking to replicate open-eye success, actually Liang and Deepsea equivalent
pretty good position to get ahead. Okay, so Brilliant Man made a bunch of money and now presumably will make a trillion more dollars. Is that the objective here? That isn't the objective here and actually that's what makes Deepseek so unique, right? They have not made any serious moves to commercialize their technology.
They have an AI chatbot. It's free to use. What I think he's doing here and from people who know him is he wants to just add to the great canon of LLM research. He wants to push this technology forward. And actually also there is a bit of a national pride here element as well, right? In interviews with domestic press, he says it's important that China also plays a role
in developing this technology and being a leader. So I think there's various ambitions at play, but he's a pure technologist. And actually, because Deepseek is not interested in commercializing their technology, right, it's like a pure research lab. People have described it to me being like the early days of DeepMind.
where you just have a bunch of engineers, a bunch of researchers working on whatever they think is the best technical pathway forward. But because they don't care about commercialization, that means that they're willing to share the secrets of how they've done that with the rest of the world and kind of enable the others to also learn from their learnings. And for players like OpenAI who are also working on the same research, but not telling the world how they got there, this is really a bit of a challenge.
Earlier this week, as the stock market was whipsawing about, we heard people asking whether or not this is AI's Sputnik moment. So they're referring to the Soviet Union launching a satellite into space before the United States, back in the 50s, kicked off the space race, a very big deal. And one of those terms that you don't hear very often because a Sputnik moment is a big moment. Do you think that this development just kicked off the AI race?
in a way? As a journalist, I'm all for fancy metaphors and comparisons. I think the comparison is not completely correct in this case, right? Like DeepSeek is a private company that has just been plugging away on AI research. It's not building rockets to send to space. But having said that, you know, US and China undeniably are in a tech war. We've known this since 2019.
and China is very, very concerned about the US getting ahead on AI and it's been providing a huge amount of support to kind of select players that they think are going to help its remain competitive and gain an edge. But really, the Spudnik element is really about the hardware itself, the AI chips. I think the kind of real race here is on
Chinese companies and the Chinese ecosystem overall trying to make Huawei or maybe one of the other Chinese competitors a true long-term and successful rival to Nvidia. Eleanor Olcott of the Financial Times invasion. Coming up next, can deep-seeks competitors looking at you open AI compete?
Support for the show today comes from Vanta. Trust isn't just earned, Vanta says. It's demanded. Have you demanded someone's trust lately? Whether you're a startup founder, navigating your first audit or a seasoned security professional, scaling your governance, risk, and compliance program, proving your commitment to security is critical and complex. And that's where Vanta comes in. You know the deal. Vanta says they can help businesses establish trust
by automating compliance needs across 35 frameworks like SOC2 and ISO 27001. They say they can also centralize security workflows, complete questionnaires up to five times faster and proactively manage vendor risk. You can join over 9,000 global companies like Atlassian, like Quora and Factory who use Vanta to manage risk
and prove security in real time for a limited time. Our audience can get $1,000 off dance at vanta.com slash explain that's v-a-n-t-a dot com slash explained for $1,000 off.
This week on Prof G Markets, we speak with Robert Armstrong, US financial commentator for the Financial Times. We discuss Trump's comments on interest rates and who might emerge as the biggest winners from the deep-seep trade. In the world we lived in last Friday, having a great AI model behind your applications either involved building your own.
or going to ask open AI, can I run my application on top of your brilliantly good AI model? Now maybe this is great for Google, right? Maybe this is great for Microsoft, who were shoveling money on the assumption that they had to build it themselves at great expense. You can find that conversation and many others exclusively on the Prof G Markets podcast. You're listening to today, explain.
Noelle King here with Reed Elbergotti, technology editor at Semaphore Reed. The reaction on Monday to Deepseek was huge. The markets are swinging around. People are yelling about Sputnik. It's on everyone's homepage. What did you think about all that?
Well, I was really kind of slapping my forehead because I think it was a complete overreaction. People knew that this company existed. And in fact, this whole idea of distilling these larger models into smaller, more powerful ones that are more efficient, this is something that had been going on really since ChachiPT came out. The biggest takeaway for me is that the market really does not understand the AI industry yet.
What's been going on with all of Deepseek's Western competitors? What have they been up to?
They're all investing massively in these huge data centers. It's my honor to welcome three of the world's leading technology CEOs. With hundreds of thousands of graphics processors, tens of billions of dollars. In fact, you probably heard last week there was a deal announced for $500 billion. Stargate, so put that name down in your books. With OpenAI,
and Oracle and MGX and SoftBank. I mean, massive amounts of money. I think this will be the most important project of this era for AGI to get built here, to create hundreds of thousands of jobs, to create a new industry centered here. And what that investment is for is running these models because there is so much demand that these companies really can't meet it right now. And what we're also finding is that
inference, which is just the fancy term for running these models, actually can now increase capability of the models a lot. That wasn't the case before when chat GPT first came out. It was just you prompt chat GPT. It comes back with an answer. Now you prompt the most advanced model of these models, and they are doing a whole bunch of stuff in the background. They're running over and over and over again. They're trying to find the best answer, and that is
exponentially more expensive, and that is just going to continue. And this new R1 model that Deepsea came out with, it's an advance, but it is not nearly a big enough breakthrough to sort of negate those market dynamics.
Can you explain why? Because yeah, I saw that as well. DeepSeek did it on the cheap. All of this money and energy and investment is for not because they came up with the little AI that could and it didn't even cost that much. Yeah, I mean, they showed that you can do some of these types of queries at a lower cost, but it's just not nearly low enough. You might have seen Microsoft CEO Satinella talking about Jevon's paradox.
Javon's paradox strikes again. Basically, as the technology becomes more efficient and the cost declines, the paradox is that you would think, well, OK, that just means that it just gets cheaper and these companies are just not going to make as much money on it. But actually, what happens is it becomes more useful and people want to use it more.
And then there was another wrinkle that appeared a few hours before we're speaking on Wednesday. There is a suggestion that deep-seak may have borrowed from OpenAI or stolen from OpenAI. What is the allegation? What are people seeing and saying?
You know, it's stolen. I mean, that's a very strong word. We saw David Tacks, who's the incoming AI czar, sort of accused Deepseek of stealing from OpenAI. And there's substantial evidence that what Deepseek did here is they distilled the knowledge out of OpenAI's models. And I don't think OpenAI is very happy about this.
So you need data to train these AI models. But what you can actually do is you can use the models themselves to create a very, very specialized kind of data. It's really synthetic data because it's being generated by an AI model. But you can create
exactly the kind of data that you want, and you can check that over with other AI models. And what you end up with is that's how you make these models much more efficient. So this is also not surprising, because that's how all of these models work. I mean, we've seen lots of companies do this. So I just, again, the process of distillation makes total sense. Whether or not it's stealing, I think that's something that's a gray area in the AI industry that we really haven't
ironed out yet. I think it's such a new thing that we'll have to sort of come up with the norms and the rules and regulations, maybe even copyright law around this.
I want to ask you about NVIDIA, the company that makes the chips that AI requires. NVIDIA is now basically a household name. It takes up a big part of the stock market. And so when NVIDIA's stock is riding high, so is my 401k. And when it's tanking, good Lord, I'm going to retire under a bridge. And on Monday, the bridge was looking like a real possibility. What exactly happened on Monday with NVIDIA and why did it seem to get hit so hard by this news?
Nvidia makes these graphics processing units that are the most powerful, the most advanced in the world. And they're very expensive. I mean, the older models, the H100s, which were state of the art, it cost about $40,000 a piece. And these data centers have about $100,000 of those. So you do the math there. Nvidia is selling a ton of these chips. Really, they can't sell enough. There's way more demand than they can even produce.
and it's all because these models take a lot of energy to run. If you can have a more efficient one that doesn't require these powerful GPUs, then maybe you don't need to spend $40,000 on a GPU now with OpenAI. But again, that's not really what happened here. What happened here is there's a bit of an advance in how efficient these models can get, but in order to get the most use out of them, you need to run a lot of inference on those. You're still going to need
really powerful GPUs. And as there are more advances in the pre-training part of the models, they'll get even bigger and more powerful. So NVIDIA is not going anywhere. I mean, certainly they have competition. There are chip makers that want to build more efficient inference chips. There are people who want to get rid of NVIDIA's advantage, which is its CUDA software that the whole AI industry basically runs on right now and creates a big moat for them.
Those are the things that I think are the risks for NVIDIA, not a person, a company building a more efficient open source AI model.
I think there was a reason that so many watchers and analysts and reporters framed this as China catches up to the US, and that is because China and the United States are in a quiet war, cold war, existential struggle. We compete with the Chinese and it raises some questions, does it not, about how worried the US should be that China beats us in the artificial intelligence competition.
Yeah, and this is where I think there really are national security concerns about China and AI. And if China does win the AI race, let's say, it will probably give them a military advantage. I mean, this is all
This is far into the future. There's a lot of debate about this, right? But I mean, I think the conventional wisdom is if you win the AI race and you get your first to AGI or super intelligence or wherever you want to call it, it becomes a military tool very quickly. And I think the US, that's the whole reason the US has put so much energy into figuring out how to curb the exports of the most powerful AI chips to China. They don't want to see
trying to be able to sort of control its own destiny when it comes to AI? After the events of this week, Reed, is there a sense of renewed competitive energy? Like everybody needs to now go back and work harder, faster, smarter, for less money? Yeah, look, I think Sam Altman said that it was invigorating on X.
We will obviously deliver much better models and also it's legit invigorating to have a new competitor. This is how research works. You know, somebody comes out with a new idea and it inspires other people both creatively and also competitively.
You know, this is a dynamic that we've seen, you know, for the past couple of years in AI or even really more than that. This is why so many of these tech companies have been publishing their research instead of keeping it a trade secret for so long because the genius
researchers who write these papers, they want to present them. They want bragging rights at the NURIPS conference. That's the big AI model conference every year. They want to get passed in their back from their coworkers. And I think there's actually probably a lot of mutual respect between the AI researchers at
you know, opening eye and anthropic and the ones at Deepsea. If they think in that world, I think it's possible to kind of put aside all the geopolitics and just say, hey, nice job. You created a really interesting model and we're gonna learn from it and try to do better.
I think the other way to look at it is, look, if the US doesn't win the race to AGI, then what you could see is a Chinese military advantage that leads to something like an invasion of Taiwan and maybe potentially a hot war between these two
superpowers, and that would be very, very bad. I think people who, you know, the most fervent China hawks what they really want is a US military advantage that is so big that, you know, there just will be no war. And I think if you look at it through that lens, then yes, I mean, this AI race is very, very consequential.
geopolitically, and really, there are dire consequences if the right outcome isn't achieved. Semaphores Reid Albergotti, thanks to him. Miles Bryan and Victoria Chamberlain produced today's show with an assist from Amanda Luellen. Amana El-Sadi is our editor. Andrea Kristin's daughter and Rob Byers engineered. Laura Bullard checks the facts. I'm Noel King. It's Today Explained.