Already and this is this is the daily This is the Daily OS. Oh, now it makes sense. Good morning and welcome to the Daily OS. It's Wednesday, the twenty ninth of January. I'm Sam, I'm Zara. What if creating artificial intelligence programs was actually cheaper than we thought? How would AI companies, governments and investors react if something faster, more accurate, and cheaper than the current AI giants like
chatcha Et suddenly burst onto the scene. Well, today we're going to talk about Deep Seek, this new company that has really shaken up the stock market and the way that we're all feeling about artificial intelligence. It's pretty remarkable, stories areah.
I think at the beginning of this week, the number of people that could name, identify, and explain what Deep Seek is would have been few and far apart at myself included. I'd say potentially you included in that as well. And now it's all anyone's talking about. It is literally dominating every single headline across the world today. Sam, take us back to the beginning. AI is a very very complicated space. But what exactly is Deep Seek and why the hell are we hearing so much about it?
So DC is a Chinese AI company that just launched its latest AI model called R one. Now when I say model, just think of a brain.
I can do that.
All of these AI companies are trying to work out how to create the best AI brain. Some are better at solving mathematical problems, others are better at expressing, reasoning, or being emotional, trying to mimic human emotion. And of course, the goal for every AI company is to try and build a model or a brain that can do everything.
Okay, And so when you're talking about AI companies, an example of that would be open Ai, Creator, Chat ChiPT exactly.
But there are many companies and there are many companies that are in the AI space without making models, and we'll talk about that later as well. Now, what's fascinating about deep Seek is that they've built a model that they say performs as well as the best American AI systems,
but at a fraction of the cost. So to put that in context, just last week we saw major US companies like Microsoft, Oracle and open Ai, which you just mentioned, pledge five hundred billion US dollars to build new AI infrastructure. Then we also had an announcement from Meta they came out separately and said that they were committing sixty five billion US dollars to build infrastructure. There is so much
money being spent on this technology at the moment. Then deep Seat comes along and they say they built their system for about six million dollars, so million versus billion. One expert I was reading called it a joke of a budget.
So they're clearly a lot of differences between these companies. But all of the companies that you were just talking about before are American companies. Deep Seek is a Chinese.
Chinese AI company that say it can do what American companies can do for a fraction of the cost.
Okay, So then tell me more about deep Seek, like, how is it possible that they're doing this?
Are they telling the truth? Where have they come from?
It's a really interesting story. So it started as a side hustle project of a Chinese it's an absurd side hustle, it's crazy of a Chinese hedge fund called high Flog.
Usually like a jewelry brand.
So these are coders who had been employed to build systems that would trade stocks and be able to predict for clients when to buy and sell. And they started playing around on the side of that with different ways to build AI products that could be used by consumers, So by you and me. The founder then spun off Deep Seek as a separate company to this hedge fund only in twenty twenty three.
Oh wow, so very recent.
Yeah, and they started only recently playing around with old GPUs, which what is one of the key tools you need to build an AI machine that they had available to them.
What's what's GPUs?
So let's break it down. I'm going to use a car analogy. Let's pretend that Ferrari were spending billions of dollars a year on new car engines to try and figure out how to win the Formula One, and then all of a sudden, engineers at Holden, you know, a really un sexy car company, started playing around with a Ferrari part made fifteen years ago, and they approached the challenge of making the fast car from a different perspective with these old parts, and actually they've made a faster car.
That's basically what's happened here.
Okay, So what you're saying is that this Chinese company, Deep Seek has created this basically beast of Nai machine, are using old parts, old GPUs, which is just the little chips yeah sure, rather than building new ones.
And getting exactly the same results.
Okay, And is that why the cost is lower?
Exactly? And so they say though that it's not actually the new GPUs, the new chips that you need to make really good AI, it's their innovative training techniques. So they say they've found ways to make AI models think more efficiently. And there was one really key part of the research that they say makes their AI different, and that's recognizing that AI has a heart moments. This is
actually what they say. They say that AI has a heart moments where they start processing a trying to answer a question you put to them, and then they realize something that might not be on the exact path they were following. They've trained their AI to take that route, so they say, oh, I've actually just thought of this, this, and this, and then go down that route. So instead of it explicitly trying to get to the answer as fast and direct as possible, it almost kind of goes
with the flow a bit more. And I'm really simplifying this here.
This train is legitimately heading.
So what you're saying here is that they are suggesting that their technology is superior because they've trained it to think differently.
Essentially, I think that we're probably summarizing artificial intelligence. Yeah, and I think we've probably summarized like twenty five years of quantum physics and quantum computing into six sentences. But yes, essentially, they've figured out how to make old technology think differently and be as good as new technology. Okay, that's what
you need to know. Okay, Sure, And so the USAI companies get wind of this, and they start testing this Chinese AI yep, and they give deep Seek and open AI's top models, so the best chatch apt models that you can possibly get. They give the both models the same questions, and deep Seek produces similar, if not better answers. When was this this week? Last week?
A panic?
Exactly? Now there is an important caveat that I want to mention here. These us AI companies have said that whilst the deep Seek model is impressive at writing in general problem solving, it does struggle with certain specific tasks, so things like you know, booking and a play, Yeah, you're you do struggle with certain specific tasks numbers. So now there's going to be a whole independent testing phase that will give us some clearer results.
Okay.
The reason that the AI industry basically is freaking out is because this new players come along is doing something for the fraction of the cost, is getting similar results, and it's non an American company.
Yeah, okay, and it challenges this fundamental assumption that we have about AI development, which is this conventional thinking that you needed the most expensive, the most cutting edge computer chips, particularly those made by one company in the video, which is one of the most valuable companies in the world.
We've talked about it on this podcast before, to build the best AI systems, and that's why tech companies like Google and Meta have been spending insane amounts of money building the infrastructure that they say they need to produce these AI systems, and investors have gone with it.
Okay, But I guess the reason that the tech world is in so much disarray this week is because that type of logic might not necessarily be the only.
Way exactly, and Deep Seek says that might not be true. And what's more, Deep Seek have released their program as open source, and what that means is they're an open book. They've taken the hold and around the racetrack popped open the hood and said, whoever wants to come and have a look as much as.
You want, Why can't they just be copied?
They say that they don't care really if they're copied. That this technology almost think about it, like you know, there's been a lot of examples in the last twenty four hours about the space race. They've kind of said, everyone's trying to get to the moon. Here's how we're doing it, hoping that the entire industry becomes more profitable. It's just a different way of doing business. It's very common in the tech world to release something out to
the public. And I want to bring up one new term here, and that's the idea of a company having a moat, a mote e moche. So think about a competitive advantage that surrounds a company and literally think about a castle with emotion around it and moat. Yeah yeah, And that's what makes a company hard to copy. So if a company has I feel that far ahead, yeah, but if you have a wide moat, then other companies can't copy you quickly and easily. And so for these
AI companies, these really valuable stocks. Their moats are not only the knowledge they have and the expertise, but just how much cash they've got, because the common reasoning has always been to build great AI, you need fifty sixty seventy one hundred, five hundred billion.
I understand.
So you're saying these companies they're competitive advantages that they've raised in dollars of money, and therefore the average Joe on the street can't go and build the same tech as them because they can't raise a bajillion.
Dollars until now, until now when DeepC comes out and.
Says, actually, you don't need a bajillion dollars, you only a little bit, only six million, and you can build something just as good.
Okay.
So what has all of this kind of knowledge and information done to the market, because there has been quite a significant economic kind of market impact.
Yeah, it's been really dramatic. So, as I mentioned earlier in the video, the company that makes the computer chips that AI companies buy to build these big AI brains, they lost a whopping get ready, five hundred and eighty nine billion US dollars in the market value on the US stock market yesterday. They went down sixteen point one percent, and the overall US market went down three point one
percent crazy. You have to remember that it's not just companies that are directly related to AI, but there are so many adjacent companies that have business in AI. You know, for example, energy companies, they would be relying on the fact that everyone's building these big data centers. They're going to need to plug that all into a wall somewhere. So their stocks went down. The entire market really did freak out about the fact that there's this new way of doing things.
And it's not just the market that responds. It is also this kind of geopolitical angle. I've mentioned that deep Seek is Chinese and Open Ai and the other AI companies are American. Talk me through that geopolitical angle.
Yeah, So the US has been trying to maintain its competitive lead in AI. They say there's a two horse race between them and China, and the way that they're doing that is by restricting the computer chips that are allowed to enter China. So, to go back to our Ferrari example, they are not letting in new models of a Ferrari engine into China because they don't want Chinese
AI companies to rival them. But Deep Seek's success suggests that that restriction might actually be backfiring in two possible ways. Either the chips aren't as crucial as we all think to building these models, or Chinese companies have worked out ways to get around the restrictions.
Yeah. So interesting.
And when I hear about that, the one thing that I think about is how Donald Trump would be responding to this, because we all know that Trump's platform is about growing specifically American companies and wealth. How has he responded to this?
Well, this has become a core focus. I mean already in his first week, full week of being president, he announced a five hundred billion dollar program in collaboration with some really big companies AI company. Yeah, and then just yesterday, talking about Deep Seek, he said that this is a wake up call for the AI industry and that we need to be quote laser focused on competing to win.
Wow, it's really like Space Race.
It's very much the Space race. Trump's close advisor Elon Musk he speaking of Space Race. Yes, he accused Deep Seak of lying about how many chips it needed to run system. So he essentially said, what you're doing is simply impossible, but there is certainly a bit of panic there.
Trump said that he might use executive orders to increase the amount of public reas sources like electricity and energy supplies and building materials to just build these data centers that USAI companies say they need as quick as they can.
Sam, I want to finish this chat by just bringing it home. And I guess contextualizing this in an Australian setting, what are the implications for Australia's tech sector? What does this story mean for US?
Well, pretty much every country besides the US has had the same reaction, including Australia, which is that this is good news. We've all known that the US has a major competitive advantage in AI, and AI is seen as a really important part of any country's innovation strategy. So
our Industry and Science Minister Ed Husick. He said yesterday that he expects to see more of these cheaper and more creative AI solutions in years to come as this becomes more mainstream, and that it kind of evens the playing field a little bit because we don't need five hundred billion dollars. We just need a side hustle in our jobs.
How did an our side hustle lead to that?
I know, but I think, look, this is still a privately held company Deep Seek, so deep Seak isn't listed anywhere. We're going to see how the stock markets responded. The stock markets in Australia that tech and AI companies were hit pretty hard yesterday, but our stock market isn't as reliant on those big tech giants as in the US. But it's a really seismic shift in the way that we think about AI and I'm going to have to ask Chatcha and see what happens. Now.
Well, thank you for explaining that story. I actually feel like I completely understand it now.
So you.
Well, there you go.
Thank you so much for joining us for another episode of The Daily Ods and supporting our independent newsroom. We're going to be back later today with the headlines, but until then, have a great day.
My name is Lily Maddon and I'm a proud Arunda Bungelung Calcottin woman from Gadigol Country. The Daily oz acknowledges that this podcast is recorded on the lands of the Gadigol people and pays respect to all Aboriginal and Torres Strait Island and nations. We pay our respects to the first peoples of these countries, both past and present,