Speaker 1: Okay, so if you've been following the tech scene for any length of time, you'll know that the USA have put tech sanctions on China. This started back in Trump's first term when tech sanctions were put on Huawei. Then they were extended to other companies and then Biden continued and extended those tech sanctions when he took office. So first of all, they banned other companies, they created an entity list and they put many companies on those lists. And then they stopped supplying them with chip making equipment. They also persuaded Japan, South Korea and the Netherlands to stop supplying China with either chip making equipment or chips. And then before Biden left office in his last few weeks, he implemented another law that certain countries also who they suspected were supplying China with chips and equipment had to apply for licenses to get chips. He set up a three tier system where all American allies can have free access to the chips and technology from the US. Then there was another level that had to apply for licenses. And then there was another level, which included China and Russia and Iran, that could not have access to that technology. Now, all these sanctions have been initially they were under the guise of national security. But it's later transpired that actually these were put in place to stop China's technological rise. And one of the main things they were trying to prevent was China's development in AI. But China being China, you know, China, Chinese people are very, very smart. They have a huge amount of very, very smart engineers here. They got to work and there's a saying in China. What doesn't kill you makes you stronger. Huawei kind of went into a quiet phase for a couple of years, but all of a sudden they came back. There's been many companies who've been developing AI software in China. You've got people like iFlytek, you've got ByteDance, you've got Baidu, you've got Tencent, you've got Alibaba. They have a lot of companies that can do this kind of thing. They don't just have one or two. But for a long time, just back about six months ago, Eric Schmidt, who seems to be some sort of consultant now to the government or works for some think tanks. Previously, he was the head of Google. And just about six months ago, he was suggesting that, oh, we don't need to worry about China. They're about 18 months to two years behind. However, when some of these models started appearing from these Chinese companies, he made another speech in an interview about two months ago, I think it was. He basically then says, oh, they're just now a few months behind. So he was actually wrong in his assumption. It's obvious that the American authorities don't have a lot of intel on what actually China are doing because he got that prediction very, very wrong just six months or so ago. Anyway, so for some time, the US thought they were really, really far ahead. But yesterday we had an event occurred that have absolutely shocked the USA and has really set the AI community buzzing. And there's a fairly smaller company, AI company in China called DeepSeek. And it's made up from the people who run it. They have, I guess, a couple of hundred engineers. It's not a huge company. And most of them are certainly the owner from Zhejiang University. They released a model which was called DeepSeek R1. And it's turned out that this is performing as good, if not better than all the Western models. Those from people like OpenAI, Anthropic, Meta, etc. And the thing is, it was trained on a much smaller amount of GPUs. As you know, most AI companies are using Nvidia GPUs. China has been restricted to the Nvidia GPUs it can get. It can't get the latest ones. Nvidia made a version that was for the Chinese market, which is less powerful. The memory bandwidth is less. And the other thing, not only was it trained on a much smaller number of GPUs because of the restrictions, it costs less than $6 million to train. This is significantly less than models by the Western providers. I think Sam Altman hinted that the latest model of ChatGPT cost over $100 million. So many in the US AI community have been kind of scratching their heads thinking how the heck a Chinese company managed to produce something that performs as good and better on certain benchmarks than ours. It's just got them scratching their head. But that's not the biggest thing. The biggest thing that it's open source and they open up their API and the API costs about 30 times less than those American models. So this has like huge implications. And I'll talk about them in a second. But as well as the main DeepSeek R1 model they've released, they've also released some more condensed models. And within hours of this model being released, because it's open source, we've had videos and stuff appear on the Internet of people literally running it on their home computers, their phones, their MacBooks, Mac minis. So this is like the developer community for AI is like absolutely loving this and like massive amounts of the AI developer community have jumped on this. Now what this means is that let's say, for example, to run some instances on open AI, let's say it costs you $10. Well, to run similar on DeepSeek R1 might cost you only 30 cents or 50 cents. So it allows developers who are selling software as a service based on these AI models to price their products significantly cheaper, significantly cheaper. And it will make a lot more of these developers products viable because instead of charging users, say, I don't know, $20 a month, they can now charge users maybe $5 a month and still probably make more money. So it has massive implications for the AI developer community and end users. And the other thing is that if you remember when AI first started, they were meant to be kind of a nonprofit. But as Microsoft have invested billions of dollars and other investors have put billions of dollars in, that's all changed. And if you look at this DeepSeek R1, it's on par with the latest model from open AI. And if you want that latest model from open AI as a consumer, you need to pay $200 a month, $200 a month. Now, you can get this as an end user, DeepSeek R1, you can go to their website and access it for free of charge. Obviously, as I said earlier, if you're a developer, you have to pay for the API, which is fair. But the whole mission of DeepSeek is to bring AI and even AGI to the masses for free. And this is why they have open sourced it. And this just has massive implications, you know, because you've got these AI companies in the US who spent hundreds of millions, billions of dollars on creating these. And they've got to make a return on that. So what my concern as a user was that, and I've mentioned this numerous times before to friends and other people I've spoke to, is that the way open AI was going and some of these other AI companies in the US, these models would have got, we've seen this with the, you know, charging $200 to access open AI's latest model. Would this have gotten more and more expensive so that the best AI would have only been affordable to sort of more wealthy people? Because, you know, for a normal user like me, if I wanted the latest open AI to pay $200 a month, that's a lot of money. I currently subscribe to the $20 a month package. But now this DeepSeek has come along. I'm using it a little bit. I'm seeing how it runs, how it performs. And I'm considering cancelling my subscription to ChatGPT because so far from just playing with this for a couple of days, it's just as good if not better. And so I think it's fantastic that DeepSeek has come along with this open source model. And that is their mission. And I think this just shows you how there's other Chinese companies that are also open source. I just think this shows how China are not trying to, you know, keep this tech for themselves. I feel these US companies are trying to, you know, shield this tech. And then they can only, you know, only certain people can have it as it gets more and more expensive. And DeepSeek have actually just blown this wide apart. And, you know, and it's not only DeepSeek, ByteDance has also released a new model which is performing better than the leading US models in many benchmarks. So to me, this just shows you that China are not at all behind on AI. And if you look at what they're doing with humanoid robots, there are numerous companies in China working on this. And I would actually argue, living here in China from what I've seen over the last few years, is that I would argue that China are not actually behind at all in AI. And I would suggest that in certain areas, they're probably actually in front. Now, I don't have any hard evidence, but I've seen factories and production lines where they're using AI, where there's almost no humans there, you know. Yeah. And interestingly, there was a memo, an internal memo from Meta leaked online. If you go on Twitter, you can probably find it. But it was basically saying that they're in a bit of a panic over this DeepSeek. And the reason they're in a panic is that their AI unit spends like huge amounts of money creating these models. And actually, this DeepSeek have spent five and a half, six million dollars, and it performs better than all the Meta models. This memo was saying that there's actually some managers or heads of department. Their salaries are actually more than what DeepSeek have developed their whole model with. So they have engineers or heads of departments there that their salaries are more than five and a half million dollars. And they've spent all this money and made these models. And then DeepSeek comes along and blows them out of the water. And Donald Trump basically has just announced a private investment of 500 billion dollars for AI in the US under a project called Stargate. And apparently, this is meant to make sure that the US are the leaders in our technology. But if I read an article in South China Morning Post earlier today that China have actually been building out these sort of AI data centres for the last two years, they will probably invest even more in those now. But I really wouldn't sort of bet against China in the AI race. What's sad is that things would be so much better if these countries could compete with each other. You know, it would be better for everybody. I do see that China have made a massive sort of, you know, impact on the AI community with releasing this DeepSeek as open source and also other Chinese companies releasing these models as open source, where the US are trying to keep them as closed systems like Apple does, like Microsoft did with its Windows. And I really think that potentially DeepSeek could become to AI or Android was to mobile phones. And it just shows you that all these sanctions that the US have thrown to China over the last five or six years have had little or no effect at all. Anyway, I'm sure we're going to be in for an interesting few years to come. Anyway, and I'll see you in the next one. Thank you.