U.S. Government Responds to DeepSeek's Challenge to Nvidia
16:31 Share

Today on the podcast, we're going to go deep on DeepSeek. I know, not a great pun, but this is something that obviously everyone has been talking about lately, and it has wiped off hundreds of billions of dollars off the U.S. stock market with a grand conspiracy from hedge funds. NVIDIA has...

collapse their stock price. So much has happened. I think it was the single biggest stock price wipe out of any individual stock that happened to Nvidia with all of this. So we're going to get into all of it, what people are saying, and why I think that a lot of this is quite overblown and some areas that I think are actual, actual big breakthroughs from all this. So let's get into it. Let's say DeepSeek is a chat GPT competitor that is coming out of

I wanted to give just a two second overview of what it's capable of. So it's got an LLM model like ChaiGPT, fantastic. They just released their reasoning model, which is like the O1 in ChaiGPT, which is like the best model. And when it came out, they put it up against all of the other AI models in the AI model arena. And it did quite well. In fact, it was like out of a hundred, it was like one point behind O1. So pretty much on par exactly with O1 as far as the quality of the outputs. So

this thing's really good. How did they get it this good? First, I wanted to just quickly say I played around deep seek and you probably heard a bunch of people say this. It's good at a lot of things. It's got internet search capabilities, which is fantastic. And it's obviously censored by the Chinese government. So those are kind of the two big things. And just quickly to illustrate that point, I asked it,

who killed JFK. It goes through all of its reasoning, which is great. Searching the web is looking at Wikipedia, a bunch of other places, and it says, you know, the assassination of President John F. Kennedy.

remains one of the most scrutinized events in modern history and then it's like the official conclusion is that Lee Harvey Oswald was the lone assassin goes through all the reasons why then it says the controversies and conspiracies and it goes through some of the controversies in regards to the CIA and the FBI and how Oswald took a trip to Mexico City where he visited the Soviet and Cuban embassies he left a threatening note at the FBI headquarters but they destroyed it

The CIA never reported to the FBI that he went, whatever, all of the conspiracy stuff that's crazy about this whole story. They put it on there. And I think that's fantastic. Great. Fine. This is kind of like one of those controversial stories.

American moments. They put it on there. Now, if you ask it about a controversial Chinese moment, what's really interesting to me about this is that it will give you the response and it will go through the whole process of thinking. So it's censorship happens after the fact. It's not, it doesn't read my question and censor it at the question. It censors it at their actual response. When I say how many people died in China under Mao,

It went through all of its reasoning, all of its research, and even the interesting thing is because it shows you the search results that it found. So I can click on those search results and see mass killings under communist regime on Wikipedia. I can look at the Great Chinese Famine. I can look at death and oppression under Mao. There's all of these different, like, Wikipedia lists. Yeah, all of the stuff, all of the data and search results are still there, and you can actually click on them. So that's interesting that it's not been censored at all.

And then it actually wrote its full response to me, which was like between 40 and 80 million people died. It was very controversial. This was during, you know, the people's revolution. Like, so it wrote this whole thing out and I read it. And then immediately after it finished typing it out, it instantly disappeared and got replaced by a story that's beyond my current scope. Let's talk about something else.

This is the same thing if you ask about Tiananmen Square, if you ask about Xi Jinping, if you ask about the CCP being anything negative about it, if you ask it about, I asked it earlier today, like who Tank Man was from China, and it gave me the whole response, and then it disappeared and said, sorry, that's beyond my current scope. So evidently, DeepSeek is Chinese and censors anything that is not Chinese, but it just does it in such a

way it's kind of mind blowing that they would allow I don't know it's just weird to me being it's just weird so anyways it's it seems like it's way too on the nose and confusing why how people could trust like what all it's it's not even a conspiracy like what other ideologies are they you know suppressing like we always complain with chat GPT being like politically biased and stuff and then like putting stuff in there

And then chat GPT is like, no, that's just the training data. And that's just what it said. And whatever. There's like all those arguments. That's the argument we're having here. We're straight up having this. Sorry, that's beyond my current scope. Let's talk about it. It's like, it's clearly censored. Okay. So that's the big controversy with DeepSeek. Now let's talk about what has happened from DeepSeek. So DeepSeek came onto the scene. Really phenomenal model. Like this thing's powerful. It's just as good as chat GPT for the responses. It's way cheaper because they let it be open source so people can host it locally on their device, which...

No, it's censored by China locally hosted on your device. The good thing about that is that none of the data that you put in there is sent to China. When you go to the deep seek website, like I'm on right now and I asked this question and I'm logged in with my profile, it's now making a profile of me sending it to some data center in China. It's like, it was me. I'm probably getting added in China of people that have asked five questions that are quote unquote beyond their scope. Right? So I'm probably added to some, um,

Chinese list, I probably can't visit the country anytime soon. You run it locally on your own computer, you're totally good. And yeah, that's totally fine. So that's the cool thing about the model, other than using it, you know, where it's accessing them. So

This all was created by a hedge fund in China that does quantitative trading. And what's interesting about this is they, when they released it, they said we spent $5.5 million to train this whole model. And, you know, we're spending billions of dollars with these top American companies, hundreds of millions of dollars to do the training runs for open AI and for offing a bunch of these other players. So it's like, how does it cost us a hundred million dollars or three or $400 million? And it only costs you $100,

$5 million. This essentially tanked the price of Nvidia. If you look at the price of Nvidia, it went from trading at around $142 a share, crashed all the way down to $117 a share. It was a 18 or 19% drawback collapse in the price.

which wiped out like four or I think it was like two to $400 billion in market cap, crashed a bunch of other stocks like AMD as well that also saw some pretty steep dives. Now, the interesting thing with all of this, AMD also saw, yeah, some big crashes as well. The interesting thing with all of this is if you look at Nvidia today,

after that big you know crash and the reason why it crashed was like well nvidia sells chips to train ai models but all of these big ai companies have so many chips now if they can all cut their costs by 30x by following deep seek because deep seek published the papers on how they did it like then they're not going to need to buy any more chips and nvidia is going to crash no one's going to buy chips anymore and it's going to crash okay

That was yesterday's news with like a 17, 18% drop. Today, we're back up 8%. And I am proud to say...

Did buy the dip yesterday not financial advice. I bought it at about 120 $121 we're about 128 dollars So I didn't get the absolute bottom, but I got it before it gave a rally I mean my prediction here is that this is all overblown and I'll give you all of the reasons why But again not financial advice just that's what I did just FYI You know full transparency so the interesting thing here recently we had

a US tech mogul Palmer Lucky inventor of Oculus and Andral said deep seek is legitimately impressive but the level of hysteria is an indictment of so many um

is an indictment of so many. The $5 million number is bogus. It's pushed by a Chinese hedge fund to slow investment in American AI startups, service their own shorts against American titans like India, and hide sanctioned evasions. America is a fertile bed for psyops like this because our media apparatus hates...

Okay, this is Paul Merlucky. Now, a couple things to break out here. First of all, I've seen a lot of analysis on this $5 million number, and virtually everyone says...

It is impossible to train with $5 million. They break down exactly how many chips this hedge fund allegedly had that trained this because apparently it's just a side project this hedge fund did. So it's like, oh yeah, we just have a random hedge fund. It's just this random side project for $5 million and it's crushing all your billion dollar companies. So that's kind of what scared the stock market. As this is becoming increasingly less credible, that's why we're seeing an 8% rally on NVIDIA today and people are pretty much not believing it. Like Palmer's saying,

So they're looking at how many chips they have and they're saying even with that amount of chips, that would have been, you know, like a hundred million dollar fund. Like so or yeah, much, much more than five million dollars. So the reason why they reported the five million dollar number, number one, this hedge fund legitimately did take out short positions on NVIDIA and others verified. So.

They took out short positions, announced this crazy $5 million number, watched NVIDIA crash, and I'm hoping, or I'm assuming, cashed in because I don't think that it will stay crashed forever. So I'm assuming they try to...

cash in on a 17% crash that they caused from this. Which is still a... If that was all the hedge fund did, if this was a side project to crash NVIDIA's 17%, and that was the only goal of this, that was still obviously a major incredible heist because they had to train an entire AI model to get as good as JGP. So I don't think that's the only thing they did. This will fund more chips. We'll just...

short them, fund more chips. What's actually happening with the chips in addition to that that he mentioned here as far as hiding sanctions and sanction evasion is what a lot of people have been bringing up, which is the fact that DeepSeek...

has apparently, or sorry, NVIDIA apparently, is selling about 25% of all of their revenue is selling chips to Singapore. Now, Singapore, of course, is right next to China. Singapore does not have any, doesn't have any sort of

stipulations. China were banning the most advanced chips from being shipped to China because we want to maintain an AI edge, but there's no problem with Singapore. So really what's happening is shipping the chips to Singapore companies who are reselling them to Chinese companies. So that's why 25% of Nvidia's revenue goes to Singapore. That's not happening. It's actually 25% going to China and companies like the hedge fund that owns DeepSeek are going and buying those.

So this is essentially what's happening with DeepSeek that is interesting. They're essentially just trying to hide the sanctions and short the company and make a lot of money in the meantime. Now, in the meantime, every single news organization, every single influencer, everyone is talking about DeepSeek and how it's going to crash the American... I've had so many random people text me asking about this. What's interesting to me about all of this is...

It's just the fact that it's become so popular that now tons of people are using it. It became the number one app on the App Store, DeepSeek. And definitely if it's cheaper, you can make a case for this. But what was really funny that happened was that DeepSeek itself yesterday got so overwhelmed by new user signups on their app that they actually stopped any Americans or anyone from anywhere other than a Chinese phone number from signing up for the DeepSeek app.

Which is pretty funny because obviously they need to use chips to run this AI model. And they don't have enough bandwidth to run the AI model. So they had to block it from people. So what would the solution be? Well, the solution would be for them to buy more chips from NVIDIA. So I don't think this is a very good...

you know, narrative Nvidia's crashing and dying when literally one of the, you know, this huge breakout AI company doesn't even have enough chips to run their stuff. They're gonna have to pay Nvidia more money. So I think Nvidia is going to keep making a ton of money. I think some really cool things that Palmer Luckey said was DeepSeek is legitimately impressive. DeepSeek is legitimately impressive.

This is true. It is legitimately impressive. They are doing some legitimate things where they saved a ton of compute. They really optimized a lot of things with how they trained their model, with how they use a bunch of these mini micro models, how they reverse engineered OpenAI's reasoning capabilities.

and built it in and it's really cool how it shows you the reasoning capabilities. You see it whereas with OpenAI they hide it so you can't see it. So DeepSeq is doing a lot of really impressive things. It's not actually like open source but they do allow people to take the model and run it themselves. It's not open source because we don't have all the model weights and we don't have all of the data that goes along with that and it's not technically open source but they are allowing you to take the model, to run the model,

giving you certain weights and information about it. So that is, you know, it seems generous. You have companies like Meta that everyone's like, oh my gosh, Meta was supposed to do this with Lama. They didn't do it. OpenAI was supposed to be an open source company and they didn't do it. And now we're relying on the Chinese company to do it. So I will give them some credit for a lot of that that they've done and the fact that they've come up with the papers that they've written on some really impressive things they've optimized. Now, the thing about it is whenever everyone's saying like,

DeepSeek is going to be the end of American AI and it's so over. We even had Trump give a press conference yesterday and David Sachs, the head of AI and crypto in America, has been tweeting about DeepSeek all day long. And in my opinion...

This is actually an incredible moment. It's going to catalyze. It's going to be a catalyst for America and American economy up until this point We've had Essentially to be more competitive. So that's that's my belief and the reason is because up until this point in the United States We've had a bunch of big players right we had and throw up. We have opening eye. We have meta We have Google all of these companies are working towards developing really impressive LLM models now

We've had this competition, but I don't think we've been as in the system best we can because we focus on a lot of things in America that are American culture. How

I mean, yeah, I'm sure some people will disagree with me on this, but my opinion, the how, you know, how biased is it? And we can't let it say this, or we can't let it be offensive to these groups. We can't let it do this. And we have, you know, like Google being accused of being a woke AI and generating historically inaccurate photos of, you know, black people that were Nazi soldiers, like,

Ridiculousness, just ridiculousness has been a focus. It's not just purely on making these models the best they possibly can be. And so because of some of this distraction and some of the over worries about safety and even I know a lot of people are have concerns about copyright. This is not an issue in China. Deep Seek does not care about copyright. There is no copyright rules in China. They have to abide by. They can use anything they want.

So there's a lot of things that we focus on that they do not that allow them to be faster. And I think at this moment, because of what's going on at a national level, we're now trying to nationally compete with China. We're viewing them like we have our local companies competing with each other. But now this is like a national thing. And so it's how do we get more compute for our AI companies? How do we get more energy for the data centers to be able to run? There's a lot of things on a national level that I think are going to start rolling out because of this. People are calling it the Sputnik moment, right?

So I'm very excited. I think that this technology is going to get much better. It's becoming a much bigger focus with a tool like this. I love the competition in the market and I don't think in any way that open AI is not up for the challenge. Sometimes they just got to be kicked into gear a little bit along with a bunch of these other companies. So really excited to see what happens. DeepSeek, an impressive piece of technology and excited to see how this spurs the entire economy forward.

forward. Thanks so much for tuning in. If you enjoyed the podcast today, I would really appreciate it if you could leave a review wherever you get your podcast, and I hope you all have a fantastic rest of your day.