# Nvidia Ceo STUNS "China Are NOT Behind..."

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=lBwh6_V_d8Q
- **Дата:** 02.05.2025
- **Длительность:** 17:38
- **Просмотры:** 11,424

## Описание

Join my AI Academy - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/


Links From Todays Video:
https://hai.stanford.edu/ai-index

Welcome to my channel where i bring you the latest breakthroughs in AI. From deep learning to robotics, i cover it all. My videos offer valuable insights and perspectives that will expand your knowledge and understanding of this rapidly evolving field. Be sure to subscribe and stay updated on my latest videos.

Was there anything i missed?

(For Business Enquiries)  contact@theaigrid.com

Music Used

LEMMiNO - Cipher
https://www.youtube.com/watch?v=b0q5PR1xpA0
CC BY-SA 4.0
LEMMiNO - Encounters
https://www.youtube.com/watch?v=xdwWCl_5x2s

#LLM #Largelanguagemodel #chatgpt
#AI
#ArtificialIntelligence
#MachineLearning
#DeepLearning
#NeuralNetworks
#Robotics
#DataScience

## Содержание

### [0:00](https://www.youtube.com/watch?v=lBwh6_V_d8Q) Segment 1 (00:00 - 05:00)

China is not behind. I anybody ahead of you China's right behind us. I mean they're we're very close. And so that was Jen Sang Hang. And in that video you can clearly see that he says that China is not behind. And even if they are behind, they're not far behind at all. And that's what I'm going to be talking about in today's video. Are China actually behind? And if they are behind, just how far are China and could they possibly catch up? Anybody ahead of you? I China's right behind us. I mean, they're we're very close. Uh and but remember, this is a long-term this is an infinite race. There's no, you know, in the world of life, there's no those, you know, there's no twominut end of the quarter. There's no such thing. And so, we're going to compete for a long time. And just remember that this is a country with great will and they have great technical capabilities. 50% of the world's AI researchers are Chinese and so this is a uh this is an industry that we will have to compete for. So I think the answer might actually surprise you because there has been so many different pieces of information online about China and the US and all of the neighboring countries and how countries are trying to dominate AI. But I was recently looking at an AI report that actually tracks all of this pretty factually. So, as we can see right here in the new Stanford report, they speak about how the US's lead over China is rapidly shrinking. Now, this was before Jen Sen Huang said what he said. This was a few weeks before. So, we know that the theme of the United States actually losing their lead to China is now, you know, it's a common theme. This is something that we're no longer surprised by. And when we actually take a look at the article, it talks about the model quality. So one thing it mentions is that the report found that you know institutions based in the US produced 40 AI models of note in 2024 compared with just 15 in China and three from Europe according to the 8th edition of the Stanford AI index released on Monday. Now, the report also found that Chinese models have rapidly caught up in quality, noting that the Chinese models reached near parenchmarks after being behind leading US models by double-digit percentages a year earlier. So, basically, these Chinese models have rapidly caught up despite, you know, not the ones kicking off the air revolution, but seemingly doubling down on their strengths and managing to follow suit. We can see here that is illustrated even more clearly in this chart right here. So what this graph shows us is that on one particular benchmark which is the LMSYS chatbot arena. This measures the chatbot on a variety of different things namely the human opinions of the model. We can see here that the United States has maintained a lead since 2024 January. Of course, with things like GPT4 and stuff, but over time, we've seen that Chinese models have rapidly been slowly but surely closing that gap. We can see that here the gap was pretty big, maybe like around 150 points or so. And then around here, we can see that the points in terms of the LMSYS arena, it's only around 20 points or so. So there's not really that big of a difference between those frontier models in terms of overall quality. Now I do have to be honest with you guys as someone who's using the AI models on a day-to-day basis. There is definitely a significant difference between models like you know claw 3. 7 Sonet and something open source like DeepSeek because there are realworld applications for these models that people are using on a day-to-day basis. So when you test the models it is definitely different from you know any particular benchmark but I will say they have definitely been catching up. Now one recent edition of that most people didn't even realize and I even made a video but most people were writing this model off is Quen 3. Now I honestly believe that Quen 3 is you know maybe not another Deepseek moment but the model when I used it was really good. Like one thing that we have to understand about these models now is that these benchmarks which I'm about to show you right here, they aren't really indicative of how the model performs in real world scenarios. Currently, what we used to have as a real intuition on how the models would perform is essentially their quality on all of these kind of exams. Things like math exams, things like the live codebench, things like code forces and live bench and arena hard. All of these benchmarks usually provide us with some kind of indication to where the model stacks up with its raw intelligence. But sometimes that raw intelligence doesn't result in the model being useful. It only results in it getting the correct answer in some sort of exam. And what that means is that sometimes certain models that may excel on exams or not even perform that well in exams actually perform better when you're just using them on the day-to-day for whatever industry specific task that you might have. And that's what I was finding with Quen 3. This is an open- source model released from China literally only two or three days ago.

### [5:00](https://www.youtube.com/watch?v=lBwh6_V_d8Q&t=300s) Segment 2 (05:00 - 10:00)

And despite the benchmarks being honestly some of the most surprising, we can see the arena hard. It basically matches Gemini 2. 5 Pro, the math benchmarks. We can see that it's up there on the same level on the live codebench and the code forces. This is essentially something that I would say is a really good model. It's actually got to the point where I'm even recommending this AI model to members of my AI community. I'm even writing quick start guides on best practices to use the model because it is genuinely a good piece of software that is basically free. But I digress and it does lack in some areas such as the coding. But the point is that China is not as behind as most people think. And China, you know, won't stay behind forever. So the US thinking that they've had to maintain this imaginary lead for so long, it's not exactly true. I mean, even for their smaller models, their smaller models are performing really, really well. We can see it's on par. this 30B parameter model on par with Deepseek V3 and GPT40 and even their reasoning models with the thinking mode. We can see that it gets so much smarter at these benchmark tests which is absolutely incredible. Now another thing that you have to you know understand is that you know China is coming out with models all the time. Often times there'll be models that you probably won't even have realized that were just released. Take for example the other day we got Ernie X1. This was a model from BU who released Ernie X1 Turbo. Think of it as the same as Deepseek R1 or Open Eyes 01 which is their reasoning model which you can see here. It manages to perform on par with Deepseek R1 and Open Eyes 01 for a fraction of the cost. This thing costs a couple cents to run per query and it is honestly rather impressive. I really do think that the only reason most people don't use these models is because they literally don't know they exist. Now, of course, you know, most models do have a nice product and chat user interface. For example, OpenAI is wrapped around the chat GPT user interface, which is relatively easy to use, and some of these are in some rather questionable browser choices. But honestly, guys, if you're just looking at these benchmarks, there's no real magnificent drop between these models. Sure, there might be a 1 to 2% here, but when I'm using some of these models, guys, there really isn't that much of a difference. the frontier models that we can all see here like Gemini 2. 5 Pro, Chat GBT4, Grock 3. I mean, it seems that we're, you know, somewhat converging around the number one spot and there's almost this real level that it seems that, you know, large language models leveled off that there doesn't really seem to be one model that has broken the mold. I personally do think that the USA is actually, you know, still pretty ahead in terms of the real world usability, but in terms of benchmark cracking, I think that has been saturated for quite some time. Now, something that was really fascinating to me when actually looking at how quickly China have actually caught up to where the US is, I want you guys to see two clips. So, in 2024, Eric Schmidt, the ex Google CEO was talking, you know, in a conference, he was basically speaking about the global relations with AI and what countries are behind. And he basically said that the good news is that the US is way in front. Now, remember, I wanted to show you guys this because this was in 2024. He was saying that in 2024 that China is around 2 to 3 years behind. And then I'm going to show you what he says in 2025. The good news is the US is way ahead of China and everybody else. And I think that's going to continue for a while. To me, national competitiveness is the challenge for the next 10 or 20 years because the Chinese are really focused on dominating certain industries and we need to compete with them and make sure we win. In the case of artificial intelligence, we are well ahead two or three years probably of China, which in my world is. Now guys, let's take a look at what he says or slash what he was saying in 2025 or in fact this wasn't even late 2025. This was actually early 2025/ late 2024 when he said this guys and this was when deepseek basically was released this week. Um there were two libraries from China that were released open source. One is a problem solver that's very powerful and another one is a large language model that's equal and in some cases exceeds the one from meta with it they use every day. It's called llama 3 400 billion. I was shocked when I read this cuz I had assumed that our in my conversation with the Chinese that they were uh two to three years late. It looks to me like it's within a year now. So, it' be fair to say it's the US and then China within a year's time. Everyone else is well behind. And the reason I wanted to show you guys that clip is because in under one year he, you know, changed his tone so much before he was like, "Oh, well, you know, China is just two to three years behind. Everyone else is well far behind. " In literally less than 12 months, he went to being absolutely stunned by what is coming out of China. And remember guys, this isn't some guy who's just randomly browsing on the internet. This is the ex Google CEO. He really does have a knack for understanding anything going on in the world of AI and tech. So this was

### [10:00](https://www.youtube.com/watch?v=lBwh6_V_d8Q&t=600s) Segment 3 (10:00 - 15:00)

something to me this was like bing bing bing. This is a wakeup call. Okay. China are moving exceedingly fast at this and they've had their AI initiative for so long. Like they've been at this AI game for a very long time. So once this chat GBT moment was here, I'm pretty sure they all just locked in on the fact that, you know, this is going to change the future and they're probably doubling down on this, investing billions of dollars. Now, the crazy thing about this is that most people don't understand why it's imperative that the United States actually win this AI race. Maybe people just think that, you know, okay, maybe they just want market share, maybe just want more customers. No, guys, this is a serious issue. Okay? And this is where the Eric Schmidt is actually warning US Congress about the fact that the US cannot afford to lose this race. And he talks about what would happen if that occurred. Imagine a situation where attacks that we cannot even imagine are unleashed by China in an adversarial thing. We have no concept of having a super intelligent opponent where we're not as intelligent as they are. Everyone assumes that it's a battle of missiles and aircraft carriers. That's not correct. It will be a battle of swarms of drones. Those drones will be highly intelligent, highly planned, and they'll do maneuvers that no one can anticipate. We collectively are not ready for that. Imagine a situation where China has invented new algorithms for military attacks and defense that we cannot even conceive of. Remember, I'm discussing a world where humans have a partner that is smarter than the collection of those people. As I said, this will happen in our lifetimes and it's important that we get there first. Uh if you take a look at Ukraine, Russia right now, you see the future of war. Um I'm assuming, by the way, that China would start by cyber attacks and so forth. There's evidence that uh these new systems will be able to come up with zeroday exploits that we cannot foresee. A zero day exploit is something we've never seen before and we can't anticipate. There's lots of people who were worried that biological attacks can be done and there there's lot there's a report from the emerging biothreats commission this week with the great details and there's a classified version that all of you should take a look at. There's plenty of evidence that these things are possible and I actually made another part of this where I spoke about this last week where Eric Schmidt was basically talking about the fact that you know this is a real issue and Alexander Wang the CEO of Scale AI also spoke about this why it's absolutely essential that we win this race to a goal that's not as clear as I would like. Uh Dr. Schmid um in 5 to 10 years every American citizen will have the equivalent of an Einstein on their phone or in their pocket. This is an enormous increase in power for humans. What if that Einstein is a Chinese one? If we fall behind the Chinese Communist Party, uh this technology will enable the CCP as well as other authoritarian regimes to utilize the technology to uh over time effectively take over the world. Um you know, they'll be able to export their ideologies. utilize as a military technology to invade other countries. um and they'll be able to use it for uh effectively spreading their regime in a more broad way across the world. Now, I was reading some stuff online and one of the things I kept seeing was that people keep saying that, oh, you know, China, they don't have the GPUs. Okay, we've got these export controls and we are basically having these restrictions to China and other countries citing concerns over their memory, bandwidth and you know, supercomputing uses. So recently you can see here this article talks about how Nvidia writes off 5. 5 billion in GPT in GPUs as United States governments chokes off supply of the H20s to China. So basically if you haven't you know been paying attention there's been this entire thing where the government has basically been shutting down China's main GPU supply. Now, this can have unintended consequences because this basically forced Deep Sea to come up with genius ways to make their AIS much smarter and more efficient on limited compute, which is, you know, kind of, you know, I don't want to say shooting yourself in the foot, but, you know, when you make do with what you have, sometimes you become really innovative. Now, basically, the H20 was a weakened version of Nvidia's H100s, which were, you know, designed to comply with earlier US restrictions. Yet it actually became really popular among major Chinese companies like Alibaba and Tencent. And despite Nvidia Jensen Huang's lobbying effort and restrictions that went ahead, the US basically, you know, they also restricted AMDs even though that the AMD is less affected due to lower sales volume. So the US of course is tightening AI hardware control. They, you know, blocking even compliant downgraded GPUs like the H20, showing it's no longer tolerating workarounds. This is part of the broader strategy that they're using to choke off China's access to cuttingedge AI compute, which is of course critical for training advanced AI models. Now, of course, this could slow down China's AI progress. And Chinese companies have pre-tocked $16 billion worth of chips, but once that supply is depleted, further growth will face bottlenecks. Now, these chips do power AI infrastructure. And without them, progress on large language models and, you know, applications like autonomous vehicles, surveillance, you know, they

### [15:00](https://www.youtube.com/watch?v=lBwh6_V_d8Q&t=900s) Segment 4 (15:00 - 17:00)

could be delayed or degraded. And you know, US companies take a financial hit, but they do maintain that strategic advantage. Like Nvidia's 5. 5 loss is a short-term financial blow, but in the strategic term, the US is signaling that it's willing to sacrifice profits to maintain the tech dominance. Now, of course, these moves are basically going to accelerate China's investment in its own GPU ecosystem, such as companies like, you know, uh, Bern or Huawei. And this is really true because look at what happened a few days ago. We saw that Huawei is preparing new AI chips for mass shipment to China. And if you haven't been paying attention, this is where things start to get interesting. We've already seen that China has already managed to basically close the gap without sufficient compute. What happens when they get their hands on enough compute that the US has? I'm not saying that that's going to happen straight away, but we have to think that this is a country that is not backing down and is really always maintaining a competitive edge. Huawei's has had, you know, recent advancements and rapid advancement in AI chip technology, especially with the new Ascend 910C and the 910D processors which are significantly altering the competitive landscape in AI hardware. And whilst these developments are primarily boosting China's domestic AI ecosystem, they actually pose a growing challenge to US dominance in the tech sector, particularly as the United States exports controls restrict NVIDIA's access to the Chinese market. Now, of course, there are some significance. Huawei is mass- prodducing these chips, the 910C and the 910D, which are basically designed to rival the Nvidia H100 chip in performance. They basically now provide Chinese tech firms with domestic alternatives to Nvidia GPUs, which of course are we, as we know, you know, they're pretty hard to get. And what's crazy is that Huawei now accounts for more than 3/4 of AI chips produced in China, disrupting the local ecosystem and emerging as the most viable alternative to Nvidia for many Chinese firms. And this actually allows them to reduce their reliance on US technology. And the crazy thing here is that while Huawei chips may still lag behind Nvidia's top tier offerings in terms of individual performance and power efficiency, the gap is closing. Nvidia, you know, the CEO of Jen Hong like at the start this video, he stated that, you know, the gap is narrowing. Okay, this these companies, you know, they aren't going to be behind forever. And he actually described Huawei as one of the most formidable tech companies in the world. So overall, let me know what you guys think about this. Do you think China will win this AI race? And as Jen Sang Huan said, this is quite the infinite race. So, it's never going to be around forever.

---
*Источник: https://ekstraktznaniy.ru/video/12886*