# Eric Schmidt DROPS BOMBSHELL: China DOMINATES AI!

## Метаданные

- **Канал:** TheAIGRID
- **YouTube:** https://www.youtube.com/watch?v=ddWuEUjo4u4
- **Дата:** 27.11.2024
- **Длительность:** 10:05
- **Просмотры:** 190,915
- **Источник:** https://ekstraktznaniy.ru/video/13670

## Описание

Prepare for AGI with me - https://www.skool.com/postagiprepardness 
🐤 Follow Me on Twitter https://twitter.com/TheAiGrid
🌐 Checkout My website - https://theaigrid.com/


Links From Todays Video:


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

## Транскрипт

### Segment 1 (00:00 - 05:00) []

China now has two of what appear to be the most powerful models ever made and they're completely open so that was Eric Schmidt talking about how China has two of the most powerful models and they are currently open source and in this video I'm going to dive into a lot of his statements now the reason why we're diving into Eric Schmidt is because he is the former CEO of Google and he had a recent talk at Stanford that went pretty vial and his statements on AI seem to be shaking up the entire community so let's take a look at some of his recent comments in this recent interview and I'm going to break them down with a lot of information so you guys can understand exactly where the future of AI is currently headed this week um there were two Li 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 which they use every day it's called Lama 3 400 billion I was shocked when I read this CU I had assumed that our in my conversation with the Chinese that they were uh two to 3 years late it looks to me like it's within a year now so be fair to say it's the US and then China within a year's time everyone else is well behind so of course one of the models he's talking about is deep seek R1 now if you weren't familiar with this model actually went completely viral you can see it's got you know 2. 8 million views th000 retweets you know 4,000 likes and essentially this was China's version of 01 they called it R1 light preview and this one was pretty crazy because 01 was a model that nobody saw coming from open a ey I mean I guess you could have seen it if you dug through the research papers but the fact that he managed to get a working version that is quite similar to 01 and even surpasses it when we're actually looking at the benchmarks in certain capabilities it actually shows us that China are not far behind at all you know Eric Smith basically says that he thought China were you know um 3 to you know 2 to 3 years behind but it seems that nii currently they are basically you know 6 months to a year behind in terms of AI you know development and I mean if this was a closed Source lab some people could potentially argue that this company is ahead you know in the sense that they're able to you know get this deployed faster than Google faster than XI and of course faster than anthropic so this is something that is truly surprising when we take a look at this entire model space and we do have more benchmarks here and you can see that the purple bar the one on the left is surpassing 01 preview which is the one that is in the green which is right next to it so all this is pretty surprising I'm not going to lie like definitely really shocking and I do want to say that this is something that is so surprising that I am wondering what on Earth is going on in terms of the AI space because this was not an update that I expected if anything I would have expected that an American company would have been the ones to you know race forward with development but I guess that somehow we're here with open source as well now if you're wondering what he's talking about with the other model is actually quite popular when it comes to you know running model models on your computer um and the model that he's referring to is quen 2. 5 now quen is essentially coming from alibaba's cloud so this is alibaba's you know company and of course they have open sourced this model called quen 2. 5 72b instruct and it's a 72 billion paramet model so it's not extremely large but it is really good at performing certain tasks and one of the things that we saw here on the compass large language model leaderboard was that the quen 2. 5 72b instruct actually managed to surpass steady the-art models like CL 3. 5 Sonet and GPT 40 so when we do take a look at this entire development cycle and we're taking a look at how quickly these models are developing seeing a model that's able to do this is one that is truly surprising considering the fact that it was only a few months ago we had a one preview and we were like wow this is something absolutely amazing now it's pretty crazy because when we actually look at AI development we know that China is facing several restrictions recently when we look at what SMC did they recently closed their door on producing Advanced Air chips for China from Monday you can see it says the world's biggest contract chipmaker acts to ensure it is in line with us restrictions on Chinese access to latest processors and basically the stating here and if you don't know who tsmc are they basically one of the most important companies in AI they produce you know basically all of the chips for NVIDIA they basically produce like I think it's 80% of the world's AI chips I don't know it's something pretty crazy but they're basically suspending production of their most to advanc AI chips as Washington continues to impede beijing's AI ambition so you can see here that there is certain restrictions on what China is able to do when it comes to getting chips from tsmc and you can see here that tsmc literally said that they've told Chinese customers that they would no longer manufacture AI chips at advanced process nodes of 7 nanometers or smaller and basically future supplies of semiconductors to csmc Chinese customers would need an approval process likely involving Washington now this is pretty crazy because of course right now

### Segment 2 (05:00 - 10:00) [5:00]

tsmc is doing this thing where they've got this extensive you know export control system but the crazy thing about this is that like this is not the only company that's facing this but some people okay and we can see here like on the leaderboard there's actually this model and if you look at rank seven which is essentially I guess you could say one probably rank four you know considering it um we've got this model called ye lightning and this is actually by this guy called Kaiu Lee and he's actually working with these restrictions and he's stating that look whilst yes previously training runs used to cost billions dollars even though there were certain restrictions companies like his in China are still managing to race ahead 01 now has is the third best model modeling company in the world ranking number six in models measured by LM CIS and UC Berkeley but the most amazing thing I think the thing that shocks my friends in the solic valley is not just our performance but that we train the model with only $3 million and GPT 4 was trained by 80 to 100 million and um GPD 5 is rumored to be trained by about a billion dollars so it is not the case we believe in scaling law but when you do excellent detailed engineering it is not the case you have to spend a billion dollars to train a great M as a company in China first we have limited access to gpus due to the US regulations and secondly the Chinese companies are not valued what the American companies are I mean we're fra we're valued at a fraction of the equivalent American company so when we have less money and difficulty to get gpus I truly believe that necessity is the mother of in Innovation so when we only have 2,000 gpus well the team has to figure out how to use it I as the CEO have to figure out how to prioritize it and then not only do we have to make training fast inference fast now of course it isn't just Eric Smith saying this I'm sure you guys remember the video I recently made where I discussed the fact that Congress is basically pushing ahead for a Manhattan style AGI project you can see here that the Congress recommends that you know in order to stay ahead of other countries they're basically stating that look they need to start providing these kind of things for current US companies and ensure that they manage to get AGI under wraps before other countries manage to get to it first and of course in that video I spoke about how you know in the article they're talking about China is racing towards AGI and it's critical that we take them extremely seriously so one of the big things that people are talking right now is of course you know with the models like deep seek the one that's currently doing better than open A1 preview a lot of people are wondering why are these models open source and if this trend continues what are going to be the kind of problems that we will face in the future and that's what they actually talk about in this podcast they're basically talking about that look it's probably not even China that's going to do anything wrong with this kind of Technology but malicious actors out there when you're able to just have a system that manages to Ace these benchmarks like future models will what are they going to be able to do and this is where we see Eric Schmidt talk about this and we're obviously you and I are not in China and I don't know why China made it decision to release them but surely evil groups and so forth will start to use those now maybe they don't speak Chinese or what have you or maybe the Chinese just discount the risk but there's a real risk of proliferation of systems in the hands of terrorism and the proliferation is not going to occur by misusing Microsoft or Google or what have you it's going to be by making their own servers in the dark web and an example a worry that we all have is exfiltration of the models I'll give an example Google or Microsoft or open AI spends $200 million or something to build one of these models they're very powerful and then some evil actor manages to exfiltrate it out of those companies and put it on the dark web we have no theory of what to do when that occurs and then of course this is where we actually see conhi talk about this and this is a guy that's focused on AI safety and he actually talks about how open source AI is going to be a really big issue AI is a problem as bigger than ukes if you just build bigger and bigger B bombs and anyone is allowed to own bombs eventually everything blows up that's just a fact about physics like this is not an ideological statement so like there's a reason that nuclear bombs are not allowed to be produced or owned by private individuals and hot take I think this is good I think it's actually great I don't think that the blueprints of the F22 fighter jet should be open sourced I don't like I come from an open source background which is very libertarian very accelerationist and when I got into it you know from like you know like Linux and like you make software like free and open and like secure so you can like help people and all this kind of stuff what I found quickly is that a lot of the people not everyone of course but like a lot of people open source don't really care about making people's lives better they just want to be allowed to do whatever they want and they don't want anyone getting in their way so I remember when for like the first deep fakes generators started like being created and like the first like voice cloning software and I was like hey that's kind of up like maybe we shouldn't open source that and people were like what are you against Free Speech it's open source so it's

### Segment 3 (10:00 - 10:00) [10:00]

good and I just think that's a very childish view of reality