Creating films, uh, when people say, "An hour and a half can be created," so before that, like, uh, the Moon.
The Chinese will win further.
The open source of their models is just, uh, the policy of catching up.
Yeah, I think we're really far away, far away from that.
I wonder what you can't, for example, evaluate, uh, the results of AI-based education tests. What's interesting.
Once again, uh, when I'm dealing with personal data laws in Europe, I think if all companies that work there would have been holding this law, Europe would have lived about you-- like a thousand years ago. There's a medieval in technology. Hello, everybody! Ah, we're on ToTheMoon. Technology sites, discussions, news from the Silicon Valley and around the world. Let's go. New edition. We're going out once a week. It's Sunday. The subject for start-up, um, I think it's good about, uh, video-generation and video-building.
And with the fact that Sasha, you and your project in this area, I want to start and discuss, uh, what's going on with the video generation. Last time we mentioned, uh, that, uh, TikTok, uh, put out his internal tools. And it feels like all the social media will be putting something in their internals. And China has shown breakthroughs in recent months, including in general in Generative AI. Uh, what do you think, um, proo, um, about the model?
And that's, for example, the Kling model that China presented as a competitor Sora.
The most interesting thing to me is that, uh, Kuaishou has released his model Kling. I wanted to dig a little why I was the most interesting. First, Kuaishou is a Chinese company that makes a competitor TikTok. And their Kwai app is probably the second largest TikTok competitor in the world. I think Kwai is better than YouTube Shorts than Instagram Reels than Snapchat et so on. If, um, if you remember, uh, I don't know, three--- four years ago, Kwai in, uh, emerging markets on the p-- on, on an equal, uh, uh, s--- competing with, uh, TikTok. In Eastern Europe and Latin America, they often occupied the first places in the mosquito. I mean, it's a huge company, they, they do, uh, I think it's over $10 billion, uh, rew-rev-- rook- revenue a year. And they still have their CapCut. So they have their own versa-sup. On the one hand, it seems to a lot that Cap-- in CapCut is not complicated and unique, but if you dig deeper and see, that's just that some companies have been able to make their own quality video broker, then What's the answer? So disgraceful that none of the western companies, either YouTube or Snapchat, neither Instagram nor Pinterest, were able to make their video-reviser a quality. I mean, uh, Kuaishou, in principle, in, uh, in a world of social media, say, the second player in the world after ByteDance. I mean, they've released a video model on video generation. From my point of view, this is the best model in the world because they have very high quality, uh, realism quality and a fairly high number of movements. I think that's why I think that's the best model in the world. And, uh, in principle, uh, all social media companies have to spend $100 million--$100 a year to provide, uh, diversity of content to stimulate the promotion of Creators. I'm guessing that all these social media companies look to the future that, until the end of this decade, 90 per cent of the picels, sgenerated, will be sgene-- 90 per cent of the picels will be generomed with AI. Iia, well, by the time Kuaishou is the leader in this direction. Now the spoon is cheap. The stitch is probably the five-six-second video generation takes minutes for everyone. And it's probably very, very high-end videotape. So far, actually, uh, GPT-3 is not yet a video game. It is not clear whether, um, videos will be developed in the same logic as LLM, because even with such weak models, five to six-second videos are about to take minutes.
But, however, it looks like a cho, The LLM will not be the same as the LLM.
What's that, uh, what does that mean? Is that what it is to do to "do not match LLM"? You mean that it's gonna-- it's gonna start really hard to get out of here, get out of here, uh, wait, or what?
Well, in LLM, that's when everyone decided to rock the models. So they were doing as a architect transformer, like GPT-3, right? Then, in principle, around GPT-3 built--- a number of start-ups have built businesses. Iia OpenAI first added tuning, ChatGPT came out, and then in principle many companies are already rocking the size of models and getting better quality. We haven't seen tuning for the pictures yet, and we don't know if it's gonna be scaled up and if it leads to better results. Here. And maybe, on the contrary, the video will be needed by little models that-- maybe another architecture will be needed, because six minutes to wait for a generation for a six-second video--- that's-- I, uh, I, I, uh, I, I, uh, I think it's not good for sosh-- well, for cryuters, for example. I mean, it might be good for, uh, social media that can generate a pack of, uh, hundreds of thousands of videos to put them on their platform. But, uh, to create a content, I think it's not applicable to the moment.
Sasha,
does that mean, uh, what you're saying now is a problem, uh, time, uh, but as far as I know, as soon as there's more power, more power, and, uh, video is gonna be faster and easier. Or is that not true? Or, for example, more-- or once you-you-go to sell, where there's a lot of people, is that gonna be a life story?
It's really hard to tell. I mean, uh, it means that it has to be understood that there's a percentage of how successful the video is, I think it's less than 20 percent, even for people who, even for people who are good in prompt engineering, Yeah, like, a middle level like me. I mean, uh, and I think that's a very long way to be, it's applicable. I'm not here yet, I don't see what-- I'm not sure that this, that is, this architecture like the Wall model, the diffusion of transformers is what we do in our start-up, either. I mean, this architecture, it's obviously giving-- it's like a big break compared to what happened last year. And in principle, we'd have the first to produce our model, but now three more companies catch up. But as long as it's in terms of the velocity of generation, it's enough, I think, as long as it's still far enough, it's , so wide skies.
And you remember,
we were just about a few episodes, like, five or six years ago, just checking out the mathematical ones that if 30 personnel per second, a minute video is 1,800 pictures that need to be. Sgenerate. And, really, this is the way with the generation, yeah, it turns out that, uh, uh, uh, well, here's the Midjourney, how many pictures you have, in fact, you're getting a little bit of a picture, and you're gonna get a minute video of 1,800 of that. I need a picture. Yeah, well, then, uh, Sasha was sending that there are anchor images that need to be generated. I mean, he's a similar order, it's not gonna be 1,800 paintings, but we're gonna have a really big hug. So, Sas, what you're saying now is really confirming, yes, here's our calculations, which were roughly, that the technology is not ready for a wide application. Computer, that is, power isn't enough to use chat-bots with the GPT almost free.
Yeah, I think we're really far away, far away from that. And there's always, uh, like-- like in science, yeah, that's the standard principle that there's exploration and exploitation. I mean, exploration is when the testiru-- how different, roughly different architectures, different parameters. Exploitation is when, uh, uh, when, uh, rudely speaking, everyone's trying to get out of the max, right? So now, the transformers are going to be e-stamped as exploitation, that everyone's more and less learning the same models and trying to get you to do the most business value, yes, they get crazy assessments, and so on. Uh, well, OpenAI, okay, from 2015 to 2020, was like a exploration stage when they found what they were doing. With the video generation, it's clear that, uh-- it's clear that the explo-- well, what's going on now is a phase of exploitation when it's a little bit of a...
It's exploitation when, um, a number of companies are putting their stakes on this transf--- this architecture transformer and raise hundreds of millions of dollars to rock it. Noo, again, scaling the model, increasing the number of parameters, only leads to longer generation and, perhaps, actually, correct, as in science, now, anyway. We're at the explo-- e-, in the exploration phase, yes, from a technical point of view. But how capitalism would encourage companies that have raised a lot of money to move on with a lot of money and to exploit that approach. Although the approach may not be right, right? I mean, we're here, we don't know that yet. I see, on my side, that there is a tweet-- that is like a VC, that is like a I-Attists, they're kind of super excited.
If-- and as for, um, regular creators for agencies, the level of awareness, optimism about videogenerative AI is very low.
What's interesting is, uh, central processors were the first time, right? And all the calculations were on them, and they were graphic, too. Then when the task, uh-a, for the graphic processors, was assigned to a separate device, in fact, there was two, yes: there's a central processor, there's a GPU. And a similar story, it seems, may happen to other specific calculations. Because the news was that, uh-a, the device that was planted under the transcript calculation. I mean, uh, GPU doesn't care, uh, general-purpose device. So it's got different calculations, so you can do something with the schedule. And if transformers become the most popular architecture for neuronets and work almost everything on them, then there's a separate device under them and possibly a breakthrough in energy efficiency, too, in the field of energy efficiency, speed, something else. Uh, there was news that even a start-up that was making such iron. I'm gonna add a reference to the skin, so I guess I can add a-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-e-h. But it looks like it could be, uh, so logical development, right? And when some equipment comes up that can speed up and cheapen the transcripts, it may also be viable.
It's really good, and thank you for bringing it up, because, after all, I'm really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really good at that, uh, start, like, a pop--- famous, uh, Exit fan-fandreis, I think it's called. They raised 120, they raised 120 million under, uh, chips that were specially plucked under, uh, transformers. There's also a c-- another Groq company. I-- but I don't see the wide application of these specialized chips yet. It's still-- it's not a thing yet for transformers where the business value of transforming is still measured-- well, how the transformers are used is still an industry on, there, Tens of billions of dollars. Most of this is, of course, like a consulting, but total is tens of billions of dollars and a-- a year. And yet, there's no special-- specialized hardware here.
If we're talking about the video generation, then, in abrupt way, Netflix is spending on-- well, that's, like, a single--- how much the video-generation market is much smaller than the LLM market, right? I mean, it's a conditional social media, they spend at best hundreds of millions a year to produce content and work with creators. And yet they're gonna need to work with the creator, because, well, c-creators are still, uh-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a Aaaa, as the order that Netflix spends on the content, it's in the area of dess-- it's dess-ten, tens of billions of dollars. I mean, and that, and it's like a market that I think, well, Video Generative AI, there, in the current, uh, view at best, can only eat 0.1 per cent of this market. As long as the technology is not good enough, and the video-generation is very behind us, very much behind LLM. And what's not even in LLM yet caste chips, tells us that we don't need caste chips in the video, and we probably need to find more effective soft-shot solutions, because so far the econ-- Because while the video-economic contribution is very small.
So, movies, like we said, uh, movies, uh, when people say, "Soon, you can create a movie in an hour and a half." Ah, so it's like, uh, before the moon. Ah, so, uh, up to the moon 500,000 times. Oh, I think in the video, of course, one of the first serious applications is a personalized advertisement. Probably where we can feel something really more or less working. Although, well, it's clear that it's gonna be a social entertainment. So, in social networks, this topic will be used. It's definitely gonna be used in some kind of thing, yeah, some stuff. Noo's technology is needed anyway to keep working with the current video we already record, put some masks on them, changes, and so on. It's cool anyway that this market is developing. Speaking of China.
Since you're saying that in China, you've done, uh, a cooling, a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a- Uh, there's some news that, uh, Chinese open source, uh, almost won among everyone, well, it's, uh, it's, uh, it's, uh, it's, uh, it's, uh, it's a certain rating, yeah, it's a question that's always someone who's studying and who's counting. We recently saw clients and users of X as the best models call everything that Ilon Mack does, so, uh, China released, uh, uh, from- to open source its own LLM. And if a year ago everyone said that there was something going on in China, they probably had something and they probably had some chips, it's getting clear, at least TikTok, with this one. Models. Uh, Kling is what it's like, Sasha, the same thing-- that's different, right?
Different producers than TikTok had a model.
Yeah, of course, it's all different producers. And thank you for bringing this up because I think it's very, very important to talk. It is clear that in American media this is not a topic-- that is a topic that has not been given sufficient attention. Ah, but it's just like the truth. So, in a video-generation, Chinese models win in an open source. They're much worse than the open source of the model, but still the most dominant models, they're k-- they're Chinese. The leader of the wrestler is Chinese too. This is a video I can tell you. The Chinese are also in the generators. We can't-- we don't know how good Google and OpenAI is, but from what we can try, it's clear that China is leading, yes, in the video. Now, if we're talking about LLM. I mean, uh, uh, uh, well, LLM is a more transparent situation because there's a, like, a-a-a-a-baccmark and a-and-and-and-and-and-and-and-and-and-and-and-and-and-and-and-and-and-----and-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- And now, the best model in a world with a big stock, there's more than 10% of it, yes, as relatively as it is, it's a Chinese model. It's like Llama after, but it's clear that Chinese people in the open source are somehow winning with a big, big reserve. And you need to understand? That with pain-- with a high probability, the Chinese will win further in closed models. Let me tell you a little bit about that. Uh, maybe it's interesting. I mean, there's a way in transformers-- that's what-- that's, that's, uh, that's, uh, let's just take a little distraction. Where will the transformers go and who wins, yes, hypothetically? The first line is a new architecture, there, a-- State Space model, like, or something. But we're not gonna talk about it yet because it's a resurrection and the sub-metros of the State Space model, they're not yet circling the transformers. So now that transformers are, they can be increased in size, and many companies have done it successfully. But we have to understand that if the proo is still in the process of implementation, very big models, they're gonna be, well, there's no trillion of parameters, and dozens of trillion parameters. They'll be very hard to launch, and it'll be very expensive. If it's about models that are-- if it's about another direction, it's what it's called, uh, how much of a start-up training is that the model in the launch process is like, like, a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a- and adapt. And it is here to make a successful model in this direction, it is necessary that the model be integrated so as to be integrated into-- that the model is to be used extensively. I mean, from this point of view, of course, the Chinese will win sooner, because they already have-- well, that models are already integrated into WeChat and so on. So they're integrated into these super-apps, right? And I think that a high probability of Consymer LLM, not enterprise, in Consymer LLM, the Chinese will win, in open source LLM will defeat Chinese, in LLM enterprise, maybe capitalism will be suppressed. Well, anyway, CEO Antropic last week said he was sure that in two or three years they would train a 10--- that would cost tens of billions of dollars. One model practice. Net GPU
One model is purely GPU. And I'm probably gonna have to let OpenAI and Google Meta do it. Noo may be, they may be, with very large models for enterprise use, a case of capitalism, but it's not likely that the koo in, uh, consumers are unlikely to be a k-- American, like the Western companies
win.
Well, that's what it says that if Chinese people in open source put a super cool model out of it, it's more like Americans' ears, it's clearly said that Chinese people don't have a open source of all kinds of peasy. Right? Well, judging by the way things usually happen. Does that mean that it means that now, in parallel, they're gonna be doing 10 different models?
I wouldn't have come here to comment on my part because Anthropic, OpenAI has been a long time since nothing is open. So, like this tasbord, which is about open source of the model, it does not include models, that is, it, uh, does not include, uh, the model from OpenAI and, uh, and Anthropic. But maybe we should see, uh, maybe, uh, maybe, uh, we could watch a dashboard that doesn't include an open source of the model. I'll try to find it.
It's really very-- well, that's when Huawei had a good videotape, and I think it's, uh, with a NVIDIA. And then, I think that the question is, uh, whether they're going to optimise, whether they're going to optimise, uh, under, uh, specific architectures. I mean, if they go, well, they can, of course, get, uh, a H100 and B200 videotape. I've checked now like a dossbody with a prophetary model, that's a prophetary, well, here, well, here, I'm gonna say, I think we should probably make some kind of a footnote, which is... if we're still going to. I'm saying, if we're talking about, uh-oh, that's-- that's-- that's, uh, that's the evaluation, I think it's basically a statistic, Chinese. And that's why we're-- and that's why we're in dashboard now, uh, the best model is GPT-4o. Noo as Chinese model, Chinese models, they're not much behind, they're definitely Naa, they're exactly at level, the pre-existing models from Anthropic and OpenAI, even in English.
A language that is not key to them.
The open source and not open source are interesting. So, Sas, you were just throwing the news, uh, Zuckerberg, who was also telling the proo here, what the AI was supposed to be, which is why I can't be there from the same company, which is likely to be A lot of companies. And, accordingly, they're in the open source, saying that they shouldn't be closed and so on. And that's very interesting. We've been discussing in one of the episodes, too, what is the open source of our models, a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a- Uh, when the OpenAI wasn't, uh, first of all, they were open, and their models were published. And that's the way to become more famous, you had more users to get to know you on the market and so on. Once you're at the top of your position, you're not opening this up, because otherwise the whole market will be using your accomplishments and how you become a closed company like OpenAI, Anthropic, like, uh, Microsoft-- Google Gemini, too, is not open. And the ones who chase, too, are Zuckerberg with their Meta, they're gonna be bragging about what they have, uh-a-a-a, all in the open source, all open. And now they're starting to catch up. I wonder if when leaders change, the same system will happen or not? So if someone is going to push the OpenAI, they're going to go back to, uh, open source or still continue their own policy, which they're following?
If someone really beats the OpenAI, then the OpenAI will sell to the biggest player. In this case, it is likely that Microsoft. Microsoft, yes! And it's fast-tracking, so it's gonna go some other way.
Anyway, when I finish with China, I think that, uh, the sign that in the open source we see cool models, the sign that they, uh, make very cool decisions in the video generator, that's-- that's the sign that we have the chorus. They have closed super-cruise systems that we just don't see, yes. And, uh, obviously, there are projects, well, super-high, uh, class. Ia will see what happens. Well, the other thing is, what kind of thing they'll be spreading in which countries. And the proo, the e-e-- and who's gonna let them go, yeah. It's like the States don't have Chinese electric vehicles. And as much as everybody feels good and lives well, they just don't drive here. Uh, just like Europe said that, uh, they'd have a huge amount of restrictions on access to AI. And the fact that Apple has already spoken, which is likely to be unable to deploy its AI in Europe, and there will be a big problem.
To be honest, I, I'm just once again, uh, when I'm dealing with personal data laws in Europe, I think if all companies that work there would have been absorption of this law, Europe would have lived around. You're-- like a thousand years ago. She'd be in some Middle Ages in technology because it's impossible to keep everything there. So if the client's registered, you have to get a written consent, and in a few years, you have to, uh, re--- ask him, in writing, it means he's in your database. I was. I mean, the idea is that all the personalized commercials should disappear, all of them, well, everything should disappear, yeah. And one of the most global advantages is the existence of your data in what you get personalized advertising. And with the technology that is now available, it's incredible that improves life, not worsening human life. I mean, I'm very happy that there's a huge amount of, uh, my data from various providers, and they're making different recommendations for me, they're giving me a bath.
Well, it's still mutue, but still.
I wanted to comment on this subject, which, uh, yeah, Eurosus, he's, uh, the only one who's ever adopted such a serious law, uh, with a wide range of applications. But you're not right about what, uh-oh, there's a total ban on personalized advertising. It's not prohibited. They first identified the risks and categorized them into four categories. I mean, there's something there, uh-a-a high-risk, medium and, uh-a, that doesn't come under this law at all, is with little or no risk. And the low-risk thing is straight, that's some kind of total bullshit, like, postal spam frits or, uh, video games. And what you're talking about personalized advertising is the fundamental principle of transparency. I mean, roughly, even if there's any video relays in the future, it's got to be something like watermark, which is, it's supposed to be written, you know, on the basis of, like, uh, uh, uh, If the advertise cigarettes, it's supposed to be written down that smoking, there, is harming, like this, kills. Uh, and on the same principle, there should be a promotional project. Uh, it's just supposed to indicate that the data were obtained by artificial intelligence, but it's not prohibited. Uh, high- and high-risk, it's kind of human rights-related, uh-uh. Uh, i.e., you can't rank people on the basis of race, some kind of national affiliation, skin colour, etc.
Uh, plus, um, I wonder that you can't, for example, evaluate, uh, the results of AI-based education tests. What's interesting is that is, in my view, the opposite is more honest, yes, the human factor is excluded. And from their perspective, it's, uh, not very honest, because the human factor must be present and how education should be made publicly available. And plus there, for example, if it's, uh, medical or transport, they're saying that there's gotta be, uh, a human factor. So someone who's watching is allowed to be the surgeon who's doing the surgery, you know, uh, AI, who's doing the surgery, uh, or there, I don't know, a plane that's unmanned. If it happens, a man must still be present., and it's kind of interesting about this law to have power distribution, because the softest regulation in China is. And it's funny, because, for example, it's kind of funny for me to see a guy with a legal buckle-bag, like, what it says is that, uh, it's a law-brainer--
this sec-law regulation must be in line with the principles of socialism. I mean, well, I wonder how AI can be consistent with socialism. Here. But it's the softest, unlike the case, in the Eurosuit, the most violent. But now, there's a lot of noise, uh, about the adoption of this new law in California. But it must be understood that, of course, different legal systems and that there is still more case law in the States. So they rely on court decisions more than on laws.
Aah, but still, it's a feeling that many people have. I've been looking through a lot of disputes over the U.S. getting on the way to a more tight regulation. Ah, and, uh, and, uh, and, uh, here's the risks I've listed, uh, that in the Euros are being addressed, and the risks that are, uh, in the American bill are very funny, because, uh, it's actually called, uh, it's actually called, uh, it's actually called, uh, it's a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a- Another category of risk, just like these Terminator movies, that AI will invent some mass-hazard weapon, some biological attack has been made on people. I mean, well, whatever it is, a law that feels like was done by more politicians and how they used such relevant concepts to attract, uh-a-a, attention to him., but there's a lot of other hard-core certifications, legal restrictions, which, you know, put a very heavy strain on start-ups and small corporations. Accordingly, many people fear that this will go as much as China, there, or some countries with less stringent regulations. Well, that's a short one.
Interesting, in fact, the subject, especially what you say, uh, that you've already gone through topics like, that's what you've given up on education.
In fact, the whole world says that everything, the lack of education, is not necessary for people, and that means everything will be generated by AI and tested by AI. And yet, a zone like that, it seems that it is technologically feasible to be resolved, is knowledge verification, yes. It doesn't have to be solved in terms of testing one test. This is a large number of different parameters, and the attendance of a person, his involvement, his unique characteristics of what he does, his analysis from different sides, from a different perspective. The perceptions of different nations, their involvement in different interests, there. Well, there's a lot of things you can check on the machine because a man can't check so many parameters. It's too much work in the deep, it's nobody's gonna rule out a man. And here, in fact, the prohibition on such verification, on the contrary, is returning very much back. Because, actually, now, a lot of things have been automatically checked, yes. It doesn't mean that now in the Euros, uh, all universities do checks only in a way, uh, human. They've been doing a lot of checks already technically. And that's the side, it's very specific. I think it is so complicated that it can cause a State to become a problem in terms of its economic development. It's like e-commerce. I mean, basically, uh, not e-commerce, but social networking. So, in fact, social networks, when they were created, were not regulated. Uh, there was no regulation because the state wasn't ready to produce such things. Social networks say, "So let's regulate, we'll work under it." The state can't still regulate normal. There are endless discussions, and in the end, we can't take the wrong decision, and just lose the crazy, big, big market, yes. And we now see this from the standpoint of sanctions, perhaps in different countries, well, or limited, or there should be a long strategy. That's how China is. China, if it takes something, takes it long, yes. And they clearly understood that if they limited Google, as long as they were, I don't know, 15 or as many years ago, too, well, they probably knew clearly what they wanted and what they were going to do. They understood what would happen, what would be problems with access, not access, who would use or not. We are willing to create our own systems or not to create. It must have been one of the possibilities to create a unique, unique player in the world on the Internet, yes. Still, China is a crazy player on the Internet, uh, with the US, yeah. I don't know which one is cooler. Uh, well, technology suggests that China is more expensive than the US, uh, but-- well, it's estimated. Uh, and, uh, strategy is very long. And that's how you make decisions?
Because still in California, maybe they made these decisions, assuming that all AIs are here, right?
Well, they say that this would lead to a leak, either to other states or to other countries. Well, that's very interesting, Sasha, you noticed, these parameters you've given, that's more about social ranking, yes, that's, that's, people's assessment, their social status. What in the European Union, there, or in the States, I think would be in violation of the human rights concepts, well, the Constitution. And in China, it's a normal practice. I mean, I think there's even something in there if I'm not mistaken, a social score for everyone, and they have, uh-e-restrictions, they're not really in their regulatory system. I mean, that's very interesting. China is really much-- they're straight, well, as they put it on, on its development, there's a lot of freedom.
No, look, well, China was approaching, uh, at least those people who work there, maybe it's, uh, micro and medium business, small and medium business I know, it was an approach, that if you were. You work in China, please work, we'll give you a lot of resources and everything, we'll support. Yet there is a technology market that is very strongly supported by the State. He's just sliding with the State, yes, very much, but at the same time, it's a separate business, but he's very much in support. I mean, if you take what Alibaba was like, there, uh, ten years ago, 10 years ago! Well, just Amazon smoking in the side, there, compared to-- or there, weChat, yes, that's what they could, that they're doing-- and what they're doing, those ins-instruments, the services that are available there, Those data tests, those soring systems, credit ratings, human ratings, yeah, what they have. Another point is that it is a problem and a complex one. I see why, uh, some things are being slowed down in the U.S. In Europe, of course, I have a feeling that they'll have big, uh, if they, she, will be very much blocking AI. Although, as far as I understand, Sasha is saying a lot about it that there are good players in France, yes, in AI, Sas? Here. And-- yeah, yeah, yeah. It's really good. What if-- yes, and it probably suggests there's probably some kind of lobbying and not closing AI. But if Europe is still closing AI, the too-- although they have the opportunity, yes, they still have access to, uh, equipment. Because there are still countries that have a hard time with it, right?
They have no access, for example, to buying equipment like that, and they'll see. Well, I wonder what's going to happen with, uh, equipment. I mean, if Huawei, like Sasha said, is really going to do something about the new chips, uh, NVIDIA, but it's just a huge resource for China, just a huge resource. for China. And we'll see who else gets out of here, uh, in terms of, uh, everything. Especially with the fact that they have a very strong business with regulation. But there, I think development is different. It feels like in a year, of course, China has been totally unbelievably stunned in this part. But the int-- very interesting about the law, what you're saying.
And, uh, if anyone else is going to study, watch these things, see what's going on for a long time, I wonder if they'll accept and what they're gonna do, and that they're not gonna change, are they? And how people will respond to it. And, of course, the question of education for people. I think people lack education in the parts, uh, personal data. Too many people worry about their personal data without knowing, uh, where, uh, there's a man, for example, worried that, uh, some system would have access to their perspainted. Although I'm saying, you're also writing the mail in a cloud storage facility, you've got all the mail and all the raw data, they've got everything on their phones, everything on the phones, everything in the clouds, right? I mean, the question of advantage. I don't think a man understands the advantage that might happen. This is a movement, a piece that can happen, a uni- unique simple, unbelievably unique, isn't it?
Yeah. It's just, you know, the legal system, it's probably the most stable and the most, like, well, it's hard to move, to react. Aah, and only in Russia, for example, I think it's still, uh, planned for four years to set up this code, regulation, and... But they're just gonna get some basic data, uh, he's gonna get some bony bony.
Trying to do some bone. So I went far. It's like a race like that, you know, that's what it's like to know that the legal system doesn't catch up. And, uh, it feels like there's a first phase of completion waiting around the world. I'm gonna have to make a little more of a story. But I don't know if any such step ever comes, because technology, well, is much faster than the legal system, and it's more simple for, uh, case-by-case systems just like the U.S., there, Great Britain, but not in Russia or, there, Germany, and
so...
It's like a transitional theme, if a car is on the street that, um, self-dr- self-driving, yeah, that's who's going to San Francisco, that she's on her day, there, and she's got some data terrabytes, and she's on her way. It's totally seeing everything that's going on in the street, and it's a way to analyze, then analyze it to improve the city, there, water leaks, I don't know, some problems, thefts and useful things, right? W-- completely different levels of utility. And here's the question always like-- what are we looking at? We look at it as a utility or we look at it like, uh, a problem. Because, of course, for 99.99 percent, it's a good thing. But, really, someone can get under a hard pro, especially when their personal data were stolen. I don't know who's looking for, but I'm at least once a month from some cool service getting a notice that they had, uh, hacker in the database and they stole the data. So be careful. A Ticketmaster operator just sent a book, the biggest salesman in the US for the sale of electronic tickets. And they wrote that they leaked the data with the credit cards, yes, in full volume. Ia, uh, they say we have a security system, services, and you'll get some money. But it's clear that in the States, if you've been stolen from your credit card, you don't care because you have insurance in the bank, yeah. Well, as much as it's mostly there, yes, credit, yes. It's hard to understand, uh, like, the Russian-speaking world. Yeah, here. Like they stole whatever it was, I'm gonna get the bank back instantly., but that's what it is, it's always a question, and it's just a matter of perception inside. But it always comes with such messages. I mean, it's-- it's gonna be, uh, I think we're gonna have these hacker attacks and some different stories, it's just gonna grow up and grow up, and it's-- it's a different, totally-shaped thing. Because literally, there was a recent attack, and there was a writing about the hackers, hackers from the Russian government. I was at Snowflake, I think, right? Hey, not Snowflake, Remote. Who's Remote Soft doing? And, uh, uh, TeamViewer, huh? Here. It's a big deal to resist such stories. And finding this line between all this is very hard. By the end of the release, what I was going to say. Sasha, did you find the comparisons?
I know what you've been looking for, uh, models, uh, Chinese and American closed.
I mean, what I said was that, uh, uh, uh, what I think is, this b-b-- other baccalaureate, he-- it's mostly in English, and Chinese models have comparable models. Level as the pre--- as the last generation of the OpenAI and Anthropic model. I mean, even in English. And in Russian, respectively, with a high-probability-- sorry, and, uh, respectively, Chinese with a high probability of poetry-- they're very much ahead. In China, as much as a benchmark in Chinese, I don't know. They're probably there, and there's probably Chinese models--
Well, it's obvious, it's clearly about the volume of data they have that are covered, for example, yes. And what they can learn, what kind of data they're wearing. It's interesting, it's a question, and they have health, for example, a lot of technology, there, and the State has a huge data set. In China, there's a huge data set. I think it's probably the biggest data, uh, actually, normalized. Normalized. And one of the problems of the model is the normalization of the data. And in China, the black-- population is crazy, with great technological change in data normalization. Thus, the quality many of their models can be just space cooler, space cooler. I wonder what's going to happen. Thank you, everyone. Until the new editions! Technology sites, um, discussion, news, cool moments from us four., the Silicon Valley, the US, uh, the world. Uh, before the new meetings. Bye, everyone!