Skip to content
Transcript

Transcript · 053 · Llama 4 Is Meta’s Weapon Not Because of Its Size, but Because It Can Be Built Into Anything — ToTheMoon

English machine translation of the Russian-language episode. Timecodes open the source video.

Episode overview
00:00:00–00:00:46Welcome to ToTheMoon
Alexander Volchek00:00:00

Hello, everyone! We're on ToTheMoon. Technological news, Silicon Valley sites around the world. Ah, fifty-third edition.

Alexander Mashrabov00:00:08

This is really happening in corporations.

Alexander Volchek00:00:10

Is that, uh, a desire to confront China or why is that happening?

Mentions: China
Ilnar Shafigullin00:00:14

Once they reach some point where they're clearly stronger than the rest of the market, what they're selling now, they'll shut it down, too.

Alexander Mashrabov00:00:21

Oh, all these companies, I think we need to ensure America's leadership. I think that's the point of a Stargate project, right?

Alexander Volchek00:00:28

Would China have a quick love for open source?

Mentions: China · Open source
Alexander Mashrabov00:00:32

It's such a big step forward that I was surprised in a pleasant way.

Alexander Volchek00:00:39

It's a very interesting and-or-historic about TikTok. It's unprecedented.

Mentions: TikTok · Llama
00:00:46–00:03:33Updating Llama 4:
Alexander Volchek00:00:46

Look, I want to start, uh, ask. So, yesterday, uh, Mark Zuckerberg announced that there was an Almada Llama 4 and in Llama 4, he had Llama 4 Begemott, some space, two trillion parameters or even a number of parameters, I'm not sure what's going on. Yeah. Somewhere in the descriptions, someone read how Llama 4 is really something incredibly cool. I understand they're not yet released Llama Begemot, and she's going to be out by the end of April, as I understand.

Alexander Mashrabov00:01:31

What's with the nemming?

Alexander Volchek00:01:34

Here.

Alexander Mashrabov00:01:35

Is that any sense?

Alexander Volchek00:01:37

No, they have good Llama, Llama-

Mentions: Llama
Ilnar Shafigullin00:01:39

Something big.

Alexander Volchek00:01:40

Yeah. Here. Or Maverick. Okay. What do you think of Llama 4? Is this a renewal, is it more because everyone needs to get up or is it really something unrealistic? Because when he was last night, I was watching this one in the social media video, he said about these two trillion parameters somehow that it was a space-- unique thing, and nobody else in the world ever did. Is that true or not? Sas, here's Moshabov, you must be here--

Mentions: Llama
Alexander Mashrabov00:02:10

Well, it's probably the most important thing to say, yeah, to start, and then, like, first, right? I mean, I can't figure out why they made a release on Saturday. All PR's rule is one, that a release on Saturday means you want to bury a release, you have to say, but you don't know how to say it.

Alexander Volchek00:02:29

Alexander Mashrabov00:02:29

You don't want to read questions, and you're hoping that Monday will be followed up and forgotten about you. But I'm gonna say it because I've been working on something. That's what's really going on in corporations, right? And what I heard from Meta that... That I heard from Meta that staff inside Meta are many who worked on the current generation, that is, Llama 4, they were supposed to be released in a very short time, because DeepSeek R2, which will be, Well, the other level is perfect. Plus Open, plus OpenAI will blow up his, uh, reasoning model in open source. I understand why this happened. Oh, that's cool, their model hits the baccalaureate. Oh, I don't know what this will affect the world, but it's cool that there's competition in open source and that Meta's not giving up. Because from American companies, in seriousness, the open source is now only competing for Meta.

Mentions: Llama · OpenAI · Open source
00:03:33–00:05:12New OpenAI model
Alexander Volchek00:03:34

Look, what Sam Altman said about the open version, is the big language models from the time, actually, there, GPT 2, right?

Alexander Mashrabov00:03:43

Yeah.

Alexander Volchek00:03:44

And how much, uh, is this a desire to confront China or why is it happening? Hold our canal, put the liquor on. This will help us build our classic independent videos and our interesting content of sites, technological sites from the U.S. Silicon Valley and around the world. Our great quarter will help move forward.

Mentions: China
Alexander Mashrabov00:04:15

I think that in terms of economic, a political, political, economic agenda in the United States, which everyone agrees exactly that AI is... the main way to increase GDP growth. I think that this high level of decision has not yet been decided, it will be through an open source or through a close source. So I think all the sub-contractors, and, uh, everyone involved in Stargate, the sub-steering companies, yeah, in a key, right? I mean, it's not a private initiative. All these companies, I think, need to ensure America's leadership. I think that's the point of just a Stargate project, yes, to ensure America's leadership in AI. Well, including the open source of the model, I understand that. And without Stargate, there would be no OpenAI round of three hundred billion.

Mentions: Open source · OpenAI
00:05:12–00:06:462 trillion of parameters in Llama 4
Alexander Mashrabov00:05:12

That's understandable, too.

Ilnar Shafigullin00:05:13

Yeah, I'd add a few more comments. First, I wonder if we have a little gap between the video and the video. And, uh, so far, we're getting a while at the recording, and they're saying that Llama 4. Literally, we've been on the other side of the day, and today is our past video, where we said Meta has refused to blow up a new model. And now they're, uh, waiting till Saturday, couldn't stand it any further, and they're all knocking off what they've done. What Sasha told me. About the trillion parameters, huh? Here, aaaa, clearly what they showed is interesting. Oh, they in the open source have released a story with experts' mixsters, yes, that is, some heads, so to speak, which are responsible for different directions. In fact, two trillion parameters are just coming from the fact that the model architecture itself is. Let's try this on our fingers. I do, so some major part that decides which expert needs to address this issue. And then she's out of hundreds, I think those experts are choosing. Where does this request go next? Yes, so two trillion parameters are not both active parameters, but one can choose the part that will be dealing with this request. For example, if you have a math request, you have one set of experts on your desk if you have a poetry request, another set. And in this respect, it's interesting because in the open source, at least, it wasn't a release in a clear form.

Mentions: Llama · Open source · ChatGPT · OpenAI
00:06:46–00:09:11Llama 4 is better than DeepSeek?
Ilnar Shafigullin00:06:46

At least what I saw. Is that better than ChatGPT, Claude and there, or anyone else? We don't know, because it's not true how these architectures are built under the hood. Yeah, but like in the open source, this story shows up, they showed it loud, some baccalaureate hit. Ah, from what I had to dig up on the Internet fast. Those who started small models from the new line say it's not that DeepSeek is worse. More importantly, DeepSeek is even better. Even the current, not even the one that's rolled out. Yeah, so it doesn't make sense how much that's really a big step forward. But in any case, it's an open source. Anyway, it's a pulse on this side, and it's really good. With regard to the OpenAI, I'd like to say a few words, too. We talked about it sometime. Remember that lazy people always pour into a open source to make some other points? And now, the OpenAI was in the lagging-up to GPT-2, and they were, respectively, licking their models and earning their glasses. But even if they weren't lax, they actually got ahead of it. At some point, they let themselves not do it just because they were stronger than the others. And here comes, well, I see, since DeepSeek, respectively, put all this in the open source and beat the closed good models that were with open source decisions. Oh, not so that OpenAI can afford not to break into the open source story. So it's gonna be great if they roll out another model. Again, we've been talking a little bit. Yeah, open source isn't always free. There's a lot of business around the open source, money's working. So I'd be more interested, on the contrary, if the models were open, and the companies around it built infrastructure, services and money on the side, not just that they're here. They grabbed the model and they won't give it to anyone.

Alexander Volchek00:08:45

Please write in the commentaries what you think on the subject is open source, what's going on. There'll be a model open source or systems that will be on something and there's no actual open source inside. As we see, most of the world's systems are ultimately not open source. Let me see that.

Mentions: Open source · China
00:09:11–00:10:38Chinese technogyps
Alexander Volchek00:09:11

Look, I read a very interesting analyst of one yesterday. And that thought, I think, is very serious about the fact that Chinese technogyps are now opening up models. There's DeepSeek. Then Baidu, Alibaba, Tencent. All of them open models have been released, and the government is actively supporting this. There's a premiere, uh, meeting with the CEO DeepSeek. They said that by calling it AI there, the AI representative, the State openly supports the open models in full. There is a question. What analysts write, and will China have a quick end to love open source? And as soon as I'm, uh, those-

Alexander Volchek00:10:00

Ah, uh, Chinese companies are gonna be coming up very fast. They're not real there now, getting involved with American competitors, so China can change its mind about, uh, rethinking about giving up. Not to give up the achievements, but the question is, what's inside. There's a big subject, I don't know, see if we can discuss it today or not, which concerns, uh, the credibility of the information on the Internet. I mean, a number of studies are very interesting about the inflated, uh, news, there, since the twenty-second year of different types of millions, even millions of false sites that have been taught models.

Mentions: China · OpenAI · Open source
00:10:38–00:13:57DeepSeek: State control?
Alexander Volchek00:10:38

We'll get there, talk about it. So, for example, in China, it was now reported that some staff, such as DeepSeek, were not being released abroad. And, uh, they put it straight, they did it separately. And Reuters put it separately on them, a, information. And here's a question, or a State holds a hand on a pulse and, uh, is basically controlling this system, because it's obviously supposed to be in control that you, uh--- that's where you're going to be so super cool. systems, and supposedly these cool systems all over the world can use free of charge. Well, it doesn't look like China. Why did I ask you about, uh, OpenAI, what was that? Is that a-- is that a standard move, a strategic move, is that a retraction? What is it? And because I'm not quite complete, like I understand that a number of things are probably cool with the development of an open source, a number of things don't matter. Oh, what do you think about China?

Alexander Mashrabov00:11:39

I, I actually have a very simple opinion on this, very primitive, that all those who do big models, they have to, you know, that it's, uh, that's probably, yeah, that's probably, like, that's probably, like, a little bit of a lot of things. result in movement restrictions. I mean, for me, that's, even though we're kind of democracy, right? On the one hand, freedom of movement, I think that in the case of these models, it's perfectly normal that, well, that people are being restricted, temporarily restrict their journey abroad.

Alexander Volchek00:12:13

Oh, you mean you can get stiff or something?

Oh, I mean, it's just totally normal, that, uh, kind of, that, well, that the country doesn't want any knowledge to go abroad, yeah.

Alexander Volchek00:12:30

Well, these people, for example, with DeepSeek, will come out, and, for example, Americans will hire them, they'll hire them to work.

Alexander Mashrabov00:12:35

Yeah, sure.

Alexander Volchek00:12:36

They'll pay a few million dollars, and they'll be happy to stay in the States, give them the papers, and everything.

Alexander Mashrabov00:12:40

Yeah.

Alexander Volchek00:12:41

But it means that serious control. So we have to understand that this is a relatively open source, including. First, is there any closed models we don't see or that don't all show up? There's a very interesting one, uh...

Mentions: Open source
Alexander Mashrabov00:12:54

Of course I do. Look. Of course you do, uh...

Alexander Volchek00:12:57

Yeah, there was, for example, interesting statistics that DeepMind, uh, was directly posted, and the report that DeepMind, uh, does not report many studies they're doing, and they, uh, don't want a pre-market at all. Nothing to say what they're working on. Well, I mean, there are companies that are probable that the market reports what they're working for, and there's someone who doesn't tell us what they're working for. So here's the story of China, there's clearly more serious models, serious stories. Clearly.

Mentions: China
Ilnar Shafigullin00:13:31

Yeah, that's what we were talking about, Sas, which is what we were talking about, that if you know that the task is being done, then repeat it for now, at least very easy. Yeah, so if you're talking about what you're working on, other companies, competitors, other countries are starting to get the money, and they can beat you up and get better results. Well, or this is what we discussed, to leak our release on the eve or the same day.

Mentions: China · Open source · OpenAI
00:13:57–00:14:23China will close the Open Source?
Ilnar Shafigullin00:13:57

And, in fact, business will also have a lot of losses. It's about whether China will be closed, not open source. Well, I see it as I see it with the OpenAI. When the OpenAI could afford, uh, not to open source its resources, that is, other competitors could do nothing, it became a closed one. The same could be true with China.

00:14:23–00:16:23OpenAI BubbleSync: failure?
Ilnar Shafigullin00:14:23

Once they reach some point where they're clearly stronger than the rest of the market, then calmly and the fact that they're selling, they'll shut it down, too.

Alexander Volchek00:14:26

But in general, the updating of some systems certainly leaves much to be desired. I've still got a picture of it in OpenAI, which they've set out, unlike Claude, this bubble sync. It's just a change between their models. And they sometimes catch him, sometimes they don't. They just made a bubble that they're supposed to choose. It looks absolutely stupid. You'll think a little, you don't think it's like that. That's how it turns out, right? Although it is stated.

Mentions: OpenAI
Ilnar Shafigullin00:14:56

It's probably another frontender who's been working three months, and he's decided to roll a new thing.

Alexander Volchek00:15:00

It's actually a nut job. I mean, this is a complex task, even important, not a complex, very important task for the OpenAI to deal with all this button and everything. We live in a world where, uh, UX, UI is done so much, uh, analysis, trends, how people do what they do. Well, I mean, you could have created a cool, cool interface, and they still tidy up a thing that looks really weird, makes it even bigger. There are no additional descriptions, sometimes turns on, sometimes not loaded. Well, maybe like Sasha says, how Meta decided to announce a release on Saturday. You want to do better? Send it on Saturday. Zuckerberg was standing, I think, in his bedroom, he had a bed and a pillow in the back.

Mentions: OpenAI
Alexander Mashrabov00:15:51

Alexander Volchek00:15:51

And he remembers, uh, like he got a call and said, "Look, this is Mark, we have to take off the view," right? And you remember right away...

Alexander Mashrabov00:16:01

Let's go.

Alexander Volchek00:16:02

The situation, the Apple situation, yes, which said, first of all, that Apple Intelligence would start working full and be super cool in March, which said we were like in Siri, and Siri would be super fucking, move. And from Siri to the GPT, I think it's like the Moon, right?

Mentions: TikTok · Donald Trump
00:16:23–00:19:47What's going to happen to TikTok?
Alexander Volchek00:16:23

I mean, it's just-- uh, well, you don't--you don't see or feel, uh, you don't feel a big integration. Let's get some news tonight about TikTok. What's going on with TikTok? Given that last week's reception closed, the bidding for TikTok purchases. But it's generally a very complicated procedure in TikTok, like him, how can he be purchased at all, right? Because there are different laws that were there, the blockage that Baden made. Then, uh, the Trump extended the block and finally, on April, er, fate balances between the sale of American investors and the complete blocking. To sell American investors, TikTok must open all of its technology, which is totally weird, they won't. Here, . What do you think is going to happen to TikTok now? Sasha, what do you think of TikTok? Given that you're in the market, it's still in the market.

The tariff war has begun. The Tariff War is needed to force the Chinese government to sell. There are troubles, troubles are raised. Now we need to get the Chinese government to sell. There's gonna be a trade. I mean, it's a joint venture. I think it's all a mascara, like all the troubles, it's a mascara that will take it, of course, Oracle, to take it to Oracle. TikTok's infrastructure is there. Larry Allison is involved in Project Stargate. Well, Project Stargate, like, all the closest people to Trampe in business. Aside from those who have direct interest in the administration, like Elon Musk, there, David Sax et al., yes, and there, and so on. So, uh, and there, including on this list of other things, we need to include Peter Tile, all right, here. And now, there's a second echelon of his business partners, and it's definitely Oracle, and Oracle must definitely take TikTok. That's a great match. Oracle has infrastructure, has a government relationship. TikTok has, uh, some DNA. How do they build such an Asian team? They also have a very Chinese Asian team in America. And I mean, I think it's just a perfect match. I think it's just gonna happen. And once Trump has agreed to negotiate with the Chinese government, he has introduced these tariffs to start a strict agreement with all of us.

Mentions: TikTok · Donald Trump
Alexander Volchek00:18:50

Hold our canal, put the liquor on. This will help us build our classic independent videos and our interesting content of sites, technological sites from the U.S. Silicon Valley and around the world. Our great quarter will help move forward. How are the tariffs TikTok? What do they have to do with this?

Mentions: TikTok
Alexander Mashrabov00:19:16

Direct. I mean, because I think it's just a way to start--- well, assuming America's in a position of strength, which may not be true, but if America is in a position of strength, then the other ones. countries should go through all kinds of communication, by the way the relationship between the two countries is built, right? So, in terms of Ki-China and America's relations, of course, TikTok is one of the main questions. I mean, Taiwan, of course, is a question that can't solve. But there's a TikTok, which, it's a question that Trump thinks he can decide.

Mentions: China · TikTok · Donald Trump
Alexander Mashrabov00:19:47

What else? Well, yes, and it's clear that India, Viet Nam, has already decided that some other countries will agree with America on some kind of new economic trade. And I think it means that

And I think that means that many countries, i.e. tariffs, will somehow work as a way to negotiate. Oh, but we don't know if China will do that or not. I mean, maybe China will do it or not.

Mentions: China
Alexander Volchek00:20:15

Yeah, but the Chinese government was always very bad about this TikTok situation. And there's a quote like that. They said, seeing something good from another, they want to get in on it by any means. The clean water is a ganglog. This is a statement by the bakers about the TikTok block. Because the question is, this asset is huge, crazy. I don't know how to deal with them. Sasha, I don't know how TikTok tariffs work here, TikTok. Although TikTok may be worth the volumes of these episodes that they are discussing between China and the United States.

Mentions: China · TikTok
Mix00:20:58

Yeah.

Alexander Volchek00:20:58

Just who might not understand, Trump somehow said that TikTok was a trillion dollar, right?

Mentions: Donald Trump · TikTok
Mix00:21:04

It's true. It's true.

Alexander Volchek00:21:05

What TikTok says he invested billions in America, the cost of the American economy can be worth a trillion dollars.

Mentions: TikTok · China
00:21:08–00:25:07China vs USA: Commercial War
Alexander Volchek00:21:08

Everyone knows, by the way, the viewers probably don't know how far the tariffs are calculated. How did the U.S. calculate the tariffs? I think a lot of people are strangers to this. And since we said, we'll mention it as a supplement to some. The US has introduced tariffs and tariffs have been calculated as follows: Some will think that the formula was conditionally Japan had provided American goods, had taken a tariff, for example, 50 per cent, so that the tariff was set at 25 per cent. It's not. The formula was next. We've got a difference in deficit. It is conditional that Japan, uh, the American product purchases less than $100 billion than the less than the sales of its goods, that difference was considered as a percentage of total trade. And then this percentage was set at half the rate. I'm pretty sure I'm shocked by that formula. I don't have my business or professional brain that's not gonna let me believe that at all. I heard that formula, I decided to over-check it because I couldn't believe that tariffs could be introduced in the deficit formula and, in fact, so, you can clear the tariff, you can just stop your goods. I'll bet you'll be there. And the tariff's gone. That's kind of crazy, but let's see what's going to happen to TikTok. I, Sasha, by the way, didn't really think TikTok might be a little bit of a solution to this war of trade, yes, which is now up. Because China against the United States has also imposed a tariff of 35 per cent.

Mix00:23:01

But maybe it won't work at all. Maybe it won't work.

Alexander Volchek00:23:05

Because I can't imagine China doing this, it's Baidu, right, TikTok?

Mentions: China · TikTok
Alexander Volchek00:23:11

No, TikTok is ByteDance. TikTok is--

Mentions: TikTok
Alexander Volchek00:23:14

Oh, ByteDance! TikTok ByteDance separately, yes. I've got it all the time byteDance. TikTok ByteDance, so they can separate the world's assets in the hallding. But it seems to be a little bit of a weakness. Let's watch. Would you write in the commentaries, please, what do you think about TikTok? Is TikTok really good to have in different countries? He's in Russian-speaking space, in Russia, he's locked up, yes, actually, without VPN, he doesn't work. There's no way in America to download the app, but it's working. And the time limit I understand is about to come out. I mean, I don't think there's much time left, yes, to... You don't remember, Sasha, what date should be?

Mentions: TikTok
Alexander Volchek00:23:58

They've moved for forty-five days. Trump's been in there for 45 days.

Mentions: Donald Trump
Alexander Volchek00:24:04

So if we've moved forty-five days, they're probably already over.

Alexander Mashrabov00:24:07

Well, he's still moved. I think Grandpaline was on Friday or Monday.

Alexander Volchek00:24:13

Yeah.

Alexander Volchek00:24:14

Well, shorter, he's still moved there until May, shorter, he's moved.

Alexander Mashrabov00:24:18

Is he going to move so indefinitely or is it the last one, or is that the last one?

Alexander Volchek00:24:22

Well, he's the president, he can do everything, I think he's in the country, yeah.

Alexander Volchek00:24:25

Oh, he can do everything. It's a very interesting story about TikTok. It's unprecedented. Let's see what happens to her. And from the perspective of world wars like this, it's gonna be very interesting, of course, that's what it's gonna be like.

Mentions: TikTok
Ilnar Shafigullin00:24:46

There's a funny idoma that's the last Chinese warning. She's got a different game right now.

Mix00:24:51

Yeah, new.

That means you're told you that you need to get to that number. And it's like the last Chinese warning. Then it moves, then it moves, then it moves, then it moves. In fact, that's exactly what Trump is doing now. That's April 5th. Then we moved to the endless, we can move.

Mentions: Donald Trump
Alexander Volchek00:25:07

Maybe,

00:25:07–00:30:00LLM-groming: AI and misinformation
Alexander Volchek00:25:07

yeah.

Mix00:25:08

Until he gets his own.

Alexander Volchek00:25:10

Very interesting. I want to go over the subject today in the OpenAI image generation. What she did has brought into the market in the last two weeks. OpenAI was verbally and accused of using a fee-paying content for training. And now there's a lot of information that they've been training their date on O'Reilly's books, only available on a payroll. And that's how it starts to open up more and more. Not only is there more Chinese models that are generally interested in providing, for example, an open source of model to countries, such as those where American models are blocked, for example,? I want to read one of those things. So it was a study now that it even appeared that was called this LLM tactics. So the tactics are the next to focus on the large language models of misinformation. It's got a case. This case is indeed linked to the Russian Government that they have set out three points of six million articles of misinformation, and this information has been found in 10 leading LLM models. Like, uh, put it like adequate. Although this information was false. Well, there's three and three million, three-- three and six of these million sources. There are thirty-three per cent of cases in which they are entering the wrong response system, and wrongdoing. In fact, if you look at, for example, Chinese models and DeepSeek to start asking about some events that have occurred, for example, in China fifty years ago, there, 40 years ago. Well, I'm not talking about Taiwan Island anymore. The different things to start asking, the model is very evading. In fact, Chinese models are evading details, such as wars. If you ask about some ongoing wars that go, the Chinese models in DeepSeek are straight, often DeepSeek says, "I'm sorry, I'm not given this information to answer," yes, and limits, limits it. This LLM sadding has always been a very embarrassing subject since the beginning, because I realized that there was a huge amount of disinformation on the Internet. There's not much media that endlessly misinformation. Well, different media outlets are making different data. Then they say, like we were wrong or we fixed something. Models are taught more often than not. All these data are not compared. I've been doing a few people's analysis, uh, deep-reser this weekend. You-- interesting, yes, Tanya, that it turns out you're ChatGPT looking for who you are and did a very big article about you, wrote a big story that you were a great designer, got there, there, there, there, that way, you did something like that, I'm not sure what I'm saying. Yeah, it's very interesting that ChatGPT was looking for people, at least in deep reservoir in O1 Pro's version, he's doing it. Now, in the current, in my last edition, I even wrote it down, which--we even showed a piece of screaming with you, information about you, yeah. And I repeat, I've been wrong in the past, the past roll. I mean, I said that ChatGPT doesn't. But I-- that's my total mistake, and I've studied a lot of people. Not much of a detail, for example, has studied itself. I've been studying myself, taking as deep a reservoir, uh, boot-- launching. He's been analysing me for 40 minutes, and I took my data, and he said, "Find it again." And then I took my data. Find another one, find another. Well, I'd like to start a whole series of studies. And I mean, it turns out that part of the system uses data and says it as objective, based on information on the Internet from very different sources. Well, I mean, it's not enough that's not exactly what they're comparing information between themselves. And this LLM grumming, well, I can even imagine what the governments of all countries are doing. I can, I can, uh, from the intelligence service, I can even imagine what they're doing to saturate the Internet now with false data. Because this deep reservoir, he's very good at analysing, including private sites, personalized sites, small sites that don't come into, generally high-risk-free. So, basically, he finds data that even hidden on small sites because there's an index? Yes.

Mentions: OpenAI · Open source · China · ChatGPT
00:30:00–00:32:18LLM-groming: AI and misinformation
Alexander Volchek00:30:00

He finds them, and he's...

Alexander Volchek00:30:00

He finds it, and he shows it. Maybe they're in serious weight, especially if there's a description. And you can, for example, be on your website. So if you, Tanya, for example, on the website you're gonna write more about yourself, you're gonna unpack a bunch of extra data, then you're gonna put an article on some other website, like, somewhere on the Medium, you're gonna put it out. The article, you publish yourself, you write it, you have the right to put it in there, you write that you're in some other subject, then you're gonna put another on another website from someone else, and that's gonna be relegated. How very reliable information, as yet reported in different sources. And GPT's chatting concludes that it's not, well, look, it's a fact, it's a fact. That's, that's true, absolutely. Different sources say that, report it. And he'll use your personal information from your website. So I think this is a study of three, six million sites from Russia that are created, well, there's a very cool analyst, by the way, super real. And the fall in, uh, systems, it's a-- well, it makes me very much thinking, because, uh, a, for me, it was a question number one for me at all. So, uh, what's going on with data? One--- one thing, that is, that the OpenAI, for example, other companies, is a crazy number of companies, I think everyone has been learning about, uh, inf-- data that were not available, that are not owned by, Some other company or people. On the other hand, there are many lies and mischiefs. And how these-- there are new types of solutions that are starting to exist, how to circumvent things like models do extra checks. But I think the more smart the models become, the more mistakes and the more difficult to check. I can't imagine. Well, that's, every new model will be created to go around again, find all these own details. What do you think about this? I'm just gonna go over to the, uh, auto-data production. People started generating broken cars and sending them to insurance, that they had a car crashed and confirmed insurance.

Mentions: OpenAI
00:32:18–00:34:41LLM-Graming: AI and misinformation
Alexander Volchek00:32:18

What do you think about this?

Mix00:32:21

Ilnar Shafigullin00:32:22

No, about, uh, training on some pay-for-all content, there, no licenses, no permits. I'm just saying, you know, never happened. And that's it again, right? We've been talking about this for a while.

Alexander Volchek00:32:32

Yes, yes.

Ilnar Shafigullin00:32:32

It's a story. Then again, the officer who was doing this has been resigned for a long time. And it was in the previous model, like they said about GPT 3. With regard to injections, well, or sadness, yes, as it is sometimes called, this may, respectively, occur at different levels from the teaching of a fundamental model to, respectively, further stories. If we're talking deep research, he's starting to search the Internet, he can just find some unreliable information. And, accordingly, this is the hard way to track. The only thing that, uh, I'm thinking again, we talked about something, remember the golden dateset or something, uh, everyone's checking out and-

Mix00:33:16

Mm-hmm.

Mix00:33:16

...the date on which you can only learn.

Mix00:33:19

Yeah.

Ilnar Shafigullin00:33:19

And the rest of it, we don't really see the subject. Some sites, some impounded or unreliable, again, by whomever.

Alexander Volchek00:33:27

Yeah.

Mix00:33:27

...commissions, et cetera. That's why the solution is unclear. The problem is obvious.

Alexander Volchek00:33:33

It's like, Elnar, a talent visa.

Mix00:33:35

Yeah.

Alexander Volchek00:33:36

In the U.S., for example. You're a talent visa, and we're like a visa, and, uh, it was just that the media were certain, right? And it turns out the media is certain. And forbes, for example, is accepted. And we were just talking about Forbes, man, publishing just a lot of unverified data, because it happened. I'm not saying that a lot of people, uh, a lot of people bought, bought, and, uh, often buys publications.

Mix00:34:10

Yeah, yeah, yeah.

Ilnar Shafigullin00:34:10

There are resources that sell publications, even those that have been verified. Even the checks. You say, wait a second, so you're just gonna be out there, you're doing it. And someone, like, uh, in the resources checked, even in the resources checked, does that story. And it's hard to check. Well, it's really hard.

That's a lot, Sasha, even in the academic environment, when we talk about scientists, research, etc. People buy dissertations and publishes this story all the time. There's a story about the exposure. So...

00:34:41–00:36:54Can we avoid a fairy and disinformation?

And if we're talking about the usual media that don't have a review, without any hard scientific checks and anything, it's a uncontrollable story.

Alexander Volchek00:34:47

Sasha, do you think technically this is some sort of job that's being solved? Or it's all-- we're still going to be in the era of even bigger fairies, more strange designs, and we're gonna have to introduce these laws. New Jersey introduced the law for the fairies. Five years in prison. For the AI-based fairies.

Alexander Mashrabov00:35:05

I think that, like I've said a long time ago, yes, everyone, there, in a private circle, I can say publicly. I mean, it's not a very deep discussion, but, uh, basically, big business will be built--- how big business is gonna be built in the field as security, because, like all, it seems like the whole security layer. At the moment, he, well, as it would be, well, obviously, he's not created to deal with AI problems. And the problems that might be with AI, they look like they're more serious, right? This misuse of AI is a big problem than any attack that you can imagine today. So, it's clear that we're gonna wait for some AI adblock, right? Sentatively. We'll see what happens, it's gonna be... Ah, bo-boo-- are gonna be companies that give-- how do I say? It's kind of a safe experience. So all companies will have to have their own, as if any company in the world has a big deal in the consumer segment, they'll need to have a child-friendly offer, right? To have a child offering, you know, you're gonna have to use AI and this is the firewall, too, like AdBlock, firewall, and so on. Yeah? I mean, it's like a big subject. We haven't even touched that surface yet, this surface is still on. We, I mean, I think that, well, there must be one tyre of two crises, and then that, and then, that's where huge investments go.

Well, by the way, locking up for kids, I think it's a little superb, which, well, is really, really, really, really, really, really, really, really big.

Mentions: OpenAI
00:36:54–00:40:30OpenAI images Generation

Okay, let's go over the subject of the OpenAI having an image-generation tool in ours. And it happened a few weeks ago. So they've released a built-in imagery, got an incredible number of new subscribers because of it. In particular, because their model has become a generating completely new things that, well, as they might have done, have been hard in other places. In particular, for example, the generation of text, the quality of the generation of a certain type of picture. And I was talking about people being generated. News went about people generating scratches on their car and sending them to insurance. Well, it looks very natural, yes, it looks very unreal natural. Before we start discussing the theme of the image generation, did you have a picture-making in your life? What kinds and apples? That's the thing I told you about the car's generosity. Maybe you had these briefcases. It's gonna be interesting to everyone to read about them. Write in the comments. This new image that OpenAI produced is a very interesting example for me because I thought OpenAI was very much behind it, like, well, not in the trend right in the top. They're doing it, right? Not in the trend with the pictures. Well, it's not supposed to be a straight market, but they just fired incredibly. And... What, Tanya?

I really liked how this thing works. There's a SketchUp program, it's a middle between 3D visualization and 2D Road drawings. And I tried to load SketchUp and describe the text as I want him to do it as a photorealist. And it was really fun. So then I tried all the collaterals to create from different interns, ask him to find common features between them, which is what you can combine. I mean, it works. Well, so far, at the base level, but it's so much more forward that I was kind of surprised.

Alexander Volchek00:39:39

A crazy step forward, especially because the text came up. Can be done now with text banners and all the skirts. And there, well, Russian, for example, works well, but English is super cool. Well, I'm not saying that this animation is, they've got 600,000 Hashtags in three days. There's something that's been killed, the animation was dead, they copied. But...

Discussion participant00:40:00

They've copied the Gibles. But I think OpenAI does something weird. Well, as in particular today's good words, there must be a new X UI, so the designer. And some things are very serious, well, they move, they throw it away. I don't know why they're not throwing away everything advanced. What do you think of images in terms of use?

Mentions: OpenAI
00:40:30–00:41:38What's better than Midjourney or ChatGPT?
Discussion participant00:40:30

I saw it now and I started, by the way, in different places, using pictures after the OpenAI pd. It's like a trigger gave me this way.

Discussion participant00:40:38

Midjourney is harder, yes.

Yeah, it's literally ten seconds, yeah. I mean, Midjourney. Everyone knows he exists, but there's not enough to use it. And when it came to ChatGPT, it was built with extra chips, there, on a text that doesn't make mistakes on the picture, it's got a lot more people, I'm sure, to use, because I turned my own off. Midjourney's signature a couple of years ago. And now that it's in ChatGPT, I'm sure I'll use it. That's why it's a huge plus.

Mentions: ChatGPT
Discussion participant00:41:11

Well, let's see what you say, vocabulary is just a breakthrough. But for me personally, it's a breakthrough because I have fifty percent of the case, like when I needed a picture, even maybe eighty percent of the case, I needed words, I'm gonna be honest, I needed to. I'm saying that. And that's why-- and plus that's still a separate system. There, Midjourney, Stable Diffusion, there, Firefly, yes, different operators, it's still a separate interface.

00:41:38–00:42:44Example of photo generation: AI restrictions
Discussion participant00:41:38

And Tanya wanted to add.

Discussion participant00:41:38

I wonder how they're dealing with their hands. Is there five fingers or six? I haven't tried to generate yet. No, I wanted to add that I'm more of a professional than anything. And my son's birthday was just about to make a school he wanted to be. I'm asking him, "Who do you want to be?" He says, "I want to be Batman." I think, uh, drawing this Batman was a handy lazy. And I think I'm just in the GPT chat room, generating my son in Batman's form. I downloaded the picture, and he started so great. That half of this picture is so generosized, and then he's so brushed and said, "I can't use pictures of kids to create a picture." And I tried, there, thinking about Ilnar's techniques, trying to get him to do this job differently. But no, it didn't work. I mean, I don't know why.

00:42:44–00:45:18Future generation: OpenAI vs competitors
Discussion participant00:42:44

Here he's blocked me all, and I've downloaded another app and I've done it in it. It's so simple. Comments on life.

Discussion participant00:42:51

What's probably the thing, yes, what's in the chat room, in the chat bot, and that inside the chat is now available and how to think of certain scenarios is convenient and more comfortable. The question will be how they'll do different things, cool things. Because of course, you need different templates, yes, you do. The user needs to explain how to create this, what to do, how to build it. Well, to get the system to understand some kind of style. It's not easy, I think, for the hundreds of millions of users we're talking about. So we're still talking about OpenAI, we're talking about hundreds of millions of people and the mass use, hundreds of millions of people. I think the hardest hundreds of millions of people is to have the system do very simple things without-- understand what you want without some super-crunch proms, right? As long as many systems like video-generation, for example, I've tried video-generation several times and I haven't. Now, the updates were different and they already gave their team. Even yesterday, I bought it extra, yesterday, I paid an extra maximum subscription to Runway. Said he had to make video.

Mentions: OpenAI
Discussion participant00:44:07

That's very bad. Even Higgsfield is better, but the best thing about Click is, I guess.

Mentions: Higgsfield
Discussion participant00:44:10

I know. No, Sasha, you have a signature, and how am I gonna buy Higgsfield? You tell me, please.

Mentions: Higgsfield
Alexander Volchek00:44:15

It's all available on the website. We're already on Monday, everything is available on the website, we can download the promocodes in Brats.

Discussion participant00:44:24

No, I'm not. I understand. Look, I understand that I, I understand that I can use Higgsfield, I can use Higgsfield fully. There's another moment. The other thing is, I'm just in the team, they asked me to go to Runway and so on. I'd love to give Higgsfield, Higgsfield, by the way. I just want to add microvideo, microvideo in youTube-video, including ToTheMoon. And to start trying. I started sending some examples last night. It's gonna be cool. We're gonna have to do it separately. We talked about Higgsfield last time. Sasha and I decided to take off the video-generation issue. Because Runway's got a bad thing. Especially with different types of mouse, but I don't know much about it, so I can't say.

Mentions: Higgsfield
00:45:18–00:48:20Future generation: OpenAI vs competitors
Discussion participant00:45:18

Sasha, what do you think about the picture?

Alexander Mashrabov00:45:18

I think we see, well, that the real challenge is that now, as the biggest gap, is exactly the implementation, that is, the introduction of technology into the sale. So we're talking to a lot of influrizer agencies, creative agencies, and so on. I mean, everyone has, like, a lot of time going to the dental. It's obvious that the OpenAI and Gemini, that is, Google models, they're already very good for the denorist generation, but nobody's introduced yet. So, as long as you can say, pictures are almost, that's a lot of progress, you know, you've done, yeah. And as much as it can be seen, all the start-ups who were doing the paintings, I think now.

Mentions: OpenAI
Discussion participant00:46:09

How do you say that soft? You can't be soft.

Discussion participant00:46:13

More lazy.

Discussion participant00:46:14

Fucking need them all. Here.

Discussion participant00:46:17

Discussion participant00:46:19

Well, now it's like the next era, anyway, with pictures, the video is complicated, it's a more complicated case. And the pictures are made, well, there's a lot of day-to-day stuff. She's supposed to be built into the usual chat-bots, right? In the usual models, and then in the process, you'll be escorted.

Discussion participant00:46:38

Yeah, she should, yeah. So it should be used, used as? Like these. I mean, just these models, they should be used for the heatings of the heatings. That's what's needed to run professional teams. Well, professional teams need an interface, some rheumat, where to leave comments, which means there's got to be a decision to do something about the middle of the GPT of the world. I think the world is in there now, I think that it's likely that a new productivity tool will be created for designers. Well, either Figma will accelerate.

Discussion participant00:47:09

I mean, there's gonna be, yeah, professional decisions that will be used inside of themselves by AI. You're talking about separate decisions. There is still a mass such a consusual market that uses conventional chat-boats and models. There are professional things. Miro, Figma, they're all gonna have to get in with someone super cool or they're gonna get a little haun in some time. Well, they're supposed to be integrated. Well, they're obviously doing it.

Discussion participant00:47:33

From my point of view, the Consymer is talking about nothing. I mean, there's distributors, I see who's got the OpenAI distributors. Well, they have a good, very good picture model and everything. I mean, I think the Consymer market is all clear with the Consymer market, yeah. You've got a...

Mentions: OpenAI
Discussion participant00:47:50

Look, but earlier, Sasha, there's always been mobile applications, yes, separate, like Prisma.

Discussion participant00:47:56

No, it's pure fain, it's a fan, it's not serious. I mean...

Discussion participant00:47:59

These will be fans, but the same will still be the fans of the mobile applications that people will use separately only for, for example, pictures. Or at some point, it migrates at a big time...

Discussion participant00:48:09

Prisma was for some of those fana effects. Well, of course, it's all centralized in the end, of course, centralized.

Discussion participant00:48:16

He's centralizing. So people will stop, stop rolling and using these

00:48:20–00:51:33Future generation: OpenAI vs competitors
Discussion participant00:48:20

applications.

Discussion participant00:48:20

Not that they're gonna stop using. It's just that if you have the same in GPT chat, right? Why else would you go out and pay for it, huh? If you know it's a normal thing to do. Oh, here. I mean, I know what's on the prostheme. That, I mean, when a man needs to design himself, he'll use all the tools, but they're gonna pay a lot more money than that. And then you have teams. The team needs this colonization, shared assets, that's all the comments, boards, mood boards, that's all. So, some caste LAW routines are routine. I mean, well, this market, there's no one out there, just empty. Zero. Nothing, nobody does anything, just sit around like that. Yeah.

Discussion participant00:49:08

Does Miro do?

Discussion participant00:49:09

I think there's nothing. Well, I mean, I don't know exactly, but if I did, we'd know. I mean, I'm not gonna be able to imagine that the company's team is in there with a billion-dollar capitalization, if it does something, we didn't know about it. I think all these companies are investing a lot in the pyre. Just because we don't know. I guess since we don't know, since it's not on the news, it means they don't.

Discussion participant00:49:28

Why don't they do it?

Discussion participant00:49:31

I think that--

Discussion participant00:49:31

You're not? Can't? Why?

Discussion participant00:49:33

I think that, like a lot of Figma or Miro safts, they're built on a very, very simple, very, very intuitive basis, right? There you go. And how easy it is to work with a complexity like that, well, how would users not get used to it, and how much team gets used to it, huh? I mean, maybe this is a situation, right? You just have a huge gap between the interactive soft and--

Discussion participant00:50:00

Between the interactive elevator and AI. It's like two different creatures-- well, of course, two different DNAs from the company.

Alexander Volchek00:50:08

Well, it's different designs. I--I had a conversation yesterday, there, with one person, and she told me that they had a corporation, you know, a company that was big enough, like owners, to share money with the other. AI separately. And the company doesn't develop a direction very much inside. One owner wants to, but the others don't. It's also a story like building an internal way to develop or create such a system. Clearly, the generation. I want to get, uh, a system that's gonna be smart. Yeah? I want something that might be smart. It's like a system that could fix the files in a smart way. I want a system that can fix the photos in a smart way. I want a system that can fix the post office smart, really, really cool, calendars are cool, right? Not-- not some kind of caesa, but that's what, well, that's what really is considered to be artificial intelligence. I want the system to do this cool. And we're planning on filming and making new videos. They'll start on Wednesday. Wait on the canal, some of the media will start. I'll see you in a week.

Discussion participant00:51:33