And when we get the answer, of course, we don't see or understand at all, because of someone he makes a calculation. He can't explain the whole chain of his own. I've got like over 100 developers in the company, and almost no one has started using, uh, artificial intelligence tools. And, in fact, everyone had to be fired, almost all, and hire a new team of people right away. ChatGPT Pro for $100. So the spirits come out and say that whatsApp are fraudsters, deceive people, and they're all inseparable and they can be downloaded. Well, now, Sam Althman has been hit a lot. We even in the past, we accidentally hit him, said he's technically untouched. Ilnar, you said that. And then there was a whole story on this subject.
Ilinar, be careful with the words next time.
Hello, everyone! We're on ToTheMoon. Technological news, Silicon Valley sites around the world. We go out once a week, every Sunday. Sometimes we have special episodes on Wednesdays. Look, Illnar, what is the subject I want to take up at first? We, uh, have a whole range of different things to discuss. I'm going to have to go. It may not be about artificial intelligence, but it still has a very strong connection to the technology market, especially since in artificial intelligence we are talking about cybersecurity now incredibly strong ways. The question is next. So the spirits come out and say that whatsApp are fraudsters, deceive people, and they're all inseparable and they can be downloaded. And so, based on whatsApp, someone's sued on them. Well, I'm, like, a user of whatsApp, I'm still important to me, whether they're sequential or not. I'm going in again, reading the news, I think what he's saying. He comes out the next day, or something, and says, "I want to explain what it means. They're not secular because, yes, they have, but messages are encrypted from the original cross-cryption, which is inaccessible to the WhatsApp Corporation. But! If a man makes a baccap, then the default of the ol's rap is not encrypted." And he says, "There's the most they're fooling people, "because even if you encrypted the baccap, the people you wrote with, the baccap wasn't encrypted, the baccaptized, and, as a consequence, access to your tags is not encrypted. "As reported to these people, "they will be accessible." I'm sitting here and I think I had a encrypted backup, and I couldn't re-establish it. I once had one. I told you in some ToTheMoon, I couldn't reinstate it because I forgot the password because there's a few times you asked him for it and everything, and then he washing it, and the backup is encrypted by your key. And they don't know that key. Well, let's just say that they don't know that key, but at least it's encrypted. But I have a question about why Douras is talking about the encryption of the Baekap if Telegram does not encrypt messages, but only encrypts them on his server and has access to it to make Telegram work with his server. different devices everywhere, unlike whatsApp. In whatsApp, you've been undone by the other device. In WhatsApp, if you're undone on your cell phone, you're being decorated. WhatsApp asks you constantly for your second double authentication password, which does not give you the opportunity to work. And the Telegram says to you, "Do you remember the password or not?" But you can't answer that question. And Telegram can't encrypt the backup at all. And in Telegram, only secret chat's ciphers are encrypted through the cross-crypted encryption. I don't know the people who know them at all and use them. There's not enough secret chat problem, they're not moving yet, yeah, from the device to the device. Illnar, you're technical, so I've studied this subject, revised, tense, uh, Pro's version. Now, a new Pro version has come in. We're gonna talk about her now, yeah. We talked about breaking up in artificial intelligence models last time. It's a very important story, how much the breakup is now going on. We're gonna talk about her now, yeah. But the point is what? Oh, you, uh, you have an explanation for the Durov speech or marketing, given that, for example, in Russia, Telegram has been restricted and Durov needs to keep an audience, for example, in part, not in his own place. WhatsApp. And he's doing these remarks.
Default! No, no!
Yeah, yeah, yeah.
Default
Yeah, by default, yeah.
They're encrypted. There's a double authentication, and there's a complex key and there's no default encrypted, yes.
So the c-day part of the news sounds like that, yes, that the default of the racks WhatsApp is not encrypted. And if your interviewer, uh, did the backup, uh, didn't encrypt it, then in that case, access to your correspondence could be obtained. Well, as you can, you can turn that news, and you can put it like you said in the first version that whatsApp is cheating on its users.
He said that!
It's not secular.
And Durov said that.
Yeah, yeah, yeah.
So everything is purely fucked up, marketing, in the hip. So here's the news, which is a good show of the world we live in. Yeah, technically, you've really done it. Here.
Yeah. I think it's just a very relevant subject, because it's like a story from the standpoint of OpenAI when people don't shut up learning about their data, don't shut down their data. There's a button. And we're talking a lot. There are people watching our graduations. I have a lot of familiars, a lot of them. I'll open their phone, they never turned that necklace off. Well, by some way, it means that the feeling of feeling is that they're still in this subject, but it's a possibility that I'm not gonna be, so I'm gonna send the data because my data will be disseminated. And that's why I'm here by a comrade saying, "I'm using secret chat." So at least you turn off the tie so that at least the basic chat rooms don't even have a tube.
I'm offering you a flashback. I've shut down my tie to study on my GPT chat data so that all of this can be
seen.
Or I object, Volkick. Or I object, Volkick. I agree with the Dura. And by the way, if someone's technically explaining it again, I'm not arguing, and I don't want to argue with the Durov because I'm not a superpowered masssanger professional, in a cross-crypted encryption, in different ways. Likes. But I've been studying this question a lot. I even think we've raised him on several occasions in Tuzmun. That's the same question, exactly the same, because Duras has some sort of periodicity. Well, Mask's been brought in, right? What Mask said is that the stashap doesn't have to be trusted. And in reality, in the Valley, the Mask and Mask team are personally communicating in Signal. I know that, right? Another aspect is that, well, there are other circumstances. And these circumstances, they're different. And in terms of whatsApp, if we say that whatsApp data can keep your code. My friend, no name, is strong in the AI world of the United States, he says, "I think whatsApp, when you put in a key to the encryption or you're injecting something, or they're the key to encryption on your phone, "They keep all the keys." He says so. Although I think that if it were to be so conservation-oriented, it would have been a long time ago, in some part of the analysis, analysis, I don't know whatsApp. Maybe it's not open. Again, I-- everything is possible, yes, everything is possible. There was also a saying that these brackets would be walking around the iCloud and the rest. But the Apple vault is actually protected, I think, more than everyone in the world is safe, and everyone knows it. And there was never a leak. And so we don't see these leaks, and we don't see the spread there, or, I don't know, the Apple says that these people are keeping the wrong data, remove it from the iCloud. Unlike whatsApp and Telegram, iCloud does not. Yeah? Here. They have no bands and no different stories, but yet. I have some troegrams in my head. The bully for their lighting these birthdays that I don't want to see at all means people. And a bait for these circles of endless people in terms of the dentist. I pay for the pay-as-you-go, right? I also have a few Telegrams with the pay-as-you-go. Can I do another pay-as-you-go version of that, which means I'm not gonna see it?
I'll have to remove it. I'm gonna have to remove it now.
Dovrov, get the body back.
I have a lot of old contacts because I have some contacts. I just started, started this disposal story, right? There's a connection, just there, 15 years ago, 20 years ago. And that some people there...
I come to see who that is sometimes, too. You forget what a man is.
Yeah, those s-- that's the same as the dentist. I tried, for example, in Telegram, well, because I've had a lot of activity in social media outlets, yes, and many followers from different channels, I tried, at some point, I wanted to be in Telegram, to put a hundred. But I thought it was, man, if I ever, I don't know, I had a pipe repaired or, uh, a doctor, they should see my moss. But it's a little weird, right? It's still another audience that signed up. They didn't come in my phone book because-- well, there are different people, yes, I don't know, some ministers to, uh, walk--- I don't know, a dog walk by some man. So anyone could be there. What kind of mosquito?
Okay, let's go to Pro's tariff. We're back in the past.
We were at last, uh, last graduation talking about, uh, that very strong starts to spread models. First, we started to produce models that individual, uh, individual tasks, right? In particular, it's just now, there's a new cybersecurity model over there. Well, shorter, it releases the inner and Anthropic, and Google, and-- and OpenAI, it produces some individual solutions inside. But, uh, I was just saying, "I think it's gonna be a while, and I'm gonna say goodbye to my free Pro version of $200, and I'm gonna have to pay a few thousand dollars. And that's the kind of thing that I said. On the same day we made the survey, the afternoon news is that ChatGPT Pro came out for $100. Let's just figure it out a little bit. It's really a big story. On the one hand, this topic, which is relevant to the desire, OpenAI, to earn some money, obviously, yes, and the amount of money. And that's not very good. Now, let's see why it's not so good. There are new statements about OpenAI, their value and their attention, there, the Enterprise, and how they want to copy, there, Anthropic and everything. On the other hand, that's the version of $100, what's she doing? She, uh, gives you a better power than a plus, but in two hundred dollars, the version, it gives more power than that one hundred-to-one version. Oh, they-- they're that way. So they describe it, for example, that in the Codex, a person gets an opportunity in Pro, there, I don't know, 20 times more requests, five or more. These numbers, they're very conditional, they're very conditional. And I'm saying that the difference is not how much power you have a little more, but the difference between access to certain models. Like, I have a model in which the version is present, Pro Extended, or a model called ChatGPT Heavy Thinking, right? And accordingly, in the Thinking model, there's heavy Thinking. And I'm in the dentist, by the way, I don't know if there's any heavy Thinking there, if he's here. Maybe one of our viewers will write off, but I think this story is gonna be separated, uh, well, a little excited about the market and at some point in time, start destroying, uh, free, well, conditional-- Two hundred dollars is a parole, a hail, a super-crunch model. So, gradually, for example, now, we're in the middle of $100, and then why not inject $2,000? Well, that would be a very good move. Now you've been in a hundred, then you've put in two thousand, and you're kind of balancing. "Ah, the point is, you're not the type of $200 you have anymore, but the point is, you have a $100 for you, please, it's a way to be cheaper." Ah, and I think that's all, of course, it's gonna be more, uh, people to get off the cool models. Although ChatGPT in its manifest, there or as it may be, for 2026 they referred to one point that, artificial intelligence must be accessible to all people without restriction, as is accessible. electricity or gas is available or Internet is available, that is, accessible to all people. They must have forgotten to indicate what quality artificial intelligence must be available. What do you think about this story?
Or is he-- is it just some kind of microfiche and doesn't matter to you?
The main story, I think, is the last few months of ChatGPT, well, OpenAI, I mean, persistently con-- copying everything Anthropic does. Maybe the Marquetologists or Altman just felt they couldn't possibly have some of their routes to pass Anthropic. And, in fact, we just started to copy blindly. Yeah, that's what you say, write it down, but not exactly. They're gone to the Enterprise, and they're moving on to the same plan Claude had. Claude for a hundred bucks, there's a single one that's pretty popular among the programmers. And, therefore, that niche not to give up, not to let go, they're doing another plan, respectively. And yet, well, like another factor that shows that it's on the side of the programmers. The difference between 100, 200 dollars, at least what they write is just a quota on Codex. And the Codex among my friends, except the programmers, is not used in any way. I mean, I hear that news is purely the audience that's gonna choose between, uh, OpenAI for $100 and Anthropic for $100. I mean, for this audience, if there was 100, two hundred and a dislocation, it's probably more towards Claude, considering his own remembrance, uh, for the best-model programmers, now it's just OpenAI. They're also in this area, and they're in the same zone as Anthropic. But I don't like it so blindly. Uh, but if there's a $2000 plan, and they'll still upload this psychotic plan as much as we should get to models, it'll be interesting. It's expensive.
Well, it's just a story here that if they just released $100, and it's like Anthropic, I think it's a very good move from a business and strategy point. I'd say at least as a businessman, it's a good move, because that decision is unrealistic for what and for whom, because the tariff itself is unexplained. Not much of a fact, I first walked in on their website, and I didn't even see two hundred dollars, and I had to go through the GPT chat because somehow I had only a hundred. Maybe this problem is being corrected, maybe it may not be corrected. I never understood what the fun was. It's like this pen they were supposed to send as a gift for using the story. That's very funny. I told you about some of the episodes. I just got a letter from OpenAI, because you're one of the first users of the story, we want to give you a gift, but to get this gift, we'll give it to two thousand people. What do you think this is for? It's a pen, it's easy. You're looking at this and you don't know what's going on. You spend $2,400 a year on the subscription of the system. You've been doing it there since day one, probation. Yeah. And even if I'm talking about the story, I'm not saying the rest anymore, am I?
Maybe there's some special pen, Sasha.
With emblem.
First you get it, then you complain. There's a self-smart module built up there.
I saw it. Yeah, yeah, yeah, yeah, sure. To say, what you've told me now, by the way.
I want to remind our viewers why the story of the OpenAI was even up. OpenAI now, well, it was, it was, they had a round of $800,25 billion, and the round was a size of $100,2 billion. And after that round, discussions began, discussions started on whether or not the OpenAI is working and working, especially in the lead up to IPO. Like everybody wants to see transparency and so on. Well, it's just a pyre, a different activity from the different sides. And who's eating bad, good doesn't understand. Yeah, it's like IPO now. And Dario Amadei and Carpathians have all told me Sam Althman is a dick. Yeah, so, well, they think they've been writing notes, recording everything there, so they've been describing, encrypting and now, so they've released an article and they've shown that Sam Altman is inaccessible, that we wouldn't give it. There was a basic package that we wouldn't let Sam Altman press the button to launch AGI. The main story. Well, now, Sam Althman has been hit a lot. We even in the past, we accidentally hit him, said he's technically untouched. Leonard said that. And then there was a whole story on this subject. So Sam Althman's kind of technically untouched and what's on the scout?
You need to be careful with your remarks.
Yeah, I hope we didn't let her in, this wave.
You're influencing--
Well, of course not us.
To the world economy.
That's the point. But. But the article came out, the article came out. And that article was just about what Sam Althman in the world thinks to be a man of the sort. Look, I think he's a strong businessman. There's no question at all. I don't have any pros and cons, but there's an illusion that people create around people, illusions, how they love them, say they don't pee or poop, and they know everything. And just incredible people.
Sasha, you have a great vocabulary today.
The first story that OpenAI had, the first so-called turnover that had occurred in the company history, was the Google first December of the twenty-fifth year. Gemini 3 is inside, so the Gemini 3 OpenAI has declared the code red. The fact that the announcement actually says we're kind of afraid of someone or something. They've declared the code red and strengthened the job. And then, in this background, five two GPTs were released. So, the second turn. I think that ChatGPT has been an incredible improvement in a year. But in general, that's what's going on in the market. The second turn was from Anthropic Enterprise. On the 23rd February, the same Reuters reported the launch of the Frontier Alliance. There with BCG, with McKinsey, Accenture, well, consulting with all companies to introduce artificial intelligence into the company ' s main processes. And on March 16, Reuters wrote that OpenAI was discussing a joint venture with large direct investment funds to enter corporate clients more quickly. Yeah. And they wrote about it, by the way. They released a whole article last week about the Enterprise, and that they even wrote that phrase that we were not just research company, but, yeah, still company. It looks very specific. And then...
Special. And they also made it clear that we were closing Sora, right? And the market thinks that if they are, I think that I'm thankful that you closed Sora, right? Because why did you go in there in this business? So they close Sora, it's also on the market that it's a very specific story. And now it's like this assessment, you know, like the market says that when it comes to this assessment, it's like a clear picture, not "We do everything." What are you doing? I think it's some kind of popularity, because when the OpenAI goes to the IPO, it's gonna cost exactly more than a trillion dollars. Maybe when he's out, he'll be two. He's gonna invest all the money, it's gonna be a crazy turn, so, because Anthropic is already, I think, almost over the eight hundred billion. Something I saw about eight hundred billion of news around yesterday. Not enough, Anthropic said he was planning to release chips. Well, they didn't, actually, say, but there's information that Anthropic is planning to release chips. Started to consider, then the fact that it was written that on April 6, Anthropic announced a new Google Broadcom agreement on the next generation multi-capacity. And then Anthropic says he's already learning and servicing, so the AWS models are at Amazon, there, and so on. And Google CPU, GPU graphic processors and tra-ta-ta-ta-ta-ta. And then they're on, so there were explanations that they're gonna start, start making their chips. I remember the xAI, and he's in the line. By the way, I think people in the market don't understand-- they don't understand that xAI is a company to date in terms of level, OpenAI and Anthropic and Google. Well, I mean, in terms of the artificial intelligence model. Even yesterday, she was one very interesting comparison, and they were shown as four directly separate independent models from the whole market. So their level is huge. The 20th-sixth Standford study says China has taken the US, but I don't see anything in China even a few of the uses that they all have. The total number of uses of their models is less than the use, I think, of half of OpenAI, yes, or half of Google. But the xAI should pay strong attention in any case and not forget them.
I want the news here about China. China is growing, as the association of Russian-speaking people with China is still mixed, it is very much built on China, like, being a village and nothing, pre-feasibility, and Many people are not fully aware of China ' s technological breakthrough. And even I remember my discussion when I said ten years ago that China, when you compared IT in China and Russia, was always looking at me like that, like, "Frying" of course, Russia, like, "Ho-ho-ho-ho-ho-ho," Which means, from a technological point of view. I told them, "Look, I'm still impressed when I'm going to China, using their navigation cards, and there's a button of choice of the level of the road you're driving, yeah, so, what level of the track you're driving." I remember that I was so impressed...
You mean how much?
Yeah, level-- not level, altitude.
Down.
Down, down, down the road.
Yeah, it was me, too, when I was in China, it just hit me.
I was so impressed, yes, the scale of this, yes. Or the number of people involved in the creation, there, I don't know, 3D models of the city. There, like, eighty thousand people were involved, they built 3D models there, right? Or the possibility of obtaining credit in Aliba, Aliba, in the interface, or WeChat in general. And that people sit in a multitude of these, in these huge applications, in huge applications, which, in fact, do not have such applications in any country in the world. When you have everything inside the same app, and people don't need websites, do they? And people are looking at all the goods and so on through the cards of these applications.
You're not doing a website by default, but a WeChat card, for example, yes, for some company. I mean, this is the one that's hitting the volume. But what do you want to say now? Still, the market is so bad now in China. And when I saw the day before yesterday, they made a charge for an electric car that put 500 kilometres into the car in ten minutes, and you look at it and you realize it's, well, it's very serious. And there's a story about robots. We don't know if anyone knows or not, but China has fifty-two or, I think, fifty-four percent of the world's robots are in existence today. I mean, it's, uh, big enough, I think the number of presences is. There, I'm sorry, I was wrong with numbers. I've, uh, opened myself up to be right. Fifty-seven per cent of the industrial-industrial--- fifty-seven per cent, you know, domestic, in this sense, Chinese market for industrial robots, right? And that is fifty-four per cent of world demand. And they do that in the twenty-fourth year, China had two million twenty-seven thousand, oh, two thousand twenty-seven million, two billion, yes, industrial robots. And in a year, two hundred and ninety-five thousand new ones were found. I don't know, this number is strange, two commas zero of twenty-seven million. I think this comma is a comma of a broken number or a mild thousand. But robots worked a lot. So they're out of the phase now-- they're out of the phase--- well, I just don't want to be wrong. Millions seem to be a very strange number. They're running a training line now. They have a line like that: 10,000 robots they do, 30 minutes after the robot comes out. Well, robots are like Tesla Optimus. Here. Elon Musk, too, I remember, said they had a factory that would make 10,000 robots. But we don't hear them coming out there. I hope they do. I'm all waiting for Optimus to show up. Although my kids, my kids keep telling me that my mom said that if we had a robot at home, moms wouldn't be home. I-- what I told them, where is she going, right?
To us.
So if I need a robot at work, where does she go, where will she go?
So in China, I recommend everyone to be alert, but look. A lot of things come out, and a great number of things, of course not faeces, totally real things in terms of their technology and implementation, and the introduction of automation and robots, and overall development. artificial intelligence and internalization everywhere, in all processes, in all processes. It's not when the Chinese say we're turning a program, it's not a joke, and it's not a political statement. So, we'll remind you, at the level, we told you about a year ago when we were just starting to discuss the strange artificial intelligence that the Chinese government was one of the first ever to discuss. The phasing out of large-scale artificial intelligence programmes, even maybe a year and a half ago, right at the dawn of artificial intelligence, yes, at the dawn. Now, there are some leaders in the world who come out and say, "Actual intelligence is a very important thing." You look like that, you think, what are you gonna do? You're out of the race. Here. But it's the way it is. Is there anything to do with this China thing?
Yeah, in general, I think that we were talking about last time, uh, Anthropic released a new model and opened it to just a number of companies. There's 40 companies, I think. OpenAI also said that their next model was too cool to give it to everyone. We'll only give it to a limited circle. On the one hand, yes, if the model is, in the cybersecurity, it's very hard to close, which is logically to close. But I think that's also China's confession, because it's worth a large company to produce a model, as in a while, all these works are already being implemented in Chinese models. Just if you don't want them to catch you, you can't show these models at all. Of course, it's likely that, but somehow, one of the indirect factors in recognizing what China is doing is, I think, a world-end closure.
And we can see now, horse-- you're just saying, yes, that, uh, open source of the model, if in the twenty-fourth year we saw that the open source of the model was almost out of serious models, I'm not saying that now, uh, that's not it, right?
Well, if you take the frontier, that's the best-name models, then, yeah, open source isn't gonna catch them yet. And if the new models actually made that jump, then the open source is still
far away.
I'm listening. Interesting, interesting, such a thing. Oh, I-- I have some business in there and where I'm involved, in general, in fighting, well, in a calm struggle, but in fighting for people to use artificial intelligence in their work, in their work and, Programmers, including those who are involved. And I see a very clear trend. And it's gonna be interesting to write our comunity because we have very different people: engineers and scientists, and entrepreneurs, starters and just different users from the market. I wonder what you're gonna write. What kind of trend do I see? That I hardly know any man who would have really applied artificial intelligence in his work, but, as long as I can. What does that mean? I mean, people who showed up, like some programming models, they started using them right away. Or some recent versions of super-cooled data analysis in the analyst. They started using them immediately, started rebuilding, started immediately.
All the processes are rebuilt. They're starting to do it. And there's another category of people you're saying, "Oh, not the fact that they're good, not the fact that they're necessary. Not that you have to program with it. Yeah, it's just a hip." And, uh, and by the way, we tend to think that people are gradually getting there. I see that people don't, not things that don't come up like a-- a layer, uh, society has been divided, and it's really very divided. This is, by the way, one of the problems many people have expressed. I think, uh, Horowitz talked about it from Andrison, Andrison. He said that people were like to split up, although he had very specific views, he had such very specific statements, and I don't know how to end up with them. So people in X, they write phrases, and you sit and you think, what does a man mean? I mean, you need to use GPT chat, I guess, to understand it.
Translator.
It was-- yeah, the interpreter, I guess.
But the point is, this story is, uh, why did I bring it? Because I see how one of these was just a case. People said I had more than 100 developers in the company, and almost no one started using, uh, artificial intelligence tools. And in fact, everyone had to be fired, almost all, and hire a new team of people right now who started using artificial intelligence. It's clear there must have been some kind of fitness, and I can see it not just in this area, in the programming area, I can see it again in my normal life. So you fight people, you tell them, you show them, you invest, you do. And that's basically like a misunderstanding. Well, people don't even get a little understanding. It's like you have to, uh, always tell them what's going on inside. This crazy break, just a real break. And that, uh, I think it's a big problem. And, uh, Ilinar, you started using it right away, right? Tanya, I think it's a little slow. And again, Tanya, you started using because you were in Tuzmu.ne, I guess. If you weren't in Tuzmu.ne, maybe you'd be a lot of things, maybe you wouldn't start using. Well, I'm not there for my own taste to buy Pro, yes, I know.
No, no, no, yeah, that's understandable. But I just find out some interesting details that I want to test right away. How would I know them?
And just like it's on, uh, in, uh, the artificial intelligence channel.
You need to keep your morning-- your internal state, right?
Well, you know, I just think that I've got everyone around me. As much as I'm saying, Sasha, 100 percent of you are right, but since we live in this place, it's very difficult to assess reality.
I, uh, when I say--
The Silikon Valley cannot be a world of peace, right?
No, wait.
Yeah, yeah, yeah.
I'm in the Silicon Valley with very few people who know how well they're used. Look, everyone's using the developers. There are no questions among the developers, but among ordinary people, I know very few people who understand the fundamentals and the difference. That's how we, for example, speak from a model perspective. I'm opening up people, you're at people's, Tanya, when you see what they're using, take their phones, see what they're choosing to have a ChatGPT model. If they're instant default, standing instant, it's not a man who uses it, you're putting a cross on it, right? And then test people, open their phones and, uh, your husband's, by the way, is not worth it, because on this Sunday, I personally chose to take another model inside.
I'm just saying.
I'll tell you why I chose. We never used to. I'll tell you why. He paid for ChatGPT because he was flying on a plane, uh, four months ago, and I forced him to pay ChatGPT. So he paid him. But the point is, at that point of time, four months ago, you couldn't pick a default model. Now, if you can, it's still on the web, on the web glue, but on the cell phone, it's starting to be okay. And, accordingly, you could not have chosen. I opened it, he had exactly instant, of course, and he wouldn't turn it off. He doesn't even know what he can do. I turned him into thinking, second mode, because he has a version plus a fee. I'm not gonna get him to buy Pro, though... Okay, but I chose him, so you'll see him, he'll see you in another mode. But that doesn't mean he's involved. He's not involved. I mean, of course, he's not-- I'm-- it's not a thing to do with him. That's what it says, the way it is. I mean, he's right next to me, even we have a business, we have a business with him.
Well, I think that's the reason, Sas.
You do all this business work.
No, he's special, he's using it for everyday life. And all business stuff I think, as you are, it's still, well, not enough to do it.
Yeah, but in everyday b-b-life, if you use instant, you use the artificial intelligence of the old generation. That, you know, that's how, uh, that's all clear already. I mean, there's a big difference right now. You're instant in charge for 30 seconds, and Pro's version is taking your picture, a kidney medical picture or, I don't know, a C.I.
or an M.R. or a 60-minute tooth. There's a difference. I had the last case, by the way. I was told in dental medicine to treat my tooth, and on the way out, uh, I was asking how much it would cost. And since I'm here, you know, in a place that's not cheap, I'm doing myself, what's that called? Gigienic, hygienne, right? There was a price so high. ChatGPT told me it was above the market. Higher, higher, higher, right? I was angry. I sent them a SM-- a message, "Send me all the pictures." So they sent me all the pictures. I loaded it in ChatGPT. ChatGPT says, "I can't categorically see what a hole is to be made here." I recommend that you consult with another dentist strictly. In fact, according to California law, you have another dentist to hold an additional consulate-- uh, consult and express an independent opinion without the use of sales and marketing, yes, without the desire of you. I'll sell something. I mean, he should be independent, although I don't understand how it can be further determined, but how to express an independent opinion. So, I started this on Pro, because Instant's model won't run two zip archives, zip photos, uh, X-ray and inside pictures. And he'll study everything and find what you meant, what a tooth. I don't even know the tooth number for the rest.
He'll study, find and tell, and he'll say, "I see everything, I know, I've already studied it, and I see it."
Still, my Pro model still keeps me on every request for it to be sent to a doctor. Whatever I say, "Don't send me to a doctor, give me your opinion." You see...
I didn't.
Not only Pro, it's also the settings, and your kind of tactics of interaction.
Yeah, no, Tanya, he should actually go to the doctor, because that's right. And he told me several times that I still can't make a final decision. In terms of practice, you should be seen by a man. Another aspect is that I know a large number of doctors who don't look, uh, now visually, and only look at pictures or pictures, right? And they use-- well, and especially you know that they don't work in the U.S., at all, without a picture, do they? They need pictures everywhere.
Yeah, they always make me, I keep putting them away, they keep making me. But we'll move on. There's a very interesting news about medicine. Why do I get that case? I'm saying medicine. A very interesting story now has been a very strong story. And actually, she might, uh, she's kind of a little bit, but I think she's the most important thing this week. That's the most important, just, just the most important. Pay attention to that. So what happened? What, is this news from Nature, from the Nature, relies on-- was a prep for medRxiv. So the pre-investment is a scientific work that is publicly displayed, but it has not yet been fully licensed by the magazine. Yes, the authors of this work examined two widely used dates from Kaggle. It's important, isn't it? I mean, the dateset, the data sets that are used in model training. Yeah? Well, how does the model know, like, how I don't know, to sew this suit, like, or, well, this suit might not make it, but she might be able to answer a lot of things, right? Or what fabric, where does it know, or is that how the model knows my tooth with a hole or no? I have a hole or no sod, how does the model know? She was trained. So, there's an incredible number of pictures, tooth photos, uh, uploaded a lot of research findings. She's been studying it, so... Research is a bunch of stuff, a bunch of stuff, a lot of data. And she can make such conclusions very quickly, right? And that, by the way, Tanya, supplementing is not the use of Instant. Because Instant doesn't do such big analyses. He's responding very quickly, giving some kind of simple, quick decision. No, not tests, not studying, right? He can tell you what, I don't know, the weather is gonna be tonight, right? And it's wrong with the probability of some kind. But the point is they found a hundred and twenty-four scientific articles with risk prediction models. Of these, one hundred and three used a stroke-and-distance, twenty--- a diabeth diabeth date and one used both at once. So the problem is, almost complete absence of birth data. It is not clear where, as in what patients, the rules for which the data were collected are, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how, how these data were collected? In the descriptions of the East-Stat-Stat-Statistics
The source was not directly disclosed in the descriptions of the dates. One of them was the downloader who wrote directly that he could not reveal the origin of the restrictions. In the case of a stroke application, it was separately stated that it should not be used for research. But the point is what? That these models have led to clinical and circumcised use, clinical circumclination. So they're practically on probation for a doctor. So why is this subject very much-- how exactly did the odds find on the strokes? It had 5,000,100, ten patients, 12 variables. He was downloaded more than two hundred and seventy-five thousand times. Researchers found a strange shift in the distribution of the average glucose level around the middle of the patient identifiers sequence. Well, it's-- it's technical stuff. Let's not go in there. But a little-- but the point is? And, by the way, diabetes. 100 thousand patients, nine variables, more than 100 thousand scaldings. And there's more powerful anomalies, yes. Here again, there's a glucose level in blood, there, and other indicators, there's medium sugar over a long period, and 6,000,900,339,250,9 duplicate observations.
So what does that mean? That we are living in the world now, we have, by the way, raised a lot of this problem, that artificial intelligence, he's learned some data, including he's learned from the Internet, including him. He's been learning about various books, the same thing-- he's learned about scientific articles. And you and I all know that scientific articles often, uh-oh, become untrue because other scientific articles are emerging. So he's learned a lot of things. Not enough, some country thinks it's a war, some people think it's not a war, he's trained in it, too. So he's learned a lot of things in his rules of the book. And when we get the answer, of course we don't see or understand, on the basis of someone, he makes a calculation. Even he makes references, as he would say, "I've referred to this data set, I've been referring to this research, so I've been here to refer to these things." There are, and there are details where he doesn't mention anything, because it's just his internal training, like his teeth. He can't explain the whole chain, like he... he's got a hole or not, he's still gonna give me some final layer. He's not gonna tell me, "Alexandre Igorevich, I've been here, so I've been re-examining, uh, eighty-nine million teeth from above two, and, uh, I already know, I've seen it, and I already understand it, like, like, like, like, like, like, like, like, like, like, like, criteria.
So you gave me a-- Though he might do some part of it. This is, of course, a very big problem with the future in terms of reporting and a very incredible problem. So it's not enough that models have yet introduced their rules and their own logic, like in the example, uh, as in the example, uh, war-- to start with Iran. One model says, "Yes, I should have." The second one, "No, I didn't. And the third and fourth, "We'll abstain from answering." I mean, we see a different-- there's no true, wrong, right? In Anthropic, I'll remind you there's direct indications for the model that you shouldn't be too Euro-pedited, right? Well, she's already got it in her neck. But they have this explanation that the majority of texts that are laid out in the modern world, the world, they have such a gradient, yes. And that's why you need to remember. What do you think, uh, Illnar? I mean, you must be in a job with this. I'm facing this very much in-- and I've always been in business, in analyst. And always said that people often draw conclusions about the conversion on the website, make them somehow prejudicial. And often, it can be proved that conversion you-- I always used to lead to this example that I can always prove that conversion has grown and fell simultaneously. Question, uh, what are we, the we're discussing, what time we've been reporting, huh? And in analytical data, deep- deep-- deep-sea research, I've got business, at least I've always seen it when people cheated, they've come, they've made conclusions based on some very very very very very good. a narrow local view. And here we are, we see it in serious calculations, and that's a clear problem, right? And she's not gonna decide. Ilinar, is she making up? I think she's indissoluble.
Look, the problem with data has been very acute for recent years, in fact, not even a year, but more because it depends on the quality of the models that learn. If I remember correctly, this is the red code you were thinking about with Gemini 3 when he came out, their quality jump was in part just because they cleaned out the date very well. I mean, we're basically collecting all the data from the Internet we could reach, then we're cleaning them out of the trash, from the wrong data, from among other things. And then, respectively, the following stages of education are being conducted. So, then Google was just able to clean up the main part, to produce data and to get a good jump. Indeed, so many companies and OpenAI, and Anthropic, and Google, and xAI, all spend a lot of money to get those data. So, I think we talked about the OpenAI paying a lot of money and lawyers and people who worked in banks in the financial field, just to learn models about the data that they're producing. experts. It's costing companies very much, but somehow it's hard to get good data. With regard to the articles, it's very nice that you explained that it was a prep. Still, the preprint didn't pass, uh, licensing. And if we take the scientific field, then, uh, why can you believe in science articles, at least math? The math of the scouts doesn't usually come back, that if someone has proved something, and that theorem went through the licensees and is still published in the high-level magazine, it's not very likely that. It'll change. How math would be in this respect, a pretty good science. And here, you see, when we're talking about a scientist, about math, he's looking at who the author is, he's heard of him or he hasn't heard the magazine that all this is printed. Has he seen any other articles of this author? Is it not confusing whether the author ' s reputation is there, a flamear or something? And there's some kind of set of indicators you're basically already named by the author, by the title of the article, you already understand that it's possible to read or better not read while there's something else more interesting. That's how people do it. If we're taking modeling, it's going to be, like, a total mock. So you're the whole Internet that is, you're reading, and there's not only scientific articles in high- rating journals, but there are preprint archives, and there's a mob of creativity and everything, all, All of them. That's what's going on. Then, as we fed the models, the company has a separate headache, so that it can be perfectly correct, yes, and then clean up all the stuff that went through the model. But it's not gonna be cleaned. Always some stuff that's coming up. Before, by the way, remember, there was a lot of talking about hallucination. Now the hallucination thing has gone so far, it just seems to be getting a little bit less sick. I think you're meeting hallucinations nowadays, if you're dating.
By the way...
Although it matters, it does. Everyone should remember that any model, she-- I've done a lot of research, and I'm bringing them somewhere. I always say, our task is to get a conclusion automatically through artificial intelligence and without human adjustment. But there's one nuance. We all have to be calm about what might be a mistake, because in any model issue, it's so arranged in the first place that she can make you a mistake. I mean, it's not guaranteed 100%, you know, it's not a calculator. That's how I understand the calculator's mistake, in simple actions, to give, you know, he can't, like, impossible at all, right? I mean, guaranteed-
If it's not broken, yes.
If it's not broken.
It won't.
Yeah, I guess I'm just not. It's just that the weights, for example, might have a misunderstanding, there's zero, zero, something, yeah, there, and so on. I don't think there's any error in the calculator. I mean, she doesn't, yeah.
Now, in the comments, we're gonna find out that when you're sharing, you're not exactly the number, but something close.
I said I said that. I didn't make a simple deal.
The domestic use is exactly the way to believe the calculator.
Yeah. In simple action. I see if you're gonna split up a whole lot of numbers and multiply, you're gonna lose some piece at some point in time. No questions. I think that everyone here understands that. Or you start using complex, complex functions different, yeah, all the floor-- but still, right? So the calculator will always guarantee you the answer. That's one of the problems in models, by the way. People don't understand that there are simple things that models don't decide. I mean, the model is winning the math Olympics, but it's still analog watch, I understand, it's bad. Although reasoning model, I think it's a good way to sort it out when they say it's not a good time to sort analog watch. There were 50 or fifty-four per cent of the Stanford study. I mean, I'm just saying, "I'm sure it's a level model, maybe not a reasoning, but, Ilnar, I guess?
All these examples, like, what are the letters "O in the word milk or "R in the word strawberry," when in English, this is, uh, a long-distance phase. Yes, it was, but as modern models are easily found.
Yeah, but again, the model might be wrong at the end. So she can give you the wrong information. And about the dates, like you said. I mean, there are things that are very strict. Well, for example, if we're measuring the breadth of my desk now, she, she's stable, she's very constant, understandable. But if we say, for example, that all people at home have tables, we can theoretically assume what the average size of the table is. But it's just a hypothesis, right? Because it reminds me of my understanding of the size of the bed as a child and understanding, for example,
of Americans.
- childhood and understanding, for example, of Americans of bed size as a child. Here we are all the grandmothers here saying that they have a very clear sense of the child having a separate room. Not just a room, but a room with a toilet and a shower. And our grandmother told us that she was like her, so dad was in the house specially, so when I started, I was in there more than five or more years old, and I got a separate room. My dad made a shower and a toilet in this room. I don't remember anything, so when I was a kid, they'd think about having my shower and the toilet. It was-- look, even if there was a place for someone to think about it, well, I mean, actually, here at my parents, my parents were building a house there twenty-six years ago, and they didn't think about every single one of them being in each of them. The bedroom must be a shower, a toilet. They thought that bedroom was, I don't know, the daughter was coming, the bedroom, the son was coming and the bedroom was a little grandmother, which means I remember now. But no one thought there was an independent shower and a default toilet in this room, actually. I mean, there's no such thing as thinking-
It's also coded, Sasha, by law. So, how we have some standards building, they have them, too. And the bedroom is considered a bedroom only if she has, you know, a closet, a pretty big wardrobe, and a bedroom master, only if there's a toilet. I mean, it's not just thinking, it's also legislating.
Go, go, go, go. And it turns out there are different concepts that are standard. Different people have different perceptions. I mean, if we're talking about the middle table, ChatGPT's gonna ask me what the table is for my purchase, and I'm not supposed to answer ChatGPT. The normal model, she has to tell us what conditions you have, how many monitors you'll have, what computers you'll do. Do you want to put a tea board on this table? So I have a favorite case, yes. And I'm at a place where, for example, a house where there's a working space, there's a small desk, and I don't have a tea board there. I'm used to having a huge table for which 15 people can sit. Of course I'll put a tea board in there. It is understood that the system, in recommendation , is based on what an abstract opinion is. And there's no question about research.
And even if someone released, and even if someone released a scientific study and said that this is considered the best table and was assured, Illnar, he's being assured, it's also gonna be averaged for a certain type. People. Here, the question arises as much as ChatGPT, giving a recommendation on the teeth as it gives it. I was studying the US insurance now, and I liked the way ChatGPT didn't try to impose some kind of insurance on me, but-- well, maybe because I asked, but he was trying to tell me about it. And the options and the difference. And he even showed real pros in super expensive options, not least. Because often when you ask people, people are based on their sense, and they, for example, are some options, super cheap or super-- well, these are extremes, right? Or super-dogs are default without explaining the details. And he explains that you can get that plus here, but you'll get the risk. You'll have to think when an ambulance is called when you're in emergency. I mean, in the U.S., for example, I've always been in the US when I was driving, doing a tourist insurance on one question. I always thought if I had something scrub-speed, urgent, I needed an ambulance insurance. Yeah? And I have this habit. And we were talking to Steve yesterday, Tanya, and I'm very much being an ambulance. And, well, we have different types of insurance and different approaches. And he says, "We're in the way because." I said, "I have-- you have an early problem." He says, "Yeah, problems." And for me, an ambulance has a bigger priority than what he has for him. I mean, I know that the ambulances can, uh, not bury me, but there's a lot of money I can get out of here. I mean, I'd rather pick something or turn around in other places, but I'm not gonna have a problem with the ambulance. I'm doing the same with the dog. I only have insurance on her to make sure that if she has something urgent and needs to pay, I don't know, $80,000, I can make it easy through insurance. I mean, I don't have to pay $80,000. I mean, I-- I don't care what my monthly payment is, it's $50,000 or $150. I mean, it's just a different approach for people. And here's the question is how models teach a good person to be backed up and ask him questions. And will they, Ilnar, do that? And Ta-Tanya, huh? Will they do that? Will they be interested in that? If they try to understand a man, and for that man, yes-- that's how Tanya says, "I want to get a response from him, and he's all asking me to go to a doctor." I said, "I got it. I signed all these consents with you that you won't recommend to me, but can you still give me a quiet answer from your databox? Yeah, I'll consult with the doctor, of course, I'm not us-- I already explained to you, I'm not stupid, I see everything, but you're gonna write me a safe answer." Yeah? I mean, how much model-- and some people can't do the same. And that's their job, models, right? I know that man can't be given. Maybe Tanya, he somehow thinks you're... I was just gonna make a joke.
I can't, can I?
That you can't.
I need to see a doctor.
Yeah. Maybe you should be in a doctor.
Maybe he's got a psychotic
idea.
Yeah, yeah, yeah. I've determined, I've identified psychopathy and I don't. Here. But it looks like a story that's not really hard to say. I finally built yesterday, by the way, before I could raise an independent agent permanent. And I'm sure I'm hit by the easy construction of a new one. I mean, I'm doing a lot of things, I know, I'm doing, we're talking a lot here. But I built a more, slightly more complicated environment yesterday, in terms of what was API Telegram, there's some sort of shipment. It was so easy. And I was in one business this morning, and I wanted to fire a lot of people. But, by the way, about firing people. Did you want to fire someone? Did one of the people watching us fire? I-- OpenAI put it in our description, we'll talk about it at some point in time, right? But the OpenAI, in his prediction of the future, has set out a story that in his prediction of the future, he has set out a story that, uh, it is very important to remember that AI should not be introduced to make people feel better. Fire them, but to reinforce them. And I'm always for that. And I'm just-- who did I want to fire? Where people fight very hard. Where you know that man, you stitch him, don't stiff him, give him models, tell him you can do something. He doesn't do that, he's not trying to figure it out, he's not trying to figure out how it works. He doesn't understand how hard it is to do it. But you say at some point, "Okay, then, that means that it's gonna be an agent. Here, you know, like a piece of work to do." Ia, uh, that was very easy to build. And I liked the Codex, and I asked him, "Can you send it to the WhatsApp, Telegram or iMessage?" He said, "I am iMessage can send it from the computer, but it's much better if I'm through API I'll get in and send you to Telegram. Then you'll be independent." I like that answer super well. It wasn't, uh, I understand it wasn't Pro's version of me, it was answering me five three Codex models, and I think even five four.
She's
good.
Yeah, I'm not. And that answer was beautiful. I mean, first of all, he showed me all the variations. He didn't even consider me whatsApp. iMessage he said he could do it through Spotlight sometime inside. That was a subject, uh, very nice. And I'm pretty quick, so, in a minute, 15 made something, wrote it. He warned me immediately about the problems, and he warned me about all the risks. I've got too much of a deal he's been asking for. I've been there for a long time, smashing a button to give him permission without knowing what I'm authorizing. That's the programmer thing, right? It's purely a programmer-
Don't repeat it at home. Professional.
Yeah. I mean, he asked me for fifty times, but I just stopped reading them. I've been clicking "yes" all the time. Well, uh, then I really, then I can tell you honestly, I started a double check on him, and he didn't get access to where he shouldn't have gotten. And that's so-- and not enough, I then unloaded everything he did and loaded it into Pro's version, Heavy, Extended, in ChatGPT, and gave it to the study of all the risks. Here, uh... But I just don't want to get a problem. Honestly, honestly, I'm not gonna want to get a problem even if there's a potential leak. But I really liked the job. Write what agents who did exactly, well, I'm not talking about programming right now. People who are not programmers, who are... who use things all the time. And I liked the ease. And here Ilnar remembers these Carpathians who wrote a sophth in three hours for the camera. I guess I realized yesterday that, well, I'd probably never have gotten any more into it, so I'd have to write video cameras. I may be a little technically aware, but at the same time I'm already back. I mean, I don't know how to turn infrastructure. All that matters is this girl and everything, I don't know about it now. Write down who--
Huh?
We'll tell you next episode, right?
Yes, sign up for the canal.
No, it's simple. Yeah, here, here, really-- I think that's--
To keep it from happening, sign up.
And he wrote, by the way.
Sign.
And he wrote, by the way...
I did. Keep your intrigue.
Here. Ahhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh
Here, please, I should have kept the affair until the next time. You told me everything.
Yeah. Uh, but I'm saying, uh, that's, uh, that's a little bit of a petty, bullshit, I'm not even gonna tell you about it, about these examples. Uh, I was just for myself, but I was-- why was I really happy yesterday? I've had this kind of pause for the past two, uh, months, I've had a pause. You, Ilnar, professionally as you walk every day, and you're going to... you're going on, so, uh, all the stuff to raise technology-cool. And I had a spousal in terms of the agent creation. I kind of stuck because I didn't understand, and what kind of agent I needed is more than just, uh, in the GPT chat, some kind of reminder, yeah, me. I'm kind of-- why would I spend the tokens, like this? And here, uh, I just launched a case that I'm really using all the time, and I immediately expanded X5 or X10 to use yesterday. And plus I saw it just now. I'm, of course, uh, thinking I'm gonna rip out for new cases now. I had a case, a stock-related thing, an analysis of certain shares, and I'm in the middle of a... where I'm checking them, looking at the investment market, there, yes. It's just like a re-- not a search, a reservoir. I'll continue to do the ChatGPT 5.4 Pro Extended. Yeah? It's just in terms of some support for me. And, uh, how to take out what time it takes to take. And I saw a lot of, of course, super-steaks that, um, I'll add. You share the case again, it's very interesting. Everyone who does what we have is very different, different ages, different people and who has tried. And what I was personally concerned about in the Codex is that there is the same ChatGPT, Codex is available on the same plus, uh, and that's where I understand it's both web and Mac.
Clearly, who uses Anthropic, you have the same instructor-- the same toolkit, right, there's a thing? Can I, Ilnar, let me in? It's coming in.
There's something to say, Sasha. Uh, it's a good time you're talking about it because until the end of May, whether double or even more extended OpenAI limits are for use of Codex. So among our viewers, if you wanted to try, the time with Codex is to move while the limits are raised, there is a possibility to directly test the whole function that is there.
That's a big story, isn't it? And while there's access to this and access to cool things, cool, uh, cool technological things. I realized that I could create a lot of things in Pro. Yeah? That's the point. Uh, probably, of course, on the idea, it should be much more expensive. I was just saying that now the Codex is, in fact, free of base, base. And yet I can be there for five-- five hours enough messages to send. There are some models that barely make thousands of messages, yes, there's about 10,000 messages, there's 5,000 messages. That is, it's a big yea in terms of the agent work of a regular staff member, a normal person. I mean, I see what I did yesterday. I see it's very cheap, it should cost more.
Mm-hmm. Here, Sasha, look, here we come to price and value. Ah, the value is very high. And the price, in the sense of cost, was a study that the lion ' s share of company costs, it was taking on new models, new purchases of new generations, the collection of dates, etc. Inference is on its own, so we've trained the model, and this model gives us the market and takes the money for what they are, there, I don't know, catering, asking questions, getting answers, and so on. This thing is very profitable. If companies could stop, uh, not learn new models, not buy some super-crunch new videos that NVIDIA is producing, which other companies are producing. If this race stopped, these companies would be profitable, because inference itself is very profitable.
It's a fact.
Yeah.
That, that's even a fact. Yeah, yeah, yeah.
Yeah. So, if you say the cost of it, it's just the company's response generation, it's pretty low. Here. So, let's see how this is going to happen. We'll stop someday and just use the current models cheap. Either this race will be forever, and the purchase of new data centers, the withdrawal, uh, computers into space and other things will be permanent. Sign on the channel.
I'll see you in a week. Bye, everybody.