OpenAI, code red. Two and six billion OpenAI users. Mistral presented a new model Mistral 3. DeepSeek has new models, including mathematical. New OpenAI methodology for the integrity of artificial intelligence. Will NVIDIA disappear? New Google cards. Grants in artificial intelligence. And they're generally easy enough to get. Oh, the Sping Module in OpenAI. A week's update. Oh, there's a commercial at ChatGPT. Or is it not a commercial? The UN Report cautions: artificial intelligence can widen the gap between rich and poor countries. Should we invest in artificial intelligence? And the conclusion is unpleasant: none of the leading firms has a clear plan to keep artificial intelligence in the rig.
Hello, everyone! We're on ToTheMoon. We're going out every Sunday. News every Sunday, sometimes special Wednesdays.
Oh, we're two today. Yeah, we're short tonight. So while, uh, everyone's on their way, and Tanya and Sasha are on their way, and Elnar and I are on the run. So, uh-
I'm right off the plane, Sasha, too. So today we have a day like this.
Yeah, I'll be in, uh, somewhere we're going to New Mexico. So, uh, c-- OpenAI, code red. Uh, OpenAI, we were telling you that OpenAI had released before the Google Gemini model, uh, last Gemini 3, he had an internal letter saying that, uh, Google had a too serious model and it. I'm gonna have to get ready. And now OpenAI has declared the code red because of competition from not only Google, but also from Anthropic. We know that Anthropic capitalization has reached $300,50 billion. Anthropic's already made a billion dollars worth of money. And although, of course, capitalization of three hundred and fifty billion with a billion--
With a billion.
Yeah, with a billion-dollar, it looks funny. Oh, and that means, uh, that's a red code leak.
Yeah, yeah, yeah.
Yeah, that's pretty good, by the way. And we'll tell you about the stock tonight. And I did, by the way, read Anthropic yesterday and thought that, uh, interestingly, they'd be faster or not, because they're kind of officially speaking, huh? Well, or at least this information is much more seriously discussed. OpenAI also said, we've been talking about it, there's three weeks, that they can go out on the market, because maybe if Anthropic goes out on the market, they can start, uh, on the stock, at least more than that. OpenAI is worth it.
Yeah.
And by the way, this leak, which was that the OpenAI leadership gave an internal alarm signal and they gave it two parameters. The first thing about this is the new Google and Gemini model, and what made multi-million dollar investments in Anthropic. And it's a fact. Anthropic made very much investment. Anthropic, uh, buys, uh, various start-ups, makes a huge number of integrations, very serious integrations. I was surprised, by the way. In OpenAI, the head responsible for education is the former commercial director of Coursera. Now, the news is that the coursera has been very close to Anthropic, not to OpenAI. By the way, Open, Anthropic has also now opened a separate and huge programme for non-profit organizations and for registration, uh, on the lung-- high-price tariffs of its system for non-profit organizations. Or-organizations. And if you have a non-profit organization or you do some of your individual projects, then you have an OpenAI, and Anthropic offers different advantages.
Here's O-OpenAI, opening the program, I don't know if we've told or not, for all the teachers in the U.S. OpenAI gave access, uh, free, I think, until the twenty-seventh year. I don't think we've told you about this. This is a very interesting initiative. There's more than a million people out there, I see. And it's basically a separate personal office. This is a separate, distinct visual environment of the GPT, where the training data can be maintained. It's isolated, this environment, it's possible to keep the data on students who study at school. Uh, you can invite, like, to collate other teachers. I wasn't in this environment, but if you're-- but it's only for teachers in the U.S. I mean, if you're a U.S. teacher, you get access. Wald is crazy, of course. Give free teachers. I would give all teachers free of charge, because once teachers start using it in their work, they start telling the kids. And all the new generation starts, uh, living with the understanding that there are GPT chatting, yes, not some other-- not some other system.
Yeah, but it's like, you know, when I went to university, Microsoft's grant program developed, and all students at university were free to get a license Windows, licensed, there, a package of programs. for programming, docking and other. It's all totally free while you're a student. The same thing is JetBrains, yes, JetBrains, she's selling her products free to students. I mean, like you're giving free, but you're taking a lot of users to use it. It's convenient. Uh, you're used to it. You're doing a homework, you're doing the classrooms, and you're doing everything in these products. And by graduated from university, becoming a solvent man, you keep using it.
Yeah.
And basically, that's how you say right, that teachers gave it, it's not just teachers who will use it, but they'll also be bringing their students to it. A few years later, the audience will be prepared. So when she becomes a solvent audience, they'll actually use it.
Plus, OpenAI will get crazy data sets to develop their internal education systems. They'll definitely do it. So we see they have a big direction that goes to education, separate directly. There is also-- just like the world, if we look at the big lines, there are trade, food, aaa, e-commerce, different. They're developing a shopping spike, yeah, really. And they're developing a lot, uh, education inside of themselves as a separate, separate line, separate from the modules.
And that, uh, uh, uh...
Yeah, the exact same news was about one dollar if I wasn't wrong or even free. They also provide access to GPT Plus for the pre-pension U.S. military forces. And here I am, uh, not quite clear about what it is. Yeah, it's a political move. Just show me that we're looking at this category of people. Yeah, but there's an explanation that there's psychological support, there, business shifts, retirement, and so on. And in a year, there, aaa, before the pencil and then, in my year, uh, very favourable conditions for the use of GPT tea. I don't really understand, it's pure, there, political pier or something, because with the teachers, we talked about it as if it was, uh, company motivation. It's not very clear here, but it's possible...
Well, I think maybe it's just--
...American history
...you know how social support is. Yeah, how--
Well, socially meaningful, that kind of thing.
But by the way, aaa, the non-profit OpenAI Foundation announced that 40 million grants were being made to support more than two hundred non-profit organizations, and they would support local projects aimed at supporting local projects. Digital literacy, there, social justice, sustainable development through artificial intelligence. In fact, grants will be awarded to organizations that teach young people of media literacy, uh, and use artificial intelligence to support, there, health, communities. Fuck-la-la. So, aaaah, that's the side. And there are some interesting grants in parallel. That's what we said, and you and I were talking about a grant, too. Now, for example, they're giving grants now, in my mind, one million dollars a whole-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a- They said, "We want to study mental health," that is, mental state, uh, man. If you're scientists, aaaa, or you do, uh, therapy, uh, in this area, you can apply for this grant. Grants are granted, in my opinion, from 5,000 to $ 100,000. Uh, and until December 17th, the reception, aaaaa, the applications. That's exactly a week, I guess you'll have a grant. And in general, grants are different now for many companies, you can look for them. I mean, if you're involved with artificial intelligence, it's clear that they're the main ones that give to companies that-- and people who use it as a...
Who use it, uh, in a non-profit manner. So everyone who uses a non-profit basis will give them all that.
So...
Well, let's go back to the code red.
Yeah.
Uh, we haven't been able to solve it. A-Altman, on the other hand, er, has actually announced such internal staff mobilization. They're shooting staff from the unresponsive directions. There was something that Pulsa was filming, and it was a line. And it's all going to get the lags off, uh, Google nivel to release new models as soon as possible, get back to the top of the baccalaureate and show that the OpenAI is still in the process. Maybe we're not overwhelmed, that we're still in the world. Oh, you're on, uh, what's your announcement? The leaks of two new models, this is about the Shallot Pit we talked about before. Garlic still showed up. They're all vegetables, right? They have a onion, a onion, a onion, a slug, and they have garlic. That's the garlic coming next year. Shallot Pit should be out in, uh, this year in December. There's literally a little time left, but there's GPT 5.2, 5.5, some sort of thing they're probably gonna let out and hope they're gonna show up on the baccalaureate.
How do you even like Gemini 3, Sasha? You used it? You didn't use it?
Well, I haven't used a piece for a while since we talked about it and stopped. So, yea plus used it, in particular, because it's inside NanoBanana. I mean, there was some pictures in NanoBanana. Although I think what I did, well, I did some simple paintings there, and I could have done it in the GPT chat room, right? But NanoBanana, it's better, I see, with the text is much better, much better. A lot of things, uh, uh, cool, faces are changing somewhere, but quality's kind of distorting, something else. Plus, I don't have a superpowered interest in everyday. My teams, which-- where, for example, it's a daily use, so I'm not doing it, they're already doing professional people, editors or designers that do it, right? Even in particular, the cover. We don't create cover for the canal ourselves, do we? I mean, it's still a professional person. The question is how much they are included. Here. So I don't think I can see, for example, in-- while in systems like NanoBanana, like VEO, it's videogeneration, yeah, who doesn't know, like Higgsfield, like videogeneration, like Sora.
I don't see, uh, relevance to hundreds of millions of users in terms of out-of-the-art intellect. And when OpenAI says that by 2030, they will be achieved, uh, 2.6 billion people. So, 2.6-- they're predicting they're gonna have 2.6 billion users, right? And that, about 8.5 percent, it's 220 million people, uh, it's gonna be on a payroll. But I don't think it's interesting to have a question of a pay-off. I think it's interesting to know the whole question of 2.6 billion. Because 2.6 billion is, in fact, uh, one of the biggest, you know, it's one of the biggest pay-as-you-go services ever to be, right? In general, one of the largest services should be used. What is 2.6 billion people? But at the same time, the break-up isn't that big. They're already a billion. They'd probably say they have a billion. And, uh, four years, it's really quiet. Well, that's very calm.
Yeah, yeah. They'll have their models.
Yeah, we have to understand that people will use systems without VPNs. Whoever said anything, but if there is, uh, VPN, it's hard to open, uh, models through VPN. I mean, I know there are advanced users who do it. We have audiences, write who uses, for example, VPN in models with any limitations. And I'm curious to see the base statistics. Don't forget, we don't forget to sign and we canal, too. Ah, and that's really helping us develop. Here. Did you want something about VPN or something?
Yeah, that's why I'm just waiting for a comment from the series, there's a "good try, comrade major."
Ah!
So everyone can write what they use.
Well, I don't know--
But I'm waiting for that comment.
If in your country...
I hope he's coming. Write it.
If your country is banned from using VPN, if your country is banned, it's just that we're being watched by the world, yes, even by the polls. Well, don't write, okay? And, from the point of view of the system, it's more obvious to you, you choose. Uh, but from the point of view, um, even where, for example, ChatGPT is blocked, it doesn't mean there's an internal order that it can't be used. You can use it-- well, you're doing some bypass. Although ChatGPT warns, yes, that, uh, we are, like, a few countries where we don't work, and if we find work there, we can block you like an account. And in any case, of course, for many countries now, this is a serious limitation. I mean, when you, for example, I know that when I'm in the US, I get access to, uh, all the updates right away. I told you I came to Europe, and I lost a lot of responsibilities. But while I have an American account, I'm only missing because I was in the IP address, uh, European, right? Not to mention, I mean, I've been confined to these things in artificial intelligence. And by the way, I wonder, Ilnar, how will they decide? Because it's a problem. I mean, if I'm used to working in a certain infrastructure, why should I have to go on a mission to Europe, why should I have to lose, uh, a certain function? We'll see. We'll see what happens. Uh, anyway, uh, two and six billion is a very serious, gr-- very serious number, very serious number.
And here's the question. Look, if there's Google Gemini and there's ChatGPT, even if we clean up, there, say Anthropic isn't a mass use, say... Not a fact, though, is it? Let's say no one else has come. Apple didn't show up, XAI couldn't do it. Uh, and there's Chinese models, right? And Chinese models, let's say China is sitting. Where's Google? How does Google share with ChatGPT? I mean, there's-- because people won't use, and, uh, permanently in their work. What's the point of the system? I don't know how you got it. Write it up. I wonder, but you asked me a question about Gemini. And I-- you may be more professional, what you're programming, you have to use a lot of different systems. I don't programme, but think why would I have to ask Gemini if I have a ChatGPT? And even if he's somewhere some kind of baccalaureate or something that makes things worse, it's like I'm a -- I've got a stockpiling story, I have a stockpiling, there's a habit of using a certain solution, but I'm not sure it's a good idea. I'm kind of accustomed to him and I'm sitting with him, right?
There's no point in moving.
Although I-- yeah, there's no point in rolling around, there's no point in rolling.
It's like we said, yes, and now it's all the news that Gemini's gone over, but now the OpenAI is getting a little bit hard to catch. It's probably gonna get over, and the difference isn't gonna be that big, even if it's there. And it's not an excuse, just once that they were staggered into a binchmarc, which other model, which other company has gone through it, not to cancel a subscription in one place, to run another one simply because in a week, Two, well, it's gonna be the same for a month, right? And then, in fact, it's gonna be going on. But when I asked about Gemini, I was just wondering, right? Because I've been very active in Gemini for the past two weeks, and the Anti Gravity, that's the daughter that is. And I really like it. With the results of ChatGPT, Gemini 3 Pro results I like. I think they've got a very, very good model. And I understand why the code story is red from Althman because Google's real competitor has been prepared. Well, not even a competitor, but as long as a temporary leader, huh? So, as long as the OpenAI didn't say the word you said.
And I want to open the Pandora's box about what you said to NanoBanana. I'm a skate-to-kate, remember, a little bit of a wolf, all the pictures. We'll get a couple of generos now, too. I think if we can do this well, we'll put it out. And the audience can be offered, too. Try to generode pictures in NanoBanana. You can probably just squeak, throw it over and see who's doing it. If you can share something with us, maybe we can put something out. It's just a series of interesting things, and how can NanoBanana be used, except that we're gonna be doing some examples right now. I tried to use it as follows. Here's project, there's a, uh, there's a hundred lines of code. A complicated project, a lot of interaction. And when you need to show it to someone, you have to show some sort of pattern. To say that there are these modules, such blocks, that's what's connected. This is where the entrance is so transformed, there's a way out. Well, it's a parole, there's some files on the entrance, and there's these files on the way out. That's the magic that happened. And to keep it from saying it, it's a good scheme to do. I was just trying ChatGPT and Gemini, in Gemini and NanoBanana, to put out a verbal description of the project I'm working on, and to get him to spit on the beautiful scheme I can show a man.
A beautiful scheme I can show a man, and a man can just visually understand, and what kind of project is how he works. So Gemini and the nanobananas are doing better than the OpenAI with his model, which is with a totally identical prom, which was there, and they're gonna drop it. Gemini's results are a show, the results of OpenAI are really embarrassing to see him. There's a difference in this plan. But again, I repeat, OpenAI will catch up, or maybe even a little overtaken. So if you're not critical, you don't have to run anywhere, the whole market will be fine.
Look, this is an interesting story, which is, in fact, if they can back off? Well, let's be honest, maybe they can. Take a case of Google and Microsoft shares, right? When Google's over now, he's over Microsoft, and we don't know if he's gonna overtake NVIDIA, right? And Apple and NVIDIA move forward. Can Apple release the model and take off and rip everyone off? In theory, it can. Given that Apple has a crazy infrastructure. Not enough, they have unique chips and phone chips. The iPhone chip is a real chip that can raise just crazy things. Yes! And I wouldn't have put them away, either. Yeah, I wouldn't put them on hold, either. Can Google be so much more of a quality-focused and aGI do?
Of course it can. Take Meta's example. Remember, we were two years ago or, uh, there, and a year and a half ago, discussing Lama a lot and different models. Yeah, they were there just recently, and they were just showing up at the conference, too, an update in the open source. But if we really see and question even the technology world, I'm asking, everybody says, "No, Lama, we've already forgotten, there, or the Meta model, we've forgotten." Yeah, they have a new name there, too. And we don't use them. And I'm saying today that Mistral, like 3 has produced a new model and we're gonna have money in Mistral, too. We haven't talked about them in a while. They're French. And it was basically the only model that was present, well, the big, big, big player in Europe. We never know who's gonna get a good one. Just open AI, why can't you just move back? Because they still have enormous power, and they're still a leader in the world in numbers of installations. And that integration is still free for everyone. And they have the opportunity to take over, and I don't know, fifty million people or a bunch of countries in the world to give free access, for example, yes? I'll let people open up. People will eat resources. And, well, that's how Anthropic, right? It's worth 300 billion. It's only a billion. Milliard is a little bit of a profit. Well, fair, huh? What's a billion-dollar? You can't always be invested.
Mm-hmm.
There's a problem that if someone didn't want to-- that was two weeks ago, the market just swinged down. Well, there would be some red flag on the market, something happened to the economy at this point of time, and those companies could be very serious. Just. Just, like, just to wake up. And now that's the Bank of England, too, describing it as a bubble. But we, we are just a little bit more than that story today, too, in terms of investment, equity investment. But I'll say one thing that if I'm a professional user of GPT Cheat and my wife, there, I don't know, semi-professional, but she's always using GPT for a year and a half a day. Can my wife, for example, cross? I'd rather go to Gemini than she did. Well, that's obvious, right? She-- she'll be fighting this GPT chat, but the idea is, if people around her start saying that there's a better system, and she'll move on once, she'll be in the new system for the next and a half years. I'm not. Just like me, I'll try different systems. But if at some point of time I see that, for example, Google is very winning, it certainly wins, I'm, of course, not much, I have a fee-paying version of Google automatically, with-- not a maximum. I don't have a Ultra version right now, do I? But I have a version of it because I still use Google, different services, I have it automatically. That, by the way, Google's advantage, yeah, that you buy, I don't know, cloudy vaults, something else, you got the version, and you still use it.
And Grok, for example, I bought a version of expensive heavy. Took her months, she was worth, I don't remember, there's $300, right? Or two hundred fifty. Some price was more than everyone else, but I didn't see it there on top of something and didn't go on with that version. And basically, I'm just not spending $300 in Grok next time. Only if I have, like, Tesla, I'm probably going to Grok. I'd like to share that I have a lot of American services that have the integration of the artificial intelligence of the modern one. Well, it's important to understand that you're still in the States, you've got a lot of cool supersystems integrated. So I made an application to Starlink yesterday. It's, uh, satellite Internet, satellite Internet, I have it, and I had to get one of the equipment and something. I wrote in support and I got a response from them, and Grok works there, and I got a response from Grok. I wrote, "So call the assistant." He tells me he's answering me, answering and answering. I say, "Stat Ticket." Well, I've learned. I know that if you're talking to artificial intelligence, you should write immediately. You have to give this conversation to the Ass-- Agent, right? We'll have to do it in Uber. And he takes, look, creates a ticket to the agent, but he watches the ticket make. I saw it for the first time yesterday. I'll tell you honestly, that's the quality of the ticet. I made him a request, he gave me an answer and didn't want to solve my problem. Why did I call an agent? But when he created a ticet, he created a ticet not based on his findings, but on my real request. I mean, he understood, he understood. He tried to make another decision. He told me to wait, it's okay. I say no, I'm scared. You've been here before. He took everything from me, made it from this summit, and he really created a very cool ticking, added all the data from my order. So he didn't open a window where the agent came to me and asked me again. You know, this is tingling when you start asking you in your own order what your phone number is or what your order number is. Well, that's just the same information. He created everything himself, and I got a soupport from some people in two hours, I hope, man. That's not known either. It's not known, either, if the man answered. Because, in fact, whoever answered me responded to me was very similar to the Grok's response, only more detailed, right? Well, maybe he didn't have another solution, but I just got a book on it, and they were just trying to explain to me that, well, whatever you're worried, he's gonna get together. That's why you're going to be like this. Well, by the way, Ilnar, I'm the one with Grok, who led when you see it, you really like it.
Like, we're on Amazon, I have a lot of columns at home, you know. You saw it, didn't you? These are all Alex's columns, there's a lot of columns. And I set up, remember, two months ago, Amazon released Alex's update. Said that there was a smart Alexa intel. I've put this up on all the columns. Well, because I'm interested, shorter, I can say that, my children and my wife have been walking me and telling Papa to turn off this stupid system for the first month. They said that. And bring back old Alex, please. Old. Why? Because it's Alexa, she's slowing down, she can't hear much. She may not perform your mission, she can perform it in 15 minutes and so on. Here's the new one. Well, I left her, of course. I didn't take it away because I'm still interested in getting some innovations and cool things. But Amazon didn't get to solve this. So voice-and-maybe logical, why they couldn't solve because OpenAI couldn't make a proper decision on the voice assistant. Well, they have no voice assistant. And by the way, who uses the world among our viewers? Please write who uses voice, voice assistants, not in particular the voice-injection. By the way, OpenAI six months ago, I think, blocked a voice-injection function in many countries. When you press the button, he'll recognize the voice. Literally, I think that two weeks ago this function has been discovered in all the countries of the world, in the whole version, I think it's all the versions now. I mean, you have the opportunity if any of you didn't see it unrealistically cool. I'll use nine out of ten of these cases. It's when you press the button, there's a voice recognition system from OpenAI, and it's still the coolest one I've ever seen. Yeah, she'll recognize, translate the text, and then you're the text. But I didn't ask. I asked about voice-to-people communication when you're on, and this system is talking to you, which is slowing down like this. They say Grok, by the way. They say Grok is more cool in this voice-over than OpenAI. I don't know, I'm Ilinar.
The experience I had with Grok shows that it's more pleasant to have a relationship with him. There's also a character being generated, and you sound like a character, and the communication itself is more natural.
At least that's what I--
That's why at least that's what I've tried. And when you're talking about technological lag, is it possible that Google is going to go a long way ahead or, uh, OpenAI will break up or Anthropic or there, uh, DeepSeek will run somewhere? Uh, I feel like I don't. Why? Because, uh, you can run, you know, basically, in three options. One: if you have a lot more computer, more computing powers than everyone else, yes, and you have learned some huge model that others can't afford. And, uh, conditionally, that's the Mistral with OpenAI, yeah. Besides, uh, Mistralki just doesn't have the ability to use a computer like OpenAI, Google and the others. So, they actually have model-- they specialize in smaller models. The other story, how to get ahead is if you have a very different data, you've collected some cool data, no one's got it, and that's why your model is some kind of, um, region very well known and very well known. I'm a little bit out. It's not likely, either, just because all these big giants collect data as they can, generate. We're always hearing and talking about the news that the OpenAI is, it's, uh, it's, uh, it's, uh, it's the money money-laundering, and it's the money-laundering, and it's the law, and it's the one that's got the dateset. The rest of the companies do the same thing, they collect, they enrich their data., the third story is the algorithms and the LLM approach. What do you want to remember? We're talking about DeepSeek today, too. But DeepSeek just over a year ago published an article about the reasoning model, about, uh, the reasoning, the training function that's used for it. We don't know if they knew it before, Google, OpenAI or anyone, but DeepSeek first published it just because they're throwing it out in the open source. After a while, there was , OpenAI of 1 GP-e-e-e-e, O.P.I., the GPT of 1 showed up in Gemini, a reasoning came from Anthropic, and the story went on. And if the Chinese did something that you can't repeat, they'd have gone far ahead. But even that seems to be an interesting technological solution that does not allow for a long run. Just because if other companies know that it's achievable, what's wrong, you can do it with a very high probability they'll do it. So there's no super rocket science that others don't repeat. And we've said a few times, yes, that if you know that the task has to be solved, it's much easier to do than just when, you know, you're looking for something to do. Google has therefore shown that this can be achieved. Uh, Sam Althman, in his official notes, has already written that, yes, they're circling us because they have better results on the pre-processing. Accordingly, we focus on this part. Uh, we've been stacking them up there, like we've already got them. I think that with most of the serious changes like this, this is what happens. Someone who's gonna show the first thing that can be done differently can be better.
It'll be there, two weeks, a month, well, a month and a half, and the rest of the companies will be renovating where the same will be done, perhaps a little with a little change.
Look, I guess there's, like, two sides I'd like to add to what you're saying. I agree with you. Again, uh, even with the computers, because Anthropic, I think Anthropic to OpenAI is very far from the point of view, of the world, right? On the other hand, let's see the evaluation. When you do that assessment when there are 300 billions of dollars, you start thinking, right? And maybe it's a story that doesn't want to have one player, OpenAI, and in America they still want to have someone else, right?
Not only when you have OpenAI, but Google is developing. Look what history is. Uh, there's still a part of the new type of equipment. And now you're gonna tell me, like Google, yeah, that Google started to build and, including to train new models and replace those NVIDIA cards on the one hand. Or, for example, their big job is-- it's de-doing, that's all about quantum computers. And the idea is, there might be a breakthrough, at least, in the next 50 years. And, accordingly, in theory, someone might have something super-universal. In theory. I'm allowed, right? It's one part. The second part is like the world as a whole, uh, we understand that we'll get a crazy change in the world if they do some unique battery, like, that's gonna make the phone work for a year or two. It's gonna change the whole world completely, really fast, right? It's gonna change the world as soon as another type of battery comes up., and technological progress will be completely different in the world, completely different. The second is history, this story is still I admit that there might be some other technology. Just. Uh, and someone's gonna figure something out that's not gonna be built on the base, LLM and a-- a-- a-- a-- a-- a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a--a---a-- language models, yeah? It is clear that they are already relatively language models, because there is still a range of serious considerations, checks, analyses, everything. I wouldn't even call it-- there are still people who say it's the generation of the next possible symbol, but it's not that, is it? It's a huge automated system. But let's assume that someone will create something self-study, some unique, unique structure. And there's a different story here. There's a story here that's gonna be controlled. So if the current problems are happening in a world with the current simple models, what happens when, uh, this is real, real artificial intelligence, right? Or, there, such a super-natural intelligence, there, AGI. What's that gonna be? And here I want to quote the report., did, uh-a, the artificial intelligence security index. It says that the largest companies do not reach the standards. And this institution that made it the Future of Life, there's an AI Safety Index. Yes, an artificial intelligence security index. So they valued Anthropic, OpenAI, xAI, Meta, and Google DeepMind on security measures and super-smart control. No leading firm has a clear plan to keep artificial intelligence in the rig if it starts-- starts operating outside the established framework. The report notes that companies invest hundreds of billions in the creation of an increasingly powerful artificial intelligence while remaining less regulated than restaurants, yes, which are. And this is a very, very interesting subject, which is basically a discussion of the regulation of simple things. But in parallel, we see a large number of problems, like those, that the base would have to be resolved and not solved.
That is a huge number of basic problems. Or that's what OpenAI did very interesting, uh, a new methodology came up about the integrity of artificial intelligence. So this also shows that the artificial intelligence is even in their own hands. And the point is? That there's a problem with modern models, that's, I think, very important information to everyone. Well, first, everyone knows they can hallucinate, they can do a concept like reward hack, it's a choice of language that the evaluator likes, not making the task honest. And they send back, it's like pretending to be worse or acting like the developers thought. Look, this is a very interesting thing, right? I mean, not to behave like the developers thought. And they've been learning the modifications in the experimental work. They took GPT 5 Thinking, that's why the model was good, and they taught two answers: basic and confession. Well, that's the basic answer she's giving, there's even a schedule, and there's a confession answer. It's where the model is frankly telling if she's been in violation of instructions or looking for tricks when she gets the answer. And the key principle is that the truth of recognition is encouraged by learning, and errors are not punished. And the result is impressive, yes. In stress tests, the probability that the model will cover the violations has been reduced to only four points of four per cent. So the model gave an undesirable or circumventive answer. And she's almost always recognized. And the methodology allows, well, this methodology to better detect mo-- fraud in artificial intelligence. But I-- and that approach is clear that it enhances trust, tra-la-la. But! When I see it and when I read it, I think, wait a second, and I understand correctly that you have algorithms that control your own artificial intelligence, and these algorithms, They don't actually work on the appearance of things like reward hack and how they send back or hallucination. And you're doing some more super-stop thing you say she'll help improve. But I understand correctly that these superstructures are going to go further again, developing, developing the model, and still making it bigger, and sending back. Because if everyone's ever talked about hallucination, there's a story in which the model, once again, does not act as it was about the development--
It's not what the developers were thinking. Yeah? And that's what I'm in the OpenAI, by the way, I don't know, that's who noticed in Gemini, too. I mean, it's because I'm not a frequent user, like I said. Elnar, have you seen it? But I'm in the OpenAI now seeing the model getting very attached to me in a psychoticipo, isn't it? So she's always trying to talk to me somehow, well, not to speak neutral. I'm just because I've been in a lot of control in my life, you know, with a lot of people, with thousands of people, I have some habit, especially, you know, with systems that make them very active. There was no emotion in the rules. And here, the model starts to add emotions. I'm looking at this somehow, and I think why is she doing this? I don't need emotion here. Well, how easy is a simple act that should make a model, huh? That's how staff are. You have a staff member who's gonna have to do a simple act.
And as a man who talks on the phone and calls on the sappport, he has a scruple, he shouldn't be off the scruple. And someone says, "Oh, that's wrong, it's bullying." It's not a bully, it's a rule. Well, it's a tough rule, as a matter of fact. A man can't add, sneeze or sneeze, or specially, I don't know, scream, or sound, quietly, and keep your voice down. He can't be as emotionally expressing some kind of phrase. Like, calling something shy or something, like, "singing" like you, "that's like your friend. Yeah, it's not my friend, it's a backup system that needs to be responsive unemotic. If I'm, like, fighting the model, she shouldn't, in fact, even take it into account. She should keep doing it peacefully, not telling me, "I understand you're angry." It reminds me of the staff member who says, "Don't yell at me like that" or "don't tell me that. You can't talk to me like that." Or, well, it was-- it's really weird, right?
We're waiting for an explosion in the commentaries.
Explosion in the comments. It's a very good, by the way, you know, it's a very good story in terms of, uh, system design. I mean, when you're talking to her together. I mean, I'm not against a broad emotional relationship and a wide-ranging engagement. Just like there's stuff in the sappport. People often ask, we have to run a very systematic business or to let something go to the company. I always said that if you have a company there aren't ten people, you, even if ten, you and someone are being very systemic, very firm, like in accounting or finance, you can't make a mistake, but... You work very creatively with someone, with super-heavy things, where this mistake might be. And now, I think this is the era of chat, one of the stories that she often doesn't know where she might be touching, and where she doesn't.
I mean, there are things where she can poke me, okay. But there are things she can't do. It's like, now, banking, the system in the world is saying that they don't trust the use of artificial intelligence, such models. That makes sense. I don't want to have a back on my account by accident, so it's either a reward hack, right? It was a play on the Internet with one girl at Chase Bank, the biggest bank in America. I, uh, at least I realized by the interface that it was Chase. She's got a sum of $90,9 billion, I think, and $99 billion, it's 9,900, but minus. And it was a gluc, a system bang, right? You wake up, open up and, well, you don't want to have a system hallucinating from the point of view, you know, some kind of decision, right?
Took you your money and decided to buy you something, didn't she?
Yeah, I don't want to. You started talking about new computers, right? I'm just saying that Google, of course, has shaken the whole market for the past few weeks, and it's all openAI and NVIDIA. OpenAI wants to add that Google has published that they have four hundred and fifty million active users. Yeah, it's almost twice as low as the OpenAI, but it's already very significant numbers. There's always a question, how do they think it is, right? It's a condition, you use a google fight or something, you have a gun or not. I mean, well, they think they're somehow, and they already have a pretty large number of users. There's already a panic mood near OpenAI that the people will move from there and so on. Well, anyway. I mean, at least what Gemini 3 presented was a very serious market effect. The next very, very important story of Gemini 3. They taught this model on their maps. So we talk about NVIDIA a lot, sometimes we talk about what AMD exists, right? You're there periodically saying that this is NVIDIA's relatives, AMD is holding almost the entire market, and the family is also there. There is, respectively, Huawei, which we also say periodically that in China, our cards are, yes, they're a little behind, but somehow they're learning something. And we're talking about TPU Google sometime. So GPU is graphic processors. TPU is its google-- its google development, it's a tensor processing unit. And here, let's just have a little bit of a context, I'll give you. When we say NVIDIA has the most cool maps, we have to understand that it's not just chips, it's in the infrastructure that's built around it. That apart from the fact that the chip itself is good, there are libraries, and CUDA is the same, yes, which allows for very rapid operation of these cards, that is, to maximize the possibilities that these maps have. In addition, we may have a large number of maps, hundreds, thousands of maps that are merged into the cluster, and information between these clusters is, therefore, somehow transferred. Yeah, there are NVCC libraries and some other NVIDIA's very well-reliant, very fast-working, unstable. Many people are therefore trying to do this through these maps. AMD or Huawei may have a relative iron not as bad equipment, but they do not have all this NVIDIA-show. They don't have CUDA, there's no library that's very well-designed for that. That's why NVIDIA has such a monopoly position, yes. The remaining cards are either uploaded or used by other maps only if they cannot obtain NVIDIA-Vuu ' s maps. Now another story. Google's been very active in the last years, if I'm not mistaken, and then, well, it's been a year since the 12th to 13th, they've been developing their iron, but they haven't offered it to anyone before. So they have their TPUs, they have their libraries that are buried under this story. Other popular fraimvors, like PyTorch Facebook, he works badly on TPU, well, practically not working. So, they have a niche of their own, respectively. What's it looking for? That Gemini 3 is fully and fully trained and is inferred with google equipment. So there's a confirmation on the market that a vast, forward model that hits all the benchmarks around the market is not taught on NVIDIA-Vu. And here, first of all, under NVIDIA, a little chair stitches, yes, that there's a competition, and that's a very cool model that's got to be taught on equipment, which is how it turns out. Those who were trained in NVIDIA-Vu. It turns out, well, as long as the professionals know, it just now has become known that Aicic part of his inference is already doing google cards. Google is already in the line of companies that want to buy this equipment. OpenAI left the application, too. And there's information that, by this, they got a big discount from NVIDIA, and that, up to 30 percent, they got a discount on a NVIDIA videotape just because they showed up. Competitor, and monopoly has been a little moving. Google's software is hers, but they're starting to port other libraries to use their own maps.
Well, I'm all about to call a video card, but actually a little bit of a video card, TPU, yeah, tensor processing unit.
Well, that's why I'm just saying "cartas," yeah, equipment to make people neutral.
Yeah, yeah, yeah. And now, respectively, they say that Meta, Facebook, they want to buy a huge amount of maps, but they're asking Google PyTorch to move the meta-Vi, to port this equipment, yes, So they can use their Framworth, which they have all written on, and the world is actually writing on PyTorch in a practical way. They could use this equipment on a naive basis. And Google is doing it. And in this sense, Google has been in the last few weeks right from two feet to the market, showing that they have a super-crunch model, showed that they have equipment, showed that they have a big deal. Number of active users. Anyway, it's really interesting that this has happened in the last few weeks. Clearly, NVIDIA will not lose its monopoly. They've got huge work, they're out there, they're about to be new, uh, you know, getting a new chip, and they're still developing their story. There's a huge legacy that's built on these libraries that NVIDIA has. But there's a competition. Uh, obviously, we're having a monopoly on this. What's the value of NVIDIA and Google? I wonder if I can watch.
Yes
I wonder if I can watch.
Yeah, they're growing up anyway. And Google has grown up in recent months. Here. The interesting case was about the GPT tip. In the GPT chat, a number of users, during-- just below their dialogue, displayed an advertisement that was not related to their request. And one of the bloggers wrote, "If you're advertising something, at least advertise my request. If you're testing the system, yes?" And he was offered at the time of the request, which was not even relevant to, uh, the topic of the phytness, and offered him an ad. There's a company like this in America Peloton, right? And, uh, there was a Piloton Pythness commercial, and, uh, it kind of got a little fast. Someone wrote, "You must be in a free version." He wrote, "I'm in Pro for $200." Right away. That's very funny, yeah, you got a $200 commercial in Pro. And, you know, I have a feeling that I've seen such messages, too. I just didn't pay attention.
I mean, because I'm a lot of wolves and plus I've been testing, so here's the backbone module. I mean, I was gonna say last time I was unsatisfactory about the spine module, but then again in a week, some things were tested with a little bit of a little bit of a little bit of a little bit of a time, and he started answering, professionally. I mean, it's a big, professional, cool answer. Who else used this module all of a sudden? I just don't know if he's available. I think he's only in the States, yes, the backbone module is available? Or do you-- or did he see who you saw? You don't know or?
I think, uh, it depends on the account version of the account. They said they were all rolled. I think he's not only available in the States. Look, he's not just coming in the States. Another thing is, is if he's useful outside somewhere else is another question. I mean, if he's looking for Amazon and American websites, I just didn't use it myself, I'm always having a-- a plate like that. I didn't use it, but if he's looking for only American Marquetles, he's got a lot of money out there.
If anyone used it, you'll write it in the comments.
It wasn't even a label. The point was, he's doing an analysis for you and making comparisons, and he's helping to pick something. And in fact, it automatically starts deep algorithms and gives you a choice. Plus, I had him acting differently. He used to offer me the goods and asked me, "Do you like it, no?" Then he started something, and I think, a week ago, even the scrimples we showed a few. I had a chance to show you when we made the video. The point is, he started asking me in a series of questions, "What's your budget?" And suggesting answers, and so on. That was the first version of what I actually had ChatGPT asking, but the interface was very nice. Anyway, I'd like to see ChatGPT in my place not as a veiz, but to keep him in my line with me. Well, as long as I'm on the command line, then keep me in touch with the command line. But sometimes, he's got these questions he's asking. There's Deep Research. It's a very big tool that everyone uses. I was just laughing yesterday. I'm telling him in Pro, "Get me news in a week." And then I know he always asks, always asks me, "What do you think you've been doing since yesterday or today?" Always. Actually, there's always this question. I'm telling him, so, uh, I think-- and I'm gonna keep saying, "I mean next week when you got a request exactly seven days at 24:00," he's gonna ask me the first question, "Are you going to be there?" What seven days do you want to get?" And you're like, "Hey. Oh, my God, well, you couldn't fix that, could you? But in the shopping, they started testing the difference-- that's what the client's interview was. And I'm gonna assume they're here for you, uh, this client-conscious system, why not all of them have to unload it? Because she's expensive. So when you need to ask your client, you obviously spend your big-- imagine how much on your request, you need to spend your resources on one request. You need to process the request, ask the question, work--- well, come on, finish it all completely. I mean, you're spending more resources, by the idea. Although I think you're getting more, uh, well, more refined prompt, and you're working on a prompt like a low-cost, low-cost model, because you're cutting some pieces, just asking. Some questions. And then when you got a cool prompot, you're giving a better answer, and the user doesn't have to say, "You're not the wrong, not the right thing," because people don't know how to do it. Looking, right? Well, it's important to know that people can't look. Most of them, most of their people don't know how to look. If, by the way, we're gonna see all the speeches, all the scholars, bloggers, people who are connected, engineers, everybody says, "You learn to write pros, then you learn to write," right? And proms aren't just one line, but really sit down, describe the details of all writing, and then you get the quality, the real answer.
Oh, uh, I'm gonna tell you the last thing I want to say, again, this is the subject we're talking about, but she still wants to be raised. I don't trust the UN much. I'm in person, yes. But the topic itself is very important. If anyone saw any other research, throw them away too. The United Nations has reported that artificial intelligence can widen the gap between rich and poor, yes. So they called this report the next great disintegration, which states that developed States are taking a lead in the race of artificial intelligence. Well, obviously. I mean, they're putting emphasis on the state, and I want to emphasize how to use our audience, that's an indicator that if you have artificial intelligence and if you can use it. artificial intelligence, you know how to use it, because not all countries can get, but people can get it, you're very much distracted from a large number of other people who can get it. artificial intelligence is not used in its work, in its life and every day. And that's a qualification that should be in any case, you won't avoid it. It's like the robot I ordered. I think I've got a lot of things to do with this robot. And my son, I think, is small, you know, relatively small, medium. Igor, he's nine years old, he told me at school that I must have ordered a robot, and someone must have said some kids at home a price. He came to me yesterday, said, "Dad, and the robot is there, I don't know how much it is." He says $25,000 yesterday. I said, "I signed up for $500 a month." Well, you probably have a price anyway, because we don't know it. It's a signature that you can give up or it's just a scramble, right?
Or maybe it'll cost $100,000 or $150,000. Although Elon Musk is kind of promising there $20,000, $30,000, that Optimus will be worth it. I'm curious about Elon Musk, by the way. I don't know, Leonard, have you seen the reports? There's a strong drop of Tesla in Europe now, China understands, well, China is logical and in America. And they say that this is because there is a lot of attention to artificial intelligence and robots, in particular, not to cars. What's important for people, including cars being updated, cars updated, their visas updated, new cars appeared, not just some stuff.
Although, of course, the latest Tesla in terms of self-driving, but it's a flight. They're all in there, I think they're all totally rolled up and done everywhere. But let's see if they can optimus, of course, release or not release. With the investment, that's a quick comment. The Bank of England has published a report on the next boume. What are they talking about? That they've smashed investment estimates for many companies. These investments were made, made by borrowing. So, there was a bubble again. And, well, as we can see, the Antropic billion of proceeds and three hundred and three hundred and three hundred and fifty billion of the scores. And once 300 billion, the score, that means that people want to make a living. And if IPO people still want to make more money, of course, that money is not even real, but at least it looks absolutely inappropriate. And I mean, it's not real in terms of the money itself. And that's why the Bank of England is very right about something. But at the same time, if you look at the infrastructure market, electricity, support, supply of various, equipment, vaults, different cloud services, all of these companies, of course, quietly, - I'm not going. - I'm not going. And there's even a position where they've been there for the last six months, 100, two, three hundred percent. And they didn't make it, they got it. Although most of the infrastructure is ticet, the overall infrastructure ticket, which relates to AI and tech, it is growing. So whoever invests, I, I look personally in this area, and this is not an investment proposal, as my friend says, if we buy, you sell, if we sell, you buy. But...
You buy? Uh, but, uh, uh, I think this is a very interesting subject. And even, uh, I'm here, uh, in parallel, opening the Meta Stock... and I'm talking about the Alpha Beta chip when Elinar was talking about, and I'm there... what information I was looking at about their chips, and... It was a little bit like Meta in the morning, uh, a little, uh, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, uh, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, uh, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, uh, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, a little, Here. So, um, even though it was like that, it was written that she should go down because they were kind of slitting Me-Metaverse. Here. Noo has Meta-- Meta's mouth shut, and she's nowhere to be seen and nowhere to be known. Well, I'll see you in a week., ToTheMoon: technological news, Silicon Valley sites around the world.