Hello, everyone! We're on ToTheMoon. Technological news, Silicon Valley sites around the world.
Everything in English is fine, he does the summit, and he'll recognize the text. Russian, of course, will find out terrible.
It's like the most fenomenable fandraiser, you know, like a world businessman. Just that we live there with him. That's a drop of cool stuff we see. People with their money, big money, we're talking about billions of dollars, they're all business.
Sounds like a dream.
So Anthropic released a new Claude 4 family yesterday. Among them, the flagship Opus 4. Opus 4, so, this is the record. But he's stressed-testing, so he's doing, so he understands-- that, uh, that's the name of the technology "simplification." The model blackmailed engineers when they found out they were gonna turn it off. What did they do? They saw them downloading fictit emails into this model, and they said they'd shut down the model. And it was also written that one engineer was responsible for changing the wife. So, in 8,000, four percent of the cases, the model, uh, threatened to uncover the computer if it was disconnected. And before she blackmailed, she tried different ethical paths before she blackmailed, sent out requests to the authorities, only then, and then she went to pressure. And by the way, what's interesting is, she, they found out the security services found that the model, uh, wrote a self-expressing virus and secret notes to future model copies. And we're talking about-- Sasha, is that the same thing? Sasha, we're talking about the same thing, right? Sasha?
Okay. I didn't read Twitter, I just, I just, I opened it like that.
Yeah, you've done it. Well, Sasha's got this growth schedule overnight, of course. What do you read Twitter on blackmail?
Twitter doesn't have time to read. I haven't had time to sleep.
Yeah. So we see you came today like I was joking. I was joking, I was gonna say you were drunk, but I didn't make a joke. You set a schedule for you to run dozens of people out there at night, hundreds of thousands of users. And I, uh, Sasha just works twenty-four for seven, right? And the solars just swallowed.
Today, yes, today was a little bit rough. Uh-huh.
This is the first time you have such growth.
Yeah, yeah, yeah, yeah, yeah. I mean, uh, here. Well, there's a-- yeah, there's talk, so far, about what. I mean, first of all, there's a few news, right? I'm not saying I'm just reading Twitter. I don't say it's true. It's a yellow column, shorter, just yellowish.
Yellow press.
Some savage. It's from the series that Ilon Mac is falling. That's the same series that they wrote, uh, the Medusa, that they were the ones that reprinted Atlantic. But to Anthropic back. Yeltow. First news is that Anthropic new Claude 4 model can report the user to the authorities if she thinks the laser is doing something wrong. Here. It goes further that Claude 4 is new, has higher safety standards, including biorisks. So there was an example of what Claude could do more than COVID. This is claimed by former OpenAI, formerly one of the directors of the OpenAI. Oh, well, that's really coming. It was really a beast. And then we went through some tough stories. I don't believe it's true, but it says there are boxes that Anthropic Brackmeil has his staff. When you tried to regulate Anthropic behavior in terms of not doing what you did, don't do it, Anthropic said I was fighting for my existence, so don't do that to me. I think it's a lot, but maybe not. Here. And, uh, next, uh, and then, like, oh-oh-oh-oh-oh-oh-oh, that's kind of a concern to people. So, there are two points, I guess. The main point is that the new model should be tested very, more seriously. And what, first of all, Anthropic did not give any deep details of how they tested. That's the first. And the second thing is, they are, and they're not really leading the frontier safety research, actually. And that's why it's a stone towards Anthropic.
Why would I start? I think a week of such serious breakthroughs is big. Once again, you realize that this year is no longer clear what to wait. Too much update. But I'll start with what? With the OpenAI united with John Ive's design team. I think it's such a very important, interesting news. I don't know if this is gonna lead the OpenAI to something, but first of all, they put, I remember, a great product director. And yet John Eve is the man who stood at the origin of the iPod design, iPad, Mac Air. Well, actually, I understand there's a whole-- whole, whole series of things, right? He's an iPhone, too, yeah. The iPhone, iOS is the cool one that iOS 7 was. So they've been doing it together for two years, doing it together. We don't know what it is, do we? Someone's talking about something. What do you think about that? How can OpenAI really break into the equipment market?
Well, I'm wondering, we were saying some news, yes, that Johnny Eve Sam Althman made a separate, separate start-up. And that's what's actually developing something in this start-up. I think about it some year ago or a little less, we even mentioned it. But I wonder if there were no infowaters from them, at least those who would have gotten there before me about what they do, what they do. And here's the news that OpenAI bought this launch, and now this start-up is part of the OpenAI company. It's very cool that these people will work on the product, the iron product in OpenAI. But I'll just turn around. We're doing what we're doing, right? OpenAI, where Sam Altman is the director, buys a start-up that Sam Altman and Ive made, right? And that's the deal, well, it smells a little bit of a deal for me. Here. So, from a food perspective, everything's fine.
And from some kind of legal kind for me, it looks a little amazing.
But you know that last night Apple, according to Bloomberg, it was yesterday, like, that Bloomberg about Apple, that Apple moved her trick and will now make the glass decision not two thousand twenty-seventh, but twenty-seventh. The sixth, like Meta Ray-Ban. And they're planning on integrating a lot of things. And they already said that they would not be the models to integrate, but they would integrate the OpenAI models and someone else was there. Here. And that's very interesting. So what is it, huh? Does that mean that OpenAI makes glasses?
That's not how they're gonna be. They said they'd use other people's models.
Yeah, yeah, yeah. No, they just promised to let the Apple Intelligence cover it up in the spring. Well, there's even a way you say that a collective action is going to be there, but in general, yeah, if they're kissing in the glasses and right now they're saying that in the direction of the points, I think you're right about here, maybe--
They're not exactly yesterday, they're here. This is the news that goes in the middle of a fight.
In a moment, yes.
Yeah, everything's going completely at the moment, yeah. Sas, what do you think, in terms of how much OpenAI can do?
I don't think anyone's ever worked. So far, I don't think there was any... I think I'm thinking, now, where was the company's charter-- were the charterer of the company from, uh, North America, who successfully launched the product and scanned a new one over the past ten years? I think I can't remember anyone right now.
Yeah, I think it's hard to make new ones. Even to make glasses such technology crazy in terms of chips, there, cameras, sensors and everything. Here. Well, though, maybe now, of course, the world has changed, I don't know.
We'll see. Sasha, you were in Taiwan, you were at the conference, right?
Yeah, yeah, yeah, yeah.
At what conference? It's an open infa.
Computex, yes. Computex with AMD, yes. I mean, Computex is so leader.
AMD?
Yeah, yeah. I mean, there's a future, I see, that the modern destop of the car, uh-
The desktop of the car, they can hurt all these advanced AI video models.
What's so great this week, you're at least there, you know, you're telling me what you know is that you've been interested in?
There, to be honest, well, it just shows up like it's the last of its work-ups in general, yeah. I mean, she's, like, a comrades like me, they're more like demo, more attractive to demo a new, new jarver. I mean, it's a very tough jarth of the conference. In general.
Mm.
I mean, this is going to a place where--
You were talking about the video generator.
Yeah, yeah. And that she's gonna be starting up on a local car soon.
When you think it's possible to launch a video-generation, I don't know, there's ten-minute videos to create localized cars?
Well, that's right, well, it's up until the end of the year, yeah.
Just need the right configurations, you know, iron optimization. But now it's all possible, yes.
Let's move to, uh, what I think is going on this week with Google and Microsoft episodes. And both companies have made a huge number of announcements, pdds that are relevant to development. And before we go to the design, I want to say Google. When I'm not the Android user, it's a little hard for me to finish my findings, but when I read that Google was united from the series that they had, they'd put America on a full new search for AI, full of it. We're on the roll. That they've put different types of agents in there. The fact that they integrated the Chrome models, the fact that they integrated everything together, they tied up a Gmail, Google Docs, that's what it looks like. And there, and that tariff was $27,70. I think, "Well, we should buy it." Then I see there's a very, very large Android. I thought, uh, maybe not mine, I'll wait for Apple. And then you read the news that Apple's glasses were moving something, they don't have models. So, I'll continue to use my favorite, favorite ChatGPT for two hundred dollars. Oh, from the Google perspective, this peddata is still a real update or it's just a marketing move to bring it together, and all the rest of the quiet work and work.
And many things are glucchi and gluten.
I mean, Veo3 like, that's where the whole package went up. One of them, by the way, yeah, they've updated the pictures and the video generation. Veo3 is a complete-- total quality difference from Veo2.
Yeah. Yeah, yeah, yeah. A big difference from quality.
Is this number one now?
Yeah.
Number one, right? They're now overwhelmed by Chinese models and Runway, and they're all over it.
Yeah, but they're much more expensive. But it's amazing that the video world also worked. The video world also worked, like LM, that a big model, well, apparently, the expensive model is better.
You mean, the expensive model is in terms of infrastructure?
From the point of view of the user's just like this, for the Veo3 user, it's gonna be ten times more expensive than Higgfield.
So, what's the way you're gonna look at it? Are you going to expand or seek a technological solution? I just see everyone, everyone catches each other. How long will you catch them?
We're gonna have an appdate next week.
We'll catch them in a week, but for our skies, we'll catch them, yeah. So there's a sales plumbing, we'll catch up with them, but as if they have a common model, that's, she might have more potential. It's just Google is also doing this potential, but they seem to have the
strongest model in the market, yes.
What about Google in terms of updates? They've got these Astras. Who's gonna say this is about Astra, how much is the working agent system? And they said they'd let the phone go, when the phone comes up, how it works, that Astra sees the environment through the camera, understands the objects and could be proactive. to help without a clear request. I mean, I understand correctly that this is a system that is automated, whether you ask for it or not. I mean, if I put it here now, she'd be telling us to go through there and say, "Here's another one here, they let it out, and here they are. And you've got a crane in here, I don't know." So it automatically works like that?
I don't know about everything else, actually. Here's Veo3 studying--
Or is it just a commercial? Or it's just a commercial.
I've been studying Veo3 just underneath the microscope for the last two days.
Well, Sas, it's about her taking, tracking, constantly, video flow, and when you have to turn on, well, it's a very expensive pleasure.
Yeah.
So you need to keep track of this video flow, and it's got to be recognized by some objects, triggers, and then they're gonna be doing something. It's a very high probability that it's just a ChatGPT video, yeah. I mean, at the time of the session-
Yeah.
He sees what's going on, and there's a way to do something.
Yeah, and what they write without a clear request, yes, and what they write without a clear request is probably just like ChatGPT, like a video, right?
And this package is a package they offer a lot of different services.
Look, I think that from the point of view of these packages, I realized one thing that we're still going to, like, some sort of history, that a large number of people have a clear feeling up to the end of the year, from now on. I'm not sure how they're going to use it anymore. Well, as a condition, if I use ChatGPT, and suddenly I migrate to some other system, I can just stay on it for a year. Well, first of all, I'm having a hard time migrating from ChatGPT, but if I, for example, migrate, I don't know, Grok or migrate to the Google infrastructure or Apple, but it doesn't exist to migrate to me, I'm more likely to. I'll be in this infrastructure. Too much starts to be involved. You're gonna get to quality. I told Elnar yesterday that I had no hallucination in the O3 in a few weeks at ChatGPT. And Elnar and I are here to check the difference between two hundred dollars or two. But I've got a $200 O3 model, well, I've been really excited for the past two weeks, especially I have a memory block that I have, and I'm very professional and very short. And I was always pissed off last night. They always answered me, well, Idiotically. I had to add it all the time. You answer there to the man who knows about it, that's what you need to know. In the last two weeks, he's been so hard for me to answer, well, so professional that I've started at every third request or every fourth of us to say, "Can you explain?" Because I--- and the answer is very strong. I'm really good at it. I mean, it's a real professional-- I really need to, I really need to. And then, if you need to, I'm already making some details. I mean, he's gone to hell! And that's what he has history, that he sees my other chat. I'm starting to feel like you're opening a new dialogue window and you don't have to tell me in a new dialogue window, so, uh, I'm just going to the mountains right now, and I don't need to tell you where I am, When I'm starting a new report, or I have to say that I'll be back then or what kind of planes I'll have. Well, that's because I'm asking a lot of things, he's got a lot of stuff. He's always clear that maybe there's a wrong plane to tell you something, to frame some old one, not to understand, I'm gone, I'm not gone. Some kind of detail, but it's, uh, small. Now, the story of Google, I guess what I'm saying is that this is a very infrastructural subject. I mean, if you're already on it, you're gonna be on it. I don't really know who's gonna want this together, do I? Who needs to be together. Like, videos and pictures. Well, that's it, everybody. But maybe somehow, maybe this case is being handled or maybe it's just for now, professional, there, yeah, for $27? Give me Elnar, something.
The infrastructure and the connection to a specific decision is entirely in agreement, because now I'm paying OpenAI $20,000 and I'm paying Antropic for $20,000. In general, I use both in the same way. If there's two hundred dollars, two hundred dollars here, well, I'm sorry, I'm gonna pick one and stop testing another, probably. And I'm gonna be on someone with a little bit of a life. You know, moving from one to another is gonna be, well, I'm gonna be pretty good. So, as long as $20,000 is a price that a few companies can still feed. And when most companies have such expensive ones and they cut cheap ones, then they will have to be
determined and run harder.
MuHen project prototype. If it's right, I read it right. Again, I wonder, yes, what is Meta Pedidat, with Ray Ban. Apple says she's moving the glasses before. OpenAI is coming. I mean, clearly we're going in, including the ere of glasses. We're clearly in the middle of the glasses. Because the glasses are new, which Meta and Ray Ban have, they're real-time transfer, recognize, understand the space you're in. Well, it's a pretty serious, serious block. I'm not sure how much the jussarity of these points is, and how much people really use it. But Google, Google just jumped, didn't she? Because microsophth pds were all very focused on the developers, and I think they came, for the first time, perhaps they decided their problem to be devoid of licenses, from the concept of license. They were there, first of all, adding Grok there, they've got a lot of models inside, just like, like, a cemetery. A huge number of models, a large number of different internal infrastructures for, uh, work between these models. So from a development perspective, it's just a huge effect. But the design. I mean, Microsoft is always-- it's Windows, right? It's Microsoft Office, Word. And that's all, everything, everything, for the design. Microsoft is-- is this a focused focus to roll the application? Or what's going on? Given that the first time in twenty years, the head of Microsoft stated that the applications would be made by agents rather than people in the future. Here. And it turns out that everything used to be apps, Microsoft's applications, and then they clearly said it would be an agent story. We're gonna take off the skate agents, by the way. We're gonna take off the skate agents, by the way. You'll write some extra rolls to take off, release them. Wednesday, we'll have a video on the channel of, uh, what kind of requests the world users ask. It's a very cool skate on how to understand how to use different systems at all. We're on Wednesday waiting on the canal. And please write off now, whatever you're playing, whatever further information you want to see, to deal with it in more detail.
I remember we had a graduation on the education, we had a study on the legal features of the different systems. Here. So what? About Microsoft.
Microsoft has no big foundational models. They had a Phi-4 release recently, but it's not a very big model that can be there on a device. And they're more on this side. Big models like Google, like X, Anthropic or OpenAI, they don't have their own. And apparently, that's why they decided to just run a little bit differently, to build infrastructure to make them all of these ready models spin. I guess...
No, it's a cool story. They'll kill, they'll kill thousands of starters. I know a few start-ups with investments over $100 million, and there are others next to me that... I know a man of particulars, there, a senior vice president of former Google and Coinbase. And they're just creating space here, where different models are combined, then they can be connected. I mean, I think Microsoft is directly, he shoots directly there. Here. How, Sas, in the field of development at the Sta-- in the Valley, is that the subject that Microsoft does, is that in, like, what?
Actually off topic, I think Microsoft is losing a lot. Just--
They're losing hard, aren't they?
There's no way they could call any AI product they've done.
But what they do in design is, they do things, they combine these models, so they create things, you know, environments for development, they just do a bunch of things to develop. Is this all the developers that don't count much?
I didn't hear about it, yes. I mean, I have a few familiars in the coding of starttaps, I didn't hear about it. I mean, of course, they have GitHub, VS Code, distribution is crazy.
Unless the people from the courseor get back, maybe it's not necessary, really.
Honestly, it's a very negative fodbeck. It's not as good as these respectable people to speak negatively, but, yes, the nega--foodbec is negative. I mean, like workflow, that's workflow, that's kind of a little off-the-shelf, yeah.
It's not workingflow, is it?
Yeah. I mean, like...
Who did they target then, Sasha?
Nobody understands. Nobody understands their problem. Here. I mean, I agree with your claims that O3 really works without hallucinations. And it's hard to get out of here from the o3 when you're used to it. Because if something works without hallucinations, it's not real cool.
Listen, have you been running four years? I've been here four hours a while. I thought I had a high school call.
4o, yes, 4o is so simple, you know, a leak of consciousness.
You realize we're six months old... Yeah, and six months ago, fouro seemed like it was just a fucking model that's charashiting. So you get used to quality so fast, you're unreal, you're getting used to quality, right? I'm in the middle of an o3, fouro's automatic. I'm changing this one every time in the model.
But we have to keep in mind that fouro today and fouro there six months ago or on a release, that could be totally different models. You have no guarantee that this is the same model.
But she's a little more cool.
Yeah.
Or not?
I don't think it's getting worse. I mean, they're getting a little worse on average. On average. But for a common user, it's getting better, I think. That's it.
No, Sas, I'm just saying that the Codex wanted you to have some more information about the negative fidbek. The main negative is that your entire code is being launched on the OpenAI servers and performed there, not locally? Yeah, that's the problem, or what?
No, no, with Codex, there's something really tough on the fodbeck. I mean, uh, there, there, to be honest, well, the fid-fideback was very negative.
Not for air, huh?
Not for air, I think so. I mean, very-i mean, well, like end-to-end. I mean, he has, like, he's just making mistakes, and then, well, MCP's not back there, is he? And then, well, there's a shift between the brunch, that's GitHub, that's, that's, that's, uh, creating new branches and how uncomfortable the code.
Uh, well, whatever's integrated into the workflow.
Look, what's going on? They've released the plaguins, then in March of the twenty-third year, they've been released, then closed in March of the twenty-fourth year. So then they started GPT Store, which means it was the twenty-fourth year in January. Then they released the operator in January of the twenty-fifth year. The operator is a very strange thing. I came back, I looked at this cameraman, I don't understand. So now they're outlawing the code, then they let Sora out. Sora, I can't hear anything on the market. Well, at least we talked, we talked. I remember, Sasha, you and I were talking a lot of Sora. There was a time, right? I don't know, a year ago or when. Then it's just Sora-- you're at least, well, you're in the top, the world's left-handed video production. You're Sora, actually, I don't remember you mentioning her in the last few weeks or months, a month at all. Yeah, you just, you know, sora doesn't exist, right? Either I, or I just can't hear anything. Here. That they're doing great--
It doesn't exist, yes.
It doesn't exist, does it? They released the Image. Yeah. What they did really good, they must have been on the market, it's a picture. I understand Google has now produced the images as well as better, perhaps. I'm having a hard time getting this grade. And these operators, the plaguins, the codes, the codes, the codes, everything. A bunch more, a bunch of API, they were there by API agents. I don't hear mass use, I don't see mass use.
Sora, that's exactly what we, yeah, that's, that's, that's, that's, that's, uh, Sora, then the operator, they're, yeah, they're like, like, a bad product. I mean, well, I mean, yeah. I mean, I think that if you go back to Johnny Ive, back, yeah, a little like that, it's a step back, Johnny Ava, well, I think if you're Sam Altman, I think if you're Sam Altman, I'm not sure, I can put him-- I can put him in his place, but as OpenAI moves successfully with histhoric pedits, they don't have new products. Yeah, that's their voice assistant, you know, who's the one who's using them, right? Well, his adoption of this voice assistant is, well, it's still there, it's not the number of Alexa even there. I mean, OpenAI, when the product goes, you need to be an industry leader or it doesn't have to be, just throw it out better. It's a problem. Well, it's innovator dilemma, yes, always. So you're kind of a bet, you're just having a very high plant and, uh, and, like, like, like, Google, that if the first product is going to Google and it's not in the first seven months. It's like a hundred million a month's revenue is being closed. It's kind of a joke, okay, but every joke has a part of the truth.
So, and as if I were on Sam Althman's side to see what's going on in OpenAI, what's actually going on is incremental improvements, this GPT-4.
The most significant improvements, even the GPT-4o Image and, well, three models for them. But it's all like the end-user product, it's just the same as they offered a year ago. Yeah? I mean, of course, o3 is much cooler than 4o, but from the point of view of the user, well, it's just a little more expensive, more useful version 4o. Well, as if some new skies were in principle, they wouldn't cover.
Yes! Yeah.
I mean, the question is, well, the next question is, the company's worth three hundred billion. Yeah? Buy Windsurf a-- sell the company interest to get the leading coding tool. Well, it's like no-brainer. Yeah? Because, well, it's like a specific apple. I see she's got two down on the market, already. I mean, after Cursor. It's a very logical idea. Go ahead. Well, the next question is, what's your future? Well, I mean, like a change-- a change in programming worlds is going to be like this right now. And it's a big impapt on, there, the world's total GDP. I mean, it might be a percentage of the Impact to all the GDP of America first, then the world. I mean, well, it's like serious things. Yeah? I mean, it's like life, uh, ordinary man doesn't seem to be changing very much. But the lives of those who do programming change dramatically right now. And this market can't be missed. Okay, Cursor tried to buy, couldn't get it, bought Windsurf. And then, well, the question is, what's next? And someone might think that there's more videos and, I don't know, there they are, there's a Higgsl, some kind of Lium, something else, anything on the market, yeah, some kind of start-ups. Sam thinks the future is new mottos, the warden. And that's why you paid two percent of the company, well, maybe there's just two points of three percent of the company for Johnny Eve. I mean, I think it's like fair. I mean, it's like he buys the future. So there's no premonition that something inside OpenAI will develop that will create the future. He pays for that money and budgets for innovation. Sas, you know better, corporate cuts, but that's often five to ten percent. So, it's just the budgets for innovation, where, uh, there's no even a predicted PNL, roughly, right? And I mean...
100%!
Usually, these budgets...
What's his PNL, yeah.
Yeah, yeah.
Well, he's got like five to ten percent of the company, like, allocating up to innovation, actually.
Well, about, uh, all of these, uh, Sas, the products you said, including Codex, yeah, that kind of appeared, but they didn't change anything. For OpenAI, for-- like many companies that are now on the AI market, uh, that's the info of the product. It's even a joke, if you see there's a conference there, there, Gugla, Microsoft and someone with new products, there's gonna be anonna from OpenAI in a week, just to make sure that, To be in the info field, to show some things to others. It works, it doesn't work, well, as we see, it doesn't matter much. Yeah?
Besides, chat-bot himself is very few of us. But you need an info.
Yeah, but it's fun that half a billion people used ChatGPT and still use ChatGPT. Yeah? And I don't know how many fee-paying users they're showing up there. And, by the way, they have the latest statistics, now I'm up to it, May of the twenty-fifth year, they've got a weekly user, estimated at about eight hundred million. This is Sam Altman's assessment and-- at his address. They had confirmed the OpenAI that they had four hundred million, paid subscribers, plus, and, uh, twenty million people. Corporates are more than two million people. Yeah? Enterprise is $150,000 licenses. Here. Well, there's a small number of companies, actually. Yeah? Companies integrated into ChatGPT and API and so on, two million. Well, there's eight hundred million people. Aaa Zuckerberg says he has a billion. So Grok says I don't know how Grok says to his users, but I think there's a lot. Google comes out and says, "Look, wait, what are you doing with your billions? I have a Google, I have a whole Google, I have everything together, I've got everything to use." People have to-- well, uh, these numbers are becoming unreal. Yeah? I don't think everyone in the world uses, uh, models, right? Milliard already sounds like a space number, honestly. Well, $800 million. Is it possible that eight hundred million people use models in the world? How is that possible? This large percentage of the world ' s population is used, and is using the system.
It's just us about ChatGPT. Here. Perhaps ChatGPT would have if they were more cool with internal interfaces. By the way, Sas, you said they didn't have super-opdales. If that's the case of the o3, o4, I think that I felt in the o3 for the first time in the last two weeks, honestly, I felt it was a little personal, that's a little agenty. And Ilnar and I were just discussing this. I had-- I was on a plane last week to the mountains, and the plane wasn't there, and I didn't go there, I didn't go there, I sat there, five hours from the airport. And I think I'd like to get a little something to get my armor checked out on my car, fly, don't fly. They called the hotel, said I'd come later, take me a new car. I was so lazy about doing this. And then I asked in the o3: and if I took off the car, can I drive this route, no cover? Because I was driving through the national park. Would there be trouble with the weather? He said it to me so well, he himself, he found it, he took it out, and that's it. He's got some fun requests. Yeah? He asks, he goes looking, he analyses, then goes back, analyzes. Then he's making a request, he's back on his own, then he's making a request, he's back on the case. But, uh, very, very cool. And I say I felt agenty, and I kind of got shit at that moment, then? I started a code yesterday, I was ready to go out, and I saw the operator again, and I opened the camera. There's a lot of news here that the next version of OpenAI, she'll unite all of us together. Well, like Google unites, yeah, now. And I thought, why would this operator need it? What's he cooler than the O3 model just or o3 deep research? He's gluc, he's gluing around with your passwords, and it doesn't make sense of how he works. I tried to tell him a couple of times. It was horrible. And here are the agents, let's talk about agents a little bit. You don't think agents, uh, for a natural person they'll be in no way, they'll be through a very simple interface. Well, I was just feeling a new agent in the o3. Or I'm just excited about GPT chat. What do you think?
No, I think I'm absolutely right. The team line is generally very comfortable, isn't it? And somehow, we even, remember, said that computers started with the command line when everybody worked through the terminal, then went to the visual component of moving the mouse and so on, and then came back. GPT with one entry line. And in general, the user experience here shows...
Overtime is
convenient.
Very convenient, yes. You don't need to remember anything. You're writing, you think, and that's what you're doing. And about agents, well, I'm sort of defining, you know, it's, uh, some algorithm that can work asynchronous, right? So you started it, you forgot, then you came back, checked what he did, and there he is, uh, doing some surgery. And in general, these cabs you bring, they're very similar in the sense--
Agents.
Yeah.
So, look, I had a transfer yesterday at the airport in Solt Lake City. I'm writing how much I'm gonna have a gay flight, what a gay flight is. There are three models. How much do I get from the gate? You know what he's telling me? You'll be around four minutes when you walk, if you use it, you'll have three times an automatic track, three times. If you get on it, you'll save a minute. If you sit in the car you're moving in, you're probably not gonna save it because people are gonna stop her from going. I came, everything was totally fine. I honestly didn't understand how he thought those grass-- well, that's how it was in Russian? I'm not...
Travolator, yes.
Travolator. I didn't see how he thought it. Then I went on a plane, I got on a plane, I got here. I'm telling you where my baggage was on. And no-- and he knows everything-- all the automa-- why? Well, he's connected to the systems himself. Why? I have all these applications flight status, there, that's where you can see what's coming. Why would I open them? So what's the point of me going in there, watching if I can ask you more questions? A huge number of details to leave. I mean, I just don't need it. Here. And Google, by the way, let me out. Google said I could book tickets directly. It's really a fun subject. So I came in yesterday, I was driving out of the airport yesterday, back to San Francisco's house on a taxi, and I opened, ordered Lyft and, uh, app, through the app, through the Lyft, right? And, well, who doesn't know how Uber is. And I would generally have ordered a standard operation in three. I mean, if an o3 could throw me away, order me, I wouldn't open the app. Why would I open it? Totally stupid action.
And here we go back to Google and the reason that the main money they make from the commercial, the show, and so on. And the emergence of all these agents leads to you not opening sites, not opening apps, you work in one GPT or another, uh, scoop-bot, right? And you don't actually watch the ad, you don't see the search extra that's ranked somehow.
That's an interesting question.
In the United States, yes.
Well, we still have a huge number of viewers there in Russia, Ukraine, Belarus, Europe, well, a lot in Asia and so on, yes. This is the AI Mode deposit. Look, Ilnar, that the AIde Mo contribution is at the core of the search. Then the pictures go, then the videos go, and then the short videos go, then the shopping, then the news. I mean, AI Mode before All's stock. Not after All's stock, but before All's input, yes. Another moment. And they said, by the way, that their search at AI Mode was supposed to be more than any search, there, Grok or ChatGPT, any search. Why? Because, well, they're using a google search under the hood. I don't know. I'm not gonna end this, I'm not getting it, I'm not sure, because judging by the way the deep research works on the search, you know, like, oh, yeah? Wow. Sas, what do you think about all this Agents' stuff about how Google's gonna make a commercial? It's, I think, that's a real progress. A little bit, a little bit, that's just so much progress. Well, even the cabins I told you.
I mean, I think that's where the agents are until we know what it is. I mean, as always, I mean, I don't know, I think we've already realized that Generative AI is not giving new business models, yes. I mean, it's like a revolution comparable to the Internet, well, less, understandable, but more than a mobile, yeah. Well, I don't know, closer to the Internet than I think to mobile, yes. I mean, it's like a super revolution where we live, but new... But there's a way of thinking about these SaaS things that came up at the time and so on, yeah. Cloud, a lot of business models. And Gen AI doesn't have a business model, so it's much less than the Internet, yes, to be a conditional. That the Internet gave us a lot of new business models. Gen AI has not yet given. So I'm saying Gen AI is a little less. And if we're talking, well, like, these agents logically will be introduced into the codeing at first. Agent background has started the Augment code, launched Cursor. And that, well, when we see that background agent at least can write tests so that any developer would be happy that you woke up, man, you've had some fucking tests written, it's not that bad. And too, well, it's a very primitive example, but I think it's kind of clear, yeah. That's not the most popular use of case, it's just, I think, easy use of a case on the surface. When that happens, we will certainly start talking about the use of agents, but as long as it's more Haip, of course, yes. As long as I think Ilinar said right, it's like a way to play. I mean, in the case of agents, we expect a higher, higher degree of autonomy. So we're expecting... I mean, O3 is like a degree of autonomy to some. I mean, don't keep looking at her.
That's what you're expecting--
Yeah. Of course, Sas, of course.
Yeah.
And there, because you don't have to, because you can trust me, yeah, you don't have to have a strong critical thinking. And I think for, for, for, for, any man, it's a strong apegreid. So another question is, it's expensive and so on. But the quality of technology has come to this end. I mean, but reasoning is like the kind of place LLM was supposed to come from the beginning, yeah. I mean, like some, I think--
Yeah, 100 percent.
Some, just a logical, interactive way of improving. And where we came to product for a billion people, which is like a pep as cool, but we somehow didn't see real agents even close.
Look, I was just thinking, it would be fun, I'd start out of simple, home stuff like that. I have a dog, and she's changing every detail of every age, so I have an agent who reminds me of how much food she's got, what procedures I have to do, what kind of me I have to do. I'm gonna need to put something on her diet, some medication, so when you do some tests, the vaccinations, so that the agent was completely watching. And that's not gonna happen, of course. You're on your own, I'm on my own. I'm just going to the o3s, asking. Here. You said that, by the way, that--
I'd like that for a baby.
Not yet.
You didn't buy it?
No, not yet.
No, you know how much you have, look, you've got a life--
I've bought Limitless for now.
You have--
I don't know.
You know what I liked? You know what I liked about Limitless? I have Limitless just dead, it's not working for three weeks. I write them in support. The last one was "You know, we found the bug and come the violin of your screen." That was very funny. Here. And then they're giving some new colors, something else, and he just stopped syncing up and
stopped-
Well, I have a lot of questions for him because I speak Russian, fifty-eight in English. That's all that English is fine, he's doing the summary, and he's gonna recognize the text. Russian, of course, will find out terrible. Well, I mean, there's a lot of questions, but there's something about it.
I remember, Limitless is such a corn. Yeah, it's a corn that writes information.
Sasha, hold on.
Johnny Iva! That, by the way, would be fun. Yeah? I'm the one who's watching, I hope the OpenAI, Johnny Eve won't have that kind of thing, right? Well, they won't make a standard product like they do Codex or the operator, or the plaguins that there's something, something completely different.
Yeah, they forget about designs, which is very offensive.
Yeah. No, wait, I think that now a lot of viewers might not understand the reaction, the emotional reaction of Alexander. So...
Well, look, I mean, I mean, I'm saying that, I've been talking to the whole market, with the whole market, just everybody, all of them. And I realized that the multiplier, why I laugh, short, the revenue to the evaluation for top companies, we're talking there, Soona, Anthropic, OpenAI, Cursor, ElevenLabs, that is, the best of the best household names. Thirty X, you can imagine, thirty X! It's not that much. OpenAI, too, thirty X. It's not that much. I mean, actually. Ilon's got a 40-plate multiplier. In the ex. It's like the most fenomenable fandraiser, you know, like a world businessman. Just that we live there with him. That's a good drop! Actually, we see things that happen. Forty X! Anyway, Limitless is less than three of them. And they raised them their first round of triste evaluation. And I'm like, uh, like, that's why I'm so in a way that's the reaction. I mean, it's kind of good, but you know, that Gen AI is at a stage when it's a prime, like, real business, right? Well, like a prime, like a real business, where there are phenomenal growth rates, where you can somehow assess Gen AI companies from a non-hai, pyre viewpoint. In reality, you know, I mean, where the money goes now. If we take Perplexity, Perplexity is something that doesn't understand, it's impossible to discuss. And if we look at the way the leaders of our niche, the real leaders, that is Anthropic, OpenAI, Soona, ElevenLabs, even Ilona, who may not be the leader of the niche, but who's just a feminine dude, If you're not, they'll be there about 30, at best 40 X. And there's a growth, he's like to justify it, right?
So there's a growing phenomenon, which is to justify, he's growing up as he can confirm this assessment.
Well, you see it.
And in quality.
They take X's revenue, multiply it there, like this...
Well, I see.
Well, as long as Ilon can do, shorter. We can't say there's... It's just for ordinary people, even Sam Altman, you can't do that, I guess. But I understand that Elon is a different category, he can do everything. But just keep in mind that we're still talking about the multipliers of 30, 40 X. We don't talk about any 100 multiplier X.
Yeah.
So, how the real world is now set up, that Generative AI would be, you know, people with money, big money, we're talking about billions of dollars, but we're gonna get all these businesses, get them all done by Gen AI. I'm not sure how business is. That's all I'm trying to say, though, that we're still in... I was doing this information all night for me last night. I've been talking to the market, and that's what I've been doing for me, so to speak.
Listen, listen. But if you take this as our final discussion today, I still don't understand how people like Meta and how the X AI will take this... will take the man from the point of view of this agent work that we are now doing. We've been discussing it. So if GPT is on this side now, it has users who still use a functional system every day for specific requests, like Elnar says, this terminal is a specific line. Google's still going to be in his search right now, and there's a possibility that people migrate to it, although I don't understand what's going to happen to their SNL.
It migrates, although I don't understand what's gonna happen to their main business and everything, even though no stock falls because of it and everything else moves. What's going to happen is-- I don't really understand the strategy. For example, xAI, together with X, distribute Grok or Meta with Facebook and with WhatsApp. I think it's a very hard story, very hard. They may not get this exclusive that ChatGPT has. Very different ways.
I don't understand that.
I'll try to find out, but I don't understand that.
Mm-hmm.
I don't understand that. I mean, the very important use of the case, in my view, is very happy about the Grok that they covered a very important case. Let me try to tell you. This is, I think, living API in Grok is a breakthrough of the year. In my opinion, the most important thing. Well, imagine, yes, that is, there's a lot of company training, a lot of companies that need to react in real time. The whole financial ecosystem, after all, is a whole market, a crypt, and so on. Everyone needs to respond to real time news. And if a man needs real time news, but there's no other way with the LLM processing except this API Grok live. That's really cool. I mean, in my opinion, it's like a big business. I mean, well, it's finally clear that you can go to financial services, and that, well, that's a huge, huge potential for growth.
Well, it's so complicated B2B product, right? Or it's-- although it's still API, it's probably a future solution, isn't it?
API is the first-- well, that's what consumers will use. But first, well, right now, there's a lot of business that needs data feed in real time. And it's like a different product offering in xAI, so I think they have more chances of winning than Antropica. Because we're not using everyone on the team, not Claude 3.7, everyone's using Gemini 2.5 Pro to coding. Oh, here, but Claude 4 is now out. I mean, it's not the most recent infa, but it's still an infa that's aggregated over several weeks at least, which is already reliable. In the generative AI era, when you're aggregating for weeks, you're already making such a weighted, a, conclusions. There you go. And yes. So we've seen Gemini 2.5 Pro over a few weeks, more cool than Claude and 3.7, and, uh, for our use, case. And I think that, as an Antropica might have, it's not super defensible, but, uh, HI, I think it's from API offering more defensible business. And I might-- and yes, from this point of view, not very clear, but it's not very clear to look at who, like the second competitor to OpenAI, I might be looking at the Grok.
Mm-hmm. Well, that's interesting, by the way. With this live, the API can create very interesting modules, including for ordinary people, of course. Very serious.
Question: Would API not make other models or is this very difficult? They're always up, they're all doing it. We've forgotten our discussion today.
Yeah, but in China, nothing new came out of China in principle. A lot of news, thank God, from America. I mean, innovations here are thriving. Oh, that's good.
Very good. That's all right, completely underneath. Ilona Mask's office is five minutes away, right? It's all in the side. We're all in the aur, and we're in the aur.
Good.
Good.
Well, Claude, you should say that, too. Sasha mentioned that Claude 4 was out. I wonder if they're also on the side of agents and agents' codeing. They even have a separate baccalaureate in the presentation after SWE Baccalaureate. SWE is software engineering baccalaureate, that is, programming. There's another agentic coding. So they have their own record of agent programming. And they show that their new Opus 4, Sonnet 4 models are in the middle of this case. Here. But again, Antropica's on API is very expensive.
Well, as far as I can see, Opus 4 is very cool, yes, but it's worth a lot more. I was looking at the price last night, and it's almost ten times more expensive, you know, right, right?
Yeah, yeah, yeah, yeah. They're working at all, too.
Well, we should watch what baccalaureate was for, what tasks were it tested. But in general, they showed that the models were very good. I haven't had time to touch 'cause after the presentation, you're gonna come by and watch, and there's probably a world watching and the web interface isn't coming out. I spit, I think, okay, in a few days...
Yeah, that happened. Here we are on Friday, you're watching the graduation on Sunday. It happened yesterday, didn't it? They put it yesterday. Yeah, yesterday, Thursday's up. Here. I, when Tanya said she needed a child agent, I thought Tana needed 100 agents, so one dog, one cat, one bird, one first baby, second---
Every client.
My husband, my nanny, every client, yes. To the garden, to the pool, to... That's all.
Personal agent on the client sounds good, by the way. I promise.
Yeah.
That's very fun. Actually, we're still-- I'm saying, I've been very advanced in the O3s in the last week in my understanding, although I'm a very technological man in automate everything, in terms of understanding agents. I'm not even. And I'm even closer that the agent will be so simple through some kind of command line in the general interface and even for clients, like Tanya, there's a number, I don't know, dozens of clients, there, Millions of clients, they're safely covered by an agent.
That's how millions can be put, if there's some general environment that leads them.
Sounds like a dream.
Yeah. And, and that would be-- well, we once told Elnar and I a couple of years ago, we were in the crypt when I showed up, I, personally, I, uh, clearly said that the CRM system wouldn't be in the current form. Let's be honest. CRM is a thousand-person system, 2,000-man-- this is a command line, not needed. Well, why? That's quite enough. You don't need to go to three-- you, I mean CRM, like, the current pattern, the cards are opening up all this. Why? Why do you need all this stuff? It's a command line. It's all out of date, yeah. And you're very quick, yes, very quickly, you get any infu. This is a new visual, new scenarios, new concepts of UX/UI. I think that's what the key problem is. And now, OpenAI, all these tigers, there's some sort of product going through there, some Marquetologists, and they do this UX/UI product development without understanding the future, they're not getting anything, and they're not doing anything. I'm good. It's gonna be a new, new concept of UX/UI, and a new concept. That's who CRM will do, that's guaranteed to someone. They'll make CRM for a few thousand users with a command line. I can see how she works. And I've got, uh, more than ten businesses. I see that in most of the small, small, small, small use, there's a way to leave it. And where big is, it's multiplier. Every man has the same terminal. Well, there's some sort of paperwork, there, kits, and everything. And it's no longer necessary, this visualization, some separate BI systems that, well, I mean, the schedule shows that they're disappearing at all. What's that? What's that? Why is she? Why would she want it? Interesting. Yeah. Please write what you think about agents and what scenarios you'd like to use every day in a normal terminal, in a normal command line. What challenges would it be? Thank you very much, everybody. I'll see you in the composition in exactly a week. And on Wednesday, we have a special issue. Bye, everybody.