Hello, everybody! We're on ToTheMoon. Technological news, Silicon Valley sites and the world. Finally, the Silicon Valley came back. So, it's a full heart. Yes. So I want to try and... I want to start with. There's a man in here putting the next case in X. He's at ChatGPT's, uh, he's got a voice assistant and said, "Look, I'm gonna run right now. Can you help me with this run and the po-- how is this run going to work? He said, "All right, I ran, how long did I run?" He said, "You've been running for 12 minutes." Well, it's been like five seconds.
All right. I've started timing. Just focus on your run, and I'll be here when you get back.
Okay, I'm back.
Welcome back. I've got your time right here. How do you feel about that mile?
How, how long was it? What?
You clocked in at around ten minutes and twelve-
Ten minutes. I s-- I swear I was faster than that.
Sometimes it feels faster in the moment.
No, like, I think I was closer to, like, two seconds.
Oh, if only time worked that way. But I promise I'm giving you the real time.
You, you timed me for ten minutes.
Okay, but, like, if you c-- if you're not able to do this, you can admit it. It's okay.
It's totally okay to double check me, but I promise I'm doing my best. If you ever want a different-
So you got ten minutes. I-- It took me ten minutes.
Yep.
And Sam Altman sits and listens to this case and says, "Well, it's okay that it doesn't work. It'll work, maybe a year later, because it's got to be a separate timer."
Yeah.
Have you seen this guy Husk online?
I don't think
so.
If only time worked that way. But I promise I'm giving you the real time. Uh, may-maybe, uh...
Do you need to show that to your product guys?
No, no, that's a known issue. Maybe another year.
Another year?
Something like that works well.
What's that known issue?
Um, that model doesn't have-- that model, that voice model doesn't have tools to, like, start a timer or anything like that.
Fair. Okay.
But we will add the intelligence into the voice models.
So what I listened to, I had this inner head. Next story. What's the timer for? Like a mini-- well, there's a simpler solution. And then I remembered the case. I don't know who-- anyone does that or not. I've written in the pre-conditions two years ago that every time you text me, you should write a date to me at the beginning of the message. And because I care when you're in the same chat room or you're in different chat rooms so I can watch the message, because you can write on different days, and the dates are-- that's not. There's still no similar construction inside to see a date there, time. That's super important stuff, actually. That's about sending messages. And in fact, this time-taking issue is very easy to deal with. It's not much of a need inside when you're ChatGPT saying, "Look, you were two and a half days ago interested in that information in that chat or talking about, like, you've got some kind of disease." I remember, we were studying with a buddy, I cut, uh, cut, gave ChatGPT a study, uh, not a study, but I gave my ChatGPT treatment. I told him, so this morning I had a problem like this, on my trail... and tonight it's like this, and this morning it's like that. I understand that ChatGPT is, it's working, it's not like I've ever understood the morning and the evening and I haven't recognized them. Although, in the idea, in the past history, he would have to compare it at least in his time. And, Ilyadar, you don't think it's weird, from, uh, first, uh, to decide, not to solve this question, Sam Altman's answer, his very little weird response that we have other priorities right now, we're kind of in it. Come on.
And to solve this, the timers must be introduced as separate applications, he said.
Oh, I heard the news, I don't know how yellow the press is, that Sam, well, a lot of people say Sam Althman is a software engineer, not a very cool specialist. Even so, as a basic, well, from a junior, middle specialist perspective, he's, uh, not very well aware of architecture or how machine learning works. He's good at another. And, well, we all know what, right? There's a big company building, some intrigue, huge budgets, product production, etc. But it's technically that he's weak enough. And if he answered without training, I can believe that, and I'm having a problem with him, he could just go to sleep without knowing what he was. There might be a lot of decisions. That's what you're talking about, if time is available at the time of the generation, and it's probably available, no extra timers are needed. Yeah, you're just like a moment, uh, a request needs to be made, and that's it.
Yeah. Not much of a voice assistant, but how does it work? Someone's gonna think that a separate session should be started during a voice assistant. But the voice assistant, actually, if you stop him, you'll see the message generation, right? And you have a message generation inside and work on it. You also have time, time to report. They're, like, I have new records, and there's 100,28 minutes to generate an answer.
They're setting up 100, eight minutes, I don't know, fifty, six seconds, right? Oh, by the way, 100, twenty minutes, eight minutes, fifty-six seconds. I have a IT business, and I was there with one partner discussing one potential, there, a project. And I told him what hypothesis? I told him that, to date, all the softs that people are developing from the outside are different applications, like some voice-assisted sales assistant, CRM system or, uh, The text editorial assistant is all very cheap models. And even if we look at the research that's rolling out, I'm already doing this research, by the way, we're not even gonna stop quoting these studies in Tuzma because one of the last ones. research, there, Stanford, very big, they were doing it on model four o's. How, for example, is the system? Why would I have to do that model four o'? We are now, for example, with the ChatGPT at five, in Gemini, in other generations, in Claude, we are already in the other generations of models. It's a totally different class of model, isn't it? It's even a model that can be... two hours to think about what to say. So we're in other resources, in general, in other parameters, everything. And they did research, they say, "We did a study early in the twenty-fifth year." Guys, the world has changed, everything's gone. And it's like people use cheap models. I'm telling him, "Because you use cheap models, the quality of answers in all these systems, it's always gonna lose." So I'm basically everywhere in business, saying, "You're not gonna have to outstay the data, you're gonna analyze it with the help, there, the versions, the heavy versions, all the ultra versions, there, Gougla and Grock, Claude. and ChatGPT, yes. I wish you were doing it all by myself, which I did. But we had a question for what? Since we wanted to create a system, I said that a qualitative answer could only be obtained with the help of the version. I mean, even ordinary hard thinking, but hard, because it's used, for example, to use a huge analytical body. I mean, you have to analyze a lot of data, and you-- you give it a really hard answer. And he's dispersed for every man, not personalized. So you have to dig a lot of things.
So let's see how many, uh, access is on API. So listen to what an interesting case was. And interesting, uh, we have a lot of eyes on both the developers and the developers of different artificial intelligence systems. He says, "I opened it myself, there, the develper of the OpenAI studio, just to test your request. Send your prompt or some kind of temporary prompt. And he says, "The first thing I've ever written was a system to use the version, walk KYC through API." KYC is knowing your customer, yes, that is, the OpenAI verification does to prevent people from transmitting models. Why would you do that? Someone's gonna say, "Oh, I'm gonna go get Grandma, Grandpa or some friend." It doesn't work like that, it's a tough story. So, modern KYC, they're scattering a huge pool of people, countries, there, and so on. I mean, it's gonna be harder to access more expensive models, first of all. I mean, it works. We heard you and I, yes, that OpenAI is launching the checks. That's the indicator. Two, he started this story, and he was curious to know that he started, uh, four O's. I'm saying, "Wait, I'm using five four pros, not four Os." And, uh, he was working 11 minutes. Eleven minutes, do you know what he was given? She gave him the API result, which I'm sorry, I can't give you an answer. And I eaten him some more tokens. In fact, by recalibing, uh, my request that I made in the prom, and, well, we were discussing with him to make an annex for, uh, the usual consumer market. I mean, it's a mass-market, yeah, that's between ten and a hundred dollars a month, like, yes. That, uh, that's clear.
Uh, it's clear that it's very important to have the cost of the request. He has a lot. And he said that-- uh, I said, "I understand correctly that the request would cost $10?" He said, "No, your request that you send would cost $550-seventy dollars." Uh, uh, uh, uh, I was kind of in doubt. He says, "Maybe more. I mean, I don't know how much, yeah. Because when you have 100 or 20 minutes or, uh, an hour, the system is analysing some kind of data, that's a whole other way of preparing to, uh, that's a whole other cost of this story." Well, I see we'll continue this analysis and continue to calculate this number.
I wonder if you'll share our views with our viewers what's going on in this part. But what do I mean? What, the world is modern from a development perspective, we'll talk now. Now, the record block has been built into artificial intelligence, just a space number. In a world, uh, in a start-up, not in artificial intelligence, uh, in a start-up world. And, well, I didn't say anything about artificial intelligence, it's got, uh, most of it. We'll talk about it. Uh, the point is, in modern history, uh, I see this problem more and more, uh, system differences. I mean, I'm just for myself now, uh, I'm so much of a difference from the plus version. I mean, I have a few people using plus, I've become their pay-as-you-go, Thinking Heavy or Pro mode. I see this huge, just like-- space difference. And on the other hand, it turns out there's a very big break in API. Whoever uses, uh--- do people understand when they're working. So if people don't know the art of intelligence, you brought Sam Altman, who-- well, we're not gonna get a believe in what you said, that man doesn't know how to put timers in there. But, uh, basic people who don't do, uh, technically, they don't have answers like that, so, so, so light, huh? And, uh, it's clear that people who don't, technically, don't get a lot of attention, then they don't understand that, uh-- and think that, for example, a voice-speak analysis on sales, he's there. supernatural. There, or an analysis, uh, or a text-processing, well, until some one of them is supernatural, there, and, um, or an analysis of some architectural or some sort of arrangement, it's, uh, some medium. Why? Because they use apps that use cheap API, and often they're not ready to pay for these apps or pay for them, like, ten, twenty, thirty dollars a month. I mean, I'm morally prepared to make myself that if I'm paying two hundred dollars here for Pro, there, or two hundred fifty Gemini dollars, there, and so on, I'm making myself morally, that the moment will come when I get here. And they say, "Alexandre, I'm sorry, there's no more halava. And now, the sum of your requests is $5,000 a month or $10,000. And I-- that doesn't mean I'm gonna pay $5,000 for these requests, because I'm using it now, just clicking buttons, reset the questions. I'll start thinking more about it. And by the way, uh, I-Ildar, you can say that, yeah, that Anthropic was just for OpenCL, uh, cut the chance. You read this new-- you saw this, uh-huh, the subject? That they cut the possibility of using just because of the cost. Because of the cost, they have eliminated the use of OpenCL, or have they removed the possibility of using the OpenCL, or have they been using the normal tariffs, or free tariffs, or have they? Because it's eating just an incredible amount of money.
And a computer that is catastrophicly lacking.
Yeah.
And that's what you're saying, besides the difference between those who use it, it's cheap, uh, API or, uh, cheap subscriptions, expensive signatures, it's all moving to what's happening now. Another layer where models themselves will be available only to a limited number of people.
Yeah.
Yeah, when we're talking about Anthropic, which has now produced their model, uh, a great, great model, but we know about it just now? That there's a double article, there's forty-four pages I think. There's a detailed description of the baccalaureate, with everything. We hear that super-crunch model that it's from a cybersecurity perspective, that she found a baga that was twenty-seven years old, nobody's ever found them before, and she's really cool. But it's only available to a limited number of companies. As Anthropic says, it's because, well, it's too dangerous to release it into general access, that there's a huge amount of business to do with it. On one hand, yes, I suppose so. And on the other hand, again, returning to the thought you were saying, uh, Chinese and any other company would not be able to copy these achievements so quickly simply because the model is only just being issued. There's a limited number of people there. So it's possible that even with the desire to pay $5,000 to some models, like-- well, it's clear that there are models we don't even know, but there's gonna be such well-known models that access to.
It'll be, well, impossible.
Look, it's like the extra Sam Altman said, I don't remember, we were talking about it here or not. Uh, he was talking about an intelligent intel, and he was back on his idea. I remember a year ago he said that, uh, there would be a situation where you could throw $20,000 in to solve the problem. He said he'd be in a situation where a billion dollars worth of money would be needed to solve the problem. Milliards! And here, look at the very interesting story. The discussion is that the question is not just billions of dollars. And in their new document, perhaps we'll take a special edition of it, maybe some part of it tonight. They put it there, OpenAI outlined some description of the future of artificial intelligence, how it will be managed within the life of people, countries and the rest. So there's a concept of time that's a little time to work on this task. Time, uh, that's gonna take some time to get the job done. I mean, not just spending two billion dollars or five billion dollars, saying something else to solve the problem, it's two and a half months. And there's a next case. Again, I'm in contact with one of my partners at IT, and I'm go-- he told me I'm using mostly ChatGPT Instant. He says, "Ty people like you, who use a fanatic system, a little." I told him, too. Now I'm with my comrade, comrade to Cairo is coming to me, and he used a system of instant fast answers, yes, rapid generosity, but with a search, quick generosity search. I said, "Loudy's not used to talking, people aren't used to thinking. People get used to what they're waiting for a quick answer right now, and they don't even want to figure it out. It's the same thing reading the headline news." Look what's going on in the world. Now, in the reconciliation of the conflict, Iran and the United States have... they say they agreed to some lists, and you watch these lists, and these lists are completely contradictory to each other in terms of points. reconciliation. And then the Vice President of the United States comes out and says, "You know there are three lists of ten points. The first list we agreed on, the second list that the Iranian side put out, and the third list that was then put out after the armistice. And we agreed on another list of number one." And you sit there and you think, wait, so that is how can it exist? And he says, "And there's still our initial list of 15 points." Yeah? And you're running a head explode at this point of time, you don't know what's going on, and you realize that you're gonna need time to figure this out. And you're gonna spend time or you're better off than getting a response right away. Instant use instant is a habit, it's like people in Ticktock. Now I'm just saying, because I've been traveling, watching a lot of people, I've got a shock sometimes when a man sits, he, he sees it, he makes a swap every two seconds. I think he's the same, he's not watching the skate. I mean, if I get a monkey that climbs on a tree, accidentally there, I rarely watch racels, but if I get a monkey, I'm interested in her, I'm looking at the end of the tape, and he doesn't see anything. He, uh, I see he had a man there, he had some girlfriends, someone else coming, so some animals, anyone, he's all over them. Whatever goes, he's always sleeping. I mean, he's watching--
I've already revised it.
He's watching seconds of content.
And it turns out that to start using real artificial intelligence systems, I repeat it, repeat this second year in a row, I can give you a name now, an executive dir-- and a right-wing. Sam Altman's hand and one of the main directors of OpenAI. He said in one of the sub-cadasts that to put in the right intellect, they should start using them all the time every day. So, uh, using is a new, new scenario for yourself. The new scenario is that you're calm about waiting for 20 or 30, or 40 minutes, for your task. So you stop treating chat like he was originally, yeah, when he just sprayed the answer. You realize that there are tasks I have, I'm starting it, and it's back there in a while. I know I could, maybe, even the next day, see the answer to this task. And Google, by the way, has now introduced a new concept of tariffs in API.
Two new tariffs. Well, there's not two, there's a new type of tariff that's coming in when you have the opportunity to deal with this problem in synchronous asynchronous tasks, that is, tasks that are being pursued simultaneously with you, there and in parallel, parallel to your process. The point is, you have the opportunity to buy a more expensive compute. And we're back on it again, aren't we? What if I needed an emergency, like I jumped the stock market again yesterday. Well, obviously, why he jumped. Because the day before yesterday, they said there was a truce, so in the war. And yesterday, he slide back a certain infrastructure market. And when you sit in the morning, if you're gonna-- you're doing this and you're doing it, well, less than half-professional, I'd be willing to pay a hundred dollars in the Pro version and wait an hour. Well, as fast as you can, pay me...
Well, as fast as you can, I'm getting my answer, so I'm gonna get a quick analysis in the morning. And that's a big story, because when you have a market opening, for example, you need to make a decision what's going on in the market. And I can't do that because the algorithms of those I need, and the amount I need, there's a thousand resources to check, to see the different blogs where someone bought, how the stock started working in the morning, I'm not sure what the new agenda is. And there was a truce. And the next morning, it's just that the truce has been re-enacted, right? Again, we're off again. And you, you know, you're moving a chaotic market from a financial perspective. I would pay extra, actually, not a hundred, I would pay more, well, there, 100, two hundred, three hundred dollars to do this, but I would have an answer instantly. I mean, we're obviously... That's the Google API, right? Is this for the developers, Elnar, the right story? Mostly for the developers.
For the developers, yes.
Yeah. You haven't seen her yet, have you tried her, have you?
No, no, no, I didn't use it.
Yeah, not API, because I think they're the one who put the story on API in the way that you can pay the money and the answer, you're actually getting kicked out of your max. I see you're standing in line, too, but in priority. And there is when you can say, "I'm ready to wait." And it's nice to see when you're actually saying, "I'm ready to wait," and you're leaving a little back because there are answers we're putting on the system, especially like me, users, mass data, and in fact, there's no problem. I started one business there this morning, there's a lot of real estate analysis and analytic large, big data, right? There are thousands, thousands of different figures. And I kind of expect an answer. Oh! I see he's here for me now. I see he's here, he's come to me, he's been working c-- well, for a while, how long is he here? Thirty-three minutes worked, but I could wait half a day. I can't say that I was in a very rush. But when you're sitting at the meeting, I'm gonna start after the tape, there's a meeting, well, with a lot of people, at this meeting, of course, I'd be cool with expensive models. I'll get a reply. It is now clear that all are leading to the example of development. Elnar, why did I ask you? Because I see Google as positioning it in terms of sophthing. So write to whom it would be interesting to receive answers faster and in which case, which boxes are possible. That's Tanya, for example, I think it's a very straight case, huh, Tanya? One thing is, you're, like, developing some drawing or model or something, and you're doing it at home or doing it when you're just not working with clients. Another case you and your client are at a specific time in a particular space, and you need to, for example, generode some sort of thing, I don't know, build a 3D model. Or you should count some kind of cost. Like, you have some cost-account algorithm, and you know you're gonna miss the story because you have to totally blend out different criteria. You asked the man, and he gave you an answer not in an hour or not in thirty minutes, and now the time is gonna increase, but you and I will see.
I think we should-- there'll be days. And it's not a story when Elnar says they're on the dock. Well, Elnar was joking. Well, it was a joke that they were writing a number on the dock. That's it, it's getting bigger because first, the systems are getting more cool, and secondly, because you're in the line, obviously, right? So ChatGPT has more gr-- they have Sam Altman asking. He's got to get a quick answer, for example. I don't know, the American government is asking for it, and he needs a quick answer. And, Tanya, you can pay, right? You can pay, there, quick, ten dollars, parole. And you say, "Of course I'll pay, I'll show people."
Yeah, I definitely paid.
No...
Well, for visualization, because the calculations are quick. There were no questions about work that you said quickly called for a response so he could think for 30 minutes. That wasn't the case.
Well, if you're looking at this, imagine if you came and showed you, like some architectural or plan, you've got a picture of these plans, a lot of numbers, some estimates, you've got a picture of them, too. You have a base of your old estimates, and you need to run an analysis and a comparison. This request, Tanya, good about, uh, heavy system will be at least 30 minutes, because this image analysis, a lot of different data, some graphics, drawings, drawings, is still a thing. He'll leave. I see, Tanya, if you ask him to count--
We need to recheck, yes, it's a very dangerous moment.
Oh! Oh, look, big system, you-- you can put it in the prom and tell her that after calculation you have to make a check, check it out and then re-check it after the check, taking into account these. And it's at the end, in the document or in the answer, you can have control checks. So you started the opportunity inside your own pros and pros to make further checks. Although they are now starting them, too, on the regime, on the regimes, on the regime, and on the regime, it's not heavy, it's called the Standard and Extended, that is, the usual extended, yes, in the pro-descript. It's heavy in thinking. But the point is, he'll waste time. So, Tanya, not a calculation, I don't know, a number of stoves that should be put on 20 square metres, right?
Well, there, he's gonna make it easy, yeah. Yeah, Elnar, you started this.
Although Google released Gemma 4, we can talk about it, too. And the third story that OpenAI presented his Spark model. Remember, right? Which is very fast-tracking code, but it's small enough. Here. And on the other hand, NVIDIA was talking about it, and we also talked about it, and we presented our conference, and, uh, chips that built a story from Grok with Q at the end, right? I mean, if OpenAI Cerebrus, then, uh, NVIDIA works with them. So there's a iron specially plunged under high-speed inference. He's not as much as the other iron we used to. So, accordingly, if you get inference, you'll probably need to pay for that equipment. Yeah, and you're not just gonna have a priority, uh, in the line, you can still roll on another iron. Yeah, until it's got huge giant models on it. But it's all about time. Six months, and we'll definitely have iron on which to roll big models very fast. So the market is moving on that side, too.
I think it's the near future. What, what do you think of models that put on the phone? Look, Google this week, there's a monster model, and Google, uh, went on the other side. She released the Gemma 4 model family. And it's a completely open source of the model, yes, they're available, if I remember correctly, Apache 2.0's license. Tanya and I talked about the license, too. And these models, what are they? They're, uh, very small. So there are options that are put on a regular home video card, yes, which many people have on the computer. There are models that even put in the phone. I mean, on the IPhone of the 20-thirty tokens a second, she's a genius. I mean, in general, these people can read slower than, uh, the model of genetics. The specialty is she works in isolation on your phone. She doesn't need any Internet, no data leaks, everything's in a closed circuit. There's Androids and IOS, and it's not as bad as it is, but it's got to be frustrated to turn it around. I think Tanya, you were suffering more while the OpenClaw was trying to turn around. I think it's gonna be a little easier here. But the question is, you think, actually, here, you and the audience, is that what you and the audience need is a model like that in the ordinary life, right? So you'd use them or not use them? You don't need it? What do you think about that? It's not exactly ChatGPT 5 Pro. Yeah, that's definitely worse. But a completely enclosed ecosystem.
Look, I'll answer the following. I think history, well, I think history is something that's a professional question. So for people who do professional systems, they're dealing with professionally different episodes, and, uh, it's an important, important story. Or for someone who's gonna make a application, well, some application that has to be in a certain isolated environment. Not just a prank. I know a man who makes models, even money attracts the alleged ChatGPT stealing your data, and he won't steal the data, will he? But I'm with it, I'm just laughing at it, over there. Or he's got a button in there that can remove all his information about himself, right? Well, you look at it and you think, you know, some kindergarten, I don't know. I mean, I'm in such, like, stories like this. But if someone thinks and says that, look, it's necessary to have a model and an app when you're in the woods, and it's supposed to work for you in this environment, for example, to scream, and, like, you, You move, for example, into the forest or watch animals, for example, in the forest, or watch plants, for example, determine in the forest, do. I just gave a simple example of this one, maybe a movie recently looked at the man in the woods killed in Yosemichi and-- a show. The plane was flying, maybe it was because of it, and I thought he was using a sophth to see his own spot. But it's a working story.
But if I have an Internet, like that,
If I have an Internet, like, at this point of time, I'm using my common model, I don't need it. Like a secondary story, rather than a model that's on the phone, it's like a kind of a backup. I mean, basically, if there's a ChatGPT that works in a cloud, and I have a ChatGPT-- I always like annexes that, uh, have a... a lot of downloaded data. Suddenly, I have Booking. If I don't have the Internet and I open the Booking in another country, I can open and watch my hotel reservation. That's super cool. And there's an app you open, and you don't have the Internet at this point of time. Not because I didn't buy it, but I just got it, it's not-- it's not available. You open it, you're not loaded, and you can't get a plane ticket, like, to see.
Yeah, yeah, yeah.
And you're like, wait! Or you came up...
Especially when you're asked at customs...
Of course!
... show me the ticket back, and you don't have the Internet.
You don't have the Internet, do you?
And you-- Tanya, here's the case. I mean, it's--
Yeah, yeah, yeah, I had that.
It's a question of customer thinking, and when it's really about the client. And, uh, in these situations, Ilnar, I think even big models are gonna come to this because, uh, it's like I have, like, my pictures, uh, in-- it's not on iPhone, it's in iCloud, Because no iPhone fits them, right? There's 100 grand or so many videos of everything. They're stored in iCloud, but it's got some volume on it. I mean, I have a balance. I mean, I think that in terms of systems, there's got to be a model that works on the phone, and she's... I'm saying, as the main point of reference, whether it's professional or something. But independent, having a model like that, but only if a narrowly specialised solution is.
It's really gonna be interesting how our viewers write and, uh, what they're gonna say about it. And I see a lot of this topic now, when I have now put up a separate paper, where he has written his corporate rules. But it's a clean policy, which is supposed to be more than just a reservoir. They wrote that they were not just a reservoir organization, but an organization to do corporate products. And then if they reread their paper, they describe the decisions for some space companies, well, for some companies that, uh, the greatest of the greatest. I mean, you're supposed to be sitting there, I don't know, five hundred programmers who'll just build what they're writing, right? And they also use those terms just when the terms are for terms. That's what they say, we're not in the era, uh, AI helpers, but AI agents. I'm sitting there and I think, uh, what's the difference? Well, that's just words people call. And even now, if someone tells me, the OpenAI can't show any normal AI agent files. Well, seriously, they say they've improved there, reduced time in Morgan Stanley, or somewhere on, uh, 90% of sales managers, released their time, and now sales managers are talking to clients. But you're like, "Hey, what does this system do exactly? What kind of agent is that who puts them in position? That's the same thing, Ilnar, with these models. Yeah? Here.
I read this news, you know what association was? Uh, a year in fifteen, 16, probably when we went to Europe, uh, swivelling cards where you're going on the phone. I mean, it's a Maps. Uh, there, uh, there's a probable, you're on your way to Prague, there's a Prague map, some main routes, a trail, so you just have it on your phone, uh, because the horn is on the road, and the maps are online, too. I don't remember what stage they were at, but somehow, that's, that's, you always have an app on your phone that's filled with cards, and that's where you use those cards. It's a similar story, Sasha, remember when we were going on a trip to the silence? Well, that's, uh, that might be, in the mountains, there, in, there, Carelia, let's say, there, yeah, somewhere else. The boys also often had maps that were already flooded, where the routes were built, and they were moving along. Uh, in town, when there's a Internet, like there's no point in it now, because you have Google Maps and other things, they're available, they're good at your location, and you're actually on your routes everywhere. You're building. But if we say you're leaving the Internet, then maybe some kind of domestic use in these models might be. Like, if I imagined I went somewhere in the mountains where the Internet doesn't catch or it's crazy expensive, and you still have a conversationalist, a searcher or something on your phone. Well, if you use chat-bot as a searcher, it's generally like you can find some use. Of course, leaving behind the bracket a large number of applications that can be created with the processing of the inside on the phone. I mean, assuming there's a skill in there, iPhone, and it can be incredibly-- oh, uh, Apple that iPhone can build, like, models, and these models will be in the background to decide some kind of models. User-friendly tasks. It's like a story. And if you just take a chat-bot, I think it would be pretty fun.
You went somewhere in the silence, the Internet doesn't catch, but you have some kind of interviewer that's pretty good.
Very rare case, Ilnar. Even when you ride in the silence once a year.
Yeah, but I, but I remembered how we had some trips to the coward when we were driving, like, there wasn't an Internet. We're five, six in the yards. I mean, it would be interesting to clarify some questions, wouldn't it?
Or for countries that have--
What data does he use, which you pre-loaded? Or how is that?
No, no, no. You've got a simple model, it's not very big. I mean, it's not very smart, therefore, in terms of the set of knowledge she has, but, accordingly, the Baccalaureate shows good Gemma 4 results.
Mm-hmm.
And here's the story that I think you can build. Tanya, your question is how to train them and how to do what, and she should know a little more. And actually, there's no Internet in many countries, or there's a lot of Internet problems, or even the Internet might be a problem when you just move in the car where you're going. I mean, it's one of the most--
Cars, yes, or other-
Yeah, big enough. It's just that you're out here, and you're out of here, and you can get a connection from time to time. Or, for example, if people work in space, in basements, like, some kind of building, where everything is completely isolated. I mean, there's a big plus in this, I think, of course, in the combination. I think that at-- llinar, when the models are combined, it's gonna be a great choice in the offline version. And maybe whoever's gonna do it first, I think someone's gonna do it. Anthropic doesn't write about it, but I think they'll do it. Yeah, that idea, it's not on the surface. I saw her like this this for the first time today that Anthropic or Google, or ChatGPT OpenAI. If they do it first, they'll get a lot of clients. I think that by the fact that you're going to have an offline model, even if it's like, uh, a little more simple, like a regime, there, a facility, and so on. And I think that, to date, given these accelerations, the programme part is now very accelerated, yes, and the memory costs are minimized, especially by using the iPhone.
I think it might be decided. I don't know, you read or read the news. I'm not paying much attention to Apple now, but they're writing that they're gonna give Siri the opportunity to use different models. Yeah? And, well, as Siri would become more than a separate assistant to his own, created, and it is alleged that different models can be intact. I don't know how they'll do it, but it's a good idea on their own. I mean, if you could use an iPhone for the right thing to create a story so you could use the power of a-- like processors and the fees of an iPhone to get this annex on the inside. And you'd have had a synchronization of GPT's tag in the offline when you don't have the Internet, and you need a quick answer. Here. That's right, this is a super cool story, it's like a quick or long answer. Yeah? When we're talking about you need a response in two hours, you can wait for the Internet to come. Yeah? Here. It looks like it works in this plane. About how we can tell today. Although there is a very interesting subject, because Google has conducted a study on model coherence through behavioural tendencies. They didn't do a study, they suggested introducing new baccalaureate. They said we don't study, we're sort of studying the series as if the model was smart, I don't know, math or programming, but we don't know if she's acting like a human being.
And they offered this research. Good things, of course, Google offers. It's important they do it all. But before we go, I want to say that, uh, the first quarter of the twenty-sixth year is a record investment in the start-up and venture industry. Three hundred billion dollars. And it was at the beginning of two hundred and ninety-seven. I like it, then it says, then it's triste. What's the important aspect to do? That it's about 6,000 starttaps around the world in a block. And-- but there's a nuance. I mean, when we see industry growth, there's some nuance. One. First, which, by the way, the first quarter is basically 70% of the total expenditure of the twenty-fifth year, right? What's the nuance? That two hundred fifty billion dollars or eighty-three percent is the US. And inside, these eighty-three percent of the base--
Interest, the main mass is artificial intelligence. And within this story, 65 percent is, uh, four companies: OpenAI, Anthropic, xAI and Waymo. Yeah? Oh, well, Waymo is a Google drone in this case, and they probably had a round. So we'll think it's Google Investment. Aaa OpenAI, Anthropic and xAI, draw attention to the area where we have been in the world, the amount of investment that is currently being spent in three leading, a-a-a-a-a-a-a-- let's keep the xAI in the lead, because I'm sure it's a company that, uh, is in the infrastructure, uh, world start-ups, perhaps the most expensive world start-up. If Elon Musk is on the IPO, there's Star-E-E, everything that sca-- is connected to satellites, right? But the point is what? That they've attracted the largest amount of money. It's kind of fun that the venture is growing, but at the same time it's not-- it's not a venture for me, it's an investment and a break-up of huge, big, big companies from all the rest. And Elnar said a very nice phrase at the beginning, that a lot of models would not be available to anyone at all. I mean, I think this is the breakup, which leads us to see the real models any less. I mean, that's how I see the story about how much the story got out. I mean, I've told you about it before, I said that you should be able to do it, test it, and now I realize that this is just another world. So if you're still thinking that $20,000 is just an app, I'm gonna spend as much as I'm gonna spend on ChatGPT, right? It's like people treat the toothpaste. They buy a ten-dollar toothpaste and say it's expensive for twenty, then go to Starbucks and buy a drink for ten dollars. I'm like, man, wait, you're toothpaste, if you use one, you use it three months twice a day. So what's the difference is, it's $10 or 100, right? Well, unless you make $100 a month, but then you don't go to Starbucks. But if you can drink coffee for five dollars or, or in Moscow, people drink in-- coffee at Coffee Man for ten dollars, right? If you can drink $10 coffee for a week, then the subject is just a pasta, you don't think so. But people who even drink coffee, they answer that. So, a lot of people, they have such an answer, it's a habit, a complicated social environment. And we have a habit of treating so much as a soft design. And that's why I just wanted to reflect these numbers, uh-oh, which makes sense that there's a lot of start-ups that are being done. But if you look at this, two years ago, we were very much speaking at ToTheMoon and saying, look, the start-up with the S-- from Sweden, the Lithuanian launch, the Dutch launch, France. We've stopped doing this. Why? Because that's what-- it's got to be done.
So you realize that such serious things have become done inside large systems, like Google. Google is a big deal. Now they're out of two new ones. We told her-- in the background, I think, since they're doing research tools. They've released a research tool to do the schemes, and they've written a special algorithm. In fact, they-- they're positioning that they have five people, there, one in the D-- to design, the illustration, the other, there, to make the scheme correct. And you have the opportunity to create a visual pattern for the performance of a man. Someone's gonna say it's already worked there, inside all of these systems. You can put down any document and say, show me the schedule. But it's not. They do it at the professional level. Well, not what's at the scientific level, but at the professional level, when the system is isolated about the analysis. Your picture, your image, your chart at the end, it's really the most appropriate thing to do with what you wanted to get from your texting document you had.
A very interesting instrument, it's called, I'll tell you what it is called, you can analyze it separately. It's called Google, Paper with Agent. He's doing scientific illustrations and text schematics, right? There's another word of "scientific" there, but I'd be a little bit abstracted from the word "scientific" because if I read it so, I read the core of this story, it's very interesting. Who needs to cook, you know, really, even if you're analyst in marketing, yeah, you're analytic in, uh, well, any absolute, yeah, there, on sale, it's a very cool subject for a man. To show in fact that there is even, possibly, Tanya in the professional market, yes, in your, in your view...
Yeah, I'm pretty interested in that, by the way. I want to...
Yeah, that tool might be useful because it's from the point of view that they've all given it, that's what I said, five specialized actors. They say we have a search for examples, that is a system that seeks references and similar images from scientific literature. Then the planner who sets the contents all over the structure. Then a stylist who's responsible for visual rules, academic style. Then the visualizer that creates a drawing or even generates a code in Python for static graphs. And the critic checker, which compares the result to the original text and, if there is a discrepancy, sends the visuallyst back. In other words, it's not one smart model, it's a mock-up team, right? Well, this team, it's clear that it's some-- that's a pyre, but I-I--I-I-I-I-I-I think it's really interesting. I mean, very, uh, well, powerful enough. Strong enough because they, uh, they did, by the way, to actually create this system, they've been generating, uh, a special set of tests, and they've taken two hundred and ninety-five cases on the basis, there, various publications and received different sorrows, and they've been sat by people. Well, they were doing a very serious job. And I'm saying myself that I just, maybe, don't have to lick something out there right now. I can even do something real, send, and people have to watch me give it to you. But they just have to, no matter what I write, what format, right? But I would-- I'm really interested in getting there. I hope I get my hands, and I'll check it out. I'll check it out, I'll study it, I'll see. And I don't know, Eldar, you're making or not cooking, uh, some sort of a combination of analytical information, that's, but it's a scientific information like a code study or something else.
I'm telling you, by code, you're gonna test and tell your opinion exactly what you're saying in terms of useful things, right?
Yeah, I just sit there and regret that in the fourteenth year, when I wrote my dissertation, I needed an illustration. It took a lot of pain to add graphics to your dissert. I've never been able to draw. I understand that it is now very easy to do for scientific articles and humanitarian and scientific mathematical. It's a lot easier to do now, and it's really awesome. That such instruments are coming, Sasha-
Did you draw yourself or did you give to someone who had some specialty-- was, like, some kind of specialist?
I was a college colleague.
How do you even do that?
Yes, there was a college colleague, a girl who was very good at Wolfram Mathematica. Is that, uh, what architects have you using? Here.
Autocad.
Autocad, MATLAB, yes. And there's Wolfram Mathematica. It's pure for maths, and there were graphic plans. And there, she just built me a few schedules, which I added.
Oh, that's still a mood, right? Not that you don't paint a new one there.
Yeah. But it wasn't just a parabolic or something, it was a very thoughtful figure I'd put on my hands. She says, "Well, let's try something now." And then I got it in my dissertation. But it's a lot easier to do now. That's a few requests, and you're all set.
Very funny, yes, very nice story. Eldar, look, I think I should get up here today and say, there's actually a really, really, uh, I'd really like to, uh, have two big, two big questions. The first is to tell OpenAI and their vision, um, their vision of the future, to take it apart. But once again, it might be a separate, separate issue.
And the second story about this Google study of behavioral tendencies. But I still want to give the moment Meta let it go.
I was waiting.
Yeah, Meta's released a new model. Yeah, I'm gonna tell you that. Why would you want to tell me that? Someone's gonna say that doesn't mean you have to run her test. It doesn't mean that, uh, you have to say it's some shit or shit. Why would you say that? What Meta is a company that costs a lot of money, and it's a company with a big resource. Like, like, Amazon. I mean, these companies or Microsoft, they need to be alerted to what they're doing. Still remember, watch them leave. It's not, well, it's not just a resource of money, these companies are with people's resources inside the life of ordinary people. And Meta is like nobody, given Instagram, WhatsApp and Facebook, the world has been introduced into people's behavioural lives. They certainly lost the market very much. If we remember, they were on the front side with Lama. And when it all started, Lama was right around and they were everywhere. But something went wrong. While Zuckerberg was interested in investing in this, hiring, watching. By the way, is this a big investment story, what's it about? Still, more of the money will be on the market, more of them will be out of the company to the company. I mean, if you were working for some of these companies, you're lucky, right? Meta's making a model. But what's the model Meta doing? Meta is making a model and saying the next thing, so this Muse Spark is, and that C-S is what the new Muse and--
News and Spark models are the first model from this line. What's the main story? That this model, which cannot be downloaded, is that model you cannot use separately. I mean, they're still betting that this model that can be used in their systems inside. It's like Amazon, yes, when Amazon is introducing, it creates some of its models inside its own devices or inside Amazon itself. And, aaaa, of course, the big question is whether they can do the quality that Anthropic does or how it does not even do Anthropic, but rather Google and OpenAI. Why am I bringing them? Because the OpenAI announced that they had nine hundred million, in the last report this week, like, nine hundred million weekly users, yeah. And these users are not WhatsApp, they're not Facebook users, they're users who communicate directly with the model. That means that 900 million people are teaching the model weekly, teaching the model weekly. That's a different approach, yes. Yet, as Meta said, and Meta had said before that that they had a billion model users, they were still people who used it periodically, like Facebook or inside Instagram. Well, of course you can put some stuff in there and use that chip. If they do, they can't turn up some space tool inside to reinforce. So Meta's obviously out of the game, and, uh, setting up a system that is paid separately, and she doesn't want to lose, and maybe that concentration, Meta doesn't want to lose, I'm gonna assume, Instagram and WhatsApp, In particular, users, so that, God forbid, companies like OpenAI will have something of this kind, yes. Google is still in some kind of confrontation with them, taking into account his systems, and, uh, making that bet. Let's see what they can do. Oh, Elnar, what do you think is so professional about this?
That's very fun. First, I was very interested in--
I don't know how much Tanya cares. Tanya says, "I don't care about it at all." Yeah? How much of this.
Look, Tanya, take this as a prom job for the people that Zuckerberg hired six to nine months ago. Remember, we were discussing that he was just about a billion dollars in backwards to the man who was in charge of his development department here, this AI And he's there for a huge amount of money to someone on his own, someone else. And I was just recruiting people, spending a crazy amount of money. That was just about six to nine months ago, if I remember correctly. And finally, for the first time since then, Meta has released new models. So these people that were hired for something, and we see the first results of their work that they've published. This is the first part. We can track how much from team design to model. It's clear there were some work, but somehow. It's the first thing. The second pillar, which is interesting, they turned back from open source, too. I mean, if they all had their models before with some more or less free license, yeah, there, with the reservations that you and I have been talking about in a separate video, but somehow LLaMA, we can have a weight. I was downloading, I could have started, and I could have been dull. Not now. Now they're on the same side as the OpenAI, Google, Aicic, all the other big players. What we have is a model, it's gonna be in our hands. You can use it in our ecosystem. But you're not gonna make it. I want to remember the funny picture I saw this week a few times. I don't know, I can't find her, but the point is, there's a cat, like an animated, on a shore with a stitch, and he's got a vet with a waterbender where he's putting the fish he caught. He's got another cat and he's fishing from his bucket and he's putting it in his bucket, and he's got another one. And that's where he's from the second bucket to his place. This thing characterizes the picture, like, well, with model training, yeah. Suspension, there, Zuckerberg has lost a lot of money to teach the model. If they had made it open source accessible, they would have actually recovered the following cotics, which were based on these data, based on those results that had been obtained, well, some of their models would have done. And here, it's just that the company probably decided it was too expensive, that the glasses of a remark they'd get for that are not worth it. Well, that's how the model gets really pretty good. I don't think she should be taken into consideration while, uh, there's a GPT issueing new models.
It's impossible to draw attention, yes, because she's not even, she's not even on parole. So she's in access to partners, limited partners within the system, yes.
Yeah, it's a last week's fashion. They, they released a story with the baccalaureate. The only thing they've painted is their whole pole. Then Twitter literally came out in a few hours, where the colors were already displayed from the work. I'll drop the link, you can add that picture from Twitter. And the point is, the pattern on these baccalaureates is not the worst, it's not everywhere, it's not all, but it's somewhere on the level.
Yeah, it said she was almost the second, second, of the images. But I don't believe it, Elnar. I don't know, I don't know, something, I have a feeling I don't believe she's the second person in the image generation.
Well, limited access is, of course, just this slogan of this week. Because we have Anthropic our Glass Wing project, and now we have only limited access to this mythical model. That's the same thing with Muse Spark. OpenAI has already stated that they have a super-cruise cybersecurity model, and it's so good that access to it will only be for a certain number of companies, just because it's very dangerous.
I want you to put her on the market. Anyway, everyone's moving this way. I don't know if they talked or talked, but the OpenAI is coming right up on the spot for the anthropics., the fact that they're starting to look towards enterprise is that they're shutting down Sora to get Compute out of this side. Now, in fact, they're letting, uh, anthropics, their cybersecurity model. OpenAI comes out in a day and says we're on a very cool approach, but we'll show only a limited number of companies. Anyway, we're moving on. Uh, maybe there's a time when the new super-crunch models are like--
Yeah, they'll be available because they're conscripts. I just got a letter from OpenAI, so as you're among the first users of Pro, we want to give you a gift, but to get this gift, we'll give it to two thousand people. What do you think this is for? It's a pen. It's just that you're looking at it at all and you don't know what's going on. You spend $2400 a year on the subscription of the system. You do it there, from the first day, on parole, yeah. And, uh, even with the Pro version, I'm not saying the rest, yes.
Maybe there's some special pen, Sasha. I'm on the grape now. There might be some form of self-feeding.
I saw what-- yes, yes, of course. It's a secret pen from James Bond, Agent zero-zero seven. Here. They sent a picture of that pen. I filled out the application, clicked, she didn't work for Safari. I had to open the Chrome to fill out the application because they had no Safari. I filled it in Chrome. I'm told, thank you very much, we'll let you know when we're-- when we decide.
When the pen is ready
No, no, when we decide. I mean, I was just doing the post office at this point of time, and it came to me. I can usually do the postal at all. I don't know how long I'm not looking. I answered five minutes, I thought, 5,000 Pro users or, 2,000, I don't remember how many people they are. Sorry, I was wondering if they'd answered it for two minutes. But OpenAI, as always, you know, I think he disappeared forever with that pen, and that pen somewhere, somewhere, somewhere, somewhere in some other world. Thank you very much, everybody. I'll see you in a week.