Hello, everyone! Technological news, Silicon Valley sites. ToTheMoon Canal. A podcast. We're going out every Sunday. What cabs do you see hallucinations?
I'm so glad there's a story that deep research can be launched on different resources, yes.
It's a little more-- a complete denial of reality. Inside is information. Inside, they're gonna have to pay billions of ten. Oh, for something, because they're really getting really lazy, right?
There's a very interesting case here, too.
I think it's the most popular question for me that all designers will replace artificial intelligence. Some people are so sure. Well.
It's more like, uh, a payment for universality. Models are so universal that they can try to do anything.
Because they don't have a car like that on the cash print like Alibababa, do they?
Look, I want to start from the beginning. Now, everyone has a lot of peddets, and there are already a lot of systems. Oh, we're coming to the OpenAI pdette last night. Oh, I want everyone to write off the audience and tell us what systems you're using for a permanent job. I'm talking about a system where you-- who you ask a lot of questions. There's Grok, Gemini, ChatGPT, DeepSeek. What do you use at your place, what do you do every day? What's the basic system and the most important, what's the basic model? Here, uh, a week-- a couple weeks ago, there was a video of us telling us about a lot of cabs, um, real-- for our normal life. And they started asking questions, but is there a difference, for example, when the same question is asking about 3 and 4 and a half, for example, or four about. I'm curious, by the way, that my sister recently literally says that, for example, I, Sasha, use four and a half ChatGPT to work big texts, not o3, because o3 is worse, for example, working with a lot of texts. Big texts. So everyone's picking up these model stories for himself.
Yeah, but Carpathian wrote about it recently. You saw what Carpathian wrote, right?
Yeah, but it's kind of unusual.
What's unusual? Tell me, Sash, tell me, tell me.
I mean, yeah. I mean, it's really amazing that it's still very difficult to navigation. I mean, how do people know that GPT-4o is like a major jusquei, o3 for something important to business. They're trying to explain that GPT-4.5 is for creative rallies, that's what they said once or twice, Sam Althman said. Well, it's like their sales-marketing, yeah, it's so special. And that's what's a 4 mini, o4 mini high, o1 Pro Mode, GPT-4.1 mini is all very difficult to understand. And yes, so, as GPT-4.1 does not have a strong marketing for the Wype-coding, either. Oh, yeah, and how deep research is a little hard to use, huh? I mean, I think people, they want to just turn on o3, and like O3 or versus deep research, I think people don't want to choose, right? I mean, I think that a normal person can remember that there are four pro-- four variations of the product, and it needs to be understood that each one of them is used. It's not like that, unfortunately.
Tanya, did you buy a version of that?
You ask me more every day.
No.
I'm just curious. Polina's using the version, and I'm walking her, so I got a fight yesterday. So, I bit my son, my son's got a stew or a bee, and it bite my hand, and I swallowed my hand, so it's three times. I say, ask the model, only O3 ask what the problem is. She says, I asked. I say, o3? Yes, o3. I'm taking her phone, opening her chat, I see she asked 4o. I'm not asking anything at all, because for me, Sasha, you just said Carpathian said four for simple questions. But if I'm used to asking questions in o3, then fouro is such a mourning for me. I just, not much, he's hallucinating periodically. Unknown, well, how does the data not appear to be the way it is. Yeah, o3 needs to wait sometimes. You start some questions, he's really going, looking, analysing, but at least he's giving super-crunch, very detailed information and supplements you with different paints that you're absolutely telling you. You don't even expect it. Yesterday was an interesting case. I'm told there's a company that was sanctioned, and I-- he showed me a leaf, a picture, I went into the o3, added a name. O3 minutes, ten minutes, looking for something, base, everything, everything. Then came and said, "There were sanctions, and there were none. And the news here says she's in the sanctions, but she's actually on the lists, she's not on the bases, and this information is unconfirmed. And that's very interesting, isn't it? I mean, you're looking at the system a little bit differently. So I'm kind of hard to get to four from one. Sasha, you're using four?
No, I'm not sure, actually. Well, I mean, to rewrite, maybe some of my emails, but, damn it, it's almost no, really. Because, uh, I need a way to make the model use some of the previous context. I mean, I always ask for something to rewrite, I am filling the context that is connected, organized by projects. And in my opinion, he still-- very much hallucinations, even in o3. And-- like I'm four-- I'm not really using it. I mean, I don't see the point why? Aaaa understands that, as all these models are going to greatly enhance the bar, well, as the quality of communication in writing, which, of course, well, is difficult for me, for example. And I, of course, well, I can only rely on O3. In this regard, a, O3 has a wider range of searches, let's say. And where she looks from Grok and Claude, she can go as deep as she can.
So she has the advantage and the breadth, and she can go in the face, but the level of hallucinations is still unacceptably high.
And you hallucinations in what boxes, Sas, what can you see hallucinations?
Well, just when you do business report, business-summary, you know, you should minimise. Well, I see. So, there's like every statement has weight, and they, and they can be totally inaccurate. And, uh, and as good as he can, he can respect the logic and create some kind of narration, some history. But what he likes to think about in the way, and it's not very and as much as deep research he's been using, it's very often to, like, ask the questions about expanding some kind of search. I mean, it's as deep as it goes deeper, deeper and wider. But I'd like to have a regime of O3 where it can be fast, like where it always asks a few questions, maybe questions, and it's the one that determines when questions need to be asked when they don't. And it can be the moments he's not sure about, because then he's gonna give him fedbek. You've pointed it wrong, you've pointed it wrong. Sign here. It's just uncomfortable, isn't it? It's just uncomfortable. I mean, if, like, O3 was wrong about something, there's no way that I can just use this to separate what I want, I'd like to make out. You do, he writes you up there, like, copies, regenerate, inspect, for example, inspect, means he's gonna recheck himself. Well, that's the iterative process, it's just not there. And like one research-
Huh?
Yeah. I hope, therefore, that, returning to the original topic, I hope that again, every time, I say it, but that OpenAI will start treating GPT chat as a product, so that it can be understood what model it is for. What yukeys, not more than four models. And that O3 and deep research would be something like a separate product, yes, which is available, acceptable, for the only thing about the lakes,
yes.
Well, that's interesting right now, news. Sasha, I think you're gonna talk about her, too, about the new news that OpenAI has. They're there, and it turns out the MCP servers were allowed to connect. And in general, that's what you're saying about making O3 interrogate you with some unfamiliarity.
You have some unfamiliar points. If OpenAI does not do it by itself, I'm sure there will be some smart-ass who will just make MCP a server that will force the system to re-examine certain refinements. And it doesn't look very complicated. If you remember, Sasha, you said that the anthropics had released an article about thinking, right? And we were just thinking that this was about thinking, and that was the article that was next. They added a tool to think, and it was empty. And thus, by giving models this tool. So she can say, at some point, okay, I use the tool to think. But this instrument doesn't do anything at all. He's just putting his mind on the pause and the next irration. And this thing, she added a lot of points on the model bench that can't talk. And that kind of tool now can be attached not only to the clown, but also to the GPT chat. I'm sure that a tool like this is to request missing information from the user or to corroborate what you want to do with the user now before you move on. It'll be very quick. I mean, for a few days, you can even do it if you're gonna have a big demand.
The question is whether OpenAI will launch this in the interface. Still, I don't want to connect the backup service, I don't want to use the different softs. Right, like Sasha said, I don't want to write a first-hand explanation of what, for example, every time I want you to do me a search, for example, for a certain author, and not add anything from me, just direct quotes, No change. Or, there, the business report analyzed, or gave me some data, and didn't add anything from it, because it was. You use one terminology, he starts this terminology, other things to use. I was just writing him about the cash flow. I'm writing him a cash flow, and he started me an operating cash flow, the residual. And I started to use the terms with very professional finances. And I understand that the company's base report, well, is not the terms used in this company, they're not working.
I mean, he can't just be on it. That's what memory is, of course, super cool. They're in a free version, so everyone knows, they're free version of the version that came out yesterday, for the GPT chat, they've freed up memory. It means that GPT chat quietly starts to analyze other chat rooms. I don't like the OpenAI that he only keeps the data for thirty days. And basically, I had chat rooms right now that I wanted to go back to, which I was doing an analysis there, he wasn't looking for them.
And the search is not working very well. But it's nice that the O3 Pro version came out yesterday, she's not-- I haven't got an interface yet, but I got something in the interface that you, Elnar, now told the outside service connection. And I hooked up tonight, so Google Drive to OpenAI via their Wednesday and launched deep research on Google Drive. And you've got to analyze the video files that he's back to me in 15 minutes and tell me the video can't be analyzed. And I didn't find a transcrib, I didn't find a transcribation. I thought, "So, you know, that's it. Well, why are you offering this service? Because if it's just files, so I'm just gonna throw you some files, you still have a limited number of files. I mean, it's very. If we see their social, they put the boxes in Instagram. I think they put the case in. I don't know who's for a de-de-deck, like, make a stylish, right? There's, like, GPT chatting has learned to make a stylish. But as if there are more serious cases you can involve a large number of people. And again, this is the food story. Now, they seem to have a very cool team, you know, from a different perspective. And hopefully, these food solutions will be coming to the fall, as Sasha has given an example of the drawing or copying of a piece of text, to indicate incorrectly or to reset this piece of text as pictures, it's regenerated. Otherwise, it's hard to use. That's very hard. And Tanya, you need to buy Pro's version. You'll feel a huge difference, you'll feel a huge difference.
Well, by the way, the pictures already have this function when you can single out something and ask for it to regenerate, but it's awfully working. He's not listening. Even if you tell him not to change the architecture, he's still changing it. I mean, let's see how it's gonna be.
So, Sas, the example you said that first, I'm really glad that there's a story that deep research can be launched on different resources, right? And in particular, Google Drive can be chosen, and then he'll be carrying deep research.
Yeah, it's deep research. It's where he's the one who's got a connection, Elnar, when you're deep research, there's a deposit.
Yeah, I saw it. Here. And, accordingly, this is the case that said there's a video or audio, and he can't recognize it. Besides, you can deep research like this, you can start with different resources. That's what I started talking about. Add some more utilitators. Yeah, it's like a constant iPhone versus Android dispute. Yeah, you take iPhone, you're all set. Android, you need to get it all back, make it work. But now, as a whole, I understand that if MCP can be connected, some service can be added, for example, on the basis of Whisper. Yeah, this is the audio transcribation system he's gonna throw down the files he found, where there's audio, and, accordingly, on his way out, he's gonna get a text document and then put it in his analysis. deep research. And in general, a comb with a bunch of pens that can be packed.
For example, for her, the real living thing is a real case, it's a dictophone audio, not talking to anyone. First of all, he's gluing. ChatGPT often glows. You say he's forgotten-- I happen to be, like, a minute, telling him the assignment, and then he's a hop! And erase everything. And you sit there and you think--
I wonder if he's still giving you a whole new set...
Yes! Or yes, it's a continuation. His favorite, favorite phrase, I have a continuation. You tell him 30 seconds, and he writes you a continuation, you follow, you follow. It's simple-- or, like, just videos would be fun. You took a little view there, and you threw a view. These are real live cases, right? He's still working on a video recognition to make this function. It'll just, uh, it's a super-involvement. And again, we used to say they were going to AGI, and so they don't start a lot of simple things, don't worry about products, about users. But we'll notice the rhetoric that the last six months have something about AGI, well, as everyone says neutrally.
Now, there's a race for a mass of services that'll get the fist out of the groin. But the question is, what kind of chips is that? So there's no chips, like Google let AI fashion. Well, it's a very strange job, very strange work. I bought access on purpose yesterday, I went to Veo 3 to try to generate proms, shorter to make videos. It's shit. I don't know how people make such beautiful videos. It's gotta be a good way to do it, to be able to write super prom. But I'm like the usual user, I'm gonna have to do something, you know, there, the retrievals or the boxes. I understand that I'm not in the system for me yet. Well, I can't. Well, I just didn't work out. hallucinations. I want to talk about this very much.
So, a week ago, chapter Anthropic said, "You guys are all talking about hallucination? But people actually hallucinate more than LLM models." And here, so I got a call with my mom the next day. I'm saying, "Mama, I'm in GPT chat yesterday, and he's always asking me to buy different kinds of plants and fertilizers for plants in the garden. I don't have them. I came in, took a picture of my garden, a bunch of photos. Said, "Tell me what to buy." All right, all right. He gave me a list, ten or 15 positions. I added it to the basket, took a picture of the basket. I said, "all." He says, "Yeah, all of it." I bought it all. The next day, I came. I've done a lot of things, so I've been working in the garden. I come and tell her she says, "Well, actually, they tell me that these systems, they often cheat." I say, "Mam, you look like bloggers from some country. They came in free and don't understand how to use it." And we keep talking to him. You're right now, you're-- that's really what I had with her. So I keep talking to her. And she says, "I have plants, Sasha, I've been sitting here three years ago, oh, ten years ago, and they're dying. I did something to them, waiting for them to get better." I said, "Mam, take a picture of me, come me pictures and write to me what you put fertilizers there." She says, "Oh, good." He sends me pictures and writes me the names of those fertilizers that he added. I'm sending a model to O3, saying, "Izuchi, study for Belarus." I'm telling him it's for my mom, it's not me. But not to think I have a samash. He's using history. It's for my mom. So you'll analyze it, look it all, give me the result. He writes, so your samashita has a disease like that. Doo-doo, you have such a problem, you have something, here. I can see it here. I've analyzed everything, so I've increased the leaves, I've looked. Do it, it's, in a week, it's, in two weeks, it's, in three. Buy some, like, fertilizers on such sites in Belarus. That's what gave me. And then he says, "All your mom did is not working. She was wrong." What's this funny story? My mom's a professional gardener, yes, she's got about 300 or so many different kinds of plants. But she was wrong.
She's gonna write me a drink.
That's all. But she was wrong. She told me at the beginning of the meeting that systems were being grazed and hallucinated, not people, yes, in terms of medicine and fertilizer. And that balance, I think we're in some zone now and where I think it's important for people to be adequately sensitive to the notion of hallucination. Well, I mean, he told you he was wrong, well, he over-checked. Just like a man needs to re-check, yeah, that he told you he was right or wrong.
Well, that's what you can do to Kant, to criticize the pure mind that the truth doesn't exist. So, yeah, I guess there's your thoughts.
Yeah. But I, uh, hopefully, we'll be able to tell you how O3 Pro works next. I think it's time for a bunch of people to get out of some different apdets. Here, I was waiting for a user, uh, this version. And we, uh, we, uh-- we started again, and I asked you to write what versions you're using. Sign it up, perhaps we need to do some sort of separate edition, which still gives examples and shows the best Grok is in the better Gemini than DeepSeek and the better ChatGPT. There's still a big difference for people, what they can do or not. We had a case, by the way. Uh, one, uh, man in the commentaries said that I was Gemini asking me to help with, uh, health episodes. He refused me, and ChatGPT did it. And by the way, I think ChatGPT OpenAI untouched their hands because they're looking for people, uh, offering solutions. At least O3 Pro is offering medical decisions of any kind, health. And that's great! Well, you can get yourself a little extra clue, I'm not saying a solution, no way, right? There are doctors there, there are many details, but as an additional analysis, as an extra clue, you can get a very cool, very big one.
So we're gonna test. We're waiting for O3 Pro. Oh, well, yesterday's news is that Apple opens access to his AI models for the outside developers. I'm just wondering what the Apple AI models are?
Yes, Sashin, Sasha must always be here with me. What are they...
Well, yeah, it's kind of something, yeah, it's some kind of...
Yeah.
It's like some sort of, you know, process.
Death of the pyrometer number.
Uh, it's kind of, uh, stupidity. Well, that's a little more than-- a complete denial of reality. So those people who, like the code, have integrated Anthropic, well, they're heroes. Yeah. I mean, it's kind of unrealistic, but it's like some kind of f-fike, right, all.
Well, your bets, look, in a month, there's gonna be, uh, Apple performance. If Apple says that they'll do artificial intelligence by next March, then no one's going to do something, well, just frustration of the fullest one, right? In terms of artificial intelligence. Although I'm not falling in the Apple stock, and they're moving, and they're moving.
They can give the same, there, phones, do whatever they want. I used Apple, like, personally, and I used it. Oh, the question is, it's about artificial intelligence. What expectations, Sasha, are there any discussions among the technical community in the Valley? Any expectations from Apple in terms of artificial intelligence or all of them? Apple, Apple will do something, we'll work.
Inside is information, insider information that they're not going to make an exquisition. And it's kind of like a good and bad thing, right? I mean, they don't make the exquis now, which means they're gonna have to pay billions of ten, and, uh, for something, because they're really getting really lagging behind, right?
Well, the exquis are saying they don't buy any company that's done or invade them now.
Yeah, yeah.
Yeah, whatever...
So we're at a stage now that there are some interesting products on the market that start to differentiate from one another and find a sustainable business model, right? That's where we are now. And as much as Apple is trying to collect as some, some specific small teams of people who couldn't make the product cheap, but they don't collect the products. Next. So they're gonna have to collect business and buy business is expensive. They're gonna have to pay ten yards for business already, right? I'm gonna be, you know, by business, I mean, it's there, it's, like, positive, and, uh, a contribution to margin, revenue over 100 million. That's what I call a business business, right? And they'll just have to pay a lot of money in the future when they're doing the exquisites, of course.
But they'll let something out of their own or us--
No.
Or is there nothing going to happen?
But I think Apple is really, really behind us.
Very strange. Something happened to the technical team. It's about the detection of your models. There's a subject now that Meta said that ninety percent of the content would be analysed by artificial intelligence. All, like, they're including this ware story. And here, from the Google perspective, in his services, it's also replaced, it's all on models, everything on the phone, there, spam, and everything. Or is that so far an abstract statement? Well, Google, Google, the things that are now showing, they're getting up, they add a lot of stuff to-- to work the device, work on the device, some kind of thing without the Internet, for example.
Yeah, yeah, yeah.
While it's, well, as long as I think, it's completely different, isn't it? I mean, I know how Gemini is, these are all local models, it's just two completely different business young people. So while Google is acting on the principle that we can have a separate team there, some experimental, single, big models, separate mouses, they're models. - You know, you know, interact, like, a little poppy. And how centralized the beam is gone, huh? I think, well, from the point of view of Google, they're trying to be in all categories of hesitation-- well, to be on a peak, yeah, in every category, to explore and look for their business, yeah, I think. I mean, I think Google is in this category. I mean, I think while Google is in everything-- that's what all the categories except search, conducts exploration, and in search of, yes, of course, they have to catch up a little, there, Perplexity and the others. Well, I think for a large number of people, like AI mode, and, uh, it's kind of a post--- well, it's kind of a
s-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-
What about history, if you look, you said that Apple didn't make a deal. As for people, Apple doesn't appear in the speech about hiring someone expensive. Now, for the past two weeks, the Internet has been playing with the OpenAI, the Anthropic pays people $10 million, because of the fact that a couple of people wanted to leave, and they've been given $2 million in shares. Right away. So Sam Althman is clearly focused on engineers who are ten thousand engineers, yes, who are ten X, which are 10 times better than the others. This differentiation that goes around the world, from a human perspective, in the world, will it continue, Sasha? And she's gonna be a big deal about engineers or not? How do you feel? How do you feel inside the market now?
I think that, I think, for business, I think that for those companies that built a business that found this magic of both the reserver and the money, it's gonna have to pay a lot, a lot, Very, very much.
So we're talking about billions closer to ten.
There's a very interesting case like this on the news, too. Uh, OpenAI announced a deal they're buying Windsurf. Well, it's for the web-coding platform. And the news is that Anthropic has shut down Windsurf's access directly to his Claude 4 model, the last one most people really want to see. There's a part of the people out of Windsurf in Cursor because Cursor Claude 4 is available and not here. I mean, uh, the story of buying some kind of business that was kind of undependent on a specific ventor, that leads to other vendores starting to say that, like, well, you and I are not gonna play anymore, Work with your models, and here's our models for other companies.
Look, I think they're changing their strategy and policy, their limitations from a month to month. No?
It could be.
I'll be wondering if we're here, I think we need a video on this subject, too. About what professions, uh, new ones, and what professions will disappear. Yeah, make a little paper, a planning analysis for people in the horizon at least three to five years. Because, of course, from a planning point of view at twenty years, I think it's getting so little hard.
And write off, please, in the commentaries, what do you think the professions will be guaranteed and what professions will be guaranteed to disappear into, in, like, a five-year-old? About what's going to be gone, I've been talking about this... interesting case. A few years ago, uh, we all realized that someone has kids, that our kids will be in adulthood.
There'll be, uh, a huge number of pictures and videos in the adulthood. It's just a crazy number, you know, just, just a crazy, crazy number of pictures and videos, at least my kids have been taking off from a certain year, and it's probably 2,000, like, 13, uh, it's been taken off. I'm just crazy. This is my big daughter, who was born in two thousand tenths, and there's a hole like that, a break. So you can't see her little. All the others, you see it as simple, just crazy. I thought about the following. They grow up in 20 years, and we-- and parents bring them in 20 years, show videos. And when parents show videos, how do you know, it's a real video or a faeque? And how do you know the baby was acting like this or acting differently? Your parent can show the baby, say, "You're all crying there, or whatever you did or, on the contrary, you've had achievements." Or, you know, the kids are making videos themselves, and they're gonna show some other stuff over there. Well, that's a substitution, so it's an absolute substitution. And how many children of future generations who are now, ten years old, how will they even treat it, how will they trust it? I'm not gonna have a story that they're using, uh-oh, to analyze, there, a lot of information and dates, and build some kind of image. So, uh, by going into this kind of reasoning, I realized that the forecast now, uh, that the things that are going to be there in ten years of occupation are getting so staggered. Everything will depend on the political, economic decisions that are made there, how quickly they will be implemented, which will be allowed or prohibited, and so on. It's very, very different. What power will be available, how quickly it all learned what it would do, expensive or cheap, yes. Here. Because it's like two years, you and I are running a canal together for a little over a year, yeah, a year and a month, aaa in two years, well, just some huge changes have happened. Though we thought two years ago, everyone said, "This is a change on the threshold." Everyone said a leather, and it doesn't smell, it's a current one, but it's a recurrent one, well, it's a phenomenal one. I'm waiting for a robot home, waiting for a robot home.
I think that's the most popular question for me that all designers will replace artificial intelligence.
Some people are so sure.
What will designers replace? Oh, people are sure.
Yeah.
Here, and that's what they're gonna do-- and they're gonna write it off. That's a very big question, by the way. I mean, the directors general will replace, you know, the salesmen, maybe they'll replace the designers, maybe the cleaners, right? Okay, uh-
Yeah, that's what everyone says. Well, it's a profession that often comes when people come after they've worked as a lawyer, economist, then on the list, and they've already wanted something, and they're supposed to think it's a profession like that. More for the shower. Here, and they, well, I'm 40, 50, 60, 80, I'm not gonna change my profession because it's gonna be a replacement for artificial intelligence anyway. That's the kind of logic of reasoning. But I don't think I know, I may be naive, but I believe that, uh, designers won't replace, but some technical parts are totally accurate and fast and soon.
It's an interesting conversation. I have a comrade here to visit me from Europe, and he says yesterday, "so I bought and I'm gonna buy buildings." Says, "Actual intelligent intelligent," and we're sitting in a restaurant in San Francisco, "and a coffee, we need a drink, we'll go to the coffee shop." But you're sitting, you think, you know, whatever the fact is. The coffee is there, there's a bunch of cars. We don't know how this works, do we? Where especially when you're in San Francisco, you see these driving indefinitely cars, there, no--
Or at San Francisco Airport, there's a robot who makes coffee. I've seen it before.
Yeah. And that's why, uh, maybe another Wyb will be some sort of thing. Maybe people are gonna go down the street, they're gonna get coffee. Maybe not-- that's what this is gonna be like, right? And therefore, I am not putting forward a long time, but there are some short-term-- some things that may happen in some locations. Again, yes, when we recall sometimes, if there are two thousand kilometres to go, even a thousand kilometres to the center of the States or to come to Moscow and leave a thousand kilometres from five hundred from Moscow.
Too bad.
Two hundred, yes. You're sitting down at this point and you think, "Well, I guess not." Here. Well, there, in Belarus, you can imagine a country small, okay, that's where they've done it fast. And here or there, take South America or Africa, right? Well, I guess, like, what do you-- what? At the same time, I guess if you invented a system that, in terms of, for example, doctors, health is very quickly degraded, well, very rapidly spreading, copied, established, very well done. I'm just saying. And a virtual doctor, for example, will become very easily accessible, like whatsApp or as Telegram in terms of installation, the installation will be very fast everywhere. And just like the growth of mobile devices, mobile applications in Africa, huh? And there's a growth of mobile devices, he's-- he's, he's been moving quite differently. I'm very interested. So, write this one out of order, and we'll try, uh, to take off, uh, and we'll find time and make, uh, the promised skate, yes, the promised skate on this subject. Illnar, we'll have a skate on Wednesday. Tell me what the rock will be Wednesday. So we have additional skates in some environments. Tell me, please, what's the Wednesday for the roller.
I've been a little sober about, uh, MCP, how to make a chat-bot so many-armed assistant a-l-agent. Uh, I've been trying to get the technical side off because we don't have the technically good stuff. Well, I tried to show you what to do, right? I just wrote a few MCP services, and I put them on Cloud chat-bot. Well, look at the video, and what can you make Cloud do with the files on your computer? I think it's gonna be fun.
I was at least interested in doing that.
Tell me-- yeah, that's great. Wednesday, Wednesday, we're out, watch it all, but, uh, depending on time, on your location. I have a question. Please tell me, there are different things that the model doesn't seem to decide. It's gradually being connected in parallel to some of the modules written. Like, there's a simple skills model can't do. The model can't read the clock. Yeah? The usual watch can't read, it doesn't understand, it doesn't. She's making mistakes.
You're round with the cypherblat?
Yeah, yeah, yeah, yeah, yeah. The reason for neurosets is that they have no built-in abstract understanding of time and space. Yeah? And these tasks require a combination of visual recognition for hours, an understanding of the arrows, logical conclusion. Yeah? And there, for a calendar, for example, a day-to-day calculation. Here's the stuff. They're technically, on the other hand, everything, absolutely all these boxes will be resolved. It's a little bit of a story. Yeah? Well, I'll have to put on some module that's gonna be separate. I mean, here. It's just that there's a lot of reasoning. It's, I think, towards hallucinations. I mean, those things that model, you know, or what the model is making it hard, or what it's gonna do, but it's just gonna be connected to the program code. It's just, I think these complaints are when people say that there's no model, I don't know, a certain word to sort out, or a certain, mathematical, math, some kind of thing, there, it's a unique one. Yeah? It's, uh, yeah, there's restrictions, but she's gonna be trained to do it. She can do things that people never dreamed. Yeah?
Here, Sas, this is not a limitation, it's more like a payment for universality. Models are so universal that they can try to do anything. And not all of this was premeditated. And it just happens when you have a totally universal instrument, they tried to put it in one place, tried to put it in another place. Turns out it's not working somewhere. But there are two options. Either they just haven't had time to do it, they're gonna have to do the instrument that's gonna solve this question, or this case, well, nobody needs it. I mean, it's in there, I don't know, a third of all users.
And there are more pressing tasks than that. I mean, it's definitely the decision.
Yeah. So all the people who... Is this just a comment that you want to make? What if you're meeting some hallucination or where the philosopher model tells you everything and gives you a false answer about how long it's been on the clock picture, or you're meeting, you know, some kind of troublesome thing, I'm sorry, I'm sorry, but I'm sorry, but I'm gonna get it done. It's like I did Veo3 yesterday, and he did videos, and he wrote me an incorrect text. Well, I didn't write, I wrote an incorrect text. He can't do it today. We know that the ChatGPT images produced a good way of generating text. I don't know, maybe there's a lot of systems now, maybe. Everything's gonna be on the way. Yeah? Generate the text. That is, that is, that is, 100 per cent of the matter will be resolved. It's just time. Yeah? It's just time. And the question is, what do you use the system again? And if you have the opportunity, I'll be recommending that people too, you have the possibility of buying some kind of system cheap, there, at least $25, let's have twenty, let's say 25, please. Try a month to use a feeble version. Yeah? Spend that money, investigate, put it in, you'll get, you know, a little bit of a different effect.
Still, new models are feeble, they're ahead of, and they're very much ahead of a lot of free designs. What about Deep-- what about DeepSeek? DeepSeek is right there. They said again they were trained at 30-- 70% of the data again, Google or so. Yeah? There's new stuff going on. DeepSeek he's--
Look, yeah. I mean, DeepSeek. Yeah? I mean, we'll be seeing each other, and, uh, we'll see how much it's like--
We'll see how his parts are going to be used, right? So there'll be some countries where DeepSeek dominates. But I think they'll be hard enough, like everyone in the market. So, like these open source models, they don't have a market like that, do they? So they, that's open source model, that's good as a advantage for application layer and for inference engines, right? Because any inference engine wants to do, right? She wants to aggregate a lot of requests, to keep them, to put them on some model that's fast, which the comunity is accelerating and all. And then reassuring this model and like ninety percent of the most popular requests to run through a distilled model, huh? It's like they want to do all inference engine and also application layer of start-up, right?
But how would DeepSeek be very difficult to compete because they don't have a car like that on the cash print like Alibaba, do they? I mean, as a bet, Aliba is helping, as Aliba really allows, as paradoxically, it sounds, but it's for the development of capitalism, the starttaps Aliba plays a big game. a role, like that, right here. What, let's just say model developers, model providers closed source, like Anthropic, OpenAI and Google. What's a little amazing, huh? So, you mean, OpenAI, Google and Anthropic, like a good way to start something and subsidize, like, like investor money to send it, right? But when it comes to making it optimization over there by the most popular skies, it needs an open source of the model.
And the fact that Aliba is giving American launchers the opportunity is that it is amazing that we live in a world like two, when there are two countries that are in a clear competition, and actually these are names, like, The creation of generative AI businesses depends on Chinese open source of models.
One second quick question. Will there be a new operator this year, a very large and the same known name as DeepSeek? Or is that impossible?
No.
I have the MENA region, the Emirates, Saudis in my head. Well, they're in the middle of infrastructure and stuff, but they're gonna be very tight with the Open-AI. I don't think there's a pattern of my own, a big model that's gonna come out of competition. Other regions, I don't know.
Here's the race. Here's the race for artificial intelligence. Come on Wednesday to the skate Elnar told me. We're waiting for you next Sunday. I'll see you in a week. Bye, everybody.