Yeah. I mean, if you know what's going on a step by step, you find where he's wrong.
I was very surprised at that answer. I think it's going beyond the rule, too.
The U.S. is allowed to train artificial intelligence on Internet data or they'll lose China.
I'm, to be honest, ready for graduation, guys, incredibly impressed by the volume of different events that happen, and, in general, by the model to look for these events. It's like a stunning quality has improved in search. Maybe a week was so saturated that the world just overtaken all the information incredibly interesting. There's a scandalous start-up that replaces people before fighting again. Now, OpenAI has sued Sam Altman. OpenAI, we're gonna tell you now, let's just say in the last edition we talked about three models, we're gonna talk about it separately. Chinese DeepSeek is doing classified information. The data leaked there was Bloomberg. They're preparing a breakthrough new model. So, a lot of interesting things. Stanford released an annual report of four hundred and seventy pages, very detailed. And of course, they're impressed with some of the analysis to draw conclusions and tell them who's ahead, what's going on, and how.
But before we come and talk, I think the phenomenal story finally came true, my dream came true, my dream came true. You, I think.
Olegsha!
Someone's gonna say that to you. That's all bullshit. But before we get there, what is the infa? OpenAI sued Elon Musk back and so they were so stunned that Elon Mack was always trying to froze OpenAI, that lawyers demanded, OpenAI demanded a restraining order, a ban on further action. The mask, saying that the ongoing attacks of Ilon must stop. So, and, well, judging by the knowledge of the American judicial system, there's certainly a way to ban him, and he can't do anything at all. He's still attacking OpenAI. Since we, we have Elon Mac, Sam Altman, a very important figure in our productions. I want to say that, including the sound of it. By the way, Meta and Mark Zuckerberg are in a huge trial in the U.S. Senate. I wonder if it's not over there, but it's a very high-profile news. It's written everywhere. And there's a difference in information on this subject. Let's see what graduation we're gonna tell about. Maybe we'll get a special release. We've got special subjects on Wednesdays, and on this Wednesday. That was last Wednesday, until that was weeks ago. Next Wednesday, there'll be special releases. You'll have to watch. Let's do it for Meta. What's there to do with that stitch? There's some interesting information that Meta was trying to give some data to China to get in, get out to China. That's what we'll know later. Did it really happen or didn't happen? Very, yes, very interesting, very interesting and very serious. And that day, well, there's a little bit of Meta's stock failure, but now it's all gone. It's hard to show something, but this investigation is very serious, catastrophicly serious. Let's see what happens. Support the channel, please, put the video on the liquor and let us take more interesting class materials, especially when we are now launching new releases on Wednesday. And sign up if you want to get our graduations.
Is there anything else we're interested in going out tonight? We'll get Higgsfield to the Higgsfield video generator system, and we'll get a promocode to the maximum version of the video generation. One of the world's top systems, and you can do quite unique things. What's the interest of Higgsfield? That there's a huge number of predetermined pre-sought-out templates. And a very interesting thing. What to do in terms of the assignment, we'll tell the video and then select the audience by number of comments or by our own discretion. So you're gonna make the comments, we're gonna tell you.
Let's go to the three model that OpenAI has. And I think it's worth talking about and I should tell you. And it's accessible to a large number of people, including a simple basic signature. What do you say about three? Professional, personal, what do you say about three?
Well, I've been doing what I've been trying to figure out as a hobby, just as a hobby, trying to understand new rules on the ceilings on NHL salaries. I'll tell you a little bit. The NHL makes the hockey league the best in the world. And it's like there's a lot of money that can be paid in total. And there are hibernary rules that count. And the wages change from year to year. And as often as it happens, there's a player's license, but you need to re-establish the contract every three years, okay? A lot of people like that. And there are some more arbitrations and some processes. So how does it work? I mean, there's a lot of logic that really can't be understood. I understand, but that's a big influence over the results, right? So there's a lot of things like accounting. And, uh, and I'm-- I just did a lot of tests. It's a whole, whole Sunday, while there was a plane on the plane, and I was so full of power after the plane, and I realized Grok was getting a lot ahead of it. I mean, it's just a reasoning when you have some financial constraints, so, uh, well, it's kind of a lot of things that are going on, right?
Well, then, much, far ahead of the Grok of all the others.
Yeah, yeah, yeah.
For my part, speaking of the 3 when she left, I was wondering if I could try. And, uh, I took one of my job assignments, so I think, let me try with him. What do you want to say? He's very good at the top of the line. That's the concept of what it would have been like to do. But when you start to get into details, or whether hallucinations are, or if the mistakes are inside. Well, you can't just get the code out of it, put it in so it can work. And it's just, you know, like, it's a condition, and, uh, it's a mistake, it's back, it's over, and it's over and over. And they've done some of the hydrations. If you don't know what's going on inside, he's lying to you. So he finds some kind of solutions that, and, you even believe in them, if you just read what he says, but not knowing what's inside this task, well, that's, uh, that's the specific task I've been working on, a man can't do that, can he? I mean, if you know what's going on a step by step, you find where he's wrong. So you're looking for documentation, like you had to do right, doing it, and it's all starting to work. I mean, that's, uh, from the point of view of the specific code writing, the use of libraries or something, not yet a genius level, like Sam Altman says sometimes. But again, in defence, yes, that's not all bad. The top-level concept of the solution was very well presented. Modern approaches are being used. I really liked it. As a final realization in detail, it is still suffering. I don't know how fast they can decide, but, uh, some step forward. But hallucinations are still quite strong.
Yeah, I think I'm in agreement with you for 100 percent.
I honestly didn't get it, I did or didn't, because I found out about his exit yesterday when I read the news list for the graduation. Because I think they're still, and I even came by the back of the head to see what all these things are like, there's a few of them out there, as I realized, right? And these are the baccalaureate baccalaureates, there are ten of these graphs. I think, God, what am I supposed to use this for my purposes? I don't have a math problem, coding, do I? Oh, and there's someplace where there's less hallucinations here, and that's where the visual material is better treated, and they're all splitting up. Anyway, I ended up closing this article and I think it's okay to hear what the guys are gonna say in the next issue, because I'd usually have our graduations, because it really helps me out of my life, I'm sorry to use and not waste your time.
As for the baccalaureates, Tanya, there are hundreds of them now, and everyone wants to show the goods face. And you're just picking up the hundreds of benchmarks that you have a model or the one you could get to do, so you can have good results. And you're showing these exactly what we're talking about here.
These are the ones we've been round here, and here's the baccalaureate, we like it, and we've got a lot of points here. And you have a big choice, you'll always find out or you can get a little bit of a little bit of a lapse.
Yeah. Well, it's still a deep research, of course, the most appropriate option for me.
Look, I'm talking about the o3. Uh, I've had an interesting moment. I'm in the middle of a request at the o3. You, Sasha, just said you said about the hockey yesterday. You still see that player's better. I mean, I've seen it this week. I'm very glad we're about to take off on a week later. And there was time to test the system, yes, because I tested it in a vast number of different, uh, caseas. I certainly didn't test the software. That's what you, Ilnar, gave an example in terms of the quality of the code, yeah. But I've been testing this on a great number of simple, routine life requests. And the first thing that hit me was when I asked in the o3, it, uh, it was my old chat and, in part of my old chat, it took the information I had already had. So I had a dog request, and he said, "Look, here you are, waiting for a certain cure, you can use it right now." And I'm screaming and I realize that it's not in it. I said, "What information do you know about me?" He says, "I can't tell the information and my context window to show." I said, "Can you tell me, you know anything about me without technical details?" And he gave me a list there. He says, "I don't know everything, but I kept a piece of information. He gave up some information. And it's just that this week, it's been shown that many people, and, uh, stirring up that, uh, the model knows some more information about them or knows their name, yes. It's often called. Well, I'm telling everyone that what he knows your name is no problem because you bought the account and filled out your name, right? But it's very strange, yes, that people are surprised, but not the point.
If you move to o3, I'm--
They wonder more about someone calling you on your name.
It's working, yeah.
Yeah, yeah, yeah.
The character behind the screen.
It's like if--
I had a funny episode of the series, too, remember, we recorded a release, I had ChatGPT, and I asked him to adjust it to the client's letter. He says, "Well, of course you can't live without me." I was very surprised at that answer. I don't know how this is, I think it's going beyond the rule, too. If he just called me by name, I wouldn't have gotten so stressed.
There, well, as he is, well, as I understand, when you're talking to chat, it's just chatting that answers your questions. When he starts calling you by name, not when you ask him, and he's acting on his own, he's already on a different level like your friend, like some kind of lively, inspired. - I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry. And that, I understand that many people are scared, that--
Yeah, yeah, yeah.
You think? You're a car. She's gonna want you! And he calls himself.
It's very interesting, then we'll add.
Yeah, that there's a dialogue window about thanks and, please, there's tens of millions of dollars. But I want to do this again about the o3 thing, but I want to separate the system for me personally. It's obvious the system can stay. So the model is suspended, reflexes over its intermediate steps, and it's out of standard mode. I'm telling you, I had a visual on my tests, what? Which sometimes sys-- the model is in 30 seconds, and sometimes I have some answers today, asking questions. I had five or six minutes of thinking. I see her moving the websites, then stops, some text writes, then re-examines, then stops, texts, then re-examines, then re-selected, stops, texts. And it's a big deal. I mean, it's clear that this is being further analysed. And uniqueness, by the way, is the model in the pictures. I noticed that last week. I gave her a picture of the pharmacy I need to buy. I couldn't find it on the shelf. I threw a shelf, wrote, what kind of medicine do I have? She said, "There's a cure on this shelf." I don't see him. I said, "Can you-- find me, how do I find him?" And it started to share a picture of the fragments, start looking for fragments and show me the pieces specific and wrote, "This is a little bit lower on this shelf, a little better. And it looked different. That was to say, the system itself had given me a number of certain steps, and it was looking for me in those steps. And it was with pictures a lot of places. I mean, as long as I've been giving her pictures, some of my drawings. There's an interesting case in piano when a man wrote his hand, it directly increases the note that a man wrote on his hand and recognises that particular text.
I mean, it's the beginning of the task, so that's the way it is to share it, yeah. And, uh, in terms of what it's thinking, it's at least more serious. What am I gonna do? I've been using the read code. I started writing at the end, and you think more, look more, so study more. And then, by the way, I'm doing deep research, too. And then deep research goes looking for 30 minutes instead of five minutes. I mean, I need more data than that, I need more analysts. It's like deep research on deep research, right? Then you'll be more like him. And there's a very significant difference between the speed of the answer in the o3. There's a minus in it, we're gonna talk about it now, yeah. But I think that, uh, that's what one of the most important differences in this model now is the full integration of all the external instruments that the OpenAI has. All of them, it's like all of their agents-- agents, that architecture has been done, right? So, during one session, he has the opportunity to use various absolute tools from the production of pictures, searching on the Internet, launching reasoning, then thinking about something, then ID' all these. Images, files. Fails and all this. And the files are on. I used to load up, like-- well, first of all, I couldn't download the file in Reason. I want to see you. Reason could download a copy of the copy, for example, on pages 100. He is now reading easily, e-e-, word-over document for 300 pages and draws conclusions. I loaded a PDF on 470 pages yesterday, Stanford's report. He wrote, "I can't analyze the whole PDF, I'm not in a position to get, uh, on it, depending on the headline, to go through a certain accent." And I've started six searches on this story. He said, "Go on the different blocks of this document." I don't understand the whole document, he's gone or hasn't passed. But what was interesting was that he had in fact learned in his conclusions on the process of reflection that he would not be able to analyse the whole document. And, uh, that was a big problem before in the crypts. You're doing a document. People thought he was analysing this whole document, text, and he didn't analyze the whole thing. He was kicking a piece out of it. And now, according to at least the extended window, they've expanded a very context window and 300 pages, at least, so I have my materials, collected by some of the things I've created, which I wrote, there, with I'm taking a lot of notes out there. Not the point. I'm in these articles, these are my articles, and I need to often combine information, help me compile some information or, there, come up with a topic for a rill, there, or for a shorts, or do something. I'm not sure what my conclusions are, are they? Ah, he's more qualitative, accurate, clear, guaranteed, better. I mean, that's what it says about his extended context memory. He's got a long context window. Nobody knows what it is. Well, they say, like, 256,000 tokens, but they don't understand. One of the stories that is about speed and scale. There's-- there's a big problem, I think, and her many people don't understand the way they're used. The system clearly eats resources very serious, but it shows, yes, in the process of reflection, details. I mean, there's resources, resources are crazy.
And it's clearly a question there at its speed, to its scale and to use it in different companies, there in API. Well, Ilnar said hallucination. hallucinations are still present. I've personally studied a large number of reports, uh, a lot of examples on the Internet when the system has hallucinated or, for example, it has complicated your question. So you're asking her a very simple task, and she's paradoxically moving her mind-playing trick like she needs, I don't know, making things difficult. I had a case yesterday, I have, there, business in education, I'm in education, and I've been studying some practice, and I've had to go through, uh, school applications, primary schools, using artificial art. Intellect, there, in China. China has introduced artificial intelligence from the first September, e., from the first to the eleventh grade. That's a very interesting subject, by the way, isn't it? They're not much of a numerical, they're making thousands of separate, and they're directly responsible for thousands of teachers on this subject. Well, there's a whole thing, like China can, right? China is beating its strength. Skyll. Huh? Skyll, skate. Skyll, yeah. Well, look, they do very serious things in there because they plan, there's 5,000 teachers to teach, right now. They are undergoing 100 60-hour certification, and by 2027, the plan is to make 20,000 predicaments. And again, it's very important. They're planning. So they have a subject they want in five, ten years later, to just have a bunch of graduates in this field. If, for example, in Russia, this topic is not yet compulsory, it is not history, uh, in schools, where it must be used, it must be applied, but China is now the number of one country. to date, which is being launched
The day that starts, the volume does. Number one, actually. So, to the model back. Back to the model. I mean, she's been in the middle of a story, and I've had one story when I was looking for it, and she was trying to make it hard for me on a very simple task, to create something... What a fucking tough answer! I always said, "Let's make it easier." And she's always hard for me. I mean, on the one hand, this is the difficulty of three, I like having a wider, advanced thought, she's building some tables inside. On the other hand, it can be seen that some answers are needed, and that's a good-bye. But I, Elnar, we're just saying that you were writing there and putting out hallucinations. I mean, in my research and data verification, I had hallucinations no more than the usual model. That's what I was gonna say. And, uh, but I'm impressed with quality. Again. And deep research, launched into three, at least again, on my test, it was phenomenal. Well, that's definitely better. I really like the OpenAI in all their kittens, it's just that I don't know how much they're doing, but the fact that they're moving towards a single model of all their systems is, well, I think that's what it is. Yes, it's great. At least I have a test like this, I have a test like this. I've started three times, maybe, or maybe four hundred.
I'm a lot of people, and I've done a lot of different cases this week. And, well, to figure it out, about the volume. About the competition, the contest. So we'll give this video-generation system a promocode. Well, actually, Hicksfield, uh, divinely cool. So what's important? What's important, uh, what do you do? In the commentaries-- we actually do not give a promocode for any bullshit, but for the utility, I think it's very much for other people. We have a lot of requests from those watching, and you have to show us the most realistic boxes we use in normal life in terms of... requests. And let's write what the most interesting requests you or there are a series of requests you've used in your life in models you ask every day, which are not business-related, not professional. This is the most important development for itself in terms of, uh, understanding and use of artificial intelligence in general, and of models, it is the use of this in ordinary life. Simple, huh? Write what you asked. The most interesting. You also vote for other requests, and we'll choose, you know, on the lycées, on the discussion, and we'll pick the best and give it. Well, that little gift is the maximum version of the video-generation. Then write it down like it was on video.
We're gonna make an interesting issue, by the way, like I said. Yeah, we're gonna do the thematic episodes, we're gonna have them, and it's gonna be okay. Write that in the comments. I want to go to the next subject, which means it matters, and I think it's very correlated with the data that Stanford's report contains. Sam Althman made a warning that the US would either allow artificial intelligence to be taught on the Internet or they would lose China.
It is important, however, that there is a study now that forty-eight per cent of the most popular web-based sources for model learning have already completely restricted access, which increases the cost of quality content. I mean, on the one hand. One of the things that went through now, I don't know if you read it or not. OpenAI limits access to its API, limits and will, uh, make a very strong test in terms of not having people who simulate its model. I mean, OpenAI has increased access to its best models. The developers now need a corroboration with the State identity card. This, by the way, will certainly limit the large number of use of OpenAI, including in the Russian-speaking environment. Because they report that 704% of DeepSeek R1 responses are stylistly recognized as the OpenAI model. And they... Seventy-four, representing, interest. But I'll be right back on this subject. Yes, that is, forty-eight per cent of the topics are closed and clearly OpenAI is showing this story after China. Why? Because China has 70 per cent of all AI patents in the world, twenty-three per cent of world publications. The US is holding a top of the 100 most quoted, there, most of the most quoted jobs, right? But, you know, it's a big deal to note that 70 percent of all AI patents in China are very serious, right? And it's clear that private investment is more than 12 times in the US. Last year, private investment was 12 times higher than in China. With over 60 per cent of start-ups being built by private, private companies, right? And I think it has, like, what? It's a very important story. Also, there was a fun thing, too, that China was fast running away. Fifteen models released in the USA versus 10 models from, uh, China, right? And-- oh, fifteen, vice versa, 15 in China and 40 in the USA. And China keeps picking up a lot, getting a turn. That's an interesting number, by the way, about the number of models. Forty U.S. models. Forty. Forty. A lot of people don't get it. This is a recent comment from one person: "You're misleading about the technological market in America. In Russia, different verticals were much more technically automated than America, for example, banking. And I wrote a comment that I think you compared the banking system ten years ago. The end of the day, you see the technology inside. Let us discuss this topic from the perspective of Altman and his fears about modeling. Is there a problem? Because I think the OpenAI was always trapped in learning from other people's models, and she's still learning.
What do you think?
That's it. Data is the main resource. He's actually the only company that fights, takes a recurrent turn, and then they say we've already got employees. Yeah, remember, sometime, the OpenAI was discussed. Here. As for this DeepSeek story and the looks, well, here, uh, on the one hand, yes, on the other hand, DeepSeek is Robin Hood. He, uh, gathered everything that was done with the OpenAI and gave all the other people free. That's what a two-way feeling about them. On the one hand, yes, they're turning, on the other hand, so, like, a lot of people are using it now. About, uh, school bans. Here, Tan, remember, you and I are on the graduation, too, I don't know, he'll be out before or after, and they said that the same Lama, for example, said in the license agreement that you had no right to use, uh, well, As a way out of the way, the results of what generated Lama to teach her other models. But, uh, it's not-- it's not enough to put it in a license agreement to make it possible to use it in some way. Because jurisdiction is different, and it doesn't work.
I'm gonna have to get an authentication on my passport.
Well, it's weird that they didn't do that passport authentication before. While it is clear at the same time that this would limit their development. When you have hundreds of millions of users, you go out to markets at this point of time and say, "We're now limiting our passport authentication, right? And everyone has to register, somehow. Although you are aware of the size of the companies now to be deactivated, what kind of companies will not be able to authenticate. Plus, when you're auditing, you're gonna get... you'll know who you are in the search for data. I mean, I can tell you, I still don't understand, I can put in some sort of illegal online request. There are many people checking OpenAI and looking for, like, a bomb, right? There, or they're looking for something else. I'm not looking for such requests because I have a Belarusian logic. So if I put this request in the system, then I'll be there in five years and say, "This is where you were looking for a bomb, which means the system, yeah. Everything, you've got problems." I'll say, "I didn't look for it, so I modeled it." They say, "all, get ready, then close, yeah." And, uh, to be honest, I don't start asking for that particular reason. Although I understand how millions of people are making absolutely specific requests, but I kind of realize that if I have a history of my chat and everything, that is, even if I do temporary chat, there is. The possibility that these requests are stored somewhere. I mean, if you look at the data stored in social media if the data are stored in the phone negotiations, well, that's a lot of data stored. Well, it's gonna be weird if the data aren't stored in chat rooms, there, the data is stored in the queue, search. A lot. Well, I mean, the data are very stored, yes, in the world. It makes sense that the artificial-- and now, first of all, the speed of search, the quality of searching in the arts. I'll notice, my endless wish, OpenAI was a little bit of a real. They've been looking for a more normal chat room. He's looking for all the chat, going through them, looking for words in those chat rooms, connecting things. I mean, by the way, the search started working. I don't know if you noticed it or not. Looking for a nice job is not just a name, a chat, a chat, a chat.
Not only by the names of the chat room, he's looking inside. I've at least been finding these billions of chat rooms and everything. I mean, uh, that's a question. They also wrote, uh, not that the complaint, he criticized the strict regulation in Europe, and said that if Europe did, then they'd have to go through the adoption of the artificial act. Intellect, then, there will be enormous problems, and companies will stop investing in the region and Europe, and will invest in regions with less regulatory environment. I mean, there are regions where the regulations are smaller. But I don't see anything that in regions where the regulatory regime was scattered and that they were developing. I mean, I don't see anyone coming in any of these countries seriously. So, serious like this doesn't show up. Well, because China, let's open source China-- I wouldn't call it open source, I don't know.
I've never, I'll never have a tongue called Chinese open source, open source. Here.
We'll see this news about the video. Oh, about the video, my God, about the location on the picture. That's the very cool news from the pharmacy you were looking for medicine.
Look, this is a very interesting story. Not only am I in the pharmacy, I say, I was looking for a pharmacy, I was looking for, uh, photographing it just in my space, I had to do it on the flowers. I took a picture of this, I had a situation, I had hundreds of plants in the garden, I planted, and some of the pots, maybe 15 in the same zone, were moved to the sun. And I came, I forgot, because the season changes, I forgot that the heat can be very strong because we have a 12-hour sunday, right? I took a picture of the mass, and he took a really cool picture of me, and you can see it. I mean, if you're in O3, you'll see how he's looking at segments, and he's going as step. I mean, the distinctive feature is, there are examples on the Internet. There were journalists, for example, preparing an article to search for information about a certain person. I mean, he's going on a step, he finds some information, processs, then he starts up the next request, processs it, then he starts the next request, then he's processing it, Then he starts the next request, he's processing it. You might think it's a deep research, but it's kind of not a deep research, right? Because, like, deep research, there's no reasoning base. I understand there's no reasoning in deep research. But O3 with deep research is still reasoning. Here. So there's a sequence here. He also collects pictures and takes pictures. I don't know how unique technology is, I don't know. So Sasha might be answering how unique this technology is, but the fact that they visually display it, they show, they're looking. And there are a lot of analysts who wrote about it. Uh, and it looked really fun for me again, right? That he's kind of refreshing some segments, not just a picture of you. By the way, let's see that the first thing O3 put out is a case. They put on their social media on the day O3 was presented, they put the case in Instagram, they took pictures of one of their employees, I think grandparents or grandparents are writing to each other. And it was written, uh, ru-- well, hand, hand written. You know-- you know or not that a large number of Americans can't read the manuscript. So, a lot of people, well, I understand it's in the world, it's a lot of people around here that have countries where people don't learn, right? It's a learning thing, but there's a place where you didn't learn writing. And, accordingly, if a man has not read in writing, he will not read, he's hard to read. And it's very difficult there is this handwriting that is unreadable. So they just showed how the O3 model is very cool in the image. Super good. Again, I don't know Grok. So Sasha's talking about Grok says it's just that I'm very hard to make some conclusions because it's got to sit right there and do cross-tests. I'm always interested in tests that are not designed to be difficult in math or to have a hard programming or some science. I'm interested in things that can be used every day. Uh, do some kind of health case there, do some kind of thing about me, like I brought, a dog or a pharmacy, and so on. It's a crazy time now, isn't it?
The FDA approved two hundred twenty-three AI devices for medical applications, two hundred twenty-three AI devices already approved.
Tanya's talking about the news where, uh, ChatGPT 503 is a parole, and where is the object? I mean, you've been looking at it, and he's talking about California, there, uh, two houses from Capitol.
Gene, it worked a year and a half ago.
But I was struck by the case, which was the guy in the library, took a picture in the library, some kind of old-fashioned bed with three books. And for these three, my God, the three ChatGPT books have identified where he is, what library, what city and all. I mean, it was just, you know, shocking.
That's crazy, yeah. That, that, that's crazy, yeah. It's, uh, it's probably unique books that belonged to this library more, right?
It's like ChatGPT told you what pharmacy you're in.
Yeah, yeah, yeah. But he's actually that, and he almost said that this was on this shelf. I think he should already determine what kind of pharmacy Sasha was in. According to the their release.
No, I'm... No, he can't. No, that's impossible. He can't determine because there's different calculations. The only question I asked him. I said I was in the CVS pharmacy and how I could find this medicine in the CVS pharmacy. He was before, I took a shot of the wound, showed the wound. He told me what the medicine was. I came to the pharmacy, couldn't find him in the pharmacy. Asked me what CVS department I'd have to look for. He told me it was supposed to be there. I went to this department, I took him three pictures of this department, told him where it was.
He said, "This is the cure standing in this area, it is, that's his name." And then I said, "I don't see him," and it helped me find him. I mean, the case was like that, yes, but it's not like he's me, yeah, yeah.
I don't remember that, discussed, talked about smart glasses. There was a video of the case where a woman--
Yeah, blind.
Yeah, yeah, yeah. With visual problems. And she just walks. She's got, uh, a device saying, "So," saying, "I'm gonna go out to buy apples. She's up there, yeah, that's the apples of the class you need. Put it in there. We need a onion now, right? And she's doing it in the shop to go through the next shelf. I mean, generally, it's starting to work at the point of the sea. So it's great that these cabins are in.
Yeah, it's funny that you're right to say that this is so absurd, because there are still things that are a little more complicated, and that's-- and that's, of course, very useful. I mean, from a human life perspective, especially adult people, well, who's harder to find, read, study. I mean, this, uh, I'm not saying that there are many people who don't read the papers themselves, are you? Just taking them off, sending them, and they find you that information, yeah. You're not doing it anymore, you're processing a whole other amount of infus.
That's the whole thing that's been fun. I don't think that's the news I've added, but there's this Swiss company that makes carpenters and tablers' equipment, which is, in general, a tip, yes, I'll add if I find, a tip, a-e-e, for equipment. I mean, you're on the benzopyl, you're putting your phone on some device, you're on, you know, AR-Up, yeah, that's, uh, that's a little bit of a reality and , that design, that's the product you're making. I need to make it. And by doing some measurements, it shows right at your setup where and how you need to drink. And the accuracy is a millimetre, probable, you need to drill this hole right there, it shows where the tree is, you move the camera, you move, responsibly, you know, the point where you need to drill, it's on one. and the same place. And that's helping you, uh, with the sophisticated equipment, make a complicated detail, but with clues. It's like...
Yeah, that's great.
Still, I was tortured with Ikei instructions that it was hard to collect furniture, though it was like everybody else was right there.
And here you're more complicated, there, you can do with the tool, with the equipment.
Well, it looks like ChatGPT started doing CAD, uh, CAD drawings finally. It's still, um, initial level, but it's really helping to get a shot.
Look, I think this is a great homework. I need to get here for one car, I have a roof off, and I'm-- I'm off the roof completely for the summer, and it's hanging. A system that puts her in a garage garage garage garage to the ceiling. And I didn't know how to diagnose where the ceiling was in the box. I didn't know for the rest of the day, you could hang it or not. And yesterday, it was a whole series, a whole series of downs. But, uh, you know what was interesting? That o3 went to search the city's records and county of my area, as usual, the garages do the ceilings and what probes and the distances between the balls in the region to make their full. He wrote to me that if you don't want to go to the attic and check these balls, and I gave me that answer, I believed it myself. I have a device that diagnoses the balls.
Yeah, there's a device.
Yeah, but, uh, it's because of what, you know, I was still afraid. I'm not-- I'm-I'm not-- I'm not-- I'm-I'm-I'm-I'm-I'm not--I'm-I'm-I'm not--I'm-I'm not-I'm-I'm not-I'm-I'm not-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm not-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm-I'm not-I'm-I'm-I'm-I'm not-I'm-I'm-I'm not-I'm-I'm-I'm-I'm-I'm-I'm not-I'm not-I'm-I'm-I'm-I I was afraid, yes. I checked that device. It's definitely found this distance of 40 millimeters between the balls, well, or there--
Yeah, it's a code. But they could have been broken, Sasha, you know what the problem is?
Yeah, I see, I see. But re-examine it again. He double-checked me, gave me all the details.
I checked again, gave me all the details, then I asked what exactly it was, uh, I twisted it, and I told you I'd use it further, where I needed to. Much more than the instructions. I mean, that's what made me, uh, let you-- make the right choice to order this backup for this roof in the garage, right? Because it, well, a separate problem for me was, like, right-- first, how to take this roof off because it's heavy enough, yeah, it's hard to get one of them off, and so it's just one thing to do. I hope she doesn't fall for any other car, this roof, right? Or children. Here I am.
Who's
to judge?
Yeah, yeah. Who's to judge? I'll be there to clap her and watch her. And I'll check, of course. I'll check it out ten times. I want to see what we're talking about today, how many decisions are in the artificial intelligence, finish our meeting with the interesting words of head Shopify. He's Toby Luteke, which means he's the CEO of the e-commerce platform of Shopify. Such a huge e-commerce platform that doesn't know. He sent an internal manifest to the staff, which was written in word: "Finding to request additional staff or resources, the teams must show why they cannot achieve what they want with the staff. "the help of artificial intelligence." So first, prove that artificial intelligence-intelligent will not be able, and then let's hire. And now, of course, but, well, in normal life, people have this unnoticed, but in business, serious, clearly visible things. Microsoft cleans up a huge amount of management to be one manager, not more than ten programmers. In Intel, they hired a new CEO. This CEO came right away and said, "What a bunch of managers, because I need to, uh, I want to..." He said, "I want to direct the programmers directly to develop artificial intelligence." I'm really cutting this moment, guys. I've lived my whole life where everyone in the office told me that you should form managers who run all the others. And the Director-General should not be involved in the development. I've always been watching this as a programmer, with a laugh-- well, like, because I've always managed it, Elnar knows it, technically, went down there, discussed databases with them, and, uh, everyone said that You don't have to do this. Now the world is going to be in a state of what we're going to do first, uh, AI, put AI-soft on it and prove that you're not. And then we won't hire people. That's a very interesting subject.
Yeah, I just liked the very news. She added a little nostalgic to me. You're like you mentioned "thank you" and "please" when you're talking to a chat-bot.
That's really cool.
Yeah, yeah, yeah.
And you're actually writing him politely there, saying, "Please." He told you. You say, "Thank you very much for your answer," and so on. And then Altman asked, yes, that's how you said that the company was spending a lot of money on it. He says there's tens of millions of dollars. And then I remember, remember, the Esemeska were before, and they were there by number of symbols. And you start to cut the "thank you" for "sorry" writing, right? There's more stuff. And the most stunned writings aren't even a cyril, but they wrote Latin when you-- because it was twice as small, uh, symbols, if you weren't writing Russians, I mean, you write Russian words, but--
Sure.
On Latin alphabet. And if you're...
So far, many write that.
Well, someone doesn't have all the caviar. Yeah. Here. But in general, if history is extrapolated and trying to save resources, then that's why you can try to cut the text you write. Ideally write English because the Tokens will be less waste. Here. But it's just...
But if you have a paycheck, you can't do it.
I didn't do that, but after my ChatGPT talked to me like that, I think I'll start. I don't know if he'll answer me tomorrow.
Here, here, it's more of a story than a pay-as-you-you-you-you-you-you-you-you-go-thinking about resources. It's like green people say that there's a need to save resources, that so on, that three liters of water or something are going to generate one response. And that is how to reduce the consumption of this resource, reduce the carbon effect.
With courtesy.
By losing your courtesy.
Elnar, carbon effect. By the way, you're an interesting, interesting topic now. Now, if you're talking about carbo-cubon effect, I did yesterday, I'll just say a second, just an analysis of this, uh, influence, that's what you said right now, right? Sustainability, right? In short, it was written that GPT-4 training was estimated at 5,000,100,880,4 tonnes CO2. That's a ton or something, I don't know. And the average U.S. citizen spends 18 tons a year, that is, 18 against 5,000, one hundred and eighty-four. Here. And...
The effect is different, too, Sasha. From one average U.S.
resident--
Of course, of course. There's also, there's actually a countertrand. The energy efficiency of chips is growing and the inference costs of the tokens are falling, mitigating, that's how it compares this curve. I mean, we still see a decrease. Well, that's right. Things are getting much easier now. Although, according to model o3, they clearly made a high-cost stake, as always.
It's probably our next issue, but we'll talk about it. But in general, this price war is slow, because Gemini is cheaper than the API. Grok's still cheaper. With all Claude and OpenAI, it looks pretty expensive, but I think OpenAI is already the prices for new models, and it cuts the mini versions too. Yeah? And there's a way to play a little, how fast you need an answer. So, we're gonna talk about this next issue, too.
Yeah. And by the way, I'm not sure what Pro is, in Pro, in Pro, I have Pro's version, I don't know what Pro's point is. So I have deep research more or what? Because o3, o3 Pro is somehow unavailable. So I don't understand now, or do I have a O3 working differently or what? So how far they're doing, they've forgotten, by the way, Sasha, they've forgotten about the users, as they always do, right? So they sold them, so it's two hundred dollars. They've done a graduation and forgot. At least they would, I don't know, just by the way. We're spending more money on you.
You're better, aren't you?
For Pro, yes.
Yeah.
Huh? You didn't see? You don't know? You don't know, do you?
I think we've forgotten. Yeah. I think we've forgotten.
Yeah. Models. You, Elnar, by the way, said it funny because Grok is the same, he's now on, now or yesterday, he's got a memory on. So he's starting to keep his memory. And you're being remembered. I'm waiting for a very good memory and a good understanding of who I am and an understanding of how to communicate with me and what I have around, that I'm surrounded by space and so on. I'm really waiting for that. They've started some chatting. If they can have this whole volume in some volume, although of course it's crazy, it's gonna be very interesting, it's gonna be very interesting when they do. I'll see you next Wednesday, Sunday we'll have a big edition, and Wednesday we have thematic episodes. We're waiting for you Wednesday at our thematic issue. Keep the channels, don't start-- don't forget to sign, put the liquor.