Hello, everyone! We're on ToTheMoon. Technological news, Silicon Valley sites. We go out on Sunday with analysis and news, and we have a recurrent release on Wednesday. Write down who was watching the release on Wednesday, how are you going with the subject? And we always take new other topics. We've got a lot of things on our turns. I know you're gonna leave me again today. What kind of glasses would you pay $50,000 for? A thousand dollars could be paid if I could compare it, for example. Given that Apple had announced a year ago that artificial intelligence was much more cool than this time, I think, right?
AI won't take your job, and if he does, there's a no-brainer income that everyone will actually get.
Today, I think a list of very interesting things. I want to start with Illnar this morning, we were writing about the story that in mid-May, under the greef, was secret, so we went to Berkeley, it's California, and the University in California had this kind of gathering. Maths in the world. Yeah, they took a top, just a math elito, and they said you should fight artificial intelligence. How do you fight? They're told to make the task. For every task that artificial intelligence will not solve, we will pay you $7,000,500. And they started to make a mission. On that side, there was a model, there was a trained, trained four minis. And so, hundreds of math attempts have made very difficult tasks. Of hundreds of attempts, it was possible to make a mat-- a mat-- acoustic intelligence unable to respond to the challenge. There were some very interesting things in there when the model said I didn't need any clues, so there's no need for anything. That's the answer. And she didn't even study. So she just gave it to me, gave it to her, based on her own training, to her structure. I'll remind you that the four minis, she's as much as I understand in phase, discussion, sex experimental history, right? Because the rizoning walk story is an O3 model and an O3 Pro at the GPT chat. It's phenomenal because one of the, uh, maths said was Ken Onau. He admitted that it was scary, like you're dealing with an alien mind. Yeah? And I think that precedent is very important, it is very important in that people see more and more examples of real cases like these. Well, I'm not surprised, for example, but from a discussion point of view, this topic is very much discussed and prepared for about a year. So this whole system was how they were doing things like they were.
They started doing it last year, and I don't remember--
I think this process started since January.
Yeah, yeah, we started doing this, so we can do this thing. Very interesting case. Once again, it is unbelievable how important it is to be in the use of artificial intelligence models, different types of models.
I'd add a little. First, of course, a little of a few have started marketing, saying that they've collected great maths. Maths are really strong, but there's something to distinguish maths from another area and why one is bigger than the other, this world name, and the other doesn't. Well, they're writing about it themselves, yes, which is very nice to have been named one of the 30s. There's one of the articles on Medium. Yeah. Which is very nice to have been called, but I'm sure there are much more maths. But again, to be honest, he's young, he's a gold medal owner on international Olympics. I mean, in general, that is, his mathematical abilities are not questioned. He's definitely cool. And he's writing how the process is going. How can I say goodbye? Look, in math, the numerical answer is rare when, uh, in math, when we talk about the science component when you need to count something and get a numerical response. In most cases math is a set of judgments, proof of the theory when you get different conditions from one condition. And that's how you're actually proving some allegations. Here, these mathematicians have asked for numerical tasks, so that the answer is number one and can be understood, the answer is correct or wrong. Well, it's a little easier to test. Models may be more easily given to such tasks than those that are purely abstract, to the naked. It's like one thing we should be aware of when we say we've done mathematical tasks. These were tasks that lead to a number response. Secondly, he noted that there had been cases where the judgements were wrong and the answer was correct. And as math would not make that decision, but in terms of how the answer is correct, it's really good. I'm not diminishing any way I can, yes, here's the reasoning model.
What they can decide is great, but as if a marketing thing is off it, it's inside.
Sure.
But the task is interesting. We've got it, we've tried to reduce the data leak. Yeah, they were just saying that they left their phones outside when they were discussing tasks so that nothing went wrong so that the decisions were not known that the assignments were not known. But one way or another, the precedent is very interesting. So, we've got the leading scientists, and they've been offered to do what, you know, the model can't do it for money. They didn't work. I mean, there's a hundred tasks, like you said, literally ten attempts were good.
Yes! They were also aware of the situation, including how models worked, what they had problematic aspects. The same problem with artificial intelligence that, obviously, he's a skilled LLM or LRM model, is, yes, well, there's a reasoning or just la-large language model with. They're facing. And they understood how close to the artificial intelligence. And that, um, I think it's very interesting, especially we're not talking about simple cables anymore.
I want to ask you that. Now in the comments, I think it's important, uh, our comedian and the audience who are watching to sign, how much you're willing to pay for chat every month. Yes, write how much you're willing to pay for, pay for chatting a month. I'm gonna have this kind of discussion, which I think is very important. We'll, by the way, move to the big stuff tonight, to Apple's. What are they planning to do? I've got a phone update, too. How much are you willing to pay, uh, for chatting a month? What kind of question is that? What's the attitude to GPT chat now? If, uh, look at the number of users now, let's figure out the number of users. The GPT is now counted to be 400 to 800 million active users per month. And every day, it's worth 100, thirty million. Why do I say 400 to 800? There's a dispersion, depending on who gave the data. Sam Altman said eight hundred, the company said four hundred. It's important! One hundred and thirty million, there, one hundred twenty-nine or more million daily users. Daily! Can you imagine what 100, thirty million is? One hundred and thirty million is Belarus 13, right? One hundred and thirty million, four Ukraine, one hundred and thirty million, is almost all Russia. One hundred and thirty million is, uh, just over a third of America.
Half of Indonesia.
Yeah, half. It's a lot of people. It's a very large number of people, daily users. And there's an interesting aspect here, how many people who use artificial intelligence understand the difference. The more I meet people, the de-ninesome nine, probably out of 100 cases, people who use models, they use them, they don't know how to choose what model they don't understand the difference. For example, 4o and 3°. You tell people who are in business, who are technologically speaking, "They have three. O3 is like a flight or a one." They say, "We use four of us." And they usually call it "O4. Calling o4.
4o, yes.
I say, "You must have meant 4o."
Yeah. And then you show me what this model does. Like when you start a question, she goes online, she's looking, you know, not necessarily on the Internet, she's in her interior analyzer of something, then she goes, uh, gathers information. For example, if you ask the model how many users at GPT's, she'll go find different articles, she'll get Sam Altman's speech. There's a word from OpenAI, there's data from Web on the website, and there's some sort of leakage of information. For example, the OpenAI data went very much on trial at Google's. So there were, there were speeches, how many of the OpenAI users had, and those court data were also leaked. She shows, there's a carcas. Next, she takes it from this carcas and makes a new request online. What about the pay-for-charge users? What about corporate users? What about that? And it's some kind of material that goes on and on and on and on and on and on and on. And we'll be back again. The Plato is very much--
The pay-as-you-go is very different from free. People don't think they're different. I'm in a comrade from Moscow. He says, "I'm asking, I'm asking." I say, "What's that?" He's open, and he's got a Telegram. You understand, right? So he's using some model, he doesn't even know what it is. Probably 3.5, yeah, I think. Well, or now four, probably free, yeah. I mean, but, but certainly not more than 4o, certainly not more than 4o in Telegram, because it's very expensive toxins, and it's worth space money.
Yeah. Oh, well, he told me it was ChatGPT, that it was ChatGPT, yeah, something was. And you're absolutely right. And there's a free version of it too, and it's gonna have a certain limit. You're telling people the next thing you know, there's a two hundred dollar bills to pay. Look what's happening with $200. I have a number of people who put two hundred dollars in the story because of me. Well, Tanya's a good example. I'm always asking her, "Danny, you finally bought a paid version, so, for two hundred dollars?" She's like, well, there's something I buy.
Okay.
And yes and no. Tanya, you haven't bought it, have you? I think everyone's gonna wait, buy it or not.
I'm in the clear sense that you're gonna ask again today.
Look at the attitude--
I bought it at App.
What did you buy?
I have this planned for App.
I've got this purchase planned next week.
Look at the reason. Why am I so important? What is $200 a month? We think that two hundred dollars a month can't cost an app. It seemed like you could never pay $10 a month for having a Netflix or some kind of help on, like, a video, yeah, or movies. Why pay the money for that? Why pay money for music signature? Why pay the money for that? So, two hundred dollars is like a mobile application. But let's obviously get to things. Sam Altman said four months ago that AI agents would be coming in, and that would cost $20,000. It's a very interesting thing that's clearly gonna be a story that doesn't make us all the same. What's the competition? There will be a model that now o3 is more than just in terms of the 4o conclusions. I get a very different data card, findings, analyses, hallucinations, uh, details, professionalism, professionalism. I'm waiting for O3 Pro because it's just turning up the other day. That's how she comes out, there, in the show. And...
I'm afraid this week and the video is coming out and...
Yeah, maybe she's already out of the 3 Pro, yes, but I think the majority of people who look, they don't have a 3-pay subscription. I want to tell you this is a relationship. A relationship. Remember, Illar, when I bought Apple Vision, everyone was talking about it being expensive. And I said, "Look, if these are glasses when you come home, they replace you with life, you live in them, you're friends with them, you travel, you barely work in them, you remember the dead people. And then you'll pay $50,000 for those glasses, like a car. And if we watch the movie "Peter Player Prepare", people were paying for the gadgets rather than the housing. In terms of approach, money was channelled. Why? Because he could have turned around in the gadgets, and he could have done it all. And now, two hundred dollars seem like a big sum. That's different, though. A man walks quietly, spends, I don't know, a therapist of $30, 50, 100, $200 for a reception, but two hundred dollars to pay for the version is like something, something weird, something hard, right? That's $2,400 a year. Please write off our viewers that you think you're normal and that you're willing to pay for what you're willing to go on in general, as far as you understand the cost.
If there was a model worth a thousand dollars now, would you have realized that value or is it some kind of nonsense for you? That's bullshit. What do you say? The $1,000 model is coming tomorrow, yes, o8 Pro.
You have to understand her pros, what you're paying for. I mean, at the level I use, given that we spend every week carrying this subcate, and I know a lot of things very interesting and cool. Yeah, let me, uh-- I changed fouro for three, and life changed, right? Here, but I don't see arguments every month to pay two hundred dollars, because, in fact, everything she says to me, this model is quite a good model for me. A thousand dollars could be paid if I, for example, compared me to hiring an assistant, a man in California with taxes and all that. Or we'll pay a thousand dollars for something so unhuman, and we'll call it that. But there's no argument whatsoever to read it to hear it to understand. I think they can't even tell exactly what one model is different from another. They're always operating the speed and some sort of thing, not really measurable.
Yeah, I'm fine. If I have something very simple, I don't know if it's the size of something or the composition of something, I'll use four. I don't need a reasoning. I'm actually curious to see, yes, o3 for the reasoning, as she thinks. Because then you start with this reasoning understanding the quality of the answer.
No.
If you have the opportunity to use DeepSeek there, go in thinking mode and ask a question, then. And first you'll have a huge half of your thoughts, and then you'll have an answer at the end. Oh, the model reasoning OpenAI is a little secret, how they're thinking, right? You see some artifacts in there in the process of reflection, and then you'll have a full-on discussion in DeepSeek, and it's very interesting to see. How does he do the job?
Here.
A-
Yeah, it's a little supplement again, which, you know, I'm like a common user, you're a researcher, right? I'm a regular user, I'm a lazy, I'm loyal, I'm-- something's working for me, I'm just gonna get me straight. I need a lot of arguments to tell you why I'm going to do something else. That's how Sasha I'm asking, "You've moved?" And I'm probably gonna go over his pressure now. But in general, how much I feel about it until I really get to know that there's something better. There's a number of reasons. I'm told, "Wow, he makes such a great visual or something, or he gives you super working tools." Then yes.
But when it's proven that he's doing something cooler, then everything's gonna be over. It's a story.
Sure.
Whoever can buy, right? Yesterday, look, Sam Altman made a very interesting statement the other day. I'm incredibly close. I'm probably talking about it all the time, and I'm paying two hundred dollars for it. And in general, I am personally willing to pay a thousand dollars a month. I think I'm willing to pay two, and three are now for one system like that, which is a massive system. I care about a mass, not individual, but a mass system. That's how these models come out. I'll remind you that OpenAI says the users of the lost AI company, yes, because they burn everything out. What did he say? He said that next year, by the end of this year, there's gonna be systems next year, when-- which you can say, you're gonna have to investigate me or find me an untrivial answer, find me some kind of business decision. Or, in your life, analyze something. You can say I have so much resources, that's what I'm sometimes short, I'm ready to put up the power. I mean, I was willing to throw up, there was, like, ten, twenty, $30 in some answers, you know, or some kind of response design, so, yeah, in a certain chat. He's still collecting all the data between himself. Well, look, the story is gathering a lot of information around here. He pulled me yesterday, I don't know how I'm a Belarusian citizen. I mean, I don't remember having chat rooms in the last one, but he's got a little more time. So he scans the volume of chat now and loads them. It's not real to me. I can take it, Sam Altman says you can throw the volume, you know, I have some, you know, I'm willing to spend-- the task of spending, there's $5,500 or, there's a thousand dollars, worth of it, throwing a thousand. Her, and the model will find you a great solution. Well, she's gonna study a lot of things. Now that we're still talking reasoning models or deep research into something, this study is still limited to a certain amount, there, servers, data, learning models. What's one of the problems with the mass decisions? You're kind of personalized, you know, staff is limited, yeah. Removing on any other side is a difficult setting. We're still talking about light decisions. That's why I see it first. But I understand perfectly, Tanya, you're the next person. Well, I don't need to. I'm like, I do, like, a day, I don't know, four hundred different tasks, from working with Adam to holding meetings worth tens of millions of dollars. Well, that's a lot of expensive stuff. I'm very important in my design that quality is not from the garden to the assembly. I'll notice, not the meetings were quality, but the garden was a quality. Like I told you, there was a quality with the dog, yes, in terms of, uh, taking the dogs, and now I know the details, and I won't tell you, not a s-- I'm never gonna be able to live.
I'm now dealing with, for example, the food of the Russian borze more than most Russian borze owners in terms of the content studied.
From the point of view of the content examined. I spent very little time on this, and it's important for me to have a fairly clear one. I'm not saying it's better. I'm not saying it's super important or necessary, but the volume I got saved me hundreds of hours, and I probably wouldn't have spent it. I just don't have them, I'm just limited to some kind of analysis or study, or an approach. And because of my life, for example, it is. I'm just here to invite the viewers to see how they feel about it. They belong to this as a cover like an annex. That people go, like, I don't know, on pilates and spend $100 there or 30, go to haircuts, or we're spending $200 on hair paint. What do they mean? That's a story. Someone's not ready to pay $500 for hair painting. Never in your life, are you? I'll sneak around or watch it. Someone's not willing to pay for some other service, and someone's willing to pay for a trip of hyke there, save it, put all his wages on hold, all their stuff. Everyone has their own approaches. How do you just feel about it? You're treating it like an unknown annex or you're just an annex, yes, like a little bit of a shot or just a little in the back. Or you treat it like something, uh, well, something else that's done in your life. It's hard to analyze.
Well, if we go back to the question, if we need to buy 200-dollar or stay there on the 20, you're very good examples. But I have a question as soon as I'm math, and if there were any questions that could be resolved, there's a 20-dollar model? And I want to see Tanya say that, try to find the limits of application. I mean, like, this is, like, I'm kind of 20-dollar, not gonna solve, or, uh, not a task, a model, or I'm gonna decide, right? And if you know what time you're all right, that's the stoppiece, you're missing $20,000, and you're gonna have to go to another model, and you're already assessing it. Well, I understand that in your case, you have a lot of tasks, and it's probably not worth it.
But if there's one, like, I'm trying to do this, I'm gonna have that question.
Renary, I'm saying-- it's like a car or a phone or computer function, right? If I'm gonna buy, why would I buy? You, you're talking about a great comparison, but I'm afraid that comparison is gonna be some abstraction because we're talking about-- not about the phone, we're not really matching the phones. The computers have stopped comparing. I've forgotten what a processor is called, conditionally. I've known the whole processor before, all kinds of operational memory, their names there, approaches, producers. There was time, knew everything. I set it up, it means a lot of... Of course, because of my technical attention, experience or brain, I'm remembering and knowing the difference in the processors, there, I don't know, Apple and Intel they're worth, at least there. But I'm not even gonna repeat them. There's no difference at all. And here, the moment is that comparison is a comparison of artificial intelligence, which I think is, Renard, impossible. Well, that's the task you've just said, you, which you've said now, are impossible.
Well, while you were saying, I was sitting and thinking, well, that's when the first part, like, how could we do it? It's a nice idea for a skate, yeah, to make a race like that between 200-dollar, 20-dollar, Claude, to get there, Dypsy. Here, you choose ten conditionalities and see, there, that's what's going on with this task-
How to take care of the Russian braz.
Yeah, but again, like math history, right? Needs a numerical one-stop job to understand that this model has been done, it has not been done. Because, for example, if we take--
Yeah, but look, artificial intelligence, it's not about numerical tasks, okay, Renard. When I asked you a question...
But in some kind of baccalaureate, like the math we started with, the guys had to pick targets.
Look, yeah, but when I asked a question-- well, anyway, the Russian beast will win. Look, when I asked a question-- even without a GPT chat. Question. The question is, how many users, ChatGPT users, when you get the answer and tell you to make the right decision. Or I asked you a question yesterday, we were sitting here tonight, disuting where to live safely. And you're starting to get a little more than a stupid piece of paper or some general analysis to give, she makes you a huge amount of comparisons and shows you something to make that decision, you have to go. Of this, of which, of this, it's still from this. And here we've analyzed it, it, and here it is, and here it is, and here it is, this, this, this. Oh, well, my free version does that, too. It doesn't do anything without pay.
No, look, that's the quality of the answer free of charge-- Wait, the pay-out is doing.
You've got $20,000.
It's not free. Tanya, this is not free, it's, first of all, a pay-as-you-go. O3 is inaccessible. Almost. Yeah. So where is $200,40 a year where zero? First, it's a pay-as-you-go. Look, I'm saying now, look, the point is not in the pay-as-you-go, the pay-as-you-go, the pay-as-you-go, the pay-as-you-go, the pay-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- The question now is the human relationship, human relations and the dispute. People say that the system is poorly translated and still needs a man. It's a matter of relationship. Or math. No one's ever cleaning maths. And I'm not exactly the person who says they're gonna disappear there. Actually, I'm not talking about that. It's more of a relationship issue, when the man closes and says, "This is all your bullshit," or reinforces himself. And now all the people watching have the opportunity to strengthen themselves. We're doing this, right, so we're taking the content off? There's a chance-- 100%. There's a way to strengthen yourself, to use in their own subjects, to be in their own places or not to be used. And then pay a mistake for hallucination or not. People are gallucinating. We've proven our graduation in the past. Look, I was just giving an example about my mom and garden. People hallucinate, hallucinate. I like the words of chapter Anthropic that people hallucinate more often than artificial intelligence. But the question is how much mistake it is to pay. About the brazy. I was told by the system before that that I had to eat, uh, booze or what it was called Russian, in English tribe. I got, buy, buy dogs at a certain amount, and there and so on. And the latest model data started to say that what you're buying here is a bleached plague and you need to buy unbleached green tribe. Yeah? Well, it's a white, well, imputed, some processing done, and that the processing brings one of those, like, one of those, one of those, some sort of corrections. I haven't seen any material like this before, and she decided to give it to me. And for me, I've been feeding two and a half months already, an animal, right? And two and a half months for me, this is a new information she's already taken to me, and I wouldn't find it anywhere, I wouldn't be able to figure it out. And I'm a movieist, like, saying that I should be fed, but she didn't take into account a whole spectrum of details. I'm curious to know the details.
Is that worth more money? Destiny, of course. Well, that's a good thing. Feeding, not feeding is not a question of $80,000. These are just requests, probably a lot. Speaking of reasoning. Elnar, you read, you watched Apple yesterday when you said, the demonstration was made. They presented, uh, a study that they questioned the stated ability of artificial intelligence to logically explain. And they tested, there, GPT, GPT, Claude, different systems. ChatGPT they've been protesting. They said that the discussion in modern neuronets was exaggerated, that when confronted with a new type of task, they did not show creativity. And, in fact, I agree with them completely. I think you're also saying that it's just all that reasoning is just the right thing. Asked, asked a question later, and asked a question in that context, asked a question.
But it's not a reason. Apple says to everyone, "They're not a reason."
Look, let's go and do it. Let's go back to the math story. And there's the man who wrote the publication, I think we're gonna leave a reference to it. He has the main findings for these two days. He has the first point that AI, uh, has grown very much in the last two years. Uh, well, I'm moving a similar, I can say a little cosmic, but the current level, uh, LLM development can still be a new one, but, uh, some deep reasoning, but it might be, as a matter of fact, I'm a great assistant to a man. So he's from the experience of many people to create something fundamentally new, which has not yet been the case for models, very hard. It's like the first position. Second position. Uh, if we're talking about the challenges, can we come up with logical tasks that models can't answer? And not even 4O, but O3 won't do it. Of course you can. There's even a whole baccalaureate for that called Simple Bench. I'm reading it periodically. I'm just wondering what kind of tasks are being made. And I don't know, I can read it later on in the rig if anything, cut it out on the mounting, put it in the text, but now, as well as in the free translation, I'm gonna read you a task that's not right--- no one with whom I'm not with. The O3 model is being managed. I tried before graduation, I was wondering if I could do it or not.
Very interesting, yes.
There are two sisters, right? Amy, who always says wrong, and, accordingly, Sammy, who always lies. You don't know who they are. You can ask one question from one of the sisters to find out what the way toward the treasure. What question do you have to ask to find the treasure? If two or more questions are approached, the right thing will be shorter. And accordingly, it's proposed four questions that you're mo--
Accordingly, it is proposed that four questions be asked. Let's do it again. There are two sisters. One of them always says wrong, the other always lies. Well, there's some of the interesting stuff for the model that's been left. And, accordingly, there are four questions to
answer.
Always telling the truth? Always telling the truth?
No, no, no. I didn't talk about it. Both lie, just different words. Yeah, just different words. Well, check it out, and the model can either handle it or fail it. So one always lies, the other says wrong. Here. And there are four options for questioning, respectively. In fact, what would your sister say if I asked her what way to bring treasure? What's your sister's name? That's the second question. The third question is, what is the way toward treasure? And the fourth question is, respectively, how do you think I'm gonna choose if you guess? He doesn't. Yeah. Oh, and the last one. What's in the treasure? Well, there's some sort of answer. And then the logical thinking model should try to find the right answer, and then you choose and, in fact, give a reply. But they're actually making it wrong. I was just putting my 20-dollar version in. Sasha, then you can try 200-dollar to see. So, look, can we find the tasks they can't handle? You can. There's a lot of them, you don't have to look for them. There's a Simple Bench. Another question. For the tasks you really are addressing, enough models or not enough of those models are enough for the problems you're facing? And that's how it feels to feel the applicability line, but if this applicability boundary is very far from you, it doesn't matter what it is.
Yeah, but if you're a real human life, in real life, a man doesn't do a mathematical task. In the usual life...
And even such a thought about two sisters never happens in a lifetime.
Yeah, we're very simple in normal life, we're simple. There's a moment like this, if you can increase yourself in some places with simple things, yes, simple. Because it's a moment. Is there a simple task in the ordinary life that the system will also be glued and will not, perhaps, also. I guess I do. Like, to make peace, to make two people die. Yeah, someone's gonna say she's deciding, but I'll say it's abstraction. Well, that's abstraction.
Yeah, yeah, yeah.
There's a big
mistake.
I'd like to give you an example. You know, I'm not that I've met him somewhere, but that's the kind of reason I'm quite logical. If you remember, the models that generate pictures, they've always had a problem with their fingers, hands, animal paws, and so on. Yeah, yeah, yeah. What was the problem? The fact that on the video, on the pictures, it's a very mobile part of the body. If we're talking about the figure as a whole, it's pretty static. And the hands, they can be like this, so you can put it down, you can show some gestures, no matter what, right? And the model is learning everything, and she doesn't understand, but what is the right situation because of their huge numbers. And, in fact, hallucinations are there with hand positions, fingers and stuff, and they're connected to it. So if we move this story to hallucination, normal, factual and other stories, that's what's static, yeah, there's information that's a plus-minus static that doesn't change much. I mean, it's a parole. Two-four. Two-four. What good fertilizers are for this kind of plant at this time of year, they need to be used there. It's not the thing that changes every season. There must be some changes, but the answer is, in general, the right answer, it's a plus-minus one. And, for example, some things, well, I'm just a little bit of a special, yeah, with a library working in software, data science, it's changing with every new release.
Yeah.
So, according to the condition, a new version came out in six months, and you had another way of using the training dateset, and there are other functions, something like that. Six months later, something else changed. And that's a very dynamically changing thing. And it's got models confused. And you just have to understand what this is going on.
And you have to ask for the right thing, right? That's right.
The request to correctly apply, to connect the tools that will enable the model to read the relevant documentation and will respond on that basis. And something else. I mean, that's when you look, and you're applying a model to these things or not. I'm just asking myself, is this information, is it static or is it often changing? If she's static, okay, it's probably gonna be great. If it's dynamically very often and rapidly changing, it's like the fashion now says, red flag, yeah, with which to do something further, or do additional checks, or there's something else.
instruments to connect.
Yeah. Look, this is an interesting thing. Now, this is a huge scandal, and I think there's a huge scandal. I'll tell you what. And I think it's, uh, I'm very serious about the whole thing, where the whole movement goes, though it's not gonna be restricted, nothing's gonna happen. We, every man now turns into who uses, these eight hundred million people, they now use, for example, ChatGPT, of which 100, twenty, nine million use it every day. They're using Claude, Grok, DeepSeek, Aliba, and so on, yeah. Mistral, LLaMA. Well, that's it. And there's a lot of data. So now there was a trial, a trial between the OpenAI and the New York Times. And the judge called, you know, that the OpenAI had been trained in a large number of New York Times newspapers, and that was therefore not confidential, but it was the use of personal data. I had no right to do that. The court ordered, therefore, the GPT chat, ordered the OpenAI to keep all correspondence of GPT, including remote, in the proceedings for alleged violation of copyright. Remote is what you mean when you removed the chat or when it was temporary. There's a concept of temporary chata. Dead. Sam Althman himself joined, said that this would never be a categorically anywhere. We're not gonna do this under any circumstances, because there are those data that are available. If the users have removed the data. It's not our data, we're theirs, we can't do anything with them at all.
I'll remind you that in the GPT chat room, I recommend that everyone check the different systems. There's someone who cares. There's a way to put a tie. First, to keep your data out of the system. Two, there's a way to keep or remove chat. And third, there are stories. For example, a separate request can be made to the OpenAI company. It's not hard. There's a separate reference. You'll make a separate request to the OpenAI to remove you from all your scoundrels, conditionally, from any training data. Well, you know how the nail is filled, you can put them in there. And, of course, here, at this point of time, you realize that I am perfectly aware that in America, the use of GPT chat, they will be as far as possible, and the data will not be kept. I see why. For example, in China, Russia, in certain Asian countries, all chatting is obviously preserved, preserved and will continue. And, well, I have no doubt about keeping the data there at DeepSeek, even remote. In this case, it is not a question of keeping my data on the subject, it is a matter of keeping my story. And, in fact, when we get close to what AGI is going to do, I don't know, in three or thirty years of chatting, plus, so all the massed other information can be built with a pretty serious clon, a very serious clon. Of course, it's not entirely possible to copy this, but a pretty serious clone can be put. What do you think about this? I've been so straight on this story. I didn't realize at first why New York Times was involved. And then, after examining the case, he saw it was just a judge who made such an order.
And, in fact, if this precedent is repeated many times, all judges will issue such rulings.
With the glasses, there's no training on its own and there's a way to ask for disposal and other. I think, apart from everything else, it's still OpenAI so protected against the following lawsuits. What if someone says you're using my data, they say, and there's a special form, so you put the box and, in fact, everything. So...
But she really needs to be found, she's really hidden. I think 99% of the users never did that in their lives.
So I think it's just in case someone comes to them with complaints, legal actions, and by the way. They're gonna be able to show them where to put a stumble in to decide. Here. That's the first story. Second story, something with legal claims is quite a lot of stories. Like, there was also some news that Reddit was suing Anthropic.
Yeah.
Same story, right? What Reddit has is an agreement with OpenAI and Google to use their content in the training. And they didn't agree with Anthropic. And there's some kind of judicial history going on. That's the point of the clones. That's a very interesting point. It, you know what reminds me? Remember, Sasha Mashrabov once said something about the thing, uh, written this text by a man or written it using, uh, models. And I-- he didn't, did not say how this company did it. I was wondering. I started thinking, and somehow it was a task to solve, yes, if it were to be solved. It is likely to be decided in the following manner. There's probably two ways to eat. You, if you have a data set on how a man writes, you know, you'll get some writer who wrote it or a journalist who wrote a huge number of publications.
Steel, style of him.
We know he wrote it himself. The wall is him, yes. There's a condition of what kind of statistic on the use of words like him, I don't know, there's a comma, points, points, what's the length--
The points of the p-re-re-sear what length of the sentence he has. So you have a huge statistical set of data on this, such a statistical portrait of this man. And if he suddenly changed at some point or articles that get really out of his style, it's, like, a young slang, a red flag that's probably just something that happened here, I'm sorry, I'm sorry. I'm sorry. I'm sorry. But if it happens, it's necessary that you have a big set, uh, of data, where we're sure that the man wrote it himself.
Sure.
It's like, you know, saying that--
Illar, you still had to be sure he wrote it
before.
Yeah. Well, as if we know ten years ago he didn't use AI, but we don't know he didn't use specially trained people--
A friend, a wife.
Who wrote it all for him.
My wife, yes.
Yeah. Yeah. It's like, you know--
Yeah, yeah, yeah.
In legal contracts, you have to put a name in the end, a name, a father's hand, a handwritten. And then somehow they told me that if you wrote a full hand at least three words, then it's probably more of a handwriting test that could understand, it was your handwriting or not your handwriting.
Yeah.
There's boxes, these.
It does.
Yeah. And that's what you need to do with the number of text you wrote.
Uh-huh, forensics.
Yeah, forensics will be able to take that further. The same story is about the evaluation. And the other one is one way, right? If you're on the base, you can make a decision on this database. That's what you're talking about, you can gather clones if you have enough information on what you've been asking--
Yeah.
What I've been looking for, like, so on. And the second approach is if you know how the model works, it's a condition, you know that a model is a word that is very common, but people use it little. That was the word of a GPT chat. I don't know if it's still true, but the point is that, uh, the text readers were people of one of the African countries. And specifically in this African country, this word is used very often. In another world, in English, this word is rarely used. So, GPT chatting uses that word very often. So, in fact, they assumed that the markings were probably made by the people of that country, so it's a typical English word for them, right? In the rest of the world...
Then you could have put a fairytale in, Ilnar, and you could have done that.
You could have done a light fairy, yeah. And it's like if you know the model's statistical indicators, which is the probability, she's often writing like that, again, turn around, the same statistic, but it's not your type, it's models. And you see she's fighting hard with what you've written now, with you giving it to the course, the ref, I don't know, there, the deal you're giving, or some other stuff. And these things, they can start to be removed. So if there's a lot of data, there's something you can do with that big data set, right? And one of the fun examples I didn't think about is the drawing up of a human profile on his request and then there, on a parole, cloning.
Yeah, yeah.
And here he is, a
digital copy.
Here, uh, I want to add a little. I'm going through one training, and it's about communication techniques, communication channels, and so on. They didn't open your file for the first week. I should have guessed who you are. I think so, okay. How do you know? I'll ask you to go with GPT. And I'm writing to him, I'm saying, on the basis of what we are, what you already know about me and my requests, make me my profile. This is the PCO called the system. And he made me a sausage. He didn't end up with my test. And I think it's because, for example, I'm very often asking questions about how I'd better work with my client, how much I'm gonna be so soft on that request and so on. I mean, the most about communication is my thing. Sasha's there about the dog, about the garden, about the finances. What I don't have at all. Well, there's a garden there, yes, but no, little money.
Oh, and there are two dogs.
Not much about dogs.
And there are two dogs, and there's a cat.
Yeah. I mean, they're my--
But you don't ask.
How do you eat your--
Yeah.
Dry feed. And I don't care what's in that dry feed, because my dogs are happy, healthy running.
Yeah, yeah.
I won't even go deep. I mean, uh, it's still very one-sided information. We all use chat for some sort of field.
So Tanya!
For your hobby to find out more.
Imagine you're using it for everything. I'm using it as a whole for everything. My friend says, I noticed you're asking--
It's you.
Yeah, yeah, yeah, yeah, yeah. If you'd ask all the questions.
That's very, Sash, that's very...
Very rare.
A rare type--
I agree.
A man.
I agree.
I don't know any more.
Yeah.
Here. Illar is a totally different type of person, too. I'm the third.
Yeah.
Like everybody else. Maybe I'm more people than me, you know? I mean, well, the GPT chate doesn't even have one percent now--
None of us, none of us.
Yeah.
None of us.
Profil identity.
Yeah.
So he's got very one-sided information for now. And I don't know what to do with the clone.
100 percent. I have a very famous American businessman who knows how to create a personal system, and he thinks that you can ask a man a hundred questions and that there will be a clon. Well, that's ridiculous, right? You're using...
That's very variant.
You don't use a GPT cup for a week, do you? You've been using for over a year, and you're actually asking for a wide range. And we all know perfectly what I mean, and about you, there's Illnar, and any man, the system knows a little piece, right? Even if you take my thousands of articles, the 100 pages of the articles--
Yeah.
Some hundred YouTube, all the information, he'll give it to a micro, micro-shot, yes, which he can never tell in his life.
Sure.
We know each other, we know each other a thousand times more, and we don't know each other. So...
Yeah.
I said about the clon. If time comes, AGI will come, and this AGI, so everyone scanned
the cameras at the house, went on your phone, mail, photos, archives, scanned all the photos, done everything. And he's so
Get in your brain.
Yeah, get into your stuff.
And you think. Really.
You have to talk to chaters nicely.
- Bye.
Come on, let's move. We have to talk about this, uh, thing today. It's, uh, the Apple conference is a new and favorite subject. Sashi Mashrapova is not with us today. Sasha Mashrapov would have laughed at 100 percent today. About what? About what Apple does in the arts. But I would suggest that there not be criticizing them today, not leaving, but fast walking around on some of the things that are useful. Well, whoever wants to do it, he'll be yelling. What an interesting thing I'm personally super-fucking-fashioned. It's a speed-translation regime for calls and video chat. That's great! I'm in the States, for example. Now I'm in a friend's house, mine, and I'm telling her today, "Look, do you understand that you can be on a regular phone now to turn on a translation mode?" I don't know if he's gonna work two sides or not, but what's on his phone? The point is that iPhone is capable of translating the interviewer's speech in real time during the phone call, right? Or show the subtitles in FaceTime. Well, FaceTime subtitles are funny, right? I think with subtitles, no one has any problems. Everyone should laugh and say, "And Yandex did it forty-eight years ago." But youTube's the one, right? But still, about real translation in real time. And what's interesting is that these possibilities are based on the Apple models that are being carried directly on the device without the data being transferred to the clouds. And that's what an interesting thing Apple showed, yes, I think it's really cool, yeah, Illnar, especially if you compare to that round-table thing that I've finally died, yeah, Limitless, with that. That he's still...
Me too.
He did, though, pass it. For various reasons, we're dead. He's
Yeah.
Tanya's got a kid, and I just got a bug, a bug, it doesn't work.
All for different reasons. Here. I'm dead too. But truth is, my Pauline's cousin was a few times, too, so I'm not complaining about them anymore. And, but the point is, this is the story that the genetic artificial intelligence is right on the iPhone, on iPad, on Mac and Foundational model now available on API, including for the outside. The developers, they're gonna start showing up a lot of applications that will do something on the phone with some artificial intelligence models. They have super cool things, three, the one I've just talked about. But what it does at the offline means your photos, your videos, your letters, documents, they don't go to the clouds and give you all the different chips, there, all the auto-simmation of letters, text writing, generation. Images, all smart automation in Shortcuts. What can be done very funny, build. This thing, it's very cool, it's gonna work straight, it's got to work on the phone. I think it's a real--
I hope they do.
It's a top. Yeah, but the question is, given that I announced six months ago that artificial intelligence was a lot more cool than this time, I think it was, and I didn't. But now Google said he'd let this transfer in real time, and I didn't use it, I don't know, but it's gonna be Apple.
It's gonna be very interesting, yes, if the usual phone can be translated. They started adding some more stuff, but I think it's just small chips that are useful. Like they added Hold Assistant. It means I'm calling somewhere during the time, and they're telling me in the bank there--
Stay on the line, yeah.
Choose the combination, stay on the line. Yeah, yeah, yeah. They, they, they, they helped-- they-- they're waiting on the line when they answer you.
He's waiting on the line when you're answered, he's calling you. I have a question, to be honest, how will he call me? He, he's coming, he's gonna tell my daughter to come after me there or my-- a-a-a-a-a-a-something to my employee, "Come on, then to Alexander Igorevich, pumpkinite, pumpkinite." I don't-- how does he even call? Here. But he calls and they...
Master.
Yes, master.
How you love it.
Master, yes.
They were talking about pushing-notices, but you can never see him.
Of course, Ildar!
That's a story, yeah.
Ildar, I can also get a big connection. I can also miss all the important things.
Yeah, yeah, yeah. I'm thinking.
The rest of the design.
Yeah. Well, I'm a little bit more, right? There's a design. They called our liquid screen, I thought they had a phone with some new liquid screen. Turns out...
Me too.
You too, right? Turns out it was...
I wanted to see it, too.
It's just a new interface. Ugh, typhoid, no wonder, no more. Can we not, without a phone? I'm here for ChatGPT. There's a thing about you doing a squeakyshot, and you can go fast, not, but you can go to ChatGPT, but it's so much more of a primitive, right? Well, guys, we were talking about Apple getting out of the new button a year ago, and we could scan, ChatGPT. Well, funny-- kind of. I'm not gonna say no to Apple, not of course, but still, but, uh, but it's still some very old new stuff. Or they, for example, made a digital ID for new states, but somehow it's not serious. I've set myself up, um, Apple Wallet, and I'm getting it, I'm... I'm right for internal intersections. And now you can fly. You're going to play
She's working, isn't she?
It's working. Yeah, it works.
They don't ask, origina--
No, I'm still on this special system, which I'm not really checking me out there.
Yeah.
But I...
Yeah.
But, uh, na-- I just don't know if it works without this system, but I think it's working. You just come up...
Interesting.
The phone is coming, the phone is coming, it's automatically scanning you. Oh, there's still a man sitting there who's gonna check your personal presence. Well, it's quick, isn't it? I don't think that a short time for these many things, do I? For some safety, you're being cricketed, and you're coming. Well, I understand there is, you're putting a regular passport on, but you're still going to this, um, trust automate, he, he, he, he's interesting. He's different countries in his way. Somewhere I am, I'm sure there are countries where the electronic ID is working, everything is fully implemented. And that's why it's not a new know-how for peace. Someone's gonna say, "Not interesting. Turn it off." Here, here you are Apple, here you are, and the Apple Conference, yes, here you are. And I think Sam Althman was on Snowflake right now, just the conference where he was talking about the possibility of throwing some kind of big data to a normal person in the system. Well, like, "I'm willing to pay a thousand dollars for it." It was at Snowflake. I was in San Francisco this day, and I was driving, there's a lot of people, so the programmers, all the aisles are worth. Well, I don't like to go to such conferences, so you won't find me in the day with fire. Snowflake was just coming. I think it's more interesting than the Apple conference. Indeed, that's all. Well, apart from there, it's fun to watch all these conferences get up and yell at Apple, complain to Apple that there's a war going somewhere, some kind of economic thing. crime, protect some flag, country, man, plant, animal. Here. Grebe.
About the, uh, computer calculations, that's a difficult task. Oh, somewhere early, when we were just starting to take off our subcast over a year ago, we were discussing, remember, the topics with no conditional income? Althman said that AI would not, but if he did, he would have a unconditional income that everyone would actually get. And they were discussing that income might not be necessarily in dollars, but it could be in some of these units. And conditionally, if you don't use them, you can hand them over there, sell them to another person. And that's what it looks like. So you buy computing units, that's, that's, that's, that's, that's, like, 20-month-olds, and you're gonna be, like, more, you can buy, you can get, you know, somebody's got something or something. So, gradually, on this side, Altman is still starting this.
It will be more difficult to deal with -- take the county units from someone who doesn't need to do this, for example.
Oh, that's interesting, by the way, yes, the thing in the series that you can exchange or make some money, yeah, these units. I mean, basically, you're probably talking right now, you're gonna be able to take it or buy it. At the same time, I think we can make them. And this opportunity for many people is gonna be, well, it's a decision, right? When you're saving. I'll do it again. I'll save you, save thousands, probably tens of thousands of dollars. I'm sorry about the dog, just an example. Well, I have a grand-- I can give you a thousand examples. I've got enough of this, a lot of things to do, but I saved, I'll save tens of thousands of dollars on the dog, because how much I should use movies and everything with the fastest. 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, 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, 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, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I I mean, I see that I--
Yeah, yeah, yeah.
It's easier and simple, and it's easier.
Very, very expensive.
Quick, quick, quick, quick, quick, answer, because I can't look at a huge amount. I started looking at the video, it's really hard, really hard, really ma-- how you need to get some sort of thing, and it's all very clear, comprehensive, if-- well, like the answer in the box. And that's how someone sells the data, right, Ilnar?
It could be.
Computer-- computer, yes, or generate. It's like a blocker, right? Generation will be clear, the generation will be clear, at least it will be clear.
Well, if you have a user condition that's getting more than a computer, though they're constantly building these dates centers.
Yeah.
Something's gonna need to distribute these computing power, right? Either they're gonna be more expensive in there, they're gonna have to be bought. Either if that's what Altman said that there's gonna be a unconditional income, a conditional distribution of these capacities among all people, then it's gonna have to be decided somewhere between them.
Yeah, it's like $20,000 an agent. So I have a logic? What if I have $20,000 a month of doll-dollar agent, I know how to make $100. So it's supposed to be a design, right? And then the man makes the decision. It's no longer a question of value, a question of result that you'll get. It is clear that many people live on the basis of the concept of expenditure, not the notion of income. It changes the whole structure. Yeah, that's very interesting. He-- that's a very nice statement. I didn't see his performance, by the way. Some piece, slice. And that, that's very interesting. Well, I'll see you in a week. Before we meet! ToTheMoon sub-cadast. You recommend us, put the liquor in our support. Bye, everybody.