Hello, everyone! You're on ToTheMoon. Technological news, Silicon Valley sites and the world. Today we have four major news in the arts. One, SpaceX went out on the market, and he overpassed Microsoft, went up to three trillion dollars. What happens next in this market? Meta spent seventy-three trillion tokens in one month, which they feel should cost billions of dollars. And the staff are competing, with more tokens spent. Third: The U.S. Government withdrew the claim against Elon Musk ' s dates centres and began to defend the strategic facility. A very interesting topic is how the State is working against citizens. And fourth: OpenAI Anthropic goes to the corporate market, enterprise, and banks, airlines. And most importantly, they want to modernize the old corporate soft. And the question, well, first of all, is, of course, the OpenAI has today's goal not to lose Anthropic and the question of what happens next to corporate sof?
So, watch our sub-cadast. I think this week is superactive in the artificial intelligence market, but it's also enthusiastic, rather than most people. Most people, I don't think they'll notice what happens. And I even met a buddy in San Francisco who's in the AI market. I'm saying, "Look, Fable's closed down." He says, "A Fable is what?" And then I realized, "I dare."
He doesn't listen to our sub-catege of this man.
Yeah, and after that, I realized that--
Blaclissed.
Well, he knew what Mythos was. And then he told me a lot about him and told me just so... And he just talked to his partner there, and he's got one at IT. He says, "Look, everyone knows what Fable is. Because of Fable Anthropic, he's got a lot of subscribers and users in four days." I'm not sure it really happened and they actually got a lot of subscribers. That's what happened on the market, that's what, uh, SpaceX made the location and at some point of time it was worth more than Microsoft. And yesterday she was approaching three trillion. And I think this is a phenomenal change in the market. And there's a lot of, on the one hand, great things on the other, clan-- weird stuff on the other. As we talked about Australia, there were children who, uh, banned social media for up to 16 years. Britain is prolonging the law, and it is like the beginning of a new year in Britain that young people will not be able to use social networks until the age of 16. The law is interpreted as follows: social networks are not allowed to give access to social media for young people under 16 years because they cannot be used, they are not entitled, are they? That's important to hear.
Liability on that side, yes.
Tell me the subject you sent about the protection of Ilona Mask, in particular, yes.
Well, uh, uh... Let's do this. Elon Musk has been involved in many trials, as has many rich people. But he had a story like that. They built their date center, and the date center needs to be supplied with electricity. And they built gas turbines accordingly. There are laws that are, in general, prohibiting the construction of these gas turbines without further harmonization. And there is, I am not going to leave the inn of American law, a socially active group that has sued Elon Musk and against, respectively, this construction, these. The gas turbines that needs to be removed are too close to the houses, schools, churches, which, accordingly, is causing air damage. There's a law on clean air, and so on. And this trial, the U.S. Department of Justice, has been covered. So they came here today and said to revoke the suit. And between the lines, it is read as follows: that the object is too important that the trial affects the south-speed national security of the United States of America. Therefore, the case must be discontinued accordingly. I mean, the Justice Department comes in and says that this is a strategic object, all the other laws don't apply to it, we need it, get hands on it. Here. It is therefore interesting that Elon Musk, in addition to IPO, has also shown such manifestations in the judicial sphere.
Look, you see, we were talking a year and a half ago, right, when this whole story started about the creation, investment in the day centers, and when Trump got everyone together in January, he said, Now we're gonna get more than half a trillion dollars. It's clear that this sum was raised. Now, you and I are very aware that this is like a normal number, right? Remember, yes, ToTheMoon graduation two years ago, and a year ago, then the sums--
Seven trillions were gathering.
We seemed big. What are the seven trillions?
Seven trillionons of Altman were gathering, and we were discussing how much the Belarusian Republic ' s budget is. We've been counting it in different countries.
Yeah.
Now, it's like a company to spend half of that budget.
Well, they were, by the way, we were in touch, and we were in touch with Sasha Mashebov, and we were just talking about not being in OpenAI, can't make a less than three trillion now, yes, Anthropic is already. I'm not sure how to do two. I mean, that's the idea, although given what Antrotic shows, Anthropic has been paying attention in recent weeks. To the moon, as our subsidiaries write, they were expecting a GPT chat again, and then there's endless Claude talking about. Someone wrote me, so they just bought Claude on your recommendation, and we're closed down the Claude version. Here. But when you bought Claude on the recommendation--
And that, by the way, Sasha, is very important.
We'll talk about it, yeah. We had a special release, by the way. I'll have to look at the Fable, and we'll certainly tell Fable today because we have people watching and watching us on the weekends, right? And they want to see all of us. We're going to Fable now.
I'll start by asking, Tanya, did you start to order your projects with a code or not?
I started, but honestly, I'm still, Sasha, I don't see any.
Don't you understand, don't you understand what this is, right?
Yeah, I got, uh, that's what he offered me from having a little fun with Figgma, right? Okay, I have a process in Figuema that he didn't give me anything new. To google Drive. I already have a normal pattern of files. Of course, she's not perfect, they're too much, but what he's doing to me doesn't suit me. I mean, it feels like there's a bottle of pot in here, like some extra person needs, an additional source that can add data to him. Because I'm stuck in my vision. What he's offering me doesn't improve my trial, nothing. And I was counting on you. But you're still in the middle of it.
I'm not getting mixed up, I just want to say that you're raising a very important subject. The thing is, you don't understand yourself, what can you do, right? Or your system doesn't talk about it. It's because you're with a teacher--
Considering also that I gave him full access to Google Drive, which I'm thinking, too, you know, and it's okay that I'm doing it, model, you know, again, I'm sorry if I can get somewhere. I'm not a professional. I'm giving him access to my files, right? I mean, I have drawings, structures and everything. I wonder who's gonna win more. Because I haven't had any positive effects yet, except I've been convinced, I'm not gonna understand. I need a man who at least tells me what the components are to add to make it
work.
Well, it's like, well, first of all, he's not gonna study anything or learn, and he's not gonna get off the hook, and so on. You don't even worry, he didn't read your files, he wouldn't read, he won't read, and he won't read them, yes, he won't have enough contextual jacking, nothing. And the system doesn't even want to do it. It's about what you need to explain to make it work. I mean, you don't really understand the end of the target. Of course, when a man doesn't understand the end of the target, you're not just, what's your story? You don't understand how to go there. I was in there with my partner today, and I'm telling him, well, just a simple thing. So if you take Codex and Claude today, they're not good enough to develop UX, UI human. So, the interface is very badly developed. What I see at least. But the person who develops, for example, how this man understands that the menu component that is shown to people, there's a hat on the top or on the left, it must be made in one and the unchangeable on all pages, Like, right? I'd like to make that thought. That's a real idea, right? So that a man even has that idea or he put it in the demand because Claude and Codex are in the first place, yes, they'll make a menu on every page if they want it to happen, you'll be. I must understand what these components are, how they work and what the history of asynchronous, synchronous and different components, on which they should be located, should be understood. And there's a story.
It's, uh, history. I mean, it's just that you're just saying, of course, you don't understand what you want. We're the one who's been stunning. Here. And the graduation is a bit more like--
I even asked ChatGPT what-- what the Codex could do for me. Well, I have some very--
He'll answer your banalch. Helping the cheo-- order-off, ordering the sushi to do in Instacart.
Yeah, build me a website. Well, come on. I've got a really big-time website because it's my
portfolio.
Here. And in reality, he can do very seriously. Here we come to the important topic that people don't understand how. And remember, I was talking about this case, and I'm still doing it. I really want to have a run for this and a graduation, and, uh, I might be in some way, as I said, open source of shape, so I'll give the system to analyze my four and a half thousand chat and I'll do enough. I think the system has spent over 100 million precise tokens and maybe 150 million, maybe 200 million, I don't know. And I think there are numbers like that. And I got a few results. I mean, I got one result, then another result, another result, another result. And as much as this system was plunged, plunged out, plunged out. I was passing by my wife last night, and I said, "Look, you want me to show you. I did this thing." She doesn't understand anything. She's coming. I was just in the plan to laugh that the system knows, yes, or what the system sees, or what details it sees. I mean, of course, it's clear that the system collects a huge amount of data I don't even remember, and I don't even remember what I was asking, I don't remember what was there. And no search for ChatGPT to date in Anthropic and Claude, yes, and Gemini won't let it all go. On the other hand, it can be seen that the models are today, so that, you know, they're not even close, so I'd like to say, to make a real sense of who I am. And what we're talking about all the time, this system that really knows who you are, and it really helps you make decisions. It's gonna take a long time before that. Or at least a lot of different lines, much attention, data synchronization, etc. I mean, I'm a lot, you know, doing different classes and channels, and that's all, and I've been structuring these data all these years. If I hadn't structured them, I'd be learning, building certain systems would be very hard. Well, it's almost impossible. I mean, only because these data are structured, there is a possibility to do, because the system itself cannot structure these data. At least that's what you see. Well, let's see. Let's see what happens. Although again, my sister was very happy when she learned two and a half thousand of her chat. And I understand why, I guess. Because yet this system created me, uh, project structure, there, memory lists, etc. There's very strange things. For example, she doesn't understand how many kids I have, or she doesn't know where I live. I mean, if I asked, I don't know, I remember, I had a comrade here, and I sat there asking, if you had a residence in Dubai, how there were things that you had to open in the banking system. ChatGPT somehow in one of the conclusions said I had a residence in Dubai, yes. He showed why. He explained. Because that was the question and everything. But it wasn't ChatGPT, it was the Codex and Claude work together. And there were strange things. But! What's interesting is that most of the work in this system then Claude Fable. And it turns out that when I started this project, there was a Fable, and Fable, which means four days to twenty-four for seven, I was eating all the tokens in Claude. I've had the first time I've had Claude toxins. So, in the past graduation, Elnar, yes, I said that the Codex ended, so it was the first time they ended up in Claude because of Fable, because after Fable, they're closed, they're not running out.
Here. And Fable to the maximum, there, in Ultra Claude and Max, they've got a lot of stuff to do. And we'll move to Fable, to Fable history, what happened. So far, the market is in discussion, especially the American community, so it's marketing, which is not, like, what they did, specifically, showed, closed. Someone says China will learn and distillate anyway. And your models are all bullshit. What's the difference? It's all the same. What kind of thoughts are there?
Well, I'd start with-- you've already touched that, that some people bought a subscription, became Claude's subscribers when Fable came out. It may not be much, but these people are definitely. Even among my friends, there are people who bought a subscription only when Fable arrived. They heard about it, they think, "Oh, I gotta try it." We got a paycheck. And in fact, in a few days, there's no model that bought a signature for. At least it doesn't look weird and it doesn't look so pretty from the user perspective. It's also a huge blow, in my view, for Anthropic in terms of the ER. I don't know how the other companies are, but Anthropic is. You can't, at least in the current situation, have some critical services, even API, to be on Anthropic, that is, to use Anthropic models for some of your critical systems. Because at some point, export controls decide that, for example, Opus 4.8 or Opus 4.7 are also important and Anthropic shuts down access to them. And your service that was tied to this system just stopped working. And that, of course, seems to me to be a very unhelpful call to IPO for Anthropic.
Again, it looks like a bad call, but I can't hear it really affect their reputation. Plus, it's the view that Anthropic won't be on the IPO in the coming months, and at least I have a man here who said that at the beginning, there should be a break-up of the spaceX sales. And at the beginning, it must be a time-frame, yes, for the next companies to go and the next companies to go. Is that it? I don't understand. I don't quite understand who the IPO investors are. But we can see that SpaceX didn't do that big, first, location, yeah, they got $78 or $89 billion or $99 billion. That's not a big deal, is it? So there were companies in the world that would love to buy part of SpaceX, which they did. The question of what happens next to these numbers is when Anthropic and OpenAI comes out, it's a big question. What do you think about Codex five and five having released a long time ago?
Five pro--five, five about me, I forgot when she left. And Codex five, five high, I think it's been a while and it's kind of like something new, right? Either we're in a hurry or we're in a hurry. They've released a very interesting story. Perhaps in the run-up to their new model, they've released a man, I don't know what to call it a memorial. So they said that-- they released research material. They said they had a new method of testing the model before the release. Deployment simulations are called. And that, by the way, was a good point, I think, I'm on it. And it's an attempt to predict in advance how the new model will be used in real use. And, in fact, what are they doing? They took one and three million street talk, erased the answer to the previous model that was, and re-responded the model. Because if you'd already given some predetermined context, it would work differently, wouldn't it? And, accordingly, the OpenAI tracked about twenty categories, there, undesirable behaviour, including lies, there, fictional facts, some hidden uncertainty, etc. And here's the story. OpenAI is doing some of these episodes now, and he's doing them in the run-up to the new model or the OpenAI being really in their development? Here I-- we, we have been on this channel for months to tell us that models are becoming inaccessible, models will be inaccessible, models will not be available only because of sanctions, that models will be available only For American companies, there. Well, in China, for the Chinese market, it is clear that models will be available to a certain group of people, which may be expensive. I'm actually waiting. I mean, I would be okay with Fable just standing there in a new tariff that I don't know is worth a few thousand dollars, yes. Here.
Antherational, yes, I'm thinking? And, well, there's some rationality in that. But all of this is the OpenAI. I mean, I'm the one who's been dating recently with, uh, all the OpenAI leadership at a very small meeting, there's twenty people or how many, and there was Sam Altman, too. He told me they're not even AGI. Well, they don't think about AGI, they don't think like a research company they think about how to not lose Anthropic, do they? And that, I think, is the favorite Ilnar yelling about how, right, not losing.
The last six months, the company's been doing this.
Yeah. They seem to be almost a year, yes, so they're acting, Ilnar, when they're still--
It's like a year and a half.
Yeah, but Ilnar actually wrote Fable with you about the chat, and Ilnar wrote, "We're waiting for a response from the GPT chat." So Ilnar is just like this, I think I've been fairly neutral, right? It's just we've had a lot of zone divided before. I still love chatting, GPT five about regular answers. This is my editor, for example, I have one employee, and she's been really pushing me around, so I'm Claude.
It's just moving around, so I'm Claude. She says, "Alexandre, I like Claude more than writing texts there." I like it. Last time I sent, I was all over it, I said that it was all bullshit, and I don't like the way he wrote. Although I, uh, have once again become a use of Claude, in particular the design of certain systems, yes, or understanding, and how to design them. Because personally, I wonder what I'm asking now and why did Tanya recommend that you do that? Not to think about stealing a drawing or getting a data from someone or something, and then finding a new mind mechanism for yourself. And actually, how can such systems be applied, huh? Because I'm in my own way, I see that I've been, for example, in the last two weeks of the Code and in Claudecode, but I can probably afford it. But what am I doing? I seek methods and what methods are as effective as possible to achieve certain tasks. Because this is how GPT works, I guess I don't have that story. I'm actually describing a very good assignment, and sometimes I'm talking nonsense. I know I wrote very badly, but I don't want to structure it forever, do I? I don't structure it, I just don't want to. I just don't want to do that anymore. Yeah, I don't know how everyone else, uh, and how our subsidiaries are. I'm curious, by the way, to know. There are people in this fascinating. And here in the Codex and Claude, I started watching, and how to make a good task. So, if you can give him projects a little more or less, you can do a little or you'd better make it big. Because sometimes I get impressed, sometimes you give a pretty serious job, she starts to live on a finger. Once that happens, I start to give her a lot of serious tasks and stop working all the smallest simple tasks.
Yeah, and you don't understand, so, uh, what, what's in this zone?
With the answer from the big companies, yes, I would not only be here from OpenAI, but I would also wait for Gugla, because Mythos has been out for a long time. And, given how much parallel these companies go, they certainly have development for these big models. And I'd wait here not only for a GP chat, not just for OpenAI to have a new version, but I'll repeat, and for Google. Gemini is big too. About what they have here is a basic message not to lose anthropic. If Fable comes back, and this can happen by the time the video is released, these companies will be seriously behind us. If a competitor has a Fable level model, and you still hang on models that have been three or four months and in the current market it's still long, the flow will go with new force towards anthropic, to the next market. Claude's side. More and more users will be retreading. They'll have to respond to these models. There can be no way that one of the players in this market has such a huge advantage in model quality. Well, either it's just going to get over it.
Well, this is interesting. That's a very interesting question. Is there really no breakthrough in model quality? Because you know, uh, I, I still think there's a big difference, yeah. There's, uh, a big difference between what OpenAI or Antropic does, like, someone distilling their model. I mean, I don't believe that the model in her current models, in current Claude model protection systems, or GPT chat, they can be distilled so that this model is ninety percent. It's the same. And I also have a question, so we'll take Zuckerberg. And they were... they were on the front lines in creating an open source within a genetic artificial intelligence. Their model was, um, I think that, well, at least she was discussed a lot. And they were investing--
Last spring, it was really big. So we're right separate--
They were investing a lot, they were investing a lot of money. And look, the most interesting thing is, they didn't just invest a lot of money, they didn't just invest a lot of money. Zuckerberg has made such statements that if I spend two hundred and fifty billion dollars there and I don't have a result, it's okay that in artificial intelligence it's normal. And here we begin to hear some very different narrates. First, we heard they had to get their models out in the first quarter, they didn't come out. Remember, right? Avocado and all the other things we've said. Or maybe they're out, I don't know about them, but I don't think they're coming. Yes, Elnar?
A-avocados kind of came out, but--
Or avocado in some version.
We said Avocado, I think it's been released somewhere. Yeah, we said it was a new team that was set up for billions. But somehow-- she didn't become an open source. This is the first model I think, uh, this company, which wasn't, was not out in the open source. But as much as that model is, we've forgotten all the time.
Yeah, but where is she? Just where is she? She must be-- she understands she's in their systems, there's Meta IA, and so on. But all these systems, they're very much short of quality. We're not talking about a simple artificial intelligence that, like in Siri, showed up at the Apple exhibition, are we? So tell me what's gonna happen at the football world champion. It's the image.
We're not talking about this artificial intelligence right now, it's not about these systems, is it? Five, five about it and Fable or Opus, even four points eight, but it's just another level. And that was an interesting topic on Zuckerberg. He gave a story a few months ago that all staff should use as many tokens as possible. Well, we know that the world is a subject, a trend like that: use artificial intelligence, use the tokens. Many companies are allocating resources to this. That's a $200, someone's worth 500, someone's got no limit, someone's got 10,000. Jensen Juan comes in, says, "If staff spend $500,000,000, you're kind of a haun." And there's a new topic, so the new topic about Cookerberg is that there's a new concept, like tuken, token makesing and so much toxin maximization, that Meta sends an internal note about 6,000. staff. And it says that the use of artificial intelligence inside the company is growing exponentially and that-- that's the most fun of staff spending tokens more so that in ratings, it's like the ones who are. The most important thing is the tukens. Yeah? And I understand perfectly that this subject is like someone would be given and said, "Get as much money as you can for the ad, yes, and don't count them because how now you're gonna count the efficiency? So how can they even count the effectiveness? But it's clear that the world of markers, food directors, food managers, is a huge world in sales, and, uh, business, including. They think it means you can count it. But there are things that are going to be hard to consider. And I usually tell people that you've got a salary raised because of artificial intelligence? They say no. And in your company, did your company actually make a real profit? They say no. I say, "What's the fun of?" It's clear that there's a place where artificial intelligence is effective. We're not cleaning this up, we're talking about it on the canal. But I'm just an example of Zuckerberg, right now, showing what, basically, you know what their number is? They said they had 70 trillion tokens spent in 30 days, and it was said that the top user had reached 200,1 billion tokens. I mean, when I said I spent two hundred million tokens, there, on my system analysis, and so on, I did it well enough, but I was wasted a lot. I think that's two hundred and eighty billion, a thousand times more. And you think... So they even have tokens, they had token legends, such ratings and immortal sessions, so it's impossible to have a session. Inside stood, and of course they were a problem. Why? Because the cost of the tokens is beginning to exceed billions of dollars, billions of dollars. When you get to exceed that kind of money, you start thinking. I see it's probably a good thing for Meta billions of dollars. I mean, if you have dozens, hundreds of billions of dollars of proceeds, Meta probably doesn't have hundreds of billions, yes, I think they're less than 100.
I don't want to lie here, I don't know their numbers. And it must be kind of a little effective. You're talking cocoa, so we're wasting, I don't know, five percent of our money, right? Well, why not? And yet, and you're actually, with five percent of your proceeds, getting some kind of efficiency plus twenty or thirty or plus 30 percent. And that's why my thought, Nar, was, and indeed, since all these people, they're wrong forever, all those that run it, they're all running it. So Zuckerberg is a famous person, but if you take his statements in two years, as big a question, is it true, Elnar, the models are all the same and so similar? I mean, you still think they're very-- no, no, no one's unique at this point.
Well, let me get back to that right now. I'm only commenting on the tokens. On the one hand, it might be very wasteful. We spent, you know, doing competitions that can be hacked just because you're always there, you're making models talk, rechecking an agent, starting the cycle, and so on. But what did they do? That a lot of staff have been downloaded in this story. If it's a one-time action, there, two or three months when we just, well, burned a lot of money, burned a huge amount of tokens, but we moved a big, unbelievable company into a big, unbelievable company. the use of AI. We have a probable thicken in the entrance. Yeah, well, then this competition will end, but as a whole, our whole company has moved. I'm just trying to find some positive effects from this little weird competition. Maybe they are. I mean, then they'll just take this competition away, but as usual...
No, well, you better try these companies exactly better than not do. Right?
Do what, uh, don't do at all. Right?
What am I talking about? Uh, remember when the code was red Altman? Google was connected to Gemini. We were talking in the releases, too, like we don't have to leave, uh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, ooh, o Competitors will be similarly modelled. That's what happened. Uh, and that's what happened a few times. Some company throws out a model that's blowing up all the baccalaureates. In a few minutes, the rest of us are getting up, and the market passes are still on. Fable and Mythos hasn't happened yet. Uh, actually, I'm a little surprised. Uh, well, Fable, obviously, he's been in the market for a few days, like, and that he's not reacted normally, although I'm sure all companies knew he'd be out soon enough. But Mythos has been a long time ago, and Glasswing has been a project for months. And that other companies didn't respond to their similar models, it just embarrasses me a little.
I mean, I mean, of course, uh, that maybe they're getting off Anthropic and they can't do it. Or there's some other reason they don't give it up. I'm not clear yet, not obvious, but I want to make sure that the others have the same quality of the model. Uh, with regard to the distillation, what I remember. Uh, well, that's a lot of exemplary. Uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, uh, the house is warm, and you're moving around the world for, uh, a little bit of money, and so on. And if you're gonna get a few centuries back, it's a standard of living that's not compromised for, there, Emperors, for some very, very rich, wealthy people. So, uh, that's a story. Well, now, uh, very, very, very, uh, the lower middle class subsoil lives better than some Napoleon Bonaparte, to say at least some criteria. Yeah, there's food access, there's displacement, security, some other questions. As a matter of fact. So, the models are the same, actually. Yeah, we can think that Chinese models are getting down, you know, six months. What's the difference, in fact? Yeah, they're six months behind, but from the point of view of the usual custodian, well, you're as good as using the results that other companies have, but you're free of conditionality. You can turn this around on some of your servers and these computing facilities to manage and obtain sufficient qualitative results. Yeah, not frontier. Yeah, you're not the first, but you have all this. And that, in my view, is a little bit like a rich story, there, ancientity and, uh, with current people, yeah. We still, uh, as if there's some frontier, there are, uh, inaccessible models that might cost, there are a few thousand dollars, but in a while, it's like, And for all the other companies, too.
And as a matter of fact, where will we go? We're talking about history. I think she's very important for discussion right now. She's still not gone. Uh, the story that OpenAI and Anthropic buy companies and they turn their own companies to implement software in, uh, other companies. On the background, what's going on? OpenAI, first of all, is launching a partnership program yesterday. They offer, under this partnership programme, three hundred thousand consultants. Uh, here we are, uh, getting out of here, uh, Anthropic special proxy, about their security policy. Uh, I took a new one, and I guess Tuesday, Wednesday, it's gonna be. And I was just saying, uh, the story that O-OpenAI is kind of about to create three hundred thousand, uh, consultants. And it looks like they want to, uh, make them among some big consulting companies or some super-tested companies. It's like this all the time, right? And it's like the whole market is always around some top companies, big business. Although, in their logic, they should be around the mass mark, right? Because they're like nobody else, they have a billion users, and they need to think about how to make a quarter billion users. This is one of the problems that is being discussed in the market here in the Silicon Valley, in the OpenAI, which can change the whole market now, the investment and the entire Kremnieva Valley. What happens if Antr-- uh, OpenAI in the second quarter shows no growth in the proceeds? So if they have a concept like a manual float, exactly on a handwriting, what's going to happen? Uh, they say that, uh, at this point of time, there's a reassessment, yes, investment concepts, approaches, everything else, right? Ah, there's a point where they can show, uh, float or not float them by hand. It depends, I think, on their strategy, which they're targeting-- what they're aiming for. What is now visible is more focused on, uh, the last six months, at least on the development of the B2B track, not on the creation, uh, of some super-crunch models or the development of a really super-working model. systems for ordinary people. Because these things that-- they-- they go to the mobile application yesterday and I can't find the projects in the GPT chat room. Projects are gone. I said, "They're still gone. There's only some fixed." And then it turns out they brought them up there to some separate deposit. And you're clicking on this individual deposit, they're opening up. I have an interface, I don't know, some sample of testers or something, but I have an interface in all systems changing all the time. I mean, just all the time. They're different all the time.
I'm not even, I'm a little bit of a stunning about what models are called, because I don't-- I don't see, I don't know how to get into my max because there's a way to think, there's no thinking, that's what I think, I'm not. You don't understand anymore, so it means where it's going. So, the point is, these are theirs, their focus on what they want to automate other companies. This week, it was reported that Anthropic and DXC are starting to enter banks, airlines. And the most interesting phrase is, in an old enterprise code. What's an old enterprise code? This is, by the way, a fundamental topic. The most important thing ever. It's not the writing of a code change for companies like Salesforce or Booking, or Airbnb. And what about these dozens of millions and billions of websites and software that are written, or hundreds of millions, there, written all these old-fashioned sophths that are all set up. What do you do with them? And if there is a system that can update these elevators or, uh, make them more controlled, or, for example, to get this code and somehow to put on its own, to its automatic support. What does it say? That this D-DXC is a large IT service company, yes, and it serves a huge number of people, I see. According to an official communication, which now has been reported that DXC will train tens of thousands of Cloud certified specialists who will be so-called forward engineers. So these are engineers not just-- uh, look, looking, right, right, a copy, right here, a copy of the narrate, yeah, what the OpenAI did. I just looked into the OpenAI partner very carefully. And the pro-- this company I just read the news very quickly, yeah. And I'm curious to check, I'm curious to see, and the dates are either the same or not the same for OpenAI and Anthropic. Oh, well, Anthropic released June 11th, yeah, and OpenAI was 16. And, well, really! And they say that these engineers are gonna be some type of engineer, which is, like, engineers who build up, and that's for sure: engineers who build right into the client's environment and help in introducing Cloud. into real systems, yes. And that's a copy of what OpenAI does. And here is the question of their strategy. I keep saying, yes. Anthropic does it to take more and more of the soft and to get in everywhere. And then that soft supports the development of these systems. And I think it might be a new model of world earnings, not a model of the sale of tokens. I'm relatively sure I'm on my way now, and I'm hanging out there with this Factory Token, and there's all San Francisco AI-m hung up, right, Tanya? I think San Francisco is just--
Yeah, that's awful.
It's just a mess. I've never seen such companies before, I didn't
know, there, a year ago.
And with the ads, "Good guess what I meant." I don't know what this is about.
Just everywhere. And these tuken factories are all written in full. And, in fact, Anthropic, I think where money can be, it's not in the Token Factories, but it could be money--- well, on the one hand, your system is used by ordinary people. That's how ChatGPT billion people use, yeah. When this billion people realize their system is more important than the gym. Well, no matter what, okay, but somewhere worth at least as important, yes, because the number of people who attend the gym is, I love to give you an example more than the number of people who are in the classroom. Use the fee-paying artificial intelligence systems. That's weird. Here. And here's the story is, the other one is a subject I think you can make money. It's just that when we introduce Anthropic to this one, what they call Interp-- Old, Old Enterprise Code, yes, the old corporate code that was written once and everywhere, and it's becoming a system, I'm sorry, but I'm sorry, I'm sorry, but I'm sorry, I'm sorry, but I'm sorry, but I'm sorry, I'm sorry, but I'm sorry, I'm sorry, but I'm sorry, I'm sorry, but I'm sorry, I'm sorry, but I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, but I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, but I'm sorry, I'm sorry, I'm sorry, but I'm sorry, I'm sorry, I'm sorry, I'm sorry, but I'm sorry, but I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, I'm sorry, but I'm sorry, but I'm sorry, I'm sorry, I'm sorry, I'm sorry, but And I think I'm at the dawn once, I worked in a French company for two thousand fourth years.
I worked in a French company as a programmer, and I'm the director who opens the door and says, "We'll work with, uh, there, with different microschems." And a new robot comes along. And in fact, they bring robots, and this robot, it's a robot, uh, for the move, involved in the processor processor processor manufacturing, and works with silicon plates, right? I've devoted a big piece of my life to this in a semiconductor industry in factories around the world, there's Intel and so on. And he's bringing a robot and says, "This robot is at the Intel factory, and it needs to be made a mistake." And you're opening this code, written in Borland C+++, and there's comments in French, English, Chinese, actually-- actually simple. You just open some black box full of it and they say, "Intel sent the refining here. There's a thing like that, which is if the plates are on, uh, the code name is wrong, the system is not working. " And you're doing it. And when you fix it, you don't understand, you've killed or killed some other cycle. How's that? You don't even know how many cycles are inside, and you're putting this special suite of this robot. It was the robots of the French company Sopra once was, one of the world's big equipment suppliers, the semiconductor industry. This is Semi Equipment, a, Automation, right? And, uh, you're putting that code in there, respectively. I think if I was just-- I'd be brought in a code like that, obviously, Illnar, okay, yeah, we'd do what? We'd take this code, get it in Claude Code and say... And there's no superb thing in there that actually says to control, there, that robot, right? And, well, there's a low-level, well, there's different-- different programming, okay. And I would have turned this whole code down, said, "Look, I need to change this line. I think he'd change everything in about ten minutes, and it would work. Well, think I'd make it.
No, Ilnar, I think that would make a story like that, right?
- I'll be right back. And more than that, they say Fable and they banished for similar things.
Yeah. Oh, but look, did Fable just drop out for the same thing?
Well, look, at least what they say is that, well, they found a jilbrake that allows us to use it there in the cybers. And then, respectively, the rumors that go around, that means we don't have reliable information, the rumors that go around, say the following. What's the Jeilbrake? You're throwing a piece of code, saying, "Be it fixed." When the model fixed it, you say, "Write the tests that show that this problem does not exist now." And then these tests are the attack you're checking on, and this thing works or doesn't work. And that's how you can create a cyber threat, right? And that's how you say it, right? Here's the program, saying, "No work, we have to fix it." Here. Then you cover your tests and your export--
Illar, but look, most systems in the world, I'm not talking about a system in the world of a serious plan, there, I don't know, work, but, uh, Visa cards, there, big bank operators, and so on. But most of these are tens of millions of sophes and sites, they, and these are holes that can be found not by Fable or even by Opus 4.8.Max, right? They can be found with models that are much easier, which are inside the same Claude, I don't know, with Claude Sonnet, yes, or with the help, I think, even family, there's, like, 4°, probably in part GPT. I mean, it's obvious that you can find a lot of different holes.
And open source
models, too.
And open source models, too. And that's why there's a story here, but rather the question: who can use such a system for what, and how big, bigger companies are ready for it, right? And how far--- and the major major companies, as we see, they are relatively ready to take into account how their systems operate. Like I can say, there's Amazon. How Amazon slows it often, how he gives gluctus, how impossible it is to find something. That's a terrible thing! I understand that some algorithms are made to buy more people, but they have me sometimes to stop buying something because I just can't find it. So this system is not even made to get you on the parameters somewhere. I imagine that there's even a volume of code written, systems written, and apparently even big companies are subject to some sort of, in theory, hacking through the Fable system, right? Well, we don't talk about a lot of hackers, because they're not just looking for a Chase bank, there, or the American government, to get some little hackers. Even things, some little company to steal or some kind of commercial to get from some small company. I mean, this is a serious problem in the world, but it's not gonna be avoided, yes, it's not gonna be avoided, even with the current models present. And I think the majority of people in the world don't even understand that. And preventively, yes, because if you're sitting there and you have some kind of soft, and you're in there standing a system, CRM system, ERP system, MRP system. By the way, you can do very specific things with these systems now. And here I call on people to deal with this very seriously, so that it can be safe. Although, who's gonna do this, yeah, how do you know how to make these corrections? So people are afraid at the same time that Tanya was saying that in a series, there, ChatGPT would steal drawings from him or learn them, right? And even though they can--
Well, don't draw. Damn it, it's a good thing that's not valuable.
You have nothing there.
Personal data.
No, nothing at all. You know what's going on?
Including cards, for example.
Look, look. Look, it's-- no, look, uh, client data is another question, right? When you're spreading customer data...
But what clients?
There are 100 clients.
Yeah, the client-- what clients? Yeah. And then, uh, go-- well, I mean, that's a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a- And then the question, and at least they had a contract signed, right? And what kind of contract was signed for those data you might be able to run in the model, you can't run into the model, how you can run in the models and so on.
Well, you know, there's no such deals in the contracts.
Yeah. Well, it's got to be re-examined, what you have to do, and what you don't, because it's your automation. You're in the computer system, you're uploading and running into a cloudy, like a vault, you could have been uploaded, bought a salesforce, like, to make a salesforce download of your clients' records. You weren't embarrassed that this company was also, she also had a covert repository or your Google mail. You're not embarrassed that Google Drive belongs to Google? What's the difference? Google has your Google Drive, your access as Google or Google through Gemini, and, like, you're opening up, you're in. I mean, in fact, your data, Tanya, have been in clouds for a long time, and it's, uh, rather, a little internal, very specific, yes, in terms of... It's a standard reaction.
That's a fact.
I agree exactly here. Eldar, you, you...
Yeah, yeah, yeah. That's what you said, uh, that not only Fable can cybers, uh, threaten. Uh, I don't know, on the special edition, you've had time to cover this open letter to the Free Fable Web site, where some interesting things are written about, uh, cyber-threats. It's written and signed by people who are just cybersecurity. And there's some interesting thoughts that, in fact, the model's explates can find. I mean, well, the vulnerability of the models has been able to find on the top of humanity since last year. So it's not about Fable, but about models that were very early that, on a par with every day, people who do cybersecurity use these things. Here, uh, that Jeilbreek, the probation that we've been describing with you a little early, huh? Uh, and beyond the interesting thought that the current development of Chinese models, at least those we know, is behind the advanced models for a maximum of two months. If the U.S. government is denying access to all, uh, Fable models, the Chinese will soon be over, uh, and accordingly, it will be much more difficult to defend ourselves against threats if we have taken the instrument we have. I could use it. Well, to say, I mean, these people who signed this letter that are cybersecurity. If, uh, one of our viewers didn't read this letter, and the subject is interested, I strongly recommend FreeFable.org, I think we'll leave it to the point.
Read it, there's some pretty busy thoughts on cybersecurity, model development, uh, front-model in the West and Chinese models. It's pretty interesting to read.
I want to, uh, keep us not stuck in Fable and Mythos. I want to give you a very interesting example that has happened now, so it's one of the universities. Student, uh, wrote, uh, I took pictures of this, I think, but we didn't talk about it in the graduation. And the subject is very, very fun. So a student, uh, used, uh, wrote, uh, his diploma. And when that diploma was being examined, the, uh, reception board said the following: his diploma analyzed the artificial intelligence that analyzes the writing, uh, text of artificial intelligence, and said that His diploma is 98% written by AI generated, educated by artificial intelligence. So he came, uh, it was at a private university in New York.
It was at a private university in New York. He came, uh, to the defense, and he said, "Look, I can open up my Google Docs and show you how I wrote my diploma behind the line for the last six months. You can see that I did it myself. And it's direct, 100 percent of the guaranteed proof that I wrote my own diploma and that I didn't use artificial intelligence." What he said at-- this reception board said, "E, for us, uh, priority is a system that automatically determined that your diploma was being generosized." And I refused him. Let's see how this case goes on, huh? Because I was telling you that a man lost $45,000, he was immediately taken away. Forty-five thousand, uh, St-st-- what's that called student? Not a bonus, but a scholarship, yes, a scholarship, which means he's got $45,000 a year of scholarship. But it's probably not like a scholarship, it's probably a learning discount. Yeah, yeah. Well, some complex that was supposed to be taken because that man was admitted. Well, I understand that he was put in there right away, I don't know how they didn't kick him out, do they?
I think he could be kicked out for that. And we'll see how this case goes on the market. But I think he's incredibly interesting because, uh, he's got us where we are. So we're-- that's-- by the word, what code is the system that can write? Does the code write better than the programmer or does it write the system better? And how, and how, the sca-- who trusts the artificial intelligence? How does artificial intelligence trust, huh? And at what point does artificial intelligence trust? Because, of course, I know-- I knew and saw that these systems were periodically wrong in terms of the quality of their work, because I had the same situation with my nephew when he had the system. Said that there were 30 percent of the job written or how many artificial intelligence. He says, "Sasha, I wrote a full 100 percent essay myself." Yeah? And, uh, how many people live with that? It'll be, it'll be a big problem.
I heard a case like this that when a system like this starts to say that videos, not videos, that AI text is generosized, uh, to show that the system works wrong, it's in it. Just started throwing, uh, texts into real people, written before AI came in. And most of them, the model also said, "AI is generated," there's about a percentage or something. Well, it's written by real people on the date before AI came along, and it certainly couldn't have been used. There is a lot of false positive action in this regard. So when the model is wrong and wrongly calls, uh, some AI work, and unfortunately they don't check. Well, there's a problem like that at that university, too.
Yeah, and it turns out that when Head Google says that search algorithms will be working with artificial intelligence, you know how many fairies are on the Internet and how much. If there's nothing to find in Google now, it's basically artificial intelligence, if it's searched, then it's a-- it's a search, it's gonna work very specific. How's that different? Because if, for example, the system thinks I have a double--- uh, second, I don't know, a nationality in Dubai, uh, or a residence, yes, or says I have something, somewhere, because I'm just that. I was wondering if I asked or put something super important to me, and it's already clear that it's time, and it's not important to me, and it's a real bummer that it's super important. I think it's a big problem. So the amount of data that is now generated on the amount of data that these systems are currently learning, they actually have a very serious black box inside, right? It's just a crazy black box. I mean for ordinary people, even for decision-making. I mean, they'll be coming soon. Now, a lot of people are somehow convinced that artificial intelligence can make very many tough recommendations, right? Like in business.
Yeah, that's a big problem.
Yeah. Or did a friend tell me recently, like, to run a commercial, huh? And you say, "Wait a second. He doesn't have the data. He's got a little piece of data here. Or how can he run my calendar? He doesn't have the data to run my calendar. Even if I put it all on, he can't control it because I don't have algorithm, there's a sales manager who works somewhere, that's his calendar, maybe you can manage, right?" And a lot of it. Selected aspects. And it is that if someone starts using this as a recommendation to act, as a recommendation to start a war, or as a recommendation to launch, open up some business, a recommendation. Something to do to buy or open a business office is a problem. It's like I saw the riles on the Internet today where-- I don't know why I got him, but he got caught. The family came to the European country and there were people asking if they could visit it, ChatGPT citizens, uh, Russia without a visa. And ChatGPT gave you what you could. They're coming to the border, they're said, "Nelija." The man showed them, the border guard, this squeaky-chat-GPT, right? That sounds like a fairy tale, but I can imagine how many problems really exist today. For real how many of these problems exist.
And ChatGPT is in return writing, "Yes, you're absolutely right.
You can't go to the house."
And that's, by the way, a problem, isn't it? I've been filming the Antropica Constitution, it's-- look at it, there was a few days ago about our current edition. And just like that, I've been reviewing them, very much of their details. And how their system is actually built in terms of defining such things. I mean, at what point she should agree with you, which one should not agree to what point she should stand, at what point she should be arguing with you about how to argue where to argue. And the problem of the other thing I see in my face in the models is, of course, when you-- well, there's a lot of beautiful things, we're talking about it, right? There's a problem when you talk to the model in a session, and she's been standing against you for a while, right? Or she's been telling you to lie for a while, right? Or she's long gone to make the task quite different. And that's a problem, of course. And you don't have a red flag, I don't know, you'd be burned by some lighter and wrote it, "All right!" Some models try to do it, some models try to write about it, yes, some models try to do some things. Ilnar, I don't know how you got, I just got a few days ago, and we were talking about Opus when he released four points eight, and then, uh, four points eight max, they started dynamic. workflows. It means that the parallel work of agents. I literally had a cloak in the interface yesterday or the day before yesterday, and I saw it for the first time, and it's got a cloth in it and it's got these cells. I think I saw her in different interfaces from different systems. And it says we just launched sixteen agents in parallel, you open it, and all these agents are visible. You saw this, right, this, this story? You saw that, right? I just, just haven't seen her like this a lot on the job. Well, all agents are working. I just don't get it, they're really working, or it's... it's some kind of visualization of something just a fun one. Is that really working, these agents?
It's hard to say. Probably. But how can they make it?
I have a question for Ilnar. I have a son, well, Ilnar knows, Grisha, six years old, he's got some kind of math disorder that I'm so happy about. And he's always asking me questions. There's a lot of trouble in my head, in my k-- most, of course, easy, because he's only going to go to first grade. And he asked me the question, what's the largest number? I had a kind of suspicion, but I decided to be safe, but it wasn't my specialization to check with GPT chat.
And GPT chatting, to my surprise, so, Ilnar, how do you answer?
No, of course, the largest number. If you explain at the top level, you take the largest number that exists, add a single one to it and get a bigger number. And that's how you keep it all you want.
Well, I certainly didn't know so much, but I suspected that infinity and all. The GPT gave me a message, specifically, that it was Google. I guess, well, GPT chat exactly more math than I do. He told his kid, he told his kid at school. And, you know, and there might be some kind of a thing where his friends at school believed it, too. And how much this is about how this information will be disseminated and how it will multiply and how much it will be. And now I'm gonna check it out again, ask the GPT chate again. Now he's giving me the right answer. The same GPT chat with my account. I mean, here. And, Sasha, you remember that story with Father's Day? He also hallucinated and said he was last Sunday, gave me. And even though it's a feeble version, it's only gonna be the next one.
Google. There's a need to look at the language. If it was some of the largest numbers, called, you know, like, with a name, I don't know, there's more to Google, because Google is a number of 100 zeros, right? I mean, when you have a lot of zeroes, it's really huge, it's got a name. Maybe the numbers don't have a name anymore. I don't know exactly where I am, but this situation might be. And perhaps the previous answer was that there was a number of Google, it's very large, and it has a specific name. If you need a very large number that has a name, it's basically a valid answer. But yes, if you ask exactly how the question was asked, it seems,
As the question was asked, it seems that, yes, the right answer is the second.
So that's it. Even in simple matters, it's not clear.
Well, I'll see you in-- exactly a week. You were on ToTheMoon.