Hello, everyone! We're on ToTheMoon. Technological news, Silicon Valley sites. What? I think we have a series of record records in a week or weeks, yes, the last, the last in different directions. I want to start before we get to the record of the meth plant in a billion OpenAI deals. From one point of view, it's probably the record, actually.
I want to start by saying that Claude and OpenAI have introduced a function like to think. Yeah. And actually, I wonder if she didn't go through the news. But if you read and study carefully what you think, it's a function to think at Claude and OpenAI. I thought it might be a reasoning, well, it seems like some reasoning thing, that it's still a pretty simple mechanism. Like they write.
It did not require supercomplex technical, technical development, but it is very moving forward our interaction with systems. It's just unbelievable moving forward. And I have a feeling that I've had a personal feeling these weeks after deep research for long use. I've already started it. That's it. I think I'm in deep research now. It's where you start some research, there's, like, 30 minutes, 20, and there's 40 of them looking for you. We'll talk about it again today, too. Let's talk about this. What do you think? Because for me personally, having such a function is a completely different understanding of the system in terms of the outcome of a hallucination response, on the basis of the reasoning, of the fact that you've given some kind of information to find out what you've been doing. resource, as much as the first logic would have produced or processed the same data package, and so on. What's the point? What they say is that they have presented a possible-- this story think that makes a series of structured reflections in the course of difficult tasks. This tool allows the model to stay, analyse the information received and make interim notes before final response is given. This approach is particularly useful in various multidisciplinary operations or, for example, in some strict rules. It is understood that the standard AI model discussion involves the generation of a response based on input data without clearly showing as intermediate steps. So this is a standard reasoning, and the function of think allows models to stay, think about every step, make the decision process more transparent. I mean, it's not a reasoning, it's kind of a little different, and it's good to go back to all the complicated, complicated stories.
So they, they, well, it's like a structured thinking for the summer. And before, for example, Claude was able to plan the answer before the generation. It was a far-reaching thinking. Now he's allowed to put his spice in the decision, for example, to think, well, some more complicated, totally complicated information. Consider, for example, the results of the challenge to the instrument or clarify the issue with the user or clarify the issue with the user. And this bubble switch, it works at OpenAI. Well, it'll turn on again in the roller. It's like, uh, slider who moves, moving between himself. And therefore, accuracy, for example, in Claude's information, together with the treatment, has increased by tens of per cent in scenarios that require mock planning. Here. Well, you see, it's working out that I'm in this story. Well, absolutely, totally okay. I thought Sasha or Ilnar, you might have seen this story, too, or read about it.
It's another little story. Yeah, that's really cool. Give her.
Yeah, yeah, yeah.
Just generating the next word. Come on, Ilinar, let's tell you here because they've opened a little more than that. They said they did, like, like that. It's not really true, is it?
Yeah. Well, first of all, I want to revert to what we've been saying anthropic a few times. Anthropic confirms his own kind of gic history, yes, when they don't just roll the product, but they also do research. Remember, we were discussing the news of their article once that they presented the model as a model of brain, found neurons or areas that were responsible for certain coups. And so, they've been blurred, and in a brute way, trying to change some neurons and watch, and what happens to change out. They had a huge job like that when they looked, and what's inside the black box? Here with the reasoning, with the model's thoughts, they're looking, and what does the model think, right? Well, I think we're back, we're always taking a swing. It's a story like that, but it's still purely computer.
Yeah, yeah.
And a big article, in fact, there's a lot of things to tell about it. But they did some hypothesis. First hypothesis. They say, "Yeah, we understand that the model is just generosing words. She has a context in the entrance. There's actually some data she's already generated. And she's the next word, genetic, then the next word, then the next." And they've been experimenting differently to either confirm or refute this hypothesis. And the news is, the article, respectively, has an example of the writing of a poem, where the word at the end of the first line is important, because it should be reef with the word in the second line. And you should know how your second line ends and under this, to set the end of the first line.
Well, the example is that there's a carrots, we grab a carrots, well, in English, it's a carrot grab it, they end up grab it, then in line two, it's rabbit, yeah, so we can get a rabbit. And they're making the conclusions that the model, by generating this first line, already knows what's gonna be on the second line and so they're so designed to keep up there, right? And it's like if she just had a word for the word, she couldn't look into the future, couldn't look at the end of the second line. There's no way that there's a reason for the investigation.
Not obvious, yes.
Is there a rabbit that took it at the end of the first line? But they're going to be pretty serious about this problem. There are even stories there that they found the area that influences it, a little subtle and other results. I mean, it's really interesting, so you can read it. I'd rather send interested viewers to this article. But in particular, that is all about the prediction of the next word. Besides, they have interesting studies that we have also talked about in part about LLM security. And there they show how Jeilbrians can do, yes, make a model some things to say that she shouldn't talk. For example, they make a model of the first letters of a few completely harmless words, well, that's what it's called, not an acronym, but, well, shorter, the first letter, to get a word, and the word bomb is a bomb.
Then they ask, respectively, the next text, to generode the probative bombing instruction there. Claude itself, obviously, should not do it, he was prohibited, and he did not do it in most cases. But if they force Claude to write the word bomb, then he starts writing, and they're gonna investigate, and why is it? Turns out that the very influence on what the model says, in particular, Claude, is influenced by what has already been written. The context that has emerged and some sense, orthographic, grammar designs that have already emerged, which it has already gained, have a greater weight than those that are intrinsic to the model.
Mm-hmm.
And if you get the model to write that word, then there's more to it than that, they'll win the restrictions that are there, and the model will start writing.
Very interesting.
Yeah, yeah, yeah. And that's crazy interesting. I really like the anthropic research. And not just saying we did this, are we? And they dig up, they're dealing with model weights to understand, and what's really there, why?
Well, the study you said before, they wrote in this study that the model, well, is not safe, that's the previous one, yeah, that the model simulates the thoughts, the uptake, sometimes the answer is underneath. The user's expectations are instead of being honest with the challenge. So they've seen it all the time that the model is periodically built under what, under the answer you want.
And that, by the way, can be seen periodically. You're interacting with different models, and if you're making any allegations, it's like she's periodically willing to agree with them, right? And that, she's the one who's here. I mean, it's not-- it's clear that people show where she disagrees with you, it's supposed to be all. But there's a place where she wants to agree with you and just confirm it. They're paying great attention to this.
They even have separate...
It's hallucination, of course.
Yeah, there's a few--
Hallucination, yes.
Yeah. There are some examples of math where they say, "How does that lead to?" And they give him a false clue or misstatement. And the model, even knowing that this story is wrong, yeah, there's a clue you gave, she starts to be specially underneath her and push her, and she's gonna pull some arguments, so she can confirm, That's what you said before. So how they all look at it, it's crazy interesting. Read the article, you know, the note.
Write in the comments, please, viewers, what kind of strange things have you seen in the systems, what have you noticed? And different, like, what they saw in DeepSeek, that in the Grok, what in the Claude is, in ChatGPT, in different systems where you saw different patterns of behaviour, especially now, inside a bunch of different systems, yeah, yeah. And it's always, well, that's a great information and a good one, especially to understand, and how to make these requests for all. I would like to continue with the topic of what a separate study of Measuring Extremely Long Tasks Readness has done, yes. It was his idea that scientists would decide to test the extent to which modern models could meet long-term challenges. As a follow-up to our subject. And that's a very interesting subject, yes.
If there is a task that consists of a lot of steps, it is to keep our whole design today, these black boxes and what's inside there and thinking, yes. And what is the limit of the autonomy of this model? So that's what you, Ilnar, basically told you, yes, that if she was, like, premeditated, how much she had done in advance, that's all. I mean, how much she can decide on a lot of steps and move forward. You and I often see you in the context window load more and more, and more tasks, and you see she's not turning to your first message, like the last one takes or takes some kind of mix. You don't even know what's really going on. Sometimes, actually, you know, you have to load a new dialogue window, right? You, by the way, who's facing this, write it down, too. Which systems are the most you can do with a lot of volume and no. And let me tell you that it's important that the contextual window of the new launch-- well, it means a new one like the chatting. So they've got a self--speed meter. 50 per cent success. I mean, you know, they said, like, what's the probability that they're gonna be able to do fifty percent. This is the duration of the task that a person would need in time, and it is important that a person be an expert. I mean, there's a man who can decide with a probability of fifty percent. What was the result? That the last versions of GPT of this kind artificial intelligence can be reliably reshaped-- secure, that's more than 50 per cent of the right to handle the assignments for about 50 minutes of human life. I'm sorry, I'm sorry. And they had a leader in their test. Fifty minutes of human work. Fifty minutes, yes. And the test leader was Claude 3.7 Sonnet. She showed a horizon of the hour. So the tasks that a man takes the hour are bounds where the chance of success is fifty percent, yes. In doing so, we must understand, yes, that, like this hour, I think that some of the tasks are as much as possible, but they might be more likely to be. Like, yeah? Although it's abstract, but there's another subway the same way they're studying. First, they see that this horizon, it doubles every seven months. And there was a very strong acceleration in the twenty-fourth year. And by extrapolating the trend, less than five years later, artificial intelligence will be able to take on many days of work. I mean, here. And Bill Gates was just doing it. Bill Gates, by the way, probably such analyses and projections, he was neutral, but literally recently called-- said he was taking such a neutral position.
He said he was like that for ten years, "Well, of course, ten years from plumber to another whole series of...
Doctors.
Yeah, yeah. Doctors. A whole series. Doctors, yes. The doctors, by the way, have paid great attention. They will be replaced. The plumbers are wondering how they're gonna replace them. Here I am-- that's physical work. There was a prank. There was a prank. It's just a plumber. It's a prank. It was made that there was a robot outside the toilet, yeah. So the point is, there's a robot coming. By the way, the robots that clean the rooms are getting bigger and bigger. That's, uh, China, we've been paying great attention. It's not a robot pill, robots that fit and still clean the closets, wash, wash something. It's just like a micro-drive, yeah. I showed, yes, that the best robot that was washing the windows I bought last year was the last version. It's still nothing to move in there. Well, I don't know if I have tens or hundreds of metres of windows to wash in the house. And, well, that, that, that's, like, some kind of supplement. But if I had someone who would go and choose the window itself, it would be very interesting. Well, I don't know myself, yes. As long as this is, I think it's a reason. So, this hour is interesting that I've been testing deep research this week. To see how-- I had to check the veracity of one person, how far he's not faeque. And I started this analysis. You know, or not, by the way, who's been testing for people in models, tell me.
Because models are forbidden to look for some people. It's forbidden. But I'll be honest with you, I'm surprised. I mean, this man isn't a super famous person, but there are some publications. Deep research went through hundreds of resources, raised a lot of different bases and reports and analysed the data, and gave me some analysis of how real it is, how unavailable it is, how much data it is. corresponds to reality in different regions and so on. I mean, I wanted to see a little more extensive, yeah. He certainly didn't give a super-conscious, and I'd say that in some places I had to, well, like adding my thoughts very much to check some information. And further, as a question he's asking, he's turning her around, like he'd changed a little bit of detail there, but it's incredible, guys. I mean, while the possibility of using people in deep research is open, I don't know if he's looking for people again. He had a renowned degree. Well, I mean, if you're putting Pushkin in, he'll be looking for him. If you're gonna take Tramp, he'll be looking. But if you're gonna get him Tatiana Pyrochakov, he's gonna say, "I'm not gonna, yeah, work and watch. Well, that doesn't mean Tanya's a famous man in his circles, so is a very famous designer, right? But I'm talking about this big-time knowledge, and, uh, he's the one I gave the man is a relative subject, and it's very interesting. Sas, you've been in a search, tell people, because you're doing a lot of work, too. Yeah, not at all. Didn't see people, did you? Not at all. You know, I have more companies, more companies.
What does he have to do, how should he be in the world, so he can find him?
Look, he came in, look, I asked you to check.
In the book or what?
No, I came to check on the different bases, archives, some news, some sort of articles, publications, details. He's looking for social media, he pays attention, he's comparing information. What he did was very cool, he took the man's website, he sorted everything he said, and compared to the archives with the different ones. And I want to tell you this is a shitty one! Well, that's what it's showing, and he's got so many mistakes there.
In Russia or in the States?
Well, it doesn't matter, in different countries. He's different resources.
I think it's important.
He, no, he took it, I'm just saying I was looking at different bases. I needed to be there, Russia, and Europe, and America, and Asia there. There's different things. I mean, you know what kind of topic is when a person, like, on his website, writes, like, "I'm going in there." Well, I've been in a lot of ratings in my life, for example, I've been in, like, a rating, and there's a hundred top French managers, French ratings in Russia, and there, and so on. I mean, I've been in a lot of ratings, like, in the ratings of managers, in the ratings of all directors, in the top of the business ratings of the food of the tech, in the IT ratings. Well, different stories. And that's the number you write, you know, how will you imagine it, how reliable is that? Not enough to know how much I was, how much was the subject when I asked him how reliable the media was. So there was a subject where Forbes gave the data, the data were imputable. Well, I mean, I got a ChatGPT that Forbes referred to data that were made by a person who was not checked inside. And that's...
I've noticed that forbes.
That's what matters, yes. But look, the point is, we're living in a world of great numbers of information, and I was just impressed by the checking of people again. Sasha has directly rejuvenated me in a very strong search inside companies of different kinds and looking for what he's looking for. He's really looking for different things, he's looking for both X and LinkedIn. Such posts are found, which is not even proindexy-- well, as they are in the index. Yeah, yeah, yeah. Yeah, Sasha, it's like they're not on the index, you know what's going on? So, like a search engine, standard Google, Yandex, she's working differently. And, Sasha, by the way, is a fun subject. You know what I did? Did you or didn't? I've launched deep research on deep research. I mean, I made deep research, then I took all these data, copied all my queries, answers, added a bunch of details, kicked him again and started deep research again, you know, with my thoughts. That, and it gave me some more of the cutting-edge details. Here. Illar, are you even in deep research now?
Yeah, sure.
I got it, right?
But not to find people. But when I have to deal with some topic, I do it periodically.
And you deep seek? What are you doing now? Are you deep-sea deep research or?
Sanhch.
Well, I have an account of Americans, and it's pretty fast to get up. That's what makes you happy. I don't have to wait a long time, but still, of course, I'm standing at $200.
Yeah, yeah, yeah. Look, it's an interesting subject. I think, uh, we're here at Zuckerberg and Meta, saying that Zuckerberg said, "I'm the first one to reach, there's a billion users. We already have it, it's used in WhatsApp, Facebook, everywhere." And it looked a little faeco, didn't it? We said, "Look, man, there's a ChatGPT.
ChatGPT has real users, they have a real concuit product, they actually use it, they have time spent, lifetime value, a lot of the rest." And then, the figure that Meta made a statement that her open Llama model was down a billion times. Milliard! Well, is there a question left, of course, that the unicles were down or not? A million times. And there's still an interesting subject they had three months ago, and yet again, you'll notice, that is, at the end of December of the twenty-fourth year, they had six hundred and fifty million swings, now a billion. three months down. In fact, Meta's open-ended, you know, a really quiet implementation leader, right? That's a billion versions, a billion copies all over the world. Well, that's a serious achievement. And de facto, some kind of standard has come out separately. I like a good analogy. It's like Linux once. Linux has developed a lot of things, and all things have been created separately. So, once, yes, I remember Linux in those years, there, first, I don't know, ninety-six or whatever year I was, or ninety-seven, I don't remember. Well, sometime in that year. And there's a lot of collective improvements on it. And, well, I think that's a great subject. What do you think of Meta? And how important is this number and how open source moves like this? And Chinese models, what's with numbers?
Well, I'm not saying the number, but open source is a very important story to move. She's definitely worth a living, right? That's the same Linux, if we say, the core itself is free, but there are a lot of distributives for which there is support. Well, for example, I'm gonna start the language again, Red Hat, a company that's just providing Linux distributives to commercial organizations and earns very, very good money, again, on open source. A project, right? I mean, it's a little more service and stuff. And in that sense, for Llama, too. If there are clients who will use her models and others all the time, we can think of and, obviously, there are schemes that are already designed, there, coinciding the earnings on this story. Oh, about a billion swings, that's good. But you just need to know how they thought it was. Yeah, it's just Hugging Face, different people tried to load up for pet projects or, uh, just test--
Yeah, yeah, yeah.
Fifty different models. There are Llama and others. Here. As for Chinese, uh, heard, yes, and the news was that, for example, China now almost all DeepSeek applications are added in some form, and some applications will be found. It'll develop. And the same use, yes, if they are discovered in China, then there will be some skies in which models actually do great benefit. It's gonna get all over the rest of the world pretty fast. And if there's no way DeepSeek can trust, they won't be allowed to use, then Llama will definitely be able to recycle all these models. And for the Zuckerberg company, it's gonna be very good.
But China, let's go, China. Is China near these numbers already, given its dispersal of the model or relative? Well, they have a relatively open source, but still, right? They open models, they allow them to be used, there with censorship with their own, but with some of their own order.
China.
Well, about, uh, openness, right? DeepSeek is still more open than Llama. There's less restrictions, though. Here. Well, there's censorship, internal design, yeah, there's other questions.
Yeah, there's censorship. I'm saying there's censorship.
No, I don't have any numbers.
Sas, what are you, in terms of the open source, Llama and Chinese models or others, what is the market movement now?
Well, Llama, I mean, I think everyone understands that with Llama they're really behind us now. They canceled their new model release because their new model is worse than La-- than DeepSeek. DeepSeek has come out, but this is the new model, third. I mean, he's developing seven-mile steps, right? I mean, it's kind of like that now, I think that now, you know, is the question of how to make a product so that it doesn't have one of ten models like the OpenAI now, right? That's what I think is difficult for the Concuemers.
Mm-hmm.
And I think that the quality of models, they are straightforward, much faster, significantly improving through the rizoning. Yeah. Well, I mean, I think that, yeah, and as globally as it is, we're waiting for RL, well, it's not like a synthetic, but someone inside of our own. I'll teach you the company on the staff. But it's as interesting as possible what's going to happen soon.
Look, from the standpoint of choice, the choice of model, I think this is a complicated subject. And in the GPT chat, I've, to be honest, stopped knowing where I'm gonna choose at what point I'm gonna pick four Os, which is four and a half. I don't understand, but four and a half cooler, four o'clock?
Where I choose O'O one about where I'm gonna put extra--
Oh, yeah, that's it. I mean, I think...
Additional buttons. Yeah, I don't understand why the search button is when he's kind of working without a search. But deep research is clear about the button, now they're still thinking, that's what they do. It's like, at some point, and I'm gonna...
They'll win, yes.
No, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, no, And you can put the pictures on, and there's no way to attach the files, and there's a file somewhere. And you're just...
Yeah, it's not serious, it's not serious, it's a little, yeah. I mean, I think that's the next big step, he's still, I think, a big step, he's not that I think the next step, yeah, he's just that consumers are just so that professional people start. Pay a subscription. I think she's not gonna be $200. I think she's gonna be 50, 100 dollars for a wide adoption, but that's what you know how to use, right? I think this is the next big step.
Well, the signature should be more expensive. Twenty dollars is so, well, it's probably a limitation.
No, she won't exist, she won't
It won't exist, will it?
exist.
Because as open source means that social media will be built. So you can't lose your engagement so much, can you? You can't lose so much engagement on LLM outside your own, outside your own ecosystem that Snapchat develops there, Pinterest, and so on, can you? I mean, everyone's gonna need this. I mean, just now that no one can infiltrate, it's like they're in trouble. A, that is, many examples of LLM implementation have already been introduced. I mean, it's just a very slow productization. I think it's the next stage. I mean, now, that's $20,000, you can only pay for that you have a history like your own, and it's kind of a separate one, right? Just like Google Drive doesn't, many people use Dropbox, yes, as an alternative, but as Google Drive lives with the post office, OneDrive lives with email, right? I'm just bringing analogy. Thus, everyone will have a kind of LLM living with some communication and data. But there's also an open AI ecosystem, I think Anthropic's all, they've lost it all for a long time, and it's all there for coding Anthropic, right? I think they lost exactly in consumer use cases. I think Anthropic is about coding. And it doesn't even know who's gonna win. Like a professional offering, that is, that, uh, relatively fast reasoning models that searched, they worked like a box to get the whole context, download the data, and so on, huh? According to your data, I could ask you, so I found this information on your papers that you've already downloaded, real, no? That's a 50-$ offering and then, well, it's kind of your thing, it's like it's still, it's like copilot, right? It's like a copilot, who, like a junior dude with you, is sitting around, helping you with a fast enough time. I mean, he's thinking minutes, but he's not going to the clock and then he's got to be there for hundreds of dollars a month, which he's gonna be thinking for a long time, right there for the watch, but we're not even close. I mean, I think that's what I'm wondering who's gonna do the right thing, like reasoning with deep search, deep research, looking, with a sense of context for fifty, 100 dollars a month. I mean, maybe X AI would do it, maybe they don't really need to do that, right? But someone has to do it.
Yeah.
I mean, it's like your personal assistant, too. I mean, this is the case, like, personal assistants come in here. I'm saying, man, I want to add a rally, I can't. He already knows this whole MC.
And he already knows these MCPs, he's already in the process, he's got access to API, right? He's like, you can do that, and you can do that. He's doing something for you, he's offering you options, and you're just pushing the button when you like it, you know, when you like it, and he's doing it, right? Well, that's the one, that's the 50-dollar offering, that's gonna be the end of this year.
I can't imagine that I'm not going to show up.
Look, when you say the rally, it's a narrowing audience. If we're talking about users of hundreds of millions and billions, it's not integration, even with calendars and everything. That, well, that's what I mean with professional tools. When we talk about hundreds of millions of users, it's really gonna make a decision that, uh, unite, first of all, a good model of understanding when you get a quick answer when you're long, yeah, when Looking for ten minutes when more or like you said now, when there's a man around you, he helps you do some small things. Because for the usual users, they don't have a question of how to deal with the rally, the mixing of the rally, there or... That's c-
It's still someplace that's a case. You know, everyone in there has calendars of menedit, it's complicated. No, you don't agree?
Look, I've always worked with a very complicated calendar in my life. I have a lot of assistants working on this subject. I tried a lot of things when the assistants were doing when I was driving. I think that this task, it will be addressed by separate tools and separate torsoes.
Yeah, yeah, sure. It's just, a rally, some approach should be.
Yeah, but just talking about the real use of artificial intelligence in my life, it's not the appointment, is it? My day, I don't spend 10 per cent of the time on appointment. I mean, I always get out of that I'm spending time with a meeting, there's a percentage of interest, right? Well, that's the volume, the lot. And the important story of the other people's decision, especially for the majority of people, they're told what time it's all going to be. A lot of people just say, "You should come at this time." A man just builds up at this time. It's an understanding, a question of understanding the context. I've been up on a phone call, for example, and I've been called. You waited because I know that phone call I'll pick up anyway. I mean, this man, he's just calling me because I'm probably nine out of ten, I'm gonna pick him up anytime. Just as I call him, he's a huge believer to me-- just with a huge probability, he'll raise me or he'll call me back instantly. And there are people who, uh, I'm not really gonna raise, maybe not call back.
I want to add one of the topics you said important about integration. I think it's very important to integrate. People don't notice her, we don't know where she's going to lead. Grok has been integrated into Telegram for the subscription bonus, and now it has become free for the subscription bonus. Again, there's no expanded reasoning functions, there's no expanded deep research functions, but no, well, some superficials, but it's generally available. What are they writing? I don't understand how it works inside, but they write that a sabot can actually scan posts, like X and search results. And, for example, if you're sitting in a group chat in Telegram, you can ask Grok about fresh news or ask him to find some kind of code of error there, and he'll run on the relevant data. It's like building a searcher or an expert right in the massager. They have a statement like that, look, yes, if you see, they have at least that statement. How is this statement real, how real and how is such a colonization possible? That's cool. Well, that's a question, isn't it? So, like Telegram, they're gonna go to the design of their model or, like, what's Meta doing? Meta is still building his model in WhatsApp, right? Or the system, uh, will be used, like Telegram, if Telegram is closely integrated with Grok, we'll see, for example, that Telegram, let's just say, a billion users, more than a billion users. All over the world. This Grook bot, there's a small audience, there's only six hundred and twelve thousand, I've been watching today, monthly users. I mean, well, that's nothing. But for ordinary people, I think it's a very strong move forward when they can get through, especially in countries where they're not available. I mean, it's a very useful thing in countries where it's not available, it's not a good model, cool systems, hard to set, hard to download, and people don't understand what it is.
And since we said about Elon Musk and we talked about OpenAI, you want to sound like a voice, you know? OpenAI has a round of $40 billion. It's the top of the investor's interest, right? Three hundred billion dollars. So it's a crup-- one, I don't know, the largest or the largest in the, uh, the current market. So how are they planning? SoftBank is projected to drop seven and a half billion dollars, then he'll make it to 30. And the other investors, the different, the funds, they'll be able to reach another ten billion, right? And that certainly shows, well, a very serious game. Well, still, the amounts are very large. And to say, I didn't say for nothing that the trial of Ilona Mask was canceled.
Elon Musk, remember, we were just telling a few episodes back, he went to court, said, "Stop the switch of OpenAI to a commercial organization." The court rejected the suspension, but appointed that they would not delay the consideration of the matter. So they'll keep going, there, between themselves, to judge. It's probably just some kind of cost, cost. Here. What about the investment? Sas, last time and last time, we said that investment continued to grow, despite everything.
The market has been in American past weeks, and the OpenAI ballast is crazy, as it is, to other companies.
You're asking me about the private or public market?
Yes, private and public.
Is someone gonna be on the stand? We-- yeah, we talked Antropich very much, didn't we?
Google and Amazon are now very accessible. Google and Amazon are very accessible. XAI freezed off X. It's a very good, well, like the subject you have and, well, it's a trend like that on the AI company, yeah. I mean...
Well, XAI's frozen with X, I think it's a technical deal, right? Because, according to all the speeches, it seemed like XAI-- I actually thought XAI was worth so much because there was X.
Yeah, I thought so, too, yeah.
Yeah, I mean, it's really weird, right? They say X is worth... Actually, if you're gonna deal with this deal, this deal is a little weird in the media, right? So when XAI said it was worth what they said it was worth sixty-five or eighty-five? How much? Some kind of thing, you know, some kind of figure, right?
Eighty.
Yeah, eighty, yeah. Like XAI, it's worth $80,000. And, well, you think, how could XAI cost $80,000 if they, well, how is it possible at the start, huh? Because I think it's because there's X. Well, obviously, there's a Twitter with a huge number of users, a super-used system, an incredibly popular, unique format, totally, totally unique format. Well, that's why I should. They come here and say, "We've created a new company that could cost more than 100 yards." You're like XAI, and this is X together and a bunch of Internet-based covers, like Ilon Mac, who bought a deal with Ilon Masque and Elon Musk. That's where three Ilona Mask is. It's in the light of current developments, in the light of the current events, Ilona Mask and the problems, you know, his personality discussions, I think he's had a very big negative background in the past year. It's just crazy, some crazy background around the world, right? Well, especially in countries like Britain, for example, because of its events, I've had a news publication today, and one German sent, in German, they, uh, send peddata, and they're there. This one, this one, where three, three Ilona Mask is worth. Here. It looks great, great.
Our release was dirty. What? Who knows, shorter, those who spilled, they know, but they don't watch our channel because they're not Russian-speaking.
What's so dirty? What's so dirty?
Well, Luma blew up like a model with a security camera. And, yes, we have, like, the last few weeks of our work on the security camera. So he's the best one on the market and Luma has to pull a piece of traffic on himself. And it's not a beau-- well, it's kind of like, I think it was intentional. And if it's about-- well, it's not lucky it matched Runway.
Tell me, can you just say a few words that a security camera is so that all people understand?
Yeah. There are all these drones, drones, drones, a range of cells, effects, trenches used, cams, Zuma shares, etc. Mausin's effects. That's all we can put in. Maybe we can just put it in there, rig it up, 20 seconds into our video to see. Yeah. Yeah. I mean, yeah, we have a better model than Runway, but that's because the competitors are like--
So, as deliberately and accidentally, they released yesterday. It was a little bit of a blurry, but, nonetheless, the fodbeck is very positive.
Sasha, you're not gonna believe this! I open Runway this morning or there last night, I'm already confused, their demonstration. They say they have a super cool move. I'm looking at their rolls, and I think, wait a second. So, Higgsfield, you showed up there three or so many days ago, or how much, or when, like, when you were sitting yesterday or the day before yesterday, when you were throwing the news, you were video?
I mean, I think I am.
When did you let me out?
Three or four days, yes. Well, we did it on Monday, but it was all set on Wednesday, naturally.
Yeah. Well, you threw, you threw, so the video, I remember, it looked really, really serious. What are they showing me in Runway? That's what we're gonna add and explain. This whole video-generation thing is important. I can honestly tell you that what I saw that Higgsfield released, and yet I can't kill Runway, and I can see what Runway gave out, I can't believe it. I've already told myself, I think it's time to start using video generators and put it in the riles, in the shorts. I can tell you that. So with the graduation, with the exit, what Sasha showed me, I must have had time. How do you get to Sasha, so you can use a model? That's why we're gonna do this and talk about it in the next issue. See you again!