By putting a guy on the Internet is very strong, it's not safe.
That's what she was trying to stop this OpenClou, and he was completely uncontrollable, he lived his life.
As for me personally, there is a zone, I'm willing to take a risk to get some effectiveness.
When the OpenClou or similar systems are installed on the top, the failure or uneducatedness of a staff member may result in data leakage.
We're going to continue our graduations and today is our anniversary graduation. A hundred-year-old, you know, except for a few more special subjects we had, yeah. So we're waiting for a lot of liquors, support, comments. And by the way, do I know what I've been thinking? It would be great for people to write that in our releases they learned super new things for themselves, that is, that they were allowed to do something really bad. Well, that's how Tanya, for example, even started the OpenClou. And we'll talk about OpenClou and the agents today. And we'll start with that, right? Or a version of it. Maybe she's already been resigned. But write what you've learned. It's gonna be very interesting. It's in years, almost two years, that you've been directly influenced and that you've been in your life. Here. Happy graduation, everyone!
Well, let's start today with a subject we haven't had the chance to discuss last time. What is relevant to agents, especially given the update. Again. In all models, it's a month of sounding that there are agents. Everybody's talking about it. I have repeatedly expressed the view that I believe that agents are working and working on some other scenarios than is now shown. But I still want to explain some difference to the end, so everyone can understand from the point of view of the agency work and what the OpenClou debate has been on in recent weeks, what it is, yes, to hear everything. Because it's not a model, and some shell, there, agents' movement. And such systems are likely to be still, or possibly, used and will not and will be used directly by models. But I'll give you some explanation. So, a brief explanation. So OpenClou is the shell that you put on, you set it up. The Carpathians wrote about it. He came to the store, he was, he bought himself Apple Mini. He says, "Why is everybody buying Apple Mini now?" He says, "All the Apple Mini buys to put it, then the OpenClou is going to be up there." So, like, "Something" is, I think, "Cremniova--
She wasn't even available, yes.
Yeah, I think history--
I should have ordered it
in advance.
It seems like a story of some Silicon Valley. Again, with Apple Mini, the computer is powerful enough to start work on models or agents, yes. So, well, that's a way to do it on Mac Studio. Logic, huh? There's all Mac devices, well, they're pretty good for that. Anyway, why did he buy Mini alone? He bought Mini separately to start this whole thing with agents, not to his general infrastructure, to his daily programmes he uses, daily files, to his operating system. Everything inside her and her computer is like some, some sort of isolated environment. So you put this dummy on, you build it, and basically, this dummy, she lives on your computer again, she's an important aspect - she's receiving messages from the Messenger. So you can write it in different massagers. She keeps working different, there, files, memory, etc. And he can call tools different, there, browser, do different things, I don't know, Slack, Discord. And then she's basically, in the case of her brain, the main one here to understand, you're connecting it to the various agency systems, you can connect it to OpenAI, you can connect to Anthropic, You can connect to Google and so on. Yeah, you're-- or you're gonna put her in a local setting. You can connect it to models that cloud-- online, cloud- or localized Ollama, vLM, and so on. So, on the computer inside, you know, for the micro there's some sort of thing, it's like, to start a certain act there. You can tell me to make the model for the agent to choose. Like, you tell an agent something, there, every day, check the volume, there, my shares, and I don't know, tell me about some risks. And he might be in the massage to tell you that. And what kind of agent he's gonna use, he can choose himself. Or you can build, there, use these agents through this system and so on. And it's already on API and it's working on API again, yeah, and it works with agents inside. So if we're talking about Codex, OpenAI Codex, that's an annex that's on the Mac. They're on the Windows now. A few days ago, it went to Windows, yeah. Congratulations, everyone who has Windows! So you can start an agent, build agents directly on your computer. It is clear that all that is, the Codex is very much linked to the programming. I mean, I'm just gonna get off the programming area, yeah. And I think the programmers are gonna figure it out. Many of them look at our channel, but still. And to go to the real zone, like, the domestic stuff, like, "I'll put myself in the OpenClou." It's a very specialized action, isn't it? So I'll put some agents in there to do something to me, yeah, and inside. We can go in the Codex. And there is an environment in the Codex, and this annex is placed separately. They're very often pgrims, by the way, I'm saying that I think the Codex has almost every day some pepheids, releases, and I'm getting an infinite update. I come in all the time, there's a button of peddith in there, and I think he's just a little bit of a sub- suspension-- he's hanging my system. And I've had a feeling lately that he's always reading something, reading it, or looking at something, some files or some kind of data. I don't trust the apple codex. So, GPT chat with the tapes, there, with all the enseabscribs from everything, there, I still trust the training. I don't trust the Codex. I think it's a law-making thing that works inside, but it's a special environment, a special application where you can start these and build a self-identified agent, yeah, inside. Write up who tried to make OpenClo. Now, Ildar, you're gonna keep going on some part, right? Who tried who tried to use the OpenClo, what did you do?
The most important thing is, there's a story here, look. There is, uh, a very strong online guy that it's not safe, is it? And now we have a discussion before the tape. Tanya was just saying that someone told her it wasn't safe there, and so on. But it's safe, it's not safe - it's a very relative zone. When you keep the emails in Google, it's dangerous or not dangerous, right? When you're just on the computer, it's dangerous, isn't it? once in two thousand tenths started running Megaplan. It was the biggest Software sales company as a Service was a cloud repository at the time. And we have the most important story that other companies should have shown, because cloud services, well, they were very small. Well, what happened? There was a post office, a parole or a storage facility, a data repository, but not a system to record, for example, your tasks, the system's record. It was all so offline, wasn't it? And you should have told people why the exposed vault was safer. I've always said, "Trust me, the servers we rent are much more serious than your servers that you have in your office. And hacking these servers is much more difficult or easy to take." Well, they were still afraid that some special services would come and get access. I say, "Believe you servers in the office are much easier to access than to get and get a piece of information in our servers." By the way, I'll notice that, uh, within my 2,000-tenth board, there was no request from any service to give us any piece of information. Although we've got tens of thousands of companies, right? And there was no request for that. Now, maybe there are some requests, I don't know. I'm not running cloud companies in Russia. In the theo-- in theory, I think they are. Maybe not. Here. But we'll move back to OpenClo.
So, with security, Sasha, is he safe or not, in your opinion?
Sasha was, by the way, one of the five commentators that told me to be careful. He never wrote me comments in the dentist, but he didn't.
I wrote to you, Tanya, yes. I wrote you, didn't I, in the dentist?
what company.
That's what she tried to stop this OpenClo, and he was completely uncontrollable, and he lived his life. I had this, by the way, too, but I thought that just my programming gaps were making me feel better. But it turns out even people who can use it--
There was a case, and now Ildar will tell us that Meta was a real case. It was a little bit of a case when the system started to delete the letters, right? She started removing the letters. She didn't just stop there, and the system took and erased some kind of data. Po-o-o-o-o-o
Took and erased some kind of data. Answering, Illar, in answering your question, my personal opinion is that, and in general, the vast number of systems I use every day are not safe. And, uh, there's always a little something personal for me, there's a zone, if I'm willing to take a risk to get some effectiveness and what risk I'm taking. So, for example, am I openclaw ready to start my own banking system? No. But I started, for example, an agent in ChatGPT a year ago in the Stripe system, like, pay-pays, because in the Stripe, I started, in fact, what could he do? Well, the most he could have come in theory is to make a man a return, for example. But to make a return, you have to confirm it through the M.S. Well, I don't think so. And that's why-- but I started it because I had to analyze the amount of data there. I wanted to figure out how the system works. There was a scout where there was an isolated contours. Why not? I mean, that's a question of efficiency. People are starting their mail infinitely now, and I want, and then they started it a year ago. I mean, here's the post office, I think it's hell to start an agent of anyone, anyone at all. Actually, from the post office perspective, I think that there's a lot of advertising now, that box that means that you get your mail, shares it with important and important, important things that show you, and not important, answering or preparing some of the summit. You guys know what, well, we're actually actually thinking what this is? What's the difference between important and important? What's the answer? I have some people who send me the M.S.A., and they're gonna make it more cool, and they're gonna put the M.O.S. in ChatGPT, right?
I'm the one person that sent me these SMSs, I told him, "I'm not answering next time." And now that I've been writing a new textbook, he asked for some help, he wrote to me, "I--it's not ChatGPT."
I'm telling him, "It's obvious it's not ChatGPT right away, right? I mean, I didn't diagnose because ChatGPT is bad. I saw it wasn't you right away. Well, I can see that right away. Such a dead system is absolutely. It's not you, it's not your slang, it's not your style, it's not your details. There's no problem with any kind of mistakes to be there or how to do it. Question is, what style are you writing? So, is the story of Openclaw safe? Well, if you put him in the mail, it's not safe if you tell him yourself, deciding that removing something from my vault is not safe. I have a problem in Zuma, I have a lot of records in Zuma, and I'm filling up my personal vault, that's personal, not corporate Zoom, but personal. And need him-- they made him special so it wouldn't be convenient to run the money off you. Well, that's a clear story. And I wanted to build an agent who would walk and decide that to remove what records, which are left, to keep on the CD. But I understand clearly that I will not risk giving him this task because I have records that are of a phenomenal importance to me, and I need to keep them for years, right? And I don't want to get into a situation where the agent is, like, "Oh, I just decided." Or he'll send a message to Telegram or to WhatsApp to some people who, uh, weird. I got a message from one of the creators of one of Russia's big IT companies the other day, so I'll tell you. He sends me a message from a man. I know him, well, he did, somehow, he asked me to do something. He's sending me, "Sashi, hi! And you couldn't profess, but my own vein in Telegram, and I'll make you a repost for that." I think, "Eh." Well, there's a certain environment of people where, well, they don't ask me about these things. I'll never ask anyone in my life for that. I mean, I can ask close people, right? But it's weird in such circumstances to ask me. I understand that this agent writes, that's what the agent's going through his entire record book. Well, because you have to be a very strange man to stop this man from writing this request, because at this point you're very likely to get into a ban and never write anything again, and there, there, You'll get into this area. Not much, you're still a man with some company knowledge. I'm not gonna call her in the toast, but I'm your partner or someone else's gonna say, "Look, he's using agents there, and they're writing me that shit." That's what we're going on. So we see a huge amount of spam in Instagram, we see automatic comments on YouTube and Telegram. We, by the way, have a man on YouTube who commentes, like, like, putting real comments, but there's always a signature? This comment is left, left by the Committee via Perplexity. And he-- my team, so when they sent me at the beginning, my team asked me a question, and there was a discussion, and that's what it is? Well, it's me-- we started talking, it's a robot that walks like that, or it's a man who looks, but he's asking, writing to a browser, and he's leaving his comments carefully because the comments are like a man, well, sat, not a man. I know there was a meditation, and I was watching video. Well, that's the real comments. Or it's just robots that, you know, walk around and make themselves a certain audience.
I mean, that's why, of course...
Robots that meditate.
Yeah, yeah, yeah, yeah. Which, uh, like, kind of looked, looked and responded to the comment, and you can sign on it. There are many of these, by the way, in Telegram. I mean, it's a swamp. Telegram became a bolot of automatic commentaries. Well, look what they're doing. They sign for you and sign the second Botov account. Then you remove it. When you deleted this, they saw what was removed, they added a new one and those you're going to be given, you remove it, and you're just having an endless auto-station in the comments. I mean, these beans factories, this is the stupidest thing to do. So, Illnar, in answer to the question, and, uh, systems like that when you're, when-- all about API, that's the models that history is, of course, uh, well, you should understand the risk area. Yeah? So you must understand that your data are really available to anyone and so on. That's what's bothering me, adding to the massagers. So in the massage, all the things that concern the m-- the dissemination of my information to the people that are outside are total prohibitions. I mean, I wouldn't give it a categorical way. Everything about the files, well, work. Well, can I be hacked individually by protocol? If they want to break me in, I'm getting hacked. I mean, I'm probably having this kind of...
even though Apple's device isn't easy to hack, but still, right?
I'd be here, uh, for my part, to break the question into two parts. Yeah, there's security of the agents' systems. Well, when we give access to an agent, and he's wrong, I don't know if there's any message to send or something. There's a problem like a sophth maker. I mean, it's a condition, and when we say we have Windows from Microsoft or there, I don't know, macOS from Apple, or some app is written by a big company with a big story, with big budgets, and so on. It's a story. And when it's actually some kind of big deal, even though it's a plus-minus opensor, it's looking at a number of people and so on, but it's like, um, nobody's gonna lose anything, If it turns out to be a bench, like, yes, there, or something, I'm just a little less trusted here. It's like giving examples. Now you can order a major company with big budgets, with a big name, with a big story, and you can find some guys that, I don't know, they'll paint you walls, they'll stick around. And if there's any problem, you've got a big company in one case, which you're probably gonna solve the question there in, uh, in the legal key, probably there's a plus-minus deal. It's a schedule, it'll be clear what happened and what to do in this situation. And in another situation, well, some guys who, well, ran away at best, and you're kind of stuck with the underdressed repair.
It's a story. Uh, well, I see her, right? What, on the one hand, if you have OpenAI, Gemini or anyone else in this system is intact, we're all gonna say that's what a great idea was. But trust in these creators, and, well, much more than some interesting project that just fired at the moment. If he proves his viability, he'll actually move just from this league to a big company league. Yeah, they're either bought or that's how we end up last-minute, and although I don't think we've had time to say it on the tape, OpenAI's got the job of the OpenCloud creator. Accordingly, it may occur that the system will be redesigned within the OpenAI, and soon we will know or update the code--
Guaranteed. Look, it's guaranteed.
Either there's some other OpenAI Clo, yes, there, or something. Here. And then, accordingly, this system will be much easier to build. There's no need to get some hacker teams out of the terminal. And it's gonna be a plus-minus working out of the box. Either that's gonna be the enthusiastes that tried, who like it crazy, who's willing to buy a Mac mini just to put some system in there and try, I'm not. I know, getting weather alerts or some other interesting events, happy that automation works, and so on. I mean, I'd be here for two planes, right? So one story is related to agent behavior. And, yes, there's a lot of problems that don't really understand how to deal with it. And you and Max talked about it a little bit about it in the security video. And the second story is about who the PO is, isn't it? So who did it? If it's a big company
And if it's a big company, I still have a lot more trust.
I, by the way, after I spent so much time, power and all the rest, mental energy on this plant, I open this dashboard and I think, what am I gonna do with it? What's all this for? Well, I'm on some crazy enthusiasm because everyone talked about it. And I even looked at some of the skates on YouTube to see how people used it. But honestly, there was no case of a case that would have caught me. There's someone there about animals, when they drive them to the vet, what their sick history is, something like that. Well, somehow I-- no, I'm even leaning on how to put it all in, all that data, so what? To get reminders. Then, yeah, well, mostly medicine, some sort of analysis of large data sets. But then I was watching an interview with one of my favorite designers in L.A., and she said she had a boutique, a studio, 60 people working, they hired a separate person at OpenClou. So she sees the potential that they even took, which means she probably didn't want to dig like me, in those terminals and codes. And in general, I was just wondering how they used to employ even a separate person. But she, for example, led to the alleged great discovery of this OpenClou, very quickly, but there's a high level of design, I just don't believe that he's really cool, I'm sorry, because I tried, I got a lot of vomiting.
You know, once again, you don't find OpenClou to understand. The point is, in this case, the cool models are different, but they're real.
Yeah. And when you go in there thinking, you start with a heavy regime,
Yeah, that's what actually
works.
The thing is, hiring a man on OpenClou, why I said it was some weird idea, because it was just some kind of softening. It's important to have people here, for example, who build up a great variety of processes. That's good. At the same time, to make a man do different things, he needs to be cleared up. And in fact, you need to be yourself, the people who are already in the company, if there are fifty different people, and then you can figure it out. Should we hire a man who's gonna build some server there or raise or some API connect? Okay, people like that can be done, but it's not the man who's built you on the job of the agents themselves. Because honestly, people who can describe these business scenarios properly, do, you know, business logic all, they've been very little in the world, yes, people who can describe companies properly. Automation. And here we have to try to figure this out, or it's just a mess. Well, at least what I see people doing this is just like phenoes are like, figs. So people do some stuff. Seriously, you're actually, basically, maybe for the sake of OpenClou. You should have taken the same on the Codex app, because you wouldn't have to build anything and ask him different tasks.
I realized that later.
You wouldn't have gotten a massager. Look, you wouldn't have gotten a massager, you wouldn't have had the opportunity.
I never got it. He hasn't been in touch. Maybe it's for the best. You know, I thought, "Oh, my God, why would I need him to get access to all my contacts?" Speaking of which, the second moment I was surprised at this interview, she said that the OpenClou was just sorting her mail, and she had a client level that I think would be very uncomfortable with their letters being sent. to some agent.
No, it's okay. Look.
There's prices and everything, addresses.
No, look, from the point of view of analysis and automation, it's perfectly normal if the post office is handled, structured around here. The question is, people go on-- one thing you're structuring the mail, getting it, like, customer data, and you've got a background that doesn't have fifty letters from one client, but you've got a file. Everything is in one place, like, structured, and you could ask any questions on any client. It's a great story. Another point: Who determines the importance of, for example, letters? Who, who are these letters for reading? It is important to understand that these agents, including hallucinating, are important to understand that these agents, which make mistakes, among others. It's important to understand how much, you know, this story-- this is all the time people understand how this system works, right? And that's how you started, for example, putting it, and in fact, it's enough for the audience to hear, you don't know the difference between what agent the Antropic or OpenAI, or Gemini. Well, let's be honest. You don't know.
No, of course.
From the point of view of him there.
I started writing again in my comments that I thought Gemini, but I thought, well, that's enough experiments. They're the developers in all the companies.
How do they know? Because they don't know your trials. The thing is, you need to know your processes, your system, what you're gonna do, what kind of person you're gonna do, how you're gonna support. I mean, from your point of view, even if you wanted to try an agent, you know, try an agent in your standard window because it's gonna be a little fun for you, probably, yeah. And start some story, make sure he, for example, collects world news on design every day, gives you a
report.
That's the first thing I've done. But then Ilnaru told me that I don't have time to read it all.
Or every day or every day, he gave you, he was looking for, like-- Aaaa, look, and when there's no time, the system is still interesting that he can collect every day, and then, once a week, another agent, that information. You, you're gonna make the most important of this. And once a month more important, you'll have to pick up...
Yeah, that was more work.
Like what? Like what? Or the other story is you, for example, saying, "Take me, for example, interesting objects that have produced these fifty designers or mentioning these fifty designers in the world." Because you can't read all the media, that's how you put it, I don't know, there, the new art director of Gucci, uh, showed him his new office, there, in his office, the main, right? And, uh, you say, "I want, and there are three people in the most famous, or there, creative directors in there, top 20 designer or top 30 or top of the 50th world. If they appear in the media or they write something interesting to inform me." And interesting or uninterested, you and I will do a test in the next two weeks, you'll send me. I'll say it's interesting, it's not interesting, it's interesting, it's not interesting. And you'll learn, right? And you've somehow put it in the code, made yourself--
Back up, turn your OpenClou on.
No! OpenClou, don't turn it on, Tanya, OpenClou.
You've given me so many ideas.
How do I know these ideas, Tanya? Look, I know these ideas not because I know technically how to do it, but because I do-- I understand less than you do. I'm good enough, well, with a good horizon and knowledge. I can offer you some interesting case files that you can help in your life. And there are so many. Look, there's a lot of these. And here's your job to start with the briefcases as you can see. And then, people who really understand these briefcases can advise you personally that you're the best person to use the instrument. Because the counselors who don't understand you from a man's perspective, what you won't build, there's a system. It's how Ilnar and I laughed that Carpathian wrote, "Oh, it's easy to programme it now. I've been sitting here, I've been in for 30 minutes." Well, it's like, once more, how do I, uh, report to PNL for anyone, right? Even for which the report is not built, even for the biggest. I mean, I looked at four hundred chisel, and you've got a problem here. It's not gonna do it faster than ChatGPT, is it? And that's why this story is, uh, from the people who advise you. But here we must all be careful. Also, tell me what you used to use, what you built, who did it, couldn't make it.
I think this is a very, very important subject. Yes, Ilnar, you wanted to, you see, yes.
About the agent-assed mailing. I'd be here, Tanya, and I'd be there for you. If I'm going to some big agency, especially some big deal about, I don't know, there's a half a million dollar repairs, discussing time--
There's more than half a million.
Yes, especially. I mean, I'm half a million, and I'm--
Yeah, in America, the other market, and there's all your personal data, all your kids' personal data, a bunch of, you know, a client's card is made up of how he lives.
So they're in the mail, they're in the mail, in a corporate one that sees a lot of people.
Well, corporate mail is usually somehow protected, right?
But you have an Enterprise version, Enterprise Model version is also protected.
Sas, when I send a letter somewhere, I'm kind of in the area of responsibility for this mailbox leaving it. And if, not to mention, the OpenClou is being held up on top of it, I think it's not very fair of me as a client, that I sent the data, assuming they're going to be processed--
I'm assuming they're being processed by a man on this mailbox.
No, I-Eldar, you're campaigning when you're running a campaign, you're in the sense of even confidentiality, probably you're usually in this campaign, and this campaign keeps your data in the box. Depending on the laws of the country in which there is. Everyone here can write what they think inside, but you're even legally in possession of data, and these data can be processed by different people completely.
I see. But look, when the OpenCL or similar systems are on the top, it may be that the failure or uneducatedness of a staff member will lead to a leak. It's clear that this leak is likely to be a good company to swing. But how would it be a problem we'll deal with later? I would not like this problem to arise in principle. And when I just sent a letter to the mail, I did, well, that's how I feel about this story, and then, uh, Tanya, you said that some very big agency is handling agents. A system that, well, for now, so, let's just say, just got out of the shop and then accidentally.
I still have this feeling so mixed.
I have a question, too.
Wait again, Ildar, and tell me, please, to join Gemini's post office. There's Gemini's analysis inside the Gmail. Is that not gonna trigger you? That's the same thing.
Not exactly the same. So we're talking about OpenCL. I don't know how he was built, who was building it, what was done, what models under the hood were. Maybe there's a cheap classifier hanging out with some model that I don't know is sitting somewhere on a Chinese server or something.
I heard you, I heard you. But I want to make sure that, in the world today, ninety-five percent of the companies you write in terms of data confidentiality and data security, there's a hole in it.
I don't have any questions. Sasha, there's no question. Just when we're talking about some cool agency and some level of service.
And the cool agency is definitely a hole because they're cool design agencies, they're programming, but they're not safe.
They hired a specialist.
Anyway, my opinion is on that.
And they gave this specialist the opportunity to make the summit of all the contracts, all the clients and the base. It's a story. It's like a system manager at a company that builds mail, has the opportunity to read the entire postal of the corporation. Very often, yes. Everybody! Not everyone, of course, there's not super-sized companies, and so on. But in general, these access is very often available, and people don't even understand it. They don't understand, they don't think that, and they don't understand who they're dealing with. This is a very interesting subject. Write your opinion on accessibility, accessibility of information. That's why I have friends. Here's the question of efficiency. Everyone always chooses, right? And I have friends who are so encrypted, so they're with the pipes, I think with the phones, with the nightlines, they're going with the phones, so God forbid, nobody gets the data. So everyone has their level, that's what they're at, that's the scale.
That's how Eldar says, Gemini is not against, but OpenAI, Open Clone.
I have a friend who still uses fax, and it's hard to buy. He's not sending anything on e-mail, just faxes.
It's okay. No, Eldar, you're absolutely, you're very reasonable. You say that this is a project made, it's not clear where he is, and how he is, and what it's done, and so on. But it's like we buy the services, leave plastic cards on the subscriptions often, which are not super-known. Yeah, you're staying, right? You're staying. It is simply that people who, for example, from Eastern Europe or, for example, from Russia, from Belarus, are less likely to leave a plastic card in other services that are less known. In America, people, you know, they're, like, left, they don't care, do they? Again, because of the financial system, yes, I've already told you that money can be returned, yes, if you haven't written anything. But at the same time, these are episodes of approach, that is the question of approach, thinking, there, access, stereotypes. Most people in the world, of course, agree-- everywhere she clicks just a glass, where she doesn't know where it is. And we know what this is about from the perspective of Europe, yes. It's like keeping cookies. In Europe, there is a law on the protection of personal data. You can't keep anything there. And accordingly, there's a lot of automation in business, you can't automate anything. Yeah, like you can't analyze the phones. In fact, it's really hard to analyze calls in general. The same is a very interesting topic, as all countries make different decisions. By the way, we had a special edition with Max about AI Safety, and there's this subject, including a very high profile, where this is the security line. And if someone made the system unsafe, does that other company have to do this unsafe system so that the Sam-Same can do something safer in the future or not? Write it. By the way, who was watching the graduation, there's a lot of things. If you didn't, look at the graduation. I think it's incredible. Let's talk--Ildar, there's something to add to the OpenCL topic, technically, maybe something, some ideas to tell and...
No, we've even been discussing how it works under the hood. And now you've told me that very well.
I think it's just that you can wait in the comments. Some of our viewers see, use themselves because someone wrote that they're making staff, which is 24 to seven, something else. It's very interesting to read that it's already real and has some benefit.
Because on the Internet and the description of these companies, there are often cabs that are totally deranged. For example, ChatGPT is offering again the case files to schedule meetings for me with someone else. Like it's faster and better, yes, in terms of the appointment of meetings. I've already said it sometimes seems very strange, yes, who you're writing with. The second story is ChatGPT history, like the case, so much as a cheque-in on the plane. I'm actually honest with you, I think who the briefcases are offering? It's like ordering sushi. I'm always cutting this case. So order sushi at the store, get a check-in. So it feels like a man is flying eight hundred flights a year, and he needs an automated agent who's going through the check-in without thinking that when you're actually going through a check-in robot, a robot can. Make the ticket back. To cancel, for example, the ticket is complete. And how he chooses, it means you're sitting, places, details. So, and especially for people who don't often fly, for example,? Or you fly every two weeks. I'm a little too often-- I can't say that very often, but I fly a year, there, dozens of times. I wouldn't give the agent any carpets to make me check-in and do my tickets because, well, it's a crazy risk, just incredible. It's gonna be interesting to see the cabs, right? As long as all these cabs on the Internet, which often fly, they're very specific, especially Eldar, if you're talking about 24 hours a day, a real staff member who's actually been effective. By the way, the study was very interesting. There's a subject to replace staff, right? And the study came out that I also met there today, and we were talking about how I think people have become more involved with the emergence of artificial intelligence. And this study just said that people were more and less working. And not that they didn't replace anyone. So everyone worked a lot. I got artificial intelligence, everybody's doing more. I mean, obviously, every day of business... I have a way to do more than before. But less I'm not working in these tasks, and I'm different, because I have a lot of business, projects, I've been differently distributed. I may have a real distribution of efficiency. But I see people who even, well, connected, I've got teams working or I've seen other business, but they just got some tools, they've got ten more X and more in there. These data are stored and operated. And so far, these staff are twenty-four by seven. Well, even a programmer, he's writing a code, he's a... no, I know there are those programmers who are working on six jobs now. Remember, Eldar, we were discussing when the pandemic started, people who were marquetologists and programmers on three, two, three, four jobs. Are there any such programmers now? Of course I do.
There are people who are burying money because they say--
They're clearly more.
They're clearly more than that, yes, they're writing the code and everything. I mean, these people, of course, have become a lot. But it's not a question of replacing people, is it? It's a question of just a few.
Sasha, I'm laughing half a day. And now, I'm laughing again. We'll add that little skate that the kids saw at work today.
Yeah, yeah.
It's just, you know, it's a reality that happens. There's a granddadline two hours before work, you didn't make it or you did it, or you were there at the third, fifth, eighth job, if you're on the subject, and two hours before the end, you're trying to get it all out of here with help. Some agent, Claude Code, Codex or something. It's a very life story of the past few years. Well, the last few months.
I'm a little bit of a fly, but I'm sure it's the last few months.
Yeah, yeah, yeah. That's very, of course. I'm here to remember the pro-- that we were doing on Wednesday about security. And we were discussing a story that was that systems would evolve and gradually they could have their own conclusions some--
Uh, your own conclusions are some of us doing what you want to develop, yes. And we have raised the question, in particular: er, if the system, uh, works and is in the process of setting up a system to avoid harming people. And in America, by the way, in OpenAI, uh, OpenAI entered into a contract with the Department of War. Uh, it's clearly spelled out the following condition: not to harm Americans. Don't hurt Americans. I'm gonna go, uh, and if you even consider the bigger stuff AGI does, which isn't hurting people, too if you want to do some bench or some other call to people to do the twenty-eighth, it's gonna be "I won't call this client anymore because we called him four times, we're bothering him. Or last time, this client told him not to call again. I won't call him anymore." So is this system entitled to block and stop doing your actions, right? So, uh, what are we gonna end up with? So, uh, who's thinking what we're gonna end up with? That's a big question. I've been thinking about AGI and OpenAI now. I read the article yesterday, uh, Ilona Mask. Elon Musk loves, uh, remake. And, uh, when people write something about his competitors. And yesterday, they put out a whole post on Sam Altman. And where Sam Altman cheated on people, said he was wrong, told him he was a scherman in the same company, he wasn't a cherrane, there's a ta-ta-ta. Well, that's a lot of stuff. So this same man, uh, well, Ilon Mack, what he did, there was a subject written where, uh, there, uh, three or four different people say, "We wouldn't let Sam Althman press the launch button. AGI." That's very interesting. It's right there. So this is a speech of the twenty-third year, three years ago, where there is a statement that this man cannot be given a button to launch the AG-AGI. So, by the way, AGI and API are going to go on inside and in companies. By the way, when OpenAI started working with the War Department. Illnar, by the way, big companies. What you're talking about Open Cloud and the dissemination of data, uh, is a similar story when, uh, they started working with the War Department. I'll just... I'll explain what I mean. A similar story in what way? Uh, a huge number of people who are very familiar in the world wrote, "We're washing ChatGPT and we're never gonna use OpenAI again." Why? Because they think that this data will go further. I mean, it's a little bit of a mistrust from the system. I mean, like you said, I'm Gemini or Google. I mean, I trust they keep the data. And, uh, it's just like someone trusts whatsApp, someone doesn't trust, someone trusts Telegram, there, and someone trusts someone doesn't trust Telegram. And then, for example, Doura says, "I wouldn't trust you. You should be an idiot to trust whatsApp." I mean, I know people who trust whatsApp, but don't trust Telegram. I mean, this one, this story. Uh, and that's the same thing. There were people who said all the OpenAI was erased, Codex erased.
And on the Internet, this schedule, uh, where people started setting up Gemini and some other system.
Yeah. Ah, Anthropic started setting, yes. It's clear that Anthropic has a much smaller audience. As a man-- a joke to Sam Althman recently: Anthropic Tehas's audience is about peace, about OpenAI. And the OpenAI schedule just went down. It's clear that these are all very relative numbers, because they're not in the l-- quantities league, uh, we-- and real daily OpenAI users, right? Still, with OpenAI, the league is in the league today, Google Gemini. Yeah, that's real. Well, there's a billion Meta. But, uh, that, um, that's... that's very funny, yeah, from the point of view of Meta. By the way, where are their avocados or what they're doing? What fruit or vegetables did they produce? Uh, this, this is-- this is the subject, it's very interesting. Look, she's got it in her.
Again, who didn't see, look at this pro-Tan issue, you didn't get a chance to see that safe thing again? You're gonna love him. Here. You really, yeah, that's your subject, you're gonna love it. Look. I want to see what the topic is today. So ChatGPT released five points three, instant and, uh-a-a, what is this theory? It's a version, uh, quick answers. I mean, ChatGPT is like, uh-- I remember once more that you have the opportunity to choose the versions because many people don't see it. And there's a car, there's a way to think, there's a regime about who the version is. Inside thinking there are different regimes, and it's inside like this thinking model. There are different ways of thinking. Not all interfaces can choose their own mode. Some things are possible, they're not, but they're choosing. And there's a regime called it, somewhere, I think it's been three or four months ago. He's an instant. That's the answers right now. So, uh, there was a renewal. Uh, let's go, uh, five and three, this instant. He's, uh-uh, basically different from his own stories. First, there's less unnecessary, rejections and less preaching. It is, by the way, very funny that the ChatGPT itself is saying that there are fewer sermons and long reservations. And we know everything that often answers, well, weird. You start fighting, you get mad. Uh, I'm thinking about my childhood, like I wanted. I had a big Sony monitor like this. I always wanted to break it when I had two hundred and eighty-sixth Pentium gulp, and it's good he didn't fight. I mean, I've been knocking on him so much, because the tone monitor would have been broken for a long time. I'm looking at my middle son, and he's just like that on the iPad, almost hitting him on the table. And I understand that this is a normal operation. I had a kid when I was a little bit slow. Uh, second story. I don't have a computer that never stops. The second story is less hallucinations. At five three insta and OpenAI, it's about a decline in the hallucinations. First, they share hallucinations on high-profile topics, search hallucinations, web-based, web-based, right? And the high-risk theme is, for example, medicine, law, finance. This is, by the way, a fun subject that is being singled out, uh, well, large companies, in particular OpenAI, are pointing out that hallucinating is more than a concrete topic in high-risk topics. Because one case you were wrong about a subject that, you know, is relatively relative, another case was wrong in medicine. Now, this is the five three Instant story, which is stronger than the letter style, that is, more live-- more art, better art emotional text, without losing clarity.
And the fourth story they're making is that it's better, uh, web search and, uh, richer answers, better tied to the conceptual, less dead ends and a little bit of corn and a little bit of caescality. Well, of course, we're waiting for five three. The Internet says five four. Yeah, the Internet says five four. Uh, but there's a mem on the Internet about why we started discussing five. Because the OpenAI has concluded an agreement with the War Department and has caused this whole surge of everything, just until Sam Althman, to be honest, looks pretty cool. There's a post like this. Sam Althman wrote that he had decided to hold an open conference on the weekends on how-- what they had concluded with the War Department. And he's in the X.O. this message several times for twittile. And he was texted, "Why are you twitting a few times?" And he wrote, "Because I'm sitting on Starlink, and Starlink was slow." And then, everybody was ruined by the Internet of Elon Musk. And he's go-- and he's written, "I'm just worried about that performance. I thought, interesting, Corpo-- it means a corporation where 10,000 people work, a lot of things are being organized. He's worried that people haven't been informed that this conference will be here. It looked really weird. And then, for the next few days, he wrote a post like, "I'm sorry, I guess we had a very big deal with the War Department, and it was like Friday night, and I had to wait there. We've been doing this for the weekend." It all looks like it means that the normal guys are sitting in a room, tea drinks, and then the starttuppers are making up who to make, huh? Although the subject is very serious, well, incredibly serious, incredibly large. And that's what it takes to me that five four are the subject of forgetting what happened in OpenAI in terms of this agreement with the War Department that needs an apdite. We've got a fuck up, and we need to release the pddata, so we're waiting for updates from them. And as we know, when the update comes out of some big players, all the other players are making updates, too. We are therefore waiting for an update from all completely companies in the next two weeks. Yeah, yeah, we're waiting. Well, I don't know anymore. I don't know where Meta was lost because Meta's not heard. Just like I don't know where Apple's lost. I'm here, Sasha, don't start, it's a long subject. No, I've stopped following Apple's updates. Well, I'm sitting here looking. Apple released a new MacBook. I don't care about MacBook, where's my new monitor I'm waiting for? I'm waiting. I open the book, and it says, "The monitor is 30, two inches more of Apple's company. She decided to move in at twenty-seven inches. I looked at the monitor and I think what's the difference? Nothing. And they made me feel bad, didn't they? And so even in the graduation-- by the way, Ilnar, we're even, we're apple pedits, even into the graduation somehow, and we didn't even want to write. What do you want to write? No, what do you want to write? Like MacBook, there's a smaller phone, and the camera's more there. Mm-hmm. The processor's back. What's the difference? Yeah, like a M18. It's in my youth that I know-- we knew, there, all the processors, operational memory, number, name of it, there, DDR two, three, there, RAM, chmm, all those names knew. And, uh, and then I remember very clearly, there's probably two thousand--
Very clear. There's probably two thousand eight, maybe, like, two thousand nine. I started to refuse, well, re-- stopped gradually taking all these additional names because for a normal person they stopped being relevant, well, they're no longer important at all. I'm not go-- and I'm not even here. If it matters to someone, write what Apple surprises you at the last meeting.
Well, then, someone might google Wiki, Andr-Android people say what Apple? There's a Google new phone out here. There's a bunch of incredible pds in there.
That's right. Well, about the OpenAI Anthropic in different leagues, there's a year's worth of cash, you know, how now? There's a present-day, yes, when you take, you're on probation, and you're on the last week, and you're on the way to 48, and you're talking, and I've got so much money in here for a year. These flashes, which occurred with the overtemphasis of people from OpenAI to Aicic, are likely to have an annual earnings of 19 billion, and the latest available data from OpenAI was 25 billion. That is why the splash and because of this formula, the Anthropic is not so far from the main league with the big players.
Well, I thought I didn't count on the manual, yeah.
Yeah, yeah,
yeah, yeah.
I counted it as a number of online users. They still have $800 million plus weekly users, even near Gemini's system of seven hundred, uh, seven hundred million monthly. I told my comrade yesterday, by the way, always look at the numbers that people say. People always choose a number. Someone says, he takes the last month of the proceeds, multiplied it by 12, says, "We have ARR, so it is." Someone's got a profit, someone who's saying a net profit to the annual accounting report, accounting, tax-- after-deposit, full-time account. That's the most reasonable, so the picture. All the rest is just some numbers, abstractions, growth. But, Illart, you said a very interesting subject. Which is amazing that Anthropic, uh, Anthropic in the company-- probably investors, investor money. And in fact, finding is in a very serious priority because, in fact, in volume, they, well, really, really, really, really lag. And, by the magnitude of the project, there's probably money in there. But their last assessment was already there, more than 300 fifty or more billion. They're just a new round, and it's a very big round. And it's very interesting, because it might happen, because if it could happen that when you just get into a league of such money, you still have the opportunity to openAI to the world a cheap start. tariffs, for example, free of charge or open more mass-based models. Here's the question who gives mass, mass market service. I mean, because if you don't give it, if Anthropic is focused, here. Maxim was talking about it again on Sunday about AI safety, who's got a scon-- well, a focus, right? Because the same OpenAI. Can the OpenAI make a four-fold increase? Maybe. He can come now and say Pro's $800 tariff. I'll buy myself, a lot of people buy, right? He can make a $1,000 a month for, uh, $3000 a month. He can raise the usual $20 to, from twenty-five, there's different prices, raise to $50. Maybe? Maybe. They can let them out a lot of things. I just don't think they're interested in getting extra cash, ten or twenty billion. Because if you look at their general economy, the question is a general economy. It's like NVIDIA. NVIDIA shows now, they have the latest reports, they show very good results from AI, there, growth, everything, everything. But the efficiency is endless in terms of money and the investment in money, net profits, there are big questions, right? And that's why we'll see. Indeed, Anthropic, I agree with you absolutely, they will. This is the story, uh, sociopolitical. They even, by the way, have a number of people working. That's the last time I talked to a staff member, there was, like, a couple months ago, yeah, they had a room. They have staff, I don't remember, more than-- a few thousand staff just sitting here, right? The main office was sitting. I'm a number, to be honest, I don't remember that many meetings. And they don't have as much people as they do in the OpenAI. Less, of course. But not, like, know-- not like, not like users, right? I'll remind you that all the Chinese models give users less than half a billion. To understand the volumes, all the models are total. What-- why do I call this number like eight hundred million users of the weekly, what does that matter?
And yet, in a mete, a billion users are illusions. Yeah, these people who, I don't know, in Facebook, maybe some kind of AI button, there, something pressed, yeah, they asked me to do something. Well, that's a big question. Oh, by the way, there are some cool things. Here in X, the chips of the-- are used continuously by the Grok in the comments, in the mete, in the chat rooms. You can invite meta AI inside the group and ask meta AI. But I have a huge number of wats in Watsap Groups. No one ever called Meta AI. And in X, I feel like if I had band chat in X, I'd call X Grok. I don't know why. I mean, I'm wondering, write down who's using, uh, Telegram. Remember, there was this story, too, which, by the way, was a fun subject. Six months or a year ago, Dorov and Telegram said they agreed to connect X, ow, Grok. And then when they wrote that they agreed, Ilon Mack wrote to him, we haven't actually signed anything yet. And Durov wrote, well, we talked. But I'll notice that Teleg-- Grok, the integration of Grok inside Telegram at a level like X is not available. Right? Well, I don't see that, that kind of integration that's the most complete. I mean, I understand she didn't happen. The question, by the way, is very interesting. Why did she happen, why not? That's, by the way, trust, uh, systems. I personally trust whatsApp, not the open chats of Telegram. I trust the tallers of Telegram, but not open. That's the question that's always with. Write it too. Who trusts where to call. Interesting candy to see. Here in groups, in group chat rooms, who uses and who doesn't use it. And Tanya, we have a family chat, right? We, uh, we don't have one family, but our families together have family chat, and we're often texting. We never called Meta AI there, yes, although we might have called there, called inside. I think Tanya, if there were GPT chatting or there was some kind of fun story, we could, we could have called it back from time to time, right? And...
If there were more, you know, a GPT agent that would give us some comments, and some backgrounds, it would
be fun.
Okay. So there's a-- Tanya, there's Meta AI. Meta AI gives a background, that's the whole thing. But somehow, Meta AI doesn't use it. What do you think this is? We've got chat called "Air with Amazon." That funny name that came out, it just...
Thanks to you and me, Sasha...
Yeah, it's coming. And I thought it was fun. We don't get Alex in there, right, in. I guess there's a code that could be written. Now, Open KLO, we need to download and, uh, yeah, and add, uh, look, to put the Open KLO, to download all your husband, my wife and yours chat. And to be personally e-mailed for us, to tell us everything, to what happens.
I have nothing to hide. Let's see who's gonna refuse.
Me too. Let's see who's gonna refuse. Let's see who's gonna refuse. In any case, the arrival of agents and the appearance, and, uh, of such mass dealers inside the group chat, this will happen in the future. And actually, this thing that's in Grok is a real case that's gonna work and be alive. It's just X is still a very specialized audience. Oh, a huge audience from the news standpoint. There are people who think X is the only thing in life. It is not yet planned, it is expanded from a financial perspective, there, with cards. Even yesterday, Elon Musk was there telling them they were planning to create their cards, there to transfer money and so on. So there's some trust in it, that's the Grok thing, yes, the trust of the Ilona Mask system. Let's see what happens in other chat rooms. And by the way, remember, there was a discussion story that GPT was planning on doing its social media, and somewhere she died in their speeches. And the GPT chat, I'll pay attention, sora's got this social media out there, too. Well, she really died there, same where she started, right? Here. Where's Sora now? And, well, we're not wrong the last time. Now, I guess it'll take months, and GPT chatting again will release a new reservoir about Sora. Like Max Grigoriev says, "You're a homoster." Look at the special prom that was there. Our graduation is 100. Don't forget to slide, uh, sign on our channel, send our friends, your friends. We're actively developing, we've got over 30 grand subscribers on the canal.
You're gonna send it.
Yeah, send our friends, too, send them all.
Here. See you later. I'll see you in one hundred first edition in a week. Bye, everyone!