Skip to content
Transcript

Transcript · 068 · The ChatGPT Agent Promises to Act for You—and Immediately Runs Into Control, Marketing, and Safety — ToTheMoon

English machine translation of the Russian-language episode. Timecodes open the source video.

Episode overview
00:00:00–00:00:48To TheMoon graduation today.
Alexander Volchek00:00:01

Hello, everyone! We're on ToTheMoon. Technology news and sites from the Silicon Valley all over the world. Crippa's got people buying and buying, too. The Government will only contract models free of ideological gradient.

Discussion participant00:00:15

I wonder if Chinese models are probably limited by the idea-- by the ideology. It's a wonder how progressive the government is thinking.

Alexander Volchek00:00:27

And it's important to shin from different angles, isn't it? And usually the technology washed without the state involvement, right?

Ilnar Shafigullin00:00:33

DeepMind was already on the math Olympics.

Alexander Volchek00:00:37

International Olympics require another approach. The model didn't work out with the open source. ChatGPT 5 will be released in August.

Mentions: ChatGPT
00:00:48–00:03:00Smartphone from Trump
Alexander Volchek00:00:48

Uh, listen, today are such interesting topics for discussion. But before these topics, uh, this morning, they wrote again that, uh, smartphone, the Trump's smartphone failed. And I think a lot of people haven't seen this subject at all, that Trump has released his smartphone. Have you all seen this story? You saw everything, didn't you?

Discussion participant00:01:09

No.

Alexander Volchek00:01:10

Have you seen it? Sasha, did you see that? You, Sasha.

Mix00:01:13

I saw it, yes. But he's all, everything...

Alexander Volchek00:01:16

Well, it's a memorial. Right? Well, that's what he's doing. Stablecoin, memcoin, phone, stupid thing to do.

Ilnar Shafigullin00:01:25

No, Sasha, maybe you have a really cool picture today, like a camera was just a little straight.

Discussion participant00:01:31

Did you change something? You're so good to see and hear.

Alexander Mashrabov00:01:35

I think it's just Christina's first wiped out in the office and there, you know, and now there's no such thing as this. I mean, I'm more dust than on the camera, I'm more dust than on the camera.

Discussion participant00:01:51

Okay, you guys, you're gonna wipe your screens, please, because Sasha can see the best of everyone. Makes sense.

Alexander Volchek00:02:00

So, the 16th June Trump publicly stated and positioned that the first patriot smartphone would be. And he also said that this smartphone would be made in the U.S. territory. It was literally in a couple of days that they noticed that the phone was not, uh, in the U.S. territory, and that it was a copy of Wingtech's phone, and there was some complex thing with him, Rev 7 Pro 5G, made in China. And then, and formally, it just put a type of brand on top of the top on T-Mobile. But what is important is that it has actually been said that he's gonna make a smartphone, he wanted to do it for his own decade of election campaigns. Well, uh, it's like Sasha says it's like a mem, but it's not a mem, it's a real story. Sasha, by the way, I think Trampe's cricket has people buying and buying for lunch with him, especially. Okay. Come on, Trampe. Ah. So?

Mentions: United States · China
00:03:00–00:04:59New law: AI without ideology, what does that mean?
Alexander Volchek00:03:00

So there's a part I think of that very important thing to discuss. And I want to start with the subject that concerns, uh, that yesterday the White House has put out a list of items, that's sixty, sixty or nine points, they want to move on on the topic that they have a new one. Plan of action in the area of artificial intelligence. And I want to see one very... very remarkable, very unreliable paragraph, which, to be honest, is very unreliable for the public. This paragraph is about the word-- freedom of speech and the ideological neutrality of models. The update--- they want to update federal procurement requirements now, and the government will only contract models, which is very interesting, free from ideological gradient. To be honest, I-- well, that's very interesting, right? Uh, free ideological gradient is, in fact, a model that should not be regulated and should be free of speech. I'm thinking, right? And I am reminded, in particular, that someone wants to close the WhatsApp and the hostesser to introduce, and yet someone lobbys the topic of free models. I think it's a very field-- I don't know how much they're gonna do, and how much it's not a political slogan, but the subject looks very useful. What do you think of this story?

Ilnar Shafigullin00:04:30

I would also know the definition of models that are not subject to ideological evasion. That thing, she's in every country, yeah. And the U.S. has a certain agenda they're moving into. I don't know, I'm just like math without a definition, and it's really hard to say something here. It looks like a political slogan.

Mentions: United States
Discussion participant00:04:52

The interesting Chinese models are probably limited by ideology.

Alexander Volchek00:04:56

No, they're not, they're definitely ideology-- they're very

00:04:59–00:07:04New law: AI without ideology, what does that mean?
Alexander Volchek00:04:59

limited.

Discussion participant00:04:59

Maybe it's like another potential advantage for American models.

Alexander Volchek00:05:08

Well, in general, I think quality will move or not.

Discussion participant00:05:12

Anyway, I think there are some legislative moments that limit, there, no, well, some boundaries that are built by law, for which the model should not even come in, even if it doesn't, freedom of speech and full democracy. I mean, it's close to the slogan, but it's still a little bit of a surprise to how progressive the government is at least in its aspiration.

Ilnar Shafigullin00:05:44

So there's got to be some kind of test bench on the lack of ideological influence. I don't know. So how will that be checked? I'm not sure yet, to be honest.

Alexander Volchek00:05:55

Well, I think Elinar is one of those things where at least you don't have to put in a model, set some limitations. Maybe it's not the way that there's a baccalaureate that checks, or maybe, in the way that there's no law that says you should limit, you should somehow be. I'll write it like China. It's not like Taiwan's not on China's map, is it? That Taiwan Island is not China, for example. I mean, it's some ideology that's certain, and there's some freedom of speech. So the model can say what you think is necessary. The other thing is, there's gonna be more there, what she can say, and that she can't talk. Here. But still, this story, it's here. Sasha, what are you? What do you say in terms of design, how useful is the market? I think it's good again. Your part, how good is this market? As part of the development of artificial intelligence in the country.

Mentions: China
Alexander Mashrabov00:06:53

Well, I mean, it's very difficult here. I mean, when we talk about how-- you can just look at AI and the importance of AI from different angles, right?

00:07:04–00:08:47USA, EU, China: three approaches to AI and ideology management
Alexander Mashrabov00:07:04

It's important to look at the importance of AI, and from some point of view, that's the influence on the minds of the population. Well, like OpenAI, it's obvious that China is already experiencing a lot of restrictions on what the model can say that it can't. And we know that, right? And that's probably why America might be, like, duplicating a mirror approach for Chinese people. And on the other hand, well, as it might be, there's a process, like technology staggered and usually technology stacked without the participation of the State, right? It's in a market format. And that led to the success of American technology and to this day, which, obviously, is a success, well, falling, huh? Because we see, yes, that all the things that can be bragged while the companies are still baccalaureates and the number of GPU online. So I think there are some kind of contentious, different points of view.

Alexander Volchek00:08:35

No, well, on the one hand, there's China, which is clearly embedded in ideology. There's a Euro-Use who wants to not introduce ideology, but all control and access very much.

Mentions: China · United States
00:08:47–00:11:32Code and algorithms control: EU/France cables v XAi / Meta / Google
Alexander Volchek00:08:47

And there's a story that now, where Twitter, X and Elon Musk, they're suing the French government on the subject of what they don't want to reveal to themselves. And there's an interesting moment inside of it, by the way, what, basically, what did France ask for? France requested access to the reference code and parameters of the recommendatory algorithm, for example, yes? Or asked for, uh, well, it's clear they have the possibility of listening and tracking all the details inside. But when they request access to the original code and algorithms and they have the same, they have different requests in both the Alphabet, Google, and Meta, and Apple. They want to control very much what's going on inside, how it says it, how it works. Probably without ideology, China ideology. Now, given this law in America, the rules, I see the rules for the first time, honestly, I see the rules. Look, not just that we're not in the mood for ideology, but we're in the public service to buy systems that are not ideology-free. That's very interesting. Another moment, maybe...

Alexander Mashrabov00:09:55

It's a Chinese model. Sasha, right? It's a Chinese model, right?

Alexander Volchek00:09:58

No, why the Chinese model? Chinese-

Alexander Volchek00:10:00

No, why the Chinese model? The Chinese model makes its ideology in systems initially and buys models only for ideology. Wait, the Chinese buy models with ideology. They're starting to hang models, ideology in. And there are strict requirements. It's just a model that's back. Sasha, there's a model here, and it's the law that's totally wrong. There's a law here that you have freedom of speech and ideological neutrality. So they say we're gonna conclude, the government will contract and support companies that only work models free of ideological evasion. So it means you can't put your ideology into the model or put some demands on it. I think it's back, it's China.

Mentions: China
Alexander Mashrabov00:10:49

Yeah, but since that's a big deal, it's gonna be a one-time thing, right?

Alexander Volchek00:10:55

That's another question.

Alexander Mashrabov00:10:56

He has an understanding of ideology, he has, like, somehow, it's influence. Well, anyway, it's a State's involvement in the process and the advent of some kind of ideology. Everyone has an interpretation of ideologies, yes, which obviously does not fit in between different countries.

Alexander Volchek00:11:12

Well, that's what I was gonna say.

Alexander Mashrabov00:11:13

It's just that it's a Chinese, purely Chinese way.

Alexander Volchek00:11:18

Would you write in the commentaries, what do you think would be interesting in The Moon to raise the subject? What are your topics more interesting and what are the focus? Yeah, that's, yeah, well, maybe interesting.

00:11:32–00:13:41AI-Memeting restrictions
Alexander Volchek00:11:32

I think it's, on the contrary, moving towards the removal of certain restrictions. Now, for example, California is again trying to impose restrictions on AI-Marking, and they're generally, actually, there's a hearing, and they're, uh, one of the main things that chat-bot has to do. to remind at the beginning of the session at least every three hours that he is not human. Well, I mean, just to remind you that you're not human, yeah, you're not working human. That, by the way, is a little resonant because you don't understand at some point in time, man, no. I had a case the other day. I'm standing with my wife and my kids, and I got some video. I say, "Oh, look, there's something going on in nature." My wife says, "This is AI." And the daughter says, "Dad, there's a signature that's AI." I mean, I was completely confused that I was sure it was a real video. And they're both saying, "This is where the mark is worth what it is AI." And the question is, what content will be marked and what is it? California's, like, a mark. But I wonder what you said about ideology. I think it's some kind of freedom, on the other hand. Well, Tanya said it was freedom, too.

Alexander Mashrabov00:12:47

Yeah, me too.

Alexander Volchek00:12:49

Yes, I dare--

Alexander Mashrabov00:12:49

On this side.

Alexander Volchek00:12:51

I think...

Alexander Mashrabov00:12:51

We split, two against two.

Alexander Volchek00:12:52

We split up. Well, because if you see, there are the main other points they're developing, they're very important in the world as a whole. It's the amount of money that is now investing in all the dates centers. There's a fundamental-- one of the basic stories at least is the development of the dates centers, and this is the development of all the factories that make OpenAI, for example, in conjunction with SoftBank. By the way, the OpenAI with SoftBank announced that they had built a lot of things, and parallel to what they were saying, they'd be less aware that it would be more than 500 billion funding, but 100. billions of funding. Kind of. At the same time, it was announced that they had almost built two blocks as two nuclear reactors there.

Mentions: OpenAI
00:13:41–00:17:25AI-Memeting restrictions
Alexander Volchek00:13:41

So let's see what's really going on.

Alexander Mashrabov00:13:41

What I was saying is, it's really super important that Dario is from Anthropic, right? He said they were considering investments from Middle East. And, uh, as Stargate has no money, Americans don't have any money, that's it. We've sort of taken it. We don't know why they went where they went. All right, well, they're not actually there, because Stargate's not here. There's money in Middle East. And I think it's probably just this, there, yelling about some government contracts, there, values, that's all. That's more about showing us we don't have any money. But, but we have a right to say that we're actually something we can figure out, so we have to be considered, even though we're poor. That's probably what it looks like, right? But, but it's just-- five weeks of money, that's all. It just happened too close. Yeah, well, I mean, if Anthropic goes up to the Emirates and, well, it's so rude to write that they think that-- they're, uh, very rude, Dario's been revoking, I think it's very not politically correct. It's one thing to say that there are other values and that he doesn't share in person, that's one thing. But it's another thing to say that he'll have investors in companies that don't bring goodness to the world, who will use his technology for those, like, compa-- investors who, you know, like, carry Like negative to the world, destruction to the world, but will be a beneficiary of Anthropic. But it's very rude, it's very not politically correct. This is the manner of speaking kindly, but speaking absolutely disgusting... but as a totally disgusting thing, it's, like, I think it's very, very rude to associate Anthropic. c, well, even if it's public goods corporation, other people must be able to determine what's good, which is bad. I think he's very much in the process of paralleling his values and values as a company. And cut it off, you know, talking about Middle East, I think it's crazy and it's not right. That's not smart, it's not right for the two of us. I mean, I mean, I mean, I'm sorry. This is my observation on this issue, which, of course, is a large number of investments coming from Middle East. Well, Andreessen has that, that's what invests hundreds of millions of every start-up that moves. Oh, well, they're the Middle East money, too. So I'm really, really, I don't know, it just seems like it's worth noting that the world is not as clear as it is as to what sources of funding are like, which are not appropriate. So I think we should be more careful. But it's my, um, look, and we haven't seen it yet, statistically, that is, we haven't seen that there is a Middle East bombing of generative AI for our religious purposes, for example, we haven't seen it yet. So I think there's some such judgments that need to be kept a little bit. I mean, yeah, China already has, yeah. So we can say that, yes, there's a lot of control over there. I think that Middle East needs a little, well, whatever, it's gonna propagate their old, that's some old baggage that's historical, forward like that.

00:17:25–00:19:55Agent ChatGPT from OpenAi: promised vs what they got in the tests.
Alexander Mashrabov00:17:25

And I think that's the subject of the guy who's giving money to gen AI, that's gonna be a big theme right now, yeah.

Alexander Volchek00:17:26

ChatGPT released the agent. So, what do you say in terms of tests? I've done a lot of tests, and I've even done a little bit of a spider about the agent. I was on Wednesday. Who was watching the release on Wednesday, write it. Write your opinion on the release on Wednesday. And by the way, you still think from the point of view of what the U.S. government is doing, that's the best move in the world in terms of ideology or that's the move towards China, so, you know, or Russia? About the agent. What do you say about the ChatGPT agent? So Gemini was, X was, yeah, heavy Grok-4 heavy. And the OpenAI is here, so it's the kind of thing that releases the agent. Yes, Elnar.

Mentions: ChatGPT · United States · China · OpenAI
Ilnar Shafigullin00:18:18

Well, I tried to run the test. I don't know how relevant he is, because I basically took a request that I might have thrown into deep research and just missed it instead of a deep research agent. Yeah, that's the tool. It's hard to sit around watching him open the browser and something pooping around and something shows up. Very often falls by mistake. Well, I mean, it doesn't fall down, but it says, "Sorry, I've got this broken, it's not opening. There's a docker down here, something else down here." But he does. The result I got. If you compare that kind of weirdo to the deep research page and the agent's performance is a plus-minus. Deep research would probably be more word-- yes, there's a number of pages he wrote that would probably be more. It might be a little more concise here. I mean, for such searches, at least on the request I've been testing, I haven't seen much difference yet. It's probably gonna be good for some cabs, that he can click on the references, go on the pagers, do something else. But so until I saw a revolution. It's gonna be interesting to hear if you found something interesting.

Mentions: Deep Research
Alexander Volchek00:19:29

Sasha, Tanya, what do you say? Tanya, did you test?

Alexander Mashrabov00:19:34

I said I was waiting for the release to see how to work with it.

Alexander Volchek00:19:38

I knew it. I knew it.

Alexander Mashrabov00:19:40

I tested after our graduation, using the position.

Alexander Volchek00:19:46

Sasha, what do you say technologically?

Alexander Mashrabov00:19:48

No, I mean, I'm the agents browsers, the brauser Perplexity of the OpenAI agent, I didn't

Mentions: OpenAI
00:19:55–00:22:58Agent Case, "Get in, order a photo": Why is it not a daily-use?
Alexander Mashrabov00:19:55

pump.

Alexander Volchek00:19:56

Look, if you take their released Agent Mode, I think--

Alexander Volchek00:20:00

They're released by Agent Mode. I think the subject is a little worthy of attention. What plan? What did they do first? Someone wrote that an agent works on a 4O base, but an agent at ChatGPT is considered to be a current independent design that causes models and some kind of function at the right time. And I'm telling you honestly, I'm in-- I'm in two stages of working with an agent. I started testing him in the first 24 hours after he was taken out. I've been in this first wave enough. Yeah, I was just in six-eight hours, starting to watch him and then I tested for a week. Someone was telling me you were late with the news about the agents. I think it's important to always have a time lag like we do now. So, write, by the way, who has the use of the agent in that time and how much one saw the difference. I'm saying I've had a very positive impression in the first place. I had a feeling that ChatGPT finally came to some normal y-interface. They say there's a system that works and makes a decision about what to do. It is the decision of the Deep Research to launch, when the reasoning is to do, it makes its own decision when it connects some side service or functionality and so on. Write up what special-sense subjects you're interested in. Now, people have written that they are interested in special subjects where it is best to get such education from the AI perspective or to apply business and professional requests. In the comments, we'll all have to work together and take these out. I don't understand the examples. All these are in the series, so I took a picture of some kind of dish, so go get me a book. I still don't understand who works in the OpenAI marketing and X marketing. I-- XAI. I'm not po-- well, I just-- they're the one that's been setting up for two years. Just go get some pictures, so go to the store, buy me those products. It feels like it. But it's an example, not a fauna, it's a fan of example, it's an example of a douche. I mean, it's not a daily use case. There is a concept of daily use of the system. So, by the way, they're on the last of their data, they've got seven hundred and eighty million users, right? Weekly. So they say that, well, people are hundreds of millions. And it's not a case that's a day-to-day thing, it's just a prank they always want to show that the agent went out and clicked some buttons. Il-ilnar, I think you're looking at it, you know, an agent, when you gave an example of what he was doing, something that's doing on the side, right? Well, I mean, because you--

Ilnar Shafigullin00:22:54

Well, it seems to show that, yes, that his main achievement is his

Mentions: Deep Research
00:22:58–00:25:45What is ChatGPT agents
Ilnar Shafigullin00:22:58

decisions, he's smart enough to click, interact with some interfaces, websites. In fact, unlike Deep Research in what? Yeah, Deep Research can just get into this story, but can't interact with the websites. An agent like that, if I understood their presentation correctly and that the anons they were making.

Alexander Volchek00:23:14

Look, yeah.

Ilnar Shafigullin00:23:15

Interaction is what? Doing purchases or what plan? Yeah.

Alexander Volchek00:23:19

Well, clicks.

Ilnar Shafigullin00:23:20

Sniffing on the tips. Yeah, buttons.

Alexander Volchek00:23:23

Yeah, but first of all...

Ilnar Shafigullin00:23:24

I mean, it's not just a collection of information, is it? And click.

Alexander Volchek00:23:27

Yeah. Or analyze, make conclusions. Yeah. But I've got a feeling once more about how I looked at their presentation that the agent is a system that makes a decision as a whole, what to do. Because sometimes you have to go on the site and get in some way, download the file, study it, which is closed, like, somewhere that's not available for nothing. In Deep Research, for example, or in the field of o3, there are models that also go. O3 o3, o3 Pro, they go too, they just don't go as long as Deep Research, for example. Of course, everything is, in fact, very similar models. I've had a feeling again about their presentation, which is what the agent thought was, that it's a subject that determines what to do, and finally there's something new that's gonna make up for you when you decide. questions to ask, which self-definition does what your logic might have, password to ask if there's something you need to know where you're in there, and asks details. And I'm telling you, in the first 30 of my or 40 agents, well, in the first bach, well, here's the block I started, he-- my agent worked much harder than the O3, o3 Pro and o3 Deep Research. But in a week, I started to notice that the agent became some dumb TMI. It feels like there's not enough that the agent's almost used 4O because I've sat down separately, and I've been studying the material on the subject that the agent is not built on 4O, because a part of the bloggers and the authors. I was saying the agent was built on 4O. And then I started to see that he started giving me such an average response. I mean, not that, uh, from the point of view, well, from the point of view of some kind of move. Because I wasn't interested in that, uh, not that I wasn't interested, but these cabins are to: go and sit down, book the plane.

Ilnar Shafigullin00:25:12

Buying food.

Alexander Volchek00:25:14

But they're weird, but they're weird. What is it that you book a plane? Like, uh, if it's about, uh, that, uh, first of all, I don't fly, like, 16 times a day, but secondly, uh, I'm much easier than--- ask him and pass him some data, I just... Open Expedia, click two buttons, I'll buy a plane, just two buttons, I've got all the cards left, all of them. And with a large number of operators, there, United, American Airlines, there, and so on.

00:25:45–00:29:12Gluxes and Agent ChatGPT errors
Alexander Volchek00:25:45

I've got, uh, so-- or Alaska's, and I've got a connection. You came in, you just clicked, bought instantly or there was a hotel booking in just a second. So you don't need to-- no problem. Plus, agents are glued. But anyway, they're not built in some super-crunch system right now. And he's half--- there was a very good story online about what-- an era when agents' beans can fill the Internet with ghosts. To date, thirty-six per cent of the Internet traffic is a bott traffic. Not thirty-six. You know what number? I'll tell you. I came in in detail, looked in terms of, uh, this transcript. He's very big, he's very big right now. To date, fifty-one per cent of the Internet is a proportion of bot-traffic, and of these, thirty-seven per cent are harmful. And by the way, now, according to the twenty-two reports, the beans have gone, uh, for the first time, through the volume, and they make twenty-six million queries in the roundabout.txt. So they can do, uh, well, cheap mass copying of content and make a hard scraping, uh, agenerative of this artificial intelligence. And there's a story with agents here. And the examples also indicated that the boots were launched. These beans are still there, they can play chess, these agents and Perplexity agent, and the OpenAI agent, but they make it curve, they make it curve, they often-- they have that kind of thing-- Ghost R. Well, he's making it feel like you're emuting him, and he doesn't really know what he's doing. Here. And there's a huge, huge amount of mistakes. Well, about the agent, it's-- at least I have a, uh, that conclusion. Let's see what happens. And I'm Elnar, by the way, what did I do? You're just saying now, starting to test. I took it straight and started asking you to start. I mean, that's how I did Grok 4 Heavy, and Grok 4 Heavy I went on doing it, too. But to be honest, there's, first of all, the quality of the Russian language worse. We're in-- I was doing a Wednesday issue, and I was telling you that in Russian, many people think Yandex GPT or GigaChat Sberbank, they're better in Russian. In Russian, all the baccalaureates are better than ChatGPT and Gemini. They're circling. And here's XAI, he loses in Russian. There's a right there. In Russian, the parameters were like eighty-five per cent, compared to 50 per cent YandexChat. Well, ChatGPT and Gemini are doing Russian language unrealistically cool. But examples, I'm being given by the OpenAI with the food order. I, well, there's smartest minds, um, planets, but it's like sometimes they use GPT chat only to develop software, not to check, you know, what's using eight hundred, Nine hundred million people, a month.

00:29:12–00:33:06I'm on the International Mathematical Olympiad: "solo”, as possible, Part 1/2
Alexander Volchek00:29:12

That's an interesting story. Elnar, there was news that artificial intelligence took gold on the world math Olympics. He's got five of the six tough ones in there. Please tell me how important this information is, or it's just, just standard traffic.

Ilnar Shafigullin00:29:32

That's very fun. Ah. Let's just dive a little bit in history. Google DeepMind had already started his math Olympics last year, and they were sort of taken out silver. But then, I think we've even told you about it, the tasks themselves have been re-formulated from the normal human language into a special language. Yeah, there's Lean, I think, used something else to make the model easier with that slave---

Ilnar Shafigullin00:30:01

So the model can be easier to work with. So the model is further there somehow, uh-a, generating solutions, generating hypotheses, testing them and, uh, making some conclusions. That conclusion was also, there, unhumanitated, and it had to be translated, right? And that's how they could, there, have four tasks out of six. That's cool, too. But you must understand that this is a fairly artificial story, because the tasks were translated into a specially convenient species and so on. Google played fair this year, as they say. Uh, it was in a natural language. So, a model was used, which was given the same job as the Olympics, mathematics, ordinary people. And, accordingly, she also gave her answer in a man-readable form that people were checking, right? And it was like, uh, five of the six tasks were completed. Ah, it's the gold medal level on IMO, yes, that's, on the international mathematical olimpiad. That's really cool., first, that the tasks were accomplished, second, that they were done in a natural language, allegedly by models that are not so far from those models that are now available. Yeah, it's different models, but they're gonna be getting into our lives. There's an interesting story with OpenAI because the OpenAI even before Google said they got the gold medal. Google them for that a little bit. There's a little anger in Twitter, uh, that's, like, that you're respectful to the participants, you don't have to declare your gold medal before it's announced. Ah, that's why, in fact, OpenAI and Google, they're at a level, scored at about one level., and both team have met five targets out of six. And what's good here? It's good that the model can generate a large quantity of hypotheses, can check them, find the one that is a worker, and, accordingly, come out with a concrete result, with a specific solution., and both team have reached a result without using any additional tools, right? I mean, it's a parole, it's a Python code, it's a computer launch, and some mathematical bags haven't been used yet. So all this was done in a natural language. And gradually, such approaches, such algorithms to address complex non-standard challenges that are not clearly a solution, will be able to enter our normal life. So it's all making models smarter. The use of these is very expensive because, uh-a, models have been thinking about these tasks all the time that Olympias have given is four, four and a half hours, I think. The process of calculation was very long, and the decision-making process was very long. So you can put some of the models burned down the tokens for a while, how much money it cost. But somehow, the result is very good. The assignments were resolved.

Mentions: OpenAI
00:33:06–00:37:00I'm on the International Mathematical Olympiad: "solo”, as possible, part 2/2
Ilnar Shafigullin00:33:06

The tasks are very complex.

Alexander Mashrabov00:33:06

Yeah, I was here to make some comments. That's where I spent a little time trying to figure it out. From what I understand, from my experience in the Olympics, yes, especially in international, mathematics, the quality of the explanation and how the details are covered is important. And almost always the difference, well, it's clear that the difference between gold and silver is one point. Well, it's kind of a big difference, because gold is, uh, MIT, Stanford, all that is free in the world, and silver is nothing. Yeah? And, and, and-- and whatever, that's, and that's always the talk about these margins, shorter. And there's a certain style of how to describe the right decision, in a way that gets the maximum number of points. I mean, it's important not only to decide, it's important how you're gonna make it happen. From what I understand, it's like a big difference between, uh-oh, well, a standard Russian math school and an international school. Because international Olympics require another approach, there, another type of task and another approach to explaining decisions. And as historically as it may have been, at least in my time, that the boys from Moscow and Petersburg received a lot more benefs because they were, well, as close as they were to understanding how they were built. international Olympics and all these national Olympics are, like, very third-grade stuff. Because for every association, yes, for every federation, in blatantly, mathematics, there in Russia, yes, of course, it is important for them to show the best result in the international Olympics. So, you know, it's like a national olimpiad isn't always that straight-up to international, and that's all pretty different things. And, well, I'm just saying that the OpenAI isn't jealous-- their decision wasn't rude, so how jealous they were those people who were prying decisions on IMO. That's very important. I mean, it didn't go through the jury, and it's super important, because, well, as for school kids, it's a huge difference, it's a huge difference. And that's what people spend on there to learn to score one or two more, because there's a difference of 35, 36 points, that's a 40-two-two-- score. And the question is, can you clearly describe the five tasks on seven points at all, without any logic or score, that is, how to show some movement rightly to the left. A task. In fact, well, all these Olympics are built. It's kind of tough, because, well, how much does it-- it's often about these margins, right? And as if anyone and how much polished, as if you can't lose scores and make scores, don't lose scores on-- when you're doing the job and making the trash when you don't get the job, it's super- important squill. There are many people, because they are exactly at the border between silver and gold. That's exactly what I can say. A lot of people are in there. And that's what I think OpenAI just shirked to get a Haip. I mean, I can understand that the open source didn't work out. Kimika2 took them, then IMO, well, they don't have better results than Google. That's why we have to be the first. Well, in gangster. And Stargate didn't get it either. I mean, I think I'm, like, a little bit of a yes. So they're saying that GPT5 will be out in August. I mean, I understand that OpenAI lost control of the narration they had for a long time, and now they're trying to find a way to get control of themselves through all the means. And I understand them perfectly. I mean, I don't-- I just say, as my own, my interpretation of the situation, but I never criticize them. They're making it clear.

Mentions: OpenAI
00:37:00–00:39:15Who's holding AI-narrate now: XAI/Elon Musk vs OpenAI/Google
Alexander Mashrabov00:37:00

It's just, I think they're prioritizing the control of the drug and aggression, not as reliable as the data. That's it. But it's perfectly normal.

Discussion participant00:37:08

Who's got a narrate control now?

Alexander Volchek00:37:10

Surveillance? Well, Ilon's. He's got this half-great animation on the girl.

Discussion participant00:37:16

But it's weird, but it's weird. If you want to get into the audience of hundreds of millions, why do you want to start a naked girl? I think a lot of people are gonna turn Grok off, not turn it on. No?

Ilnar Shafigullin00:37:29

How many animosexuals in the world are there, Sasha? I think there's a lot of them, so your audience is big enough.

Discussion participant00:37:34

If you're doing it for the audience, if you're doing it for TikTok, maybe. But I mean, it's a weird little thing. So do different avatars or different ones, but it looks very specific. It looks very specific.

Alexander Mashrabov00:37:48

Yeah. I mean, it's a pretty special situation. I mean, I think that Elon is about to be winning the baccalaureate, triggering a viral effect, saying he's doing AI Vine. Plus, I see Epstein's a sheet, right? And there's a few, there's something he said that Stargate won't work. I think that as he is now, he is actually controlling the minds of a broad, broad population of AI enthusiastes. Here. Well, maybe it's a quick change.

Discussion participant00:38:18

But what you're saying is what you're saying is, it's what you're saying is, it's the AI enthusiastes. This is designed for technical conditions. I sometimes watch some presentations as if they were designed to be technical, not to the mass market. You just say OpenAI lost his narrate. You always watch it? Look at the tablet, on the number of surgeries.

Mix00:38:37

On the pill? AppStore, yes. Yeah, yeah. There, of course.

Discussion participant00:38:39

Sure. Look, look at the number of downloads, the number of users and the use of these conventional boxes. I think that's what matters. Here.

Yeah. It's hard to say, it's hard to tell where they are. Aah. I mean, it's a little hard to say here because there's a point of view that you have to win, well, that there's a lot of AI educators and so on. AI influensers who are suspicious of AI content. And if you win a technical audience, you win over time and a wider audience. There's a point of view.

00:39:15–00:40:10Strategies to speak to the consumer: why AI-companies do not have

So, Sasha, I'll make sure there's no AI company that learned how to talk to the consumer. That's how Apple tries to talk to the consumer through WWDC, right? Tesla speaks to the consumer through his own conferences. None of AI companies can talk to the consumer like that, and so, well, as a bar, it's not very tall. You don't have to talk to the consumer, you need to talk to AI. If we're looking at this low bar, then Ilon, I think he's in control now. Here. But it's a low bar, right? Because nobody talks directly to the audience.

Discussion participant00:39:51

That's right. Clearly, from the point of view of the subscribers, the mass even the subscribers, though, it is personally making a very large deal of the envelope.

Alexander Volchek00:40:00

Personally, a lot of things are converted because he's got a major focus in XAI. That, that, that's-- that's what I think is a huge plus.

00:40:10–00:41:26Which would give it to me.
Alexander Volchek00:40:10

No Gemini, no OpenAI does that even around. More like technical stuff. There was some news that Apple wanted to buy Mistral. And I like, uh, discussion, you know, what it's supposed to be, that kind of data. It's French, French AI model. I'll remind those who didn't see our viewers. I'm an inter-- I like the discussion. Apple generally has a price of 15 billion. I'm looking at it and I think it's not a price issue. Apple now, even if we look at the company's evaluation, they're back-- they're back from Microsoft, and they're $400 billion. And I think that if Apple has anything in artificial intelligence, they just make hundreds of billions of dollars in evaluation. Here. But we won't talk about them for long. They're gonna do some more push-notices in the middle of something smart. But these statements, they're very funny, of course. Here, push-notices are very funny. Look, I want to raise the subject by the end of the release. Ah, important--

Alexander Mashrabov00:41:15

Wait, wait, wait, wait, wait till we ran Apple. Actually, was just trying to, like, explain this little view here, right? To make sure it's like it's all heard, right?

00:41:26–00:44:45Meta and AI in sales: what is really launched and that is not
Alexander Mashrabov00:41:26

I mean, I think it's super important to cover it a little deeper than it's a passing. At least I think it's more important than these, there are, the Trompa policy. So, Meta now has no AI in the sale, well, it's just not available. I mean, Meta AI, who, there, was a generation a year ago, even a year and a half ago. I don't know, Meta AI wins Character AI's engagement, doesn't win. Who--who's-who's gonna need this Meta AI, doesn't know where it goes, doesn't understand. In addition, AI agents for advertising that are not close to the market. I mean, what's going on in Meta, I don't know. But since they buy people a lot, they spend billions of dollars there, everything looks cool. Stock just all time high, huh? And that's because there's no AI at all, well, just missing for now. And, uh, and as much as that, well, it's pretty obvious that, like a volume, even these bass billions of dollars have, like, a very big impact on capitalization, right? And Apple is worth two hundred and fourteen, at the peak, they cost two hundred and sixty. I mean, there's, uh, a lot of growth, yeah, 15, 20 percent, that's exactly what it is. And as much as it's costly here, it's impossible to understand that the market now looks like it would have appreciated these influencings in AI, and that could have improved the situation, too. I mean, why they don't do it, it's impossible to rationalize. On the contrary, they lost their chapter of their multimodality here in Meta.

Alexander Volchek00:43:05

Yeah, yeah. And why aren't they doing?

Alexander Mashrabov00:43:09

I don't know. I mean, in my opinion, that's impossible to explain. It's, uh, it's a nons.

Alexander Volchek00:43:14

Well, it's for me...

Alexander Mashrabov00:43:15

It's crazy because--

Alexander Volchek00:43:17

Alexander Mashrabov00:43:17

Because, well, it's kind of a rule of c-capitalism, right? I mean, aaaa, well, I think it's just a rule of c-capitalism, if you're being honest, right? What, you know, you're, uh, the market, he's pushing you, you know, to make some decisions. And I don't know why Apple is fighting like that.

Alexander Volchek00:43:35

They'd have to buy, buy some super-crunch players, just take them back and everything, right? So someone else doesn't take them, and move forward. Well, they have a lot of free money, a lot of free money. Clearly they can hire, strengthen themselves with this, uh, technology team, or at least two or three teams.

Alexander Mashrabov00:43:55

100 percent.

Alexander Volchek00:43:56

I guess that's the same thing...

Alexander Mashrabov00:43:57

100 percent.

Alexander Volchek00:43:58

Google, in Microsoft and even the team, the parallel or there in Meta should be teams that work side-by-side. So these are companies that should allow... That's what NVIDIA is. I know exactly what a huge number of parallel teams are trying to do in NVIDIA. There's no one in your department doing, like in XAI. Well, NVIDIA, Jensen-- Jensen is trying to do something straight, uh, something different, from different stories, so that it works. Maybe it's not a best strategy, it's not always the best strategy, but if you can't do the cool stuff like Ilon Mac, what else do you have?

Alexander Mashrabov00:44:34

Well, in my view, that is, in my view, that is, how obvious they pay a lot of Google, and they need some kind of search, they need to experiment.

00:44:45–00:47:31Anthropic: risks of AI abuse and MENA investors
Alexander Mashrabov00:44:45

And there's You.com cheap, there's Perplexity in the dark, but... Yeah. And I mean, uh, it's impossible to explain. Impossible.

Alexander Volchek00:44:52

I want to comment on another story that I've put forward, and, uh, chapter Anthropic, and she, at least I think, is very important to the perception. He's a sca-- he's gone some leakage of his correspondence with internal staff, and he said that artificial intelligence could serve bad people. So it was written that unfortunately, the principle of our technology should not benefit the bad people. Well, the principle that our technologies should not benefit the bad people, will be almost impossible to implement. So it makes it clear to everyone once again that, in fact, by developing cool systems, these systems will be used, uh, for abuse and for-- and for different very specific designs. Well, I love quoting him, I think he's good, good, normal, right, right, saying that nobody's making certain illusions for themselves.

Discussion participant00:45:53

What's this got to do with it, Sasha? It's because they just started taking money or they started taking money from Men in the region, there, I don't know, from the saudits, from the emirates, from someone else. They had refused to invest before, and the rest of the companies, the same OpenAI and other companies, in fact, attracted huge amounts of capital from the region. And if they refused last year, he says, "Well, we're either on the side of the road because we're not gonna have money, we can't get it, or we need to get involved." And then this is the correspondence that we refused to take money from the bad people. And I'm curious about, well, how do you feel about it, how do you respond to those who they refused to take last year to what kind of well, now give us the money, we're ready to take it. That sounds, of course, that news is then--

Mentions: OpenAI
Alexander Volchek00:46:43

No, look, wait, wait, but you're probably the one with the subject. I also said that he, he says that bad people can abuse the system.

Discussion participant00:46:51

That's the same thing. So they didn't let their systems take to people they thought were bad.

Alexander Volchek00:46:57

Oh, you mean, one of them.

Discussion participant00:46:58

And now they're like... Yeah, and now they're like, well, give us the money, we're ready to give you the money. So, what happens, then? Our models will go to the wrong people, and they will be used by other not-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-so-like people.

Alexander Volchek00:47:10

Very, very interesting.

Discussion participant00:47:12

I understand that this news is about attracting investment from the Men of the Region, Middle East, North Africa.

Mentions: Middle East
Alexander Volchek00:47:21

Thank you very much, everybody. I'll see you in exactly a week and watch our Wednesdays. Bye, everybody.