Transcription
Sasha, I really like your antlers. >> These are not my antlers. Tanya, >> I have, uh, uh, above the bed in this motel, a whole series >> of antlers hanging. >> Of antlers, yes. And they are made, uh, in an Indian style, decorated. And I lay down yesterday. Well, I can't say it's pleasant to sleep with this. They are not just hung up, but just hanging. And at the same time, I remembered an idea yesterday. My son Igor, about a week ago, found deer antlers on our property. >> Exactly. >> Remember, you came over and suggested hanging them on the wall here. No, here they are very coolly incorporated because they are in the Indian, in the Indian style, yes, they are painted with leather laces, they are made with something else, so they are incorporated as some kind of decoration. Yes. >> Yes. And Elnar, Igor found them, so he runs in from the property, happy. Deer constantly move, so, >> yes, deer antlers. >> Deer antlers. >> I suggested he make a candle holder or hang it on the wall. He somehow, you know, out of politeness, like, yes-yes-yes. And then Steve tells him, "Make a slingshot out of it." And you saw, his eyes lit up automatically. He ran to look for a rubber band. How? Yes, >> a rubber band. I said, what rubber band? What slingshot? >> Yes, that was funny. Okay, so what? The heads of major American banks have stated that artificial intelligence increases productivity but will lead to job cuts. Meta plans to release a new large language model called Avocado in the first quarter of '26. China has reached 60% of global AI patent volume. Deep has released open models that are comparable in quality to the most advanced ones, like GPT5, Gemini, and are an order of magnitude cheaper. There are figures like 10-30 times. Certification, heating, and much more have been introduced. Hello everyone. We are on the To the Moon channel. Technology news, insights from Silicon Valley around the world. We release episodes once a week, although keep an eye out for announcements for the upcoming Wednesday. There will be an episode. Sasha Mashbov and I are doing a special episode. It should be on Wednesday. I hope the editing will be done in time. So, if not this week, then next. >> Write comments for it to come out this week. >> Write comments for it to come out. Regarding, I read a report again where a whole series of banks, like JP Morgan, Wells Fargo, City Group, PNC, they all said that they checked everything, that artificial intelligence increases productivity, it's super cool, but we will reduce jobs. And I immediately remember, there was a JP Morgan report in which JP Morgan said that, uh, jobs will not be reduced, but everything will improve instead. And what I wanted to say with this comment is that if you listen to everyone, if you collect all the comments from everyone in a circle, from the point of view of their statements, then there is a lot of, uh, absolutely unstable information. Nobody fully understands what will happen, how it will happen, who will be the leader, who will not be the leader, which platform has been lost or not lost. And I just saw very interesting information about Deep. Someone posted the difference in the description of Deep's cost, from the point of view of Deep's prices for their models. And there is a very interesting comparison there, that for input, Deep is four to 10 times cheaper, and for output, it's up to 20-30 times cheaper, if you compare it to GPT 5.1 and plus it's almost free with cache hits. Well, that's when you work with agents. And it's very interesting, of course, from the point of view of pricing policy to see it within the framework of a Chinese model. And at the same time, you read the news that Anthropic plans to reach $70 billion in revenue by 2028. Yes, $70 billion. $70 billion. I'll note that Anthropic's last recorded revenue was a billion. And, as far as I remember, it was an error, meaning it was annual recurring revenue, projected. That is, they took the last period of time, multiplied it by 12, for example. But each of them also has their own calculation methods, and they showed this revenue. It's very interesting how accurate this revenue is when they show it. We were discussing a tech company yesterday, where a person said that their company was valued at $200 million with $200 million in revenue. And you sit there and think, maybe they received such monthly revenue once, and then stopped, right? And they had no more revenue. And, >> well, like, they showed it on Black Friday, multiplied by 12, and said, like, we have a billion. >> Listen, there was a lot of discussion about this, by the way, regarding American AI startups and how someone exploded on Black Friday, sold, and then continued to do so. Yes. Even some startups we know did that. By the way, speaking of Black Friday. Tanya, about shopping. Several times, several episodes ago, when ChatGPT released its shopping module, you were clearly right, because a report came out on Black Friday online sales, it broke all records, $11.8 billion, largely. And it explicitly states, largely thanks to artificial intelligence recommendations. >> So artificial intelligence, >> yes. They didn't miss, you see. The main thing is to be on time, yes, >> the right time, the right place. Yes, it is said that artificial intelligence greatly influenced the volume of goods sold and increased its volume. Yes, increased. Well, it's like the appearance, I think, of marketplaces like Amazon, yes, or in Russia, Ozon, Wildberries, they created a situation where people started consuming more, yes, due to the ease of ordering, the possibilities. And now you also have, you know, endless recommendations. >> Well, I think because the tool is still new, I mean AI, people trust it more than others. Others have become too familiar. It's like before, if it was shown on TV, it meant it was definitely true. And they were a big specialist. And now, I think AI holds that position for now. But of course, it will lose it over time. Well, by the way, speaking of the 70 billion. A big specialist was shown on TV. So, it's the same now. It's no different now. 70, 70 billion dollars in projected revenue. Not revenue, profit. Excuse me, profit again. So they are projecting correctly, Elnar, they are projecting profit now. Yes. >> Yes. Now, look, this is a very important story. Now they have a billion dollars in revenue at Anthropic. A billion dollars in revenue. The last valuation was fixed at $350 billion. At this valuation, moreover, you won't buy this company, right? You won't be able to get into this round and nothing, and no one will sell it to you. And the revenue was a billion, and now it's shown here that they want to make a profit of $70 billion. Perhaps these statements are for IPOs next year, maybe for something else, I don't know, and how they will make $70 billion in profit when ChatGPT wants to make $100 billion in profit. And here come out models that are very cheap. And I think that the example of these prices, what does it show? Especially there was a very interesting report from ChatGPT on how companies use it and how much the use of ChatGPT in companies has grown in terms of API, tokens, GPT modules, all sorts of projects, well, and everything else. And we, I think, will talk about this separately. Maybe we'll make the whole episode about it. There are very interesting things there. And where will all this money be? Well, where will this profit come from? That is, where is such a volume, how can such a sales volume be generated? That is, to generate such a sales volume, it requires sinking a huge number of other companies. Not just sinking, well, like, sinking some mammoths, but this is a new market. So this money should appear in a new market, or these companies should replace other companies. That is, to make such a profit, Salesforce must disappear. A huge number of Microsoft's corporate products must disappear, because, well, how else can you generate profit? And even remember OpenAI, they said last time that they predict, and, I think, for five years, right, they predict user growth from 800 million to several billion, and yes, up to 2.6 billion. So even if you multiply everything here, where will the profit come from? Uh. And this is a very, very interesting emphasis. What do you think about this? And at the same time, Anthropic, well, Dario Modena said in one of his speeches that they are very conservative and very cautious about equipment purchases. And he mentioned there that, I think, they say that in 90% of cases, well, that is, they take future forecasting, they take that, like, 90% of cases that can arise, Anthropic should buy this equipment itself, meaning it should have enough money to recoup these investments. And only in the 10% of the most negative cases will it lead to Anthropic not having enough of its own money to pay for this equipment. In this sense, OpenAI is taking a much more expansive path. That is, remember, they have, I think, $1.4 trillion in equipment commitments, they've already bought a lot. Well, one way or another, yes, through companies or something else for the future. >> Yes-yes-yes. And Anthropic in this regard is going very, very conservatively, that in any case, we have enough money to pay for this equipment and so on. This simply leads to the fact that, if everything goes well, there will be a huge number of computers, and Anthropic will have significantly fewer. And how they want to make $70 billion in revenue with such a gap in the opposite direction from OpenAI and most other players, that's certainly an interesting question. Profit, profit, Elnar, profit. >> Even profit, even profit, yes, even profit. $70 billion, yes. >> The most important thing, the most important thing is how they will make $70 billion in profit. That is, they clearly have enough money now, because they are simply selling stakes in the company, right, selling, selling, and that's it, you have money, capital has come in, and you are not taking borrowed funds, loans. You are giving up a stake in the company or exchanging it, maybe giving up a stake in NVIDIA for them to supply you with chips, and so on, right? And there are a huge number of memes, reels on this topic, like, like NVIDIA is sitting there, exchanging stakes with OpenAI, with Anthropic, they've all exchanged in a circle, they've all grown in valuation, and nothing has changed at all, just nothing has happened. That is, no one has done anything, and a plus trillion dollars has just appeared, right. That's the market, that's the market. We will watch this. And on the one hand, it's just a statement, and on the other hand, we see development and trends, and how these companies are developing forward so grandly and doing a huge number of projects within which we don't know, and how to find out about them and how to understand them, it's unclear. So, by the way, regarding, yes, >> at the intersection of your first news about banks, that JP Morgan and other banks made a forecast that, yes, we will have, uh, labor efficiency will increase and at the same time the number of vacancies will decrease. And Anthropic. At the intersection of these two news, there is a funny post by Andy Jones. He is one of the first researchers at Anthropic. And he provided such interesting analytics that in 1700, the first steam engines appeared, and accordingly, they developed, then internal combustion engines and so on. And, if you take the number of horses per person at the beginning of the 20th century, well, that is, when cars just started to appear and so on, and accordingly, after 50 years, almost all horses disappeared. Well, from everyday use, as transport, as draft power, as something else. So it took 50 years for engines to make horses disappear in New York or elsewhere. And then he gives the next example with chess. So, uh, in the early 2000s, well, in the late nineties, a strong grandmaster won 90% of games against a computer. Literally 10 years passed, and accordingly, the situation changed, that a human against a computer can hardly do anything anymore. Then he goes on, he says, "Look, I don't work at O, I work at Anthropic and I was one of the first researchers hired. And in 2024, the main part of his work was answering questions, technical questions for employees who were just hired. He has been working for a long time, knows the entire architecture, etc. And employees constantly turned to him for advice, for how it works, how to do it better. And Claude appeared, well, he spoke at a seminar with this report in the summer and says that Claude appeared literally six months ago, that is, last winter they started actively using it at Anthropic, and in just a few months it took almost all these tasks off his hands. And he draws the following conclusion: horses had about 20-30 years. I hope we will have at least 20 years too, >> because the replacement is happening very quickly. It won't, >> it won't. Yes, >> soon there will be some negative mark. >> You were replaced a month ago, but you don't know about it. >> You haven't realized it yet, right? >> In general, we will leave the link. Go and read it. On the one hand, it's funny analytics, on the other hand, it's a little sad, but >> you want to cry. Yes, yes, yes. Yes, yes, yes. This is at the intersection with the H2 news you mentioned. >> Yes. By the way, speaking of big, uh, numbers, Elon Musk's SpaceX plans an IPO in 2026. And the numbers circulating are about $1 trillion. So, by the way. Yes, again, it will pull Elon Musk away from any rankings. I don't think anyone can catch up with him. I don't think anyone can catch up with him under any circumstances. >> I kept thinking: "What will happen when he's gone? How will everything stop?" It feels like people will fly to Mars. Sam Altman would argue with you. >> I would gladly argue with him too. Let him contact me. Well, by the way, the changes that SpaceX is making to the market are indeed very significant. Remember, we even discussed once that if, uh, delivering cargo to space becomes much, much cheaper, then space will start to be used in completely unusual and incomprehensible ways, yes, which we can hardly imagine now. And here I want to recall, Google had news on November 4th, yes, that they are launching the Sunkatчер project, they want to build data centers in orbit, and there, accordingly, place satellites with solar panels, because the efficiency of solar panels in orbit is almost eight times higher than on Earth. At the same time, the solar panel can be positioned so that it always faces the sun, yes, so that it doesn't hide in the Earth's shadow. And I don't remember the exact figure, but if the cost of delivering cargo to orbit reaches a certain threshold, building data centers in orbit will be significantly cheaper than building them on Earth. And from the point of view of electricity, which is generated there, and from the point of view of availability. In general, the technology is ready. Moreover, Google has already tested its TPUs, which we talked about last time, of the previous generation in the calorimeter. Well, that is, roughly speaking, they irradiated them with radiation and checked that the memory does not degrade, at least within a 5-year horizon, and so on. So it may turn out that, thanks to Elon's reusable rockets, the cost of delivering cargo will become much cheaper, and directions will appear that, well, even a couple of years ago, it was hard to imagine that data centers would be built not near nuclear power plants or hydroelectric power plants, yes, or conversely in the north, because they need cooling and so on, but would simply be lifted into space. Now this looks more or less like a real story. >> Yes, so this is very cool. The head of Google has been stating in every interview for several weeks in a row that they will build infrastructure in space endlessly. That is, the entire X there, V there, generally, yes, and that infrastructure will appear there. And in Japan, Japan, I think, was the first to manage to return accumulated electricity from space. And in the plus, as far as I understand, in the plus, yes. And literally yesterday, I think I was flying and saw data that Japan is super innovative, super innovative in terms of space exploration in terms of electricity. It looks very specific for now. We'll see if they can resolve this issue. Well, by the way, speaking of cost, and I remembered, you know, the story about Neuralink. The latest, you know, there have been a lot of reports lately that concern, uh, well, companies, like SpaceX released and now an interesting one. And literally a couple of days ago, Neuralink, also Elon Musk's company, accelerated neurochip implantation. And there, you know, I like this news, and generally the information that is written, it's written, you're talking about cost, and it says it accelerated neurochip implantation to 1.5 seconds and reduced the cost of components by 95%. That is, Neuralink announced that its surgical robot is now capable of implanting one electrode in 1.5 seconds, whereas before it took minutes. Well, and accordingly, I think, it's like speed matters. Here it's already like a factory, right? >> Yes. A minute or a second, like, >> of course, of course. Like, it seems like, look, it seems like, look, what's the difference. But if we immediately look at his statements regarding the release of robots, that he is now making a factory that will produce a million robots, then 10 billion million, and somewhere a billion, he directly states and says, "Well, then we will probably be able to produce a billion a year." And they are building a factory for a million, I think, or even for 10, in Texas. And again, there was information that they will start mass production in the first quarter, and so on. But we'll see what happens. But regarding this Neuralink, about 10,000 new people have signed up as volunteers for future trials. Viewers, by the way, what do you think about this? Would you sign up as a volunteer for trials or not? And if we have volunteers among our viewers, 10,000 people have signed up as volunteers, and, uh, for the implantation of the implant for future trials, yes, and, uh, well, there are a lot of tests shown, yes, what these people can do with the implanted Neuralink now, yes, how they can control something in a game, how they can move something, how they can move something, and so on. Well, it's clear that there are still a large number of different startups that are engaged in this topic, simply because of Elon Musk. We will announce, we will show Neuralink. And probably Elon Musk will have one of the largest resources in the world right now to invest there. Especially this is definitely a technology to which he pays incredible attention. Well, and 1.5 seconds raises a question for me, of course, what do they plan to do? Do they plan to implant Neuralink in billions of people or what? And here the words come to mind, if >> multiply by 8 billion quickly, everyone will be gone. Yes, yes, yes, yes. And here the words of, uh, Vanka or whoever, who is the head of artificial intelligence at Meta, come to mind, who says that children should not be born now, but a little later, when a chip appears that we will quickly implant in the child and immediately teach him. Well, that is, he will learn everything, all the math, like all the physics, he will know everything completely, he will be a super-intelligent person. Write about this topic, it will be very interesting to see your thoughts on this story. What do you think? Who would want to do this? Who would want to do it? >> Surely, if this even happens, the child's consent will surely be needed for this. That is, it's time to give birth. Because within 18 years, they will definitely do something similar. But it sounds terrible, honestly. >> Yes. And will there be the child's consent, or will they introduce some law that allows this? That is, it sounds very, very, yes, it sounds terrible, actually. Well. By the way, speaking of Meta. Since we mentioned Meta. And I think this news is important because in the last episode we said that Meta has fallen behind, Meta has disappeared, Llama has disappeared. Although we did post data in the credits in the last episode that its open models still have great weight and significance in certain tests and uses, but Llama has quieted down. And here, insiders at Meta, Chinese insiders, well, Chinese and then American media are writing that there is a codename for the model. Uh, this codename is Avocado. And, uh, that it will be the next generation of Llama, that supposedly this model should appear in the first quarter of 2026, and that this model will be closed. That is, at least, what is described there, that it will be released as proprietary and that access to it will be through an API without open weights, and accordingly, it seems like this will turn Meta around. Very interesting. Very interesting, probably with the resources, again, with Meta's resources, and Zuckerberg's statement about, Zuckerberg has already stated that they are ready to spend any money, including mistakenly, on artificial intelligence, even if it's $200 billion. says that they are clearly making efforts to stay in the race within artificial intelligence, and to get into this list, as Llama used to be. Just before, we discussed Meta a lot, remember, it was constantly everywhere, right? To return to this race and be alongside OpenAI, to be alongside Gemini, at least to be alongside Claude, at least from the perspective of American models, yes, if we're not even talking about Chinese ones. >> So, we're waiting for Avocado. >> Interesting, Avocado is a continuation of Banana. Will the next ones all be fruits, vegetables, shallots, pitanas, eggs? Yes, they are, they are, you know, by the way, this is an interesting story that today with Elnar, when we called, Elnar says, "Where are you?" I say, "I'm in Santa Fe." Elnar says, "Hyundai." We actually know a huge number of words, like before there was Apple Big Sur, yes? And and yes, and you hear some or there was Monterey, or they had, well, completely different names were used. Uh, and you treat it, you think, what, what kind of words, yes, beautiful or Sequoia, they use, yes, you have, like, some names, and then you, considering that you live in California, and you, well, Santa Fe is not California, it's the state of New Mexico, yes, and you understand further that New Mexico is not Mexico, yes, and it's not located next to Mexico, as it might seem. >> And there's snow there. >> Yes. Yes. As it might seem, yes? There is snow outside in general now, a little, but there is. And it's very interesting, of course, when you understand where that or that word actually came from, yes, sometimes additional depth appears to everything, >> yes, a lot of additional depth, yes, even, accordingly, when you have a different pronunciation of a word, that is, you understand that it's not Zuckerberg, but Zakerberg, and not Nike, Nike, yes, and you, being in a huge number of different brands and surnames, everything, >> but BMW, not BMW, >> yes, >> let's start with Livar. >> I still can't switch. I just said something, what >> So it's also interesting. So it's also interesting, Tanya, pay attention that if you are with Americans, then, well, you understand that you say Nike, yes? Well, that is, you say, you use, well, obviously. But if you are with Russians nearby, then you already don't, or I even speak with people who don't live in America. I don't understand, should I tell them: Nike or Nike? And this story, I have a feeling that if you say Nike, then they'll say, Oh, well, it's clear, they're trying to be like Americans, >> yes? Yes, like you're pretending, >> yes. But on the other hand, you, well, you get used to it, yes, and you understand that, yes, and you can say it automatically, >> yes, and there are things, you can't come into a store here and say, give me a sweater in size S. In California, they
They won't understand that this is often, they need to be told, right? And Polina often tells me this: "You speak in the store, they can look at you, stand there, and say: 'What are you even talking about?'" So, everyone has, well, habits, right, some cultural traditions, names, words, a lot of things. And Polina constantly promotes to me, says that in Russian, strictly categorically, only Russian words should be used. Yes, categorically do not use English ones. And lately, she herself sins with English words. >> Of course, you start sinning with English words and preserve. We met with one person here, he has lived in America for over 20 years, and when speaking Russian, he does not use, well, any Anglicisms. He doesn't use them. And that's very cool. This was clearly not Sasha Masha. A fine line. >> And Sasha, now I'll remember everyone, Sasha Masha. No, I'm saying that at least, at least >> Yes, well, just Il'nar, you know, this Nike and Nike usage, right, especially when you say, I speak with my mother, well, to my mother, Nike will mean nothing at all. So she won't be able to. I want to say that in general, even people our age, Nike means nothing. Well, you're like, I remember, the first time I heard it, it was like I was hit >> with a knife in the heart. >> Yes, with a knife in the heart. Yes. >> Yes. From the point of view, Sasha, there is no clear boundary. A borrowed word and a native Russian word. Well, for example, is "buterbrod" a Russian word or not a Russian word, can it be used? Or "kaloshi"? Yes. Yes, yes, yes. "Buterbrod". Therefore, yes, bread. >> But from the point of view of habit, >> yes. Well, from the point of view of habit, especially with all the brands, with everything, how it's pronounced, there are such concise words in English that cannot be translated into Russian, which you want to mix, because the word hits 100%, and this one hits 97%, so >> it's not quite the same. >> Regarding, uh, therefore, avocado, avocado. So, what do I want to say again, we somehow talked about patents, I think, six months ago. And for viewers to note for themselves, this is a very interesting thing, a general understanding in the world of the balance of power. I roughly, well, when we started a year and a half ago, Tuzmun strongly promoted the idea that pay attention in the countries where you are, that a huge number of countries are no longer participants in artificial intelligence. They will never catch up in their lives, right, they have lost the market, and lately, it seems like there's no need to even talk about it. And what's the point? And last time, I think, we cited the UN, uh, yes, the UN. We talked about the UN, that they did a whole research article, a research study and an article on the topic that, uh, artificial intelligence is very strongly separating countries, and many countries have already lost 100% in this matter. And they will never return to this topic. And here's a question for China, right, we know that, you know, there are leaders, the USA, there is the leader, China, that it is clear that in general, American models have a sufficiently strong weight, have a huge amount of money, valuations are high, and so on. At the same time, China has an incredible thing. Firstly, the number of cool engineers they produce. And this is due to some settings. For example, many consider that the coolest university is this one, or, for example, I don't know, Russian programmers are cool, or Belarusian are cool, or these ones, or those ones. Once again, about the Chinese. Nine out of ten top world universities are Chinese in terms of producing engineers in artificial intelligence. And 60% of patents in the world are now, 60% are already with China. Here's the new figure that China has reached 60% of the world's patent volume. And, of course, to be honest, this figure is astonishing, especially when you understand, especially when you understand that they, by the way, have grown, uh, in general, the Chinese industry has grown by 24%, but and in the world, they don't occupy such a large volume. They currently occupy 15% of the world market in artificial intelligence, right, the world. But 60% is very impressive, considering that, probably, the Chinese themselves don't care about patents within their own country. Well, probably within the country, from the point of view of international law, >> yes, the most loyal to content >> it seems, but maybe not anymore. But maybe not anymore, yes, because China has long been a completely different market, right, Chinese products have long been completely different. And judging by the latest news, when, like, America, China blocks, and then Trump allowed American chips to be sold, >> China itself refuses, I think, >> Chinese companies. And immediately, imagine, this is very interesting, right? So, first, you know, everyone says that American chips should be distributed, then they say: "We won't, so they don't steal technology." Then, you know, they say: "No, we will." They allow them to be supplied. And China internally says: "And we won't buy some of your chips." And we will very strictly control the purchase of chips. Very strictly control the purchase of chips, because you might have embedded some virus in your chips and want to store some data. What, what, look, in principle, if the American government suddenly decided to open up chips, it says something. That most likely, something is embedded inside these chips. >> Well, the chips, to be honest, are of the previous generation, H200 is used. But not the 200 that are current, right? >> First, second, China, I understand, has fundamentally decided to gain some independence in this regard from chips, right, and forces its domestic companies like Huawei to actively develop, forces them, in fact, to train their models on these chips, but so far unsuccessfully, right, so far Huawei has not coped with this task. The same thing. Remember, we discussed with you in August that they are trying to build training for their new model on Huawei chips. They can't get inference to work yet. That is, subsequent use, and specifically training the model on Huawei chips, cannot be done. What Google managed to do, which we talked about last time. And now, too, these DPC models, which I started talking about, are also trained on new Nvidia chips, which were brought into China through various schemes. But such widespread distribution. China still wants to have its own chips. Well, because now the US has allowed, I don't know, in six months, let's say, no, we changed our minds, right? >> Yes, yes. Yes. Therefore, accordingly, it is very important for China to have its own chips be as competitive as possible in this regard. And it's also very interesting, why allow chip supply now, if even for American companies, even for American startups, everything is very strict in terms of equipment issuance. It is very limited and large volumes are issued with government approval. And everything is very strict, and it's not so easy to buy. I remember someone wrote to us: "What's the big deal? They bought it in Russia, chips. Well, buy them, buy them. You're just not in Russia, and you won't be able to buy a normal volume, right, it's very, >> yes, it all depends on the scale. You can buy them individually without problems, but in large volumes, when you need thousands, several thousand cards, it certainly becomes a tricky task. Continuing, Sasha, about what you started talking about China, about talents and so on. Last time we talked, right, that PAI was the first to publish the GRPO algorithm, and now they have, accordingly, published how they train their PAI model. Here. Mathematics V2 came out, 32 came out for the main PAI. And they use very interesting approaches, they don't spend much effort on data labeling, which is the most expensive, right? And they show how they build the process in layers. That is, first they make the model generate answers, then, accordingly, people just say, good, bad, good, bad, good, bad. They start training another model that will try to predict whether it's good or bad instead of people, and so on. That is, there is a whole matryoshka of different models, but with this, they managed to get very decent results without spending much money, at least they claim so, and without spending much effort from live people on data labeling, right, about which I spoke at the very beginning. That is, they have released several models now, very good results, and at the same time, they publish how it was done. That is, in a sense, China is also driving forward such, well, I wouldn't say science directly, but its open part. That is, I am sure that, of course, OA and Google, and probably Anthropic, have their own developments in this regard that are not much worse. Although, who knows. China, for now, is the one that shares, publishes interesting open things. >> Well, and we also know perfectly well that China and Chinese domestic technologies, they don't reveal them at all. That is, a huge number of things are definitely hidden. No one understands what is being done there, how it is being done. If here they openly say: "We cooperate with the government here, so we issue laws here," then 100% everyone cooperates with everything, 100% everyone does everything, 100% is aimed at implementing artificial intelligence everywhere. Moreover, China's automation in terms of the concept of automation itself is at the forefront of the entire planet. I'm talking about scale now, I'm not talking about individual cases, when they give examples, say, we automated Singapore, well, and Monaco, you know, when you automate, well, some micro-small region, this task is very simple, right, when you need to bring a huge country with such diverse infrastructure in order, this is a serious issue. But China has always amazed me. I even remember in 2013 I had a trip to Tibet, and I had been to China many times before. And we were crossing the Tibet-Nepal border. And you feel this difference, right, when you drive in Tibet and there are cameras everywhere, and super-roads are made. It's clear that due to the occupation of Tibet and all sorts of things. And then, when there's a bridge, you, you know, approach, your passport is checked in China, smiley faces everywhere, how you were served, the quality of employees, you know, everything is covered with cameras, some scanners, you know, whether you are healthy or not. You then cross to the bridge, and in the middle of the bridge, I remember, we approach, there's a young guy, about my height, holding a sign with our names: "There are three of us, we crossed the border on foot." We come out, you just cross, there's a village, such a chicken coop, wooden buildings of some kind. We walk, there's such mud everywhere. We walk, even enter some wooden building, where some uncle or someone just stamps our passports. Then we go out, get in the car, and drive forward on such a bumpy road. Well, it's clear that China has flown very seriously in terms of technology and everywhere internally in terms of coverage. And it's clear that when you consider that artificial intelligence is being tested there on huge volumes of real data. By the way, a big question from the point of view of state data, probably China 100% and human, well, real human data, there is access to everything, because in America everything is classified, closed, systems are different. The president can declare: "I want to intervene here." They tell him: "You won't get in here." So, the struggle of all different groups of states is very serious, the constitution for data preservation. In China, obviously, you can scan anything, all people's correspondence, build it. They scanned everyone like that in chats, right, we talked about this, by the way. And imagine what a mass of data there is, which can be analyzed, which can be looked at, which can be understood from the point of view of people's lives. Even, for example, the US does not have access to training on correspondence of live people, on modern correspondence of people. They cannot take and train all these unified arrays on correspondence in WhatsApp messages or on correspondence in iMessage or correspondence in Facebook Messenger. They cannot do this. In China, firstly, the number of people exceeds everything possible. And there, of course, there is crazy training internally. We'll see what happens with it. Yes. This, of course, is very interesting. And it's interesting, by the way, we always talk about AI, that whoever releases EGI will surge forward and, well, break away from the stratosphere, right, stop peeing, pooping, that is, break away completely. What if China does it first and, well, or something that can be called AI, Asi, and all similar smart, you know, and cool words? We'll see. It's very interesting. But there are very many cool engineers there now. And I, again, met on Monday, was in the Anthropic office, met whom? With whom? With a Chinese person. In ON, I corresponded last week, corresponded with whom? With a Chinese person. Here. >> Yes, there might be a Chinese person who was born, for example, not in America, but in Canada, and someone in America, someone else, but a Chinese person. >> About 15 years ago, when I was more or less involved with mathematics, publications and so on, you read an article in mathematics in English, and in the authors, there were definitely Chinese people, or all Chinese people, or partially. So, even at that time, they were very actively advancing science in completely different directions. In mathematics, for example, it was very clear that they had really gotten into this territory. And they are the first to implant a chip in their heads, right? You know, children, and they won't have problems with the law. I, by the way, think that there won't be problems with the law regarding implanting chips in children. Perhaps they are also allowing more children to be born, right, because of this. And as soon as chips appear, >> right? The second child with a chip. >> And, by the way, this is interesting. Do you know interesting data? If you look at reports, then in China there is now a shortage of people for work. This is often worth looking at global trends from the point of view of economics, migration of people, facts. Firstly, in Russia, there is a huge, huge shortage of people for work. That is, 100% very low unemployment, there are not enough people in all spheres. In China, there is a huge shortage of people, and they are interested in people coming to them, in people staying with them. Although, again, the brain doesn't process this. It seems like it should be different, >> yes? We just talked about it four or five episodes ago that a talent visa appeared in China. And, accordingly, >> they use all possible means for this. >> I wanted to ask Tanya, you, Sasha, also, I think, said that you haven't used the ChatGPT integration with Adobe products, with Acrobat for working with PDFs, with Photoshop. No, haven't had time to try anything interesting with Photoshop? >> I want to try it today. Today I plan to try it, because after Nanobana, I completely forgot about Photoshop, but I didn't really need it at such a professional level for interiors. And then you sent your things. And I want to try playing with some presentations today, because, I think, it will be more serious detailing. >> Uh-huh. If it works out, let's add this short video. There, I just asked Photoshop to highlight people in the photo and make the background gray. And it, accordingly, in addition to giving the processed photo, also provided the tools that are in Photoshop, so that you can independently adjust how much the gray shades should be, or, accordingly, leave it in color. Uh-huh. >> I, Il'nar, I think this is the future of all the design programs that I use, and I don't understand why this hasn't been done yet. Just in AutoCAD. You can actually sit in voice mode and say: "Draw a line, you know, 12 inches, wall width five." You don't even need to tell him the wall width, he knows what the wall width is in America, conditionally speaking, if you've already chosen a certain plot style. That is, why isn't this here yet? It's so easy and logical, right? And then ordinary people will really be able to use these programs. Or, for example, very, I don't know, how many commands are in AutoCAD, you can still create codes, invent something else. And they all, I myself sometimes sit and remember what that command was called, which you rarely use, right? And there, for example, not fillet or something, I don't know, uh, I remember it visually, round the corners, you tell him, and he does it immediately. Why hasn't this been created yet? That is, this would be really cool. Just a dream. >> Tanya, it will be very soon. And now I'll talk about this a little too. A very important thing happened that will lead to this very quickly, most likely. A small clarification, right, the difference between Nanobana and, say, this tool with Photoshop in ChatGPT is that Nanobana generates. Yes, and remember, we posted in the last video where Sasha and I were in the image of cowboys and English dandies, etc. And here, it turns out, that is, in any case, it tries to preserve facial features, but it still generates, and distortions and other things still arise. And here, in this mode, you send the photo to Photoshop and use Photoshop tools for this. That is, for example, >> yes, if you need to highlight an object and change, I don't know, color correction or something else, the image is not regenerated. And in this editor, Photoshop uses tools that process it the same way as if you, for example, did it yourself in Photoshop. And, uh, very significant, right, speak. >> I'm afraid you'll lose your thought. You know, before Nanobana appeared, designers used Photoshop precisely to get when there was no possibility or time to do 3D visualization, Photoshop was used, and quite photorealistic images were obtained, a combination of these layers. So for me, this is all a bit on one plane professionally, so, yes, it will be interesting. And then, I think, when you've Photoshopped it all, you can send it to Nanobana, and it will arrange everything, create perspective, and everything else. You need to try it. I recommend it to everyone. I know some designers are watching. How to call this mode? Write Adobe Photoshop, do this and that, and accordingly, attach the photo. The mode is buggy. I had to attach the photo several times. It says: "I don't see your photo for some reason. Can you attach it again?" And so you attach it, but then it starts working, they'll probably fix it, but for now it works a bit strangely. >> And the tool itself, you just >> Yes. Yes, yes. Yes. Just like with Canva, with other tools, you just call it, give consent once to connect it. And then, accordingly, in this chat, you will work with this tool. And now, Tanya, what were you talking about? These are very cool things. Uh, this week, the Agentic AI Foundation was established. This is, in general, a foundation, it's open source. It's under the Linux Foundation, I think, under the management of the Linux Foundation, I think. And large companies have pooled their technologies there. Uh, for example, Anthropic gave their MCP protocol there, saying: "That's it, now it's public domain, use it if you want, right? Develop it further." Also, accordingly, for all this to work. OpenAI gave their AgenticMD there, right, it's a file that tells the agent what to do in this situation. Google sent their developments there, and accordingly, there is a whole circle of companies that have pooled their technologies there. So, for example, MCP used to belong to Anthropic, right, you could use it, but it was still company property, and now it's all pooled in one place. What does this lead to? It leads to the fact that these open protocols, if everyone starts using them, they create standardization. And it turns out that, for example, if AutoCAD wants to add something like this, agent behavior, connection to links and so on, then there is a protocol that works. Conditionally, Open AI doesn't suit them, they replaced it with Anthropic's. Anthropic doesn't suit them, they replaced it with Google's. And all this happens almost seamlessly. Well, of course, here there are always quotation marks, because a lot of things will arise, but it will happen much more smoothly. Also, the news that Google has wrapped some of its services in MCP. This is most likely a consequence of MCP now being in the Agentic AI Foundation. And, for example, I clearly remember that Maps, I think, something else was there. Now it will be available for calling by any systems through MCP, that is, you don't need to wrap it separately and so on. And with the development of this foundation, more and more tools will become available for use anywhere practically. >> The same Photoshop, now they've collaborated with OpenAI, and that's how this story with ChatGPT arose. But if this story with the Agentic AI Foundation develops and, accordingly, the Adobe company wraps its tools according to these protocols, then absolutely any other, whether DeepMind, or Gemini, or something else, they will be able to add this functionality. It will no longer be a connection between two companies, it will be such that Photoshop provides these tools, and everyone else, accordingly, starts integrating them. This will lead to, I don't know, editors in phones starting to use Photoshop directly, whether paid or free, that's another question, but it will be much easier to do, simply because the connection starts to work. And what you're saying about AutoCAD, draw me a wall of this thickness, so many inches, and so on, I understand that in six months, a year, such a thing may appear in many tools. This is a very cool thing from the point of view of integration, of course, from the point of view that they provide the opportunity, create protocols, so that all systems can communicate with each other, so that there isn't a mess like with smart devices, right, when you have a serious limitation, you are always tied down, you don't understand what's happening. At home, I think I have 10 systems or 15, a bunch of subscriptions. That is, this is a very, very difficult topic. Yes, Il'nar, tell us what you wanted to say. I'll add something about unification too. Yes, yes. Yes. This, Sasha, is what you're talking about? This removes the fragmentation of this technology market, right, that you have a fragment that, say, Opena Aa makes, there's a fragment that others make, and they are very poorly connected to each other. And the emergence of such open-source protocols allows this to be smoothed out, and within each tool, all the others become available, this is an ideal world, of course, but all the others start to be available. The last thing I would like to say on this topic, I recently watched a couple of episodes of a series, I don't remember its English title, Pluribus, something like that, but the Russian translation was something like "One of Many," something like that. In short, >> due to the emergence of events, >> yes, yes. Due to some events, it happened that people's minds merged, so to speak. All that I knew, now everyone knows. All that others know, now I know. >> Hot, awesome. >> The idea is very interesting. And, well, what does it lead to, conditionally speaking, a girl from, say, a Thai village now sits down, takes the controls of an airplane, starts flying simply because, well, in the general mass of people, there are a huge number of pilots. This knowledge has become available to her too, she can use all of this. And when you were talking about, uh, that many people will start using AutoCAD, even those who didn't use it before and so on, this is already happening, >> because you don't know how to do something, right, you conditionally ask GPT or something else, and now you know how to do it. In essence, you know, without a neuralink yet, not directly into your head, but you still have access to data, it becomes much faster, much smoother, simpler, right, and essentially we are moving towards this. I literally watched a couple of episodes, >> but I can't get rid of the feeling that it has actually already happened. And, Sasha, what you said at the very beginning, right, increased efficiency, banks made a report, and in their report, it was stated, I think, that people note that they can now do things that were previously inaccessible to them. This is precisely about this, >> right? Well, this is, by the way, one of the main stories that we convey. No, well, for example, I couldn't do research like this before, which I get now, or I couldn't process such a volume of information in 10 minutes, or I would never have had the opportunity to process such a volume of information, for example, >> yes? Or as you say now, Tanya's pictures, right, techno, just pictures, changing them. Yes, this was not, this was impossible, or to create some micro-video. This happened a few months ago, how to use Photoshop, and now you can do it through chat GP, like this, this, this, and it's all ready. >> Yes, >> well, for now, it's a matter of time. Exactly. Yes, this is exactly the story of what we talked about driving, right, when I have a quarter of the children, I understand that my daughter, who is now 15 years old, is now starting to learn to drive, and all other children will most likely not learn to drive. Well, most likely. Or maybe for fun, maybe for the sake of it, they'll learn to drive, to supposedly go somewhere where they'll need to rent a car. But by that time, most likely, if we're talking about the states, not most likely, if we're talking about the states, Waymo, by the way, is expanding to three more cities, again, a huge amount of everything. Well, most likely, there won't be, uh, well, there will be self-driving cars, right, and there will be self-driving cars. We flew in yesterday to New Mexico, rented a car. Well, most likely, you'll just rent a car, a car will arrive without a driver. That's it. You just rent it full-time or part-time depending on what you need. Conditionally. I wanted to add a story about unification. Now there's been an announcement that Google and, I think, this topic, well, complements it, Apple and Google are cooperating for seamless data transfer between iPhone and Android. I really want to smile and say: "You'd better set it up properly so that it transfers normally between two iPhones. It's still buggy." But still, they don't, well, it really is buggy, right? It really constantly doesn't catch AirDrop and so on. So, but according to insiders, this is not public information, they are developing a new function for direct migration from iOS to Android devices and vice versa. And the first tests are underway on Google Pixel. And in the upcoming beta version of iOS 26, the function will also appear, they write, right? And, well, this topic is, of course, very useful and very good. Very good. But again, when you see artificial intelligence and models like GPT, into which Photoshop is integrated, my question to Photoshop is: what will you do when a person gets used to using your model inside? And ChatGPT will offer that besides Photoshop, you can also use these 40 tools here, right? And who will pay for your service? Or will you pay money on a micro-model basis, will ChatGPT send you a penny for using your service? That is, how will prices, how will prices be determined? This is what we are seeing now, by the way. Remember, Sam Altman promised a year ago that ChatGPT would appear this year, which would cost $1,000. And they also said it would appear in the fourth quarter, a topic where you could throw in $10-20,000, and it would solve some task for you. But we haven't seen the appearance of this yet. We only see the reduction in the cost of models via API, improvement of models, improvement of performance. And for the same $200 that I paid for Pro a year ago, I think Pro has been around for about a year, if I'm not mistaken, well, when it appeared, I've always had it. My characteristics are only improving, right, I can do much cooler, much cooler applications than before. Well, we'll see you in exactly a week. To the Moon, and on Wednesday, a special episode. And we are waiting for you at the special episode. Bye everyone.