The Turing test has become an everyday problem. It's getting harder for people to tell a human from a model. The Pope said that artificial intelligence should serve people, not concentrate power in the hands of a few. OpenAI and Anthropic are diverging in their narratives. "There will be no jobs apocalypse" versus warnings that, well, crazy global changes are coming. Anthropic and their security module found more than ten thousand vulnerabilities. Is that harm or benefit? KPMG, it turns out they have two hundred seventy-six thousand employees, and they're rolling out Anthropic's Claude to all of their employees. And why are they doing that?
And you're on the ToTheMoon channel: technology news and insights from Silicon Valley and around the world. We're now putting out episodes many times a week, and our main podcast is on Sunday. Tanya, hi. There's, there's one big-picture piece of news I'd like to discuss. I think it's important. From the outside, it seems to me, the world doesn't understand what's going on with artificial intelligence. Because I saw a speech just a couple of days ago, I mentioned it in one of the special episodes we had during the week, that in the Vatican, they announced the Pope's encyclical, where artificial intelligence is viewed through the lens of the common heritage of humanity.
And I saw the Pope's speech. He called for limiting artificial intelligence when it comes to data on conflicts, or on weapons. And I understand perfectly well that there are things that can no longer be undone. And artificial intelligence is going to keep developing. Indeed, right now this system, artificial intelligence, relatively speaking, is still a system that performs a certain task. It acts according to algorithms. There's a big emphasis there on the heritage of humanity. And the Pope, obviously, wasn't discussing which chatbot is better.
He placed artificial intelligence in that long, what do they call it, the long tradition of the Church's social doctrine. And he started with, well, it's kind of an old story. It said that artificial intelligence forces us to talk about human dignity, power, the concentration of control, and about who makes decisions instead of people. And what's interesting is that representatives of Anthropic were at that meeting with the Pope, they were there inside. And they discussed all these topics. And logically, the Pope spoke out, well, directly, directly on the subject of artificial intelligence.
And here's his phrase, yes, exactly: "The development of technology can be either a new Tower of Babel, or a city where God and humanity live together." Yes. Or: "Artificial intelligence must be disarmed, freed from the logics that turn it into an instrument of domination, exclusion or death." Well, here's what I'll say for my part. When I read that, I got the feeling, it came across to me as populism. Well, my attitude to it is very particular. Why? Because all these statements, well, they have... he, he carries a good context, about people living together with God, about the right way for a person to develop.
That artificial intelligence could enslave everything, everything that exists, everything around us. But at the same time, no matter how much he calls for it, no matter what does or doesn't happen, this can no longer be stopped. So I took it as, well, well, these are just statements, and you have no influence over any of it. That's what I wanted to say. Not to contradict him, not at all. I'm not saying who's right and who's wrong here. But you have no influence here. Tanya, what do you think about all this?
I think that, well, you know, I agree with you that the statement is rather idealistic, slightly out of touch with reality, but that, I think, is the essence of religion: to set some high ideals that are basically unattainable in earthly life, so that people at least strive toward them somehow. And the second thing that comes to my mind here is that, well, nobody really knows or understands or can even guess where all this will lead. We're all living in complete uncertainty. And how big all of this will be, it's already clear it'll be big, but how big, nobody knows.
At the same time, you know, we were discussing this before the episode, you asked a question that probably worries a lot of people: who is going to be making money? And it seems to me that this whole human need, the need to earn money, will go away altogether, and some absolutely new currency will appear in the world. Like, you know, back in the day people, I don't know, sold cows or weighed gold on scales and so on. And now, gradually, first the money physically disappeared. Here in Silicon Valley, for example, there are already lots of places where you can't pay in cash, which sometimes even bothers me.
Why? It's money, after all. So. And I get the feeling there's a scenario that might come about where all people will have some kind of, well, like everyone, many people, obviously not everyone has food right now, but many people already have some basic set of food. Some countries are still, obviously, without water, without food, without anything. But if we're talking about those who have internet, that's probably not the case anymore. And I get the feeling that money will go the same way, and some completely different realities will appear that we'll have to get used to.
And probably, really, the people who'll do best right now are those who keep their finger on the pulse and who are, well, ready to adapt. Because if people don't adapt, sooner or later it'll break them. And these two scenarios the Pope named: one is apocalyptic, and the other, on the contrary, is very idealistic. I think, as always, the truth is somewhere in the middle. The question is which way it's skewed. Well, again, we live in interesting times. Although recently I was also thinking that everyone kept telling us that in two years we won't recognize anything.
But I was trying to reformat a file with ChatGPT Pro, Excel, or rather Pages into Excel, it took it an hour, and it still mixed up all the pictures. I thought, it's probably not a matter of a year or two after all, but a somewhat longer distance, but still, it's amazingly interesting to live in this time.
Listen, you're saying interesting things. And before we move on to, I think we need to dig into the topic today of who can actually make money with artificial intelligence. And in the last few weeks, I don't know, on this question I've somehow found more answers for myself, or some understanding. That doesn't mean it's a complete understanding, some understanding. Here's what I want to say: you're saying that a lot of people promised things. And what bothers me a lot is that the heads of the biggest companies developing artificial intelligence, essentially, very often sound in their talks like what the Pope is saying.
And you sort of justified the Pope's statement by saying it's the Church, and the Church is the Church. I'm not against that. They have their own policies, their own rules, their own history. I talk about this a lot on some of my other channels and so on. This isn't, isn't the place for it, right? But just so we don't clash here. But once again I got the impression, about the Pope, that he doesn't understand what he's talking about. You know, there are cases where people don't understand.
Sometimes you listen to the heads of companies that develop artificial intelligence, and we think they're superhumans who understand everything incredibly well. And then you start running into the details, like recently there was a whole, a whole big message.
Sam, remember?
Yes, that, yes, Sam Altman, it turns out, doesn't really understand the technology all that well, for example, which doesn't mean it's bad or good, but simply people, they make statements, and they, they either stand behind their words or they don't. And so here Sam Altman says this week: "I'm glad I was wrong. I thought that by this point more entry-level office jobs would have been eliminated." And then he says: "The human part of the interaction really matters to us. I can't imagine handing that over to artificial intelligence." And you're like, this sounds like some kind of baloney, right?
You know, it's as if, so, he just won the lawsuit against Elon Musk, they're preparing for an IPO. They need the right rhetoric. Why not, in that rhetoric, start speaking a little differently, right? I mean, people make an endless number of promises in all directions. You can't tell what's going on. Like last week I said that in America they want to introduce a safety law, so that ninety days ahead they check, well, receive all versions of artificial intelligence, test them in advance, and then release them to people.
There's talk that Trump is pushing a bill so that absolutely all American models, ninety days before their release, give access regarding updates, before the update, to, well, the politicians. And they would check those models. That's the first big issue on the table. I don't know if the US will actually let something like that pass, and whether it'll work, because if it does work, it's a huge blow overall. What does it mean to hand a model over to the government ninety days before an update? Can you imagine what that is? How can that even work, that you have to get their sign-off?
And then this week that law has already been turned around. Or, one moment the US is protesting against China's development, the next it isn't, as far as their top politicians are concerned. And you realize that this question isn't settled at the country level at all, because the country is split here. There are people who believe China needs to be restricted very hard. There are people who say artificial intelligence needs to be restricted. Someone says: "If we restrict artificial intelligence, we'll lose to China."
Someone says: "No, we need to supply China with all the equipment and all the parts, supply everything, otherwise they'll build their own." And it's an endless debate. The camps are split there: Anthropic, Jensen Huang at NVIDIA. The camps are shifting a lot, a lot. Another interesting case.
There's this person, he came out of OpenAI, one of the senior, one of the founding researchers there, inside. Daniel Kokotajlo, I don't know exactly how to pronounce the name, but that's not, not the point. So he just gave an interview. He had an interesting statement. He said: you don't understand that right now, the people who own the artificial intelligence companies, and the governments where those companies are located, they will never stop, under no circumstances will they stop, from deploying artificial intelligence in their own interests, in the interests of those companies, and in the interests of the state.
That is, they'll keep expanding. And I've said in many episodes that OpenAI will block countries where, for example, using artificial intelligence is prohibited, not because of sanctions or domestic laws, I don't know, it won't roll something out in Europe, because of GDPR there and data has to be stored differently, but because it isn't interested in that, because it's interested in generating profit for its own company and for the government. Because, after all, companies of that level, worth trillions of dollars, are, well, they're, they're the country's companies.
I mean, of course they're private, in America it's private property after all, right? But at the same time these companies act, among other things, in the interests of the country, because the country is run, among others, by certain people or groups of people, clans of people, and also companies, yes, companies that own all of that, move around there, and so on. And so he said the following: neither the US nor China at this point... since there are two leaders in the world, no other leaders exist, nobody else is even close.
The head of Mistral, the French AI company, recently said that the European Union has only two years left to avoid becoming, like, bots of American artificial intelligence. But I think they have no time left, they've already lost everything completely, they're already bots either way.
That's the story with the horses. Yes.
They're already bots, yes. You can tell the story. I'm not, they... by the way, we have super news today. We have super news today. My ChatGPT pen arrived. I forgot to mention it.
I forgot about it. Tanya, ChatGPT sent me a pen. So, it's this little thing.
Oh, it even has a stand.
Yes, it's, it's heavy.
Did you check it, Sasha? Maybe that's not a stand.
Maybe it is a stand, what does it have to do with... this stand...
Yes.
You could kill someone with it. It weighs more than a golf ball, a lot more. Maybe two, three times, that's the kind of pen it is. So, there's this beautifully made pen, and it says on it ChatGPT Pro Since 24. Meaning that I've been a user since 24. That I've been Pro since 24. That I'm one of the first users. So, well, it's funny of course. I-
I think your grandchildren will be able to sell it at auction someday.
Well, maybe. I think there are ten thousand of these pens.
So take good care of it.
Yes, there are ten thousand of these pens.
Just imagine!
Well, it's silly and funny, of course. Especially giving a pen like that to me. But, but the point, but, but what's the point? The point is that... No, well, well, it's OpenAI after all.
At least they kept one promise.
I had an assistant, and when she was leaving me, she said: "Alexander, I'm such a good person, I want to give you a pen." Anya. And she gave me a pen. Not a cheap pen, I must say, quite a serious one. A Montblanc. She gave it to me because before that, someone had stolen a pen of mine that a certain person had used to sign several multi-billion-dollar contracts. And it got stolen from me. So she decided to give me a similar pen. But, but anyway, they sent me a gift. So, glory to ChatGPT, OpenAI, many thanks to them.
Somehow I ended up in the sample of people who managed to click on the email in time, and so this pen got delivered to me. Thank you very much. So, anyway, what is this person from OpenAI saying? He's saying that, as it turns out, there's the US and there's China, and the center of power in the US is the private AI company, clouds, chips, capital.
And the Pentagon, obviously, whereas in China it's the state, the party, industry, data, control. And it's important to understand that each country has different goals. There was recently a comparison of Stalin with a certain well-known politician. And one person said, said that Stalin's goal was to build the party and build communism, whereas the other person's goal is to make money. So it's important to understand that all people have different goals. And China, their goals are very clear.
The people at the top, the people in charge, they're not about money, they're not about money, they have completely different interests. And so, in all circumstances, it's the state, the party, industry in itself, and control. Obviously, at the end there's control. So for the US, for example, it's not a question of control; for the US it's first and foremost a question of huge private AI companies. All that infrastructure, who owns all that infrastructure, capital, capital, and control for the sake of capital, right? And obviously the center of power in terms of the Pentagon.
So what's the US logic? Invent faster, deploy faster, monetize faster. China's is: build the state, production, control and strategy faster. And, and, for example, the military logic in the US: artificial intelligence as an accelerator of command and planning. In China, artificial intelligence as part of a centralized mobilization system. So, and the US has a weakness: there's a big gap between corporate interest and public oversight. And in China, obviously, the main weakness is the rigidity of the system, the risk of self-deception, when you have centralized data.
And in general that's a big problem when it comes to centralized data. And in China we're now going to end up in a world of false correlations. I've always told everyone, over the last ten or fifteen years, especially when machine learning got started, big data, that in big data there's a huge number of fake correlations, false correlations, which lead to false conclusions that people draw. It's like my Whoop draws the false conclusion for me that if I drink coffee at night, I sleep better. Or if I look at my phone in the evening, I sleep better.
I mean, I log what I do, and it shows me things like that. Your sleep is seven percent better. That's a false, fake correlation. It's not true, these are secondary things that just coincided. And so for it, whether I drink coffee or not, supposedly that somehow means it's better, like everything has an effect. So go drink and you'll... but that's not how it is. Well, it's clearly not like that, that's a fact. Or the phone, right? What phone? If I look at my phone for thirty, forty minutes in bed in the evening, it has a positive effect on my sleep. That's not true. It's just that, possibly, on that same day I would have slept even better than I did.
I mean, I was sleeping well anyway, right? As my sister says: maybe you were in a very good state, and that's why you picked up the phone. And when you're in a bad one, you don't even pick up the phone.
Those are things Whoop doesn't measure.
Yes, those, those things.
State, mood.
I've always dug a lot into the algorithms of astrology, into the math that's in there. There's this concept of false correlation there, right? And people, not fully understanding everything that's going on, start drawing these false, false conclusions. It's the same as: if our child, if our child gets straight A's at school, then they'll have a good job. A false, well, a wrong, fake conclusion. And now artificial intelligence, there are simple ones, and artificial intelligence is now going to start drawing a great many false conclusions, and because of them a huge number of false laws will be passed, and a huge number of people will suffer.
Obviously some will win, but a huge number of people will suffer. And this Daniel says that on both sides, if I've got the name right, yes, I think so, Daniel, that on both sides of the Pacific Ocean in this case, meaning China and the US. These opposing countries are building systems to govern the world. And he says that... ah, what's his main phrase? His main phrase is that the deepest danger he sees is not an evil artificial intelligence, but the disappearance of the human scale.
I mean, he's saying that artificial intelligence is starting to think faster than a human, plan deeper than a human, know society better, write texts more persuasive than people, create videos, images, ideologies of fear, of hope. And that a person is gradually losing not their job, but their inner position as a subject. They no longer ask what is true. They ask what the system considers optimal. It's a transition from wisdom to optimization. And in this case artificial intelligence starts running everything. And that's his main story.
And everyone keeps saying: here's an ad for you, a product, some sunflower, a person will lose their job. But let's look at what will actually happen.
It's all happened before, yes.
Yes. And what's happening, on a mass scale. And so he says, as a visionary, that two machines will appear. On the one hand, there will be big machines of fate, big machines of fate, which will be the US and China. And he lays this out in great detail. And here I want to bring up the Turing test, the Turing test. It used to be that you could very clearly tell a human from a model, right?
And now people already, right now in the Tech Explorer research they reported, they did a study. They reported that in one version of the experiment, the model was identified as human seventy-three percent of the time. And that's an old model, Tanya. They ran the study on ChatGPT 4.5. And I see how there's a huge number of systems, for example, that determine whether a text was written by a human or a robot. And I know of a lot of cases, my sister told me about this back when my nephew was applying to Stanford, well, to good universities like that, you know, in America.
And the essay he wrote himself, personally, the machine determined it was written by a robot.
Uh-huh.
Or it flagged some high level of issues. And now I see how texts written entirely by artificial intelligence, the machine determines they were written by a human. I see it, it gets published and so on. And what's interesting is that artificial intelligence itself, if you go into the algorithms, if you go in and study it, you'll see that artificial intelligence will write the following: for example, the probability of detecting me is twelve percent. It wrote that to me recently. It says: "I can't rule that out."
But, essentially, when you have twelve percent, you can't determine it, because it's not a probability, in this case in terms of text, because after all it's text. And although there are still a huge number of people who believe a human can proofread a book better, a human can write some piece better and so on. What does it mean to write better, and what does worse mean? We know people write completely different pieces of different types and everything else. And clearly it will come to that.
Look, Sasha Mashrabov's "Higgsfield" feature film was shown in Cannes. Well, in Cannes on one of the stages, but they showed it. And it was, like, alongside, not... alongside it.
Not the main programme.
Yes, well, the film was ninety minutes or however long, and it was made by artificial intelligence. Obviously there are nuances, in terms of quality, the quality of details, the polish of certain moments and so on. But obviously all that will get fixed, and there will be films made with artificial intelligence. But for me, it makes no difference to me. As of today, I don't know about you, Tanya, but it makes no difference to me. I'll explain what I mean by "no difference". I'm already stopping looking at whether this text was written by artificial intelligence or by a human.
The question is whether this text is real, in terms of meaning, position, logic. Whether someone is trying to deceive or manipulate me with this text, and if they are, I want to understand the logic and structure of how it's put together inside. Is it great to have a certain art behind it, to see real people, visuals? Of course, it's super great, it's super cool and it's super important. But a huge number of sci-fi films are made by computer, and the scenes there aren't real.
And the fact that a person worked hard to make them, well, now a computer made those scenes, so did it get worse or better? Not necessarily. What's more, the script of that same Higgsfield film, a real director of a serious level worked on it, as far as I understand, there are serious screenwriters there. I mean, some serious people were there too. And in new films they'll be there too, those same serious, top-notch people. The most important question becomes how much this really matters to a person. And it's like, what have we come to?
To the point that in America, you know, you come to some fair, and they tell you, this especially applies to Native American fairs. When you come to Sedona, somewhere in the mountains, there are Native people standing there selling various things. They say it's handmade. And you look at this handmade thing, and you're like: "I don't want to buy this little pendant for a hundred dollars just because it's handmade," because, it's like you're supposed to buy it from him just because it's handmade.
And we're starting to see how heavily this gets exploited, because a huge number of people who can't do anything creative took a stone, painted a little dot on it, and put it out. Like, I don't know, my five-year-old son sat down with a pebble, painted a dot. That's handmade. Because they're Native people, because he's a unique Belarusian in Los Altos Hills, well, or there are twenty Belarusians living there, that's why people are supposed to buy this pebble from him for a hundred dollars.
But that's strange. And with handmade, I think we've moved into a very, kind of peculiar zone. And the same thing will happen here. What do you think about this? Especially you, from the point of view of, uh-
Yes, how-
your own approach.
Sasha, you suspect I have the opposite approach.
Well, how, how do you survive with this? Look, look, you and I agree very much on visuals, on certain desires of real people.
But what, what do we do with this truth and untruth? The untruth in the offline world that exists, and the untruth in artificial intelligence that exists.
Well, first of all, we've been living in this untruth, it's just becoming more mass-scale now. For a long time already, I think. At first it was just in a person's head, they built some wrong logical connections and made generalizations. Now it's as if someone is doing it for them. And you say: if, let's say, I'm reading this text, I want to know who's manipulating me and whether anyone is at all. Well, I think you should assume a priori that someone is manipulating you. The thing is, AI is already very good at manipulating people. I already notice it, for example, in the answers it gives me: it knows what I'll like.
And going forward it'll get even more intense, I think, because it keeps learning more, it understands human psychology more and more. And, for example, for all my loyalty to our other Sasha, and the fact that I'm very proud, and I think they really are making history, I'm not a fan of films like that, in general, regardless. I mean, I'm not that into sci-fi, for example, all kinds of action movies and things like that. I realize I'm in the minority here. I, for example, prefer arthouse, I love live acting and so on. Those micro-emotions and things like that.
I'm always interested in what story stands behind everything. And, for example, even if someone offers me a pendant with a little dot for two thousand dollars, but tells me some amazing story that I'll remember every time I see this pendant, and it really touches on my values in life, right? I mean, it'll give me something, I don't know, confidence, faith in people, a reminder of someone's strength, that someone put in a story about how, I don't know, this Native American's grandmother behaved beautifully in some way and so on. I'll buy it.
If I have that kind of money, I'll buy it. Rather than, for example, a pendant made by some super machine. But it's not the same, you know, because there's no soul in it. A soul, after all, you can feel it, in things-
I wouldn't buy the pendant either.
In films, in books.
I wouldn't buy an artificial intelligence pendant either, and I'd buy a pendant, a pendant on the second-hand market, where a grandmother a hundred years-
Yes! You see, if it ended up-
Like the rugs, I remember when we first came to Silicon Valley, to Stanford. We arrived, and there was an elderly couple moving out, selling things in their house. He says: "My grandfather brought these rugs from China." We're like: "We'll take them."
Can you imagine what a story is really behind that, and what value? The thing is, I work in that kind of field. I mean, what you just described is the most common objection I hear from clients: "Well, why can't we buy a chair like this for such-and-such? Why do we have to buy a chair that's five times more expensive?" I say: "Sure, you can buy a stool, it'll also have a seat, four or three legs. Well, the functionality is the same. It's like comparing a bicycle and a motorcycle, right? It's not quite all the same thing.
And it's not always necessarily about price, it's about how much something resonates with you. I think that what will resonate with a person is still something human, but what will manipulate them better is precisely something created by artificial intelligence, because it's already starting to surpass us on many parameters.
Write in with what you think on this topic, our viewers. And it's a very big, very broad topic. And on the one hand, in some places we agree very strongly, in others our perceptions differ a lot, and I'd like to, well, understand that and see. In any case, I have a feeling there's no escaping artificial intelligence anymore. It's like, you know, in customer support, whether a robot answers you or a real person. If a robot gives you the solution, no big deal, I'd rather talk to a robot than to a person who sometimes doesn't understand what I'm telling them, or I don't understand them. So.
And the next point is this. Let's you and I really dig into this topic today. The question we were discussing: where do you make money?
The thing is, just a couple of days ago someone wrote to me. Well, I saw the message, there's an endless stream of them coming in. I apologize, I may not reply to them, because a team goes through all the messages. If there's something, you have to break through somehow. So this person wrote that he's built more than a hundred agents and knows how to develop them, knows how to develop them, knows the logic of how to build these agents in different areas, for business and not for business and so on.
I mean, he understands artificial intelligence. And so this person presented himself as someone who understands artificial intelligence. And for me, well, for me it immediately becomes, for me that's not interesting, for me personally that's not interesting. And not because the person is bad or good. I mean the offer itself isn't interesting to me. Just like people sometimes write to me: "Hello Alexander, you have such a good profile. We watch your episodes, let us make you an avatar."
Yes, and you immediately realize the person doesn't even understand who you are, what you do, what teams you have, businesses, people, where you even are and what you yourself can do, and what kind of systems you actually build. Or I recently published an article in a certain Russian-language space, in one publication, and a person wrote there that I'm talking about some untrained agents, whereas... well, he was clearly trying to get across the idea that he knows how to build agents, that he can build them for us.
The person doesn't even know who I am, what I do, what I build and so on. I mean, for him, people, people come in through their knowledge of artificial intelligence. I mean, they want to show that they can do something within artificial intelligence. What does "can do" mean? Well, I can, I know, I don't know, how to set up Claude, say. What does "set up" mean? I know what the menu items are, or I know the menu items in ChatGPT, or I've opened it ten thousand times, or I've set up the API. That's the kind of approach.
And I realized... and the second story, I have another side to this in terms of programming. I mean, having run into this myself personally, well, obviously I have huge teams that build things, there are teams that call themselves AI native teams, for software development, there are tons of directors and everything else. I'm running into the realization that some of the people who do certain functions, for example, certain analysts, certain marketers, certain, I don't know, even sales managers, if they're doing automated actions, that I would somewhat move away from some of these people and replace them with other, other means.
That doesn't mean I'm against hiring people. But there are people you want to strengthen. You try to strengthen people, and they don't get stronger, well, they don't. You try, and they run away. I've been switching on systems in a lot of different systems, and literally in one of my projects the head of the sales department writes to me: "Oh, well, something needs to be tested there, something needs to be looked at." And you sit there thinking: listen, no problem at all, I'll just do all of this with my own system without you. Well, how else? What am I supposed to think? What am I supposed to think then?
Either I'll hire some other person who'll do this, or you'll only be doing offline actions. I mean, you'll start doing what machine automation can't, can't touch, for example. Which is actually what's been happening in the world. Well, I've been implementing automation intensively since two thousand two, two thousand three. Well, three for sure. And since two thousand four, professionally, seriously, even since two thousand three, for serious money. And I understand that, well, a lot has been replaced in terms of people. And a huge number of people supposedly lost their jobs, or retrained.
I mean, people started doing completely different work, well, completely differently. And in this discussion I come to the conclusion that, essentially, the person who'll win now isn't so much the one who knows how to use artificial intelligence, they do need to know how to use all these systems, but the person who genuinely wants to figure out and understand what needs to be done, and how it needs to be done, how to help the client, or how to help the company make money, or which metrics exactly need to be hit.
To have a very broad horizon inside. If I go out to the market today and ask who'll help me improve my sales department, a thousand companies will say: "We'll help." But in the end, at most one will genuinely offer something worthwhile, what you actually need. And they all still take on helping you. And then at the end they say: "Well, it didn't work out," for example. And as my friend said yesterday, people who offer to comprehensively teach you something, well, in a specific area, for example, I don't know, to make money with OpenAI or make money on Amazon, specifically offering to make money, they're frauds, right?
Because none of them will ever reveal, well, a person who makes money in their own business, there are businesses where money is made. You don't, don't go to a person and tell them, guaranteed, the exact structure or some specific precise scheme, some tricks that bring that specific effect in your specific place. You don't do that, and that's not how you find it. And so the people who won't just understand artificial intelligence, won't just show up, right now there will appear, as the world says, a huge number of entrepreneurs who understand artificial intelligence, or, well, who will get access through artificial intelligence.
I don't see any effect from them right now. Right now there's a pretty strong debate on the market about jobs.
Some talk about mass displacement of people, while others believe it's hyped-up and even convenient management rhetoric. Some CEO just spoke out, I saw it literally yesterday. He wrote that, it's a chief operating officer, actually, he wrote that, well, I don't see any efficiency from artificial intelligence so far at all. No matter how much we implemented, it had no real result. And it's kind of a populist phrase. Or someone comes out and says: "We laid off five thousand people because of artificial intelligence." And Jensen Huang says that you, generative...
well, artificial intelligence in real productivity appeared only recently, so it's strange to use it to explain layoffs that started earlier or have financial causes. I mean, he's saying that you're saying-
The facts are being fudged.
That you're laying off, yes, laying off ten thousand people because you were doing a crappy job, by saying that you're simply optimizing your structure properly now and implementing artificial intelligence. And are you really implementing it properly? Because if you look at many companies, like Meta, they endlessly lay people off, hire, and so on. If you look, they make an insane number of mistakes! Well, fundamental mistakes internally, and crazy mistakes. So isn't this just an excuse?
There are people who, on the contrary, take the opposite view. I mean, there are people who say, Andrew Ng, for example, says: AI will change jobs and create value. And obviously, he says, all jobs won't disappear tomorrow, but still. So, there's this endless rhetoric going on. And now, whatever news you open, that's what's happening. Another, another aspect is that work really will change, right? It will change, Tanya. Right? Well, a huge number of people really will disappear. And I see it, I tell a lot of people about this.
And people who are very tied to their work, you know, there are people who are fanatics about their business, they're ready to go into the red, they'll go into the red, but they won't believe their business no longer exists. I think it's very important for people to, well, be in a state of awareness, just genuine awareness. And to see for themselves the usefulness of artificial intelligence, not specifically in the sense that they need to retrain. I ran a company in Russia from two thousand twenty. We were merging two holdings there to take them to an IPO in America.
Well, obviously there was no American IPO, and there never will be one for a Russian company, well, not in the next ten, ten years for sure. And there was this rhetoric, at that point in time, in twenty, twenty-one, the rhetoric on the market was that everyone needs to learn to become programmers. And when we were creating the training program, including together with Ilnar, and Ilnar was the director of the educational programs related to IT. And obviously we had hundreds of thousands of people studying, studying simultaneously, millions of people, communities, big traffic, it's big traffic, a large number of people.
What always bothered me, Tanya, was: how can you become a programmer in a six-month course? Or I remember coming to the first meeting, we had just taken over the company, and I started running it. You come in, there are twenty directors standing there. I say: "Which of you understands DevOps?" Well, infrastructure deployment in IT. Not a single hand. I say: "So how do you create programs? How do you even learn programming then and understand it?" And Sasha Sugun and I kept trying to push the idea that in IT you need to study and get a fundamental education, and with Ilnar too, that you need to learn, including math, you need to learn algorithms.
I mean, you need to broaden, among other things, your horizons and your knowledge, a certain cushion of yours. And layer the technologies on top of it. Obviously, this was done simultaneously. Your business has to grow at the same time, and you have to make money. You're not a state university that's a hundred and fifty years old, living on subsidies or something else, where it makes no difference to you. You have to survive, right? The profitability of these companies is terribly small.
And you also have to stay in the market hype in terms of marketing: get your first job in ninety days and start earning two thousand dollars a month and so on. I mean, in this, in these hype-
Or we'll give you your money back.
Yes, or we'll give you your money back. I mean, you have to stay in the hype, in all that. And so you have to keep a balance. How to learn, how to learn, how to seemingly give something quick, but at the same time lay down the basics, serious basics, fundamental basics of understanding what to do, like in your profession, interior design. Can you learn it quickly? It seems you can learn it fairly quickly. The question is whether you can quickly learn to feel space, to feel people, to feel the harmony of this place, the country, this town, to understand, to immediately put together in your head the money, the children, the future, the past, and the present.
A huge number of different aspects of perception, details, nuances. To see all these nuances, to feel these nuances, to understand the importance of these nuances.
Wow, Sasha, couldn't have put it better.
The importance of these nuances. How, how, how do you perceive that?
Amazing!
Of course. How do you know whether you can wear a torn sock in a sneaker? Is that normal or not? Yesterday I was at another concert, well, another one of the kids' concerts. Once again, in Silicon Valley, in Silicon Valley I'm just, I'm in awe of the concerts, and the schools, at least my school district, my previous one too, by the way. That's where the school is located, the school district. I'm just amazed. Yesterday I was at a concert, sixth, seventh, eighth grade, choir.
In total there were about two hundred people, sixth grade about seventy, eighth grade some, and so on and so on. About two hundred people who study in the choir. These are kids, in the school itself there are about a thousand people, I think. That's sixth, seventh, eighth grade. Because schools are split, this is middle school. In the choir there are about two hundred, probably. And these kids are incredible, how they sing, how they try, and do a ton of things. And I, of course, came to this concert in the evening and again looked at how people looked.
Well, all the parents came. Not a single man, not a single one. Even if there were a couple of men in jackets, they were in sneakers. And, and then there are still elements there anyway. Well, the eye, your eye will spot it immediately, right? The belt, jeans or not jeans, what it looks like, what kind of jeans, and so on. It's an automatic detection. Not a single, not a single person. And so how do you develop that in yourself? And how do you develop it in yourself, even in terms of understanding business, if you're building a business, so as to see something in the future like a visionary, or to truly understand that you can break away from your competitors and make more money.
For real, not in the abstract, not from having read a lot of ChatGPT, books, having gone to a million courses and so on. And even if you have gone, having added that to your collection, because it's fine to go, it's great to go. There are lots of interesting, great courses that get called "successful success", infobusiness, whatever. A lot of cool things are taught there. The question is how, onto what, onto what base to graft them, right, Tanya?
Yes, yes, yes.
And they get grafted onto a person, not onto an ant. There's a big difference. Onto what base do you graft them? I mean, what's happening at that moment in time, what does it look like, what is it, really, what is it really? I think that's the main conflict, on the one hand, and on the other hand, a person's task in order to develop in the future in terms of artificial intelligence and not stand against these giants that are emerging. What do you think here?
Well, I think, you know, that time you're talking about. I remember talking to some, well, kids isn't quite the word, teenagers.
Back then they had this very popular notion: why even go and get a classical education now? Well, at least, I was living in Russia at the time. When everything is changing so fast and you need to go get as much practice as possible, maybe finish some courses, those six-month ones and all the rest. But what can I say here? I can only go by my own experience, right? As you know, my first degree, I basically don't use it in practice at all, thank God, because it's a law degree.
But this university structure, those five years, how they first taught us how to learn properly, and all sorts of subjects like philosophy that you're required to take at university and so on, all of that simply builds certain connections in your head and helps you later, even if with that university background you come to a course on successful success, like how to launch an Instagram, you absolutely do see the connections, the structures behind it, and it's much easier for you to take in the information, much easier to put it into practice.
I mean, well, here too everyone surely has their own opinion. But I'm a big fan of classical education. But maybe that's only because it's my experience and it helps me. There are surely completely opposite cases.
Very interesting, it's a big topic. I'd also like to touch on something in today's episode that concerns the model as a judge.
And there's this thing, I think it'll be a very useful case for many. And often, when someone hears it, they say: "I did some research in ChatGPT, for example." Someone says: "Let's ask Anthropic." And there's, there's this kind of test where you check one artificial intelligence with another artificial intelligence. And so a fairly big study was conducted. Actually, a series of studies came out this week, and they reminded us that you can't just assign one AI to check another AI and consider that reliable oversight.
And in the privacy block of these LLM judges, two topics matter: leaks of sensitive attributes, and the weak reliability of the judge model when evaluating complex research agents. Well, it's obvious that when you evaluate a complex research agent, that's a whole problem. What is a complex research agent? After all, we're now in systems that don't just take and give you some answer, the first one produced, generated straight ahead as tokens, right? We're in a system that thinks and performs a whole chain of specific actions. And now there's this notion: you were given an answer that's correct, but useless for you.
And there's been a lot of discussion about that too. And for example, Sundar Pichai, he was asked, the context was that he's the CEO of Google. Pichai was commenting on criticism of Google's AI search. Google admitted that AI search can interpret a question too confidently. And he replied: "I think that for such a query, the answer probably came out more judgmental than it should have been." But it turns out that today, when you ask questions in search... and by the way, AI Agent Mode in Google search has exceeded, guys, a billion weekly users.
A billion is the new figure disclosed by Google. And you can get a really great answer, but it's not for you, it's for someone else. It's whether the system gives the answer for you or for, for my wife. We have to understand perfectly well that the information differs a little, and it can suit one person very well and be a plus, and seriously harm the other. It's like, remember, there was the story when ChatGPT recommended what I should do on the weekend, and recommended I attend three events on Saturday, right?
I mean, for some person that might have been a useful recommendation. By the way, it reminds me of recommendations from programmers who live in Silicon Valley. It's the same kind of recommendation as making sushi at home. So. And I mean, you can see this system didn't understand me, and it can make my life worse. And accordingly, when one model evaluates another model, it's often the same mistake as when one person criticizes another person or evaluates them, well, or not criticizes, but evaluates them, right?
And it confirms, confirms a position without figuring out who that person is. And, well, I think it's a very interesting study. It's important to remember it for yourself, to keep it in your life, because we'll be using other systems.
So, I took Codex's code, loaded it into Claude, while building a certain system. Honestly, Tanya, this week I've started to feel nauseous at the word "agents". Everyone uses "agents" left and right, the internet, ChatGPT. Agents, agents, agents.
It's hype, yes.
Yes. And we had an episode where I talked about how, well, an agent is somewhat of a past story, it's just, well, some little gimmick. Think a bit differently, think in terms of software made by artificial intelligence, real, full-fledged software. It's not an agent. And, well, it's like how people kept trying to replace the assistant. Remember, one of the main gimmicks was: we need to build an agent that will manage my calendar. We need to build an agent that will read my emails.
Look, all these people in Silicon Valley, especially the tech billionaires or millionaires who told me about this, I've never once heard that they use it or have implemented it. Although after several years of artificial intelligence development, it's very important, a very important topic. It's much easier to build now, it's very easy now to write a really cool system that will go around and ask people about things, well, interact with them. And interact quite seriously, and work through various things. And I might even add a system like that for myself on certain questions at some point.
But we'll talk about that. So what's the point? I loaded the whole big core written by ChatGPT Codex into Claude and told Claude: "This was all written, written by Codex. Analyze what I should do." It gave me, it gave me a bunch of errors in the architecture, and in some other details, in one project. In the second project it said your project is quite seriously, strongly prepared, and so on. One project it trashed completely. And so it... by the way, that was also an interesting thing, because I was deploying two parallel projects, and the very first project turned out to be more seriously prepared and built in terms of infrastructure.
So, Tanya, its phrases. It says to me. I say: "What are your suggestions? Because I'm not sure all this should be written in Codex." It says: "You're thinking correctly. Since Codex created such crap..." It says it just like that. Well, almost: since Codex created such a mess, I'd suggest that first I build everything, I'll do the whole architecture completely, and then Codex can fix up the small things. And I'm thinking: something smells fishy. So Anthropic is pushing its own line.
But right away, the second chat, when analyzing, when analyzing two other repositories, doesn't suggest doing that, but suggests that everything continue to be done in Codex. To which I tell it that I don't want Codex to do it. It agrees, but it doesn't use that kind of narrative. That was very interesting. Obviously, I was using the most expensive version of Codex with the maximum context windows, that one-million context window, 4.7 Opus, I have 20x or whatever it is. Well, the most maxed-out version I have there. But the point is, of course, when you see two answers so different in their essence, it's very interesting of course.
It's very interesting, of course, and to understand-
I wonder why that is.
Yes. And another interesting thing it gave me, wrote to me, and, there were discussions, I wrote some articles, and people there said: well, what do you expect if you... ah, or in our comments on YouTube, on ToTheMoon, I think people also wrote. Write comments, we read all of them. Someone there wrote: go open the code that was written by, for example, Codex or Opus, they wrote this code. If an agent wrote the code, then you can't work with it afterwards, right? I mean, that code is already unfixable. Only a system can develop it further. And indeed, that does seem to be the case.
And here Claude writes to me that it's immediately obvious. You know what it wrote? Traces of Codex. It's immediately obvious that this is Codex architecture. I kept wondering whether it meant Codex architecture or artificial intelligence architecture. But I didn't waste time asking it that, because I had my own more important tasks after all. I want to show off the pen once more. So, their tasks, the tasks were more important. But its reasoning and these topics, they're very interesting, of course. And I, that's it, I recommend you watch the special episodes I've been recording lately.
They're very much on this topic: that AI native software will be created, well, that is, software fully created by artificial intelligence. And essentially, there will no longer be a question of what the code is like. We'll stop caring about the code. Just like Tanya Tsvetkova isn't interested in what code and how the software is written in Figma, or in Nano Banana, or in Archicad, or anywhere else. She's not interested, what difference does it make what the code is like, and what the code quality is. The main thing is whether the system solves her problem or doesn't, or doesn't solve her problem, right?
I think that's fundamentally important. Write in too with what you think on this topic. It's very interesting, a very serious topic. So, what now? See you in exactly a week with our full lineup. Ilnar will definitely be there. And watch our special episodes during the week, there will be more and more of them. Write in with topics you'd like to learn more about. Bye everyone.