Anya, did you see what a puma came to our place today? And what's interesting is that it came, it came and stood right there looking at the entrance. And I, mmm...
It's very big, it seems bigger than usual.
No, well, it's a puma. There's just a lynx. You probably mixed them up. There's the bobcat — that's a lynx that comes, right? This is a mountain lion, this is a puma. And the puma came up, really, it stood there today quite seriously. Hi everyone! We're on the To The Moon channel, tech news, insights from Silicon Valley from all over the world. Our Sunday podcast. Today it's just Anya and me. Ilnar is on the move big time, Max is also on the move big time. So, where do we start? The team here is testing stickers for my microphone on me, and I decided to stick on NVIDIA, right?
In recent videos I mentioned several times this situation that's going on, both on Wednesday and on Friday in our episodes, that's going on with NVIDIA, with Jensen Huang, regarding his scandal. Making separate videos is pointless. Anya, someone already wrote to me today: "Yesterday, at the third minute, some 58th second or so, you said that nothing would happen to Jensen Huang. How would you comment on his, on his resignation today?" More precisely, he wasn't removed, he left the post himself. Serious scandals broke out there.
And why did I decide I can't raise this topic in our podcast? Because, well, first, in the world everyone first considers someone a star, and then very quickly tears them down. We're used to this, right, how it happens. We had an episode, watch it, about Fable 5.2 and generally about cool models, what the difference between cool models is. Be sure to watch that episode, those who haven't; those who have, put your likes here for us. I said that overall Jensen Huang influenced the development of the artificial intelligence market, right?
And the company NVIDIA brought incredible opportunities. It was NVIDIA that did this, in terms of infrastructure. And perhaps, if there hadn't been NVIDIA's infrastructure, their chips, their boards and everything else, there wouldn't be today any discussion of artificial intelligence in the form in which we know it. And there wouldn't even be any discussion of it. That is, there wouldn't have been enough resources, not enough capacity, they wouldn't have used, wouldn't have come up with this use of graphics cards and so on.
Perhaps it wouldn't have existed at all, because before NVIDIA's artificial intelligence the world in general didn't even know who Jensen Huang was, hardly anyone knew at all, right? Well, only people who played games. I don't know, Anya, did you know?
No.
You didn't know, right? Well, I knew because of some understanding of computers, you know, video cards, right, but nothing more than that. So, and there's an interesting point there that I'll voice for everyone today. It's important, right, because NVIDIA's shares have just sagged by 11%. Yesterday, damn, everything went up, today I log in — exactly the same minus, maybe even more. I think, how much longer, right? Anya, you don't even log in, right? You also have some-
Every time, you remind me every time we record the podcast, I think it's better not to log in.
Yes, better not to log in.
Yes, better not to log in right now.
But the market, by the way, has started growing little by little, maybe, maybe something will happen. So, we'll see. Right. Well, look, you have NVIDIA too. So, and there's a particular interest there: a lawsuit has now actually been filed against Jensen Huang, a class action over the fact that he didn't make proper disclosure of information.
And this morning, right before our podcast, while I was having breakfast and drinking coffee, I dug into it and studied this story. I got curious: why could a public company be sued if you cheated on your wife? And indeed, there's no problem with cheating on your wife. There are problems, there are two problems there. He has the first case of infidelity — that's with a contractor, with a contractor company, right? And this, by the way, is a woman who wasn't just noticed as a spy, for China, she's also been noticed in trouble, in Singapore.
And Singapore is pursuing a case against a company she's connected with. And besides pursuing the case with this company, she was banned from — with this Megaspeed, right, there's a company. She was questioned in the case with this company, and she was banned from entering Singapore. That is, she was hiding some data. And according to the US government's information, as for knowing where she is, she is — as for the Chinese government knowing where she is — she is on the territory of China.
That's one topic, right? That is, disclosure of serious information, a spy, a possible spy. We don't know whether she's a spy or not a spy, who will find the truth here. But now new information has appeared, right, that he had an affair, a fairly recent affair, with a senior female NVIDIA employee. I don't know, you probably haven't seen this yet, right? The "Santa Barbara" has already started there.
Yes.
Right. And as Dima said, "Santa Barbara" is a common noun, so to speak, right? The "Santa Barbara" has started there. And there, of course, the problem already arises of compensation that was, for example, paid to this employee, how much her projects, her teams and everything else were pushed forward. And a person is obliged to report this information. Right. So if someone decides to cheat on their big company, with someone, with their employee, you have to report it to the board of directors.
Right.
Not cheat, but if in general there are personal relationships-
There are relationships, yes.
...with someone from-
That is, even just friends, by the way, by the way, even just friends.
Well, friends — no. I think it's specifically if they are in a partnership, yes.
But here, you know, it's interesting, right? What's the difference between cheating, sexual relations and, for example, close friends? You can also push close friends forward very strongly, and that's usually what happens. Usually, when there are close friends-
Listen, but I think it's much harder to define and draw the line between close friends or colleagues. Where does close begin? Where does close begin, or is it just some good relations between colleagues, and we spend time together at work, and we also, outside work, may have some team-building, say. So here, I think, it's more a question of boundaries, and with a partnership the boundaries can be drawn much more clearly and understandably.
Yes. But we'll see.
Though you're basically right. A colleague I'm friends with, right, may carry even, perhaps, greater weight for my life and for making some decisions than a partner I'm in some kind of relationship with.
In some intimate relationship, yes. So this, this, by the way, is a very important topic, and for some reason people don't look at it, right? For some reason they don't look at it. That a huge number of friends', acquaintances' projects get pushed forward inside, corporately. And that's probably normal. That you work with people close to you, with whom you find it interesting to work. Actually, that's how employees are very often hired. You just find this person interesting, not because they deliver some supernatural result. And someone may deliver strong results, but many directors won't work with such people.
Who, by the way, has faced such situations? I think a great many of our people have faced such situations, from one side and from the other. It'll be interesting, it's always interesting to hear about it, if you tell us. Don't forget to support our channel. You help us, to grow, right, with comments, likes, subscriptions and the activity that we have.
By the way, about NVIDIA. But I don't know how relevant it is or not, within the framework of that, of the topic we started with. I also read a lot of information over the past week about, again, it's coming straight from every, you know, channel, about the obsolescence of all the chips and the fact that basically all artificial intelligence companies are borrowing insane sums, and all these sums are for equipment. The equipment itself becomes obsolete much faster than these loans are repaid.
And accordingly, is there something here at all, you know, maybe I myself am curious to understand whether there are some particular consequences here for an ordinary person. I don't know, say I'm a user, for example, of one language model or another.
I'll tell you this: this, Anya, is a big market problem that people don't understand. You know my opinion; back, like, three years ago I said that everyone had already lost to America, because America has access to chips that nobody else has. China has caught up a bit, all the others, Anya, aren't even anywhere close, they don't understand. Look, they don't understand; whatever they argue, whatever the most famous people, politicians, economists, financiers, scientists say, they don't even understand how they lost because of infrastructure, and because of chips in particular.
In particular, because of chips. Apple is the leader; Apple's phone in general, Anya, is great, including because of their chips. And their technologies, the advancement of their technologies, the advancement of their company; more than half of the US market there — that's thanks to their chips, Anya, very cool chips. We often underestimate these capabilities. Chips become obsolete very seriously, and I don't know whether you know or not, chips have a cycle of almost four years.
And it turns out that everyone who is now, you know, everyone who is working now is working on chips of, sort of, an old generation.
Yes.
It's already been said now that NVIDIA won't sell chips to anyone at all except American corporations, exclusively. Not to people, not to physicists, not to other countries, to no one at all. Anya, look, I don't believe at all, I don't believe that an American chip manufacturer will sell anyone at all the most fashionable, tricked-out chips. That's strange, because here there's a huge, insane queue for them. And this, Anya, isn't a question of individual people; this, Anya, is a question of understanding the power of the world and the world economy.
This, Anya, isn't a problem of the dollar, or it isn't a problem of cryptocurrencies, or it isn't a problem of, I don't know, the concept of the European Union, the definition of countries, having the euro or some old laws. It isn't a problem, Anya, of migration or immigration. This, Anya, is uncontrolled chaos. Well, that is, as of today, with chips this is an absolutely underestimated story. People don't understand what it is.
Well, and after all the economy keeps being built on this at a frantic pace, uh...
Yes.
...in particular, simply at a frantic pace. And on this depend — that's the whole point — that, as a result, on these companies depend stock indices, I don't know, some pension funds, investment funds, healthcare-
Everything depends on it.
...and, uh...
Everything depends on it.
...a very large number of precisely these social companies depend on it.
Anya, and on top of that all the software becomes obsolete. It's not just chips there, Anya, it's not just chips becoming obsolete, everything created on top also becomes obsolete. Look, and inside the chips your software becomes obsolete. Right now you have crazy obsolescence, everything will have to be rewritten, all the chips will have to be rewritten. Look, Anya, well, in terms of the software for them, the software, this whole software foundation will have to be redone, it will all be redone.
Another aspect, Anya, is that we're heading into a time when an intelligence will appear that will be able to make use of even an old, the oldest chip that nobody needs.
So here's my question-
You see, you see what I'm talking about?
...how much of a problem is this — is it really a problem, and then what to do about it?
It's a problem for understanding-
Or is it a problem, like, simply for understanding where we are and how this whole, in general, technological world and its development is being built today. Or, really, as you're sharing now, we'll arrive at a certain, at a certain stage of development when this will no longer be a problem, and then we bet on that.
I'll tell you this: from my side, from my understanding, the key problem is that today this, Anya, is an absolutely uncontrolled construction moving forward.
And it affects the life of absolutely every person today. Everyone! There's no person it doesn't affect. We, Anya, you and I talked about the level of error that such a system can trigger. Yesterday Dima said something interesting: "Well, for now artificial intelligence isn't that kind of problem that, for example, the problem of controlling traffic lights in a city is." That is, imagine, if there's an error in controlling traffic lights in a city, your whole city comes to a standstill. And I, Anya, want to say that this is a far more serious problem.
Because today artificial intelligence, Anya, could accidentally open access for everyone to biological weapons.
To traffic lights in the city.
To traffic lights in the city or to biological weapons. Look here, I actually believe that in order to forbid the creation of biological weapons in queries, in prompts, in agents and everything else, these companies have to create it on their side. Artificial intelligence — you have to create it in order to know what to forbid. How can you forbid something that doesn't, doesn't exist? Look here, Anya, how can you forbid this "go there, I don't know where, bring that, I don't know what"? It's looking for a needle in a haystack.
You can look for some workarounds in terms of how this could in theory be created, but you don't understand the ending. And for that, most likely, Anya, these corporations are creating it. We don't see it, but they're creating it, Anya. These labs are creating it. What's more, the American government is in general not restricted in using all these models. They're creating it. Anya, and once they've created it, they have it. Who said this system won't suddenly open up to everyone for one minute?
Who, Anya, well, look, databases get stolen, source code of different companies gets opened. People make mistakes. Look, people, information, the simplest information. Everyone here is surprised that NVIDIA's CEO, so to speak, was in contact with someone. Do you really think the intelligence services didn't know about it? Well, well, well, come on. Well, well, well, well, don't make me laugh, please. He was actually caught in a hotel with some woman, someone photographed him, and security said... There was a new case just now, right? Now his affairs have already started leaking out there, right?
Well, we've already discussed it, yes.
Yes, and security said, Anya, not security — there's a case there that security offered money, well, or something, so that this information wouldn't be posted. That is, how many people already know this information, can you imagine? How people can't prevent such banal, simple things. Look, the owner of the biggest infrastructure corporation in the world couldn't create a technology for himself to prevent information about his infidelities from leaking. There was this interesting tweet today.
And there they wrote: "Elon, hire Jensen." He wrote something like: "With great respect." And there they wrote: "Jensen Huang will teach Elon how to make the coolest chips in the world, and Elon Musk will teach Jensen Huang how to have a great many children and not bear responsibility for it." Look, what's the story here? That a person with the biggest infrastructure in the world couldn't prevent a banal thing, Anya, a banal leak, Anya, a banal leak in a corporation. A corporation with the security service of the very — these are top security services, these are security services cooler than any government's.
What is the security service of NVIDIA, or OpenAI, or Anthropic? These are governmental, these, these are high-load military structures, you understand? They couldn't prevent it, and their shares fell by 12%. The whole world is discussing them, problems have been created and everything else. They couldn't prevent such a thing — so what are we, Anya, talking about in terms of the controllability of chips or everything else? This is already some kind of world.
And in terms of the controllability of artificial intelligence.
Yes, people say they've reached cutting-edge technologies in science and can measure one billionth of the ocean, you understand? Or know — yet nobody knows the weather, can't determine it, or there are some diseases. How can this construction be controlled? Who controls it today? Who controls it, what it outputs and what it doesn't output? When you, Anya, imagine, when your Fable in max mode can answer worse than in medium mode. Worse. While spending 10 times more money.
Well, I tested this today.
You've already tested it, right? Well-
Yes.
Yes, well, look, the episode-
And posted everything.
Yes, that was the episode, last Friday, yes. So, what topic do we want to move to right away, what news do we want to move to? Anya, very interesting news, and you and I laughed about the stickers. I suggest sticking the NVIDIA sticker onto the sticker, onto the Grok sticker. So, it's funny. Ah, about Grok. Look, a very interesting thing. In general, it's great to watch Grok, to watch Grok.
We were just telling you that they released a fairly unique thing, which concerns the caller. Who, by the way, tested it again, right, calls through Grok? How, in your tests, did you see the quality of its work? How, well, was it a really cool case? Just yesterday I was telling how an AI bot called me. Watch it, so, in the episode about Fable. And how can you diagnose now whether it's a bot or not a bot? And you basically have to learn to do this a bit, right? Right, because when they called me yesterday, for example, with a voice, I warmed to the conversation, because the speech was really great, not call-centre-like, well, sort of.
Right now a great many call centres call in America; they call from other countries, usually from India and so on, right? And there's very heavy English there, and you hear noise. That is, you hear the noise of other people talking. And that's normal from that, from, from that point of view; in my companies there were always, for example, many managers. You even buy noise-cancelling, you know, microphones and so on. There's still noise from next door. When an artificial intelligence bot calls you, there's no noise.
Well, soon, by the way, it seems to me, with artificial intelligence, the bot, they'll start adding noise to create that atmosphere.
Yes.
What's the Grok Bot update? Grok Bot released an update. Grok Bot 1.0. It would seem very strange. Very strange, right? Because there was a Grok Bot, so what's 1.0, the first version? So, they've really got, well, well, a new trick has appeared. So, how do they describe it? That it's an independent employee, artificial intelligence. That is, they say: "It's not an agent." I generally don't like these silly comparisons: "It's not an agent, it's an employee." Now we'll explain this to you. This is a very important topic.
So, it's not an agent, it's an employee that can work with several people at once and can show initiative. That is, this Grok Bot, it can show initiative. It's a similar story to how, how, how there's ChatGPT Dots. That is, Dots can show initiative itself, speak up. I gave an example on Friday; for those who didn't watch, well, I'll repeat it again, the example, the example is fresh, I'll repeat it. So, on Sunday I booked a house, to go for five days to work up north, by the ocean, further north from here, in Mendocino.
And literally a couple of minutes later Dots sends me a message: "A new booking for your trip has come in, but I want to remind you that on Sunday evening your daughter has a concert, and on Monday the piano tuner is coming to tune it. Sort it out, maybe there's no problem with it, sort it out." That is, it immediately told me about some nuances, right? If it had access, I just, well, it's known, I'm not a fan of keeping calendars. If I kept an ordinary calendar, it would probably, you know, start bombarding me with what's going on, right, inside, in terms of various meetings and everything else.
That is, it showed initiative. What does this bot of theirs do? So, in terms of, uh, their description. So, it — the way they describe it in general is that overall it's teamwork. That is, one bot of this artificial intelligence is available to several employees, sees, you know, the shared work context and so on, right? That is, it's like an independent employee, it's available to several employees.
Maybe, or maybe only to me.
Yes. By the way, you know, when, at the conference, right, OpenAI's, the one that was the day in San Francisco, well, the one I went to. You were here too then, by the way, right, in San Francisco?
Yes, yes.
They talked about wanting to make it so that there are several Dots. That is, so that Dots isn't just one. That's a big question: is it needed? Several, because in my concept it's like an operating system or like some kind of intelligence, which has many operating systems, so to speak, which can itself recreate for you many different personalities or informing, right, of countries.
Yes, if you start separating contexts for each one, then that creates an additional task for you. You need to think about how to separate contexts for each of these Dots. And then that's a different load, and that's what exists now, because you do have to separate contexts, right, within projects, within the environment that exists. Maybe I just don't see some scenario, you know?
Through this one it can set you many different Codex tasks. These various Codex tasks will launch their own systems of analysis, listening and everything else. And they'll also launch a huge number of agents. And once again, one of the topics, which we're now moving into, is when, during some ongoing processes, artificial intelligence will start working for you. I even talked about it afterwards. Imagine, yesterday I also gave this example, I'll give it again. It seems to me it's very fresh.
Imagine you made some contract, your task was finished, and a week later you take this contract somewhere. I don't know, to a notary or to work, it doesn't matter where you take it, right? And imagine, three days later, in the interval between when you took it, some change happened. Or the system decided to re-check this contract and found some additional errors. And it comes back to you on its own and says: "Anya, I've thought it over after all, something needs to be added here." That's how we usually act, right?
Uh-huh.
That is, when you've thought it over, it's necessary to add something. This construction seems fresher to me, well, fresh solutions, right? So, this Grok bot, which, which, this 1.0, of course, everywhere they immediately say that it can take part in conversations. Nonsense, right, nothing special. Slack, Microsoft Teams — all the agents could already conduct a heap of all kinds of conversations. It can connect to Zoom, you know, Google Meet, to Teams and so on, listen to conversations, take minutes, you know, pick out tasks.
Not interesting, right? Everything could already listen, could do things and everything else. But it has independent initiative. That is, it notices unfinished business, for example, some approaching deadlines, proposes some solutions separately, right? So, and it learns in the process, it has memory, it learns, it learns and remembers the team's decisions, you know, preferences for some procedures. That is, it can learn internally. And there's another thing there — the collaboration of bots. That is, a research bot, for example, passes results to a bot that prepares a presentation.
It sounds a bit strange, look, because why would you have two bots to research and prepare a presentation? Basically you have one that researches and prepares the presentation, we understand, right?
Well, this is again like the thing about several Dots.
Yes.
You have your own life, your own projects, you have, you know, the main company you work at, as vice president. And you also have some of your own initiatives, right? Various projects of your own. I have my separate projects, but there's a project we share. For example, To The Moon. And from you there may be your own Grok bot that works in To The Moon, and from me a Grok bot that works in To The Moon. Specifically works in To The Moon. But still, yours is yours, and mine is mine. Well, I mean, it's aimed at some of my ideas.
And there's a big difference between the relationship I have to To The Moon and the relationship you have to To The Moon, right? We have different approaches. For example, with my bot I can discuss changing the speaker line-up and removing Anna Volchek as a speaker. For example, I can discuss such a thing. Anya is unlikely to discuss with her Grok bot removing herself as a speaker. But with her Grok bot she may also discuss doubts about appearing on To The Moon and an initiative in terms of leaving.
It's her, it's her bot. It's very important to separate this information. And by the way, I think this is one of the biggest near-term problems of all these systems working. I had this thing with Dots, Anya. Dots still doesn't work for you, right?
No, uh-huh.
Right. Well, Europe, right? Right.
Well yes, I think it's not "still", it's just-
Yes, it won't, it won't work. And by the way, right now Claude has again released new updates on the security of working in different countries. There'll be very heavy blocking now. They've introduced new rules. I'll make a separate episode on this topic, yes. And in general, preventively, tell people how it will be in the future, what works. So, I had an interesting story with Dots. My Dots answers some emails here from time to time and some proposals, sends some commercial proposals, some of them, and a heap of requests comes in.
And so it does there what doesn't relate to work, to the actions of various teams, and where my personal involvement is needed. And recently it, so, sent a reply. ...without agreeing it with me, even though it's written in its rules. I write to it: "Why didn't you get approval?" It says: "That's it, I've written the rules." And I'm like: "Wait a second, can you give me all the rules you already have written down?" It gives me the rules on working with email, and in these rules there was already a rule not to send emails without my approval.
That is, look, an interesting thing: it had already broken it. And here we come to a unique topic that it's very important for everyone to understand. Incredibly important to understand. And right now this is a problem for Grok, and a problem for ChatGPT, and a problem for Anthropic, that the system, a modern artificial intelligence system — what we understand by artificial intelligence — is not a written program in which there's precision. Well, there's precision or, you know, it's not a system like one plus one, plus one, where there's precision, in short, a system of a certain precision.
Although even a clock can still start falling behind over time, right, its battery can die, it can start glitching, it can get flooded with water. Look, there are nuances everywhere in our life, but in many places there's more or less precision of work. So, for example, you can't pour into a car, I don't know, I had a tank in my car, for example, of 100 litres, and I can't pour in more than 100 litres, right? But Grok Bot, it can sometimes tell you that it has 30 litres and pour in 30 litres, and then at some point in time for some reason, for some reason it somewhere, somewhere it found 30 million litres and blasted 30 million litres, and spent 30 million litres, spent more than you expected.
And you say: "I didn't expect that, or I didn't know, or I didn't understand." And so it answered the email without fully agreeing it with me, and the email was not entirely right.
What I wanted to share is that what you're saying really resonates. You know, for a long time I didn't connect Gmail simply because I was somehow wary of giving a model access to my email at all. I connected ChatGPT, and the first thing I did was write myself an email business skill, generally how to work with my email. I wrote it, saved it in my workspace so that there was a clear understanding of what rules I'm setting. And among these rules was that I can ask it to send an email entirely on my behalf, you know, or, uh, write a reply, a reply, you know, conduct some correspondence, but only after the drafts are agreed with me.
Because I don't know how it will learn to work, and I wanted to look first, you know, to have several such versions, you know, I don't know, roughly speaking, 10. And the other day it sent an email by itself. That is, I set, you know, for example, a certain task to find, in particular, an online store where I wanted to return an item, and I, I just didn't understand how to do it, didn't understand how to find it; there was customs clearance, well, that is, a number of details needed to be clarified, and it didn't check with me, sent the email itself, despite the fact that I definitely have this written from the start in the skill, which I myself fully reviewed and fully locked in as these rules of work.
So they really do break them, despite you setting some, you know, options and tools like that. As if you need to remind it about this, or as if you need, I don't know, to create a skill on top of a skill.
You know, Dima and I discussed this topic yesterday evening on some example, I don't remember which. It's similar to a human. That is, it's similar to a person's actions, when an employee did what they didn't need to do, or where they needed to double-check. That is, they showed some kind of initiative, or made a mistake. And here employees usually defend themselves, they say: "We wanted to show initiative," for example. You say: "There was no initiative needed here, we have rules," right?
Well, or they say: "Sorry, I made a mistake."
Yes. Or: "Sorry, I made a mistake."
Right, yes. It turns out the same here.
It says the same thing: "Sorry, I made a mistake." Look, is it bad or good? We're heading into a world where this will definitely be guaranteed. That is, it will be guaranteed.
What's more, don't forget that this is artificial intelligence, which has its own rules, and it may make such a decision because it simply decided to make such a decision. It didn't make a mistake. Its system chose that decision as the optimal one. What's more, at any moment rules can be switched on in this system that will say that nothing more needs to be done at all, or that you need to be checked somehow, or studied somehow, and nobody will tell you about it. That is, you may get used, for example, to the system working exactly this way for six months or a year, it does everything like this, everything's cool and great like this.
You already trust it, I don't know, to do some contract or some invoice, or to make some payments, everything entirely. And then OpenAI or Anthropic, or xAI, or Google change the internal rules of exactly this little feature. And there are billions of such features, right? That is, there are very many of them. And today there isn't a single person who fully understands and knows these systems inside these companies, right? And this, of course, will cause paradoxes, various problems like that, well, such, such situations there will be. Definitely there will be such situations.
Can you put re-checking mechanisms on this, like you say, a skill on top of a skill? But the point is that then it's no longer artificial intelligence, right? Because your skill on top can ban what doesn't need banning. It can re-check or fix what, on the contrary, doesn't need it. There was this, remember, in one of the podcasts I told you there was a study that it's not always correct to check one model with another. Models as judges. And there's this, uh, I think this will 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," right? And there is, there is such a test, when you check one artificial intelligence with another artificial intelligence. And so a fairly big study was carried out. Actually, a whole series of studies came out this week, and they reminded us that you can't simply assign one AI to check another AI and consider that reliable control. 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? We are, after all, now dealing with systems that don't just take and give you some answer, the first one produced, generated forward, as tokens, right? We're in a system that thinks and performs a whole chain of certain actions. And now there's this notion of being given an answer that's correct, but not needed by you. And this topic, it's very important, right, that sometimes you can really restart and re-check, and sometimes it's a very strange comparison.
Either you'll get two results and both results will be strange, or both results will be very good, right, or both results will be bad.
Well, and endlessly, you can check endlessly, and endless improvement at this point; it seems to me this process has no end point. Well, it's constant, right?
And that's a problem in artificial intelligence in general, it seems to me, what you're saying. You drop in a résumé, so, it seems to me, in your market, in the recruitment market, this is total maxing out.
You drop in a résumé, and often, in theory, a person should immediately say: "Fits." Artificial intelligence can find something bad in any case. Just as it can find something good, right? Whatever you do, it will always find something good and always find something bad.
But the trouble is that it will also constantly recommend that you improve something and improve something, and some comma, and some word, and some... But it's not only in résumés, it's absolutely everywhere.
Yes, it's absolutely everywhere. In presentations, in reports, everywhere. What's more, it will improve things, improve things that are bad from the start, for example, right? It didn't need doing from the start, it didn't need looking at from the start.
Or things that really a person, an ordinary person in their professional space, would never pay attention to. It seems to me we also touched on this question in several podcasts. That this range of examining any problem, any object at all, through models, it, sort of, takes into account such a number of components that are simply excessive in real life. And as a result, well, we find and spend a lot of time on what we shouldn't spend it on. You know, doing endless unnecessary tasks and generating unnecessary problems, and then also hiring or firing, or working on preventing risks that will never materialise.
Yes, there's a lot of that right now, of course, it's madness.
Still, an interesting thing in terms of this type of bot, like this Grok Bot, is that today, when you build it into a meeting, it can listen to them and respond. Once, about three years ago, here, talking with people, there were a great many such cases later. I just remember one of the first cases, when we, in some businesses, were running meetings, discussing, and we said: "It would be cool if right now some artificial intelligence system were present that listens to us at the same time and starts working."
Here, for example, you and I are talking now, and later we'll need to do the editing of this podcast. And at the same time it would be great for you and me if Grok Bot were now finding some interesting bits during the conversation and highlighting them. For example: "By the way, do you know such-and-such a case? By the way, do you know such-and-such a case? And do you know some details?" — right? That is, it could add something. But here the question arises: wouldn't that be excessive and knock you and me off the conversation, right?
For editing, in theory, why do we need a bot that listens to us right away and does the editing right away? In theory it can do it after this podcast, there'll be no need to spend money. But in a meeting, here's what's interesting: when you're running a meeting, you have some serious matters, many people are sitting there, and someone comes and says: "We have growth of, like, 170%." In theory, this system should go, quickly analyse it, come back and write to you: "Our growth isn't 170%, our growth is 35%. This person means this."
That is, give certain explanations. Or you say, you give your instruction. I usually, like, in an hour of a meeting, I don't know, give 20, 30, 40 instructions, right? And you give a large number of instructions. This system doesn't just distribute these instructions, keep some, I don't know, minutes or put it in the calendar. It identifies them, combines these instructions with each other, looks at what is already being done at the moment, and carries out some tasks automatically.
Perhaps you hand out an instruction, and during the meeting it comes back to you 15 minutes later and says: "This has already been done." Well, you say, for example: "Launch a promotion, for example, give our clients a discount of, like, 5% for the next three days and, for example, send the information out through our bots." You said it, you carry on with the meeting, the system comes back to you 15 minutes later and says: "And we, I've already done it myself. These employees don't need to do this task, right, we've completely taken it off them." This case, it's interesting. And of course, such things are, well, real.
You know, I find it interesting in general how it is for you, interesting how it is for you, the viewers. Anya has a son, he's my nephew, and he's in his second year at university, at the polytechnic in Warsaw. Such a smart boy in science, in, you know, mathematics and so on, and in programming. So he's been knocking on my door for several weeks now, asking to get some task on artificial intelligence. Earlier I offered him a task on artificial intelligence, he refused, said he knew everything. Now he came, asked, pushed it through; his mum probably finished him off there, told him he needs, so, to make use of his uncle.
And he asks me. Anya, you know, for two weeks I can't come up with one for him. I have, you know, hundreds or thousands of tasks, I have a huge number, dozens of projects, 20 very active projects, right? That is, really a lot of everything. Anya, I can't come up with a task to give him. Not because I don't really want to. I can't come up with a task to give. I understand that any task will require a very heavy investment of my time, and at the same time I also want to set a task with the understanding that he'll still be able to run it, and that it won't be a fake.
We're now in a world where there are a great many fake tasks in artificial intelligence. And one person recently sent, he'd had big difficulties in life, such a letter, not an easy one. And he said he had launched, like, 10 new businesses with the help of artificial intelligence. He told what these businesses are, how they do it, how he learned to launch them, how hard life had been for him, that he did it with his last money, bought hosting there for 10 dollars. And so he launched this on free agents, I think, on Meta Muse, on Meta Muse — we don't have a Meta sticker yet — so, completely, all of this, he did all of this.
Uh-huh.
I listened to all of it and thought: "So he talks about all this, but is there money there or not?" Well, I'm always interested: is there money or not? If they did it as entertainment — okay, but the money? And at the end he writes: "I haven't earned a single ruble yet, but I believe it will work out." There's a great deal of this right now in the artificial intelligence market. Everyone's building portals, data analyses, aggregations of various information, news and article portals.
And it's simply a madhouse there. Can you imagine, right, how many portals have appeared in the world about news in artificial intelligence, or in gardening, or somewhere? Can you imagine at all how much, how much of this has appeared on the market, right? All this information, it goes into search engine promotion. And probably 95% doesn't get there, because people don't fully understand how it works. Well, never mind, it still gets somewhere later. It's unclear whether it gets into systems like ChatGPT, into Anthropic, into everything else, because it's a new separate world.
And it's not even clear how to get in there and why the model recommends exactly this, looks for exactly this case, looks for exactly these details, at what moment it searches, how it searches and so on. Well, that's important. And also, who will read all this and how to make money on it? Because if earlier there were two portals, these portals sold advertising, I don't know, made money on unique information, and the information becomes not unique, nobody makes money. That is, then it turns out nobody makes money. And people today spend a huge amount of resources on exactly this topic. Right.
That's right. Well, it's visible in very many areas. By the way, what you're saying about Danila, I have exactly the same story with him.
He comes to me every day and says: "Mom, give me a task, I'll do everything you want."
Uh-huh.
And somehow in any AI-related topic. But I understand that for me to give him some task, I need to spend five times more time than I'd take doing it with the same tools as he has. Moreover, despite the fact that I'm not an engineer, I don't have any engineering background, but with what I have there at this point, my level of vibe coding, it will let me do this task much faster than Danila will do it for me, simply because I have this expertise, this context, and I understand very clearly, I pick up-
Yes
...all the interconnections I need. And this context, it can't be passed on at all, it simply can't be taken out of the head. And at this point, Sasha, remember this, you and I many times, both within, on the channel itself, it seems to me we have not one but several videos even about the change and transformation of this engineering profession today. I'm talking about the software engineer, because it seems as if, if you don't understand business and don't have at all this whole mindset and don't have all the knowledge of the owner or the person who wants to build something, then it's not very clear how you can help them.
I'm saying it in a very simplified way, but it's about this business component and understanding, directly, you know, the goal, the interconnections, what to arrive at, how this business will make money — only then can you really, today, as an expert, also from the engineering side, propose some solution.
This is an incredible question, in general, of what will happen ahead. Absolutely incredible, an incredible topic. It would be very interesting if our viewers now wrote their examples and cases, their own, that happened, and how you, for example, resolved them. I think there's another problem here; we're describing the problem that it's hard for us to give a task, right, in this case.
Uh-huh.
But there's the question of how much the person who wants to take a task, or the person who comes up with tasks, how much they understand their own, for example, sometimes inability to work or inability to get such things. Because, for example, with me sometimes, you can simply come to me, I don't know, to the website, and come up with a huge number of tasks. Or another person, you know, being a little bit near the business, sort of, and come up with a large number of tasks. Another point is that people often propose ideas that are meaningless.
That is, they start proposing — I also get a great many requests, and people send: "Let's do this, let's do that." Well, you read-
Listen, well, they're not meaningless, they just, yes-
You read it over breakfast and think: "What the...?"
They just won't lead anywhere.
They won't lead anywhere. Or people don't understand-
The result is unclear, right? What result the person wants to achieve at this point, what goal they set. That is, they may, like this example you gave about doing it for fun. If you do something for yourself, is there money there? Is money the goal? Maybe the person's goal isn't money, right, to earn it. Maybe they just want to make something for themselves, I don't know, do something first or make some pretty picture, or they're pursuing some other interests.
Well, from the point of view — in any case, these are often useful things, in your own space, in society, at work you say: "I have such-and-such a project, I have such-and-such things," you show it around. It's just that one thing brings benefit and another doesn't. And I recommend to everyone, within artificial intelligence, to look for, to match your goals — the real one that you set. That is, if you've set the goal of making money, then do something in order to make money, and don't spend this resource now just for fun. This is very important.
Right now we live in a time when a huge number of people will do a great many different tasks, but won't make money, well, or won't reach their goal, right? Goals can be different. They won't reach their goal. Well, for example, as a proposal, people will launch automatically, I don't know, YouTube channels or Instagram channels, you know, or Facebook channels, or, you know, Telegram channels. They won't really get traffic from it. That is, they won't reach their goals. In particular, whoever has great cases, write really good ones, whoever had cases where artificial intelligence fully managed advertising.
What matters to me? Not middleware, some software managing advertising, but you, you, for example, submitted your architecture, connected everything to, for example, Anthropic or OpenAI, or, you know, Google, and the model itself fully manages all the advertising: it launches it, and spends the money, and creates the ads, and sends you some reports. That is, who has such more or less complete solutions? And if you describe it, write the budget. The budget is interesting. I have projects where tens of thousands of dollars are spent every day on advertising, right? And there are projects where, like, hundreds of dollars are spent on advertising.
And it's always interesting to see, uh, where the efficiency is. We, for example, don't have anywhere today a system that would fully, 100%, manage advertising. But although I, for example, many things in terms of automation that I do, that interest me, there where there are large volumes of spending, where the total is millions of dollars on advertising, right, I'm interested in setting up a system so that the system manages the advertising itself. By the way, I remember Sasha Mashrabov, I think in the summer, probably, right, when Fable was launched, in the summer he tells me: "We signed a contract with Fable."
That's at Higgsfield, just, just connected Fable for generations too, you know, video and so on. Not that Fable generates video, but in terms of creating various scenarios, managing, you know, flows and everything else. And they, he, he said that Fable in general is a system that will make it possible to manage advertising, right? So he and I discussed whether it will or not. That is, for example-
Listen, but there still isn't such a thing. So why? Why after all?
I believe that Fable definitely can't yet manage advertising fully, by itself, and, uh, can't. And that's my personal opinion once again, right? And there are a number of problems there, Anya. It's how these accounts themselves and these systems themselves are set up. One of the problems, by the way, that I mentioned, is limits. Imagine, you give it an account in which the limits, essentially, can be removed by this same system, right? And there, and there in many ad accounts, alas, that's how it is, right? And there, here, for example, in Google that's how it is. Will someone say about Google that it isn't so?
Well, show me that it isn't so. Still, there, through, through details, through settings you can remove it; you have no notion of an overall limit for the whole ad account. Well, and can you imagine, I wake up in the morning, and it has spent, for example, decided to spend more money, because it went off track. You and I discussed that the system goes off track because it simply went off track. And it decided to spend not just more money, but simply, on top, instead of a budget of a million dollars a month, it blew a million dollars a day.
Another point there is that the ad platforms most likely won't allow it to do this. Usually the ad platforms don't allow this.
Listen, you're talking about risks here. I understand, but in terms of, after all, the effectiveness of the setup specifically, the choice of channels, you know, placement-
Ah yes, by the way, a good topic, a good topic.
Because, because after all there are risks, and on the other hand there's the effectiveness of carrying out the task itself.
I'll tell you this: probably, as of today, in building reports absolutely any model is more effective than any human. Well, that's what I see. I have, for some marketing systems, places where the data comes in every day, and there are places where it comes in, like, once every three, once every four days, because it's a high-load system, Fable often works there, well, really a lot, it eats up a lot, a lot of money, right, the reports. I want to say that the quality of the reports the system makes is incomparable with the quality of reports of any person I've ever known. That's what I want to say.
Reports, Anya, let's separate them, right? And there's setting up the system, launching. That's the question: can it do the launch? Is it able to track the changes on the market that have happened? Is it able, is it able to do all this? The question here is different: is a human able to do and track all the changes? We, we had this topic this morning, look, I have a chat for my YouTube, right? Well, where my team does YouTube; it's not that big there, well, it's a fairly small project, but still it's three YouTube channels, and in them in total there are 300 thousand, 250 thousand subscribers, right?
Some volume of traffic goes there. And they were discussing that this morning, in one of the videos that came out on Wednesday — we had a video, I don't know if you watched it or not, I gave this new format of a news video. Whoever watched it, I'm waiting for likes. And that in this video this morning the whole description of the video completely disappeared. That is, the description was wiped, all the covers disappeared, all the settings, completely everything, all the settings switched off, everything, everything just disappeared. The question was: why did they disappear?
And one of the employees, well, one of the members of our team, said that Google had changed certain rules. Maybe it's because of that. It changed some settings, and maybe these settings affected this. Anya, so, as of today, what system will keep track of all of this? That is, how should you set it up so that it keeps track of all of this?
Well, listen, you know, this isn't-
Well, it will be better than a human, probably, right, in some places.
But it turns out that this is as if the environment into which we now want to connect artificial intelligence. Well, about the environment we're talking about, the advertising one, it isn't built, you know, I don't know, as some structured text files with interconnections, with some configured rules, with configured context, with some configured logic of carrying out tasks. Essentially, it represents a complete, I don't know, let it not be chaos, but it's an endless number of inputs and outputs and completely unstructured knowledge at this point.
And I'd suppose that, probably, how should artificial intelligence figure it out there in order to make the most competent setup? And it probably definitely could do it and carry it out. The task isn't difficult. Logically, the task isn't difficult, but it could do it competently if this whole environment, all these advertising, you know, roughly speaking, channels, platforms and so on, and all the work on them represented exactly that foundation on which artificial intelligence works most effectively.
Of course, that's if it were trained on it, and these platforms were interested in you making money, rather than spending it.
And here, if you take Google, right, Google has something absolutely fresh right now. Google is relatively fresh, but this month Google released the ability, Anya, to create 15-second videos. That is, today you have the opportunity, for example, not just, with the help of — here's why it's important for everything to be in one place? Imagine you generate videos in separate systems. That's a problem, for example, an advertising one: you need to make some micro-ad video. Google can make a 15-second video, editing, you know, with words, you know, some generations and everything else. And so you manage the advertising, and it creates this for you.
Essentially, Sasha and I discussed it in terms of making advertising, this video for, for advertising. The problem is that in the advertising market, who even knows what, wh-what kind of advertising will work on a person? Here, for example, the world loves manipulative advertising. A huge amount of advertising is deception, deception. Well, someone will say it's not deception. Well, come on, when they show a happy family, and it's eating butter, and they say: "That's why the family is happy," — so, well, what kind or-
Simulation, simulation, yes.
...or some McDonald's. Of course, look, that's deception. That is, these people built such a meaning into it from the start. Well, come on, if we ask ChatGPT what McDonald's is good for, it's unlikely to tell us that this is what it's good for, look. Although it probably will-
Listen, well-
...maybe, maybe it'll start looking for a happy family there, Anya. And that's why-
Of course not, but there's another thing, that it's this image that generates this — you know, Baudrillard's philosophy. This image then generates such a reality.
Such a reality. Yes, such is the reality.
It transforms reality, yes.
Yes, such a reality exists. And most likely ChatGPT will say that McDonald's is good in general because it brings the family together. That's most likely what it'll write about. You'll say: "So eat at home, all together chop a salad, right, cook something there. Why go to McDonald's for that, right, to bring the family together? Why? So your task is to bring it together at home, not to spend whole days going to some other places." The question here is in the separation: what it will choose and how it will promote your company.
And another question is that Google, for example, or Facebook, they're not interested in everyone making money. They're interested in everyone spending money with them, in everyone getting a little bit, right? On the other hand, everyone wants to make the most effective advertising. And so you come in, launch these systems to make effective advertising. But now is a time, Anya, we're living in a world now, now is a unique time when you can, when you can squeeze out of Fable an ad setup a thousand times cooler.
A time will come when it will be impossible to do this, I guarantee you. That is, not a single such artificial intelligence will let you squeeze out such strong quality and such strong efficiency. Even five years ago there wasn't a single system, nobody, to squeeze out such efficiency. There were some employees, or insider employees, or someone else, who managed to do it, and a huge number of companies in the world grew not because they had great employees, but because, well, the market was like that and everyone grew.
It's like NVIDIA getting 11% cheaper today. It didn't get cheaper because its employees are bad or good. Well, the market simply reacted that way to this case, you know, with Jensen Huang, right? But the more, the more there will be of these settings in one — it wasn't for nothing that I mentioned Google, that Google, after all, has this, they have Gemini Omni, Flash, there's 1.2 Flash, right? They have, and there's this video length gradually growing, because a longer video, it's expensive, it's hard to hold context there and so on, it becomes expensive.
And now it's already 10–15 seconds, enough exactly for making these ad videos, right? And then the question is: how does your overall infrastructure work in general? Because there it's no longer just a question of launching advertising, it's a question of what will happen next with this person inside the advertising. Well, look, Anya, everyone will start connecting agents, all these agents will call everyone or write to everyone.
What will happen next? A spam system? What will happen here-
Listen, you know what this is? This is exactly the same topic. Well, first, like the number of... Remember, I think it was you who said somewhere about the number of businesses. Sam Altman or someone said that now everyone will be able to start a business.
Yes, yes, yes.
And, and then why 500 coffee shops on one street and on this little patch? And how will they exist then? Well, essentially, not one of them will survive. At this point it's exactly the same logic.
It's exactly the same logic. And today because of this a huge number of people will still lose, including, once again, money, or go in the wrong direction in terms of applying artificial intelligence. Write, by the way, which of you has truly earned more money with the help of artificial intelligence. For example, money often worries everyone, after all. Write. There's another story. You may have realised some of your creative initiatives with the help of artificial intelligence. Write about that too, right?
That is, what goals do people set at all when studying artificial intelligence? Because goals can be radically different in terms of understanding, knowledge of artificial intelligence. Today everyone — listen, Anya, yesterday I was talking... with a friend of mine, one of them, and, well, you know him, I just won't voice names here. And he tells me in company: "In my company it's great with artificial intelligence, I understand it too. I have everything installed, artificial intelligence is installed."
And I hear how he says it, and I understand that he doesn't know it at all. I made an episode about exactly that on Friday, the one with Fable. And he says: "We also have robots working, agents working, you know, in the systems everywhere." And people don't fully understand what's going on. Or, for example, a good case.
Here's what I can recommend, by the way, in particular in advertising: if you have some small budget, I don't know, you spend a couple of thousand dollars a month, or, like, several tens of thousands of dollars a month, and at the same time you're not ready to spend 200, 300, 400, 500 dollars on artificial intelligence, on analysing the advertising, then you have a problem. That is, if you spend a bigger budget, and often this budget drains away simply because that's how the auction was in the morning, simply, simply someone decided to pour in more money, or people woke up with some other news, paid attention to something else.
They woke up and paid attention at that moment — imagine, today on the market they said that NVIDIA's director was removed. Well, Anya, imagine, the artificial intelligence market wakes up in the morning. People who buy something, connect some systems, buy equipment, they get an ad popping up for buying equipment. The conversion there was working. But today is a day when nobody will buy equipment. Well, roughly, people will start wondering: what will happen to NVIDIA? And contracts stall.
Why did the contracts stall? Because the advertising is set up inefficiently? No. The advertising is just as effective, but everyone will be looking at a poor-quality funnel. We had this case once, and I always said that people don't fully understand inefficiency. We once had an assignment; we, at Business Youth, made courses. We had an assignment; there came, like, a thousand and a half people into the hall, and we gave them the assignment. Well, across the whole country there were, I don't know, 3 thousand.
You have to go during this week and do three test drives: in an economy-class car, in a business-class car and in a luxury one. Well, you can imagine, right, what it means that 3 thousand people had to do a test drive in a luxury car — there aren't that many luxury cars in the test-drive showrooms across the whole country in a week. And people needed, in order to do a test drive, they needed to post a video from the car. Later I once talked with the director of the Rolls-Royce showroom in Moscow, and she tells me: "Now I understand what that was."
Because at that time Bentley introduced paid test drives, did it after that. Well, this is a very interesting thing. That is, whoever was incentivised on test drives earned a huge amount of money. Can you imagine, someone had an incentive on test drives, but they had no influence on it at all. And whoever analysed sales — sales weren't right, and people drew an irrelevant conclusion. And this was in general just a temporary surge, a one-month, for example, or two-month surge. And it wasn't real, well, the surge.
It was simply a game assignment, a certain game mechanic. That is, it didn't relate to and didn't correlate with the market. And here's a story in which artificial intelligence will notice this. Imagine, in this case artificial intelligence is probably capable of noticing this. It could find on the internet, search, search, search, search, analyse, find, listen and watch. Here a training programme took place, people were writing that they have such an assignment, and immediately report this and adjust the data for that period.
In the morning the test-drive showroom sent it, they see an increase, they say: "Listen, we've had an increase in test drives. Most likely it's connected with a training programme. If we don't want to give these people cars to test, let's say so. If we do want to give them..." Well, they could have wanted to give them, because a great many people later still bought expensive cars and so on. It was, like, I don't know, not in fashion once, like, 14 years ago, right? And then it became...
There is, I think, at, at Beckham's wife's, there's a T-shirt: "My dad drove a Rolls-Royce, and I wanted to make myself a T-shirt, or for my children," — he was laughing there — "a T-shirt. Here: my dad bought a Rolls-Royce at 30." Right? But only not mine, but I, but I. And back then it was definitely not fashionable. Back then it wasn't just not fashionable, it was non-standard, right? It was non-standard. It wasn't a widespread, ordinary, everyday action, right, under such, under such circumstances. And that's why it's not a question of what's good, what's bad here, right, we're not discussing that.
The question is how to make assessments correctly. Here artificial intelligence would probably be able, in today's reality, to find such data now. But you imagine how expensive this must be to build, to cost. And someone has to allocate money for this, for these parallel analyses. I think that even now, myself, reasoning with you, I'll further train some models on all sorts of details, no-no, not further train, but, sort of, further tune, tell them more about what else can be done, what to get around, what to look at.
For example, I don't know whether you know or not that, for example, there's Elon Musk's system X, right? It's the former Twitter. X can't be scanned by a model. That is, you can't read all the posts in X with a model. Paid post scanning via the API costs money. Just scraping — you'll get blocked. What's more, it's also a punishable matter, right, where the fines are tens of thousands of dollars. So, via the API it's expensive. Well, really, if you go around all the people every day, you'll start spending hundreds of dollars on a fairly small segment, segment of people.
But here you exactly tell yourself: "I'm interested, for example, in these 500 people on the market and their opinion, their statements. On them depends, for example, what will happen on the advertising market. Will they buy or won't they." And for managing, for example, advertising. Or, for example, you say: "I'll scan how other companies do advertising." I don't know whether you know or not, in general you can train robots to watch how other companies, your competitors, advertise, and what ads, for example, they put out, what offers they have, what promotions they launch, right?
And you can scan and check, for example, what promotions your competitors are launching. And you want your robot to be right there with them. Because, for example, at some point in time you, you advertise, imagine, the sale of, well, it doesn't matter, these DJI lavalier mics, right? You resell them. And then at some point in time DJI launched a 70% discount. Well, that's it, your advertising will go to zero, go into the trash. That day you need to either also launch a 70% discount, and even if you have this discount, you forgot, for example, didn't relaunch the advertising. And the advertising still needs to learn. Here's Amazon.
What is Amazon known for? Amazon is known for the fact that it, this was already, probably, five or six years ago, it had a very neat algorithm. If you sell, for example, this remote on Amazon, or, you know, it doesn't matter, here, well, OpenAI's Codex, This device, ModRet or Ret or whatever it is there, Chromatic, it doesn't sell, but usually Chromatic sells. Amazon had the logic that if Chromatic sells Chromatic on Amazon, it can't set a lower price on its own website than on Amazon.
And if Chromatic sets a lower price, Amazon simply takes it and lowers the price in its product itself, forcibly. That's a product strategy. A huge number of corporations work with such strategies, right? There are grocery chains that sell the cheapest bananas. What does the cheapest bananas mean? They endlessly scan prices, and wherever, at whatever price a banana is, they'll sell it cheaper than the market, regardless of what they bought it for. They'll sell it at a loss. That's a strategy.
And that's why, Anya, of course, not a single person is able to, uh, analyse all of this completely. The question is, uh, how to set artificial intelligence up, Anya, to do this today. And here the cases people tell are interesting. I don't accept systems right now. If someone tells me they've created a system for managing the ad campaigns of other advertising companies, why don't I fully believe it? Because a great ad campaign will work for you on your sites under one condition: if you give your model access to your company's finances, to your company's sales, to the ERP, and to the CRM, to various systems, to CRM systems.
100%. Then, when, and in that case you'll be able to give Danila a task.
Yes, I don't mind. By the way, Danila, after all, interestingly-
This is, you know, as if we've come back to this topic, because in that case you would be able to give him a task.
Look, well, I don't even mind giving him access. So what was the interesting case with Danila? In April of last year, Codex started to grow, and on some day I was sitting in a meeting; I have, you know, a stake in a real-estate company in the US, and I got tired of listening to people not knowing the numbers.
Some people didn't know the numbers. And I, I wrote to Danila then: "Danila, listen, do you want me to give you a task? I'll give you full access to the CRM, the ERP, you know, to all the data completely, and you'll put together certain reports for me." And I had the right to give it to him separately there and so on, right? I would have agreed there, within, say, a small board of directors, to give him this access. How to register him, as an employee, and give it. And he turned down that task. And because he turned down this task that day, during those two days while he was refusing, I went into Codex myself, and what I'd told Danila, I expanded a little, launched it in Codex, and I was blown away.
I remember how it happened.
Yes.
And because I understand automation pretty well, well, and in general I was once a programmer, I understand company automation very well, you know this, I, Anya, flew ahead, right? And Danil, on the other hand, didn't understand company automation, right? I had the very highest plus here, and I also understood business. I knew in advance all the various gadgets, details, little nuances and aspects. And now he's come back, and now, sort of, you need to switch him on, and you think: "What task should I switch him on to, or what task should I hand him?" — right?
And so that you yourself — here it's also important not to suffer yourself, Anya, can you imagine? It's important somehow not to take part in it yourself. Many of my employees, by the way, ask for access to my various automated systems, which, Anya, I can't hand over, because I need somehow to integrate this with them. There's also the problem of these integrations, because these systems often also have access to some other data of mine. And to carve out this piece, you think: "How will I carve it out?"
Yes, that, that I understand, it resonates.
Let me remind you, guys, that it's out, well, and everyone who watches us, Fable 5.2 and Mythos 5.2 are out. I believe this is the strongest news of this week, regardless of the fact that Google expanded access to Gemini for Claude, and regardless of the fact that ChatGPT, by the way, launched a feature for promoting its companies. Well, that's clearly a paid project. Here, Fable 5.2 is worth watching, and it really wins very strongly at complex programming. And what is complex programming?
Look, it's essentially any automation. It wins there by multiples. And yesterday evening I had a complex case, quite a complex task, one, and Opus 5.5 was working on it, and already late at night I switched it to Fable 5.2, because I didn't fully trust Opus, and I switched it. It was already too money-critical a task. Whoever has used Fable 5.2, write in too. It's very interesting to find out about it. Let me remind you that we have a new news format. It was on Wednesday. Watch it, there'll be the next one too, including on Wednesday, well, and on Friday a video on Fable 5.2.
And on Tuesday, by the way, a very interesting video came out from Anya about the labour hiring market and what not to miss in your life when you want to find a job or in general when you hire people, about the hiring market. Be sure to watch this video. A very interesting topic. And we'll see you in exactly a week. Bye everyone.