What should you be learning today so that a few years from now it doesn't turn out you spent your time on the wrong thing? And what do we keep paying for, even though with artificial intelligence the task can already be solved differently? Hi everyone! You're on the ToTheMoon channel. In the first part, a few days ago, on Wednesday, I went through ten rules on how to set tasks for AI, give it context and check the results. Ten rules that a person really needs to know well for themselves. And today, ten more key things that you absolutely can't afford to miss in this age of artificial intelligence development.
They concern what will be valued in your work, your profession, your life. And I suggest you look at it all more holistically, even if you already apply some of it or know some of it. So in total there'll be twenty top thoughts, right? Well, as for super-geeks or superhumans who think they're extremely, incredibly well-versed in AI — you can probably, maybe, skip this episode. If you truly understand where you are and what you're doing. I'm really glad this second part has come together.
Let me remind you how all this came about. We've made more than 160 episodes of ToTheMoon, and there are certain ideas that I personally keep putting across from episode to episode. Obviously, some thoughts and ideas change. For example, yesterday I was in San Francisco at OpenAI DevDay. It's a day for developers, but overall it's simply OpenAI's conference. It's a closed event, and obviously when you go to places like that, when you talk to different people, your worldview shifts in some ways.
At the same time, what matters to me personally is that you don't live in hype mode. For example, the conference took place yesterday, we'll get to it on Sunday — watch the episode about it, Ilnar and I will tell you all about it. But what I specifically don't want, and what doesn't interest me, is for us to come and say: "Look, Dots appeared at the conference, or Spaces appeared, run and get it right now, this is something super cool and it'll somehow solve something for you."
And living in endless hype. I want you to be able to look ahead and forecast your future life, and to truly understand where you stand in terms of AI. I keep saying that these episodes, these two episodes, are the ones you can send to people in general who want to really get to grips with it and understand what to do and how to move forward with AI. Sure, there are a million or so people who'll say they're extremely well-versed in this. We'll see, right? By the way, yesterday OpenAI's Sam Altman announced that they have 1.2 billion active customers.
So, point number one: the gap between those who truly use AI and everyone else will grow very seriously. It's a recommendation of sorts, and I'll repeat it once again: these are my thoughts — in places rules, in places ideas, in places a vision of the future. And you'll see that today. And this first point of mine is very obvious, because just this morning my sister and I were discussing that, for example, even my wife, in the things she understood, she was, in terms of AI, a fairly strong user a year and a half ago, right — when there were, say, 300 million users, she was in the strong cohort.
Now the gap between me and her has grown catastrophically. It was huge back then too, right, but now it's grown catastrophically, and it'll be harder and harder for people to understand what's going on. You can see this in yourself, or in how you get information, or in the people around you. And be sure to watch the previous episode. Everyone needs to be living by these ten thoughts, or ten rules. For example, using AI — as one of the thoughts in the last episode was, right — using AI every day not just in your life, but also at work, plus in life, right, for different tasks.
In other words, it's a critically important necessity to keep looking for different use cases for yourself, and not to limit yourself only to what you're actually used to, and also not to limit yourself to your own micro-ideas. By the way, a lot of that came through yesterday at the conference, when engineers were talking about their ideas. There was a lot about what they like, what they want, right — even in terms of building websites with ChatGPT Sites, an engineer said that every employee in the company already has their own site, but it's not a company's employees who need websites, right — websites are needed by companies out in the world and businesses out in the world. A completely new process.
And here, of course, this story will only keep expanding, right, and the gap will be catastrophic. Elon Musk talked about this a lot, and many people — actually, many industry leaders — have talked about it. The gap will be catastrophic, and you won't even be able to notice when it happens. By the way, why else is the gap catastrophic? Because models — good models — are very heavily restricted for people in terms of access and cost. And as OpenAI employees said yesterday, on average an hour of running a normal task costs $30, and there are a huge number of such tasks running in parallel.
So you understand that it's tens, sometimes hundreds of thousands of dollars a month to build serious systems.
The second point for us today is: don't chase the concept of agents, right? And this isn't just hype. It's a story where people and companies lose a lot of money. Today a lot of it concerns the advantage at work, right — that kind of business topic. Although, as I always say, it matters for the ordinary person too. Where you're a user trying to figure things out — say, you've started using Codex or started using Claude Code. Don't chase the concept of an agent at all. I've said many times that I don't like this concept. I think it will go away.
I think it's very abstract — especially, again, yesterday one architect said: "So, we launched an agent in Astra here. This agent used 8,500 subagents." What is that? Or, like, 8,200 subagents. It's like counting tokens. That is, it's not the future, and the concept of an agent is not the future at all. OpenAI has released this thing called Dot — watch the episode about it with Ilnar on Sunday. Can you call it a super-subagent? Or can you call it your own AI? Or your own superintelligence, right?
The US has introduced a new concept: in government bodies there's no longer the concept of artificial intelligence, there's the concept of superintelligence. And can you call it your own AGI, the one OpenAI wants to build? And all this making of agents — I think this mindset of agentic work is catastrophically flawed, and it doesn't let you fully grasp what artificial intelligence is and what the system of the future is. That is, you're always making an incomplete project, and you always want, roughly speaking, not to change the whole room but to change a small piece of it.
The agentic concept makes it really hard just to picture this construct for yourself, because a huge number of people use this word without fully understanding it, or riding the hype on it, right. It's like the talk about agentic browsers and about browsers inside an agent. But somehow we don't hear many people talking about that, right, even though they said what a space-age world it would be. But the ToTheMoon channel, both Ilnar and I, never bet on that at all. So, the third point for today: what matters most now isn't a profession, isn't a diploma, but the ability to understand a task as a whole, right.
Let me unpack it for you. I've shot a number of episodes before this — you can watch them — about hiring people and the future of professions. And I've very often talked about this story that you need to learn, say, Prompt Engineering. And a huge number of people went to these courses. In general, I know a lot of people who buy courses on how to use AI models, and they buy courses not on something serious for themselves, to figure out, I don't know, how open-source models run on a particular processor architecture, or, say, setups with NVIDIA, for example with their hardware — rather, they're bought by people who aren't in a technical environment at all, and they want to cover some micro-layer for themselves, right.
This whole story around Prompt Engineering is deeply flawed, because the skill of writing — thinking that you'll learn to write prompts properly, or thinking that you'll get the profession of a prompt engineer — well, that's an absolute illusion, because modern systems, for example models like Astra, are basically built on the idea — or even Sonnet 5.5 at Anthropic. Astra is in ChatGPT, right, for those who don't know, and in Codex — that's OpenAI. And models like these aren't about you knowing how to write prompts; they're made to understand your task, your context, as fully as possible.
And in today's world, learning to write prompts is the same as what my nephew told me: he studies at the Warsaw University of Technology, a very prestigious university, in a prestigious program, and in their first year a lecturer is teaching them how to use Windows 10 and how to adjust the settings there. For example, which — how you need to set things up to change the screen resolution, or how to take screenshots properly. So I voiced this thought — it may not seem very positive to you. I voiced the thought that, frankly, such people should be put in prison.
Obviously, it's an abstract thought — I think very well of people, and I have plenty of other videos on other channels for that — but I'm deliberately highlighting this edge, the way I sometimes say such things about doctors, or about scientists. Because how can anyone, in 2026, even teach modern people about Windows 10, or in general teach a skill that's absolutely useless these days? Everything has changed, everything has become completely different, right. It's the same with cleaning an apartment.
But if a robot vacuum drives around my house all day long, then teaching my kids to perform some operation like vacuuming the house would be strange, right, if a robot vacuum is driving around. That is, there are other tasks that need doing, not some abstract task. But teaching Windows 10 is, of course, total absurdity, bordering on madness, absolute madness. Write and tell me what you think about this. Even if the people teaching this are competent, great people, it's still very strange, right.
It's probably the same now with Prompt Engineering: teaching how to write prompts properly is an insane thing, and I don't want you to get caught up in it. And so, understanding what to do in terms of a profession or a diploma is a big question, right — it's covered here on the channel, in different videos, I cover it on my other channel, I talk about it very often. I'll try to shoot some episodes on this topic too, maybe do some special episodes about it as well. It'll take more than one meeting, right — more than one meeting with people who are in these professions, more than one meeting with parents who are choosing professions for their kids.
It's not a simple construct in this day and age. So, number four.
We're going in order, right, and our editorial team will definitely be putting these points up on screen for you today. Here it is: choose one main AI system. And I consider this critically important. I talk about it a lot, and many people have said about me that I love ChatGPT way too much. Then they started saying I love Anthropic way too much, that I don't talk about other systems on the channel, even though I very often mention Grok, I often mention Gemini, I mention Meta's system — the new Muse, for example, that came out, or the previous ones — and I even mention Chinese models from time to time.
Recently in an episode we talked about both Moonshot and Kimi. And, mind you, they didn't buy any advertising from us, even though there were questions about it, offers. Or about Alibaba's models. I don't want to tell you about the part of the market that's in the third or fourth tier, sometimes even the second. Someone wrote in the comments: "Russia has now taken on AI, we'll get it here too." I don't tell you about Russian models for one simple reason: because they're not even at the second level, not the third, not the fourth, not the fifth, not the sixth, right?
Once again, we'll show it now — there's this picture of the models: A, B, C, D, E, and S, SS and SSS levels. They're not even anywhere close in these systems — they do carry out and solve certain tasks. Just the way Siri, as it's been built so far, solves tasks, right — even though it's supposedly based on Gemini or some other systems. Apple has done an update, but it's a lag, a very serious lag. And as for catching up with OpenAI and Anthropic today — well, Grok obviously has a chance. Elon Musk actually said that in a year Grok's agents will be number one. Possibly.
Google, Gemini. So that's the top four settled, and we stay with them. Everything else is secondary as of today. If something comes along, we'll definitely talk about it. And why choose one main system for yourself? Because then you'll be able to understand as fully as possible what's actually going on, what's actually happening with the systems. At least you'll be able to test different things in that system, choose different versions — at least you'll be able to figure things out there somehow.
Although, in reality, it's impossible today to figure out even OpenAI's systems. They have a browser version, a mobile version, a Codex desktop version, a ChatGPT desktop version. All four of these systems work differently, they have their own features, often even their own menus, right. It's unclear how they're connected to each other. It's unclear which one has access to what. OpenAI tends to launch some solution and then shut it down. Yesterday it released Dots. Before that there were Pets, there were some kind of roles.
You don't understand what's going to happen next and how it'll develop. That is, there's no unified system for developing the architecture within a single company — look, so that, in essence, we wouldn't get Dots, but it would all be together, it would be chat, and voice, and everything together. That is, all the projects are developing, a lot of the development happens independently. OpenAI is turning into a zoo. Anthropic is also pretty much a zoo, and, well, Google is quite the zoo, right?
Well, Grok is probably less of a zoo, or Meta is less of a zoo, but Meta is somewhere out there anyway, somewhere mid-league. And when you stick with one system, that's a huge, really big benefit. And there's also the case where you can choose a second system for yourself. For example, my second system is Anthropic Claude. I don't use the chats in Claude for basic, simple work, as such. I use Claude Code, the same way as Codex. So I have two systems, Claude Code and Codex, for building different solutions, for automating my processes, for working with different interfaces — to the point that, look, yesterday they released Dots.
I'm already sending messages to my wife in iMessage through Dots, even though Dots can't send messages in iMessage — I send them through Codex. So there's this kind of in-between layer, right? And there's no direct connector there either — you send it through the interface on the Mac, sort of. Well, these kinds of in-between layers can work. And there's the second system. Of course, I have subscriptions to other models, I have subscriptions to other systems that I use now and then for some micro-needs, but I have a lot of different projects, and I need that.
The fifth story, the point that's up for today — and let me remind you, we have ten of them today. Last time we also showed ten. And the fifth point. A point, again, that concerns your professional development, your work, where you live, what you do, right? The fifth point: rebuild processes around artificial intelligence rather than bolting artificial intelligence onto the old, right? I want to say that this is a global problem, of course, and a problem I could observe yesterday at the OpenAI conference in what developers were saying — not all of them, but many.
It's a story where a person tries to use an elephant where there used to be just toys. And there's a huge number of processes we live through in life just the way we've gotten used to. Once there wasn't much of an internet, right, and there were no mobile phones and no way to communicate so quickly in messengers. There were no such habits. Not everyone even used email. And in that world, if you'd shown the possibility of creating what today is called an agent, which, I don't know, checks your email for you every day.
The kind of simple use cases people love to name. It would have seemed like some space-age world, and people wouldn't have believed in it and couldn't even have grasped it. That is, people simply wouldn't have been able to grasp it. And our task right now, yours and mine, is not to be left in the back ranks. In general, a person's task in terms of professional development, and I'm talking now about the people who are connected with AI, is not to get stuck in this failure to understand what's happening, and in studying systems that won't actually represent the future.
When you, for example, automate some process, you start automating part of that process. For example, you need to improve a sales manager's efficiency. You start building, bolting some interface module into the CRM system. Or you need to make a website faster. You start bolting some AI module into the CMS. Or you need to automate some block in finance. You start working with your existing Excel files or Google Sheets and automating them. Yesterday a young woman talked about this very example — that at OpenAI they have Google Sheets. I'm honestly amazed they have them, but they do have Google Sheets. I'm quite the critic, right?
And she says: "So we made a site with ChatGPT Sites that lets us visualize things better instead of Google Sheets." And my point is this: why do you even need Google Sheets, and why do you need a site, if you have your own Codex, or you have these Docs, which can tell you that a status has changed, which you can ask what projects there even are at the moment, and which can make your very own visualization for you, one that isn't needed as some external world of development, right?
So the task in front of us today will, of course, not be understood by many today, but I really want you to lay this foundation within yourselves. And not understood — not because I think someone's stupid or not stupid, mind you, right — but in terms of actually living through it, because you'll be hearing it in your companies from an endless number of smart people who'll be talking about all sorts of things. What it means, of course, is that you need to rethink your life in terms of the processes that happen in it.
To understand that many things will happen in a completely different way, that many systems that exist today won't exist at all — they'll be gone. I mean, I calmly accept that there may not be such an interface as the iPhone at all. That is, the iPhone will take a completely different form — maybe like in the film "Her," or maybe some other way, right? That is, it'll take some form, possibly like in different episodes of the series "Black Mirror." But I understand that these hardware interfaces will change, and we'll interact with them in different ways. I allow for that from the outset. It's like with cars.
We need to stop assuming that our kids for some reason have to learn to drive a car. That's — because there'll be cities where you won't be allowed to drive on autopilot. But obviously, if I'm walking around town now and at the Tesla store on Santa Ana Road there's a Cybercab with no steering wheel, and every day near my house and on Page Mill — given that Tesla's office is here and xAI's office is here, right on this street, five minutes from where I live — I see these cars endlessly, so why learn to drive yourself?
It's some very strange construct. Or debating whether a human drives better or worse. Why even pay attention to that? Although my daughter, again — she's just turned sixteen — did of course learn to drive, because these companies still don't have a license. But I think all my next kids, and even my next-oldest daughter, who's twelve, certainly won't be learning to drive anymore, and she'll have a car like the Cybercab. I think within four years they'll work out the licenses, including for kids to ride in them. So, point number six.
I think this one is exactly right for a discussion in our chat. Don't forget to support our channel — like, write comments. I see a lot of people write comments and never like. And they even write "thank you" in the comments. Don't forget to hit like — it helps a lot. Those who don't like it, who watch every episode and still don't like it — go ahead and hit dislike, no problem, write comments. That's also active engagement with the video, and it's needed. So, an important point: models are already smarter than any expert. It's just an endless loop of debate.
It's like, you know, that debate about how America's economy is about to buckle and the dollar will cease to exist. And you're stuck in this endless system. Or in different countries people say: "A war is about to start here." And you're in an endless discussion. Or that there'll soon be an earthquake in California, and it's coming any day now. And any moment now it'll happen. And you're in this world of these discussions. Any model today — well, any decent model — is smarter than absolutely any human, any expert, in absolutely all processes.
It comes down to certain unique particulars. And obviously a model won't go out in your place and start digging a hole in my garden. And a model may not know some of your inner details, certain sensations, but overall, by all the tests, it's smarter than absolutely any human. Not a single person today would beat it. And more and more robots of different kinds will appear that a human will never ever catch up with. Just as a human will never ever be able to clean a space as efficiently as a robot vacuum does. It's not for nothing that I give such simple, dumb examples.
A human can't do it, just as a human will never ever be able to watch and monitor a space for security as effectively as a camera does, and as modern AI does, or modern software, or even earlier machine learning models. Never ever will a human be able to monitor and recognize everything with the same quality, so incredibly well — maybe spot some unique details, but at scale a human can no longer do this. Just as, if you ride in a Waymo taxi in San Francisco, it obviously drives far more safely than any other person's car, because it has a constant, 100%, 360-degree view.
No human has ever had it, has it, or will have it. It'll only get better, it'll only get cheaper, it'll keep getting higher quality. And you need to admit that to yourself. In no way am I saying here that AI replaces humans or somehow surpasses you, or — that it'll replace you at home as a dad or as a mom, as a husband or as a wife. That's not what this conversation is about at all. It's just a point to pause on. And over the last few years I've had to spend a great deal of time talking with people, when people just kept launching into: "Oh, well, ChatGPT is dumb.
Well, and Anthropic answers wrong here. Oh, this model here doesn't understand this at all." You need to forget about all that, especially given what modern models can do when a lot of money is spent. That was visible yesterday. It's interesting. Yesterday Sasha Mashrabov — he's the creator of Higgsfield — speaking at OpenAI, put it nicely: "We made a clip, a ten-minute video, nobody in the world can make it, and people can't make anything like it. Well, we just spent $30,000 on it." So it simply costs some extra money. A cool video, or cool clips, cool games, cool results — they cost money.
Just as OpenAI solved part of the mathematical problem of the Navier–Stokes equations. Well, they simply burned a million or tens of millions of dollars on it. But the system was able to do it. This is very important. People could burn as much as they like, gather in any numbers — they wouldn't have done it. Of course, what's going to happen now with olympiad math, with education everywhere, is a big problem for humanity, of course. Can artificial intelligence read any X-ray, any scan, better than a human? Of course, at scale, yes. Beyond that, it's about individual specifics.
Who can do what better, who can do it more efficiently, and who knows how to do it in terms of volume. Today, no AI has learned yet to work in parallel, to run hundreds, thousands of different processes across different systems. And this Dots that was released yesterday can't figure out its own OpenAI systems. So what's there to say about it figuring out a person's life — but people themselves often can't figure out their own lives. And not just often — always. And people can't fully figure out the companies they work in, the professions they practice.
They keep using marketing methods that are forty, thirty, twenty years old, keep programming systems that don't need to be developed at all anymore, keep using the wrong motivation system, and so on. Drawing analytical conclusions that aren't real, making erroneous forecasts. That's why here too, by the way, it's very important how you relate to AI — and to understand, of course, that AI can make mistakes. I've said it, I say it, and I'll keep saying it: AI can make mistakes. And I think that among the previous ten points, this was one of them.
So that's why you need to stop arguing and not react to other people who try to prove the opposite to you, who constantly try to show that AI is somehow dumb, undeveloped, doesn't know things, and so on. I often reason on the channel that someone didn't do something here, something's missing there, but that's more like nitpicking at some small part. It's not nitpicking at the progress itself, the insane progress that has happened in AI. Look how many people were saying it's the same kind of hype as crypto.
How many of those statements have there been? They're still around, they're still out there. So those are the six points about how your work is changing and about the human advantage. We'll highlight them for you here once more and move on to the four points for today — this last four out of the twenty — where I'll tell you where all of this is heading and what to get ready for in general.
Point seven. I actually think point seven is fundamental in terms of… Oh, by the way, I noticed this on my desk — yesterday at DevDay every attendee was given a Chromatic. I remembered my old Game Boy, and my nephew told me how, back at school in Minsk, literally not long ago, when he was at school in Minsk — he lives in Europe now — he and his friends used to play on it during breaks. I'll ask my mom to bring the cartridge from Minsk. My Mario games, from '91 or whatever year it was when I got it. And these came out yesterday — here's what makes it interesting.
It's connected to Codex, and Codex easily programs straight onto the cartridge and loads all kinds of games. I don't play anything at all. It's not of the slightest interest to me. And by the way, they put them up on eBay yesterday — they're already selling for $3,500, $2,000, $1,500 apiece. Since it's some kind of unique thing. But I just loaded a couple of games onto it through Codex for fun, to show my kids. But obviously I'll limit it and load the concept of screen time onto it. So, point seven.
I think companies have lost a huge amount of money on this one. A huge number of startups — thousands of startups have already been wiped out because of it, and maybe tens of thousands more will be. It's something where a huge number of different mistakes have been made, from not fully understanding where we're actually heading in terms of software development. And the point goes like this: there will be no search, browsers, websites or programs as we know them, and what will remain is some kind of dialogue with artificial intelligence. What kind? That's a big question.
And even Dots, which OpenAI released yesterday — this Dots — is in some way close to a sense of where this is probably heading. But it's still just a tiny beginning. I wouldn't be surprised if OpenAI kills this project, and most likely, because it works in a peculiar way, most likely they won't solve this problem architecturally on their end — or maybe they'll manage to develop it. We'll see. Even though yesterday Sam Altman stated more than once — on stage, and later to various media, and on the sidelines — that OpenAI isn't planning to kill off the software products of various competing companies, or to build every software product, I think that statement is very much politically and economically motivated.
And my forecast is that, of course, people will work through some kind of command line, through some kind of interface, sometimes by voice, in different ways, somehow interacting with a form of artificial intelligence that will be called OpenAI or Anthropic or Grok. And this form will then be represented and expressed in various hardware-electronic-physical systems. That's what I'll call them. Why physical? Because we don't know how it will be represented or in what form it will be represented, how easily we'll be able to integrate it pretty much everywhere, how easily we'll be able to roll out all these different microphones, cameras, various sensors, how we'll be able to put them everywhere, what we'll carry around with us, what kind of device it will be.
Because obviously sometimes it's convenient to type, sometimes it's convenient to speak, sometimes it's convenient for the system to just listen to us in the background — how it will be integrated across different systems everywhere. We'll see how it plays out, but the intermediate layers will of course be stripped away, and the ones stripped away won't be like Booking — those will most likely be corpses anyway. And the intermediate layers won't be ones like Google search or Yandex search either, right? Far more serious intermediate layers will be stripped away.
The question is: will there be an operating system like Windows, or an operating system like Mac? That remains an open question, because if OpenAI releases a really cool device tomorrow, one that will — Codex will be running inside it, and it'll make calls and connect to any service. My big question is: why would you need a Mac? Because look, there's a very interesting case today. Today there's Codex and ChatGPT, integrated with email, with Gmail, for example. Say your email is with Google — it's integrated with Google. That means I don't open Google, I stop opening it. I wasn't really opening it anyway, because I opened it in the Mac and iPhone interface.
So it turns out I'm already no longer opening my email, and Codex — or this Dots — scans all my emails, tells me about them, deletes some, tells me about others, sorts some out, applies various rules. So I no longer open the interface. So what am I actually using? I have a certain email address — that is, there's my name, which I can migrate every now and then. And finally I can migrate to anything I want, whenever I want and wherever I want, in terms of additional systems. There's my name, which people are used to writing to.
I have different email addresses — I had different ones anyway — and I couldn't connect many of them to the Gmail environment anyway, because there are restrictions. Some of my email is in Apple's iCloud infrastructure, some is in Gmail, some is in various other corporate infrastructures or somewhere else. Essentially, it makes no difference to me. Well, not that it makes no difference — your emails are stored somewhere, and you can choose for yourself. Whether you want to keep them in a certain country or on some service, paid or free, secure or not secure, at home, and so on.
You connect some particular mail service. It doesn't matter to you what kind of service it is. Just like when I order a delivery from Amazon now: Amazon might bring it, FedEx might bring it, UPS might bring it, some other delivery service might bring it. I don't care who. That's Amazon's problem — who it's made a deal with to bring me the goods. But I ordered everything from Amazon. And here it's a question of Codex's architecture. And now Codex has started making websites. It's strange that it isn't starting to make its own messengers. It will — it will start making its own email agents, its own email systems; they'll start doing all of that. They've said so, by the way.
At one point there were statements that they'd launch their own social network. They tried to launch one based on Sora, on video generation, and it didn't work out for them. Lame independent projects like that. But it's obvious that I don't need email providers, just as I don't need providers when it comes to SMS. What do I care who sends my SMS? Beyond that, I use some interface to send SMS. And all of you use the SMS interface that's on your phone by default. I have the Apple app, and I send through it. People rarely send through other apps. That's usually some geeks and so on. Sure — well, if some of you do, write in.
And so, if the operating system is Codex and Codex can connect to my SIM card — my mobile carrier there is, I don't know, Verizon — and it doesn't matter which mobile carrier you have: AT&T, or Orange somewhere in Europe, or MTS, Beeline, or — what are they called in Belarus — A1, or something else. It doesn't matter what interface you have or what carrier you have — you can send these SMS messages through some system. What's more, you can connect additional carriers to send messages.
Just as today there are eSIMs, where you can easily switch different eSIMs on and off, buy them and use them on your phone. Without even fully figuring out whose eSIM you're using, when nowadays a huge number of people fly somewhere and hook one up through Telegram, not understanding what provider it even is — who issued it to you? Did Orange issue it through someone, or how does it work? Is it its own network or not? It's like Apple rolling out a big project for sending messages via satellite — whose satellites are they actually using: Amazon's, or will they even use their own, or SpaceX's, right?
Whose satellites are they? Starlink, for example, right? This, you see, is a way of thinking that's very important for your own perception when it comes to the existence of familiar software. And I'm still betting that OpenAI won't be a marketplace, won't be a platform. By the way, yesterday they announced they'd opened a new marketplace — I think it's already the fifteenth, right, in some form; there have been various ones, for plugins, for projects, for GPTs, and so on. Yes, they'll be an operating system. They have no choice but to become an operating system, because someone will become the operating system, just as Android once did.
Google, with Android, took over a simply colossal market, and Android devices appeared, right. But it's a completely different model altogether, a completely different approach, a completely different discipline, completely different things, and so on. A completely different sphere of perceiving everything that's going on. Point number eight. Point… write and tell me what you think about the operating system.
This probably concerns more professional users. If professional users are with us too, it'd be great to hear about it. So, point number eight. The real… once again, I'm voicing my thoughts and my opinions, my rules, my recommendations. These are the things I see. It's not for nothing that I give some reasoning. Obviously, it doesn't mean that I dreamed these things up, or — somewhere I heard something from someone, somewhere I saw something, somewhere I combined something and brought it together, and brought it into my knowledge.
This is my current perception of reality. If tomorrow the whole world gets cut off from the Internet, my reality will change. And thank God for that. So, the next point: real artificial intelligence will ask you questions on its own and remember everything. And this is a topic that's very close to my heart. I believe that Anthropic, OpenAI, Grok will get — and Gemini — whoever starts doing this will get incredibly strong growth, when artificial intelligence finally starts taking an interest in you, finally starts asking you things, finding things out.
Even yesterday I asked Dots to create a service — this is in a chat inside ChatGPT; well, roughly speaking, it's built using Codex or ChatGPT. Dots is just a new feature, available on Pro subscriptions. What's more, I don't know if it's on the first Pro tiers, because my sister — I think she has Pro for $100 — didn't get it, while I got it on Pro for $500. And it was just yesterday that they released Pro for $500 — it showed up, I have it, I have it switched on. So I launched a service in Dots and told it to sort through the emails I get from my four schools, from my different kids.
American schools — anyone who lives in America knows — love sending something like a billion emails. And to pick out the information I need. And it told me: "OK, task set." And then I said to it: "Listen, can you…" and started telling it what to do for me. That is, it didn't start asking me anything, offering me options, saying, "What do you consider necessary or unnecessary?" That is, it didn't go through a process of understanding as fully as possible what's going on with me. What's more, it had already seen and knows that I have a wife, for example; it could have split things up and asked: is there a difference, for example, between what information comes for which child, who's older, who's younger and so on, on which days of the week, the details. To ask: who actually takes care of the kids?
Who usually keeps track of this schedule? Who drops the kids off at school? That's really very important, because I'm, for example, the secondary person who reads these emails, since in my case it's my wife who primarily reads them. I mean, she's more primarily involved in the activities at the school, specifically in terms of the school's day-to-day operations; she's a little more involved in all of it. She drops them off far more often than I do, for example, or picks them up more than I do, or deals with the school and reads things from it more than I do when it comes to organizational stuff.
Write a check for something, fill out some form, sign up for some special program, and so on. And she comes to me and says: "So, tomorrow there's an open school event, so that's where we're going," for example. I say: "Well, OK." And the system didn't ask me about any of this. And of course, I'm sure that in the future the systems that ask questions will win. And here a question even comes up, by the way: can you actually build services like that, which question a person? I'll be honest with you, I've rarely seen humanity have services like that, and I'm trying to create them myself.
I probably got into this seriously around 2014–2015, probably even from 2015. They require a deep approach, an understanding of data normalization, automation, an understanding of both business and development at the same time, and so on. I'm not sure it makes sense to build these services now, because artificial intelligence will question any person far better than any person can question another person. And these systems will appear in any case. And you and I will be interacting with an artificial intelligence that learns to ask questions.
And as soon as that kind of artificial intelligence really starts to appear, once they bet on it — we'll hear about it — it'll be a strong move, strong progress. Recently I've allowed that maybe the reason this isn't being done isn't that these companies haven't gotten to it — although that's also a big part of it, definitely, given what they sometimes do. I think that, among other things, there may be a restriction on giving all systems the ability to ask a person for all their data, for two reasons.
First, in terms of cost, it would increase the load on their servers. They're not interested in that. That's exactly why they keep feeding people cheap models, right? Even where a person could use a more expensive model, when there's no limit on it. The second thing, of course, is safety. Because if we give artificial intelligence the ability to talk with every person, we don't know what that will lead to or how far AI will go. Because it's one thing for AI to scan books, and another thing for AI to get really good at scanning a billion people.
And by the way, I want to tell you that if today — if ChatGPT started asking questions of 1.2 billion users, the system would get answers not just from tens of millions of people — I think it would be 100 million plus people, very thoroughly questioned and with their profiles filled in. And by the way, the system has the right to do this if it gets the person's consent, of course, right. It's surprising that this isn't being done, but it will be. So, number nine.
Access to strong artificial intelligence will become expensive and limited. And yesterday two things happened that I consider fundamental in this area. They brought back the $200 ChatGPT Pro plan, but with the caveat that its extended limit, which was 25 Pluses' worth, will be changed. So the $200 Pro version will become simpler. And for that they opened a $500 version, and the $500 version got new functionality that neither the $100 nor the $200 one has. That's the first area. The second area is that if you want to build real systems, you run out of tokens, and API tokens are very expensive.
And yesterday one of the architects at OpenAI said at a closed session that, roughly, they'd estimated Astra's work at $30 an hour if an outside person were paying for it. Accordingly, they have a huge number of such sessions running, and it easily adds up to $100,000 a month per person to build something. And my own work — I've always estimated it — my work with systems like Claude Code and ChatGPT, if I worked only through the API and used models this powerful, would of course cost hundreds of thousands of dollars a month.
That's obvious, because ChatGPT today — the Pro version doesn't limit chat, it limits the use of certain models; it used to limit them in the $200 version, by the way. On the $500 one, it won't limit at all how many chats you can ask in Astra, and so on. Not for now, anyway. We don't yet know how it'll all turn out. Of course, artificial intelligence will only be available to a certain number of people. That's exactly why I recommend everyone work with it, understand it, use modern models constantly so as not to miss this… to be at the cutting edge now, while it's accessible to humanity.
For now it's still quite accessible to ordinary people. But for some reason, when company executives and ordinary employees come to work, they work with intermediate layers of software somewhere, they use all sorts of middlemen. You ask one of them: "What version do you have?" He says: "Oh, the $20 one is enough for me, or the free one is enough for me," and so on. It will be very expensive, and it's already unavailable to the world as a whole. All the top systems today are available to certain corporations, certain organizations, certain people, clans — they're not available to the world. This is very important, right?
They're already unavailable to most countries. The US — if we're talking about countries — and China have the systems; everyone else doesn't. Accordingly, the internal agencies, the internal corporations of these entities have them, and everyone else doesn't. They don't! Even if they're in a partnership with someone, those systems are of course limited, right. They're definitely limited. And point number ten for today — a great finale to our top twenty.
Once again, hit like on our video, and on our channel in general. Don't forget that both the like and the subscription matter. You won't see our videos more often just because you've subscribed, but it matters to the algorithm. And of course, any comment matters too. And of course, don't forget to share the video with your friends and acquaintances where you see value in getting those people involved in artificial intelligence. Point number twenty is, of course, that artificial intelligence is developing faster and faster. And for a long time I've been saying it's no longer years, but months — and then not months, but weeks.
But attending yesterday's conference, I realized that the companies themselves no longer understand what's going on. Because they just released 6 Astra, and just last week they released 6-point Sol, and then yesterday they release 6.1 Sol. And you no longer understand what kind of system it is. And someone shouts at you from the stage: "This costs a bit less, that costs more." You see your tokens getting eaten up in some places and not in others. The changes have, of course, become daily. You and I will be living in systems that change every day.
And when the system reaches the level of this operating system, its own operating system — when artificial intelligence itself creates this environment, itself creates the operating system. And by the way, it won't be just an operating system. You know, I've talked a lot about how, when artificial intelligence starts controlling robots on its own and starts creating and programming various microcontrollers on its own, creating hardware on its own, designing it. Just recently in China they opened one of those factories where only robots work, nothing else, right. Factories like that will clearly keep appearing, right — factories where there'll be only robots, and the factories will be run only by robots, right.
They'll get to the point where they gain the ability to create hardware with the help of artificial intelligence. Then it's game over. In terms of the overall picture — then it's game over. Right now it's not even a matter of weeks anymore, right? Even though on the site our editors will put it as weeks rather than years, changes are happening every day. And that doesn't mean — sometimes someone writes that we're lagging behind on the channel. The news came out four days ago.
Systems like 6 Astra are worth living with for months. Otherwise we won't be able to understand them. Systems like Fable are worth living with for months. And what has started happening now shows that it'll be harder and harder to track what's actually going on inside. And you can also see that these systems have been given clearance and given permission, and no laws will stop them. And it's no accident that the American government changed AI, artificial intelligence, to superintelligence — superintelligence, ASI.
See you in our next episodes. On Sunday, watch the episode about Deep Research. Bye, everyone.