Claude Opus 5.5 is out, and once again we're seeing new records, new claims from Anthropic that artificial intelligence has gotten even stronger. At the same time, we can appreciate that in recent months both Fable and GPT-6 Astra came out, and now the arrival of Opus 5.5. But what exactly does that change for us ourselves? Because between how fast the models themselves are growing and how much our life, work, business, our everyday affairs are changing, we see, of course, an insane gap.
So where do the new models really deliver a different level of results? Which of them should you even choose today? And what do you need to change in your own work so that all this progress starts giving us a real return? Today we'll break it down using the new Opus 5.5 as an example. For the last three years we've been watching endless updates to various models. New versions of ChatGPT appear. Back then, just a few years ago, we saw the appearance of, well, Anthropic appeared, Gemini appeared, later Grok appeared, Meta was also in the picture.
Various Chinese models started appearing. Then the updates started within them. There are versions 4.0, 4.5, 5.0, 5.5, 5.1, 5.6, and now we've reached 6.0, right, in ChatGPT. At Anthropic there was a whole line of updates, and over the last six months the changes have become colossal in terms of quality. I mean, on the one hand, if three years ago everyone was debating that the model makes mistakes in, I don't know, punctuation, can't look at a picture, for example, at certain, recognize certain objects or compute a big spreadsheet, now the discussions are about completely different questions.
And what's happening is, on the one hand, we see discussion that with the appearance of Mythos-family models like Fable, a huge number of hacks have appeared. We hear about some incredible robustness of systems, and about writing websites, programs, mobile apps. On the other hand, you can observe a certain parallel effect where the new models seem to be degrading. I mean, they're supposedly meant to be improving, and then we see some tasks being done worse. What examples are there? There are examples where a person has something set up, for example.
I talked to my sister today. She says: "I have a task that's been running on a schedule for the last six months, and I see how the new version of GPT-6 Pro — that's Astra 6, yes, in Pro mode has started working worse and not doing the task that not just the previous version but practically the one before that was doing." I also saw this in Claude Design, and to this day in Claude Design I use Opus 4.8 Max, because when I started doing design with Fable or switched on Opus 5, I had real problems, as if the system didn't understand what it was supposed to do.
And in that same Astra, in ChatGPT 6 Astra, you can see it when you ask it to make some complex reports, which, it seemed, I think a year ago, or six months ago for sure, 5.4 probably, well, or the previous versions there did quite well. The system seems to glitch, it doesn't do them.
Today Claude Opus 5.5 came out, and I'm not a huge fan of filming separate episodes on specific model versions yet again, but I do want to give you a direction for your attention. Not just to say that some model has appeared which supposedly surpasses Fable in quality, which costs less, and is supposed to work even in Medium mode. I mean, not on the maximum tiers, Anthropic's tiers, supposedly better than the previous models worked on the maximum tiers. Rather, I want to draw your attention to how models are being released in general.
I mean, today Codex wrote to me that it no longer supports version 5.6 Sol, and it'll be interesting, and that's in Extra High mode on the Pro subscription, and offered to migrate all my tasks to GPT-6 Astra. I didn't want to do that, by the way, because in 5.6 I had a certain stability on certain tasks, while in 6.0 different scenarios started happening. And it's strange in general that OpenAI is doing this, because OpenAI released this line of models, Terra, Luna and Sol. And here, before people have even had time to test them and actually understand their quality, we're sort of moving on.
I mean, what's happening is that companies like OpenAI, like Anthropic, are of course interested in their own result of achieving AGI or ASI, or the singularity, right, which they talk about a lot, the technological one, or of building systems where they'll get some very serious benefits, in particular, they'll learn, I don't know, to hack states or countries' economies, or to analyze some volumes of data in order to push their business forward more seriously. That, by the way, is a big question here: will the new models that have appeared over the last six months really help a company like Anthropic or OpenAI push their business forward?
Push forward — meaning increase the number of users, get users to pay more money, make it so they earn money from them more effectively. From what we see today, there doesn't seem to be a direct parallel result of that, right. Even though the systems have improved incredibly. I mean, we won't see, we don't see a result from them in terms of some different marketing or some, some different visuals, probably, if you can put it that way. Although, once again, the quality of the systems is improving significantly. And there's the question: what's the race even for? I mean, what do they want to achieve? What is it they want to achieve?
Do they want to achieve the ability to, I don't know, create businesses and multiply them? Or do they want to achieve the ability to improve people's lives, only improve people's lives through what? It's not very clear through what and how. People have undeveloped, undeveloped knowledge when it comes to business. And you, say, give the whole world the ability to open a business. Well, what's going to happen? There's going to be a certain chaos, right? I mean, there's going to be mass competition, and a huge number of so-called quality companies appear.
It's unclear whether that's a positive or a not-positive effect. It's very unclear how this will affect the economy. We know, I did an episode at the start of the year, that over the past year the influence of artificial intelligence on the actual economy, on countries' GDP, is small, it's there. Obviously, it's normal that changes will happen, and the changes are already irreversible, right. I mean, still, with the arrival of artificial intelligence and access to information, there are irreversible changes not just of an economic nature, irreversible changes in terms of people's behavior patterns, the system of how these people interact, how these people communicate with each other.
I mean, the market is changing substantially, incredibly. Now I'll tell you a bit more about Opus 5.5 and about the application of models like these. At the same time, I'd really like, we have a lot of viewers too, including new ones on the channel, for it to become clearer about, clearer about this channel. First I'd like to ask the bulk of people, ask you: what's your feeling regarding the models, what's going on right now? Again, what are you using now, where do you see improvements, where do you see downsides, where do you see upsides, right? What movements are there in terms of models?
I recommend, by the way, that you watch my episode from exactly a week ago about plugins on our channel. I think it's a wonderful episode on whether you need any skills, whether you need any connectors, whether you need any plugins, and where we're heading at all, where we're going. Indeed, today we're in a situation where standing up websites, building reports, fairly complex design of apps, of various systems, is becoming a simplified task. At the same time, people stay the same as they were, and their heads stay the same as they were.
I attend a large number of different meetings and interactions with different companies. And honestly, I probably see maybe one in twenty or one in thirty cases, in terms of people, where people have truly changed in terms of their approaches to their activity and their work, and where there's a truly positive effect, right, from systems like ChatGPT. And on the one hand, you could say, take my wife, she doesn't work, she takes care of the house, we have four kids, herself, the family, the space, and so on.
I mean, well, a pretty large number of different things in a person's life. And she's been using systems like ChatGPT quite actively since the first day they appeared, right, on paid subscriptions. What I want to say is, well, if you look from the outside, has her life truly changed because of it or not? Has access to this information really brought some substantial changes or not? It'd be good, by the way, to get this information from the current viewers, new and old, and it'd be great to see how many new people there are on the channel now who joined recently or are watching an episode for the first time.
Has your life changed? Have you, for example, started earning more, or do you truly have more free time? And if you do, where did it come from? Well, for example, you used to do the work in eight hours, now you do it in one hour and don't tell your employer about it, for example. Or how did it happen, right? Because the bulk of employers expect that you won't be doing your work in an hour and then not working the other seven hours, or ten hours. The bulk of employers expect that you'll do what you did before in an hour or half an hour, and then for another ten hours do a huge amount of other work at a different level of efficiency.
So imagine, are companies ready to optimize their processes like that? I mean, if a company, for example, had a hundred people working, and a huge amount of the work got an efficiency gain of, say, five times, and this exists in many areas, for example, in finance, in analytics, in marketing, in editing, even in production operations. Well, in many places work could have gotten a result like that, an effect of five times, ten times, fifteen times with the help of ChatGPT. And the company, imagine, had a hundred people, and now it's as if it has five hundred or a thousand people in terms of processes. I mean, has everything really changed, really been digested?
Have people really started working at that speed or not? Or have people sort of started generating more documents, more text, creating more systems of some kind, while in fact no improvements as such have happened? I use a huge amount of software. I don't see that this software, well, or these programs, have truly gotten some super results in their development. I'm talking about serious solutions now, right? For example, companies like, I don't know, Booking or Airbnb, right?
Or companies like, I don't know, like a person working in Google Drive, sitting there, or we work in Gmail, right, in email. Or Google search itself. Has anything really started working differently? I mean, it seems that over these years search should have turned into some kind of miracle, and a person should be getting the right hints, getting some additional results, getting some filtering parameters, seeing some interfaces and so on. But none of that has happened. We really don't see any of that. But somewhere inside, some system really is being built that's capable of analyzing an entire state.
And today ChatGPT, Anthropic can analyze any number of people on planet Earth. That's obvious, right? I mean, it'd be no trouble for them to analyze a hundred or two hundred or three hundred million people in terms of certain records or data analysis. Clearly. And this gap is, of course, colossal in terms of where the bureaucratic systems, certain bureaucratic processes stand in the life of countries, economies, politics, states. And between this efficiency that's emerging. How do you feel in this efficiency? What's going to happen for you in this efficiency?
What's going to happen, or is happening, in terms of hiring people? Has hiring really changed in some way or not? I see all the companies: for everyone, it was hard to hire decent specialists, and it's still hard to hire decent specialists. For everyone it was hard to get people to work, I don't know, super well or more effectively, and it's still just as hard to get super-quality, more effective work. I mean, you supposedly have the ability to do more, but what's the result? Even if you look at video editing, the videos being made on the channel.
Well, some simple tasks, right, these are simple jobs, small businesses. Has something substantially changed, or have some substantially different results started to happen? Or, say, there's now the opportunity, I shoot content for ToTheMoon, there's content, the opportunity to shoot more. Or I shoot content, I have two more YouTube channels, or there's Instagram, or Telegram appeared. The ability to do something differently, or what? Well, fine, there's the ability to update the website more, create some additional solutions, create an unbelievable number of chatbots.
But in fact, what does the efficiency from that turn out to be? Or for the offline businesses that exist, I don't know, are more people coming into the coffee shop, or aren't more people coming in? Well, yes, people have started choosing restaurants differently, choosing cafes differently. And they've started choosing them so differently that it's practically impossible to track. Someone recently asked me a question about this, I was saying that aggregators will disappear, a huge number of aggregators will disappear. And, well, I'm deeply convinced of that.
By aggregator I mean platforms like Booking, for example, right? Or let's make it simpler. A restaurant booking aggregator, for example. Or not even booking, but restaurant search. Ratings of those restaurants. And someone wrote: "But how, Alexander, can they disappear, if the system will need to get its information from somewhere anyway?" I mean, the system will have a task, it will need to get information about these restaurants from somewhere. Where will it get it from? I said this: today, systems like OpenAI have far more information about those very same restaurants, those very same cafes, those very same hotels than any aggregator.
I mean, we think that's where the people live. They booked something through Expedia and then, I don't know, left a rating. But in reality, in terms of communicating with the system, when I go to some hotel or fly on some plane, I ask ChatGPT far more questions than I ask in Expedia or Booking, or directly of the operators who even sold me that flight. I ask ChatGPT which seats are better on the plane, where the better exit is on the plane and so on. I mean, I communicate with it endlessly.
I have, again, for those who haven't heard this, let me remind you, I believe the system will start developing very strongly, and we'll see revolutionary changes, when they start surveying people. And it's fundamentally interesting why this still isn't happening. I mean, it's not like it's some clever, some super-genius idea of mine. But it's not happening, on the one hand, because, most likely, these companies don't care about people at all in general, they don't have the goal of helping people.
Because if there were a goal to help these people, then people wouldn't be reasoning about how some super-powerful artificial intelligence will be created and get out of control, but would be dealing with the ordinary life of ordinary people and putting out metrics and results on where it's stronger, not in programming, not in terminal programming or, I don't know, some business processes, right? But would put out, would make some tests that evaluate the life of a real person, and would try to help, at scale, at scale, the real person.
And helping this person, of course, is possible not only, it's not a question of a computer or a phone, it's about interaction with a huge number of different devices in general, right? In several years we still haven't gotten any changes in terms of smart home systems. I mean, smart home systems worked terribly, awfully clunkily, and so far they keep working terribly, awfully clunkily. I mean, we don't see super improvements, we don't even see cool improvements in car assistants.
Well, you can see slightly more improvements at Tesla in terms of its own assistant. And even then, it's all slow answers, not fully, not fully good in terms of the speed of processing all this. Where these models are located, how they're set up, how well they can work in a distributed way. And if the internet drops at that moment, how well does this model work or not work, and so on. I won't presume to speak for all cars, for all systems, and for those who have local models, whether, if there's internet access, they connect easily to the cool systems, if there are cool expensive subscriptions, at two hundred, at a thousand dollars, whether these systems can process information very quickly and so on.
Probably the systems depend on the different cars, but I have different systems both in cars and at home, and I see it with friends too, from their feedback, and from various reviews. The system works very peculiarly so far. Very peculiarly, relatively peculiarly in terms of real value. Although, once again, we see a very, well, insane lead, OpenAI too. This positioning of OpenAI's Astra, it was still a somewhat new positioning.
They did this positioning where they said that this system is precisely supposed to communicate less with the person. But the funny thing is that their previous system also didn't really communicate with the person and didn't ask them questions. I mean, on the one hand, again, they don't ask the person questions because that isn't a goal for them. On the other hand, they don't ask the person questions, possibly, because they're afraid. I made this assumption recently, by the way, you know, a few episodes ago, I think.
Possibly they don't ask because they're afraid the system will get super-access. I mean, if the system gets the ability to survey a billion people, well, or, let's say, a billion is relatively active users, right? Let's say it gets the ability to survey a hundred million more or less active users overall. I mean, not just a person who logs in once a week, but, well, slightly more engaged users. Or, say, ten million highly engaged users. After all, people who are highly engaged have a completely different level of engagement altogether. Yes, we have to understand today that the gap between people is colossal.
That's not one percent there, even ten million people, but zero point one percent, probably. That million people, they have an absolutely colossal lead over those hundred million people who use it. Even if the system got the ability to survey a hundred million people — that's very serious. I mean, what volume of data you could collect, what information you could find, what, what analysis you could build. The thing is, it'll cost even more money, and even more money to maintain these systems, even more complications.
I mean, whatever we say, look, Opus 5.5 came out, and they say the system spends, like, thirty or forty or fifty percent fewer tokens, for example.
To this day people don't understand what tokens are. Probably only the creators of all this understand it. And even that's a big question, whether they understand. And how can you even compare systems to each other in terms of tokens, when what you actually need to compare overall is the end result? That's obvious. A person asks themselves the question: was the result achieved or not. Now, as you all know, I've talked about a lot of tests. I have a huge number of different tests, right, a lot of different systems, dozens of different systems. And one system, remember, was very simple, where I did analysis of heart-related data and hooked up special sensors.
Then I collect and analyze that data directly. It runs for me straight from the chat. Remember, there was one case where I very quickly wrote software for the Mac so that it would read from a sensor I have, the Polaris sensor, I think, or whatever it's called. And this sensor is for professional measurement of everything heart-related. Or sort of semi-professional measurement. Twenty-four seven. And I really didn't like that I had to launch one piece of software, then launch MATLAB. And essentially I was interested in getting data straight from this sensor and passing it to ChatGPT.
And I wrote a program to start the sensor. Now I realize I don't need the program. And there were always problems with the program. It was eating up the wrong kind of memory on the Mac, the sync was somehow off. I realize that all my syncing works through Codex. I mean, I just tell Codex: start my sensor for the night, it just starts it, since it's right there nearby, and that's it. In the morning I tell it: turn it off, download the data. It downloaded the data for me, and right away I tell it: build me a report, or compare the readings for me, or do whatever I want.
For example, go into my account, say, with my current provider and send this data to my doctor. Or compare this data with three years of cardiogram data. Or find me in my mail, in my notes, in all my files, whenever I've had data on how my heart works, and make a variety of it. My system is pretty healthy. Everything to do with the service. But I was curious. I was studying things related to sleep, related to various detailed, modern health-analysis indicators like that. Right? So the software isn't really needed. And now, with GPT-6 Astra in Pro mode, to analyze a few days of the report, I had to make three or four requests to it, ask it.
It spent hours of work, literally hours, and the results it gave were very so-so. Even though the question is very simple. And before, it seems to me, in terms of compute spend, the systems were, you'd think, worse. For example, ChatGPT-5.5, right, or 5.0. Clearly the costs were much lower, and the effectiveness was more or less the same. That, of course, is very, very seriously confusing. I also had an incident today. I launched one task in Claude, and it ate all the tokens in two hours. Subscription ones, I mean it ate the subscription tokens, not the paid, purchased ones, right?
With paid ones, what does "eat them all" even mean? You hook up a credit card and spend money. But it ate the subscription tokens. And there, to be fair, it was set to Fable. It was Fable 5.1, and it was probably my mistake, probably, to run that task on Fable. But I didn't expect it to eat tokens like that. And so it's no accident that many people write in. Someone writes: my GPT-6 Astra or Fable eats tokens instantly, and someone else writes: I can't even use up all my tokens on the free version.
I see people like that often, and my question is: what are you even doing in the free version? What do you mean you can't use up all the tokens? Is there no option to pick an expensive system there? Well, okay, there's no option to pick an expensive system there, but if you launch reasonably serious tasks, I think the free version should run out instantly. Well, unless you're sitting in certain weak versions. Anthropic, of course, with the release of this 5.5 version and the fact that, look, terminal work, there's this metric, for everyone it may not be super important, but I'll say it anyway.
It's sixty-six percent. Its test result is sixty-six percent, while Fable's test result was fifty-five percent. Astra's is something like fifty-seven percent. Programming — fifty-four percent, while Fable had fifty, and Opus 5.5 — fifty-four. So what now, all of YouTube should be flashing with headlines? A new Anthropic model has appeared on which you can do crazy things. From a single prompt you'll create a website, from a single prompt you'll generate a 3D world, from a single prompt you'll generate eight hundred presentations.
But it seems you could generate those from a single prompt before too. I think for the last six months ChatGPT in the Pro version, Extra High mode, Pro, could generate anything at all from a single prompt pretty well. There are things, of course, that have improved significantly, yes. Like the quality of search, the quality, probably, of document visualization, creating documents, say, in HTML format. The quality probably keeps getting better and better and better. And of course the system is getting better. Look, yes. Of course they're getting better.
The only task where GPT-6 Astra gets a significant win over Opus 5.5 is scientific tasks. And even that is by seven percent, that is, fifty-eight percent versus sixty-four, so 5.5, Opus 5.5 seems to be ahead of Fable. And that raises a big question for me, essentially. If Fable is a system that's capable of hacking anything and everything, everything that's out there, I mean its level of data analysis and its level of building various systems is colossally high, then this Opus 5.5 released to everyone, which is ahead of Fable, what is it actually, really? Or did you make Fable worse?
Because there was a moment, and many viewers here know, that Fable was made worse in July. By the way, a few days ago we ran a poll on our channel. Who hasn't used Astra, and more than thirty percent of people there hadn't even used the new Astra. I mean, people don't fully understand what video models are coming out and what's being released. So people use, obviously, and will keep using some one version or one model. It's strange to even have a choice between these versions. We've talked about this a lot.
Maybe some feature where I'd want to spend more money to get this task solved for me. That, yes. But these sliders with a choice of version and a choice of thinking or engagement modes, they are, of course, significantly, significantly problematic. By the way, I don't know how ChatGPT will solve this problem at all, which very strongly sets them apart from Anthropic. In ChatGPT, tokens in the chat aren't counted together with Codex tokens or Work tokens. It's unclear how they'll solve that at all.
And it's unclear what they'll do about the fact that there's Code, there's Work and there's ChatGPT, what they'll do with these three systems, how they'll merge them together. Let me remind you that Greg Brockman took over this project, I think, at the end of spring, or right at the start, or in June. Our editors might dig up that information, when they said outright that they'd be making substantial changes to merge everything into a single app. But you can still see that there's the chat and there's Codex. The systems are kind of disconnected. And there are people who use Work and who say: "Work is a great thing."
And you sit there and you don't understand, what's that even for? Exactly the same as people who, in Anthropic, separately pick Cowork, Code, or just the chat where they leave messages, just Claude. And these interfaces, essentially, seem to carry some different meanings, or different functionality, or different skills, or connectors. And on top of that it's called something different everywhere. And you don't understand why the system goes into one mode or another. Is this, at this point in time, a kind of pause in the development of these systems, or is it after all some substantial, substantial movement forward?
By the way, Opus 5.5 has some interesting numbers regarding its settings, right. In Claude there's, like, Low thinking, Medium thinking, High, Extra High and Max. So, between, for example, Extra High and Max, there's this notion of an intelligence index, right, a sort of mind index. So, for Max it's fifty-eight percent. Fifty-eight — it's a number there, not percent. Fifty-eight, yes. And for Extra High — fifty-six. I sit there thinking, it's not very clear what that means. Because Anthropic's index is their own scale for analyzing artificial intelligence, right, and a certain kind of intelligence, you could say.
I mean, it's not a percentage of correct answers, it's not some IQ. And the cost, by the way, for Max is obviously higher, but Extra High mode doesn't differ hugely from it. And recently someone wrote in the chat: "I've optimized all my requests and can't spend money at all, because I just work in a small context window all the time." Maybe you really do need to work in small context windows, and then the tasks will be, then the resource becomes insane. As for small systems, various open, open-source systems or models that you install on a computer, that's a whole separate topic.
And how relevant are they at all, and is that relevance even needed today? Or are we still moving toward a story where cheaper, more mass-market models, super cool ones, will be created? Then how will open models keep up with that, or the various chips for running these open models or small models of your own? Or will the cloud and these systems still be far more cost-effective? I mean, this model, for example, Opus 5.5 Low, its price is ten times lower than Max's, and the intelligence level is forty-two versus fifty-eight.
So the intelligence level still remains pretty serious. And the model costs almost nothing. The question is: how then will other models coexist with this, and what's going to happen? One more thing, by the way, one story that seems to me quite important. We don't really see any competitors to these models.
I mean, over the last six months Anthropic and OpenAI have gained a simply striking lead. I mean, there was a moment when Grok was still visible, there was a moment when, I don't know, Moonshot Kimi was pouring something out, people were looking. There was DeepSeek, right, we all remember that hysteria, when it was visible. But now, of course, as for Anthropic and OpenAI, who is even anywhere near them at this scale? There's Google, obviously, and Google will be there given their Google search and everything else. But the question is whether Google will really be able to hold parity with Gemini.
Because if Google didn't have its own search and a huge number of services, Google Gemini would be in huge trouble and difficulty. And we all understand, with the release of the new iPhone, no matter what new systems Apple makes and however it presents artificial intelligence, it's of absolutely no interest, nobody is interested in that Siri. And, well, write your thoughts on this, yeah. And I've got the new iPhone 18 right there, but I don't care how artificial intelligence works on it today.
Even though it supposedly has twice the power for running artificial intelligence. You just stopped using it, because it's not functional, it's not interesting to you, it doesn't do the tasks you set, it doesn't solve your problems at all. And today there isn't, there's still no system that could work in different places at the same time and switch between them easily.
I mean, to this day, even to set up Codex to work between one computer and another, there are nuances. There are nuances everywhere. In Claude Code you still can't, if it's connected, say, to a repository, a cloud one, you can't upload files to it for some reason bigger than, say, twenty megabytes, for example, a single file. Or you can't tell it: work with my, with my file system on the computer. I mean, there's no certain universality. And they haven't fully learned to work really well with system folders. Today I'm doing an episode, again, not about, it's perhaps more of a philosophical one.
And you know that I generally like to make these short philosophical episodes. We make practical episodes too. In particular, as I told you, the episode about plugins that came out recently. Or this Wednesday we had an episode about finance, whether the systems lie or not about finances. By the way, watch it, it's astonishing. In the latest research from September, from the end of September at that, how the system deceives on financial data. One more astonishing aspect. I don't know whether you've noticed or not that the systems have become very slippery on many topics, and they seem not to want to talk about those topics.
I mean, when you start asking the system whether, say, my computer could be hacked, the system seems to start hiding things on those questions. I mean, it's clear that it hides these things, in particular, from those people who use such questions precisely in order to hack a computer. But, on the other hand, it can tell from a huge number of chats or from the details, from all the circumstances, what you're actually doing, what you're really doing. Or at least show a message and say: "Listen, you are, for example, in a risk zone, you need to go into, say, a certain isolated environment or do something, or go through some additional verification to get access to this."
By the way, remember that situation when Anthropic put out Fable, then blocked it, because the US government issued an order that this model couldn't, couldn't be used by non-US citizens, right? I mean, even employees who worked at Anthropic and weren't US citizens, and there are plenty of those, of course, they couldn't use this model. So which model are we using now, as far as Fable goes? And how did they come to an agreement among themselves? And what did they discuss, and what did they decide?
I mean, are there actually restrictions or not? Or are there companies that don't have these restrictions? Or the government, for example, has no restrictions. And if Anthropic has no restrictions, they have unlimited compute, or OpenAI has no restrictions and they have unlimited compute, then why don't these companies launch a cycle of creating separate business lines or businesses for themselves? For example, why doesn't Anthropic separately stand up a huge PR department, stand up a huge separate marketing department, one that's so singular, a singularity, technological-singularity marketing department, AGI, right, or ASI. Why isn't this happening?
Why don't they stand up and build and take on whole industries? They could be capturing an incredible number of markets, right? They could be creating a huge amount of software, they could be creating an incredible number of apps, but we don't see any result from that. And you don't need to allocate ten percent of your compute for that, it's enough to allocate one or two percent of compute and take these markets straight for yourself, take these markets, bring in other companies and so on.
You all know, ChatGPT has this system called Sites. It's a system that lets you create websites inside the ChatGPT infrastructure. Many people, by the way, don't know about it. And host those sites. I mean, these sites aren't just created, they're hosted there. And Greg Brockman, I think, said more than ten million sites have been created and are hosted. Our editors, let them put up that information too. I mean, this is an example of how to capture the CMS market. Well, obviously, yeah, how to capture the website-building market. Because, well, let's be honest: in a year it'll be strange to build websites using some CMS systems.
Old sites, or migrating old sites — okay. But all the new things being written will be written outside of CMS systems. Since people aren't super advanced in terms of perceiving reality, they'll now keep using various CMS systems to build new pages on a site or new functionality on a site and so on. But at some point in time that will start to change, and people will stop doing it. Just like people use development teams in-house. Many development teams still don't use artificial intelligence all that heavily, at the level of effectiveness these companies need.
They're not hugely interested in it, or somewhere they don't want to, or somewhere the knowledge is lacking, or somewhere the business can't pick it up. But the world, this is going to be changed. I mean, at some point in time, I don't know, marketing departments will stop ordering this from developers, they'll do it themselves. Sales will stop ordering this from developers, they'll start doing it themselves. I mean, at some point in time they'll start building these systems themselves.
That doesn't mean developers will cease to exist. Again, I made an episode on this topic, the one where I'm driving in the car. If anyone hasn't watched it, watch it, we'll put it up on screen now, that episode.
I was driving in San Francisco just then, filming on this topic. I mean, I'm sure there will be and will remain a huge number of people who will be developing these systems. It's just that right now a new world of new people with new opportunities is emerging. The question is whether they'll be able to enter and get into the competitive environment, and how they'll fare in that competitive environment. Today, of course, for capable people, for a certain slice of the world, an opportunity has appeared to suddenly make a lot of money, suddenly gain a huge number of connections, gain a huge number of opportunities.
Just incredible. I mean, some percentage of the world, say, I don't know, zero point one percent of the world got some kind of super-opportunities. Yes, these are people who have enough brains, at a certain point in time, to come up with the right solution, to make the right decision. Although everyone wants to do that, of course. Well, or a lot of people want to do it. Although at the same time I see, of course, how to this day eight out of ten people, probably, resist, and to this day artificial intelligence is very much a marginal thing for them.
I was just flying, coming back from Europe to the States, and saw how on two planes, well, or three, probably, some people are sitting and working with big spreadsheets. You see a person working in a spreadsheet for the whole flight. For example, you fly for ten hours, and he works for ten hours in some particular spreadsheet. There's a huge amount of information in there. And he doesn't open ChatGPT or, say, doesn't open Anthropic. And you're like: "Why isn't he opening them?"
I know what the interface looks like in these systems, and I can see that he doesn't open these systems, he doesn't launch these systems, he changes something in these Excel files by hand, double-checks, writes, reads some emails. Often it looked like some accounting data or analysis of some accounting data. You realize that this accounting data is presented in some super inconvenient way. I know very well what inconvenient Excel files look like. That's how they're presented there. I mean, people aren't using these systems. Will these people be able to use such systems?
Will they be able to adapt to it? And as I said at the very beginning of the video, the question isn't even whether they'll be able to adapt, the question is whether it will really give them an effect. Don't forget to subscribe to our channel. Here we talk about artificial intelligence, the world of technology, and how to really use it for your own life. And of course, today I'm expecting a huge number of comments from you, including on what you think about the new Opus models, well, and all the other new ones, and all the other models.
So, see you in the next episodes.