Hi everyone! We're on the To The Moon channel. Tech news, insights from Silicon Valley from all over the world. Everything about artificial intelligence. So, look, Anthropic has released Fable, the new version 5.2. And it would seem like such a small update, there was 5.1, now 5.2. What is it? What's the difference, right? Or Anthropic also has Opus 5.5. You can get confused in all of this. Today I want to tell you directly, so that you really get into it: first, the difference, understanding the difference between the Opus and Fable models, and understanding the Mythos family of models in general.
That's where Fable comes from at Anthropic, compared with, say, ChatGPT or other systems, on the one hand. On the other hand, I want to talk about some things for when people think they understand artificial intelligence. I'll try to show you that topic today too. Just yesterday I went to Napa. It's a region a bit north of San Francisco, known for its wineries. We just went there, it's a very beautiful region. And on the way I was talking with a friend of mine, and as always, basically, the conversation turned to artificial intelligence and the fact that I have a channel, the To The Moon channel,
and as always, we ran into the person saying that I use artificial intelligence, and I have various agents, and we analyze clients, and our service works with it. We'll touch on that topic too, right? And I told him I'd give him the best advice of his life that I've ever given. And he has done a lot in business and grown thanks to my advice, among other things — he really believes that and says so explicitly. He mentioned it again yesterday. I said this would definitely be the best advice of his life when it comes to artificial intelligence.
Plus I sent him a couple of videos on our channel, where, remember, there were videos where I talked about the 10 best trends, and then another 10, just recently. And I told him that after these videos he should give himself a score from zero to 100 for how well he knows and understands artificial intelligence. So right now you feel you're at 60 points, right? Then give yourself a score afterwards, based on how you feel. Well, obviously, after that you should essentially feel that you understand and know artificial intelligence at zero. Look, there can't be any other feeling, right?
If you're new to our channel and you watch both videos, then basically a person should come away with the impression that they know it at zero. Unless, of course, they're super involved, a professional user, who has just come to us here. Such people really exist too, right? And you should calmly figure this out for yourself here. So, look, Fable. Fable 5.2 is out. Let me remind those who don't know what Fable is.
Fable comes from a family of models at Anthropic. There's a family called Mythos. Mythos models are basically not available to people; they're available to special companies or specially selected people, because these models were originally developed essentially to serve security systems. That's what they're called, right? They're artificial models from the security standpoint — cybersecurity, finding various vulnerabilities. And one of the best-known cases was at the end of spring: these models were used to analyze a large number of open-source systems and found more than 10,000 different bugs and errors there.
So this model can not only find vulnerabilities, it can also create vulnerabilities, right, and create various problems in different systems. Thus, essentially, Mythos models are probably the most high-class weapon in the world today in terms of cybercrime and in terms of software. The highest level of weapon in the world, right? Essentially, if you run even the Fable model — Fable is available to people — and tell it to analyze your own software, whoever has software, for example a website, or
some CRM system, or your own home-grown system, or even your own server at home, or your own computer, and you say, find various vulnerabilities, it will give you, I have to say, quite a large number of, well, serious problems, right? And it will let you close holes that programmers in the world, as humans, could never do. Never. I mean, so immediately, so far in advance, at such a scale, they couldn't do it, right? Now maybe a mountain of comments will pour down on me. By the way, whoever thinks what about this, especially current or former programmers, or those who really, really understand software development.
But as a programmer in the past myself, and someone who understands building various systems quite well, I don't think a human was ever capable of doing that, right?
So this family, this Mythos, Anthropic's models, was in closed use. And at some point in the summer Anthropic releases the Fable model. And the Fable model caused an incredible sensation in the market. But what was it known for? For being the first model that the American government officially blocked, right? That is, the American government announced that this model could pose a threat to national security, a threat to people, to US citizens. Clearly the American government, well, thinks about its own country. And this model had to be shut down. Why?
Because information appeared that supposedly with this model people could create various kinds of weapons — cyberweapons, and offline weapons too, right? You have to understand that today models like Mythos make it possible to create, among other things, serious biological weapons. Not only weapons — they also, of course, make it possible to make very serious discoveries in medicine and biological research. And everything to do with human development. For example, a certain company, people used, for example, these models from China, and they were blocked, because it turned out that this lab was in the Ministry of Defense of China and so on. So there are cases like that.
And people have learned to fake accounts, people have learned to buy and resell various accounts online registered to different people. There are systems that fight this more harshly, and some that fight it less. By the way, we'll definitely have a special episode about blocks at OpenAI, ChatGPT, Anthropic, Anthropic, various systems. It's a very current topic in particular, I'll tell you about it in depth: what people get blocked for, why they get blocked. And this information is needed not so much so that you don't get blocked right now, although that will also be useful, but so that you understand what will happen in the future.
You work with this system — what will happen to you in the future? For example, right now we're coming back with my wife, with Polina, from Napa, and she's sitting there having to listen to this content, which is probably not easy for her, maybe somewhat complicated, or maybe it'll be useful to her, right? You know that I like to say my wife would benefit from using Codex from ChatGPT. Maybe she'll start using it. And what would be interesting and important for my wife to know?
She has been using ChatGPT for three years now, right? And ChatGPT has accumulated a huge amount of data about her. How well does she really understand what data is stored in there, right? Well, how well does she understand at all what data is collected around her on her various devices, how well does she understand what is stored inside this system, and whether she really needs sometimes, when asking various questions, to adjust, let's say, let's say, if one could put it that way, right, to adjust her reasoning, in particular, in using systems like ChatGPT or like Claude Code — not Claude Code, but Claude from Anthropic, right?
Or you use DeepSeek, or you use Gemini, or you use Grok. By the way, with Grok, for example, since these are Elon Musk's models, with Elon Musk that's a big question altogether, because Elon Musk has one development strategy for Grok today, and it can change, right, which is what actually happens, for example, on Twitter and X. So, on the one hand, the strategy is maintained there, but on the other hand, he may come up with some new opinion about what people can ask Grok and what they can't, right?
So let's go back to Fable. The American government blocks Fable by order of the president and the Department of War. A big topic. At first there was talk. Look, for those who are interested, I'll walk you through various strategies and various conspiracy theories. First, Anthropic is supposedly a company unfriendly to the government. Why? Because when President Maduro was captured, information came out that Anthropic was used in the soldiers' work. And, as we know, they captured this president instantly.
I mean, the operation was incredible, right? Anthropic was used, and clearly many people don't like that. People thought Anthropic could be used against people. And supposedly Anthropic's leadership spoke out that this isn't good, you can't do that, it's wrong altogether, and we're against such things. And right away statements came from the White House: "And who are you anyway, Anthropic, to forbid us to do this? We'll exclude you from all our systems altogether." By the way, just a few days ago the US president said... this is very interesting.
He said that supposedly there was some person, the head of one of the most leading artificial intelligence companies, whom they called from the White House and told: "Listen, pack your things, kiss your wife goodbye, you'll be wearing orange clothes. Meaning you're going to prison." This is a real topic he talked about a few days ago. And of course, by basic reasoning, who should it have been? It should have been the head of Anthropic. Well, who else could it be? It's definitely the head of Anthropic. It's unlikely they told him: "You should be, like, more accommodating," right? And then everything changed.
And that's what he's talking about, that then things changed. It's unlikely to have been Jensen Huang, the head of NVIDIA. Now, given the scandal with Jensen Huang, several people have already asked me whether I'll make such an episode.
Well, it's strange to make an episode on our channel about someone's love affairs. But this topic simply drew the whole world, because everyone loves discussing rich people. But what makes this topic interesting? The well-known story is that Jensen Huang, the head of NVIDIA — and I'll note that NVIDIA is number one in the world, the company that made it possible for us today to have the artificial intelligence we have. Because it's precisely thanks to NVIDIA and their most unique hardware, their most unique boards, that they managed to create such artificial intelligence, right?
And all the super-cool infrastructure in the world is built on their chips. And that's why I've always said that catching up with America is practically impossible, right, even if China has its own production of certain chips, and even then the quality is a big question, because they all try to buy NVIDIA. So what was the head of NVIDIA suspected of now? I already posted it, many people don't know. Information came out that he was cheating on his wife, cheating on his wife with a Chinese citizen, an employee of a Singaporean company.
I think she's a citizen of China or Singapore. She'd been in photos with him for several years, a year, two, three. And supposedly, according to the New York Times, she's an agent of the Chinese intelligence services. And she was leaking information to them. Well, and the whole internet was flooded with videos. I don't know if our editors will show it or not — AI videos, where the president of China, you know, recruits this woman and watches everything. Including videos where they recruit Asian women for Elon Musk and so on, yeah. And the whole internet is flooded right now. I don't know, Polina, did you see?
There's a lot on the internet: whoever has Asian wives — spot the difference, right? Well, it's sometimes hard for Europeans. People look alike, yes, that happens, just as we look alike to them. It's hard for them to tell, sometimes it's hard to tell who's who. And it seems like it's the same woman. Well, those are the jokes going around. Obviously these are jokes, and people differ a lot. And I, for example, by the way, can tell pretty well between citizens of China, Japan, Vietnam, Thailand and so on.
Although of course I can just as well be wrong. We often have discussions here where we guess: is a person Belarusian, Ukrainian, Russian, Polish, or French, or German. It's very hard to tell, yeah. By the way, can you tell, those of you on our channel? Don't forget to support our channel, share your opinion, leave comments. And I'm especially interested whether there are, for example, people here who had access to Mythos, who used Cyber, who used Fable. Write your comments.
We have a big, great community. People will happily read who has what where. Well, you can write about Jensen Huang too. So there's a whole debate going on now that Jensen Huang must by no means be removed from the board, otherwise NVIDIA will have a problem. But in short, the US president is unlikely to have meant him. He certainly meant the head of Anthropic. One of my theories, though, was that maybe this is marketing for Anthropic. Let me remind you that Anthropic is a company that came out of OpenAI. Yes, and all these people originally come in a chain from Google.
And notice that Grok comes from there too. And here we have four leaders in the market. It's Google Gemini, DeepMind, that's Google DeepMind, it's OpenAI, it's Anthropic and xAI. All these companies are incredibly closely connected. And these are the four leaders of the global artificial intelligence market, yes. Once again, xAI — in the Russian-speaking market it's often strange to hear this — xAI is obviously a world leader in artificial intelligence, obviously a world leader in artificial intelligence.
I was just saying recently, you know, right, that in the model lineup the Tesla Y has now surpassed sales in China in the auto industry, well, in a certain category it surpassed other cars — that was September. We'll show the statistics now. And Tesla sells incredibly well in China, even though it costs much more than its competitor counterparts, almost three times more expensive, somewhere two, or four times more expensive than Chinese competitor counterparts. So xAI — and xAI is the company SpaceX, and there are essentially discussions now about merging it with Tesla, right.
This is all the infrastructure of Elon Musk, the richest man in the world, with the biggest — essentially these two, these two companies give one of the largest capitalizations, not the largest, but they're in the world's top ten today, right. Let me remind you, SpaceX is already at two trillion, plus they were floating two trillion somewhere on this topic, but that's together with Tesla. And the artificial intelligence inside is the artificial intelligence of this corporation, right. That's very important to remember.
So I do, of course, have one hypothesis, which was that there are certain arrangements there, collusion in the market, and Amodei plays by the rules of certain people, among others. We'll see what happens in the future. So of course, in the summer the US government unblocked Fable back then, it was released after several changes. And there was a very cool moment there. There was a moment, before, before it was blocked, before the Fable model was blocked, I remember I was in Santa Barbara at that moment,
and the kids were swimming in the pool, and I launched Claude Code from my mobile phone, and I was blown away by the quality of its work. And for me it was the first time a system worked that seriously. Not some small stuff, but it really worked seriously and did really big, serious tasks. And at such a level that you'd just describe a task by voice somewhere, and it, knowing your infrastructure, understood what you wanted, right, and carried the task through to the end. So it held a long cycle of work. And that, as we know, is very important, right, being able to hold a long cycle in development. That is, to hold the context ahead, over length.
By the way, I want to tell you all that modern models are not models that work only with your one context. Modern models, especially a model like Fable 5.2, are models that, while working, in parallel with the main work, constantly re-check themselves and launch new sessions. So you have to understand that you don't have just one session running, right, when you work with Fable. You have parallel sessions launching, something else is constantly moving. And essentially, in the future, the task will even be done, and the session will keep working. That's the point. This is the technology made in ChatGPT Dots, essentially, right.
Dots, by the way, is of course an unreal thing. On Sunday we rented a house here with my friend, we'll go north for five days, shut ourselves in and work. And as soon as I had rented it — Dots — imagine, Polina, an email about my Airbnb booking comes to my inbox, and five minutes later it immediately sends me a message and says: "I want to point out that you've organized a five-day trip here, but you have conflicts. First, your daughter has a concert on Sunday." And second, a piano tuner is coming to you on Monday. So it instantly — I didn't ask it to do that.
So it's a system that works in parallel, right? And it's cool when a system can work after you've done the task. Imagine you've made a website, imagine you've made some module in some program, or some app, or made a design for greeting cards, or drawn up some contract. And the system tells you: "You know, here you, Polina, for example, I don't know, signed some paper or some contract and agreed to hold something, and you've already prepared it, done everything, and a week later the law changed."
And the system tells you: "I want to remind you that you have, I don't know, an event in a month, but the rules have already changed, the price changed, something else happened." So Dots lets you keep track of your work further on. Of course, right now it doesn't spend any super-large compute, so this tracking is still relative for now.
And so back then, when I tested this Fable in the pool, it was of course incredible. And back then Fable was cheaper, it wasn't limited, they hadn't restricted it yet, especially if you had an expensive plan — well, they launched it on the expensive plan, and Fable showed an incredible result. After that it was blocked, opened again, restrictions and limits were introduced. And many of you will have limits now. And for people, I want to say right away, for people on a cheap Anthropic plan, you basically can't use such models, but it's important for you to listen to this, because this is the whole point.
Why now you often need to pay an extra 100 or 200 dollars, right, or pay 300 dollars, and sometimes get a task done with such systems — because it's a system of a different class, a different class of work altogether.
So then we had changes, right? At that point in time they release Astra. First they release the Sol model from OpenAI. It causes a certain sensation in the market. Its specialization was precisely that it can hold a long context window. Let me remind you that at DevDay — two months ago I was at OpenAI DevDay in San Francisco — at a closed session some of OpenAI's strong developers talked about having sessions that run for 60 days. That is, you work for 60 days in one context window. You don't reload the chat, essentially, right? Then, after this Sol, they release Astra.
And Astra is essentially a model that was built as the strongest competitor to the Fable family of models. Changes were happening at Anthropic too. They released Opus 5.5. And I believe, of course, that today it's a truly top model. In my view, for mass development, building various business processes, mass automation, it seems to me it's number one, right? At the same time I remain a daily user of ChatGPT itself, for chats, various basic life tasks and everything else, and of Codex, right?
But for building various systems Opus is, of course, very serious, very cool. If you can afford even two plans at 100 dollars, it's in both, buy yourself ChatGPT for 100 dollars and Anthropic for 100 dollars. You'll feel a really strong effect. Here a very interesting thing happens. Because we've all been using a large number of models over the last, say, three years, all people stopped understanding models altogether, right?
Polina, do you know what version ChatGPT is? She doesn't. She shakes her head, doesn't know. Why? She knows what ChatGPT is, right? But Polina is one of one point two billion users, I don't know, one of 50 million active ChatGPT users, right? Before, she was, I don't know, among five million, now she's among 50, right? Because she hasn't started using it in some super-cool way now, she uses it the same as before, right? But at some point she was, of course, in some narrower group, but still, she's in the 50 million. But which model, which number, she doesn't know. Does she understand what modes there are?
Do you understand what modes there are? Well, she hardly does. Here she taps, she has Medium open, and she has the paid version. The paid version has this Medium. That's the maximum you have, right? You have the regular paid one, twenty dollars, right? So there's Medium, GPT-6 Medium, right? But I usually mock all of them here — her, her friends, or people who don't choose the version — and I say: "Set it higher." So she has ChatGPT-6 Medium. Do you know the difference between Medium and Small, or something? You don't, right? There's Pro, High, all that. Well, for her it's some kind of nonsense, right?
Of course she doesn't understand it. Just as you don't know how many megapixels your iPhone has, right? You don't even know what a megapixel is, right? I mean, what normal person would know that? And here a very serious confusion has arisen in the market, and it persists. An insane confusion. What's more, ChatGPT also has different modes, these Medium, or Ultra Code, or High, depending on whether you use ChatGPT or Work, right, or Codex. They're different. There's a merger happening there now, but on the whole. So I want to bring you some clarity here with examples.
Imagine you're comparing Fable and Opus. And we've decided to determine which is better. But to determine which is better, even if we're doing... Now the question is what task we're doing. Even if we're determining which is better, we need to understand what mode we're working in, what mode we're using. Are we using the simplest Fable mode, Low, or the Medium mode, or Extra, or High, or Max, or Ultra. Well, in ChatGPT, I think, Ultra Code is the maximum mode. Again, I won't say for sure, they may be changing them in different countries now.
But there's one nuance: when you run a task, for example, on a simpler model, in a lower mode, like Medium.
Once again, there's the model, and there's the mode. Usually. There are actually other things too. Usually. Then this thing thinks longer. And, for example, a task may take three hours. The task will be solved only thirty percent, or not solved at all. While another mode can solve the task in ten minutes and solve it one hundred percent. And in the end you can spend on a more powerful model, which costs much more, whose tokens cost much more, get used up more and so on — you can spend less in money.
If you take cases in the world that people talk about — for example, I don't know, a system copies the Shopify website. Well, imagine, you live, you want to copy and make a full copy of some e-commerce platform or online marketplace, right? Or a full copy of some website, I don't know, doesn't matter, Booking, or some store, I don't know, Zara, any topic at all, I don't know, Ozon, if someone lives in the Russian-speaking market. You want to build some online marketplace, a fairly serious store, right? These systems do have serious stores, after all — so these models will behave very differently in terms of the end result.
And in examples around the world it's written somewhere that one model spent three hundred dollars in one mode, another spent nine hundred, the same thing, or the other way round. And the most interesting thing, look, incredibly interesting: if you take all these cases and compare them with each other, we won't see any correlation as to what is more cost-effective and what is more correct.
So if you watch some blogger who tells you ultra-rules about which mode to choose, when to choose it, why to choose it, which model, how tokens get used, that he checked everything completely — that's a conditional truth. And even for that same blogger, the same task may work differently today and the day after tomorrow, because models change, and you don't know which group of these tests you landed in. It's unclear what you had, how this task was solved, in which country you solved it, by the way. Whether you were, for example, in Europe, in Asia or in the US.
And here's a well-known story, when I spent, I think, ten billion tokens in ChatGPT — that was, I think, July 2026, exactly when they released Sol. That's when I set a task and left, and it was solving it for me for two and a half days. Two and a half days later I came back and said to ChatGPT: "Well, did you solve the task? I think you're stuck." And it tells me: "I'm still only solving the first item." And I realize it simply burned through these ten billion tokens of mine. Just burned them for nothing. It was horrible, of course.
Fable, by the way, has never spent that many tokens for me. I mean, I haven't had that with Anthropic, but OpenAI is known, first, for giving away tokens, simply giving them to people. That's the first thing, right? And the second thing: OpenAI and resets, for example, giving and so on, right, doing that. And Anthropic isn't known for that. Right now we all know OpenAI had a whole month of various resets of everything. So once again, look, there's no fully verified correlation that, for example, Fable 5.2 in Max mode will do better than Opus 5.5 in Max mode.
Fable 5.2, or 5.1, in Max mode — that's their fourth mode, I think, which costs three and a half times more than High, or not High, than Medium. So Max mode can do the task worse, and spend five times more money than Medium mode. So imagine, the same model... You might wonder how this happens. How can better reasoning be worse? And that's the whole point: better reasoning can launch extra agents, go the wrong way, start doing things you don't need at the moment. By the way, we went to Napa, of course, so write in, who knows this region, who's heard of it around the world.
Yesterday, as I was driving, I talked with various guys, and Polina tells me that everyone knows what Napa is. Today I asked two more, none of them knew — neither Lyosha nor Vitalik knew what Napa is. And who of you knows this region of the US, and especially who of you has been there? And who has been to California at all? Let's write in, who has been to California or who's planning to — well, those who don't live here, of course. Those who live here, it's also cool to say so, how many people live in California. Again, by the way, it would be cool to do a roll call and see how many of us are in America.
I think about eight percent of our viewers — eight, nine, ten, seven, depending on the video — watch from the US. Clearly the Russian-speaking space worldwide is a decent volume, right? Out of, for example, a million views, it turns out a hundred thousand are from the US. And this thing I'm telling you now about modes and models — everyone who wants to really figure out artificial intelligence, it's very important for you at least to know about it. Just to know about it. And it's not even a matter of experiments, because many experiments may not show you this, but it's important for you to remember it.
If you don't burn through tokens, then I recommend, of course, using the maximum modes, because in most cases the maximum mode will give you a more impressive result. Look, why is it important to know about the Fable model? The thing is, when building various systems and various business processes — and you will be building business processes, every ordinary person has to build business processes. Even my wife has to build business processes, right? Because soon it will be less and less clear whom to trust to carry out a given task.
That is, for now you can still buy something without manipulation. For now you can still see something without manipulation.
We are going to move into a time when manipulation becomes far more serious than it is now. It's a very serious underestimation of the world and of different people in general. You can't compare the time when I was growing up, say, thirty years ago, with how my children are growing up. They're growing up in a time of super-manipulation, simply super-manipulation. Today there's an incredible number of channels of influence on a person. Very many channels, yes. And we are entering an era when artificial intelligence will truly know who you are. Everything before that is kindergarten.
I started saying this about fifteen years ago, and I deeply disliked the market of deception in marketing in general. And Russian-language marketing, American marketing — it's a deception not of marketing for individuals, but of the professional marketing market. Because marketers kept saying: "We know our clients, we've studied them, we've analyzed them." But that doesn't compare with what it means to know your client, understanding them the way artificial intelligence does now, right? Let's go back to Fable, to Fable's uniqueness.
There are certain tasks you should let Fable try. And you can give Fable the task of drafting a certain kind of contract you need. It doesn't matter whether it's an inheritance agreement, a will or a divorce agreement, or a contract for the sale of a house or an apartment, or a car. So in certain places Fable can do something super-high-quality. It's like connecting a unique brain to this task. Basically, for example, if the task is everyday or more or less small, or different answer options would suit you, there's probably no need to bring in Fable.
But if you need a maximally verified result with maximum final effectiveness, then of course Fable is very interesting here. Very interesting, right? Especially, once again, when setting up some process, to find some error in that process. For example, what you didn't take into account in your own schedule, or in keeping track of your own finances, or to see some subtleties, some nuances, some additional patterns.
Fable can be interesting in analyzing advertising, in analyzing sales, it'll be interesting in company strategy. Because when you build a company strategy, there are always many different options in company strategy, right? And you'll say that there are also many different options in marketing. But the thing is that marketing, in this case, what's interesting for marketing is that it will calculate the final effectiveness, so as to calculate all the parameters, to calculate the financial accounting, right, of course, for real.
Look at where you are in life in terms of finances in general. Your life in finances — what is it like? By the way, I made a video, take a look, last Friday — a very interesting one on organizing personal finances, on organizing business finances, right? I think it's an important video. And there was another very useful video, by the way. It's the video about the evolution of ChatGPT's systems, about the difference between systems like Codex and an ordinary chat. I also strongly recommend you watch that video, right. And it's exactly in organizing this area that Fable is very interesting.
If you have tokens and you're on a good plan, then for organizing this finance task I recommend using Fable. If you can't pay two hundred dollars every month, for example, for Anthropic, you can buy it once for two hundred dollars, try this system and try doing something with it. But I'll tell you right away once again: first, Fable is limited on many plans, it has fifty percent of the plan. On the two-hundred-dollar plan there's no such restriction. Second, Fable eats tokens very heavily, right, and if you're on a twenty-dollar plan there's no point using it.
But it makes sense to understand that a very cool model exists. You're unlikely to be able to use Fable via the API. Some will, but be prepared to allocate tens of thousands of dollars to it. That is, you can push, for example, an analysis of your marketing ad campaigns through the API. But if you're spending millions of dollars, then that's effective. That is, of course, it's effective to connect systems like Fable to analyze such ad campaigns, right? So that can give a really substantial plus.
Basically, if you use some systems for analysis — once again, of marketing, sales or somewhere else — they will never use the Fable model, because the input and output tokens there cost astronomical money. So it's a model for a subscription or for people who are able to allocate, say, a million dollars to solving some task and so on.
By the way, more than a year ago Sam Altman described a very interesting case. I think this case is very viable. They haven't got to it yet, maybe — and maybe they don't even want to get to it, right? He said a time will come when you can tell yourself: "I have twenty thousand dollars to solve this task." And you give artificial intelligence twenty thousand dollars, and it solves the task. Why haven't they got there yet? Because right now you spend after the fact. That is, you can spend twenty thousand dollars, and it won't even start on the task, or it'll go the wrong way. Or maybe you could build a thousand.
So the system itself doesn't know in advance how to calculate anything for these twenty thousand dollars. That is, today artificial intelligence can't plan these costs of its own, right — the planning of these costs, well, it will be unreal. By the way, where could such a case be used? Imagine you want to start your own business. And people very often come to me about starting a business.
I say: "You don't need to start a business, it'll be a problem for you." I say: "You can pay me ten thousand dollars. I've just given you good advice, because you'll save a hundred." It's just that people in the world aren't used to paying like that, but on the whole it's the right fee. If you're launching a business in which you invest a hundred or two hundred thousand dollars, and your probability of losing is about eighty percent from the start — that you'll lose, and it usually is like that — then paying, say, five thousand dollars to artificial intelligence to work through your task is the right move.
People don't do that because they think in terms of expenses. Yes, they'd rather take five thousand dollars and go on vacation, rent a hotel, buy themselves a bag, whatever, right, make some payment. No, it doesn't matter, they'll buy a sofa, no, it doesn't matter, right. They won't think about how they could invest in a task of this type — for example, working out the right financial accounting strategy for their home, right, or a financial accounting strategy in business, or figuring out a marketing management system. Figuring it out, really allocating money to it.
And you often meet people and ask: "Who runs your advertising?" They say: "We run it ourselves." You say: "But you don't understand advertising." "Well, we can't hire a person right now." Well, then at least pay two or three hundred dollars for ad analysis to such artificial intelligence systems. And the person says: "I can't pay three hundred dollars, my whole ad budget is fifteen hundred." So pay three hundred and spend twelve hundred on ads. But, alas, people don't think that way.
By the way, for analyzing ads worth twelve hundred dollars you don't even need to pay three hundred — that's roughly ten percent of the two-hundred-dollar plan you have.
So Fable lets you solve very interesting tasks, a bit differently. I, for example, right now have a lot of tasks and a lot of spending, well, I have to make a lot of decisions every day, hundreds of constant decisions. And of course I started choosing where to use Fable, because at some point Fable started eating up my two-hundred-dollar subscription in four hours, or five hours, eating all the tokens completely. And then you start buying more, and buying more becomes very inefficient, because with a subscription you're kind of ready to eat tokens inefficiently, but when buying you want to eat them very efficiently.
You want to know that you spent a hundred dollars and they needed to be spent, or that you spent a thousand dollars and they needed to be spent. Right now an automatic task of yours can eat a thousand dollars, when in fact it didn't need to be done at all. It was a pointless thing. It just kept launching random unneeded agents, for example, right, or kept doing what wasn't actually required at all and was, for example, completely unimportant, completely uninteresting. Speaking of the case — I started by saying that yesterday I was talking with my friend, and he said that he understands artificial intelligence pretty well.
I told him that I understand artificial intelligence at zero. And so do most people who even just use artificial intelligence, type in some queries, even do something in Codex and so on. By the way, I asked what he uses. He says: "I have a ChatGPT subscription, I use DeepSeek." And I think he said he has Claude Code. When I heard these systems, I asked him: "How much is your most expensive subscription?" He said: "Twenty dollars." That's it, zero right away, yeah. And he's a businessman, with monthly revenue in the millions of dollars.
But what twenty dollars? I mean, you can't have a subscription like that, right, for twenty dollars. And then I also heard DeepSeek, and I got bored right away altogether. And an important point, and it applies at least where this can be used. An important point: you need to realize that today the level of knowledge and understanding of artificial intelligence differs radically. Well, how do you measure this understanding? For example, there are certain parameters of understanding for a surgeon, when assessing a surgeon. You say: "How many people have you operated on?
How many hours of operations have you had?" Or: "What was your longest operation in hours?" Or: "Which universities in the world did you study at?" Or: "Whom did you assist or work with?" Or: "What type of operations do you do?" And so on. There's some kind of assessment for an airline pilot, right? How many hours has he flown? Although that doesn't say anything either, right, especially in today's reality, where the plane takes off and lands by itself, but still something matters.
Or, for example, in which countries of the world the person has flown, what difficult stories they've had, what types of aircraft they've flown and so on, right? So... But in artificial intelligence today, assessing how a person perceives some artificial intelligence for life is very hard to grasp. And by the way, I want to say that all the people who create artificial intelligence today, top people like Sam Altman or Dario Amodei, will understand artificial intelligence very strategically, visionarily.
But when it comes to new, modern, various use cases, many of them will find it hard to understand somewhere. Or they have to be fanatics who develop some systems themselves every day, create some cases. And from these people's interviews you can see whether they're fanatics or not. And most, many of them, or most of them, are not fanatics of such systems. And you can see the same in developers of artificial intelligence systems. They often build some very cool, high-tech things, but then they don't apply them anywhere much.
It's like inventing some high-tech material for building the wall of a house, but not understanding what houses exist today, for which people, in which countries, what circumstances there are, how it affects prices and so on, right? So. Although we're used to listening to or trusting people exactly like that. Especially if a person comes out and talks about artificial intelligence, and he has a hundred billion dollars, then everyone thinks he understands it and knows it.
So your task as of today is to figure out at what level of development and perception of artificial intelligence you yourself are. That is, if you want to dive into it, if you're interested in your own professional development, if you're interested in, well, developing your business, if you're interested in the future in terms of your education and technology, then you should look at where you are now and constantly check against it. Having worked as a programmer for forty years doesn't mean you understand artificial intelligence. I recently read a phrase — it's hard to see that from one programmer.
He wrote: "I've been working as a programmer for, like, twenty years, but I don't want to and won't use artificial intelligence, because it harms the technological world," right? And that's it. So the person doesn't understand and doesn't know artificial intelligence. Please write what level you see in yourself in your perception of artificial intelligence, for real, right, and in your knowledge, your real understanding of artificial intelligence. I'll be interested to read what scores there are here. I won't share any of my preliminary conclusions, but I'm curious what description you'll give next to this.
And see you in our new episodes. Bye, everyone.