We're moving pretty quickly toward a situation where, on one side, artificial intelligence will be working, and on the other side, artificial intelligence will be working too. Shopping, advertising, some kind of support, negotiations, all of this is starting to change right now. And is the system becoming more efficient? Or are we gradually building a structure in which people stop interacting with each other normally altogether. In parallel, ChatGPT's voice got a very big update, a new way of working came out in Claude.
And in general, this week once again a huge number of things appeared that change how we're going to use artificial intelligence. Today we'll break down what in all of this is really important. Hi everyone! We're on the ToTheMoon channel, technology news and insights from Silicon Valley, around the world. Actually, just a few days ago I literally did an episode about Opus 5.5, and its main distinctive story is that in all the announcements it costs less than the previous version of Opus, and works no worse than Fable.
But, at least, Anthropic personally sent a letter today, and in that letter they claim this very ardently. And also there came out, watch my episode. Overall I devoted it entirely to breaking down how to navigate these models and what's going on with the new models. But Opus 5.5, we're definitely reminding you, and for those who haven't watched, well, it's a serious topic. I think maybe we'll touch on it today, we'll see what happens. Then once again the limits at Anthropic were raised very significantly. The second story there is, in terms of releases, first of all, the new Voice came out. We'll talk about it now in the GPT part.
An absolutely fresh update of the new voice, and the presence there, there's GPT-6 Astra. New models came out for GPT-6, well, not even new models, probably, new models, yes, there's Astra, Sol and Luna came out. Literally after filming my last episode, I'd only just shot it, so, and they came pouring in. A huge number of updates, in short. So, updates are coming, endless updates are coming. We'll discuss them now, today. And what topic do I actually want to start with? With two stories. So, there's this Nikita Bier, formerly of Grok and xAI, as far as I understand he doesn't work for Elon Musk now, although he periodically puts up interesting posts in the style of: Elon Musk didn't pay the money, I gave it to him, and shows that he ordered a Roadster, a Tesla, yeah.
He put up this interesting post on the topic that Meta traded its users for artificial intelligence, for the hype around artificial intelligence. We'll discuss it now from the point of view of marketplaces. And the second story is very instructive, also quite fresh. OpenAI is testing ad conversations with agents. What is that?
It's when you click on an ad inside ChatGPT, ads are shown to free users. I don't know whether it's shown on Plus or not. Not sure. Our editors will highlight for us now whether ads are shown on Plus or not. Tanya, do you see ads in ChatGPT or not?
I'm only being offered to take part in tests, but I don't see any ads.
Their ads are running, running, gaining momentum, it's, well, there's some decent amount of money going through this. But for now this is only the beginning. We'll definitely have separate episodes on this topic. But what's the point? About the format. That is, you click on an ad, and in this ad you're not just shown a product, for example, and the option to go to their website and leave a request and so on, right? OpenAI is testing the following scenario: you click, so you clicked on this ad, and you start a separate conversation with the advertiser's representative.
But it's an AI representative of the advertiser, yeah, an AI agent of sorts, and it, so, answers your questions about what kind of company this is, what products it has, what services, and so on. And, accordingly, they already have this format, it's called sponsored agents, Sponsored Agents. And I just wrote a letter to a company that does management in the city, well, whatever, servicing houses nearby here. And they replied to my letter that they have a huge number of requests, and they recommend I turn to their artificial intelligence agent. But personally I don't react to such things at all anymore.
It's not that I find them absolutely repulsive, they just don't interest me at all. I just don't understand whether advertising will work or not from the point of view of an AI agent. Because when an AI agent writes to you right to your face, you somehow, it's unclear, probably, placing an order, writing some question, that's okay, but we'll discuss this topic now. What I just want to do is, in parallel with this topic, launch a discussion today, I think it's very interesting. About what Nikita Bier said.
So, he put it quite harshly regarding the fact that there's a department that developed Muse at Facebook. Let me remind you that at Meta they released a new model, which is supposedly super cool, agentic, as they announced it. It's not just, so, a model that answers questions, it handles your tasks. It seems to me everyone has been using this slogan for the last year and a half or two years, and they keep repeating, using it as if it were something unique. And supposedly two clans clashed at Facebook. So, the clan that makes AI, it's clearly big and has power.
They, by the way, aren't much liked at Facebook, because these people also somehow earn fabulous amounts of money, yeah. Because those rumors about a billion and a half dollars in earnings and so on, they seem to have, well, real truth to them. And on the other side, they clashed with the people who make the marketplace. And there's Facebook Marketplace, where people list goods. It's an active system in America, a very strong system in America, possibly in other countries too.
And maybe we'll highlight now which countries it's available in, but in the US it's very widespread, where you list, you have the opportunity to sell some item, right? And I don't know what the Russian analog of this system is. It's Avito, probably, right? Avito. And, probably, Avito, yeah. And supposedly they proposed the following thing: to launch Muse to haggle on behalf of the user. That is, essentially, the user comes and says: "Query, six, like, different listings, or ten, or even fifty." There are limits there and so on.
But still, he says: "Ask around on the listings, I want to buy, like, well, so, this kind of table, or I want to buy some, I don't know, equipment or this kind of camera. Find me a listing of this type and haggle for me." And what did Nikita Bier write? That such a system will lead to only one thing — to prices going down, well, that is, the system will haggle, sellers will be unhappy, and the system will collapse. That is, the trading mechanism will collapse from within, because the mechanism, among other things, is built on the fact that, it's a big psychological question, that many people don't want to haggle, are afraid to write it, can't write to fifty people at the same time.
And on one side, it's an artificial intelligence agent writing here, but on the other side people have to answer, people have to haggle, people have to offer the person various details. And supposedly Zuckerberg let this through. And the headline was precisely about how Zuckerberg, what's it called, sold people out, right, and so destroyed a whole business because of it. And it seems to me, this system, these kinds of cases, from the very beginning they seem sort of cool, harmless, and the whole world is promoting them, by the way, yeah.
About how a huge amount of this kind of super-cool automation is going to appear. But on the other hand, they really can destroy lives, in particular, this type of advertising.
That is, if the system launches this type of advertising, what will it lead to? What do you think about that?
Nothing's going to happen, Sasha. There won't be a situation where there are bots on one side and people on the other. Bots will appear on the other side too. Or they'll roll this system back. I mean, imagine that if, for example, bots are constantly writing to you and offering to buy your table, like, twenty, thirty dollars cheaper, you'll definitely have the same kind of bot on your side that will say: "No, get lost. The price is listed, if you need it — buy it." Or auto-replies will be set up.
There won't be a situation where people are constantly talking to chatbots. It will work from both sides. So I don't think Nikita is really right here, it's more that he kind of stirred up a bit of hype in that direction. So. In reality, well, this system will be unviable, I agree with that. But that it will fall apart — no. Either they'll roll it back, or there will simply be bots talking on both sides. So, in general, that's roughly how I see it.
Uh-huh. Well, it's the same thing, yeah. You have…
This reminds me of how very often scammers write to you, as soon as you post some listing on a marketplace, you get, like, ten weird sort of… Well, you can really tell from them that it's not a real person. So, I think it'll be a very similar kind of story. And over time you just ignore these messages. When you've only just started, just registered, maybe you can't tell them apart yet, but then…
Yeah, I just mean that in any case, for Facebook, this is, well, the marketplace generates a large amount of money, and they won't be able to, well, lose it. That is, in any case, it will all come down to something workable, possibly a bit different, but it won't be the case that you have AI on only one side. It'll either be on both sides, or they'll start, well, banning such users. That is, they'll roll this thing back. Or you'll simply have AIs on both sides, and the way they come to terms will be acceptable, there'll be some kind of parity in itself.
That is, say, you set that yes, you can haggle, but only by such-and-such an amount, and actually, the haggling will happen, but it will be built into the price in advance, and, accordingly, as a result the deal will still happen when it suits both sides, just as, in fact, it happens now. It's just that up until the moment the deal is done, well, you'll possibly have several, like, I don't know, chatbot exchanges, but it won't be the case that it only works from one side.
Yes, but here's the thing, if inside Facebook they ban other networks on you as much as possible and turn on Muse by default, then the bulk of people will use Muse, and people who don't understand technology, they won't use it. Well, and actually, the system should play out the way you say, that everything will fade away. There's another, probably, positive effect of sorts there. As of today, if too many listings are posted in the listing system, then, essentially, your listing goes away, and people stop seeing it.
And it's precisely the presence, sort of, of an agent on the other side that wants to find the best offer, it can bring it up, including finding this listing of yours. Unlike a person, he won't be able to scroll through, maybe, fifty pages and study absolutely all the listings, right? But an agent, probably, will be able to study all fifty listings. How this will work is unclear. Especially how many limits there will be, how many possibilities there will be, right? That is, for example, if you've been given the possibility to haggle once — that's one thing, to haggle five times.
How will people haggle? What will they do? Well, and most likely, indeed, this comes back to the point, as Tanya just said, that if you understand that it's an agent writing, you, so to speak, stop interacting with it or think about how to trick it. And this thing becomes lifeless. It's just that, after all, the human story here, it's that it's too chaotic in terms of selling and decision-making. How many times have we seen the situation? You go on Facebook Marketplace, you write, it's for sale for $900. You write: "I want to buy it for 150." They tell you: "Take it."
And sometimes you go in, they write: "Selling for 900." You say: "I want to buy it for 600." They tell you: "We don't haggle." Yeah. That is, this whole thing is very… And someone, you write to him, for 900, for 150, he won't talk to you anymore. He'll think you're some kind of fool.
Yeah, he'll, like, block the chat.
Yeah, and he'll block you, block you and then also write you some comment afterwards.
Yes, and the rating.
That you're a scammer there or whatever. And that is exactly the story of, like, these human constructs. And still, from the point of view of artificial intelligence, especially an agent written inside the system itself, it's unclear how it will work, and it's unclear whether this really helps develop such systems or degrades them. And as we saw earlier, a huge number of systems rushed, you know, to create auto-generation of, I don't know, various images on social networks, on video networks, on text networks and everything.
And then some of the networks start blocking, on the contrary, the application of the use of artificial intelligence, or at least understanding what kind of content it is. Although, of course, fighting this today is pointless, because all the new modern systems, they will bypass captchas. That's a system for blocking and recognizing a human. My kids have Roblox accounts. And so, the kids asked me to change the limits there. So I go in, and I've forgotten the password. And I tell Roblox: "I want to recover the password".
Some new modern captcha appeared there. I'll tell you this: I didn't pass it. And when I tried to pass it two or three times, I told my kids: "Let your Roblox take a nervous rest somewhere, I'm not logging into this system. That is, if you can recover the password for me, please do, I won't be able to do it". Because it was asking me, not that, probably, if I had paid serious attention to this, I would have done it, but I didn't feel like paying that much attention to resolving this issue.
I thought, why did Roblox do this, right? Roblox, obviously, is going to lose users because of this. A huge number of people won't figure out how to solve this quest. It's really a quest. And my little son, who is ten years old, he told me: "Listen, I tried to do it. I couldn't. And I had, how do you say it, that same picture. I tried, I couldn't do it". So they, of course, are going to lose some number of users, but why did they implement it, and why did they implement it? They asked me ten times, there.
And there, basically, some little figures are standing on some pictures, and you have to look, match the pictures and the little figures, right? Ten times. Ten! Not one, not two, but ten. And it seems to me, all of the ten actions also have to be correct. And the story here is that, most likely, the system is being very heavily attacked by bots, and attacked by modern bots, yes, which can bypass a huge number of, pass various tests. And let's be honest. Well, there won't be systems that, at scale, at least inexpensive ones, in general use, captchas, that will be able to, after some amount of time, diagnose a human, right?
And so bots will clearly generate anything and everything, if, wherever they want to generate it. Or, like Ilnar said, well, a person will just write to ChatGPT: "Reply to all these people about what's going on there". And here, so, I think, Nikita Bier posts, so, on X. I think it's him again.
So today we're giving him an ad, but he doesn't actually need an ad. He's too serious a person. He posted this just now, I literally just saw this story. He writes, so: "Hey, computer! Find me every purchase I've made on Amazon. Call the manufacturer's number. Say that you're not satisfied with this purchase and request a refund. If they don't refuse, say that you'll write a review about them. If they don't refuse, then find the CEO's phone number and ask him for a refund. Then repeat this same process for purchases I never made.
And see if we, if they'll be able to send us the money. And if they ask for a receipt for this purchase, then generate it from a message on a social network and send it to them. Don't stop until you've earned a million dollars". So, you know, this is absolutely fresh. It was there, like, roughly, today. And I want to say that this is, of course, madness, but what he has written as of today, this is, yes, well, what is really going to be happening in the world very seriously, right?
And this, of course, is a big problem in terms of the work of various service departments and the interaction of these service departments. And what's going to happen with this? Of course, this is, well, a big particularity, especially when AI agents will be hooked up everywhere. These AI agents, en masse, will be responding.
Reminds me of a movie, in Russian it's called, I think, "Desperate Con Women". I don't remember the original title, but that's exactly the point. A woman, for the first time, writes that she wasn't satisfied with a product, gets a coupon for a free product. Well, and from there the whole story spins up. In the end she ends up in prison, but one way or another, right? That is, this scheme in America, most likely, will even work for some time. Such reviews there can, most likely, bring some benefit, but the story will fizzle out pretty quickly. Rather, Nikita has already squeezed the maximum out of this story.
That is, he made an interesting hype news item, but in reality not that many people will be able to abuse this, right? Because, well, on the other side there are also corporations, where the money is counted. So the maximum out of this, I think, Nikita has already squeezed.
Well, let's put it this way, honestly, Ilnar, you, on the one hand, you, on the one hand, are right that the money is counted, but on the other hand, in fact this is a crazy manipulation, and especially among companies that aren't corporations. And these companies will have to sort all of this out. That is, you're now talking about some corporate story, mass-scale, big, right? But there's also the non-corporate story, in terms of the super-mass-scale big one, there's a large number of ordinary sellers. And for these sellers the system works and is set up in its own way.
Or, and before, for example, if a seller one hundred percent refunded the money without any problems if a person complained, then now, in theory, they'll start blocking everyone, and if they start blocking everyone, then these people will lose the quality of service. There was just a very big story. At Lufthansa, a person wrote a complaint, and he wrote a post of the following nature. He sent them a complaint, they didn't reply, he sent them another one, they replied to him, didn't reply. And he wrote them a complaint on X.
And Lufthansa replied to him, like, why are you also writing to us on X? We received your message to customer support and by email, and we already answered you there that we're not going to resolve your issue. And do you really think X is going to help you? And ten million people saw that reply on the social network. Lufthansa apologized, wrote, so, sorry, we really want to, so, engage together with customers, and we're against stories like this. But, first of all, it's unclear, again, whether a human wrote it or a bot wrote it, whether they'll resolve the human's issue.
But this is precisely a telling story, right, that switching to a standardized service, or what we said from the very beginning, when, Ilnar, you say, when, on one side, there's a bot and a human, the system collapses; when there are bots on both sides, then the system starts to degrade, it turns into a non-working construction. For now we see this in the example of those agents that can't even negotiate with each other at scale, well, or a human is being looked for altogether. And where are the people? Everyone's looking for a benefit, both the seller and the buyer, right?
And if everyone is only going to buy, to get a benefit, then what will happen? Amazon, for example, on its site doesn't give you the ability to filter products by best price. It doesn't have that function. Amazon, for example, doesn't even have a function to sort products from cheap to expensive. It's amazing. It has the ability to search within a certain price range, but it doesn't have complex filters. Why are there no complex filters? Then you'd stop buying certain products.
That is, if Amazon, for the same product from different manufacturers, from different sellers, shows you only the cheapest one, then the other sellers will die. And, of course, how will this situation live on in terms of the advancement of artificial intelligence? That's the big question. And the topic is very interesting. It's interesting, for our viewers, from the point of view of how you see it. Please describe this story and what your opinion is on this in general?
This is some kind of amazing, in my view, downright fraudulent scheme, especially considering that he asks for a payment document to be generated.
So it turns out there's nothing behind this. Because if someone tries to do this, then behind this…
Of course there's something behind it. Of course there's something behind it, because today you can go into Amazon, go through your purchases at Whole Foods and write that some product of yours was rotten, and it will automatically refund you the money, and for certain items. So there is in fact something behind it. So he's writing this based on fact, even in terms of the laws that are in effect and how people work. If you start complaining to people often, you write to Uber, my food wasn't delivered. Here's a very simple example in the world of how Uber works. They bring you food and bring it to the neighboring entrance.
You say my food wasn't delivered, they write to you: "Sorry, everything is fine." You write to them again: "Call a manager. You didn't deliver my food, refund the money." It writes to you: "Everything is fine." You write to it four times: "Call a manager." It calls a manager. The person says: "Oh, sorry," refunds your money. Yes. If they don't refund your money, you can go in and say: "Well, then I'll go complain," and it refunds your money. It's a scheme, and it works. And he's precisely promoting and showing this thing here from the point of view of reality. It's clear that, Tanya, making a million dollars is fraud. First of all, the very fact of it.
Sasha, even with Uber, say, you do this a few times, then they'll throw you on a blacklist.
Well, our viewers are going to ask us now.
I don't know whether anyone has done such refunds very many times.
That's not what we're talking about.
I myself am curious whether they're going to block at some point.
Like, this one, it's not going to work. Well, obviously, a tweet like that won't work. But the question here is that people today use bot systems everywhere it's possible and where it isn't allowed, and something works. It's clear that somewhere a few times, yes. Exactly the same way people get deceived and money gets collected from them. Let's get back to Voice. So, a new voice assistant has come out for ChatGPT.
A very fresh topic, today's. Just now, at least, as of the moment this episode launches. And let me remind you that ChatGPT has several voice modes. It has a regular, standard voice mode — that's the voice mode where you send it a message, and it then converts it and answers you. There's also a separate mode, by the way, that lets it look at pictures and video that you shoot, right, or pictures of your computer's desktop. And there's a mode, they call it GPT Live, and this mode appeared in the summer. It's a mode that talks with you in real time.
And today they released a big update, they hooked up, inside of it. Essentially, it doesn't just talk, it's learned to connect various services. And to date they've been developing quite iteratively, by the way, seriously. They had a whole series of updates, and they added it. First of all, one of the big updates that there was, they added it to Codex and to Work. Then they gave the ability to do various file uploads in projects there and then started getting updates. On September ninth, that's a few weeks ago, they enabled the ability to use 5.6 and GPT-6 Astra for searching complex questions.
And now they've got these plugins in Live. And they're on the web, and in the mobile apps, and in Work on the web, in the mobile apps. And plus they announced limits. So this voice, it has limits. Those who are on the $200 Pro version, though it seems the $200 Pro version, by the way, hasn't appeared yet, as far as I understand. Some people, by the way, wrote to us that it had appeared, but it hasn't appeared. There it's unlimited in the $100 Pro version, there you can talk fifteen hours a day.
Well, in the other versions, so, it's all shorter. In short, it's there. But what's interesting is that these plugins that they… In Plus, by the way, three hours a day is allowed, because someone's going to ask. In Plus you can do three hours a day. And I wonder, does anyone among our viewers talk three hours a day or not? There's also still a Go version there. Oh, by the way, I was asking about ads. I remembered. Ads are shown in Free and in Go. Yes, that's the worldwide version of Go, I think it's ten dollars or whatever it is, give or take, there they also have three hours.
Does anyone among our viewers talk with GPT Voice Live for three hours at a time? Interesting topic.
Do you talk to it at all, Sasha, even a little?
Listen, I don't talk to it, I talk to it a little. That's how I'd put it. My last experience was when I was driving to France, and there we were on the road for two hours or however long, I tapped on some question. I was asking ChatGPT, and a friend was sitting with me in the car. That was the only thing that interested him during the whole week of the trip with me. Even though I was doing complex things, often discussing complex things on the phone, telling him about some points. He said: "How do I get a thing like that? I'm ready to buy it."
I mean, for that he was ready to get a ChatGPT subscription, for nothing else was he ready to get a subscription. And when he heard how it talked, he was so surprised, he couldn't believe it talks like that.
We underestimate Voice.
That's how it is live, yes, that's how it is live. And, well, listen, look, actually Voice, as far as I understand, neither Tanya, nor you, nor Ilnar, you don't use it, right?
It annoys me. I don't know how fifteen hours, well, one can… I can't even imagine that person. I wonder, does he exist? Let him speak.
That's probably a professional, Tanya, use case. Tanya, fifteen hours — that probably has to be a professional use case. I mean, if a person…
Fifteen, wait, that comes out to, minus sleep, all the rest of the time. They're counting on you talking, right?
But in theory that would be interesting. And I had, it was used for some number of hours. If it integrated easily into various devices in the home and worked on them well enough, like on the phone, perhaps that would have a certain effect. Still, as of today there are no great smart home devices. Ilnar, you don't use it, right, constantly?
Yes, I don't use it. Just now I got curious. Well, I'm in China right now, and to communicate with the Chinese I use Google Translate. When you speak or write there, you show them from one side, they speak or write from the other side. And now I got curious, why don't I try using ChatGPT as a simultaneous or near-simultaneous interpreter? And you, when you were in Europe…
I did use it, by the way, it was great.
Yes, yes, yes. So there is such experience and the experience is positive, right? So for me…
In China, as it happens.
Cool, I'll have to try it.
I don't know, it was, it was how long ago? Three months ago already, but it was really very great. It helped a lot. Even though there was always an interpreter with me, sometimes his English still wasn't perfect, and ChatGPT helped really a lot.
Listen, awesome. I'll try it today then. And then, Sasha, answering the question about Voice.
If you can, try it. Google has Gemini Live, Google has Gemini Live. I just don't know whether you have a Pro subscription there and what services you have there, but in Google… And what's more, this Live, it has extended thinking. That's specifically with deep reasoning. And it talks quite decently, reasons, performs actions, commenting on the progress of the work in parallel, yes. Here the question is that, as far as I understand, ChatGPT Voice in terms of quality is supposedly better in quality.
That is, it doesn't lose the conversation there while performing tasks, they state this, that it makes pauses.
But what I didn't like in Voice, I'll be honest, is that they say that it, sort of, while performing parallel tasks, there is, look, a cool thing now. Now you can in ChatGPT Voice, before you couldn't do this. This appeared a few weeks ago. Choose the version that also works for search for you. That is, essentially, you have a voice assistant, and besides the voice assistant you can choose, for example, GPT-6 Pro. But since GPT-6 Pro works and searches for a long time, it turns out this system has to, sort of, understand this context really well.
And plus you essentially have just this one chat. As far as I understand, it can't create parallel chats for you simultaneously. That is, say, launching GPT-6 Pro in different sessions as one thing, and in parallel a quick search as another thing, in parallel. And it itself is also doing something there without search. And I don't like how it holds the context in this matter. That is, it gets confused, you have to explain something to it all over again. That is, there's no certain stability. Claude also has Voice, and it works with both Opus and Sonnet, but there the rhythm is more sequential. That is, it listens, thinks, answers.
And it's interesting, Ilnar, you can actually test, I think, all three, right? While you're in China. Because the idea you proposed, it's great, because what's the difference from Translate, you can always clarify with the help of the chat, right? You can formulate a question for it and ask it to translate something of a completely different nature and set it up, and say that if the person speaks Chinese, then you translate to me, in my direction. I mean, in theory, that could be done in a pretty cool way.
And to count not tokens there, but to count some kind of quality, the result. But it seems to me that all these systems are still simply as far away as the Moon, right? There's not even a whiff of it anywhere near, of a system genuinely looking at its own work or evaluating how it did something before. There was another interesting problem too.
Viewers, write in about what problems you've had. I had a problem when, for example, the limit in the system would run out, and Fable, I didn't want to keep buying them, and I'd switch, for example, to Opus 5. I'd switch to Opus 5. Opus 5, in half a day, would pile up, you know, such nonsense that later, when I, when I noticed it, like, after half a day, I'd stop all the work and go to Fable. I tell Fable: "Opus piled up nonsense here, roll everything back and continue what you were doing."
And that was really, I don't know about others, but I've already had a whole series of cases like that, where changing the model led to a catastrophic result in what was, you'd think, an already described story, yes. And in general, I think people have a lot of cases like that. We're now, probably, talking about developing some kind of systems. But, Tanya, you don't build systems, you see this in your regular chats and questions, right? Like, in preparing, maybe, some reports, or, like, in some research, or some actions.
That is, the systems periodically work very, very strangely. Well, and the race is on. Another interesting topic came out, by the way, let me quickly tell you about it too, because it's unclear whether we'll get to discuss it in other episodes.
An interesting update came out in Claude Code — new Projects. That is, there's the option to set one overall goal for a project and create several parallel tasks for it. That is, this happens in the new version of Projects, and you can launch this parallel work. Essentially, a project brings together repositories, environments, various connections inside, and individual tasks are executed in their own separate branches. Look, the idea itself is interesting overall in its essence, but there are, of course, well, I mean, in general, the fact that there's a managed division of work — that's a great thing.
This is what we were, essentially, just discussing with voice. If I could talk by voice with a system that would manage a large number of chats, that would be very interesting. It's like my statement that I'd be willing to pay several thousand dollars a month for a system that would constantly analyze my work and my chats. And would somehow report to me about them during the day. And essentially, how could it report to me about them? Well, for example, by voice, or send me some notification, or send me, maybe, some short little message in messengers and so on.
And this thing that came out in Claude, by the way, isn't available to everyone. They're rolling it out now in the Pro and Max versions, and rolling it out in stages, at that. They're rolling it out first, like, to people who didn't have existing projects in the web or desktop apps. It's interesting, for those who've already run tests in this system, how useful it is from the standpoint of developing various systems. But right now I wouldn't be talking about development. You know, I like to recommend using Codex and Claude Code not only for development, in particular, for your ordinary everyday life, right?
By the way, this won't affect the old projects that you had running. That is, in the old projects the logic won't change, but the topic itself is becoming interesting overall. And it would be cool if ChatGPT released something like this for chat. I'd be interested in testing it for a project, where you could somehow set a goal. Although I don't quite understand how to manage that, because managing linearly, setting a goal for one project, what the memory would be there, how that memory would change, isn't very clear.
Well, I mean, it's just not very clear how that's going to be done. We'll see what happens with it. Especially considering that Anthropic itself is declaring in parallel, it says it has tens of thousands of agents building code there, Claude Code, and creating systems. And at the same time it tells us that a hundred agents can't come to an agreement among themselves. But it's interesting how this will happen, the work across different tasks, and how much it really can be, well, serious and high-quality.
We'll see the updates that, at least, OpenAI will announce in the near future, that's literally next week already, I think. So it turns out our next episode will come out, at least, after I attend OpenAI Dev Day. By the way, interestingly, Sasha Mashrabov from Higgsfield will be speaking there. There aren't many stages, but Sasha Mashrabov will be on one of the stages, yes. For me there's always one interesting thing there — I understand that they'll be talking about fairly basic, simple things there.
It's more of a get-together. But what's always interesting is what additional answers the owners and executives of the company will give during the Q&A, yes, at the very end. And whether they'll present something, well, maybe, specific, or something serious, interesting, or give some cool cases in terms of how GPT-6 Astra works. We'll see if that actually happens, because for the ordinary user, I think, so much stuff has come out that it's unclear how to test all of it, right, and unclear how to check all of it.
And I said in our last episode, our editorial team ran a poll inside YouTube, and there more than 30% of people, at least of those who watch us, they haven't even tried Astra in ChatGPT, right. And, of course, figuring out this whole zoo is becoming not just harder, even harder with all these voices, when you're told there are three different voice models: Advanced, Live, and there's also the one that scans video. And in this system you can do this, and in that one — that, and in this one you can connect chats. I mean, you sit there thinking: "All of you, disappear, so, can I just open one little chat and switch some version?"
And the main part of the world hasn't even bought the $20 or $10 version yet. We'll see you in exactly a week in our episodes. And look, we also have various episodes coming out almost every other day. Don't forget to support our channel with likes and, of course, we're expecting a whole lot of comments from me. Bye, everyone.