This week I was in San Francisco at OpenAI DevDay. It's OpenAI's main conference. At the closed sessions. Sam Altman was there, and lots of their employees, the people who actually build the system. OpenAI showed the new ChatGPT Pro for $500, showed GPT-6.1, showed Dots, the Ultra Fast mode, Codex Cloud, Spaces and a whole bunch of other updates. Some of them, of course, we've already managed to test. Today we won't just retell the announcements, we'll figure out which of the things shown at this DevDay really change how we work with artificial intelligence.
Which of it can truly change our lives? What will we have to pay more for? Which new features are really worth using, and which announcements aren't worth spending any time on yet? Let's go. Hi everyone! We're on the ToTheMoon channel: tech news, insights from Silicon Valley. It's evening in Silicon Valley as we record today. And for Ilnar in China, since Ilnar is in China right now, for Ilnar in China it's morning. Today our topic is really very clear, a precise topic, right?
All the updates that have happened at OpenAI related to Codex, to ChatGPT and to the company in general over, you could say, over this year, probably, right? Given that DevDay 2026 has taken place at OpenAI, it's their main conference, a closed conference, it's positioned as a conference for developers, although of course it's more their conference in general, right, a conference of sort of…
Of announcements, among other things.
Yes, their announcements, because the announcements there, probably the main announcements, that's their main session. It's not that easy to get in, right, you can't even buy tickets for it separately. Yes, it's registration, sometimes a lottery, sometimes a separate invitation. So I went there and even brought back some merch. And as Ilnar put it.
Like you've had cupping done on your back.
Like I've had cupping done, yes, that's exactly what Dots is. By the way, I brought back an absolutely unique thing from there, and it's being sold. OpenAI handed out this Chromatic, and OpenAI handed them out. It's a special Chromatic made for Codex. And of course you can play Game Boy cartridges on it. And I want to ask my mom. I got my Game Boy back in '91, '92, and I have a Mario cartridge lying around at home, right? So she can bring it to me from Minsk. But here there's a special cartridge, and the whole styling is Codex-themed. And they even made their own framework so you can load games straight from Codex without any problems.
Actually, the idea is cool, right? Ever since GPT-4 they've constantly been showing that you can vibe-code some game, but before, you had to play it either in a browser or inside some sort of…
Yes! The idea is really cool, yes. And I did it yesterday. I did it yesterday, gave everyone a power, handed out superpowers. My younger son — screaming, my wife — a robot vacuum. Basically, I gave everyone a superpower. We joked and laughed about it, but it worked. And really, you can make a game, and the games that get made are pretty cool. I don't play games. In general, I'm not interested in that at all. But…
You've got two game testers right at home for that.
Yes. Well, yeah, with limited usage. So, what I want to say is that my visit to DevDay was an interesting one. On the one hand, it's kind of cool to get right into this community directly. There are, of course, lots of people from OpenAI there, employees. Sam Altman was there, of course. On the one hand, lots of different partner companies. On the other hand, lots of different sessions were held, closed, open, in different formats. And there were announcements like that. Well, the announcements were made right from the start, large-scale announcements overall, about why OpenAI is updating.
On the other hand, of course, for me the very, the very quality of processes at OpenAI as a company is a big question mark. And you and I both love to nitpick about how OpenAI comes up with something, releases it, and then it disappears. And, probably, the quality of how the conference itself was organized. You can see that this company, this company isn't AI. Well, I mean in the sense of AI first, as they say, right? Or infrastructure built on AI, that's very clearly visible. For example, two hours before the event I got, in honor of DevDay, thirty free days to try the $100 ChatGPT Pro.
Right? I mean, things like that. And my sister said something funny: "Well, for a company, $100 is, you understand, they have different processes. $100, are you serious?" I said: "For them it's nothing, not serious at all." And the very next morning they sent me $2,500 in credits, right? And so, simply as a gift for having been at DevDay. They sent it to everyone, or not everyone, I don't know. But that's the gist.
Sasha, did they send it, I mean, to your account, or some kind of activation code for any account?
No, that's the funny thing, they sent it to mine, Ilnar. And of course it looks very peculiar, and it seems like some kind of minor oversight by the corporation, but actually it shows how the corporation works in general. And today we'll go through exactly what they talked about and how it looked. And in today's rundown we have a really serious, well, serious structure of all the things at DevDay. I really want to discuss them today together with Ilnar and tell you about them. Both from me and from Ilnar, right, to tell you about them.
I haven't had a chance to poke around in it yet. I didn't have time.
Yes! Because, for example, it didn't roll out for my sister, who has the Pro version. And… Well, here the question is who's where, what tier users are in, and how it's going to appear, who it's going to appear for. We'll talk about that today too. But what I want to say is, well, first of all, I think that, on the one hand, my dream has come true: plans have started getting more expensive, and a more expensive plan has started to appear.
On the other hand, of course, it's funny, because they released ChatGPT Pro for $500. I upgraded to it right away, right there, sitting on the stage. On the whole, it didn't matter to me what's inside it. There wasn't really any reason to upgrade, right? Because essentially this plan basically gives 25x Plus, while the current $200 Pro gives twenty times more than Plus.
Sasha, everything has actually gotten much worse.
I know.
That's the main disappointment of the whole DevDay.
I understand. Yes, but they said the following…
I'm really very upset.
Look, but they said the following: with ChatGPT Pro for $200, for all the $200 people, they'll keep the current tokens until the end of October — not tokens, but the current volumes there are, the perks. Although I think they've already started disappearing, and in the $500 version it's only a 25x upgrade. I mean, you can see that, of course, as you and I said, ChatGPT Pro in this form will cost more and more. Now it'll probably start costing $500. A cool Ultra Fast mode has appeared there. I'd probably say this mode is very interesting.
They showed this cool thing: a rocket being built with Ultra Fast and without Ultra Fast on Astra, right? And the rocket with Ultra Fast had already taken off, while the rocket without Ultra Fast was still being built. So, I really liked that, and, honestly, I'll tell you, I bought it because of this thing. Well, I'm telling you in all seriousness, because this thing is great for me personally. You can't use it, Ultra Fast, by the way, in ChatGPT; you can use it in Codex and in Work.
And you can only use it in 6 Astra. Now we'll talk about the 6.1 updates, we'll talk about Dots, we'll talk about real-time, all kinds of vision, about the Decisions API. A lot will be covered today. What I heard at the closed sessions, and there are interesting things there, by the way, in terms of some, some aspects, both positive and negative. But still, Ilnar, about Ultra Fast. I like Ultra Fast in itself. What I probably don't like is this part: as an ordinary person, I'll obviously have to pay more for everything.
At the same time, I understand there's no way around it. I mean, this is probably their plan. They brought back the $200 plan. If any other $500 subscribers are here in the chat, let us know. I just don't know who it's available to, because how available it is, for example, outside America, or who can switch to it at all. And you, Ilnar, said you didn't like it. Tell us.
It's very bad. DevDay, in my view, was awful. I mean, on the one hand, a ton of announcements, a ton of everything we're still going to talk about. But there's this opinion that it's OpenAI's funeral. That's too categorical, but they've made things much worse for the overwhelming majority of users. Remember when we were talking about, like, which model to choose, all that. I was saying precisely that besides a model's quality on benchmarks, you also look at other things: availability, by geo in different regions, limits, and a set of other parameters.
So, the limits were practically one of OpenAI's main advantages over, say, Anthropic. I mean, the limits in ChatGPT were much more generous than in Claude.
Let's, let's remind them. Let's remind people that ChatGPT essentially has three internal solutions: ChatGPT, Work and Codex. And ChatGPT doesn't use up the base tokens that belong to Codex. For example, in Claude, the Claude limits, the Claude Design limits, the Claude Code limits are all pooled: if you've used up your limits in Claude Code, then with the chat you essentially, well, we're talking about the top versions now, right? Because the cheap versions are always available, you, well, they're nominally always available, you won't use them. I'm just reminding you, right, so it's clear.
The key word, Sasha, is "was". Because yes, these…
Well, in my case, Ilnar, I'm telling you, I basically paid an extra $300 for 5x.
Yes, yes, yes. And now, well, they brought back the $200 one. Even before DevDay itself began, Tibo wrote that, look, we'll bring back the $200 one for you, but in reality it's the $200 one instead of 20x. What does 20x mean? There are limits on the Plus version, that is, the $20 GPT Plus. Well, everything is counted relative to it. If we take as one unit the limits available on GPT Plus, then, accordingly, the $200 one gave 20x, that is, twenty times more, and the $100 one gave five times more.
And in that sense, buying two $100 ones wasn't equal to one $200 plan. And the $200 plan was a very good deal in terms of the limits it gives. Now it's becoming the case that the $200 one gives 10x, and accordingly the $500 one gives 25x.
We don't know for sure. Yes, we don't fully know yet. They may still change things somewhere. And I also want to say, the only thing is, there are, admittedly, a couple of upsides there. For example, in the Pro version the five-hour limit has been removed, Ilnar, right? In the Pro version you get cool modes added, for example, things like Ultra Code. In the Pro version you get the whole notion of Pro in general, and ChatGPT Pro, for example.
They're all Pro. I mean, Sasha, the $100 one, the $200 and the $500 — they're all Pro versions, just with different limits. Well yes, and Ultra Fast is available on the $500 one, yes.
And it's unclear whether there's a difference. But there is a difference. Look, in the $500 version they removed all the limits on using ChatGPT 6 Astra. I mean, even the $200 version had limits. You couldn't create more per day, more than a certain number of hours. And also, by the way, it's been lifted in Voice. Obviously, there are no limits at all in Voice, but there weren't any in the $200 one either, right? Ilnar, here we are discussing 200 bucks. Most of the audience doesn't have $200.
But this is my pain and my colleagues' pain. I mean, a lot of people from my circle…
Got it.
On the $200 version. And these limits, with the release of Astra, you know, you already had to keep an eye on them, and you're losing them. And overall, yes, you stay within them, but you already had to keep track of them. And now…
Yes, that's it.
So for the same money you get half as much. I mean, the main result of DevDay is that they simply gave you half the limits you had before.
Let's put it this way, let's be honest. When you first bought the $200 version, and I bought it too, there wasn't even Codex back then, right? Back then there was ChatGPT. And let me remind you, Codex basically didn't exist at all, and you and I used ChatGPT for $200, and it seemed fine. Obviously, now that people have gotten used to having Codex, that's a question. What I want to add for you is one very interesting aspect. There was one session there, a small one, and the speakers there, I actually think it was the best session, at least of the ones I attended. And I definitely didn't watch all the sessions.
I don't even know if that session was available online. I want to say it again, once again, for me it was one of the most, at least, useful ones. To hear certain things, although it's relative, right? There was a discussion about the following things. There were architects there, they were telling us, they look after CI there. Well, everything to do with deploying different versions inside OpenAI itself. You could tell they were two, you know, serious in-house programmers with teams. And they said they had calculated how much it costs to use, in general, Astra via the API, the kind that keeps moving forward, well, works on its own, carries out tasks, because inside OpenAI they don't pay money, they've got unlimited there.
And I know, I've often voiced my own opinion. I've said that if a corporation has bought Enterprise so that everyone can work together, because people buy themselves separate paid subscriptions and program individually, but inside companies they have Enterprise, and in these Enterprise plans people have to buy tokens, and on tokens you can't really program all that heavily, really hard-core. And he said: we calculated that one session costs — first of all, by the way, an interesting, interesting thing. Two things that are fundamental.
First: an OpenAI developer said that he has sessions that have been running for more than three months. And I really liked that. That's what, Ilnar, you and I discussed, that sessions have become, that you can sometimes leave them running. They said they run absolutely smoothly, the context gets compressed and so on. And this applies to development, mind you. I mean, a session runs for three months. The second thing they said, they said that Astra 6, it was hard to talk about versions there because the versions kept jumping around, but overall it can work 300 hours without a break, roughly 300 hours.
And this is an interesting story overall. Let me remind you that this appeared, probably, with Fable, right, Ilnar? That systems started doing everything on their own for a long time. Then came 5.6 Sol, which took a long time on everything, but there was no result. And it could usually work for two or three days, and there was no actual result as such. Now things seem to be getting better with Astra, but they said Astra can work for 300 days. An hour of work used to cost, they said, $32, they calculated it.
Now it costs, so, $16 an hour; 6.0 came out and it started costing $16, they said it's supposedly cheaper now. So there you go. And then the guy said: "Well, and obviously we have lots of sessions running in parallel." So, guys, if you multiply $32 an hour by a whole month, and it's running, that's $23,000. And when you have many sessions running in a month, say ten, that means you're spending $230,000 a month on development. So, Ilnar, I think that this whole story with $200 and $500, it's all of course going to lead to cool systems being expensive to build, and people being left with some kind of interface for everyday life or for their own personal affairs.
But even, probably, what, what we have access to today may become insanely expensive. We'll see, right, what actually happens. Our viewers, it would be great. What do you think about… we're going to go through things now, and it would be good if you rated the different things. What do you think in general about the $200, the $500, the $100 thing?
I'd also add something about Ultra Fast and the $500. For $500 we get 25x, so it's not very different from the current two hundred, from the current $200.
We don't know yet, Ilnar, what's going to happen with Plus; maybe they've lowered Plus, and it'll be even less than it was.
That could be too, you can just, this…
It's 25x! It's 25x Plus, Ilnar! Ilnar, it's 25x Plus, right?
So it turns out…
And it's unclear relative to what.
Ultra Fast, budget-wise, cost-wise, is six times more expensive than a regular request. Most likely it'll also eat up limits six times more expensively, six times faster than a regular request. So, accordingly, we get eight times faster for six times the cost. And these 25x in the $500 one, this seemingly big limit that almost matches the current one, will just be gone in an instant. I mean, on the current $200 I don't use the regular Ultra, which takes something like 1.5x or 2x.
But a six-fold Ultra Fast — I don't know, for now that doesn't look very usable, to be honest. So overall, my impression of everything they presented is a bit negative. What I want is: leave it the way it was.
Ilnar, you know what there is… I'll give you an example of a great use of Ultra Fast. You're absolutely right about the cost.
Right now I've got a lot of resets, you know, these gifts, limit resets. So I really have to spend it all and seriously burn through all the most expensive models. What I want to say is, there's a use case where I, for example — here's why I bought it for myself. I often end up in situations where I need an answer very fast. And, for example, I'm using or doing some analytics, and the system goes off into forty minutes of thinking. For me, speeding that up eight times, or ten, or six, roughly — for me that's a really big deal.
So not an hour, but six minutes. Well, I'm speaking roughly, proportionally, because it's all very relative. And the fact that it's not in ChatGPT is a bit of a minus. First of all, this version is only available in 6 Astra. The speed-up isn't available — in 6.1 Sol it isn't available, right? And I think it's available in 6 Astra, or it's available in 6 Sol. Well, anyway, it's not available in 6.1. And all of that bothers me too. They'll probably polish it, finish it, and this Ultra Fast mode will probably be in lots of places.
So for me, Ilnar, it's more like this: if I've got some meeting going on right now, and right now I need to quickly do some analysis or get some data. I can, like, throw it into an expensive model — I don't know, in Work, for example — and ask the expensive model to get me some result. But by the way, to this day none of us knows the difference between ChatGPT 6 Pro and, in Work, 6.0 Astra or 6-point Sol. I mean, nobody knows. Ilnar, do you know what Pro is? Pro — is it Astra, is it Sol, is it a combination, is it something more serious?
Because, for example, Codex 6.0 Astra is different from ChatGPT 6 Pro. I mean, Codex will stop, or it won't produce that kind of mass of information, and so on. Whereas ChatGPT can go off for an hour and a half and give you a result that'll blow your freaking mind, right? Usually, for Codex to produce a result like that, it has to devour a huge number of tokens and sit there for five to ten hours. So, Ilnar, maybe — you don't know the answer to this question either, do you?
For me, it's just that because of the huge amount of all these analytical files, spreadsheets, reports, information — really heavy digesting — I still, of course, throw it into ChatGPT, the big one, yeah. Although I've discovered a lot of new use cases for myself, including with Dots, including with Codex, and so on. But you're absolutely right that, on the one hand, the masses don't need this Ultra Fast mode, but I think, Ilnar, there's, like, a whole horde of people who do need it.
Remember, we talked about how Sam Altman said you'd be able to spend $20,000 on a task, and it would, you know, get solved for you. That kind of functionality doesn't exist right now, but this is somewhere along those lines: you can spend some money and get something solved faster, for example. So that's what they showed: the rocket took off faster. The question is whether that rocket will fly faster, or whether, say, Codex will make a mistake and not find a solution. And we all know about that, right?
Sometimes you run something in Codex and it delivers, and everything works great for you, and sometimes it works for eight hours, comes back to you, and things have gotten even worse, right? And you have to roll it back. That, of course, bothers me, it bothers me. And by the way, I personally stopped using Fast mode in Codex. Viewers, by the way, write in and tell us who uses it, or uses it in Claude. I stopped using it because, well, in Claude — first of all, you couldn't use this mode in the cloud. I don't know if you can now.
I stopped using it because — I think you can't, right, in the cloud, yeah, well, in cloud stuff. I stopped using it because I keep going over my limits, I have to buy extra and so on. And I thought, well, no big deal, let this task take a bit longer to solve. So we'll see whether I use it. I think in a month, two, three, we'll figure it out. But still, of course, I bought the $500 one simply as a geek, an OpenAI fan, and I probably want to get the next pen again as the very first one who once bought Pro.
And also get a paperclip for buying the $500 ChatGPT while sitting on stage. I don't think anyone else did anything like that. Actually, I wasn't sitting at all, I was standing. So, let's move on to the next ones, the next cases — not cases, but what they released in general. And of course I want to move on to the story. But by the way, Ilnar, this topic — about 33 months of running, and 300 hours of Astra, and $23,000 of API — it's interesting, it would be really great. Our viewers who are more on the professional side. Those who work more, do more. Share your experience with complex sessions, single sessions.
How much money you've spent on the API, what your max is in terms of the spending you have right now, or how long your latest sessions ran in, say, Opus 5.5 or in Fable 5.1 or 5.0, or in ChatGPT Sol, 6.0, or Astra, or 6.1, right? Because 6.1 has appeared in America. I don't know if 6.1 is available to everyone now.
Also a very strange thing to me, right? Before we get to Dots, once again. Why do you release 6.0 Sol last week instead of waiting and properly releasing 6.1? And in general, why didn't you just call it all 6.0 instead of messing with people's heads? We hadn't even figured out what Astra is. Then they said Astra is 6. We hadn't figured out what 6 Astra is. Then they told us there's 6 Sol, Terra and Luna. We haven't even had time yet to look at what kind of system this is, how they work. Then they come along and say there's 6.1 Sol, and there's a price comparison table up.
And it says 6.1 Sol costs ten times less than 6.0 Astra. And you're like: maybe it eats more tokens? Maybe it does a poor job? Anyway, at their presentation Sam Altman specifically had that slide up. That, of course, really bothers me. And I think it's very hard to discuss 6.1 Sol as a whole today, because not enough time has passed; these first days everything is bound to be glitchy. It's been twenty-four hours for me — more, it's been twenty-four hours since I got back from DevDay, and today is Wednesday, evening, and even though our episode comes out on Sunday, Dots has already glitched lots of times; in Codex huge new updates kept coming, and some things weren't updating.
For example, Codex Cloud became available. Meaning you can fully use Codex Cloud from your mobile phone. I'll say right away: it's not fully usable. And Sam Altman says from the stage: "Now you can make full use of Codex from your phone." Well, it's the way it was. It got a bit better, sort of, but it's just as much of a pain in the ass, yeah. And a huge number of features aren't available. None of those integrations are there. And honestly, again, this is probably just for geeks, so let's not dwell on it too much.
Well, you made a great point that literally one week passed between 6.0 Sol and 6.1 Sol, and it feels like they wanted to beef up the presentation, this DevDay, so there'd be something to talk about. So they just rolled out the next little model, just so there'd be at least something. Because otherwise it's: we cut your limits. They also showed some other small things. Here's a little model on top for you, to make it all look beefier. That's the impression. And, well, the limits most likely won't be enough for Astra. So they said: "Use 6.1 Sol.
It's very close to Astra on benchmarks and much cheaper. For this one your limits will be enough." That's roughly how I see the release of this model.
Most likely it's also that Anthropic released Opus 5.5. And, as we know, on a huge number of benchmarks Opus 5.5 now sits above literally all systems in terms of quality, cost, speed, token gluttony and everything else. It's still unclear whether Fable is better or not. As far as I understand, Fable is a bit less restricted safety-wise, but Opus 5.5 just rules — some unbelievably cool model. And then this comes out.
Sasha, wait, we'll get to it in a moment. Against this backdrop, against the backdrop of Opus and 6.1 Sol, Google literally just announced that they're back in the game and are finally releasing their Gemini 4. Well, I mean, they announced it, but nobody has access yet — there's a waitlist and all that. But maybe on this turn Google will try to overtake OpenAI. We'll see. We might have to get a third big subscription to compare them.
Maybe. For now, Google is still lagging far behind. And we'll show you the ranking once again. I showed it this week in two episodes, on Wednesday and on Friday. A ranking by usage classification — models A, B, C, D, E. Google has its own classification, and S, SS and SSS. By the way, the ones that came out on Wednesday and Friday are really cool, I think. I waited a long time and I'm very glad we managed to make all the episodes. They're about trends. Ten on Wednesday and ten on Friday: rules, thoughts, ideas, trends you need to know when working with artificial intelligence.
You can see a kind of playful, game-like mechanic there. It immediately reminded me: didn't OpenAI release Pets at some point? And OpenAI, I think, made different personalities in all sorts of voice assistants, or in the ability to choose what tone to answer in. But Elon Musk made a Dots Easter egg and bought the domain dot.com. And now when you go to dot.com, Grok's bot opens up — and specifically a bot, like a chatbot, which is already connected to all of Grok's models as well, and they can communicate with each other.
We'll now figure out what Dots is and its particular pros and cons. I, at least, gave it a pretty good try over the last twenty-four hours. Very intensively, I'd say. Although obviously it's an early version, and I hope our editorial team will show what it looks like. And all these memes — you can see Spider-Man, and Spider-Man without glasses sees Dots, and with glasses he sees Grok Bot. So yeah. It's surprising that OpenAI didn't buy that domain and that nobody spotted this at all, and so on. To me, of course, Ilnar, it raises a question. First, a bit of criticism.
I'm tired of the complicated interfaces of both Anthropic and OpenAI, of these endless menus, of the huge number of gimmicks they roll out that then stop working, and they hide your projects so you can't find them anymore. You don't understand what's important and what's not, and they announce something and tell you: "This is super trendy." And you have — by the way, Sam Altman said ChatGPT has 1.2 billion users — and you have 1.2 billion users. Why are you showing them a Sites tab when, out of all that, 8 million sites have been created? We'll discuss that separately, because I was at a session where the creators of Sites, the leads of this project, talked about Sites.
Basically, at its core, as the stated idea. Obviously, they showed some examples. What they had was kind of kindergarten-ish, but overall the idea isn't bad. So there's a central interface that can read your various chats in Codex, in Work and in ChatGPT. I'll say right away: it can't read everything. It doesn't have access to everything, it can't change all the settings. It can't do all of the functionality. As of today, it isn't integrated with WhatsApp or iMessage. It's integrated with Slack, with Microsoft Teams, with something else.
Obviously, the integrations will probably keep coming, if this project doesn't croak the way — the way Pets or something else croaked, right, shoved off somewhere to the sidelines, or like the Sora social network. But you can open this thing on your end and take a look; it's available in Pro. If you have Pro, you can create it from Codex. And then this pet will show up in your mobile app. And essentially it's just a single mechanism, so to speak — as I said, a single agent. It can create tasks in Codex on its own; sometimes it creates them, sometimes it doesn't.
Sometimes it says it can do something, sometimes it can't do something, sometimes it has access, sometimes it doesn't. And I loaded it up really hard these days, to the point that I said: "Print something for me on my printer. Send my wife a message via Messages on the computer. Sort through my email and reply to these people, connect to my Amazon." Anyway, I gave it lots of different tasks. It looks pretty good; interestingly, by the way, it runs on GPT-6 Astra, if I'm not mistaken, that's what was announced. Overall it's neat, it works fairly fast, it can take on many sessions in parallel, holds on to different tasks pretty well, doesn't forget them, can set recurring tasks inside itself — and I don't even know where those are stored or where they show up.
It remembers all kinds of information. So it's essentially a single window. And I think that's a plus of the system. Someone will say that some other systems have this or something more, but for OpenAI it's a big plus, because to me this functionality is something approaching the point where you start having one single window. Obviously, this single window can't take the form of a chat. And they're trying to do it. For example, they said you can call this dot, you can somehow integrate this dot into a website and leave it there, you can bring up this dot in Codex. So.
I immediately asked: "Can I write to my wife's dot, so that my dot writes to my wife's dot?" And it said: "You can't." Yeah, for example. Things like that, of course, seem very interesting to me. When it becomes possible to operate across these systems at all, and when this system isn't just a chat interface but is rather available everywhere and can really control everything. But I'm not sure that OpenAI — Ilnar, look, I think OpenAI won't be able to solve, right now, in its infrastructure and in its…
in the way the corporation is built at all — they won't be able to solve the task of reworking four completely disparate systems. They have ChatGPT on the Mac, there's Codex on the Mac. And they're different systems. And, for example, only the ChatGPT chat has the Record button. Let me remind you, it's a unique thing, very cool: if you're in a meeting, you press the button, it records everything and also transcribes it all right away, right? Then on mobile it's its own app, with its own settings, its own system, and on the web it's its own system and its own infrastructure.
And they all differ a little from one another, with different interfaces, different perks and features. And I'm not sure that in this zoo they'll be able to push this Dots forward somehow, because I have a big question: why didn't Dots evolve out of Voice? Or why doesn't Dots have the cool Voice inside, but instead has some kind of phone call? I mean, there's a button, Ilnar, to call — I can call it — but there's no button to, like, bring up Voice. It's very strange. It'll probably appear, it'll probably be there somewhere. It's kind of strange in terms of connecting some plugins, right?
I mean, you have to, like, open ChatGPT and scroll through the chat in the ChatGPT mobile app until you get down to this dot. Or, for example, in Codex it also hangs at the top, just as your dot, under the name of a new chat, right? You can't create several of them, you can only create one. I think it's a very resource-heavy system for OpenAI; in a way it's getting closer to monitoring — going through your chats and looking at them. This is that idea — let me remind you, just as a memory — where I said: "I'd easily pay several thousand dollars if there were a system that constantly checked my chats."
So this is a similar system, and yesterday I gave it a task: go through all my emails, find the ones that concern my kids' school, filter them and send me only the important stuff. What comes in there — I don't need the junk, it's mostly junk, I can't read it. But send me the important stuff. And it took on tasks like that in parallel. Again, look, it seems no different from ChatGPT, but there's a nuance. If you have one bot, it counts as a kind of single messenger, right, this ChatGPT, an entry point like that — then for me it's a cool thing.
I realized that if my wife also had a dot, and they could talk to each other, and it was a built-in messenger, then my wife and I would only talk inside ChatGPT. I wouldn't talk to her over WhatsApp or iMessage anymore. Well, those are the two systems where we talk. I wouldn't talk to her anymore. Yesterday I sent her a message via the dot. She said: "Next time you send one, I'll block you." I said: "But you won't know I sent it via the dot." She said you can tell from the message. I just asked my dot.
The dot sent me the school status for the kids this evening and told me what's going on at the schools. I said: "Listen, keep my wife posted on this too." And I just went into my messages with my wife, right, literally in parallel with you, and I see she got that message. So now our dot will both help me and send, send messages to my wife. And what do you, our viewers, think about this system, this topic? Who managed to test it, who didn't? Obviously, not everyone is on Pro.
What are your feelings about it, right? And do you see this system being useful to you personally, or developing a lot down the road? We'll watch that together going forward. I'm waiting for the dot to show up on Alexa, for the dot to show up on any device, so I could load it everywhere. And then, I think, it'll be really cool. So I'm walking by and I say: "Hey, dot so-and-so, tell Ilnar's dot something," and so on. So this probably looks very interesting, and this kind of future movement should kill off a huge number of startups.
When our Tanya was trying to install OpenClaw four or five months ago, I think — well, now there's no need to install it anymore. OpenClaw has essentially appeared inside ChatGPT. The only difference, right: on the one hand, it's easier, easier to set up than OpenClaw, Hermes or something similar. Or, on the other hand, you're completely locked into the ChatGPT infrastructure. I mean, any third-party solutions can only be done through what's inside the ChatGPT ecosystem. But for most users that's more than enough.
Lowering the barrier to entry matters more here than being able to, I don't know, call various models via the API. And inside, there's a way to dive into the settings a bit. Yes, I mean, it can run on a little virtual machine. You can go right in — there's a desktop — look at the folders, look at some scripts, you can drop some scripts in there yourself. So there's a bit of an opportunity for geeks too, to dive into this thing and work with it. But again, far from everything is open. It's not an open source system.
Actually, getting under the hood of Dots won't be that easy. But again, 99 percent of users don't need that anyway. Most likely, this is the result of OpenAI buying — well, poaching — the founder of OpenClaw. He works for them now. Maybe it's sort of his little graduation project that they've now rolled out here. Do I think it's something useful? My first impression, from the presentations I watched, is no. It's like I never had a need for an extra character like that to interact with.
Yeah, I mean, I'd still rather have, I don't know, the problems I'm facing right now get solved, instead of extra features showing up that I then have to figure out. Like you're saying, the UI is getting insanely complicated, turning into a garbage dump. You've got Sites, you've got this Dots, you've got some other extra stuff on top. There's confusion between Work and regular ChatGPT too. Some places use tokens, some don't. Same thing inside Codex. So in that sense the cognitive load seems to keep growing.
The number of models — remember, we once dreamed they'd cut down the number of sliders and settings, but the number of models has only grown, and it's getting harder and harder to make sense of them. I'm sure some use for it may well turn up, but you absolutely can't rule out that this thing will die off just like Sora, which you mentioned. The plugins from the very, very beginning, the GPTs people were creating for a while. It smells a bit like that. Yeah, expecting a finished product from OpenAI that will definitely work is, well, unlikely. Their models are good. Codex turned out great.
Everything else is still sneezing and coughing a bit. I won't go into detail about the use cases, but from what I've seen, it's a buggy system for now — though they'll most likely polish it, yeah.
Yeah, look, again, a lot of good stuff works for me, including as a kind of entry point, a single window — it is one window after all, right? And I even like that they don't let you create many dots, because many dots is a fundamentally different concept of working altogether. Essentially, by its very… and there's probably no need for it, because, yes, you can make lots of chats. Yeah, because basically a chat does, well, roughly similar things — or basically a project in Codex, I guess.
And so this kind of single window would be cool if they released it as some separate, additional application, well, a separate app, where you go in and there's only that — like, for example, there's Grok Bot, there's Grok Bot on your phone. Yeah, you open it and there it is. Although that's probably the same thing as ChatGPT, right? I don't know. I don't fully use Grok here, but if there were just an app or some entry point into it, and it only sent you all the updates, and you worked with it directly.
Right now, Ilnar, I have a certain feeling that this — this is a cool thing. Again, if it gets access to everything — because its access is limited, it often can't do a huge number of things, and you have to go directly into, say, Codex to get them done, or directly into ChatGPT. And of course we're waiting for voice to be integrated into it, waiting for video recognition to be integrated into it, yeah, waiting for integrations there, and waiting for this application to be installable in all sorts of places.
By the way, a stated advantage of the $500 ChatGPT Pro is that it can be used through partner apps. And Sam Altman announced from the stage that there are already dozens of apps there. And what does that mean? It means that, supposedly, in Vercel, for example, by logging in with your ChatGPT Pro account, you'd get access to the compute and the tokens of that account of yours, fully your own. But I don't fully understand this advantage — how well or badly it'll work, everywhere or not everywhere. We'll figure it out.
But Sam Altman did seem to say several times on this subject that there'll be this end-to-end subscription-sharing thing. Whether it'll work that way or not, we'll find out at some point. What else important was announced? They announced there'll be a marketplace. How this marketplace will work, what it'll do — nobody has a freaking clue.
I don't think we'll even spend any attention on it today, because Ilnar will immediately say: well, there was the GPTs marketplace, there was the plugins marketplace. And in general, what marketplaces haven't they had, right? It looks like the news is just that someone's doing some project for the enterprise market, like, say, Amazon or AWS decided to sell some of their subscriptions. And OpenAI will let them sell their subscriptions. That's kind of what this thing looks like.
And in general, I think ChatGPT has been saying a lot in recent weeks — and the company's leaders have been saying a lot — that OpenAI won't build all the software solutions, that it's a platform. I'm still sure that OpenAI is an operating system, not a platform. And this "we won't build various solutions" sounds kind of very peculiar, right? Because even something like — on the one hand, the way Sites directly kills CMSs, first of all, if you look at it honestly, or GoDaddy too, hosting providers too, Vercel and so on, right?
And on the other hand, OpenAI is still building things like Dots, and Dots, in essence, should at some point even kill messaging apps. Obviously, that's if it's messaging apps within some environment. But what I want to say — again, once more, take the example of communicating with my wife. I mean, if we had dots, our ordinary basic communication would end altogether. But I don't mean communication in person — in person, in person we obviously talk and will keep talking — I mean communication at the level of iMessage or WhatsApp.
So if we're talking about things like that, it would probably end, because in terms of scenarios for our family and a huge number of different tasks and everything else — well, and we do see each other all the time, we live together, right? So it's definitely a very interesting topic. And so whether OpenAI will — whether it'll remain some kind of platform that people can build solutions on, or an operating system, we'll see in the future. My bet — my bet, well, you know it.
Well, Sasha, as a concierge service it looks cool, right? Like within a family, I don't know, keeping track of, I don't know, a parent-teacher meeting, some concert and other things like that, and sending notifications about it. It looks interesting. But on the other hand, they give the following example. Say you're working, and in Slack you've got some — you get some message, and you write there: "Dots, pick up this task for me."
Yeah, that looks pathetic.
If my colleague did that in reply to my message, you'd go to his office, give him a smack on the back of the head and say: "Look, either we work together, or I'm going to just…
So, you know what the funny part is? The one showing it was, like, one of the architects — she showed it, and that's how she replies to her colleagues. And I thought: well, your colleagues will answer you in exactly the same way. They'll ask Dots exactly the same way, right? They'll answer you the same way. And then your Dots are talking to each other. So then the question is, why do you need Dots?
Because if you work at a company — here, Ilnar, I'll give you a simple example that was shown at the closed Sites session. I actually think it's a perfect indicator of how modern corporations work: when it comes to automation, modern corporations pee, they poop, they're ordinary people, and many of them are working decades in the past. So there's a young woman standing there saying: "So we built Sites." First, let me remind you what Sites is. It's a ChatGPT feature for creating your own website, hosting it entirely right there, publishing it, basically doing absolutely everything there.
And the person presenting it said: I'm the architect of this thing, I worked there as a programmer, and I came up with the idea to build it. And my very first thought was: do you even understand what websites are in the world — corporate websites, sales for all kinds of businesses, small, large and so on? And I had the distinct feeling that the people presenting don't understand this. Why? First of all, the young woman started by saying that there are 8 million sites, and that, finally, every OpenAI employee is able to create a site for themselves.
And I'm like: oh, cool, so people can now create sites. What do they need them for? Then she showed an example. At their company they have Google Sheets. At that moment I froze and thought: hold on, you're OpenAI, you can't possibly have Google Sheets. I mean, they have a tabular view of their data. She actually opens this spreadsheet — on how they're preparing for DevDay. And she says: "Before, we had to open the spreadsheet and look at the report. And now I've made a site from a single prompt to view this report." And I'm like: what difference does it make whether you open a spreadsheet or a site?
So she — she says: "Look, I changed it on my end, and it changed for him too, in his account." And they're doing this for real, Ilnar, dead serious, during the session — the room is packed with developers, everyone sitting there with their laptops, all serious, from serious companies. I think there wasn't a single person there making less than half a million dollars a year. More like a million and up, yeah, somewhere around there. I mean, serious, really serious companies. She's explaining: "Look, his data updated too." I'm sitting there thinking of leaving.
I wanted to leave about three minutes after the session started. I thought I'd go take a little walk, drink some water. But I decided, fine, I'll sit it out. I was already sitting in the third row. I looked maybe a bit extravagant, given how I was dressed, so I decided, okay, I won't get up, I'll sit it out politely. And then she — she shows this whole thing in terms of the functionality in general, well, what's been built with Sites and with Google. I'm sitting there thinking, well, in theory, if you change a status and… and the other one, the site's creator, says: "Look, I can go to the site and change the status.
And it's changed everywhere." And you're like — and it, like, changed in Google Sheets and somewhere else too. And I'm sitting there thinking, why do you even need all this if you have Dots? You can go into Dots and say: "I've finished the task," and Dots will notify everyone everywhere, change everything itself across all the architectures, and do a whole bunch of other stuff too. Why this visual stuff at all? Why is it even needed in terms of… why all this complexity at all, why this management, why these tasks?
Then in Sites they showed this thing where you — where, for example, a person says: "I like the color blue, and I can pick a different widget and a different size for this widget, say, so it's bigger on the site." Ilnar, I'm sitting there watching this and thinking: guys, what is this, 2010? In terms of CMS development, this is of course a fail, just the most brutal fail of everything going on, because, at the very least, they showed functionality where supposedly everything lives in the ChatGPT infrastructure, and you don't need to set up access levels and permissions, and you can log in through ChatGPT.
And I thought, well, actually, not every employee at a company has ChatGPT. And actually, clients come along who don't have ChatGPT — how are they going to log in then? I mean, you need an external permissions service that doesn't depend on anyone. All modern permissions services let you log in from every possible system. You have to let people log in from Claude, from Grok, from Google, from wherever they want. By the way, today I downloaded Grokbot, since I went to dot.com — and because I had Grok separately, I guess, well, yeah, I did, but not this Grok bot — and it offered to let me log in via X, via Apple, via Google.
Basically, same as always, right, Ilnar? But here you kind of end up in a system where you have to become a ChatGPT customer in the first place, you have to become an OpenAI customer. And when it comes to Dots and its presence inside the infrastructure, I'm sort of okay with it here, if I'm an OpenAI user and a billion people in the world use it, right? And, say, 500 million are communicating — that's a big system, right? Like, I don't know, Telegram has a billion users, or two, roughly speaking, right?
And obviously, at some point OpenAI will be used no less than Telegram or WhatsApp and so on. So there are these systems, and I can send a message to a person inside them. But a website is a completely different infrastructure — these are completely different tasks altogether. And the examples they showed are silly. My sister puts it this way. She heads HR at a company — a software development company, a thousand employees. And she says there's this term, home pet. You'll know it as a developer. Damn, I forgot — someone will remind me. It's pet, like the animal.
When someone comes in for a job interview, you usually ask them: "So what kind of thing are you doing…
You mean a pet project?
Yes, yes, yes, yes, yes. Pet project. Like, what's your pet project, right? And honestly, I'd just never come across that term. And she says: "It reminds me how at OpenAI all these smart people — I mean, they're supposed to be hiring smart people there — they're all making pet projects for you and me." Well, because all these projects look like pet projects. And it's exactly the same with most of the presentations. They showed some 3D visualization, animations, games. And usually people would say: "I love Formula 1, so I built a race track and burned a zillion tokens on it. I love games, so I made a game."
So that's roughly, roughly how people structured the whole thing, right? And you look at all of it and you can't tell what's cool and what's not. By the way, there was — I think it was Tibo presenting for Sam Altman, I don't remember, from OpenAI — and he was the only one who did anything even remotely interactive. He says: "I arrived yesterday and hooked up a robot here." He set down a little robot and says to it: "Show us what our hall looks like." And boom, it connected to the hall in real time.
He says: "Now create a site that shows, roughly, everything that's happening right now, the conference, and also raffle off three spots among the current attendees." Now, that looked great. I mean, at least it showed what Codex is, right? By the way, there were about three or four fails there. I think all of them were fails related to voice control of dots, or voice control in general. Everything that was loading — everything, Ilnar, died. And again, I'm not looking down on them.
I really like OpenAI, I liked their organization in terms of style and everything. It's fun being there and all, but when it comes to projects like these, it's really troubling, seriously troubling. And you voiced it — Ilnar is more conservative here. He probably used the right word, right? He said: "OpenAI's funeral," as people put it on the internet. Maybe I wouldn't say funeral, but of course there'll be a ton of enthusiastic bloggers saying: look, 6.1 is out, look, Dots is out, look at this other thing.
Of course, we live in a world where everything changes every day, where a huge number of new projects keep appearing, and it's very important to figure all of this out yourself, to use all of it yourself, because we, Ilnar, as it turns out, are evaluating this from a more or less, well, professional standpoint, right? After all, we're, well, pretty decent professional users of all these infrastructures and everything that's going on. And of course I feel sorry for people, in the sense — in the sense that you can see OpenAI tests everything on people. Sites — Sites is horrible, Ilnar, Sites is horrible.
Yes, yes, yes.
Horrible.
I think this DevDay will be very good for competitor companies, if they take advantage of it. Like how Elon Musk bought the domain, and that's exactly how you went in and downloaded Grokbot — downloaded Grokbot for yourself.
Right. Same thing — I think Google hustled and sped up their release a bit — well, not the release, the announcement of their Gemini 4 release. Maybe we should expect something from Anthropic, because DevDay, in my view, was pretty raw and not very strong. And for competitors, I think now is exactly the time to play off this position. Hopefully next week we'll be able to record quite a lot of interesting stuff in addition — other companies' responses to this whole story. Although there are certainly some interesting directions, there's nothing, in my view, revolutionary, so to speak, right? Because…
There's no central line. You know, it's as if there's no… there's some… look, there's a central line of building an AI model as, as an AI model. Statements that we'll build a platform, statements that we'll build AGI for people and so on — those are just statements, right? So there's the creation of, like, a smart, cool artificial intelligence. But everything around it is, of course, pretty sad so far. And in this respect I like Elon Musk's statement. Within a year, the most — the most fundamental agentic system that everything gets built on will be Grok. And in two years, I think he said, nobody will ever write a single line of code anywhere.
And, well, our editor will correct me if anything's off. And this statement of his about Macrohard becomes more or less clear, because he's obviously aiming precisely at them building their own operating system, their own platforms — which is clearly obvious, because at some point Android appeared, so this will clearly have to happen. I mean, at some point it has to happen. By the way, speaking of which, the other day I fully tested Siri AI, and it's available in English. I don't know if it's available in other countries. It's available in America. You have to get an invitation, and it opens up for you almost right away. It's crap!
I mean, well… I don't know how else to put it. For example, they did say it can search internally. Just now I asked it to find contacts in my contacts — I'd decided to delete some contacts. I've got, I don't know, thousands of these contacts. I decided to delete old contacts that I started creating back as a kid, when I bought my first phone in Belarus, a Nokia 8210, in 2001. I told it: "Look at all the contacts that have seven digits or fewer." Yeah, well, those are some landlines or the very first phone numbers in Minsk back then.
And it tells me: "Done, I found five contacts." I say: "Can you find how many there are in total?" It says: "Here, eight contacts." I say: "Can you find all of them for me?" It said: "I can't search everything. You need to open Contacts and search." I opened Codex and said: "Can you find me all the contacts I have that are like that?" It said: "Can do." A couple of minutes later it gave me: "You have 324 such contacts." I said: "Wipe them out." It said: "I'll save everything now, set everything up, double-check everything, and leave this thing for you just in case."
And off it went. I thought: "Goodbye, Siri. I most likely won't ever open you again." Although it did show up in Ask, it showed up in that system. But you get how surreal this whole thing is, right? And the fact that they again opened it only in English — that's just, well, some, some, some kind of nonsense. English isn't a problem for me. I'm just sitting here thinking: what's the point? I mean, in my view, the language thing has been solved absolutely, radically, incredibly well.
By the way, Tibo — I think it was Tibo — when he was showing this robot on stage, there was a cool thing.
In his examples he used everything together: the real-time voice recognition system, voice processing, and video recognition and processing. All these technologies are great, of course. So the question is how much money it costs, because it's very expensive. I mean, I just thought: if I hook up video recognition for myself now, not through an open-source model but directly through ChatGPT, how much money will I be paying when I have a camera running 24 hours a day, right? And what if there are lots of them — right now I only have one camera connected through Codex, and there are twenty-five or however many, thirty cameras just hooked up, right?
And I thought, if I run thirty cameras in a 4K stream through this system, I think I'll have to sell my kidneys, and my wife's, and I don't know whose else. And my own future.
Our viewers can write in somewhere — when it becomes a normal working system, we'll definitely tell you about it, how to work with it together. For now, I personally probably wouldn't recommend it.
Well, Sasha, in a nutshell: with Dots they dragged in OpenClaw, and with Spaces they basically dragged some kind of Notion equivalent inside. That's exactly why I'm saying there's nothing revolutionary, in that they didn't show anything new. They said: yes, we've dragged this inside our infrastructure.
It's not even Notion, it's, it's a crappy Notion, right? It's not Notion, no?
As best they could.
Yes! Because, because look, Sites isn't a normal CMS system, it's sort of… Yeah. Look, another cool thing came out.
I think it's very interesting for developers — the Decisions API. What do I mean, cool? The announcement is cool. What is it? It's a special API that lets you, roughly — you write code and you say: "And here I'll call the Decisions API, which will decide where to go." For example, if this email is marketing — that is, determine, say, the type of email that came in: marketing, sales, products and so on. It'll decide about that email on its own. I argued with ChatGPT for a long time: what's the difference from just asking about this via the API, picking a model, because the Decisions API only works with Luna and it's unclear how much it costs, whereas via the API you pick any model and ask it for a solution, or a Decision.
It said: "Well, overall there's no difference, it's just something purpose-built for that." And again, of course, that's questionable.
By the way, Ilnar, I was really, seriously struck, and in a negative way, as a fairly public person who's been doing public speaking for a very long time, you know that, right? Well, my public talks started, probably, back in 2008, since 2008. But obviously, from Megaplan on they started, well, moving forward, further. And everyone spoke reading from a screen, every single one without exception. Even the person who did the robots, Tibo or whoever he is, who told everything emotionally — he was reading it all.
Some read professionally, some didn't. Sam Altman read everything; sometimes he'd say "Thank you" — that he didn't read. It certainly looked very... It's awfully weak, just awful. I mean, the people coming out should be ones who talk about OpenAI, show some gigantic things, sell the unsellable, right, in theory. And if you want to sell a lot of Pro modes, then do something serious. I mean, roll it out seriously. It looked, well, really very weak, just like, actually, the use of artificial intelligence at the conference.
Everything was sent out by email; you couldn't, for example, hook up WhatsApp. To find out how to get there, I had to ask Codex. The QR code came on the last day. I mean, you know, that's how it all looks. There's no face recognition system. They could have done all kinds of cool stuff. Sign up for the face recognition system, sign up for all this stuff. You walk up, there's an iPad standing there, and you have to turn this QR code toward the iPad so it can scan it. So, very strange things, when it's essentially a geek party — it's DevDay, after all.
Only developers sit in the front rows, with black developer wristbands, right? Tables are set up, all with computers; everyone everywhere is on their computers. I kept thinking — these people, what are they doing with their computers? What are they up to? Are they taking notes, what are they doing, right? So everything was read out, everything was read, and that's probably a minus. Once again, I love OpenAI, I'm sure it'll be worth $10 trillion, they'll grow, keep moving and so on.
This is exclusively my personal opinion, one I've reasoned out myself. But for all that, there are strikingly strange things, like offering me the $100 Pro version. And I want to say: "Come on, you can see who I am. I filled out lots of forms, I gave lots of details in the forms. So why? Why did you ask me about those details? Why ask, then?" About the details. Why do you ask people what they do? What do you need it for, right? Why not do something real, well, like — not even something interactive, but show this movement, where if Sam Altman declares that we're already in the technological singularity, if the US government yesterday — by the way, you know, right? — yesterday changed the term: in government you can no longer call it artificial intelligence.
Now it can be called superintelligence. Did you hear? It's "superintelligence" now, right? And an executive order has been signed — that's how it's to be referred to at the government level. So everyone's discussing that AGI supposedly already exists. Maybe there's no ASI, and maybe there's no real technological singularity, but for OpenAI the technological singularity is here. Of course, all of this astonishes me. And I really want — I'm, I'm not, not protesting here at all — I really want people to genuinely get into understanding the profession, because a great many people will lose their jobs and money out of ignorance.
A huge number of companies will go under because they won't start doing what needs to be done. So I want people to figure it out, including with the help of our channel. Yes, Ilnar.
That's exactly why — about what you just told us, your impressions of DevDay — that's why my main feeling about it is that they took away half the limits I had before and gave nothing in return. That's why I was really upset by this DevDay.
Listen, look, I, once again — I hear you, I hear you. I think a lot of people have that question. On the one hand, we're moving forward, they're expanding our functionality, and we live in a world where functionality will only get cheaper. They keep saying that everything's getting cheaper and so on and so forth. Whether it's getting cheaper, I don't know. And really, Ilnar, probably, what you used to be able to do for $200 and what you can do for $200 now — those are different planets.
I mean, really, seriously different, right? Different planets. I don't know how much money's worth I've programmed by now — I've programmed a lot — but in tokens, and I'm still laughing, I've got 50 billion. Right at this moment it's 50.3 billion. Yesterday it was 40.9, right? Or what was it? Or the day before yesterday. So I've gobbled up 400 million tokens over these days. I've got 50 billion tokens, and that's like, I don't know, a million-plus dollars. And I, I, so — that's been spent. In Codex alone.
And so I — that's without ChatGPT. Of course, I'm deeply grateful to all these companies in general and to everything that's going on. On the other hand, of course, we're in a zone of absolute uncontrollability, and in a zone of uncontrollability like this, I'm someone who never whines, I live, I always say: live your life, move forward, raise your children, love everyone else around you. But of course we're moving through an unknown that updates every day. Just today I was filming an episode — watch it on our channel, it was out on Friday — about the top, most interesting things for the fourth quarter, for understanding artificial intelligence.
And there was this, this thing I've always said: that systems change not over years but over weeks. Well, systems change not over weeks, but over days. And you and I see new updates coming out every day. And as always, there'll be someone writing in the comments: "ToTheMoon, you're not much of a channel, because you filmed this too late. That was five days ago." So. Everyone who writes comments like that — write them, we really want to see them. Don't forget to leave a like, definitely, and reply with a comment.
You can hit dislike if you really didn't like it. And we have about 96 percent, 95.8 percent likes versus dislikes. Those of you who write comments, don't forget to leave likes. I've seen lots of people who thank us but don't hit like. Someone's written, like, twenty comments, and zero likes. And you read the comments — great, they say thank you, all of that. Don't forget to like, don't forget to subscribe. Share this episode with people who'll be kindred spirits to us and to you.
Don't forget to do that. Yes, Ilnar.
There's a story on this that the lecturers at the physics department told students back in the early 2000s: roughly speaking, a couple of thousand years ago you'd acquire some skill, and that skill lived on for generations.
Say you learned, I don't know, to hunt, passed it on to your son, and he to his son. And so, I don't know, for twenty, thirty generations that skill doesn't change at all. Just as people chased moose with a bow, they keep on chasing them. Then, closer to the Industrial Revolution, skills started to last for one generation. So, say, you learned, I don't know, to type on a typewriter, and you work as, say, a secretary. Your kids won't need it anymore, but for you it's enough. The skill you mastered is enough to last you through the main part of your life.
Then later, closer to the 2000s, people started changing professions within their lifetime, right? They'd study for one thing, then maybe even work at it for a while, and then everything started speeding up. And now it's exactly what you're saying: reality changes not over years, and not over weeks, but already over days. We're getting closer and closer to that. And what was the point of that whole story they told students at the physics department in the early 2000s? That the main skill that still hasn't lost its value and is still alive is the ability to learn.
That is, if you know how to take in information, acquire new skills and so on, that's the most important thing you need, because knowing in advance what to learn so that it'll be useful to you in the coming years is, well, practically impossible. Everything about these changes with AI confirms this more explicitly than the professors told us in the early 2000s.
Like, you definitely don't need to learn prompt engineering. And you know what Danila told me? You know Danila, we went to Kukash together, he's my nephew. He told me — he studies at the Warsaw University of Technology, in a pretty serious science program. And in their first year, Ilnar, a lecturer taught Windows 10. Windows XP had been crossed out and changed to Windows 10, and they had lab assignments on how to take screenshots and how to change the resolution. That's our world…
Didn't they have to create disks too?
Our world at this moment in time — I said such people should be put in prison. Our world today, it's, well — not people like my nephew, but those who teach that kind of thing. Our world today, guys, is, alas, roughly like this. I mean, a huge number of people are telling you that you need to learn how to take screenshots. And the funniest part — they take screenshots at companies, right, in order to automate processes. So. Watch ToTheMoon, don't forget to support us, and learn to learn.
And most importantly — apply it in life, stay good people and love other people. Bye, everyone. We'll see you in exactly a week together with Ilnar, and with me in a day.
Very soon.
Bye, everyone.