Hi everyone! You're on the ToTheMoon channel — tech news and insights from Silicon Valley and around the world. I'm heading from Silicon Valley towards Santa Barbara right now and decided to record an episode for you, because just now, five or maybe ten minutes ago, news came from OpenAI that they are, after all, planning to merge ChatGPT with Work — chat with Work and with Codex. This is a substantial evolutionary influence on how these systems develop. And we know that this system already consists of ChatGPT, Codex, Work and Dots. That's important — yes, there's Dots.
There are also various new systems. They had a system, by the way, a strong one for researchers — they're winding that project down. The day before yesterday, or in the last few days, they launched a new system dedicated specifically to biologists. So they have different ones — I think they have a system for the military. There's already a system, there's a system for education — they have a whole line of systems there. Our editorial team will highlight them. But I want to tell you about this substantial, substantial nuance, right, the one that's going to happen here now. So, look at the situation, at what's happening now.
OpenAI... is planning to merge ChatGPT and Work with each other within one to two weeks. For many people this will seem like an unclear story, because people may not even know what Work is. And you rarely find people who understand what Work is, basically. Because, well, people are so focused — they know what ChatGPT is and what Codex is. Let me remind you once more: ChatGPT is OpenAI's basic, original chat, where you can give absolutely different tasks, including tasks of what's called agentic work — for example, for this chat to send an email.
Or for this chat to study your finances, or for this chat to set up a weekly recurring task. That kind of basic functionality. Then there's Codex. Codex was originally developed as a system on which you can build various solutions and work together locally with your computer. And this "together locally with your computer" often causes some confusion, and people think it's for some geeks or for programmers. But in reality Codex was a new format, a kind of upgrade of ChatGPT. Essentially, recurring tasks that work much better.
More convenient work with files, with various reports, with editing presentations of different data, processing that data. I want to say right away that the models worked differently from the start. That is, if you take ChatGPT and Codex, you'll see that there's a different choice of models and of the models' working levels. It was different then, and it's different now. For example, in ChatGPT you choose a default model — there's such a notion, so to speak. Then you choose its working level. And it's strange there: for example, in the medium working mode it may put on the 5.6 Sol model, and in the maximum working mode leave 6 Astra, for example, right?
In Codex the story is a bit different. You choose a model and, choosing its working mode — its working mode doesn't have, for example, the notion of Pro, right? It has a mode — Medium, there's High, there's Max, there's Ultra Code. But there's no notion of, say, a Pro working mode. And what was visible from the start? Look, Codex didn't work as brilliantly in a single session — well, it didn't want to work the way ChatGPT wanted to work from the start. It was as if the quality of the answers was a bit different. Even if you launch an expensive model and ask a question, it's as if Codex doesn't search the internet the same way, as if Codex doesn't analyze its common, unified base the same way.
That is, as if Codex were just a model deployed locally on your computer, wanting to solve your tasks. Along the way, they also released Work. What kind of environment was this Work, in ChatGPT's view? That if Codex is for developers, then Work is a system that lets you work. Well, it was actually a copied story of Cowork at Anthropic — probably the approach was copied. By the way, Cowork at Anthropic has disappeared too, right? And what was OpenAI's idea? To create an environment where people can work with various documents, files, presentations, email — but not with programming, so to speak.
well, the teams that built them inside didn't understand what these systems would actually — well, what these systems would actually achieve. Even if you listen to the leaders of these companies now, you can see very strange examples from everyday life. And, by the way, I want to say that it seems to me these people very often don't use Codex themselves — or their own… or Anthropic. They don't use it daily as much as advanced people use it in various businesses or when building systems. Essentially, they can't use it.
Where would Sam Altman find so much time to sit and figure out how to build various processes? If he doesn't do it himself, then essentially he isn't living through the new era of new use cases at all.
It often amazes me why people, so to speak, seem not to talk about these cases. I was watching recently — a friend of mine came over, and he's actually helping me with filming today in the car while we drive to Santa Barbara. And he was watching an interview of a blogger with Greg Brockman. He's the number two person at OpenAI. And the blogger says to Greg: "Look, I've developed a system in which, like, like an application, right — it's an app that lets me gather all the information about who you are before our meeting and put together questions and answers for you.
And I really want to fine-tune this system so that it helps, understands how I personally, as a blogger, usually conduct interviews, what my processes of interaction are and all that. What do you think about it, Greg?" And Greg says: "It's a great system, very interesting. Such systems are useful." They carry on the discussion, and I'm sitting there thinking: "What great system?"
Today you can go into Codex and say: "Codex, build me a system in which — gather me a report in HTML form." That is, essentially, it will immediately deploy an absolutely working tool for you, which you can then publish anywhere, even as a website, right. Well, it's not about HTML. Again, HTML is enough at the very start. Gather information about Greg Brockman for me as follows. Take all the interviews he has given over the last seven years. Download all these interviews from YouTube, transcribe them with a local model, spin up a local Whisper for me, even if it's free, so to speak — transcribe all these interviews for me in full.
Then take all the publications in all the media where Brockman has ever appeared. Comb through no fewer than a thousand different sources, study the policy that exists at OpenAI everywhere. Open social media, look at what he posts and look at what his agenda is. Compile all the trends he has singled out. Compile all the things he didn't manage to do. And where he made wrong predictions, where he got predictions right, who he had conflicts with, who he has friendly relations with, what his position is on the development of artificial intelligence and on the twenty most important questions about artificial intelligence on planet Earth.
And show his opinion on these questions. And questions of the kind: which professions will remain and which will disappear? How much will a person earn on average in their life? Will people fly off to other planets? Will robots be present among us, or will there just be some additional technological environment among us, and so on. I want to tell you that… ah! one more thing. And now, Codex, please analyze all my interviews over the past year that I've conducted with people of this kind.
Work out the overall style of these interviews, look at what questions I asked as a person and how you see it — how interesting or uninteresting these questions are. Also take for me all the bloggers who asked him questions, and suggest some exclusive, unique, new questions, some hype questions that are on the current agenda right now, and questions I usually ask in my style, and deliver this work to me. So that you don't make mistakes, Codex, break it all down into stages. Make no more than ten stages but no fewer than five, and start doing them in sequence.
But first carry out a general audit of how you'll do it, what you'll do and so on. Recommended model for the build: GPT-6 Astra in Max mode. So you've given such a prompt by voice — I gave it, it can be copied, this prompt here, you can copy it and paste it into Codex in full. It's enough, yes. You can do this for any person, by the way. So I've given the prompt. That is, essentially, if Greg Brockman were giving an interview, he'd say: the fact that you made a system is some nonsense, because even what can be done now, the way I just said it, is nonsense.
That is, it's the old world. What I just said is essentially the old world of artificial intelligence. After all, look at the problem that's happening: people were making an incredible mistake. And by the way, we have a list of trends on the website, and this topic is marked there. I first got into automation — I got hooked: in two thousand three I got hooked on data normalization, and at university I had a course on databases. There were the first, second and third normal forms of data storage. And I liked it very much. And I started, started, really got into it.
That subject was very interesting to me. Laying everything out by entities, structuring all the data. I got really hooked on it. And it so happened that I devoted a big chunk of my life to automating enterprises of absolutely different kinds. ERP, XRM, CRM, various systems.
And the fundamental mistake everyone always made in automation: all people, when automating, always wanted to automate the processes they already had, without changing them. That is, people wanted a computer to help carry out existing processes, but they didn't change those processes. The same people stayed, the same salaries. Well, maybe everyone wanted it to save money, but the processes remained the same. That was a fundamental mistake altogether. Why? Because basically all these companies that did automation would come to the management and then go to the employees to find out from them how the process should be automated.
Well, obviously, employees wanted to automate their own processes so that things got easier for them — not easier for management, not so the company earned more, more money and so on. Right? And this violation, it — but it still exists in the world, it's extremely widespread. And cool new automated systems are usually created basically from scratch. Someone comes from scratch and proposes a new idea. From scratch. Like Airbnb was created from scratch, or Uber was created from scratch, I don't know. They create something from scratch, they come in. And that gives the market an understanding of a new business process.
And old companies gradually start catching up with this new business process, for example. That's how it works. Support our channel. I'm very grateful for every like, every comment. And, by the way, I really like that in the comments — our audience writes such real, living things in the comments. Even if, well, you're shy to write something, it would be great to always see what city you're from, what country, what your profession is and how old you are. And, well, are you a man or a woman? And who in your family interacts with artificial intelligence, and how.
It even broadens my perception — not just of the audience, but in general of what kind of people encounter what kind of information. We have a large audience, actually a very grateful one, at ToTheMoon. And I see how diverse this audience is. There are many around me — well, since I'm still in the professional blogging market, well, a relatively professional one, I'm still in the professional education market, yes. For many years of my life, fourteen years, specifically in education, fourteen years. And many people say, for example, that our content is very complex.
But at the same time, when I see people of different ages coming who have never encountered artificial intelligence in their lives and are grateful to the channel, saying it's better, you understand that all these outside comments that our content is very complex are, well, rather specific. So we're waiting for your support — likes, subscriptions and sharing with friends.
Now, with the coming of the era of artificial intelligence, a more serious problem has arisen, because artificial intelligence and what the models do today make it possible to bring incredible changes into a company's life, yes. Simply crazy changes into a company's life, colossally different from this standard automation. And what used to be called, or is called now, XRM-like systems, or CRM, ERP, then MRP and so on. And, well, of a simply incredibly different class. Moreover, today, with the help of artificial intelligence, we're able to create new systems whose names haven't been invented yet.
Look, they haven't even been fully thought up yet. That is, what will the system I just, well, suggested to you for searching for information about Greg Brockman be called — what will such a system be called? It's unclear what kind of system it is. Right? That is, evolutionarily it's unclear what this system is, how it will move and how this system will fit in at all. And you understand, right, that this is an elementary story, elementary. And we're entering a situation in which we study processes, people, life as it happens, cause-and-effect relationships differently.
Everything starts going differently for us. Clearly people will get to conducting interviews like this, well, a bit later, yes. And to preparing like this, people will get a bit later. Few people now understand how information can be gathered like this. Even those who consider themselves advanced in artificial intelligence won't actually carry this task through to the end. It seems, look, the request is very simple — the one I, well, said. But at the end I specified, first, the model. I said GPT-6 Astra Max.
You may ask me: "Alexander, why did you choose exactly that?" Right? And the second story, which may be an interesting aspect, is why I said it needs to be broken into stages, no more than ten stages and no fewer than five. And then how to keep track of these stages. That's a big question altogether. Well, to really implement it there inside. I want to point out that this doesn't mean the model will do everything in five or ten runs. It — you'll have to adjust the execution of this task a little along the way. That is, the task seems easy to implement.
At the same time, there are a number of nuances in implementing this task, yes. But these are new systems of a new approach.
So, look at what happened at ChatGPT, in the era of Codex's development. Codex essentially became a system that made this kind of thing possible. And because OpenAI and Anthropic didn't foresee such cases from the very beginning — that is, their staff lacked the skills inside to spot such cases. Their staff and the people who build these systems — there are many of them. Look, there are central people who build the core of the model. These are the most brilliant teams. They are, as it's considered, at the frontier of artificial intelligence.
And indeed, right now Elon Musk liked a post saying there's a very strong difference between the people who develop artificial intelligence models at the frontier and basically ordinary people who use them. But if you look next to that at the projects that appear at Anthropic and OpenAI — for example, projects for mailings, for communication with their own customers, projects for building their own websites, projects for building learning systems around them — it all looks as if it were built in the 2010s. That is, it's all very old.
And essentially this split that happened into Work and Cowork — well, into Work, into Codex. This ChatGPT, Codex, Cowork and — oops — and Work, and Claude Code, Claude and Claude Cowork. Now there's Claude Design, even Claude Duet, right? Or like ChatGPT Dots. This split comes from the rather narrow perception of many project managers who work there, people, in terms of their understanding of the future. That is, they often try to recreate processes that someone somewhere on the market has described a little. Right? That's the reality. I… there are good people there and all that.
But let's look at the fact, the fact as it is. And this case I described with Greg Brockman is a case of the consolidated work of chat, Codex and Work. And these are exactly the systems that lie ahead of us. Us… although in fact the case I told you about is a case from the past; it's already a very simple startup case to implement. People often write to me on the channel: Alexander, you talk about very complex things, and you use some expensive models we don't have access to, and so on.
Look, it's a question of thinking, a question of thinking. Would you have been able two years ago to launch a case like the one I just described with ChatGPT? Simply, would you have been able to foresee it right away? And are you able to foresee how, essentially, in the future, ahead, these models will work, how systems will work in the future, what they will do? Tens of thousands — actually hundreds of thousands — of startups in the world have died, and more will die, that develop systems for artificial intelligence. And what is the problem with these startups?
That they're trying to solve some local task that, basically, it's clear artificial intelligence will solve. And this startup doesn't need to be launched. Well, some tiny minority might get lucky with the right arrangements with investment funds, with the right contact for raising money somewhere. That's called the right politics. But a tiny minority. What's going to happen at OpenAI now? Why did I decide to bring up this news for you today?
Because OpenAI is planning to remove, essentially, the notion of unlimited work — to remove the chat. As we all know, OpenAI's models, unlike Anthropic's, in ChatGPT were essentially unlimited. Certain models were limited for certain users and so on. But quite good models remained available for people to keep working with. And on the Pro plans in general there was essentially no limit at all. The Pro plan at five hundred dollars, for example, is completely unlimited, even in terms of using the Astra model.
What's incredibly cool is that you could always in ChatGPT — you know my cases, right — upload tens of thousands of pages, hundreds of pages of articles, or upload a million phone calls to analyze how a company works, or upload, I don't know, a hundred thousand calls, two hundred thousand tickets from the customer support system, upload code, programming, upload marketing, sales and financial data in order to analyze a company. Yes, you know my case. Now, if I come across some business and want to come in as a shareholder, or I need to sit on the board of directors there, I say from the start: "Give me absolutely all the data, I'll analyze the whole mass of data, yes, and look."
That is, they're now going to remove, merge these versions. They've already sent it — just now, ten minutes ago, the email came. I decided to tell you about it right away, well. Of course, a few days will pass before the episode comes out, while I'm filming now, but you'll see it — it's going to go away. What do you think about this, in general? Because this topic is, of course, well, very, very important, very serious. By the way, I'm driving now — my friend and I are driving to Santa Barbara.
And once, for me, Santa Barbara was a kind of byword. In particular because in the nineties the TV series "Santa Barbara" appeared and, of course, the post-Soviet space ran into this series in a big way and lived its understanding of America and Santa Barbara through series like that. And also through similar Brazilian series like "The Rich Also Cry." By the way, Dima, you know, right, that the series "Santa Barbara" wasn't filmed in Santa Barbara. I think even some of the opening credits come from there. And it's a most beautiful place.
We'll get there in, already in two and a half hours. It's a four-hour drive from home. And there's a most beautiful hotel there on the ocean shore. There's a very good place there, The Ritz-Carlton Bacara, very different from other Ritz-Carltons. If you're ever in these areas, I highly recommend it, because these are amazing places. Here, as we drive, you can see on the map where exactly we're going. This is the ocean, here's the Pacific Ocean, and we're moving from north to south, from north to south towards, towards, well, Los Angeles. Further on, Los Angeles begins.
Well, here is already the border, where San Diego is — that's already the border with Mexico. A most beautiful area to drive through and to record a special episode about artificial intelligence for you. By the way, over the weekend we filmed — well, you know, right, that our videos come out many times a week. Mostly, well, three or four times. That is, our main release days are always Wednesday, Friday and Sunday, and sometimes Tuesday too. And on Sunday and Tuesday the segment is hosted by Anna Volchek, right?
And on Sunday we have the weekly podcast. That's the original ToTheMoon podcast as it was. We created it and talked precisely about insights and news from Silicon Valley. That was really our focus. And around the world too. It began, it started with four of us. And we have two irreplaceable people — Ilya and Ilnar. And for a long time we were even four, then three, then two. Now, from the very beginning, Sasha and Tanya were with us, and then only with Tanya. Now, on Sunday, we filmed, and on the episode we had Anya and Max together.
So watch that episode. I think it turned out to be a really great discussion of what's happening in the artificial intelligence market. There are so many topics to discuss now. We'll also film for you the story that's now unfolding on the financial market of these companies, right? Given, after all, that Anthropic, well, it's going to have its IPO and go public, the company. This will most likely bring certain adjustments to the market. Ordinary people may not even notice it. Why is this topic interesting to me? Because we've all been talking about it for two years.
We've been talking with you about the IPO, about how these companies are financed, about what will happen on the market. So stay with us.