Transcription
Why has Open AI, over the past year, not released a number of certain products in terms of such cool solutions? ChatGPT is not a product, it's an experiment. And we are not clients of Open AI. Their secret weapon is Sam Altman. Someone told me a story about him, that Paul Graham once said, one of the smartest people I've ever talked to is Paul Graham. He's the guy who founded Y Combinator. And Gemini – is it just a technological solution or is it a product? Now Google has a foundational model. Everything should work on top of Gemini. >> What about Anthropic and their culture? >> Anthropic is a kidney that was moved from Open AI. Knowing how Apple is structured internally, I can say for sure that the iPhone is an exact copy of the organization that built the iPhone. One of the initial investors in Open AI was Musk. Hello everyone. We are on the To the Moon channel. Tech news. Insights from Silicon Valley around the world. We have a special episode today. Let me remind you, we broadcast on Sundays, every Sunday, and on Wednesdays, and sometimes we have periodic special episodes. And today we have a special, special, special meeting with Maxim, my close friend, who, Max, how long have you been living in San Francisco, 13 or 14 years? >> Well, the thirteenth year, >> right? The thirteenth year, he's been living in San Francisco. He's very deeply into artificial intelligence, technically, with knowledge and contacts. And we were recently discussing the topic of the influence of the culture of various companies on the products they produce. And while discussing the questions of why, for example, Open AI releases certain solutions, and some things don't work out for them, or why XI gets what it gets, we came to this topic of culture and decided to launch a special episode. I think it will be very interesting for you. Don't forget to support the channel, like, and also send greetings to Max. One of my fundamental questions was, why has Open AI, let's say, not even in the last 2 years, but in the last year, not released a number of certain products in terms of such cool solutions? For example, they endlessly move the model selection slider, or launch Pulse, which they then don't develop, or release a research version and then don't develop it either. Or they release Sora, then forget about it, then release a new upgrade, then forget about it again, then release GPT, then say it doesn't work, then release plugins, released them, then they don't work, then, they say, you can make your own applications, then they release agents, then they forget about them, release reminders. That is, Open AI has made a huge number of different solutions, but it's as if they don't bring them to completion, right? They released shopping, yes, they didn't bring shopping to completion either. And why does this happen? >> Well, this is not a unique situation. Google was criticized in exactly the same way at one time, that Google constantly launches some products and then closes them. But here, as we've already discussed, I think this is a typical application or a typical situation described by Conway's Law. Conway's Law is, well, it's not exactly a law, it's an observation by an engineer, which, the law states that any organization that develops a system inevitably creates a design that mirrors the communication structure of that organization. That is, the organization, in essence, in the form of the product it's working on, creates a copy of itself. How the team is structured, so will the product be structured. Yes. If we look historically at what companies here in the Valley have built, probably not just in the Valley, I'm just much more familiar with the tech companies located here. I can say that this law corresponds to reality 100%. That is, if we look at how, for example, Apple builds products, and knowing how Apple is structured internally, I can say for sure that the iPhone is an exact copy of the organization that built the iPhone. Right. That is, at Apple, essentially, product managers run the company, not engineers, not managers, not some people from above, as everyone thinks. No, it's product managers, who of course were hired by people from above, bosses, right? So the culture determined who was hired, and then these product managers manage the entire company, and they, accordingly, build the product. That's why, if you look at how the product looks, it's very coherent, it's very, uh, seamless, so to speak, right? Because the product managers agreed on how everything would look. And if something doesn't conform to this general vision, which they agreed upon, they will fix it, right? Then let's look at Google. Now, Google is a completely different situation. There are several products that are simply technologically so far ahead. Take Gmail, for example, no one can build another Gmail. Yes, still. How many years has Gmail been around, yes, it was released in 2004 or 2003, and for all these years, no one has been able to replicate Gmail's success. But at the same time, the product itself is not so polished. And, again, even if you look at how Gmail is released, when was the last time you saw a Gmail release, what version of Gmail are you on now? Everyone knows that we have iPhone 17 now, right? In September, Tim Cook comes on stage and rolls out a new version of the product. What is the current version of Gmail? Nobody knows. There's some internal build, right, but nobody has any idea because Google is run not by product managers, they are not product people. Google is run by engineers who, essentially, set tasks from a technological point of view. They have an idea of how to do something. They've paid for this idea. This feature is integrated into Gmail or into search or into something else, right? But as a product, there isn't one. It's a living, engineering, living engineering system. Of course, there are product managers, but I worked in search at Google, right? I just saw how it all happens. We had one PM for 50 engineers, who, accordingly, was concerned with something, like the color of buttons, but essentially all decisions were made at the level of engineering leaders, right, the tech leads and so on, who then, accordingly, directed the movement of the entire product. Well, and now we're approaching, accordingly, what's happening inside the AI corporation now, right? If we consider them through the prism of Conway's Law, right, it's exactly the same picture. Who founded Open AI? Five researchers. These are people who were doing research on AI models at Google. The people who are at the origins of the company set its culture. That is, they crystallize, they create a culture that determines what the company will work on, what interests people will have the most. Most importantly, whom they will hire. They hire people who are like them. They hire people who conform to their values, who think the same way, who see the world the same way. So they hired a bunch of other researchers. And I can say for sure, I have many friends working there, this company is run by researchers. The research organization of Open AI is a separate, elite organization that, in essence, sets the pace for the entire rest of the organization, the entire large corporation, although it now constitutes, perhaps, 10% of the entire company. Right. And now, if we look at the product, right, it's an experiment. ChatGPT is not a product, it's an experiment. And we are not clients of Open AI. We, if you pay them money, or even if you just use ChatGPT, you are simply providing data for this experiment. Well, and if you provide money, you also provide some other resources, right? No one wants to create the best chatbot. And what's more, the leaders of Open AI talk about it, people just don't listen to them. They say that we are not here to build chatbots and Sora, we are building artificial general intelligence. They are building artificial general intelligence. This company has stated this from its very inception. They talked about it. True, they initially intended to, let's say, open-source it, if you can call it that, right? That is, Open AI was founded as a company that would build AGI as a counterweight to Google, because Google had already, in the mid-2010s, started working on these models and so on. And several researchers realized that it would be dangerous if Google were the only company to have this AGI in its hands. And to create a counterweight, to create some counterweight accessible to other people, not Google, for this technology, they went and founded Open AI. They didn't intend to build a chatbot, they didn't intend to build a video editor, they didn't intend to release any products. It was a laboratory that was engaged in research in the field of artificial intelligence with the goal of building artificial general intelligence. They are still doing it. That is, people just don't listen to them, they talk about it, right? They don't have the task of releasing some product. And therefore, when they don't release another product, you shouldn't be surprised. Everything we see is an experiment. It's part of a big experiment. These are micro-experiments that are part of a larger overall experiment. >> Listen, I want to ask some clarifying questions here. If they built AGI, is the torch over? >> Yes, because one of the main properties is that it's self-improving. It will continue to build itself. As soon as you get a self-improving system, you don't need to deal with it anymore. It will improve itself much better than you. >> Well, they will continue with other research, they will build, there are new three letters ASI, all sorts of things, and so on. Or are they real, it doesn't matter, or what will happen? Will it be a product? Will AI become a product for people? If they are building AGI, are they doing research, or since they are a research organization, will it be the same research product? >> Electricity is a product. >> Well, it's unclear. Unclear. >> Yes. It has penetrated so deeply, it has penetrated the very essence of creation at this moment for us, right, that we can't imagine life without it. That is, it's so pervasive that if you lose electricity at home, you don't know what to do with yourself, right? You can't function, you can't cook food, you probably can't do your usual work, right, for most people. Right. And, uh, AGI will likely occupy a very similar place in our world. That is, it will be a thing that allows us to solve certain problems, right, but it will penetrate the very essence of the world around us so much that we won't even think of it as a product. Listen, very interesting, and if we go back a little to Google, you've grasped a very cool point. We'll move on to Gemini now, but I still want to talk about Gmail. >> You've grasped a very interesting point. You say, look, there's an amazing product, just incredibly cool. In terms of global usage, right, and at the same time, they couldn't make something more out of it. For example, they couldn't make it into some kind of infrastructure, I don't know, like WeChat did, right, or they couldn't launch a messenger next to it, or they couldn't even launch video conferences next to it. It's still a separate application. That is, they couldn't do anything based on this application, right? It remains this old, incomprehensible application. I myself have my main inbox on Google, although I use Apple's interface. And I don't like email interfaces at all. For me, it feels like they haven't changed since I got my first email in the nineties. So for me, once again, you've articulated it clearly now, probably. Did Gmail have such a task? >> Yes. No, again, Google's teams always have technological tasks. When they decided, Gmail was also an experiment, but it was a technological, not a scientific experiment. The idea was, if you remember, before, if people here are as old as I am, and are watching this, then they probably remember that before Gmail, a free email account usually, well, you could save about 25 MB, for example, >> or 30 MB of mail, right, and Gmail was a technological experiment. Is it possible to create a free email account based on Google's infrastructure that would allow people to never delete emails? They initially offered 1 GB of space, which at the time seemed like just some gigantic, gigantic space. People weren't sending 100 MB email attachments, videos, etc. The internet wasn't that fast yet, but then they added 2 GB, and eventually they made it 10 GB. Now you can even expand the size of this mailbox for money. That was the essence of the experiment. To create an email mailbox that would allow you to never delete emails. This experiment was completely successful. Again, no one at Google intended to build any products. It was a technological task to build a scalable, very well-scaling technological solution that would allow people to never delete emails. And because of this, some interface capabilities were built around it. Then, good spam filtering was added on top, and that's it. Gmail essentially killed innovation within email because they made such a good solution, right, >> that it was impossible to compete with. >> They solved the main problem, it turns out, they solved the main problem of a person. >> They solved their main task, but at the same time killed Gmail because of it. Then they killed, killed innovation within email, because Gmail is a good monopoly, a powerful, strong, good monopoly. But since they made the product free, right, it turned out to be very difficult to make money from it. This monopoly turned out to be a very strange monopoly that creates a lot of value but doesn't capture any of it. What is a company's task? To create some value, and then part of that value should remain within the company. And, accordingly, you multiply the value by the amount that remained within the company, which the company calls, well, in English it's called captures, meaning it managed to capture some part of this value, and your company's success, the company's size can be estimated by this value multiplied by the percentage it managed to capture. Gmail created a lot of value, captured the entire market, right? And at the same time, it couldn't capture any of this value. Essentially, it's very, very close to zero, a very small share, right? >> That is, no profit, right? But why? Because they couldn't create a machine for capturing this value. If you look, for example, at search, then the situation was completely the opposite. That is, Google indeed built a very, very valuable product, but then with the help of advertising, by copying AdWords, right, then by copying other advertising products, they were able to capture this value of search, the value they created. And because of this, Google became such a valuable company, right? With Gmail, it didn't work out, with Maps, it didn't work out, because, yes, in fact, with search, it was just a coincidence. That is, when Larry and Sergey decided to build a company around search, their task was to build the best search. They didn't think about how they would make money from it. They said it out loud, that they didn't know, they weren't thinking about it then. They would create something very valuable and then see. And this idea remained in Google's blood, that let's create something technologically valuable, very technological, and then see. Right. And, well, in a sense, Open AI followed Google's path. Well, indeed, five Googlers founded the company, right, these are people who left Google, who created a company with a similar culture in some sense, right, but focused not on technological experiments, but on rather scientific experiments. These are people who were primarily focused on working with RL, Reinforcement Learning. And they looked at the task a little differently. Now, of course, AI is also a very complex technological task, right? But the point of the whole story is that neither Google, nor ChatGPT, nor Open AI are product companies, they don't build products. First, it's a technological company. Google was built around a technological task, namely, to index the entire internet and then run the PageRank algorithm through it. At that time, this task was not being solved, meaning there was no technological solution for it. Google assembled a large number of very good infrastructure engineers, and they were able to solve this task. And from this, the DNA of what we know and love as Google today was born. And ChatGPT is a research laboratory that continues to be engaged in one big experiment and periodically releases mini-experiments, which for some reason everyone tries to dress up in the guise of products, and everyone sees them as products. Well, it's just an experiment. >> Okay. And if we take Gemini? Is Gemini just a technological solution, or is it a product? They took it, they made, for example, Gemini, and around it, it feels like models, for example, like the video generation model Veo, or like the image creation model Imagen, or like the product NotebookLM, where you have the ability to upload a huge amount of different data, and work within AI. Is this built around Gemini, or is it again just different technological tasks? And that's why the interface is so clunky in this regard. Well, Google's interface has always been clunky, so to speak. That is, I don't know a single Google product with a good interface, but that's just not their focus. This is not their area, this is an area where Apple excels very well. That is, graphical interfaces, of course, Usability at Apple is designed very well. And in other companies too. But it's not about Google. And the whole story with Gemini and related products is all a reaction to an external stimulus in the form of ChatGPT. Yes, Google had all the technology. Google had people who wanted to work on all of this 10 years ago, right? And many people were upset that Google wasn't working on these things, and they left, including for Open AI, including for Anthropic, right? And what is happening inside Google now is primarily a reaction to the popularity of these AI experiments, AI products outside of Google. And Google realized that, in fact, well, frankly, if you look closely, and squint a little, what is ChatGPT? It's just the next iteration of search. It's just how people will search for information in the future. So even if we don't achieve AGI, the amount of value created is gigantic. Instead of clicking on links and integrating them into some understanding of the problem, you can just start asking questions about the problem, as if you have an intern who has already done all this work and is ready to answer any questions. And as soon as ChatGPT, for example, can integrate advertising effectively, and this will happen sooner or later, of course, it's even starting to happen gradually. So, Google might start having problems with their core product, which generates the most money, right? And this is primarily a reaction. That's why Google quickly started integrating AI answers into search. Maybe earlier than they should have, maybe later than they should have, I don't know. But one of the first integrations was precisely this integration. And what Open AI and Anthropic should primarily fear is primarily integration into search. Because we are talking about literally billions of queries a day. And on the largest search engine in the world. >> And we are talking about the largest entry point that people are used to. >> This is the largest technological product in the world, >> if you think about it, right? There's nothing bigger. That is, people, probably, well, with operating systems, interfaces, the operating system, right? >> Well, yes. But no one thinks about the operating system. No one thinks, "I'm going to boot into Windows now." Everyone just thinks, "I'll turn on the computer now." Yes, this has also become a background element. Microsoft, of course, missed a lot of opportunities in this regard, but that's a completely different story. But it is Google that is reacting to, essentially, the threat to their core business, right? That is, search. So ChatGPT has indeed removed the need for searching. Listen, where are they going? Look, they've responded to the threat, they've launched a cosmic product. In terms of quality, right? They didn't lag behind and didn't concede to Open AI. Okay, they've already lost a big chunk in terms of weekly usage, although we understand that they still have search. >> They still have search. And Open AI hasn't yet taken a huge share of search, advertising, all of that. Where are they going, Gemini, where are they going now? Are they also just doing something? >> No, well, everyone believes in artificial general intelligence. Google also believes in it very strongly. And if anyone has a chance, in my opinion, it's one of these two players. So why Open AI? Because they have Sam Altman. Their secret weapon is Sam Altman. Someone told me a story about him, that Paul Graham once said, one of the smartest people I've ever talked to is Paul Graham. He's the guy who founded Y Combinator. Right. And Sam Altman worked for him for some time at Y Combinator, and he said that this is a person who, if he lands on an island with some natives who don't even speak English, within a week he will be their leader. Within a week, right? This is a person who can sell snow to Eskimos. This is a person who can sell water to people sitting in boats on the Amazon. In short, this is their secret weapon. And thanks to Sam Altman, they might raise enough resources to move forward, because at the moment it's a pyramid, essentially, right? So both are demonstrating some results, not financial ones. Mostly, the results are primarily technological and ideological. Right. And then, based on these results, they raise the next round, the next amount of money. And Google, on the other hand, has a very serious business that they can rely on, right, to invest, reinvest this money in building these increasingly powerful models, right, and they also believe in artificial general intelligence. They already believe in it, right? And secondly, you can still think about how the models can synergize with the businesses that Google has. We were just talking about search, right? How much better is search made by these generative models? Search has become a much better product. That is, people want to use it more now, right, for things that people didn't use search for before. Do you mean the quality of the regular interface, or the new interface, or the old interface? >> No, well, either I can spend 20 minutes clicking on links when I'm doing some research, or I can make one query, and all my research appears right in the search panel for free and so on. And then, of course, advertising is also attached to all of this, and better than before, because I know exactly from your guiding questions, right, what you're interested in, I start to understand you much better. But if you think about synergy in other products, I was talking to a friend the other day who is involved in video generation, right, and he says that there is an incredible demand for high-quality video generation, because people who are currently making videos for platforms like TikTok, for Instagram, for YouTube, right, they see an opportunity to spend much less money on content production. Look, you and I are sitting here talking, we had to spend time and
And so on, and that is very cheap content in a sense. And if we could generate all of this, why, why, why would we sit and set up the video first, then sit and waste time? Ah, well, and Google has the largest video platform in the world, right? So imagine if they energize Veo, right, and integrate Veo into YouTube, >> that would be incredible. >> I really want to say that they are moving very fast there, and they are already pouring out a huge number of internal solutions. For example, in particular, they have started cutting videos into small clips to generate, they will definitely integrate it. The question is the amount of money they can allocate to it. And they clearly have money, they clearly have money. >> With money, with money. There's an interesting story. About 10 or 15 years ago, everyone was measuring and talking about the huge amount of cash that these big tech corporations were sitting on. Apple had something like 400 billion dollars just in cash. Well, that was, of course, a rough estimate, >> yes, these remaining amounts. It's not such a big, uh, big pool of gold coins where Zuckerberg or Larry would jump in and swim like Scrooge McDuck, right. Well, in some sense, liquid money, right, that was somehow parked. However, this was un-reinvested money. And the point was that the Valley didn't know what to do with the money. They didn't know, they had no ideas about where to build what, how to develop. And everyone at that moment looked at Amazon, which always had practically zero remaining. And everyone was surprised: "What's happening? Amazon reinvested. Amazon reinvested all that money. They built a giant logistics system. Now they are reaping the rewards of these reinvestments." Ah, for example, Apple and Google had no idea what to do with that money. Now they know. >> Now they know what to do with that money. And all these remaining amounts have gone in the right direction. Apple, of course, wasn't positioned correctly. They are not the kind of company that can build general artificial intelligence. They don't have it in their DNA. Well, we were just talking, we started talking about culture, right? These are people who don't think in terms of experiments. They don't know how to conduct experiments. Every product must be perfect at the moment of release. That's why they are so slow. There are rumors that Apple will finally release a foldable phone. Yes, we can buy the Samsung Fold 7 now. Seven generations of foldable phones have passed, and Apple has finally decided to release a perfect one. And it will be very cool, at least in terms of hardware. Yes. The foldable phone will be perfect, polished, and so on. But it's no longer an experiment. They waited until they could make a perfect product. The same was true with the iPhone. People pestered Apple for several years and said: "Please release an iPod phone." There was such a story. Someone even made funny renders. At that time, they weren't called renders, but pictures, so to speak, collages of what it might look like. If you Google it, you'll find old pictures. Yes. But Apple spent time, they had a breakthrough in touchscreen quality within the company. And then Steve Jobs said: "We will build the phone around this." They spent time, they made good software for it, and so on. But it wasn't an experiment, they want to release a finished product. And in terms of AI, it's unclear how to make a finished product out of it. What is a finished AI product? It's always probabilistic systems. They will always give an answer with some probability that is not entirely correct. And Apple cannot release such products. Everything must be deterministically 100% perfect. They cannot release a product that is not perfect from their point of view. Listen, but it turns out, well, okay, but Apple, this story with Apple Intelligence and the button on the side, and that they said that this year in March they will be super cool, they missed everything. Well, look, you are still an Android person, and I am Apple. Yes. For me, it was just, I don't know, of course, I will remain in the Apple infrastructure for now, but I definitely am. But for me, it was a bit of a strange story. They promised, promised, promised, they didn't have such a direct failure. I didn't see it. Why did it happen? They were afraid of the market. They were afraid of the market. In fact, this failure has been going on for over 10 years. They once bought a company called Siri. This company grew out of Stanford Research Institute, a corporation that was involved in artificial intelligence research a long time ago, starting in the seventies. Yes. And from this institute, the company Siri grew, and they bought it and named it their assistant. Ah, so SRI Stanford Research Institute gave Siri >> this assistant, right. And it was also a research project. The company was a research company, in a sense, it resembled OpenAI. This was before deep learning models, and therefore people tried to build everything completely differently. It didn't work very well. And I know the founders of the Siri company, and I once asked them: "Why did you release such a raw product at some point?" When Siri came out, if you remember, it was a completely useless thing, it didn't work. >> Yes. >> And I was surprised: "Why did you allow this to happen?" And the story goes like this. Apple, when they bought it, the agreement was, Steve Jobs was buying them at the time, and the agreement with him was that we will release Siri when it's ready. That is, we ourselves, the team itself, will decide when the product is ready, and then we will release it. And several years passed, the company, the company was already integrated into Apple and continued to do research internally, essentially. And at some point, Steve Jobs was gone, he was no longer there. Some big boss came to them and said: "Guys, that's it, we're launching on the next iPhone version, we're launching Siri." They say: "Well, we're not ready, this thing doesn't work." Yes, they say, we're not interested. You have about 12, 13, 14 months, right, you're working on releasing this thing as a product now. Think about what features you're launching, and so on, because we don't have a launch feature for the next iPhone. And we need a launch feature. And we decided that Siri will be the launch feature. So Apple can't. A pattern emerged that they need some kind of launch feature, that is, it's some new feature that will sell the next iPhone. Think about why I should upgrade from iPhone 16 to 17. Yes, yes, yes, 100% need a launch feature, and Apple last year, >> yes, last year it was definitely a feature of such an upgrade. It definitely was. Although this year there don't seem to be such features, but their sales have been very good this year. Yes. Question. The question is still, ah, ah, but it's their mistake, what is this bet? Here, >> they, they perfectly understood that this is a very important aspect of their future product, right? They bought it in April 2010, 15 years ago. Apple already understood the importance of artificial intelligence in its products. Yes, I'll repeat again, Apple doesn't have the cultural DNA that allows it to work on such systems. Once again, AI products, the way they work, at least the way we are building them now, are probabilistic systems, they are statistical, meaning they don't guarantee a correct answer. When you press the settings button on your iPhone, the Settings app definitely opens, right, it doesn't crash, it doesn't open another app, it will definitely open that app. And Apple is used to building exactly these kinds of products, where everything is carved in stone, so to speak. Yes, everything must work perfectly. AI is a different product, it simply won't pass internal checks. Even if one of the PMs, who, as I've already explained, essentially manage the company, asks a question, and this assistant answers it incorrectly, they won't launch it because it doesn't work perfectly. They can't put an Apple logo on a product that doesn't work very well. Unlike, for example, OpenAI, which is a research lab, they don't even have the concept of something being finished. They are working on something, right, this is their experiment. They will continue to work on it. It doesn't matter to them how well it works now. It's work in progress forever. As they used to say about Google, that everything at Google is in a state of constant beta, right? Well, it's not even beta, not even alpha, right, because alpha implies that at some point there will be a release. No, GPT is an experiment altogether. We haven't even started formulating what the product will look like. We haven't started formulating the prompts for what the product should look like. We are just experimenting. >> Does this mean that there are a huge number of projects being done in ChatGPT, even where they make agreements? >> In fact, all these agreements and partnerships can dissolve at any moment if the research suddenly goes in another direction. I don't know, then you'd have to ask lawyers, right? But, for example, if you look at how everything went with Microsoft, with whom they had a love-hate relationship, right, and honestly, I was sure that Microsoft would eat OpenAI at that moment, because Microsoft is a very serious political player, right, and they have repeatedly partnered with many companies, and then these companies dissolved. Microsoft remained in the end. You can recall at the very beginning Lotus 1-2-3, right, you can recall IBM and so on and so forth. Such giants. IBM was a giant, it was the computer corporation, right, where is IBM now, and where is Microsoft. Yes. But Microsoft didn't account for one thing, Sam Altman. Sam Altman managed to outplay them somehow. I don't even know how it happened. I'm not involved in politics, so it's hard for me to understand what really happened in that company. But Microsoft is essentially winding down its partnership with OpenAI, right, and they are no longer their main partner. And the same thing didn't work out. >> Why didn't Microsoft manage to create a model? >> That's also, well, they did something there, of course, but that's also a company that, it's not in their DNA. The company has long ago turned into a corporate software provider. Yes, they are very good at selling what they have, right, they are very good at working with corporations. They managed to convince the US Army, they convinced the US Army to buy virtual helmets. I was shocked. Especially since these helmets were absolutely useless. I used one once, and it became completely uninteresting to me. Yes, they somehow managed to convince the US Army to buy thousands of these helmets. That's their DNA, right, they don't know how to build experiments either. At the same time, Microsoft invested a lot in the development of artificial intelligence back in the 2000s. They had an office in Cambridge, and they were able to hire a significant number of people who were pioneers in machine learning. The same way Google was able to hire Hinton at one point and based on that create the first, make the first convolutional models work, right? Similarly, Microsoft hired people who laid the foundations of machine learning, on which all artificial intelligence was later built, right, but it wasn't in their, I don't remember what it was called, Microsoft Research, I think it was called the organization, right? They, I don't think they made any products out of what came out of that organization. Well, except maybe for some things for compilers. So Microsoft is not the kind of company that will, again, build experiments of this scale. They simply, if you look at their organizational structure, again, Conway's laws apply. It's a company that builds very well, very complex products, right? For example, if you look at Word or Excel, they are uneven, non-homogenized, very, very complex products. If you look at any Apple product, it will be uniform, it will be as if it consists of, it's flat, right, everything works the same everywhere, the interface is the same everywhere, everything is very, very clear for users, but the complexity of these products is not very high. If you look at Excel, then each menu is a world, right, it's a world in itself. In Microsoft, in Excel, a whole team of dozens of people is usually responsible for a submenu, who build everything separately, who practically don't talk to other teams. And that's why when you go into another menu in Word or Excel, your hair stands on end from the complexity that it all contains. But at the same time, if you want to do something with a document, Word can definitely do it. It's not guaranteed you'll find it, it's not guaranteed you'll understand how to do it. And Microsoft >> now has help part >> GT, >> yes, for this, there is definitely GPT or Copilot for this now. For this, there is Copilot, yes. Yes. >> Ah, but, uh, Microsoft's organizational structure, uh, it consists precisely of these teams, right, which are very isolated from each other, which fight for resources with each other. And that's how the products look. It's like a large number of small castles, right, like how Word is integrated into Windows, right, not at all, right, there was some story about some components that could be dragged from one application to another. How is Excel integrated with Word? Not at all. There were also some attempts to do that at some point. >> Normal. No, normal. Normal, >> yes. Normal. There is no normal integration. And Microsoft is a very political corporation, and they won't be able to build anything like that simply because of their internal organization. It's needed. Google is now conducting an experiment at the company level. It's a giant company. I spoke with a PM from Google the other day, and he is very unhappy that everything within the company is being shifted to these transformer-based models. That is, we have, Google now has a foundational model, right, it's Gemini, and everything is being shifted to it. Everything should work on top of it, right? He, for example, is very upset about the fact that Google Translate is being translated to it. >> Uh-huh. Because, well, in his opinion, it's very difficult to improve the quality of translation by working with these foundational models, because it's not the goal of the team that builds the foundational model. They have a common goal, >> and translation quality is not even their thirtieth goal, right? And now, when you become such a user, uh, who is downstream from this model, right, they've thrown it at you and you have to make it work. You don't have the ability to adjust anything and make it work, >> yes, you can, in fact, you can, on the one hand, seemingly super-improve the product, but on the other hand, you can lose the product, you can spoil it very badly. This was an interesting aspect. We updated Alex about two months ago, or how long ago, and they released AI internally to closed users and updated it. My children came literally a week later and said: "Dad, throw out this stupid system, can we revert to the old one?" Because it's slow, buggy, doesn't respond on time. I kept it purely for my own experiment, but I can say that setting an alarm has become difficult. That is, before, you say: "Set an alarm," and you instantly have an alarm. Now you say: "Set an alarm," and it might not respond, or respond later, right? This is a disaster. This is the same in translation, right? In translation, you want to get super direct, standard quality, and people are used to getting words, these examples, showing some variations that exist. Ah, yes, it lags behind the translation quality of, say, ChatGPT or within G9i, but there is still a standard interface. It's interesting how they will solve this, of course, >> yes. Well, uh, as I said, but Google is ready to put, as we say, a farm on it, right? That is, they are ready to shift the entire company to the rails of these foundational models, because this is Google. At one time, they had an idea about Moonshots, that is, they even created an organization that essentially shot for the moon, so to speak, right? That is, they built experiments that could do something big. They were not interested in experiments that could be successful with high probability, but do something small. They were interested in experiments that would do something big, but with a small probability, right, that these are risky experiments with potentially big results. Now Google is ready to risk the entire company, essentially. They are now shifting search, advertising, a bunch of other products to these foundational models. Expecting that foundational models will improve so much and that they can be launched so cheaply, that all of this will eventually turn out to be the right, correct decision in 5-10 years, right? Ah, well, and this is a question, because, for example, the cost of running queries to these models, right, is still much higher than, say, translation or search, right, the amount of money that Google is now spending on generation, on creating these generative answers in search, right, they are naturally not recouping it now. It's an investment, of course. >> It's an investment. It's still money. No one pays for it, it's still money, right? And there's also the risk that people will see fewer ads. >> They will see fewer ads where Google makes the most profit from impressions. >> Microsoft will never risk this. Apple, maybe Apple has risked it several times. It was under Steve Jobs. Tim Cook is retiring, and he is not Jobs at all. Yes. Under Steve Jobs, the company was put on the line several times. This happened several times. And each time he won. And because of this, he gained that respect and even deification that still remains a bit in the Valley. Yes. Google risked it, Google had this MacOS M or iMac moment, when they decided at some point to put the entire company on the line in connection with these models. They were reacting to an external threat, but I think they made the right decision to respond to this threat. Essentially going all in. >> And if we take Anthropic? What about Anthropic? And their culture. >> Anthropic is essentially a kidney that was moved from OpenAI, the same culture there, right, but it's the culture of early OpenAI. OpenAI's culture has changed significantly internally when a lot of money came in. And all this non-profit and all this human-oriented story was essentially thrown overboard. They still want to build this general artificial intelligence AGI, but they want to build it for a different purpose. They want to make money. Anthropic is a company founded by people who still retain some ideals, right, and they still want to build AGI to ensure the future of the human race, because it's not a given that we will be needed. Of course, this is the argument of the Luddites, right, if the audience is not familiar. A Luddite was a movement of former weavers in late 19th century Britain. At that time, the first automatic spinning wheels began to be produced, which created fabric automatically, based on power, based on steam engines, right, and the weavers rebelled. We are no longer needed. Why will people be needed if machines do everything around us? And science fiction books were even written about this, right? The world was imagined like this, that everything would be replaced by machines. The steampunk culture is from there, right? It turned out that the Luddites, the Luddites lost their war, of course, right? And we are now surrounded by machines, but people still have enough to do. Here's a modern idea, in my opinion, it's very similar to the Luddites' ideas, saying that after AGI arrives, people will become useless. I don't believe in this, right? First, I don't believe that AI will look like everyone imagines. Second, I don't believe that people will become so abruptly useless. Of course, how people spend their time has changed a lot, right? Even before ChatGPT appeared, how we spend our time, how we do our work, has changed significantly over the last 20 years. It has changed a bit again, it will change again. Undoubtedly, how we work, how we spend our free time, how we find information, how we consume it, will change again. Yes, but I wouldn't take the position of the Luddites that this should be stopped immediately. Anthropic are people who believe that AGI must be built, not in one place, but there should be sources of technological expertise in AI, there should be several of them. And at least one of them should be oriented towards the interests of humanity, and not just the interests of the capitalist system. Yes, and Peter Thiel has an excellent book called Zero to One. It's very, very popular in startup circles. And one of the main ideas of the book is that any company wants to build a monopoly in the end, that it must build a monopoly. That is the task of any company in a capitalist system, right, to build a monopoly. The idea of the people who manage Anthropic is that in the case of AGI, it will be very dangerous, that if we create such a strong monopoly in the sphere of AGI, it will be very, very dangerous for humanity. Yes. And they work primarily based on the idea that this must not be allowed. Yes. And it's still a research laboratory. It's still a set of research nerds who are primarily engaged in experiments. They don't build products. But they approach what they release a little more thoughtfully, let's put it that way, right? That is, their coding models, right, they are much more attentive to what these coding models generate. And, for example, Trust and Safety, that is, the safety aspect of what the models output, what Claude answers you, it will be stronger, because it's part of their DNA, they worry about it. >> And XAI >> to conclude, to conclude, add to the conclusion the culture of XAI. I think this is the cherry on top. It's a very interesting story. Few people know, but one of the initial investors in OpenAI was, was Musk. Musk was one of the first investors in OpenAI. And even because of this, there was a lawsuit when he was eventually kicked off the board, right? Because of this, there were even several legal disputes. Nothing came of it. And at the moment, I think that XAI is several things. First, well, he is the richest man in the world, and he wants his toy, he wants his AI, right, the best AI in the world. Yes. And if the richest man in the world wants something, he makes it. Second, we are still in a bubble, no matter how you look at it, right, we are in a big, big bubble. And he saw an opportunity there. He is a person who sees very well, he has proven it several times. He has built several fundamental companies, right, for his industries. And he saw that yes, this is indeed a big opportunity. This is a big opportunity, and he was able to raise giant money for it very easily. Yes, he was able to assemble a very strong team quite quickly, but like all other Musk companies, they are tied to his personality, and they revolve around him. That is, figuratively speaking, as I say, so it will be. If you don't like it, the door is there. And I know several people who were essentially in the top ten. Uh, such a founding team, XAI, they are all no longer working there. This speaks a little about the culture of XAI. It's an opportunistic company, right, which, essentially, if there's an opportunity, you have to take it now. They don't have any idealistic ideas and so on. These are people who try to raise the maximum amount of resources as quickly as possible and do something with them as quickly as possible, qualitatively, so that all of this can then be sold and somehow earned from it. That is, I wouldn't even call it a product. Is PayPal a product or not? Yes, it was an opportunity at the time. People saw that electronic money needed to be created. That was Musk's first company, X.com. Musk's first company was called X.com. It was merged with PayPal. Yes. And the idea was exactly that. So the internet needs electronic money. How will we do it? Yes. And it was simply an opportunity, it was a chance. Musk is an opportunist, he saw an opportunity, he built it. This is XAI, it's an opportunistic company. In my opinion, honestly, they won't build anything interesting, it seems to me, because idealists are building a grand vision around it, right? And XAI is an opportunistic company, Grok is an opportunistic model, but it doesn't stand out in any way for that reason. Plus, they are a bit behind everyone else. Well, the only opportunity they have is to integrate the model into Tesla. Yes. Well, if they start releasing robots inside robots. Tesla, how many Teslas are in the world, right? Well, some millions. And how many >> that's it? Yes, that's not billions of installations and not 2 billion users there. That's not 2 billion or 5 billion. Not yet. And it's not yet the time that a person spends all day on their phone, right? So it's not >> who has a giant opportunity, who is sitting on it, is, of course, Apple. They, at one point, significantly changed the world by making their own chip. And to this day, no one can repeat this result. These Apple chips, right, A8, and so on. And then they were able to transfer their computers to them. So Apple Apple Studio, for example, is actually the fastest computer available to users that can run a large model. If you are not an enterprise, if you don't have your own data center, and so on, the largest computer on which you can run an AI model is a large Mac Studio, because of the technical solutions they put into these chips, right, and that anyone can verify, you can install LLM Studio on Apple Studio, download quite large models, and run them very efficiently. Yes. And the fact that Apple is not using this superiority in computing, which they currently have very seriously, right, that they don't have a solution that would work precisely on the device and use these models, is simply a loss. An incredible loss for the market and for Apple, of course, too. >> Well, that's great. We'll see you all on Sunday. Be sure to leave your comments, like, support the release. Everyone, bye. Thanks for the conversation. Sasha.