There must be a lot of things to say about. It seems that by the end of the year, there will be less than four companies. Emotional intelligence. Communications are very weak. It's time for an evil, evil Alexa. It's a level, it's very much lower in efficiency on the baccalaureate. That's a very deep question. And Microsoft is the biggest company ever.
Tanya, have you ever used-- used her once? Or is it useless?
Well, we've got you in the past, I think we've shown this very strong subject, which is social engineering and how human devices work at all to get out of it. information. If you didn't, look. There's a lot of, in fact, comments from the point of view of what levels have been taken. So, in this, maybe-- maybe this is where we'll put it on level six. We've set low levels, six, seven, we'll not put it in this issue. Where do you want to start? Before we go to the emergence, well, to such a very strong launch of the new OpenAI model, I want to start this whole thing from history.
There's a Deep Research. Who doesn't know, it's a tip inside ChatGPT and, by pressing this button, the system goes on the Internet, takes five, ten, twenty minutes and does some research. Well, for example, when she came in, I said you could pick up rice for yourself, I don't know, good rice, right? Or do a study of how much traffic is worth in different countries. Well, I mean, do some research. I didn't use it much for the first weeks, but I was just passing by the time the new systems were developed and then four and a half showed up. We're gonna go over there now, right? This new edition from the new model, from OpenAI, the largest, yes, in their history. They say that at least. I wonder what Deep Research I've noticed. These days, literally. I mean, I bought a dog, and, well, it's gonna be at the time of the release. Well, maybe her picture will come up, so at the time of the release. And de-- I don't have kids knowing about it yet.
And I was looking for this dog because it's a beard, a Russian borze, a Russian dog borze. I was looking for a coach here in-- considering I live in Bay Area, this is the U.S. Silicon Valley. And I was looking for a-- how to find a coach who knows a Borgoke, working with the beasts, like dogs. And, uh, we could, uh-- yeah, it wasn't necessary for a Russian batter, but at least worked with those dogs, even though it was desirable for a Russian borze. And it's hard to find. I mean, I couldn't really find it in the search. It's not enough there's a club of beards in North California, and even one of the chairs at the club, I'm in touch with her and I've been in contact with her when the dog was chosen, it was very hard to buy. She can't find coach. Well, she found me a veterinarian, like, an hour to go for a braze, there, and so on. And I couldn't find it. I asked for a Deep Research request, "Find me a coach who was with a borze." And surprisingly, he's back in ten minutes with an answer. He gave me two options, just two options, but he found me one option I couldn't find in Google at all. He's been scattering hundreds of resources and the gift he gave it to, it was really cool. I mean, basically, what happens if we're at LLM now, in chat rooms, do some kind of request, find a good restaurant or pick up a country to go, or tell me, please, where I'm going, I don't know where I'm going to go. Barbershop or where to buy flowers. Those answers are very primitive, no details, and often so stupid. Not enough, they're not real. Well, you know, you just got some three first restaurants out there, you know, or something. And the answer was very-- well, first of all, he's been clarifying some details of what I need the coach, right? And then he's asking questions. So you ask the question, he asks questions. He... He gave me a very cool answer. So he wrote why he offered me this man. I gave it to me, I found information on it not just on his website, but I found Yelp, found me his rating, found me external reviews, found some feedback, where they wrote about the beast, wrote what they were, the beasts, and so on. I did. And made a second recommendation. She was, uh, so medium. And yesterday we checked the Deep Research search test for the exact same reason, my sister wanted some sweets that don't contain, there, sugar, etc. I mean, she's-- she's here, she lives in Europe and has some kind of restrictions. And it turns out she's in ChatGPT, and the delivery, of course, again, well, nothing. I mean, it just gives you something. And this Deep Research, again, he did a normal conclusion analysis. I mean, first of all, you realize that this isn't hallucination, but it's not a hallucination, either, because Deep Research is saying it might come up with data, but at least what I saw isn't hallucination. And it was a conclusion, well, it was clear why you were talking. I mean, of course, development, that's a move-- quality, improvement of quality, it's not real.
And, I guess if you don't use this system every day, there are people who don't use anything in the world, you probably won't even know what's going on in these systems in these models. I mean, I don't know if you're doing a request once a month or every two months, you won't even understand. And if you're doing this every day, there, using these systems for ten, twenty, thirty, then you start seeing this incredible improvement. I want to ask the audience. Write in the comments, please, what exactly are you using every day? What do you use every day? Who uses Grok, who uses GPT chat, who uses DeepSeek, who uses some, maybe a combination of applications that are the aggregates of other different systems, huh? Whoever uses and, uh, where Meta comes in, that means that stuff does. So what are you doing, what are you asking about? And then what is the amount of your requests? Write some examples. I would like to discuss this subject in the comments. It's important to see what kind of people use. If you don't use anything or come in once a month, write it too. It's very useful. Please do this. Even in terms of discussion and examples, and conclusions. Well, we'll get a little more understanding of what we're going through, yeah, and what to tell. What do you think from the point of view of this improvement of quality, huh? Again, before-- we're now, before we're four and a half, there are some of our new features.
Yeah. There's a moment, you know, Sasha. I'll tell you right now, yeah, I've been seeing a very strong separation, that we're already with AI, and we've quickly slide this wave of what, you know, there's no technology that's improving, yeah, and it seems that it's... Stop. I think we've already done this, and we've seen advanced technologies, like deep research, that are used by business executives, business professionals. And we see all the Consymer applications, they're moving to deepseek because they're cheaper. I mean, I think it's, I think, super good for the market, because it shows that the market is a lot of stuff and you have some ways of developing. It used to feel like OpenAI was the whole market and everything, so that's it. Well, it's kind of better in the code, something else. I think we're seeing a better differentiation now. And when a clearer differentiation is visible, I think it's a way to show, in a way, how much room for further improvement, right? It just seemed like, you know, all the Indians are indiphpherical, like, and everybody just uses GPT chat 'cause they're GPT chatting. And now it's clear that different skies are different. I think it's super positive.
I have, you know, I have a feeling that these simple things I've been bringing up now, but it's... it's clear that I used deep research in my business requests, you know, in my business, I'm not sure how to do this. But the fact that I did it was that I came to the point that I had a very strong feeling of deep research in the usual household stuff. I told my wife, literally yesterday, that you could try some things, press this button down there, and somehow it would help you shape something further. I realized that, with the capacity-building, they'd probably open up the data to ordinary people and forgive. Just that we see that GPT is the last version of four and a half million tokens worth $775, right? And, accordingly, you have $150 there, a million weekend tokens. So it's you, you know, almost 100 times more expensive than 4O. Well, it's a parole or what? Well, it's a lot more expensive. I mean, it's not for simple tests. So no one now, in fact, will use these systems in the API in the creation of AI as assistant. Well, I mean, AI's an assistant or connected to this system, just generating an endless bunch of content. That's not gonna happen, because otherwise you're gonna spend a thousand dollars, right? So, the data are very divided.
For that deepseek. It's just that everything's going a little bit.
Yeah. Or he'll go to deepseek, right? But, from the point of view, again, deepseek volume, look, I'm gonna assume that to start a deepseek that's gonna be looking for 20 minutes, it's a lot of money, too. Up-- right?
Well, Perplexity, as far as I can see, is worth it.
I understand that my deep research was on the basis of deep seek, if I remember correctly.
Yeah, but you have to spend a lot of money anyway. I mean, it still depends. You're really gonna do some simple things and watch them. I'm just saying, well, that's, uh, that's a hearing, but it's just, uh, Sasha's talking about one division, one type of separation right now. I think there's another division in terms of different, maybe quality standards, I don't know, or some approaches or details. What do you trust more in terms of potentially things? Here. Although again, I don't know how deep research works, well, or technology like that, if it's done on, deep seek, how cheap it is, how much it's gonna be as good as it's as good as it's as good as it's as good as it's. I'm not even sure. Because the answer was here. That's what's actually going on. Again, I remember Lukeun, who says that in two or three years we'll forget everything that LLM is and forget these approaches. I mean, what I saw personally looking for this dog coach, I'm honestly saying, it was, I guess, what I want in general, what kind of information I want, I wish I could get before. I just couldn't, yeah. I'm gonna have to put some snacks on it, re-examine, check it out, get on the Internet, do a lot of things. Plus it's been here a long time. Like you should wait, there, parallel, in the parallel screen, to do it. Then you come in, you see, he returned this information to you, returned it. You're gonna have to remember that. And so on. I understand that this will accelerate, right? I mean, it's doing 20 minutes now. I'm sure it's about the version I pay for $200. It's probably a quick topic. Remember, there was a story that OpenAI ChatGPT could cost $2000 a month. I think Sasha, you thought about and made that kind of a straight-up bet a year ago, right? Well, obviously, I want it to be done all the time. It's just not gonna cost $200, but it's $2000 a month, like, or a thousand dollars a month. It'll be totally different, different type. I was just gonna say I'm gonna pay $2000 for that. I thought it was a fucking $24,000 a year, like $20,000 a year, maybe it's kind of a little bit of a little bit of a little bit of a little bit of a little bit of a nuisance. I mean, now, $200 is $20,000, $400 a year. Okay. Yeah, you can have an assistant alive. You're sitting there, you think. For home requests, $24, thousands of dollars a year, it's kind of something. At the same time, you understand the acceleration, right? But I'm here to remind my sister again, who says, "Look, I'm using ChatGPT just crazy, making a lot of conclusions." But she's taking a lot of things from me, too, learning about these things. But he says, "I don't see anything that I'm getting less work." Says, "I used to spend the whole day writing a free-body pip in rallys. I'm going to the airport now, I'm gonna talk them out, they're automatically formed, and then I come and spend the whole day on something else." Though, the idea was that you should've been out of work, right? You're gonna save time, you should've never worked. No way! You're going to a bigger job.
So, what's the point, Sasha, with the movie the answer was finally right? Did you get him to sleep or find out what it was?
I wrote him, he wrote. No, he's real. It's a real subject, a real story. Yeah, that's totally adequate. I'm telling you, it's a uniquely suited to me. I mean, I'm doing a lot of things. Even yesterday, we went to San Francisco for dinner and we wanted to have coffee there, right? Give me the coffee. It's trash again. It's just that the series found a list, filtered, issued.
Well, that's not a real answer. The answer is very alive, yes, very alive. And again, before we go to the new GPT model, we have to talk about it today. I think that if you say Sasha. The news just came that Meta wanted to release Meta AI as a separate annex. What are we doing? It's a very logical step, isn't it? Because no matter how many Zuckerbergs say he has the largest number of users who, well, are looking inside whatsApp, in Instagram, on Facebook and so on. You're clearly gonna lose the market if you don't have a separate infrastructure large. It's clear that there's no question inside your apps that everyone will use your AI. Well, logically, whatsApp will be the Metow AI, yes, inside Facebook Metow AI, but playing to compete. And last time we said that victory is not clear at all now, except for GPT, like a side operator, not like Google and Apple. What do you think of a meth?
Yeah, I guess there's a lot of things to say about. In the plan, DeepSeek has released very much technically, technically repositories, and it shows that their level of work is very deep. And actually, what they were able to do is train the Zero model directly with reinforcement learning without supervised fine-tuning is also very interesting. I mean, they made a number of improvements, both in efficiency and in science, right? That no one's done before. That's impressive. And for Meta, it seems to me that it is now very important to move on to some interesting decisions, for, like, conscripts. I think it's gonna be a year when we see a very strong differentiation. I think there's gonna be very strong companies that get inference on some, you know, untouched irons like ej devices, phones, etc. I mean, I think so many players, you know, like players on the foundational market actually gets a little smaller. DeepSeek is not a new player. I just wanted to, yeah, say. So they had a coding model 7B used in the top start-ups there a year ago. I mean, I think we understand that Mistral is not gonna be a full foundation company. They can be a sovereign company, yes, the European Union, but they can't be a foundational, just like Here. And it will, I think, Meta will need to find applications that are better suited to their objectives. And in particular, the practice of super big models, like Zuckerberg wanted, is not in line with their objectives. And I think we'll see more press on Zuckerberg, right? Because, why do you train the model, there's a trillion of parameters? Why? I mean, they won't be servouring her in India. I mean, they can't do it. It's not gonna be a million of them. The three-billion-dollar model won't work. The goal is to get the most popular AI assistant, it is achievable. But it will be with a model of seventy billion parameters, with a distilled, yet another mixer of experts, and where the total number of seventy billion parameters is. So I'm saying that all of this is that you're gonna have to know what I think. I mean, just have a general point that we'll have the best foundational model. I think that by the end of the year, there will be less than four companies, probably less than three. And I wonder if OpenAI will be one of them or not? That's interesting, too.
That's the target. I think that--
is a specific way out of OpenAI.
Sasha. Moreover, they had one news on our news, they had written that they had a target and as a gold standard, they had the OpenAI GPT models. GPT 4o, they had to overwhelm it, and that was like a target for, uh, reservers. I mean, they even in their internal documents, they point to the main competitors as the main goal to be achieved.
That's interesting. What's going to happen? They're the ones who'll give the same thing that everyone has a part or something. Do you think that from the point of view of Meta AI, they can do something serious, or are they actually just gonna roll the line? And if they roll the line, why are they taking it out for so long? Why are they gonna stop her now, so she's not gonna get kicked out fast?
Meta is a very difficult thing to say. I think it's for them. I think it's for them, I hope I hope I hope it's again, right? I mean, they focus on building an open source of infrastructure. In that regard, the food companies still have a need to make Euluishen. They have, they have a job to do the closet and so on. And, well, I mean, better than a long chain still isn't. Well, that's pretty sad. And by the way, esteemed viewers, if there's anything better than a long chain than you use in a model-show shopping, please let me know. Here. But I think there's nothing to report. I've got nothing on my experience. And there is. And there, because it's not as clear as it is how to make money. I mean, it's just a foundational like a layer, yeah, for a business. And I wanted Meta to do more innovation at this level.
I mean, again, they could have made some agents' interactions, and they could be a very leader.
Look, you said something interesting. Again, before we move to a new model from OpenAI and an update, which is clearly now, of course, everyone will have a little more weekly or every two weeks of ChatGPT, and many updates are going on. What will be different this year will be a great deal, including devices. And you said the devices, some devices can be added. Quick bet. And I want to make sure you're here. What's the box? If you all remember, the audience will remember everything. A year ago, I, or more than a year ago, bought a pendant from a company. Then they changed the brand, it was called Limitless AI. Then they sent me, "I'm sorry, we all remember, we'll send you everything." I'm getting a reference here a week ago. I just came in parallel on this reference.
I just got in parallel on this reference, I'm gonna make it all right, I'm gonna pay. And it says, "Pay the delivery of the fist." I thought, I mean, I paid him a year ago, and now I have to pay for the delivery, and I might be four years later, or what? But I think, okay, so, you know, you know, you know, you should just keep it up. And they write that they're going to have shipping in the first quarter of the twenty-fifth year, which is the most innovative in all. The world's most, then wearable AI. So, what's there, this little thing that you're... you're hanging out, walking, it's not just-- it was originally planned that this thing that keeps data. So you go with the pendant, he keeps all the data, and then he's sort of analysing them. I mean, I was curious. I mean, when I spend a lot of different things, there, meetings or discussions, details, and they kind of made it with something, like, a little talk, supposedly. Although I don't know how she's gonna do what she's gonna do, yeah, she can hear. We'll see. We, I-- we're gonna have to make a reference here, as it is. And they have a slogan like that: Let AI summarize your life, right? So let AI, uh, make a summary of your life. Sasha, what do you mean by the appearance of some kind of extra, uh, devices, if I heard correctly, if you said that?
If so, what type?
Additional devices in the context, and I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, I, uh, have said more about the interesting use, rather, yeah. I mean, I think that--
Yeah, about the application, the use.
I mean, it seems to me, yes, that it would just be a lot more differentiation. Iia, to be honest, if we still have a software level that doesn't have that clear differentiation, it's very, very early for hardware. And it makes sense that we are in a situation now, in my view, that is very logical that we are in a situation where Limitless is gradually closing. They're running out of leadership. And that Humane sold. That's very logical. I mean, it's, uh, it's really hard to get ahead of, uh, uh, oper-- well, just, they're ahead of their time, yeah. I mean, it's very talented, maybe one of the most talented entrepreneurs. Consumer tech of our decade, probably both. But they've taken over an ambitious task that hasn't been easy for. But both super-talents.
Yeah, that's great! Because release this device and so much money to attract and make it worker, sell it to consumers, do all these things, though. There's a Pinay in there, and they're not selling, they're shutting him down on a parole, right? Or who they are, I don't remember who bought them, he'll be developing it--
HP's on the way, yes.
Will he develop it, huh? And, well, or HP's just gonna die there, uh, somewhere on some shelf, in some project. I just wonder if such things would be, you know, big companies, right? I think that's a thing like a corn that writes and so on, so maybe oura should do, right? If everyone wears their rings, she probably should-- they should probably have done, uh, a story that keeps, remembers, I don't know, information.
Although it might be technologically, it's possible. Okay, let's move on to the appearance, uh, the biggest, so powerful apdite, at least what OpenAI says. On February 21, at the end, just a few days ago, the OpenAI released a new GPT model 4.5 a week ago. She has a name-- on O, yeah, what's it called?
They're actually called Orion.
Orion, yes. She has the name Orion, and, uh, this is a very interesting addition, including today ' s discussion, that the main focus they place on this model, at least in their description, in their website description and that, They say that, that this model offers better opportunities to interact with the user. So they're saying it's not a story, like, that's better in math, or better in programming, but it's a topic of better interaction with the user. I mean, I mean, most of the baccoons and comparisons that are often made, they're made on a basis that's smarter in terms of, uh, programming, doing some scientific research. A lot of these tests. And this is a very high rate on, uh, simple, how, human, right? I mean, improved-
Emotional intelligence.
But they say, yes, that's the improved nature of communication, right? I have this model, available. I'm just-- it's hard for me to draw conclusions. We'll make different conclusions. Why is it hard? Not because I don't know, I didn't spend time with that. Because I need a little, a little, like, some time interval, to be a few weeks or a month to test different case files. Because those cabs I tested are hard for me to say that it's, yes, how much it's improved. Because I saw people writing that it was kind of cooler, like a book. But this, again, is not just a user-friend relationship.
So they're still talking about their knowledge base, that they have a decline in the number of hallucinations, that they have a high emotional intelligence, there, creativity, etc. What about this model? Is that a routine update? Do you think it's a routine update or is it a break worth 4.5? - I'm not...
No, I think the communication is very weak. It's time for an evil, evil Alex. Anyway, well, communication is very weak. What did he say? He, he, "We get criticism that we have many models, people don't understand how to use it. Now we've produced one model for everything, and the Consymers are clearer than to use." And it's like the biggest model there, but it's not the best. What's the argument? What's the difference? You-- if you say consuit, the Consymers like, why say that's the biggest model? That, well, if you're talking the biggest, you're gonna put these down with meters, there's some new technological thesis. He's blocking everything in one. It's just that even OpenAI is hard to stay, it's hard to get up to the point. I mean, just, well, something's ready, something needs to be explained to the world. Well, you should get a release, you know, they're getting a release. I think the standard corporate process started, right?
So you're considering this model just like they started shooting O1 mini after 4O, then O1, O1 pro, O3, O3 mini-hyi. Then they said that O3 would not be fully released. So then they said they'd put GPT 4.5. I mean, 4.5 is just something. I never understood what was important.
An important team.
They also put it out, they put the button out, and yes. And they also have a Deep Research button. I still don't understand if I press Deep Research in O, in 4O, what's the difference between Deep Research in O1 and Deep Research O1? Is there a difference? I can honestly tell you, I have everything, you know, about O1. I still don't understand, the difference with O1 is either. And they, Sasha, what you're saying now is all in one, I think it's just another model for me.
I don't understand what it's like to say.
Oh, and I do. I mean, it's still unexplained. And what I was trying to say is that you're looking, that's how corporations are, right? There are different work streams. OpenAI often says that, yes, we have, there, well, there, up to fifteen people working on something. They write people there often, in their releases, it's up to 15 people. Company, well, there's a company of six hundred engineers. I'm just saying, a little of my head, two hundred of them we'll never know because it's either infrastructure or a bet of that bee. Well, we're just saying that they're not gonna exist. There's four hundred left. And, well, there's a management, another fifty, and there's three hundred and fifty. And, uh, speaking of the way, when we say, and that is, they're organized, let's say, ten, fifteen, twenty teams, ten, fifteen, twenty, right? Well, let's say it. And of course, these teams, they all work on their direction, but at some point, some team is getting back. But you can't throw a table, right? I mean, well, that's what they are. That's why we just need to release.
You think something's up?
I think GPT 4.5 is really lame, yeah.
From what?
On the level, it's very much lower in efficiency on the baccalaureate.
What's she got behind? She kind of got back...
From time to time, she's back from time to time.
Yeah, but it's something medium. No, wait, four, what's she missing? She misses, like, good rizoning inside or something?
She doesn't have risoning. Rizoning is a bigger increase than the increase in the volumes it learns.
Yeah. I'm saying she's missing a good rizoning. Is that right?
Well, look, rizoning is just a process that's superficial to other models, like Sasha says there's a lange lane, a tox tip, that's, like, a consistent thinking, a few requests that Sash says, We're having a 4.5, it's a model, as they say, it's the last model that's not rizoning. So they started training her once, a long time ago, on a huge set of data, and they spent a lot of money on it. Finally, they've trained it all. But at this point, it has been shown that with the rizoning, much better results can be achieved. And that's why they kind of did it, they spent a lot of money, but it's worse than it was with the help--
In fact, they put a type of pdddith on 4O, and the world has already slipped away in the last six months. And if the pddette was to be put in, then 4.5 should have kept all the O1s, for example, and had them inside the rizoning, switched on when the rhythms had to be added or made very fast there and so. Go on. They didn't, and they also did, $4.5 worth $75 per million token.
That's why you don't know why to use it, right?
And that's because the model is huge, it's monster-ass, big. I mean, well, we don't know how many parameters there are, but it feels like there's more than four. And, respectively, when you are
lies. And, accordingly, when you run it through reasoning, you'll have plenty, and many times this monster will work. It's very expensive.
Yeah, they have a comment that the model has a better knowledge of the world, and that allows it to provide more precise and informative responses to a wide range of questions. But what, what, what, is that so much knowledge of the world?
That's a very general phrase, yes.
Yeah, yeah. What is this?
And coming back, Sasha, to your dressing room about, remember, you started where you were from? That the baccalaureates, many of them are tied up, there in math, to programming. It's just easier to check. You always have the right answer at that end.
Yes!
And when we talk about emotional intelligence--
AI.
Peace.
Right here, yeah, well, here the guys are, they're probably just, you know, they didn't win. Then we went to some subjective area where no one can check you out, and they won't grab your hand, say, "We're better here." No one's gonna fight you, but--
But there's not enough. Yeah. Here, but there's a commercial-market in the Consymer market, and there's actually a lack of someone who's really expanding now, well, with big leaders, something to use mass-use with some adequate stuff. Not just like Apple made that button that's in, pressed and photographed. You don't understand what the script is to do. I mean, you're like you're not doing a user scenario. Or we've got four and a half of them out here. You don't understand, so I use it, or I-- I'm sitting here, using it, I'm gonna use it, I'm gonna use it, O1 Pro, I'm gonna use O3? So what do I end up using? I'm gonna get Deep Research on the button, I'm gonna get Search on the button? So what's the end?
Yeah, I mean, he, you know, he has something he says, but, but he's not logic anymore. But in principle, all corporations, right? I mean, you just, you know, you're not mo-- you've got too many priorities, too many different promises, and you're just trying to get some designer to pack up. But they get clowns usually.
And the OpenAI, well, this is the first of their anonosment, which is unrealistic, just unrecognizable.
Illar, listen, you think the baccalaureate can be built not only on these things, very objective, yes, scientific, clear, reliable, but on the basis of some sort of analysis, you know, another one.
Well, as long as it's almost all built so you have to have an answer, right? I mean, you have to understand whether he's right or wrong. And you either need a model where you can check all this, yeah, that's how it was with chess. I mean, it's a parole, chess, ho, any game, yeah, we just take, play a party. We won it, we're good. We lost, we're there, we, we're, we're, like, wrongly going somewhere, looking for where we should have been strengthened. Either there are specific examples of programming. The task has been solved or not solved. And then we know, the best answer or the bad answer. All that subjective stuff is, we're still learning the model to do what they like. So you still have this subjectiveism, he's going to go out there... Now, it's stuck. On the side, he'll go to the side of the OpenAI markers. So how they've been tagging the data, like they said, that's the answer that's more emotionally good, that's the answer that's less emotional.
You still have a right answer, and that's where the model will be built.
Look, if I move to, I want to touch the block today on Microsoft. Yeah. And Microsoft, well, they have some of these, which means they have some, you know, to show the world. First, they're planning on closing, right? And they've already announced it. I was the first time I set Skype for me in two thousand four years, and I even had a French name because my French boss registered it and wrote it. Alexandre to French manners. Yeah. And I still use Skype. I'm talking to one American, point, English language training, Skype speech. There's no other Skype left. I think they have some very important news. I don't know, maybe we'll devote a few separate time to her. If you remember, we were talking about the most important trends that move from the twenty-fourth year to the twenty-fifth. One of the trends was the development of quantum calculations.
Microsoft, including now, has a part of the achievement. But I want to go to Microsoft, what kind of question I wanted to ask. Microsoft will build now, do something separate, serious? That's when we're called ChatGPT or this Meta says, "I'm gonna let Meta AI out." Or there's a Grok separately, so they'll be single out, developed really under their brand, under their brand, going to the Microsoft brand. They're the Bings, they're doing it. They've got a lot of things. Will they be producing such a separate consumerian thing on the mass, on the whole market and make a big brand out of it?
Uh, for Microsoft, yeah. I mean, well, they're just like a phase right now, I think... That's a very deep question. And Microsoft is like the biggest company ever. I mean, it's clear that they tried to make Microsoft AI under Mustafa after the acquisition, and just like the person who's been dealing with it, I think it's about a year, right? So there's a period up to a year that people in corporations are ready to wait for the company that bought them to sit on its own hands, right? But then we need to get the delivery. And from Microsoft, I feel like zero, right down zero. And I think Microsoft, Microsoft, will have to do something.
Well, they're gonna make some kind of decision, some kind of independent decision? What are they gonna do? They're in the process of integrating with different systems and... Or don't you understand yet? Well, because they're gonna have to, obviously, they're gonna have to do something really cool with theirs. Well, with their infrastructure, well, it's definitely gonna be coming. And if they're late, if they're late, they don't blow the market. That's a story with a skip. Well, it's just that the skip had a lot of opportunities, right? Just a crazy number of possibilities. And I still don't understand the Microsoft strategy from the point of view, they're probably going to Microsoft Teams and so on, but it's just that, and it's a serious product, it's very much used by everything. But the story with the skip they've been blowing up the massage market, for example.
You know what's too hard to say in the way that I thought they were screwing up, too. But now that I'm looking at Applovin, who built a successful ads-net without their applications, that's, on the contrary, they decided to focus on one thing and, you know, sell their entire market, all their business. Annexes. That's very, very interesting.
They invested ten billion in OpenAI. Maybe they think they're in this market already, what they've already come to? I don't know. On one side. On the other hand, I'm all just gonna have to be kidding. It's an old joke. There's a God that created all people differently, and Colt made them the same. Well, like DeepSeek came and all the company's reservoir plus-minus made the same. So if Microsoft has a desire to market some product and has an understanding of what product can be demanded, the foundation of the model may not be so much needed to be in it. - I'm not. Besides, there's an OpenAI. And how they're working together, not very clear, really. I mean, maybe they don't need to come into the foundation at all.
Look, well, from a mass market point, even an announcement, Stargate and Stargate, there is no Microsoft, but there is an independent OpenAI company. There is an independent Oracle, independent, still present, which is in large quantities of products, including the Competitive Microsoft. ARM is a separate company, a separate company. I mean, there are companies that are listed independently in terms of funding, but we still don't see much of the Microsoft statements in terms of their financing, do we? And such space-based contributions to development. So that's where Meta says a lot. Google, I think this is a very high priority. Microsoft's attention is also, in fact, being paid. There's no question. Before, I think, for the twenty-second year, Bill Gates was actively claiming that AI would be super evolved. I was just watching some interviews with Bill Gates now, where he was asked so much. You were talking about AI, how do you feel about it? Well, it's all in there, I think, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really,
really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, Because Microsoft still has a team running, a lot of things to do. So it was definitely impossible to get anything from him and no comment to see. We'll watch what happens to Microsoft. Especially I think it's based on Meta AI statements. Very interesting. Please write in the comments, what do you think from the perspective of the players in Microsoft who uses Apple, who uses Windows? And I think with the existence, we'll still remember the treatment a year ago, there's a lot of discussion that would show up on the descope that would work and process all the documents, files, a lot of stuff. The rest. So, in infrastructure, this is the combination of equipment, Apple and Microsoft software, well, they're gonna be very serious in their environments. All their graduations that were still there, they're very small-- well, it doesn't look like some miracle AI. Maybe Microsoft's done great. Ildar, remember you once told me that Microsoft was giving out some sort of stuff in terms of the computer files or not? Or they're like
Or not? Or they've been like this a while, haven't they?
They had a story when they did Copilot PC. Yeah, so when this whole hat just started with models, they said that now we're running Windows, and we're gonna have a laptop, and there's a LLM inside the laptop. And, accordingly, they can analyze what you have--
Yeah, but it's for the developers, right? That's all for the developers.
Yes! And even those Copilot PCs even got sold. But as I hear, it's not that everything works really well and there's some sort of killer chips that you have to buy them for.
Sas, what do you think from the point of view of becoming a comp, uh, some real serious decisions this year, they're waiting for something to come up with serious destop?
Oh, yeah, of course not. Of course not. Well, I mean, we, you know, have to go to the comp so that it's meaningful, you need to be super cool. I mean, you need to get a deep-screen level. You need to get deep research to understand the context of what's going on. I worked so fast. Suddenly, well, not deep research, but you need a O3-mini model that's really gonna hit you right now, and that's a visual understanding that's good at what's going on.
What year is this?
Ancestors. I think, more and more, I think nobody's even serious about it. I don't think anyone's serious about this. But...
It's actually good. Why is that? I think it's a good question, because a lot of start-ups actually started doing two more years ago in this story.
I think, well, there was a plan configuration wrong, the whole plan went to practice models.
You need to do something more specific, Sasha, to decide. Either there's a problem that's already in place, yes, that's like deep research. Without him, you had to go in there on a bunch of deposits, find some information yourself, analyze it and gather some common impression. Now you've asked, button's on, and it's been here in 20 minutes, like everything's ready. Yeah, and there's some sort of script that understands the pain that you're getting off. With the computer on the surface like it's not that much to show, well, at least I don't see. And either they need to come up with some sort of scenario that we're not using yet, but that's gonna be super comfortable.
I mean, you can't see that either. I mean, it's like an instrument, but what do they want to do? Not that I'm very clear.
But there are simple scenarios that don't work. For example, if IPhone is there or MacOS photos are coming in and start asking for some interesting queries or filters or questions, he's not doing it yet. I mean, it's all very banal and stupid, yeah. And it was a lot, I remember, fun when you were looking for fish six or seven years ago, and he found you a fish, and then it was surprising. Yeah? But now, it's kind of funny, it's kindergarten. He can't do serious things. I mean, looking for pictures. I have a lot of pictures there, and I need some kind of data search. There's no one who can make me a good, cool analyzer. Well, I don't, he might be the Apple. And even a simple case, it's not solved. Or the case is in the notes, or the case, it's really good with the post office, even with the postal, it's not resolved. Well, we all know that ChatGPT is impossible to understand, it's impossible to find. He's still not working the right search.
Yeah.
So there's no simple scenario. Though it seems like this button is new in iPhone. By the way, who has 16 iPhones and who used this button in the last six months. Have you had any scenarios where you see this button is working? Or have you heard of these scenarios where you work? Because every time I make myself, it happens to push this button, check the script, I know that some stupid interface is made uncomfortable. I mean, I'm waiting for another one, you know, not to... I'm not going to work real. I'm not doing this. I don't know, Tanya, you've been using her once or for nothing?
Oh, you're old. Sasha, have you ever used a desktop?
No, I'm old. There's no point in buying. Well, if Apple didn't explain to me what skies. Yeah, yeah, yeah. That's the same thing.
No, the only reason you can buy, then that Apple Intelligence is intact, which is now checking the proofread, refread. It's obvious that it's a little bit of a thing, it's a little bit of a thing, but the buttons... There you go. That's also an interesting story from the point of view of the real, real skies that showed. I think Apple did it, it's huge to get Apple to take a new button on his phone. What do you mean, create? They've done so many tests, scenarios. Or it's like Apple was in a hurry again. I mean, Apple probably got a four-and-a-half OpenAI with that button. And it feels like it was their fault, and they had to keep the button down, yeah, but it's better to work it out properly and then to clean up how really some know-how is now. Although maybe the market will swallow, as always, and it will be, and it will grow up.
And everybody just thought that someone didn't do it, then did. It's all good, as usual, in a big world, in the economy, in politics, in big companies, yeah. Illnar, last time we started a new shirt. We still don't know the name. What do we have today?
First, I want to separate the comment under the previous video about this story. If anyone didn't see, I told you where Grok came from, why is it called a model from Elon Musk? And we have a great comment on the video. A whole story about Alice in the country of miracles about how our viewer and his kid taught poems and how it went. I read it with skill. And thank you so much for all those who write such comments. It was nice to see. And today I want to touch the subject in our shirt with the license stories that are for different products. Now we say sometimes, yes, there is, uh, closed compa-- closed products like OpenAI, and there's open open open source of the model. This is DeepSeek, Lama. And we're all gonna have to go to a lot of shit, right? Which is like two sides: there's a closed, there's an open source. And all that open source is the same. Everything that's closed is the same. And here I want to remember, uh, the quote from the movie "Assing ghosts." I was really into him when I was a kid, and there was a funny quote that shirts could not be divided just into clean and dirty. There are a lot of intermediary layers between them, such as ten minutes on the balcony, and you can already wear.
Same with licenses, right?
There's no room or open. There are a great number of intermediate steps between them. Let me, uh, whip the screen. We've already talked about the LM Arena service, the chat-bot of the Arena, and it's got to be seen, uh, what licenses are. First, I remember, yes, it's, uh, an arena that competes with different models. They have places they're taking, some glasses they have. And the most interesting thing for today's graduation is what's on the right. Look, different models have different licenses. Let's start with the pro-Very models. What do you mean, a pro-prietary? It's just a closed model. You've bought access to OpenAI chat-bot, you can use it for personal purposes, but you can't sell it, you can't make a commercial product on its basis, and you can't actually just, underneath it. I'm not paying for the OpenAI. So you're just using it, buying the right to use this product. This is for closed models. You're the one who's the most important. If we look at DeepSeek, just what we call open source models, it's MIT license, it's the most accessible MIT license. It's when you can do anything to do with this product. You can make a commercial product out of it, you can change something on it and not show anyone you've changed and, in fact, you're gonna keep using it. You can make money on that. The only restriction you have on you is that you just have the copyright to preserve, yes, to say that you use the model that was designed by someone. And if you look further, the whole zoo starts. DeepSeek R1 is licensed by MIT. V3 is under DeepSeek license. It's also some extra set of restrictions that you're being placed on. Uh, if we look at, like, the next popular model, we have Lama. Lama is also one of the open source models, on the other hand, under Creative Common Licenses. What does that mean? It means you can use this model freely for your own non-profit purposes. It will not work for commercial purposes. If you start doing some product, earning money on it and using that, uh, product, that's how you're gonna have to negotiate with the company that's making these models. And if we see, we still have licenses separate from Qwen, separate from Mistral, there's a Creative Common License with additional restrictions on, uh, non-profit. use. For example, Vickipedia is under such a licence. Everyone has free access to Vickipedia. You can read it whenever you want, you can use it for any of your research. But you can't do a commercial product based on Wikipedia. You'll violate the rights that you're given under this license. This is the little tour today to show that not all that is-- not all that is called open source is the same open source.
Yeah, there's a lot of intermediate steps that you need to know if you want to go deep in this story, use these products in your own way.
for some additional purpose.
Well, I think it's very useful. I know a lot of businessmen who don't know very much what you're saying, actually, they don't understand what you're saying, but they want to create some kind of start-up, do something on the basis of something. And they usually have a team of programmers there who do something. I think it's a very useful story to make any man who wants to start a start-up or start-up or introduces something in his company to see and understand where you are. Now, you're acting under the law now, or you've already broken it, or you're gonna break it in the future. To understand at least it's incredibly important, especially if you're making a start-up, you want to make money on it, because maybe you're gonna do something, and you can't make money on it. Or you start selling it, and then you'll have to pay a lot of fines or some restrictions, and so on. Very, very useful information. She's kind of complicated, but at the same time she gives, uh, the next such a platitude in general, what's going on behind this line? And especially you're right that in open source, people don't understand that open source, yes.
We just opened the window on this side just to get our viewers a little familiar with this story. If you're interested in getting this in more detail, maybe we can just do a little more detailed on this subject. Write in the comments if you're interested. We can actually deal with this issue and prepare a separate little video on this subject.
Well, thank you very much. Ah, ToTheMoon is technological news, Silicon Valley sites all over the world, from the USA. We're going out once on Sunday, we're always four people. See you again!