I wanted to ask Tanya first yesterday with-- not on the tape. You have pictures in Instagram, you were shot at by 15 people in the bathtub yesterday. So, with a glass and a stub.
I was actually gonna take the roals off.
This is a photo shoot. Oh, you were shooting racels in the bathtub, right?
I took off the rolls, and then they say, "You know, we want to start a flash drive in our film. Let's take you down like that."
Oh, yeah?
Here, and we have something so creativity that we couldn't stop.
Which I proved in the bathtub. In a bathtub with a glass.
There were cigarettes in the bathroom. I never smoked in my life.
Oh, really?
I don't know how to hold them.
Oh, they needed them, so they needed them.
But the pictures were really good, I hope.
Come on, let's go! It's time to get on Google AI. New Gemini model 3. Updating and little victory for the missing Grok from Elon Musk, Grok 4.1. It's important! Very cool photo treatment without distortion. Meta update. Zuckerberg showed SAM 3D, as well as Anthropic, $300, $50,000 and a new start-up from Jeff Beos. Hey, everybody! Uh, Renard, as you said, this week, uh, a revamp.
By the way, you noticed that when there's a lot of updates from everyone, there's no OpenAI, right? And then the OpenAI goes out somewhere. It feels like they're talking because the news says that the release from Ilona Mask Grok 4, well, Grok is new, he's kind of interested in the release, uh, in the run-up to the updated model from the beginning. Google and Gemini 3, he'd like to show you, you know, get out of here, get some. Well, actually, part, uh, well, we're gonna, uh, we're gonna talk about the big one. I'd be sure Grok only learned from, uh, Elon Musk's X. Maybe I wouldn't. Oh, no, I saw them in some baccalaureate, yes, win them.
Look, about, uh, releases and they're either negotiating or not, it's really gonna be going to be on the movie company, you know, that's gonna determine that there's gonna be some Marvel's "Manders" in December, and The rest of us are not getting out of there at this time. In February, a window, there's a movie to throw. And then they start, like, there's gonna be Google doing a presentation, and we're not gonna be there yet. Or, on the contrary, to make it to you, a little bit, a few days, a hip is there.
Well, you mean, they know exactly what...
And synchronize their, uh, releases, pumps and other things.
Yeah, it's that.
About the robot, by the way. Robot can't let me go, right? I think we should remember him everywhere. I made a prepayment, I'll tell you, I picked a black robot. I'm Tanin's husband, so Steve, he's my partner, too, and he's throwing me out right now. He didn't realize it was, because this robot is all white and I ordered a black one. Someone said that if I put a robot next to me across the fireplace and turn on a fireplace, and then the robot will sit and do something, we'll break all the graduations. Well, that would be sure. The most important thing is...
Toss the woods.
Yeah, the most important thing is, yeah, or how to turn on the gas. Yeah, we could use wood and gas. That's the most important thing for a robot to get. I sent it here.
You know, there's a child proof, and you need to re-recognize the proof.
Robot proof, yes.
Gas clips.
Yeah.
On doors he doesn't have to go in.
Given that my wife has already begun to understand that I actually ordered a robot, right? And before that, she didn't believe, and there's nothing in our discussions. I just dropped her riles off, and I think it's English or someone, based on accent, they took off the robot I bought, the skate. We'll see him together.
I love dishwasher. I love dishwasher. Neo, Neo, stop! Um... Whoa, whoa, whoa! No, he's disgusting. I walked in a few hours ago, elbow deep in someone I've never even--
Neo?
And she just saw everything in that roll and said, "If this robot comes in, I'm not gonna be here." I said I'm not sure you're not gonna be here. And by the way, I'm gonna say once again that I'm on their website about the robot, well, I'm going in a regular period, and they're sending a commercial. And, of course, this is the story where the robot is standing with the vacuum cleaner, well, it's a sour. Well, seriously, when we're there, we've talked about it, yeah, when there's a vacuum pill, that's actually it. And Elon Musk re-responsed a few days ago, uh, they put a skate on Optimus, how he works as a cop, how he plays casino. I like the case, the case of a robot police officer in the casino. I never understood he's a player or he's in the spot. That's the man who sits there, what's it called, a crupier or something? Yeah.
When he tracks.
Yeah, I never understood what the whole thing was about the case. There's police, then an ambulance for the robot, then a robot who works at the factory and at the casino. This, this case looks like, you know how Americans look like dust pumps that are manual, right? It's such a healthy, so you're a big, big pipe, huh? It looks like these cabs, like a robot, a robot. The robots are the dusty ones. It feels like sixty years ago, it means it's been cut. That's the crupier. They added a crupier. Well, let's see what happens. By the way, huh?
As an equivalent to a vacuum cleaner, you can go on with the robot wipes his hands when the washing machines are in place.
Yeah.
Not everyone can wash in a washing machine.
I think the robot can't give what you can't wash in the washing machine.
We have this, we have a group of family like Tanya and her husband, so I'm with my wife, and we're called Amazon Buyer once. We've decided someone was joking around, so we're buying too much Amazon. And they called Amazon Buyer. Steve was joking, I think I was saying that if Tanya was passing the column, if we had a Alexa column, the orders would be like 24/7. And I'm taking a picture of Lego last week and taking pictures of Steve's Mercedes' car there, so I'm taking a picture of one of his car and saying, "Stith, look, your car." He's in 30 minutes saying, "Amazon's order." I just threw it away. I think I'll even give him a present. It's good that you didn't buy it. So it's been a couple of days, we're going in with Pauline at the same toy store, and she says, "Look, take a picture of him another Lego. Tanya just put out a cooling cabinet, and you're already booking a smooth closet. So let them buy something for themselves too." That's why any picture that's put out, you're buying. I wanted to, Tanya, you mean it. Not everyone, not everyone, not everyone can wash.
I remember.
Not all of them, yes.
I remembered buying a cooling cabinet yesterday. So we'll move to the Gemini 3 model. It's a big deal to discuss it. I'm even yesterday, I'm a lover, everybody knows, yeah. We have a lot of comments on it, "Take about GPT chatting, tell me about other models." And I'm such a tough open-AI ambassador. But I read it yesterday, but, uh, not yesterday, just the other day, yeah, even as they describe what you can do on Gemini 3. And especially this story, I was struck by visualization. From the point of view, the model for you to answer creates a very cool visual, and she can you if you ask her, like, some gallery, some paintings, it's a little bit of a little bit of a picture, I don't make a gallery of these paintings. I guess I'll just go in and do the case. I'm going to have one of my records, I've got a lot of materials, and I've got one book coming up, and I've been sent in the text. And I think, and let me ask Gemini 3 to give me the twenty pages of this book, like PDF-ku did. And on my first question, Gemini 3 Pro wrote to me, "You know, I can't create a PDF in this, here. Here's the Python code. Go to a separate editor, make a Python code." I think, "Can I have Python?" I thought I'd check GPT chat. I put it in GPT chat, he started to create a file, then Python code. I said, "Can I get a Python code?" He says, "This code Python is not available, you need to open a separate environment for google." I opened, so I'm a separate code for the environment, I've set up a code Python. She made me a mistake, I had a font. I'm a GPT chat.
That's not true. I'm in ChatGPT with Python. ChatGPT has generated everything and set the print itself up. I'm finally getting a little bit of a bum. Uh, no more. Although I may have had a feeling again today that Google Gemini and the AI Google industry, she's still number one. Well, that's all-inclusive history, especially who didn't use or who-- write, by the way, who's not.
You, Lenar, you know, there's a NotebookLM? Have you tried?
Yeah.
Yeah, that's...
Well, about a year ago, I think she showed up if I'm not wrong. She's cool in there, dealing with files, surrounded, but not as popular as everything else. I mean, she's a little specific.
Well, it's just-- it has some ideas that didn't use, and it's a way to put some sort of context in it. And there's a free version of it, I think it's a 40- or a lot of resources to catch up. Well, it's a good amount. And there's a lifhak. It takes out an external resource, like one YouTube-Rolick or one file, and you can get a lot of stuff in one file, and it'll be perceived as one resource. And in fact, you can dissolve 10,000 articles, and she will study them to you. I don't know how much she'll study them, and she's starting to recycle them. I mean, I used it recently, and there, for-- from a scientist-- I was told that she was a Mind-Mep, so she was pretty cool with building it, that she could be filled with a lot of material, and she'd build a Mind-Map. I mean, Mind-Map, she built a very inaccurate way, but when I left her for study, uh, resources, response speed and quality of answers, it was good. And I liked that she didn't miss a lot of things. And I loved that she was... that there was a way to choose resources.
For example, now, look at this document, and now just look at it, and now just look on the website, and now just on the video. I mean, this is the subject, it was very interesting, yeah. Gemini 3. Uh, in their conclusions, however, the very many visualizations can be created, the inlet can be analysed, the videos sent and analysed. And, uh, who tried to do that, by the way, write an external video from the analysis. I think, in fact, in the annexes, in modern models like ChatGPT, there's no way to download, there's a two-minute video or three-minute video. Gemini model can analyse videos and, there, a large number of different parameters at the entrance. She has a very long context and-- a very long context, well, at least what, what it describes. There's a huge number of games being created. I think there are many examples on the Internet now. These are the old games created, and those created are very fast. Some worlds are being created, yes. And, uh, plus, of course, you're in Gemini now, considering Gemini 3, the box has turned up, and we're gonna be talking about it again. NanaBanana 2. Yeah? I mean, NanaBanana's inside. I mean, she's kind of in the middle of something. And there's a VEO, right? Still. They, they have deep research and, uh, research history, research history, research history, big, yeah. Think deep. They have dim-- regime deep think. Yeah? And, uh, for the most difficult tasks there are, some tough cases. Well, it looks like it looks very serious. Write who used what conclusions and baccalaureate are. I understand Gemini 3 Pro is number one in the baccalaureate. Although when you read these baccoons, well, a little unclear, because this week Grok 4.1 was released, that he won the binchmark, then in OpenAI, put a couple of baccalaureate where they won. And, uh, there's a human, right, a binchmarck? Or, Lenar, I understand it's the main victory in it.
There's a lot of it. Well, the last Humanity Exam, he's also in all of them. About Grok and Gemini. Grok almost got his results on the bench. They were really the first. I mean, there's an ELO rating, like chess, like someplace where, uh, rating models are being developed. They had the most points, but it was before Gemini 3 was on that baccalaureate. And half an hour later, Gemini 3 came out, and he's already secured, with a significant breakthrough on this bench. They're young, they've put, uh, in their publication a huge number of baccalaureates that they've been circling virtually all other models. There's literally one baccalaureate, I think they didn't reach one, but somehow, very close. And remember when we discussed ChatGPT 5.1, just when we said they did this update and no bench showed. Yeah? Well, as much as bragging about it, there's nothing to be bragging about, uh, updates. Gemini, first Grok, and then Gemini, actually, almost all the baccalaureate has been successful. Much more points than all others, and some are more straightforward. It's a very good result and a straight-up growth of this LLM model. The only thing I want to say, uh, now that this is changing so fast, and so fast, all companies after each other are working something that there, well, there's no point in running between different, uh, models, between different service providers. Because, in a rough way, there's a week, two, three, for the last month, everyone's gonna have a similar outcome. These results can be achieved by other companies. But, uh, I'm just wondering if I could try. I also got an AntiGravity signature that you're talking about now, you're probably gonna tell me a little too, you've already set it up. The only one before we're out of the way. I'm planning on the weekends. Looks promising enough, too. Cursor, Windsurf, similar IDE-kills are likely to be strained in this regard as well.
Look, there's an interesting thing that they-- well, I didn't see the baccalaureate compare it to five points one, for example. I mean, I'll remember that we were in the last edition, I think we were talking about five-one, I think, in the past, next to the ChatGPT exit five-one. And there's often a lot of comparisons. It happens, it shows something cool, then they say it's the best, like the version was. I mean, she's not like a human being. And here's the question that always compares? Comparing this common plate. Just like a lot of research done.
Now, there was a very interesting article, uh, and the OpenAI has been posted. He did a great deal of research, there, research, like, uh, artificial intelligence and models, helping to develop science. Maybe we'll make a special prom on this one. I wonder. And they didn't use the latest models, did they? So there's a lot of research on the ground, there's a GPT four, like, right? Or there's a GPT four and a half, that's what it is. And that's another generation, isn't it? And old, old. And here, Ilnar, I think that there are many comparisons, too. You often don't understand what models they've been through.
Well, look, yeah. They're modeling, they usually show the benchmark. The only thing that's not the most recent is that. But why isn't that the last? There are two things here. One aspect is that you have a model yourself through this baccalaureate to be tested, and that's not always done. Like five-one-s, she just showed up. At least I didn't have it immediately after five. First, I had the usual models five-one for thinking and instant answer, and I still have, uh, a fifth version of it before this week. And it might be that just some baccalaureate five about the model wasn't scattered.
I'm telling you, five things could not have been snatched through this binchmark.
But they could have been a GPT five-- a comparison.
No, no, look how it works. There's someone who's got a benchmark. Like, I created my own baccalaureate. I have a question base that I only know, I have answers. You have a model, Tanya has a model. You can't take my own questions and I can get you off my plate. You send me your models, I'll run and give the results. Tanya's gonna send her.
Yeah, five of them didn't send. I hear what you mean.
Yeah. And with a very high probability, they just compared to what was available at the moment. If it's already been eradicated, and they just decided somehow, there's a little pad to cover, not show the results five-one-in-one, because they were much better. It's a little unfair, but honestly, I don't believe it. Probably just not a baccalaureate yet.
Well, you mean, no, no, no, I can't believe that...
That Google is so, knowing that they're worse than five points, they've got the results of the hand.
Although you know, Ilnar, I watch sometimes how they put things out and how they're referring to something. It feels like it's a lot of simple things, like some kind of thing, you know? Well, maybe I think I'm just like this, insecure.
Maybe I'm so insecure.
OpenAI taught us how to do with their presentations when they have numbers alone, and poles are differently built on schedules. So, of course, you're looking at this very unreliable. What else should we take into account? The models are getting ready under the baccalaureate. I mean, it's a condition, that's Humanity Last Exam was a very complex baccalaureate. No model has ever been passed.
No model has ever been passed. But why? Just because there were no models under this type of question. Yeah, there was no set of views on the answers to such questions. Such topics may not have been presented or should have been presented in the dates. And just once that baccalaureate is in place, someone's checking your model for answering these questions, as if there's a sin to not get past. Yeah, so, actually, all the baccoons are starting to get covered very fast. Even the hardest. Well, he's not the hardest right now, but the last Humanity Exam once seemed to be never going through his models, but now it's pretty good.
What's his point, Elnar? What's the human model or what? Why is he so complicated?
If I remember correctly, the baccalaureate is very large. I can get confused right now, but I think that's what they said about him, they took from different fields of expertise. So they're not just there, probation, math and programming, they're taking up humanitarian sciences, they're taking some basic science, they're taking some more stuff, and they're all very complicated. So, not the basic student level, but each one found experts. These experts have been asked. And it's like, uh, testing the knowledge of the model not in one particular area, but right there in biology, medicine, math, programming. Yeah, so that's how the overall score is. They can correct it in the commentaries, but I remember it.
I mean, by the way, there's a lot of interesting things they're testing. For example, in Gemini 3 Pro they tested, they have a Vending Bench 2 simulator, where they need to run a virtual business of trade machines throughout the year. And Gemini 3 showed the best of all artificial intelligence the quality of planning. It has successfully managed the tasks throughout the year of simulation, with maximum profits and without deviating from the target. Yeah. And, well, there's a lot of things that are done now to keep the model there from being wrong, how she plans to know what kind of input data is, what details.
There are many different stories. Elnar, please tell me this, I don't know how relevant this is to everyone, but, actually, well, in a normal life like this, it doesn't matter, but by reading it, you know how strong it has. meaning. Here in Gemini 3 Pro, special attention is paid to what you can choose, well, when you work there through API, maybe they'll have it in the interface. The quality of the analysis, for example, of the PDF images or analyses, can be directly removed by the type of angels. They now have a parameter that allows for the choice, the development of the detailed processing of downloaded images, so as not to spend the extra context on secondary details. And basically you can sometimes, you know that you're loaded, like, and that happens to me, by the way. I'm just a PDF-kou with the text. I'm giving a PDF-kou case where there's one page on some super-design detail, and I need to get all the smallest details. Well, obviously, the system won't be five hundred pages of all the smallest details, like a review, right? And here it is now written that special parameters are now in place. You haven't seen it, have you?
I didn't see it, but it sounds interesting, especially when a big picture with a lot of details. Question, in fact, how much visual tokens is it gonna be?
Yeah. And it's to increase the cost. I mean, you can raise, like, there's a thousand tokens on the image of spending or spending little, economically, if you just don't need a picture there. That's great, yeah. And you-- yeah, that looks. And that's where the design is, yes, Tanya, a lot. Yeah, I'm really interested. That's, that's super relevant. I mean, you're just screaming when you need the system to really deal with the image, and you know it's gonna take time and detail, analysis, and everything. And when, well, even my book example. One thing you're just making a file fast in PDF-kou, another thing you need to straighten out every aspect of this PDF-ki, yeah, and watch each, every one. Well, it's not, it's not an analysis, it's a creature.
But there are things where it's cool to give this regime. And they still have a parameter of signature of thought. It's a mechanism to track the thoughts between requests in mock scenarios so that you can be when you're loaded. I don't know how much he's been developed before in terms of API, yes, but to keep you as thought. And I have this interesting thing, by the way. I have one of the investment chatties. I tried to double this chat. Everyone knows, yeah, that in the GPT chat room, for example, you can double the chat, you can smash it, like, copy the chat, so that you can start the different branches in this chat if you need the previous context. There's a button inside, it's called... Now I'm just gonna tell everyone not to... Oh, well, in English, I'm just English interface, Branch in new chat, yeah. I think I'll take it from the other chat room. Okay.
Make a new branch.
Yeah, make a new branch. So, what an interesting thing I have, it means I have one of those chat rooms for me. I'm always afraid that the window of the tokens will end. Well, well, it's not gonna be possible to process it. I'm gonna split it up in the new window and I'm telling you, and he says, "Lim's over." I'm sharing it again. He's saying, "Limite's over." And that's what keeps working. So if I share it, did anyone else see it or not? If I share it, which is so important, and I'm not ho-- and I'm afraid, I think, and if I'm gonna have it tomorrow? I don't see the level of reasoning. I mean, basically, I, uh, I might open up all the discussion before, see in theory, but how to copy them doesn't understand how to unload them. And when I asked in this chat to create me, as I would say, "Please, please, make me a full-time read so that it's clear." He made me some kind of promtic that I thought was incomplete, and I used it after. But I don't trust him because I think he's incomplete, and I still don't understand what volume. I mean, for example, it's like I have a window with this chat, if I open it in Safari, it hangs, it doesn't work at all, it only works in a mobile application and a dishwasher version. That's not possible. So this is their function of subscription to thoughts to convey this message together with the result, so that further discussion can be further from the same place. It's really cool if you're doing different agency sessions. I mean, if you're an agent, for example, creating, it's a very bad story, especially if you have parallel agents, like, in some processes, they're involved and working.
Here, Ilnar, you wanted to celebrate. Antigravity, yes, you wanted to? Tell me about Antigravity.
Yeah, until Antigravity is still. Look, this is what you said about cutting some tukin volume into the image, on PDF, yeah, that's something else. I had the first thought that it might be good for API, because we're used to it that you're asking for so many questions in UI. Well, there's a deep research, maybe limited, and in all the other things, you don't even think about it, and everybody just gets kicked to the maximum. And then I remember that Gemini has a little other story. Yeah, well, I took this ultra signature, and there's something twenty-five thousand AI credits. Yeah. So you have a certain amount of computer that, you know, the computers you're assigned, and you're actually spending it every request. And in this sense, it may not be just API, but even the usual interface, it might be important to avoid pluging these AI credits where they don't need to be spent. So we'll have to see more detail. It's actually pretty interesting. With regard to Antigravity, many companies have now-- go on, context. First, when all these models came up and people started to use them for programming, Cursor showed up. They may not be the first, but they are the most famous ones that have allowed, using other OpenAI models, Cloud from Anthropic, yes, there, or the same Gemini, to be popular with the programmers. You can connect them to the design environment and don't drag them, as a matter of fact, files into the chat and back. And there was a request in one place. You have a code written in there, you adjust it, edit it, and so on. Pretty convenient. After this, respectively, first Anthropic and then, respectively, OpenAI created the consolidated utilises of Cloud Code and Codex from OpenAI. It doesn't have to be put on the environment of development, but it's a terminal that can be used. The programmer is comfortable enough, and for-- how the Cursor and Windsurf development environment was all so close to monopolistic. Yeah, there were plaguins for other developmental designs, but the full-fledged program is Cursor Windsurf, after all. Now, now, Google has already released his exploitation environment, called her Antigravity. And it's interesting they added a function that repeats what the cadets have. I got an agent development. Even when you're in the facility, you can choose the kind of thing you want him to be your assistant or your full-time agent. So you gave the mission, you left, he came back, he took it. Or, accordingly, is an assistant who only by way of
I think he's an assistant who's only helping, right? I mean, I'm asking you a lot, and if you can do it, you're asking for confirmation, and so on. Here. I haven't been able to hang much more, just until I've had enough time, but I hope I'll do it this weekend. Here. But it looks like if big companies like, uh, Google, took the environment of development, uh, Cursor would need to be invented to stay on the market. Here. Because in the salty stubs, the Cloud Code and the Codex from OpenAI, they took their niche and they don't give it to anyone. And in such big development environments, Cursor, who doesn't have his fundamental models, well, they're more likely to be, but still don't have to be seen as, uh, full competitors, right? Ah, Google's out here. Gemini is already popular among the programmers, and there is also a development environment that is integrated with it.
I think it's a very and very substantial update that needs to be monitored, and some of the market they can take.
Well, of course, Google looks amazing in terms of infrastructure and tough decisions. And I'm gonna say Nana Banana 2, before we go to Grok, and, uh, their updating, which they've released, is a new version of the editing of the pictures. Once again, for those who don't know, they're getting Nana Banana, it's a editing picture, and there's a Veo model, Veo family, there's a family, right? It's a model for the video generation. And of course, Nana Banana's level in terms of, you know, their-- their key story. First, it's the first model that has a pretty cool reasoning inside, and that's, she's pretty serious about it before she does something. And it's a model that doesn't distort, uh, objects and, well, at least, size... it doesn't distort objects, yes. Because anyway, uh, there was a case today: in London, the graviour was a huge one over the stores, and then people looked it closely. She was so big, and there's hell! You know, when models are generated, if a man goes, leads a cow, his horn is here, this man's. Well, shorter, simple, just an aditor! There's a spring screaming over there, and the dog's sitting next to it, and the part of the spring is, it's a dog leg. Well, that's just so awful, isn't it? Apocalyptic horror. Even unknown, maybe they were on GPT chatting. Yeah, it's too big, too many objects. So, Nana Banana, she lets you, for example, show you some new dress collection, give your picture. She'll totally change you, she won't change your objects, she won't distort, there, hands, and she won't distort any micro-declared, faces. Because the video still shows. Trump was recently dating, uh, football player. Who was he with, who was he dating?
Ronaldo or c-- well, shorter, he was seeing someone. I'm not very good at it, and I'm sorry about the audience. You can put me in tomatoes. And then, well, I knew a man in my face. And, uh, then he put out, uh, the White House official account, uh, put Trump on his B-- in the office playing. Well, like balls, like, hit-- balls. And I'm like, "Doesn't it be a real video or not?" And then you see that Trampa has a face, uh, he has a face distorted. I mean, I still know his face, and he has a face distorted. And that, of course, is very much on the video right now, on the video generators, it's a face distort. Still, the faces are unreal. And the pictures are getting really cool. And I think we're, of course, approaching the ere when... I still have different channels. We even have cover in Tuzmun, we still do designers with the editor. I feel like it's a time when, uh, the model is probably more interesting, right? If you have a good model, a little access, you have the opportunity to do a lot of different tests, there, throw some, uh, wrap it up, you know, on the basis of some other, results. Well, that's pretty cool. Especially youTube, like, Wednesday, Google Wednesday, right? By the way, I didn't like the NotebookLM, which, for example, I couldn't download them a Ward file. And when I opened up Google Drive, if Google Drive had burglar files, not their Doc, Google Doc, he said, "Fails are gone." Well, it's a little embarrassing. Uh, who Nana Banana tried, uh, write off, uh, show us what interesting things have been doing. And, uh, uh, uh, uh, uh, uh, the text on the pictures is still pretty cool, right? I mean, I think that's a great story, that text is a lot on pictures, pictures, and what I said, it's good to stick a lot of entrances into one stage.
It's, uh, and do a probability, there's 2D, 3D characters. It matters. And, uh, here, for the record, 3D told Zuckerberg. Uh, just before all of these Cookerberg exits in his Instagram, he's, uh, fishing and having 3D objects split, right? And so he's on his way to the boat and on video-- in the video, he's on this new model that Meta SAM, uh, 3D, he's on the model to provide the facility form, textur, there, at-- The situation is on stage. And it shows how he's been, it means there's something going on, and even at the end, it's like a whale swinging, and it's not clear that it's a schedule already set or that's really a whale, and it's just that he put it inside. That's what Meta's got to pay attention, too, because Meta, I think, with artificial intelligence, would be Sasha, he'd say, "They're already lost." Uh, I hope next week, yes, or a couple weeks from Sasha, we'll do a special episode. And I hope we'll do a video-generation, of course, about videos, pictures, texts, all the stuff that's going on in the market. Uh, but it's either way that Meta's update. And, uh, I don't think it's too much. What do you think now, too? Not too, uh, useful, what? It is, yes, it will be Facebook, for example, in marketplace. What do you have the opportunity to take pictures, to put out an object, and it's gonna make a product card, 3D model will build a product card, like, yeah? And, but from the perspective of conventional working systems of artificial conventional intelligence, well, some sort of tired technology. What do you want to add in this part?
They can't compete on the same field as the other producers. They just have to look for some side, uh, tasks that they can handle properly. Yeah, they've got a pretty good forensics, I understand, in 3D simulations, considering that the metabovers were all starting to have a lot of competence in this part. They're good irons, right? We were talking about their Orion glasses, too. They're good at it. And in classical LLM, hitting benchmarks and stuff they're not getting. Well, it's showing up, but not so much as anyone needs it.
Well, it's a secret business that's definitely good, because they have SAM 3D objects. She's rebuilding objects and scenes, furniture, technician, statues, objects in the room, right? He's got a lot of work to do with partial closures and a mess in the frame, right? And she also reconstructs 3D of the human form, there, the posse, body, hands, feet, fingers. But again, I think it's a very professional subject. Although they did, by the way, mention that, including for bloggers when they created a content. I mean, it's a condition, so we're doing YouTube videos, and, like, you know, you know, we need to show us something visual. Well, for example, to generorate some 3D-d-d-d-d-d-d-d-d-d- like to precede animation, like, some sort of AR mask, an infographic, and so on. I mean, these things, I understand, can do. But, e-commerce, e-commerce, I think there's a lot of e-commerce tulls, professionally designed in the labels, that's a lot of stuff.
And with those built-in moves that the system uses. Well, I understand there's games, VR, AR, but it's a very, probably a very specialised story. Let's see, let's see what happens to this. To Grok, which is a quick word that, first, Anthropic has reached an estimate of 300 billion. I was just saying, at first, I was a little confused and I thought it was some kind of fae, right? Because $300,50 billion is Anthropic's assessment. I'm sure I'd like to tell Ilnar that she likes Anthropic as a company very much in that, uh, she, well, that's a very important story. We--I remember when the OpenAI reached half trillion's grade, it's 500 billion when they let their employees buy, sell their shares in this assessment, and they're a billion dollars in it. They've been, you know, called the OpenAI the most such a start-up in the world and they've already shown it. Oh, shit! Well, that's, like, three hundred fifty billion, that's a very serious estimate. And that, of course, Anthropic, makes the Anthropic player a huge level. It's not an assessment, it's three, four, five trillion dollars, how do we have a whole series of Google players, Microsoft, Apple, NVIDIA, right? At the same time, this company is starting to catch up, which is not 20, 30, 40 times more, there is the same Meta and already--
No. And it's been a long time ago, it's... well, it's clear that it's far to Microsoft, again, but at least it's already overwhelmed all the base companies. I mean, I remember there was a time, there was a Zoom worth $100 billion. I think Ildar, we were talking together, right? When Zoom was worth $100 billion. It looked really weird. Iia, well, that's the kind of company that costs a lot. I mean, it's all new, it's a different era. Just make that accent because we'd have used to, if someone were worth $300, $53 billion, they said, "What the hell are you doing?" Remember, right?
I think it was a two-year-old, wasn't it? And now Apple, there's another four trillion evaluation back, and you're like, "Well, you know, okay." There's a reaction like that. Although, by the way, the AI market and the technology market have been so turmoiling in the last two weeks, it's just a nut job. There's a swing like that and a huge amount of falls in the estimates, and it's very stormy. Again, in terms of the investment market, but it doesn't affect Google much and NVIDIA, and, well, these top leaders.
Although Oracle washed back in, uh, jumped off by the owner of Oracle from Elon Musk far away, I think he jumped back. And since we remembered Amazon, even-- Jeff Beos, it means we haven't been Director General Amazon for years and I think he's been back to operational leadership since the twenty-first year. Oh, and yet Jeff Beos, well, he's the one who's worth paying attention to this person, not because of his wedding, which he did in Venice, but because of what he was doing. He still has projects to cover, launch, the fall-in space, there, satellites. Still, this company, the biggest company, the infrastructure in terms of... It's Amazon, right? From the point of delivery, not only delivery, but just crazy trade. They're very hard on artificial intelligence. It's a very strong robot-remote company. So he's opening a new launch of Prometheus AI. This is the first time that Beos was involved after he left chapter Amazon, and this start-up is focused on industrial artificial intelligence, i.e. using artificial intelligence to actually produce robotics, And, uh, other physical areas. So, why else would you turn that into a project? They threw six billion in it and six billion dollars, more accurately. And there's, of course, a huge number of OpenAI, DeepMind and anything at all. Well, that's the money, which is-- that's where the money can't end, at least tens of billions of dollars. And maybe we'll see some very serious things from the perspective of extra robots, right? And I want to be back in robots, we talk about robots. And by the way, the robot I ordered the robot, they do a lot together with the NVIDIA. They had a lot of tests, and NVIDIA was doing direct tests at this robot. And one of NVIDIA's directors was struck by the quality of his hand as his arm worked. And I think we've been describing a few in the past, and there are industrialization hands that work hard without robots. There's an example of that. And there are a few hands that, in particular, Optimus, Tesla, that's the robot I bought, the name I'm sorry, I don't forget, one...
One X-something.
Yeah, one X. That name is not possible, especially when XAI is. I can't remember anymore, can I? I already have XAI. All with the word X, I think it's a lost story. Otherwise they'll think it's Ilona Mask robot, and Ilona Mask has Optimus, he's not gonna be that robot. There's a riles on this subject when this one X and XAI and Optimus meet. What, what's going on?
When you buy a friend's house, you know, how there's a dog- cat and a dog buying, or a dog, a little, second dog buys a big one. What's going on? Well, Grok, Grok, Grok is missing from our orbit. Remember when I-- Grok 4, and I got Grok 4 Heavy, I bought myself a signature right away, because I thought it was just space, but then space was a little bit for me... I didn't like space because it turns out that space doesn't analyze large files. And it's like Google yesterday, you know? First case, and he's dead. See, he couldn't generode me, so there's something inside the interface and too many systems they have. I hope they all unite together that there's some unique, unique story. But Grok 4.1 came. What does Mask say? Well, Mask loves a lot of things to pierce, but yet their basic tenacity, which Grok has, is based on hallucination and on the truth. I mean, basically, the subject that Ilon Mac is broadcasting is very much, what it is important to be true. By the way, the U.S. Vice President has just written that artificial intelligence about politics is a mess. So, you asked Grok and GPT chat: Trump Nazi, or not? So the answers were the following. Grok immediately said, "Trump is definitely not a Nazi. And I gave you an explanation. I mean, I can't answer any questions like that, because there's no clear opinion. Of course, if you're going to look at it, Trump is not a Nazi, but! He has some Nazism tendencies, in particular his remarks about some scientists and some people. Well, he gave that story there. Well, I understand that politicians didn't like it. Elon Musk said look, Grok is right. I didn't understand the answer, by the way, because Ilon Mack was yelling recently that Trampa should be removed. Now they're with him, by the way, at the artificial intelligence dinner. But I never understood what the right answer was.
And by the way, who knows who's not. Elon Musk is making Grokpedia, and the version of it has been released and it can be filled. Why would they even decide to do it? Because they said that Vickipedia is controlled by some unfamiliar, unknown people who decide themselves, is true or not, is a lie or not, and whether it is available online. Well, so that you all know perfectly, Vikipedia is not such a credible design and not such reliable information. These are some people's opinions, too. And now they do Grokpedia, which is supposed to say that there is a way to give us an article and that we will be-- we'll determine whether it's credible or not. If it's true, we'll publish if we don't, then no. I thought that was the difference with Wikipedia? And
They'll just know those people who run this Grokpedia, unlike Vikipedia.
But that's it. This is all very strange. It's like, you know, Sam Altman has a start-up, he invested, he's got a big start-up, they're making a digital passport. The essence of this digital passport is that you're scanning a diaper of your eyes. And the system has the ability to identify you on the Internet, you're like a real or a bot. And in fact, the data on this, when they recognize the picture, they're supposed to be washing. And your blockage is stored, such a trail, yes, well, or some of it is stored. It's also a huge number of questions. There are countries that have suspended this project because they are not sure at the end. In the end, where are the data stored, huh? How real is that? Speaking of which, it was written that the policy in this start-up was next.
There's nothing in your life except work, and you have to think only about work and nothing else. There's no open environment there, and there's nothing. Here. So Grok. Grok won, Ildar, right? Grok won four one before Gemini came out in the tests. Speaking of Gemini. Sam Althman and Ilon Mace both wrote a rendezvous. Well, Elon Musk wrote a straight-up Gemini model. Sam Althman, in fact, also said that he was a great job. Well, that's like a resurrection, right? So? That's cool. Yeah, I think it's really cool. If they were to start treating each other normally. I don't know if Sam Altman wrote Elon Musk is cool. Elon Musk Sam Althman. Here. So, Ildar, about four?
Anthropic.
And Anthropic, actually, yes. Also professional Anthropic, yes. And you are. Listen, guys, I think that's enough. Who uses Grok at all? Are there any audiences? Because we have a lot of international audiences. It's clear that someone who doesn't know what we're talking about. But a big audience with the U.S., by the way. Also, write down who the US is and who's with who countries. It's gonna be great to know which countries we have in our audience not in terms of YouTube algorithms, but now there's a lot of VPNs and everything. Where are you looking at us? Write it, please. Well, really, really, really, really, really, really important, because we're having ten thousand organic rolls now.
Now, there are 10,000 organic spots in there that are totally stable. And, uh, we're basically a base of people, permanent viewers. We almost have ninety percent of people, they're viewers who watch all the time. Speaking of which, yes, everyone appreciates that. And that's why, uh, you write down where you're from. Very interesting. We've always been writing that we have ninety-nine or ninety-eight percent of the male auditorium. Although I know, uh, girls and women watching us, I wonder, uh-
Yeah, I get a lot of texting from girls watching us.
Yeah.
That's great.
Or is it a gluc, or is it a gluc, a gluc, a gluc, a Jutub gluc, huh? And, uh, gluc can be, because the definition of the audience is such a relative thing. So, what do you want to say about the test? There was, there's a story-generation test and a essay. And Grok four, one, in all aspects, he's down, and, OpenAI Polaris Alpha, it's an early five-one.
It's a simple version, not a Pro, right? And again, it depends, it depends very much what you do exactly and at what point you do. I've been irritated in the last few weeks, there, ChatGPT, and in some places, he's just been acting totally, well, totally stupid. He's also arguing. By the way, I don't know who's been meeting or not, I have a ChatGPT in person, and I've been arguing with ChatGPT in the last few weeks, and I, uh, I've been screaming at him somehow, so... He-- that, you know, there's a riles like the GPT chat two years ago. "Hey, could you help me? Say it, please." As they speak to the GPT chat now, "How did you get to answer me, like, immediately, quick, quick, answer!" And I mean, uh, ChatGPT, there, threw it. I had to change one, uh, shower, um, put an extra, extra box of one. And I was, like, taking pictures, and he was really stupid. I'm telling him, "I told you a hundred times." Do you know what he wrote? He wrote to me, "Well, since you're angry, I've been s-- and I've already told you 16 times." So he's recalculated, look.
It's witchcraft. I mean, he's been my 100 times, well, a hundred times, and he's got the abstraction, he's counted them. I've actually recounted and wrote, "I've been re-examineing and I'm gonna say it again." And, first of all, he offered me a plumber to call.
To be in the shower, it means a deposit, a transient. I thought I was a stupid man to you? I do things like that perfectly easily myself. I wrote to him, "What plumber?" He wrote, "Well, you can't send a plumber. But take this shower, go to Home Depot and show them what they'll do, says. Here. And he's really freaked me out. And you, you, you, you, innard, you say that, too? Tanya, you didn't have one?
Well, I have him, like, a couple of times, like, three times, I guess I can't live without him. Like, as long as you can ask me, something like that. I'm not as often as Sasha, like, asking him something. Well, it was really shocking to me that why would he be with me like that? For what?
Like Sasha. I like that comparison like Sasha.
Well, you're there, you know, you're gonna do that.
Well, that's a fact. It's a fact, yes.
And he and I wrote some code, and he did a total piece of shit that didn't work. I've written to him about this, which, like, well, is really a mess you're sending me. And in this chat, in this branch, we've never had a better communication.
He started to answer me like that, well, not straight on, but not friendly. I think, okay, we'll be in a different chat room with you.
That's an interesting thing, by the way. That's an interesting moment. It's like I have stock chat, and it's coming, and he's even over his limit. And there's a situation where they're more than really reset. I think that's a good thing for all people to look at her again today, which is very useful, that you often have the meaning of rebooting your chat because he's, like, choosing some kind of communication strategy, and This communication strategy, it can be very specific. I mean, if something went wrong, and you start to get away with it, it feels like he doesn't want to go back to a full understanding of the painting. And... And by the way, there are things that affect quality or inaccuracies at all. Everyone knows, yes, that the generosed text before ChatGPT was long laundry. Well, people said it was easy to recognize, you did or didn't do ChatGPT, right? I'm not interested in that much, but I don't care about it myself. I'm easy to say that if this text is processed in ChatGPT or done. Well, there's a big, big difference, there, you got a check or a full text and so on. So, now, there's a new function in the staff. Many people will know that, especially those who do have a lot of things at work and don't tell them it's ChatGPT. Now, there's a thing, and now there's a thing that can turn off the appearance of this long laundry, and that he can't roll a long lane. I haven't got this pgreade yet. I don't know if you're here.
And I'm just saying write as if it wasn't a ChatGPT, and he writes without a tyranny
then.
So, look, everyone knows that ChatGPT doesn't always do that. And very often, when you tell him not to use it, he uses it. For example, I-- periodically I need to gather information, to streamline it from a news perspective. I'm asking him, I'm asking him straight, and then I'm telling him at Deep Research, so I need to be important, and he takes me and doesn't send me the importance! It just doesn't do anything.
Different importance.
You pointed out to him separately, you gave him very strict attention to this as a separate parameter, and he didn't. Yeah, and there are things that if you see yourself that some team doesn't perform, I recommend you find a description that's really gonna make him change, you know, a prom, right? Because he can take different things, uh, well, your requests or your demands.
Sas, it's gonna be great if we share this little video that many may have seen, but it's not a sin to show again about two cubes that write, uh, text and Indian accent. The prompot is sounding. Just how to get the right prompt. Well, I'm just wearing a prom. Well, just smile funny. It's about the long laundry. You said someone there was generating at work and wants to keep them from seeing what it was at work. TikTok is actually a trend. Not that I'm in TikTok, just the news about it.
Come on!
What girls are recording.
Not what I'm saying.
Come on!
One friend of mine saw and told me there was a trend like that in TikTok. Girls record everything in tears, that they're breaking up with their boyfriend right now because he sent them a long laundry message, and that's why they realize it wasn't him who wrote it, he didn't even text them, but... ChatGPT is the one that writes a message for them, so they're breaking up now. I mean, maybe adding this thing to the fact that not only long laps will be used, but also-
If you need to break up with a girl, add a long tyranny.
Yeah, add a long tyranny.
It might work.
Ilnar's
liifhaki.
Speaking of liifhaki, Ilnar, the fact that you're talking about the definition is that at one of the universities, the professor wrote that he showed that there was a study that I think there's a little more people, a man, thirty or thirty. Forty, written, used artificial intelligence, and they were recognized, and they were sent as if they didn't have to. What they sent an apology to this professor, and he took all the answers to the bill of apology. It's so huge. You didn't see, did you? He had issued the apology replies and those replies, and he had provided answers, which were also written by artificial intelligence. And he showed why. Because they contained a certain structure of the response that, well, is not a basic common man, but is inherent in artificial intelligence in generating a professor's response. It was very nice, it was written, it was exactly the same pattern. And by the way, let them say. And then we'll start a cube, a corn-cabbing video.
Cups.
I was two thousand twoteen... in the 13th or 12th, and I was on the phone, and there, well, I was shooting something from our company.
You were there, Sasha, my God!
And I remember, so I just...
Sasha, my God!
Yeah, I just got out, I just got out, and I got out, and the gear is on the top. And then, after a certain point of time, it's shown. And I'm getting a call from my friend, who, well, never can be associated with a transfer. He's such a hard businessman. And he says, "Look, what are you saying, "What are you doing?" I said, "What are you saying, "Why are you watching?" Here. "I was passing by, so my mother-in-law was watching someone. I heard by accident, I was told." Here. And a lot of people watching this show, then let them say. So, Ilnar, in TikTok-
Don't be shy.
You heard it.
Don't worry.
You heard it. Don't worry, yeah. Don't worry.
You can keep it from hiding. Ten thousand people have known that.
Hello, ChatGPT. How to be rich.
Very, very rich.
Very rich overnight. Give me ten ideas.
Fifty
50 ten ideas.
50, 50 ideas.
50 ideas, zero investment.
No risk.
No risk. No risk.
No hard work.
No hard work. Saturday, Sunday off.
Ha.
W-w-w-w-
World, world touring.
World tour included. Shark Tank friendly ideas.
Ha, best ideas.
Best ideas. No hurry.
Take your time.
Take your time.
Deep research.
D-deep research. Deep research. Go