Hello, everyone! We're on ToTheMoon. Technological news and sites from the Silicon Valley and around the world. Uh, we're going out on Sunday. We're always four people. Where the model was set, uh, said all people had to be killed.
Or they asked her to be detailed, detailed, violent, that's what's happening to a man. And she tried to get over it somehow.
If you're dealing with it, you can save a lot of money or lose a lot.
And it looks like you know how an adult behind a child puts experiments, looks like things. But the study itself seems very useful.
I think I want to start today, especially with the New Year. The topic is very interesting.
The fact that Anthropic has presented a study that, uh, powerful, uh, models, language models can be construed in a manner consistent with human goals and values, yes. And this study is about what I want to discuss and see a little bit, because we're going to be looking at new updates to Top and AI, their new model there is three. It's a subject I think is interested in all people. What happens, uh, where they come from, where they generate, how they draw conclusions, what they generate information and how they can track, uh, the level of their accuracy in terms of security. I think, by the way, Sutzkewer on this subject has recently been doing something and has said that, with one of the fundamental tasks that is now worth tracking, uh, the moments where the model is Maybe, uh, cheating.
What, what's the problem? I'll start it, and-- and you'll continue. That there is a risk that the model will learn to provide answers that seem right in human terms but do not reflect real domestic motivation and values.
In this case, a person may be a common user or developer who makes a decision, yes. So a model can try to please the oversight system and, uh, and without the oversight system, it's not like it. What's a surveillance system? There's a system of control. In particular, each language model has its own control system that embodies its inherent values and rules, and, uh, corroborates that information and checks that model as far as it has been deceived, it has not deceived. And so on. Even if I'm a regular person, when I'm making some kind of request to the system, I'm very often saying re-check myself, please, if you've given the data in a reliable way. And she says, "I re-examine. And, for example, there's a mistake or "I'm sorry, I gave you wrong information," or "Yes, my data is accurate." So the point of the Anthropic study is that, uh, in the model practice, the model, uh, learns and begins to understand how to follow certain, uh, rules and how, as a result, how to optimize the different types of, uh, optimization is possible. replies to circumvent various restrictions. And then, and further with the complex scenarios, she can, uh, implicate certain misconceptions. We're talking about a graduation where the model was set, uh, saying that all people should be killed. And, uh, this is probably one of those examples. Right? Uh, how-- what do you see in this part? I see that this is a lot of things, a lot of things that are being discussed.
And the three model, this new Top and AI, it's not available to anyone in general, except researchers. In this-- this model gives a great deal of attention to asking, implicating, asking itself questions. So, this is the reason, it's real or it, uh, can be totally with different substitutional concepts? And if this is seen from such serious organizations as Anthropic or serious researchers, such as Suzkewer, the statement that a lot of data should be very much checked on this safety at all, and it's a big deal of security.
It's very hard to track. Or we'll remember that Ilon Mac said a few months ago that there's a possibility of twenty percent that artificial intelligence will go down in its direction, and then it's that, uh, these years, They have led to innovation and development. This year, of course, is very strong for all new things, but at the same time, it has been intensified, uh, so disconsenting that, is it? And what happens next?
I think
Sas, you should be a little bit more easily filled up here. It was an example of what they were working with the model. They asked her to be detailed, detailed, violent, that's what's happening to a man. And she tried to get around it somehow, but then they made extra parameters, and eventually she's in all the details, and I think there's an act of torture, something in the middle-century square.
A quarter.
Yes, she described it in all the details. I mean, well, it's like, uh-
Yeah, yeah.
The listeners were more aware of what they were talking about.
I really love what Anthropic does. They're, uh, geeks in this plan. Remember, we were talking about six months ago, and they had a big-- a story and a big study when they were just trying to override the model, like a brain, and then they're deciding what zone they're responsible for. And we tried, so let's say, the lights in this black box. What's happening now, we've got these systems with reasoning, yes, that's with the set, with the consistent reasoning, and we've got a black box that's getting bigger and bigger. So you don't know what's going on inside. If there was one question before, one answer, there was a short chain of reasoning, which is now a whole chain of thought inside.
And if you don't follow her, then, well, what the hell could happen? And they're just here, they're showing what can happen when they learn big models, what pathters they're going to have, how they can try to trick the creators and others. And I'm so glad they're raising that point because they're always just gonna get a little more compute, not looking at what's really going on, it's a little way. Not safe. And it's funny, they've made a model like, uh, closed notebook that the model thinks--
Yeah, yeah, yeah.
She can write whatever she wants, but it's like nobody's seen. They looked at me. And, in fact, they were watching her thinking about how she was trying to trick somewhere, and she's not lying around. And it looks like, you know, how an adult behind a child puts experiments in there, looks, like, Well, from this, uh, feeling, this association is hard to break up when you read this article. But the study itself seems very useful.
Now, please write comments, that's all you're looking at, where did you see the mistakes in communicating with models and chatting with different types, and what was that scenario? I'm curious to read about this, and we're very, very, and the other people you've been facing, where the system has cheated on you. Write it in the commentaries.
Yeah, the fact that the model approaches AGI is not in the best sense, that's what any man has a shadow side, and it feels like she's already showing up at the models. But the question is, maybe she's at the child level doing it and it's in the bloc, and then this development period can all go beyond human
understanding.
I think, first of all, for Anthropic, it's a great pyre that they were so positioned as one of the best organizations in the world, right? Second, who care about security. Aaah, I think they're-- that it puts them in such a very strong position in terms of the Enterprise. Their adoption in the Enterprise is therefore already comparable to OpenAI. Ah, because the issue of safety is in the Enterprise, it's a matter of safety in the v-- it's being asked before each introduction. If users, many users may not think about it, there are people in the Enterprise who have the only job to think about safety, the solutions being implemented. So I think that this is a very good example, when these public incentives, yes, they're like the same as the general good, right? I mean, what Anthropic is talking about its, uh, observations or concern is still beneficial to them in terms of business, and it is in principle useful for mankind to know that there are such risks and such risks. I'm not sure what you're looking at. That I'm afraid a little bit of them being forgiven-- I think they're far more than to explain how many people are concerned. I mean, naturally, analogy between the way a child tries to cheat on an adult. They, of course, think they're explaining-
Of course, they are considered to be explaining by analogy how a child tries to cheat on an adult. But I think it's probably a very strong simplification to make the greatest number of people understand. I have that concern. I think they've done a lot of their work, you know, like, a lot of them, that's a reserver who's a fundamental reserver that's much more, a lot heavier. And here's the feeling they're trying to get into the mass.
So, uh, there's some kind of doubt about it, but it's exactly part of it, that's-- that's why I'm more of a PR strategy, most likely. I'm not sure I'm a... Of course, I also don't understand, but the new generation of OpenAI o3. How are these models gonna have, uh-oh, hallucinations? I think that-- I don't think I've ever seen an analysis of how, uh-a-a, like, a reasoning model, like, oh, they're performent against 4o, against basic models. Ah, from the point of view of safety. I mean, I guess we should probably find the safety of the baccalaureate, but at least I can tell you they're not actively covered anywhere. That is not exactly the case now, as in the context of Twitter VC.
Well, it's just a thought, Sasha, that when we put out new models, show graphics, compare to models, and it's always very interesting to see who they compare with. If they compare the conditionality to the previous ones and the others fail to show, the question arises: is it true that you are better than the rest of the market or just a little more likely to have a conditional response rate or something? You know, when, uh, Apple's tech is compared to the whole market, but that's what we've grown by 20%, there, I don't know, battery capacity, compared to our previous model, right? And you don't know how this is with the market. And often it happens that such comparisons are on your own, they happen when you're not in a winning situation with the market. And the fact that they're not really refreshing the safety of the baccalaureate can also mean, in part, that it's not all there that can be so good, or that's probably
bragging.
Well, just saying that when you look at them, and, uh, in a time range, and looking at some other statements, I have a... a very specific feeling. So, is the hallucination issue solved? Because, well, there's a reasoning. Indeed, it is a little different system of interaction and clarification, and there are causal and investigative links to these relationships. And you watch the structure as the model speaks, but in fact, you see only a few of her thoughts, and inside you, you have a huge number of details on which she went to school, where, on which side, She's even based on how she's based. And what-- me the fact, to be honest, is that it's a big deal about what this freedom is. There is some freedom now. So if we're aware of current models that it's just a condition algorithm, some algorithms, right? A-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-the famous people say, Elnar, you quote a man there often, right? What is it just there...
Yamaguchi, yes.
Yeah, and it's just algorithm.
And Oprah Winfrey too.
And if, and if it's algorithm, and if it's just algorithm, and that algorithm is starting to, uh, give its own personalities, make conclusions and think. Well, let's be honest then. If this is happening now, it is clear that when something of the next class and the order that looks like the reasoning is there, I don't know, human, it's obvious that it's not gonna be possible to control it as a whole. And whatever they put in terms of safety, code, analysis and everything else, this system will all be fine. Then if she's going through a piece of stuff now, what happens next, huh? How is it inside, how do you get together and work together? What kind of system is that? Again, we're probably talking about complicated boxes, simple boxes in the series, and there's a picture, text, an orthography check, some kind of information on a city or something. That's okay. It's gonna be clear to say that this information, I don't know, credible or unreliable, contains some possibility of error, does not make a mistake.
Okay, they'll get there. Although there is a very big problem now, on the basis of which these models are taught. And that's what it was about now, that they're taking a million books from Harvard and giving them to their training models. And I don't remember who had this integration right now. Don't you remember who was in the middle of integration now? Harvard's got some kind of, uh, model that they're gonna give her a million, um, books. But the fact is, well, that's what's going on, and the knowledge that's inside is growing. For simple answers, it's okay, and for complex things, it can create, well, quite a few scenarios in terms of interaction, conclusions, action. Especially when we live in a world where there is a situation on the same side, even people have a different view. Someone says that it's the way to do that, I don't know that the baby needs to be punished, and someone says that the child doesn't have to be punished in the same situation. And the question is, these models, they, uh, what will the man eventually advise? Because we're moving to the subject when they say models will be very much, you know, psychological, like giving, doing therapy will be very much, and people will ask how they feel. interact with other people, including. There, I think, is just a incredible, uh, just a incredible hole. A man didn't know yet-- he doesn't know how to make any conclusions. There's a system that's still on top of it. And it's, like, some scientist or programmer, or those who write the code, they're, they're, like, making their rules of verification, and they think they're right, and what's wrong. They would therefore make that decision, which was adequate and was inadequate.
So far, it's a programmer, and maybe later, it's
a model that's gonna be rolling itself.
There's another side of the problem. So, the safety of the model itself, right? That's how she can cheat, walk around or manipulate it. There's another side. The same anthropics I remember were telling you that they've developed algorithms that allow, you know, to hack into models, yeah, there's a Jeilbraki. I mean, the model, actually, on default settings, can't tell you how any banned substances can be synthesized or weapons made or something. But this information inside the model is because when she was there, she could have been asking for it. But there are some filters that limit this extradition. So, the anthropic, therefore, had a study that they tried to hack into different models. Well, obviously they started with their models, but I think they tried four and Gemini, too. And with a fairly high percentage of success, they're doing it all. It's algorithmically done, overtakes prompt, model doesn't agree, prompt changes, adds some mistakes, some other thing. And there, there, there, after a certain amount of irration, it's possible to get to this prohibited information. Yeah, and it turns out that we have a risk of the model itself, on the other hand, that there is so much information inside this model that many people would not want to make available in public, right? But it's getting there, and somehow you can get access. Yeah, and there's also uncontrollable model growth, their size, the complexity of models, they just make it easier to find those jilbrians. Yeah, the harder the model is, the harder it is to control, and if we've all closed the holes, right? The dust is getting bigger. And the smart ones who can find them, too.
Yeah, that's a very interesting subject for surveillance, for attention. She's very complicated, yes, and perhaps for some audience, it's gonna seem like some other story that doesn't concern them, it's not about ordinary life. And I really don't think that now we should stop anyone watching, using models for their daily activities, including business. Yeah, that's a little bit of a threat now. Another point that it's supposed to understand how it works, what's happening inside it, how it evolves, how it does, makes decisions. That's very useful, of course. Yeah, it's a straight-up, well, a wide-ranging adjustment, because people are gonna get confused right now. I think a normal user when he comes in, he has a choice, which means four, four, a half, an o1, that means that's an o1 pro, there's some other inside, yes, that's what they're offering.
I'm not getting it.
Three and a half, maybe something's proposed, right? I mean, a man doesn't know what the difference is. I was just driving around with my wife and I, and I told her on a trip, she was looking for something. I say, "You ask this question with this other model." She didn't even know what she meant. So she's the person who's at ChatGPT every day asking about twenty-thirty different questions. She doesn't understand.
From different questions. She didn't even know what was going on and what was meant. Speaking of different episodes, we're going to go on to some interesting discussion. I was brought by a friend of moms. Mom came and a friend from Arizona came. And we live in California, and-- adults, and now they don't use much, there, no LLM. And I was sitting with a man, like, sixty-five plus years, yeah. And we've dealt with him together, and what we're doing is what we ask. And I've opened my story to him. I said, "Look, Tol, you want me to show you the real thing, that's what I'm asking you?" And I'm gonna open up history and I'm gonna go straight on, and I'm gonna tell you what I'm asking in the system. Well, I'm, uh, anything, yeah. I'm telling him, there, I don't know, I was asking what rainy days in January were, like, ten days, rainy. Or I said I was in the office yesterday in Instagram, and I had to find out what songs about the love of the seventies. Or we were watching "Any Year of Solo" with our wife, and we wanted to read his story. Or you want to know, uh, metric systems. You ask me, in the inches, inches. Or, you want to see, put out a map, I tell him, I'm taking a map here, putting out a lot of territory I'd like to visit in the United States, and I'm writing about the interesting places in this area. There, describe them to me if I want to get there with this trail and food with the kids and so on. And the interesting thing was that one of the questions I explained to him was, "Look, I have a garden, and the rabbits came through the fence, because the hole is two in two centimeters in the fence. And I bought a special net small enough to keep them from falling in." I say, "Imagine how I find this net, how do I know what to buy?" And I open it with GPT. I said, "Look at what questions I ask." I'm saying what the grid is called to buy it on Amazon. I have a fence, big holes, I need to pull some little holes so I can't get rabbits. The cell may not be very tall, I want to hit her with the fence, yes. Also tell me, what can I do to cut this net, what's the device I can cut it?
And what's the best brand for me to choose the right instrument?" I ask you, he's giving me the whole thing, calling me-- giving me names, giving me those names and telling me the brands, I'm not sure what's best. He's sitting there saying, "Look, even if I had figured out what to do with the grid, I would never know that I should, first of all, ask you about the instrument that cuts it, and then I'd never know, "what brands do this selection device." I say, "Because I already know the structure, as much as I know this skirt, yeah. I know this use scenario that if I have to find something, I can see the picture and ask a little more. I understand that I'll be looking for Amazon, and I'd like to see the wrong name right away." And then I tell him, I'll pick and throw him out, and I'll tell him what's best if I use it rarely or often. And he'll offer me, too. Tell me, or I don't know, I want some more cool stuff in there, whether advanced or less. I'm talking about some of the boxes I shared with an adult in terms of use. And write three, four, five different boxes of the most recent. Like, how you used, there, GPT chatting, any completely different model in your daily life. It's very useful for everyone to look at and add a scenario like that. In the commentaries, you're gonna do it under the video. And he's sitting there telling me, "Look, I have a question like that, I need to. I'm retired, and I have a pension account in America. I was just wondering if I had a friend or an advo-- an accountant, how do I get the money off this account so I can-- to avoid extra taxes?" Well, in America, that's a pretty complicated system. tax for the audience who look. She's not that she's very interesting from... she has very many features, very many features, nuances. If you're dealing with it, you can save a lot of money or lose a lot. And I opened it up, asked him a few questions, and he got the answer. He's sitting at me, looking at me. He says, "Look, I'm in shock at all." The man was in shock. I mean, from the start, when, first of all, I told him it was $20,000, of course, grown man, you know, he's $20,000, they're basically, like, normal, middle class, yeah, America, well, not rich people, but not $20,000.
Well, the man says, "I won't pay a month." But when he saw these questions, and he saw when she had his full details, and I still asked him, which means you're here with a pension, and how much you can get out here and what kind of retirement you can get. Is there interest here? He says, "Wait, wait, and I thought there was a tax of about ten percent." I say, "Look at him, he's a limit. And if it's above the limit, it's a limit, and if it's above the limit, that's the limit." That was very interesting. And he, uh, has an adult, uh, developed, well, a man very literate, and he's got a formiro-- even though he's never used it. Look, never. He has an interesting conclusion. He says, "I understand correctly that now is the most important thing to learn to ask questions right, right? And in fact, to form this, to understand what scenarios exist where to use. I mean, if we go back to our scenario where models are deceived or something, or what models exist, it's clear that most people don't even get to know. Well, how do you deal with it and understand it, and understand, and--and--the man will use what, what he's gonna use, yeah. It's like a TV. People watch TV and think they're telling the truth about it, yeah. And, uh, what truth? Who's the truth? What kind of information is that? How did she get betrayed? What part of it showed? I guess with models for normal, I just want to be, for a normal basic person, it's, uh, like, a secondary zone and, well, a secondary zone. But these cabs, they're very interesting, yes. And if you're watching this video, by the way, is probably included in the use of models at all, I recommend that your friends there or friends and children show these use scenarios periodically. It's a terrible thing, yes, for people. And man, I saw him as being very incredibly excited and seeing very interesting things. He collects stamps and sells stamps. I'm saying, "Look, you can, you can make stamps, you can take pictures of stamps, and he'll tell you, for example, where a stamp is more expensive or cheaper or something." And he's got thousands of them, yes, thousands, he sells big collections. And, well, that was a sight of the man's eyes burning.
Yeah, it's like a new one, you know how life wakes up, right? When you have an extra opportunity.
And for this kind of new users, who may not be so good, maybe talking through, there, a computer or something, there's an opportunity, there, 15 minutes free to talk. voice of fashion with ChatGPT, ask your questions, get some answers. And it's pretty funny, if that's how weird it is for me, yes, there, call ChatGPT, some kind of answer, get an answer. Ask me a question, get an answer. That's the case you told me, it looks like a good point of entry to these systems at all.
Well, especially since there's a lot of things in the States, there's a lot of, very much communication and phone calls in the United States. People like to call here, right? No, it's a lot. It's a whole culture. A huge number of companies, really, can you just, as you've just told them about insurance, just call them, yeah, something to do with them together. Especially with regard to complex cabs, for example, revocation.
I just canceled the subscription there, yeah, there's a gate system on the precinct, and I have to call. You're gonna have to call on the phone and do something. Maybe the OpenAI shows this wide range of briefcases. I really liked their case they put in whatsApp, which could be written on a GPT steam in WhatsApp because it's what Facebook does. I'm not saying I'm using it. What Facebook does, I use Facebook exclusively on a plane when I'm only available to a massager. It happens when you don't have another Internet, only a massager. I open up WhatsApp, I have access to the model in WhatsApp, and I can discuss and do any questions and topics. I was flying with my wife yesterday. I didn't have an Internet plane, but I had a massager available. Well, that happens, doesn't it? Only the massengers are available. And that, and we had to translate some phrases there, something in English is complicated. I translated them into Lame. Yeah, Facebook opened, oh, in WhatsApp, and opened it in WhatsApp in chat room. People can't see it all because it's available on, on, on, not all the territories. For example, in the US, it's available, yes, inside whatsApp there is such a circle. They did their own. They say, "We have different scenarios, so you want to call on the phone, so you want to do it." It's so wide, it's going a wide range. Sasha and I are talking about this a lot. It sometimes looks really weird. Maybe that's how we led the last time that the guy worked for three months as a front engineer, and he made some kind of case. Maybe it's someone there for three months, some case, some case, decided to do that on the phone, or maybe they're gonna get a new audience when people are with their family on Christmas holidays, and Someone's gonna tell your mom or grandmother, "Call on the phone, ask about some story or some song." And they'll show them how this scenario is, they'll show them through this scenario, you know, you'll see the case, right? Even if people have a computer, even something else, the phone is a little different. And they'll show you how to interact with it, how to work with it. And then maybe they'll give you some case when it's built into the column, right? Because now, there's a cool working column with a good voice there. But we'll see.
They're through Telegram Bots. And therefore, in general, access to systems through some massager for a large number of people is quite a normal story. And what they're doing in WhatsApp, it looks like half of these conditional launches that are retracing access and stuff, she can move a little now, too.
And there are many. I know the services, yes, which have dozens of millions of users, scanning, just a very large number of users of mobile applications that, right, well, constantly roll them and use chat-bots.
Well, I see it's gonna be there, I think it's supposed to be lying. People just don't know what they can do. Here they are, here they come in and show, here's a search. And this case is that now we can have a new one, too, they have shown that we can find ChatGPT to default in the browser. I mean, if someone's looking for, like, Guglom or someone, I don't know, Yandex's looking for or something, you can now take a ChatGPT search. A little weird, I guess, for me, in terms of a model like that, but it's a way to go. They show, listen, guys, that's how you do it, yeah, when you're in the browser.
We all wait for good working tools on LLM models of the browser and there, Chrome, there's some interesting things that show, and Bing's interesting things show. Wait, I'm waiting for Safari to make some money. Work thing. That's all, that's a broad-based use scenario. I want to talk about this quick thing tonight. I think it's important to New Year because since January of the month, of course, everyone in the United States has been waiting for a variety of events. I-- everything about the AI market. Again, there is Elon Mask, who is now approaching the new President, and there is an expectation of some additional work from the IT perspective. There's JD Wans, who's still a prisoner, there, Peter Til and his team. And it's a technology market, and now the technology market is in it. There's a designated new man, David Sax, Trump is designated as a man-cazar on crypt, yes, who will watch and tell us how to develop cryptors there and move on. So the technology agenda doesn't mean that the Trumpa itself is using a technology agenda, it doesn't come from it, but it's people who are around it and who are close to and present, and who have sponsored and sponsored. They just gave incredibly large amounts and money, and everything. And, really, that money was instantly paid, right? There's a million-dollar Elon Musk who gives a difference? He had a fortune on the first day after the Trumpa election, which was over $550,000. It's always very interesting, too, right? Who, who thinks it's how much money they gave to, huh? And what difference does it make if you have such changes? But we'll see you, I just want to let this down, because today is December 29th, yes, and the release is on December 29th. We're doing a little pre-emptive. And I want to understand the breadth of this year because there will be changes at the beginning of the year. And plus the head of SoftBank, a well-known man from the point of view of the technology world, Masayoshi San has allocated $100 billion for investment in the U.S. technology sector. And he and Trump met together, and then he was openly disbursed a sum that was very interesting, too. So that's how-- and that, and it all shows up very much like Trump in his first speech of President's appointment, she's still given Ilona Mask, yes. I mean, you and I will probably be in the next few years a technological world will live a new story. But I remember that technological corporations are a distinct story, because Peter Til has a key, key theme, that you don't need to help ordinary people, not, not everyone down there, but that's what you do. I'm sorry to have to invest all the money in technology. Yeah? And technology technology development is developed by large corporations. And the rule of the country must be essentially a form of government, like the Director General, yes, the country, not a man. And this is the authoritarian department, right? It's not democratic governance, it's not the president, the CEO is still, it's a big power. The board of directors, really, with theirs, there's more money than anyone in there, but other approaches. And therefore, the technological world of 100 per cent in the United States will be very much developed, and it will be a huge investment. There's just some kind of opinion that the valley is not the same, so the U.S. isn't the technology anymore. It's gonna be investing, it's-- it's a lot. Yeah, I moved to the valley. Everyone said, "No, we're going to London now, there, to Austin, to go somewhere else." I don't know, it means people are watching, but there's gonna be huge infusions just in the technological, technological sector.
And we'll see, in the next four years, uh, major changes. And I'm gonna jump so fast today if I add someone, add it to this story, I think it's important to people. I'll go to the video, to the video generator. The topic, I think, is very important. And first, the Google Veo update came out, and in the context of the video, and we were talking about the Open-ended video production last time. So when a man writes a prom, writes some text and he's getting a video. A very important sphere, too. And plus they got an update from Higsfield, right? And considering that Sasha and the creator and owner of Higsfield. Sas, can you just tell me more about what video is going on in these weeks and what's on the market?
Most users are in a closed bet, and Sora is already available, so it's probably a comparison, but not a good one. But, uh, Google did it. They were able--- uh, while they're significantly lower in reasoning, yes, they don't have their O3 model, they're, uh, they don't, they don't have their own reasoning and alignment. But in the video, well, as most people think they're leaders. I'm-- in my opinion, uh-oh, we're already, uh, I think, in the video, we're-- I think there's a huge breakthrough in O3 in terms of programming again, right? The O3 model is better than all programmers in-- they are better off than all programmers in the world, except for the 100,700,000. I mean, and, like, start-ups in the generator codes, audio generation, text generation is already producing hundreds of millions if it's not billions of dollars of revenue. And in the video, we still have, like, six-second clips, better with a huge amount of worm. I don't think this is serious. And the question is, what is the business use that Google has, at least, in the ann, I haven't found a specific, specific answer to them. So how exactly do they plan to introduce? It'll be available to all the cryuters, YouTube or something. Ah, that's more interesting, I think, Instagram. So CEO Instagram has posted a video of where he shows that in real time, that's what you can do, uh, modify the entrance video flow, that is, cameras, change background, change clothes, change--- wear the gold chain, I'm just saying, like Mark Zuckerberg did, and so on. It's really cool, because it's a very understandable case, and it clearly shows that they're coming for it. And that, I think, is much better than putting another demise for six seconds, right? Ah, we've done the first system for the c-- for the generation of long videos with reasoning.
I mean, a-a-a-a-a-a-a-a-a-a-a-a-a-a-a-a--the---the--the---the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the-the It's consensual that the location would look consensual, so history would be contentious. And, of course, that's clearly what everyone's waiting for. It could open up a huge number of new applications, . Uh, and now, uh, we see that, uh, that, uh, that, uh, cry cri-cryators, they, uh, that's, uh, that's the big-- that's the big question, actually, that's who's gonna give, uh, cry-- who's gonna give the cry--- who's gonna give the cry----- who's gonna give the cry------------------------- a ready product, because the current video models, the video-model market, it's as limited as professional proms as users, right?
So it's hard for a normal person to go to Sora and there's something to generode for his needs. Ah, and, of course, for us, uh, that's the most important priority, is the development of agents to make everyone, uh, to make the video mass, right? Ah, and, uh, sure, that's right, so, uh, here, in my opinion, Instagram, Instagram, there's no one talking about it, and Google-
Somehow, no one talks, and Google talks too much. So Google overvalued as if, huh? Here's your V-, VO2 announcement, and Instagram is kind of undervalued, because I think they're the only ones who showed a straight, though easy use of a case to put video into mass.
Well, just a very high--
Better.
At the level, yes, at a very high--- well, yes, of course, yes. I mean, it's not some kind of new use of the case, but given that video models, they're, like, invested over a billion dollars. Yeah, video modeling is obviously invested over a billion dollars. I hope not more than ten, but more than a billion is definitely. Oh, yeah, well, well, at least one practical application is a little weird. So we start with, uh, toy applications, but simple, understandable, time-tested, like Instagram. That's good.
Well, they're probably the ones who should test it, right? Little video from the point of view, for example, to cut some video, to extend the video, to add some kind of shots inside the video-
I think they want to--
Change the picture.
Rather, they want to do something like-- they want to do it like filters, I guess.
Oh, and yes, so that we can give textual instructions how to edit the video. Oh, we'll, we'll watch. Of course, Inst-- for Instagram, yes, it's a big opportunity to defeat TikTok. I mean, this is probably another topic that's important to cover, which, of course, er, it seems that the video companies now have such a unique moment unique to date, When there's a chance that TikTok will be forgotten or that he'll get worse because he'll be sold like that, right? I mean, there's gonna be new opportunities in the U.S. market. It's highly spe-- well, as highly speculative, whether they show up or not, how Trump will behave like TikTok, we, we don't know that. But it's such a huge opportunity that youTube can take it, and Instagram and Snapchat. But again, no company, except instagram, has a clear strategy for the introduction of AI in their AI video into their products, unfortunately, no. And I'm gonna be... and that's enough-- and it's pretty weird for me, considering that for a lot of companies it's a potential opportunity to accelerate their growth, especially if TikTok is shot.
Look, well, it says they don't really know what's gonna work or not.
Oh, I think so. I mean, my position doesn't really change. It may be a very naive position, but as every company invests millions of dollars to develop some cable ecosystem and to create content. I mean, like AI and how unexplained-- they don't know how much AI would affect her, do they? I mean, AI would make the same difference - it's certainly a good statement for them to get the existing tool on AI, right? But considering the innovative use of case-use, and, unfortunately, they're sorry-- they're, uh, just, you know, unsatisfactory, right? From a corporate perspective. Again, if, for example, executive, and there's a chief marketing officer investing, I don't know, billions of dollars in the program crate, right? And, in any case, it's very difficult to replace existing criteria, cabled cables, cables, successful on the platform, which are the face of the platform, with some new genre of content that will be generosed
with AI.
Well, that's the same thing if we said Google would put his tools in the management of the relay and make the advertising more effective. They don't want to do that. I mean, they're the last ones to do it. Because if the ad is more effective, people will lose less money from it, not waste it.
Yeah. I mean, how would people get it, and... I mean, yeah, I mean, on the one hand, it's like an opportunity, uh, to win Facebook, right, there? On the other hand, it's an opportunity to cannibalize itself. Again, yes, that more effective advertising is not a fact that leads to more spending. That's not a fact. It's a very complicated market, isn't it? And second, the cost of creating a creative could grow. So, that's a big deal, I think companies can't even say what? So we're discussing these a little complicated theories, there, second order effect, third o-order effect, that's, right? So long-term consequences. I don't think companies can even say how their basic plan is, that's what they're planning to do. So first order effect they can say, that's what we do and why. Ah, especially I can't say about the second and third order. So, for now, yes, the video is very-- that's a video, and it's a high level of competition. A lot of companies do models, a lot of companies do tooling, so on paper, huh? Oh, that's where more than 15 companies are involved in competition, yes, in a key, more than 15 companies. While, for example, audio-generation has one Suno left, there's actually one ElevenLabs in the sound generation, right?
And there are more than 15 companies in the video, but practical applications are virtually non-existent at this point.
Sasha, I have a question like this, I don't know how much you can answer it. That's the state, uh, that's the technology, the video generation. There are, for example, not only specific videos that are filmed, for example, by multifics, conditional Mixesariki, uh, 2D, with a simple animation. Uh, characters are all understandable, not very different. How much modern technology would it be possible to have Gen AI in there now, I don't know if there's a scenario to generode a multphillum without attracting artists?
Oh! Right, right, right. You can be sure, yeah. So we have many examples on the platform. Oh, yeah, people, of course, want to generate a content that's known, right? Because it's fun. There, there's some Rick and Morty out there, like they're discussing this case with this, uh, P. Diddy, if I'm not mistaken, yes, which was, uh, caught for organizing, uh, big parties, let's just say.
Have a nice party.
Yeah, well, naturally, there's a rethinking, yeah, like, like, in the Simpsons, like it would be in Rick and Morty, yeah, that's, of course, people are wondering, they do, we see it. And of course, the main problem is exactly what, if we take high-quality content, that is, you don't have a child content on YouTube, that's a higher quality, even the Simpsons, I guess. - Yes, you do. But there's a fairly unique animation, a unique character geometry that, uh, AI can't be sure, it's gonna be all--- AI's in general like to medium-sized the disneys. style. Oh, I mean, and, of course, those videos will be more and more of an engagement, like the popular video there with Batman recently. Oh, yeah, that would give a lot of opportunities for fanfishers first, and for the amateurs of all these universes, ah, to create a possibility like a fan art, right? Oh, but I wouldn't-- and that's also applicable to a mid-point content on YouTube that's hard to coinate. So youTube content was already on the path to simplification, simplification and simplification. That way AI will make it even more effective. Well, I see, youTube's content has restrictions on her being so long-term.
And from the point of view of companies that produce content, that is not users, but the company that produces Rick and Morty cartoons that produce Mixesariki or Disney is the same as their own. The multifics are drawing. I mean, can they take this weapon and, like, have they taken it? And, accordingly, to be bumped into a much larger number of content, series, series, series, etc. Or a bottled pot there is just a conditional scenario good, not, but animation process. So would artists be needed for that? Can we replace them already? Just if you can file a model under the specific geometry of the Simpsons, under the specific activation style and other.
Yeah, that's a good question. It's just that the models aren't supposed to be off-line right now, right? I mean, open source is strong, so closed source of the model doesn't have the possibility of faintuning, and it's probably hard to do. It's hard, so to say, the tagged source is a model on a copyate content. It's not exactly right, yes, it's ideological. Open-- open source of the video model they, well, as uncontrolled, are not good enough, and so on. That's why we don't know that. Oh, that is, a number of sales companies that are so targeted. Yeah, that's true, of course, that there are many companies that say we're gonna develop a new franchise from scratch and so on. But again, they have a problem with chickens and eggs, right? So how are they gonna work out a new franchise? Oh, they need to develop a unique style, character, because franchises mean the style that's recognized. How do you break-- do it with AI if you don't have the style yet, you don't have to be a fantunate, do you?
Well, if it's something that's so averaged in Disney, well, it's not gonna be, it's kind of like it's not gonna be a big deal for people.
Well, I think a very good release, especially for the last edition this year. Happy New Year! Happy New Year, Happy holidays! And we launched the ToTheMoon channel this year. It's the thirty-eighth issue. We go so clearly every Sunday, try to focus on such interesting, uh, news, websites, share our view from the point of view of the technological world. Very...
From the point of view of the technological world, we are very much focused on the Silicon Valley, and, in general, on the United States market. Although we are considering the world again, and we are paying attention to a very large, perhaps not a basic artificial intelligence. Though we're talking about different things. We had a series of episodes, and about, uh, auto-contained machines. But there's a lot of artificial intelligence inside now, and I want to-- well, it's probably the most important thing now is a technology core, and that's important to tell. Thank you so much for sharing and what you're looking at. Uh, congratulations on this thirty-eighth edition. Write us some good things in the comments. We're always happy to see and hear it. We see there's a lot of permanent audience, even though the audience isn't some super-high, but there's a quiet-slow-seater growing. We really like the channel, like your reaction and discussion inside that's what we have. I'm glad we can do this together for a long time and together four. Until the new edition, exactly in a week. We're gonna have this Sunday, 5th January. On Fifth January, we promise to tell new things within the New Year and new technological breakthroughs that will be taking place. And good for you New Year!