Hello, everyone! You're on ToTheMoon. Technological news, Silicon Valley sites and the world, USA. Hey, guys, everybody! We're going out on Sunday. We're always four together. You find the places he might be interested in.
It's just that everything will lead to everyone starting to use it because you either use it and work or you don't use it or work.
And if he gets something interesting to do, for example, he'll say, "I want to do a new search." Yeah. For example, Amazon is not interested in selling the sellers of good quality. The Yandex is not interested in Google, and it's not like people are doing the right thing to make the commercial.
Why don't you deal with the model right away, I don't understand.
That's if I walk through this Olympic.
I'm thinking about starting up with the subject today, I think it's very interesting. It is stated that one person is reported through AI services, automated services, such as epling, registration, yes, or applications, applications and a summary of his or her work. Jobs, two thousand eight hundred and forty-three times through this automatic service. And I think it's such a very life-long, interesting precedent. These are, in fact, the things of mass stories through AI, there are many of them. And I guess competing with a lot of things is gonna be useless soon. So it says there's a service called AI Hawk. In fact, those services, their entire, well, there's enough of them. Yeah, that's not the only one. I've met a decent number, discussed, discussed and even co-examinated with Alinar, and I've been doing some things that are related to the automatic creation of a accompanying summary, and so on.
So when a person responds two thousand eight hundred and forty three times automatically, what happens? So, and on that side, there are very often automatic services that recognize these responses, well, from companies. They start using systems that automatically recognize, you know, or select candidates for work. It is clear that the letter will become smarter and more smart from this man. This is the process. So the man came in, he checked out the type of work he wants to respond, made a number of actions. There, at the first hour, he did some 17 counts there, 17 jobs. And then the system starts to build up and starts to make automated responses, that is, to find, to find the vacancies that might be of interest to it. I don't know exactly what I'm doing. And then...
There's a--
This system, yes, yes, yes. So they'll automatically sign. I mean, it feels like you've been personalized writing a letter for your work. Here. And as we know, there's a lot of discussion on the market that you want to get to work, you can get back to 100 different jobs. Well, I've responded to two thousand and you're looking for feedback or not, and the answer's not here, and he's still a 10,000 response. Clearly, there are restrictions on all these systems. Once again, systems are large, yes, all of these systems have some limitations on the number of responses. And, in fact, these systems are interesting because they are often integrated with job-seeking sites and with people like LinkedIn, for example, in the US. And through these services, it's easier to get through this automation.
Although it is clear that some kind of service can be written by itself. By the way, I have a feeling about LinkedIn, I'll give you a word. I have a feeling that LinkedIn is for me personally, and it's a kind of a washing, because I get an endless number... people are signing up for me. Once I'm making the exept of these people, which I agree, I'm getting an automated email from them the same minute. I mean, it's an automated chain. And now LinkedIn is just making a simple, very cool automation. But again, it's becoming what? I'm not even reading this content. Sas, you read LinkedIn, considering that you might have some kind of interesting contacts in theory, you read messages that send?
All the messages that write to me, yes, I read them on LinkedIn and email. I just think it's something middle between the inbox and the spam. That's what I'd say. What's in LinkedIn getting. In general, LinkedIn now is a yes, not a very good reputation, I guess. I mean, from there, I think it's a very difficult LinkedIn to sell. People don't turn up news from LinkedIn. I think they're much less than they're leaving Twitter. Well, at least that's SF Tech Bubble, that's exactly what it is.
What do you think of these summaries and the vacancies? How long will these services be alive or is there any kind of transformation of this hiring story? Because it all starts to look very specific, yes, when you can answer two and a half thousand jobs.
Well, I'm just wondering if I could remember the equilibrium of Nash. And if you remember, there was a John Nash movie called "Man of the Sense." And if this is a mass, m, a mass response to the vacancies gives some advantage to the prospector, for example, yes? So with a service of 2,000,800 responses, you get a job, and without it, you get a better job or no job, that's just what's gonna make everyone start it. use it. Because you either use and work or you don't use or work. That's on the one hand, right? I mean, if it's real, uh, it's a good thing, it's actually gonna be a little bit of a little bit of a little bit of a little bit of a little bit of a time-consuming. On the other hand, you've already started saying that, uh, the second side of the process, it's gonna have to react to it, and probably just the same automatic dropouts, yeah, there, if there is now, maybe. Be there, in key words, and most likely, many are beginning to use LLM to assess, and what response is it to be, was it time to spend, to intrude it, and so on. That's gonna get bigger and bigger, too. And then there's news that some HR has been working 10,000 in the day, uh, applications for his job, of which three have chosen. And that's what I'm talking to.
I have a-- that's the question, and what's the point, actually?
Because there's a man sending two thousand applications. Well, first of all, he has to have some super universal profession so that he can get two thousand jobs under his request, right? I think that if I, for example, publish this application, I don't have so many of my potential positions. And the second thing, well, the difference is that, like again, I have more experience in Russia with HeadHunter, that you're hosting an announcement and that you have hundreds of people there, you know, maybe not thousands. The third point, that when I was in the USA, filled a summary of a position I wanted and eventually got to do, is a very standardized form of this summary. So there's no creativity. And as much as I don't know what the point is to make something like that, to take it very hard, because it's like, in the end, my interview was five stages. So the first step is, well, this is the resume. So it's the base, it's practically, well, it's just, like, spam down. And then four interviews, either taking you or not.
The point is very simple. If you, for example, are looking for, you know, first of all, you'll find a system there, too, thousands of jobs. The question of the criteria that are. If you find that you have a million dollar-rate criterion per year of salary, you understand you'll find little work. And if you set out the criteria that you're going to work as a designer, a software manager, a sales manager, anything, a mass, a lot of jobs, a lot of work, he'll find you a lot of work. Doesn't mean they're all right. The point is, if 100, ten, 100 thousand people or there are million people who start--- they use the service and two thousand times respond, then they'll get totally unrelevant applications on that side because The system will be clearly responsive, including to work that you would not have responded. I-- because today, all, uh, all the vacancies, they are actually standardized with a very minimum number of data. I mean, really to determine a candidate, you really need to date this candidate, give him some tests there or even talk to him. And still, maybe in the Hindu-- in the industry, I see, mass industry, where it goes, I don't know, but, uh, some standardized process, a standardized process, a standardized process, maybe it's gonna be possible on the n--- well, that's how it goes. I'm sure you'll win. Ilnar said that if I was supposed to win...
What if, conditionally, only the one who comes back in mass. I mean, if a man-- it used to be a thousand times-- you can't make two and a half thousand calls, like, for instance. I should have called before. You opened a newspaper and called--
Well, it's been a while, yes, it's gone past.
Now, I, I say, and now what is two and a half thousand vacancies going through, responding? That's a problem. And there's a service restriction. Like some service says I can, you can answer 30 times, 50 times and so on. And here you-- external tubes give you a lot of push-ups. If someone did two and a half thousand, then someone could do 50,000, or some crazy amount, including filling standardized forms. It doesn't say you're doing some know-how. It says in general that a man did this, and it caused a certain resonance, yes, because, well, it's a very special special special special, yes.
Well, plus, Sas, look, you're the man who, I understand, hired people, right? Anyway, when you publish a request, you come in, you know, there's a percentage, maybe 80% of applications that you don't like, you can already, at the resume stage. Right? There's still some kind of, uh, spam, I suppose.
Yes, I do, but you can turn into a tough spam. We're talking about this being more of a tull that-- why is this about LinkedIn? I brought this example because it really became a notion of a tough spam. So they let it do, so they let them do such automated vortexes. When you made an automated vortex, I mean, I just have a very large amount of stuff, just a spama of automated. When they come to you, you agreed, you just--
You have it, I don't know everyone.
It's clear he's somehow filtering. Okay, I understand the story that, uh, I'm filtering some spams down there in LinkedIn, okay. But when his number gets too big, you stop using that tool. I mean, if from the employer's point of view, if I just have a huge amount of unrelevant things, those services will in fact stop helping. I mean, on the one hand, like Elnar says, they help if you have the opportunity to respond. On the other hand, if everyone starts using such mass services, they're just tens of thousands, people will have to change things on the part of their employer. I guess now for people who are watching us, it's a very good, super-crunch tool.
Yeah, well, that's a good job, you know.
A win, because less than one--
Less driving.
Yeah, I think less than 1 per cent of people, not that there's 0.1 per cent of people using this kind of service. And I guess in some professions, in some cases, I don't say anything about this subject in some professions, but it seems like it's a good idea, uh, to draw attention.
I want to go on. We've been discussing models a lot. Clearly, we're focused. The last time we had, uh, a lot of stuff about, and, uh, Apple, what he was doing, and about XAI. Last time, Ilona Mask was very serious about his company, saying they had a big prospect. And here on these days, uh, NVIDIA, uh, mass shows that they have an open source of Nemotron over 70 billion parameters. And I had one AI engineer at home yesterday, and he says that the interesting thing is, in fact, uh, it seems to us that some players can't do anything. Sitting NVIDIA, which is a hardware guy, says, "Look, we have the largest number of capacity in the world that we have and play a part." And Sasha Masrabov always says that we have a role and power, yes. They, they, we, they're gonna let us out of anything, uh, anything more massive, more cool. And yesterday, NVIDIA wrote a post about, uh, or the other day, that he thinks that something very serious and interesting can be done in, uh, industry, uh, health. And if he gets something interesting to do, for example, he'll say, "I want to do a new search," yeah. Here. Or I, uh, want to, or I want to get in there, I don't know, in a mass, I don't know, in an e-commerce, that's, like, some kind of field. He could use physics too. What do you think of Nemotron? Sasha, what is technically discussed on this Nemotron 70 billion parameters that are supposed to be the same as me here, like it's hard for me to say, like, uh, totally like GPT-4 on quality.
Yeah, that's the truth, yeah. I mean, I think that training these models is just commodity-based, yeah. I mean, it's been a long time since you've said that the world is working, it's a conditional thing. Like every CEO with half a billion dollars, he can easily get a state of the art model. Well, somehow, yeah, maybe half a billionaire is so fair, yeah. Well, aaaaa understands that for Zuckerberg and for, uh, impoundly, Zuckerberg has one of the biggest clusters in his own right. And, to be clear, Jensen's ow, too, I assume, is an unlimited number of H100. They can do what they want. I mean, uh, it's just, until, uh, let's just say, that's what's not working out for anyone, and nobody's got a little model in open source and take on, you know, like, get usage. That's not what anybody's got, is it? What am I saying? By the fact that, models are less than seven-and-seven billion parameters, uh, it's hard to get impressive results. I mean, there's a Mistral, which is like a separate start-up training models. They decided to go into small models, but for now, they can even start on the phones.
But I haven't seen the same practice yet.
But it's, uh, it says that, uh, now big players, they have the resources and power to do theories, to make some very cool decision in the mass. But until no one can and no one can do anything, uh, mass in terms of the usual users. And even the same ChatGPT that was released by OpenAI. OpenAI is clearly a research company, it is not directed towards ordinary private users and ordinary private users. So once another company releases something that is superior to the ChatGPT itself, then people will forget and use another. The question, uh, is this fundamental here is whether you can knock or can't do it?
Well, you know what's not clear? While it's not real that anyone can, someone will come and be able to take ChatGPT a palm, you know, just pick up the users. I don't know that yet. I mean, it's clear that anyone can have a half--- anyone who has half a billion dollars for training can make a model of comparable quality for the baccalaureate, right? That doesn't raise any questions. But, uh, jump in, uh, get-- what, but, uh, but at the same time, n-n--- it's not clear what you have to do to get away from, uh, ChatGPT's palm of the first thing. Then, well, that's the problem with a lot of baccalaureates, too. So they don't reflect the baccalaureate. Benchemari is... very academic, isn't it? I mean, but, in principle, it's the same thing that happened to the recognition systems, which is like a lot of secret sauce, how-- it's like, basically, it's like machine, it's basically all systems in there. The ranking of the recommendations is roughly the same, but each company has its own special sauce. But Meta has the most delicious special sauce, so they make so much, uh, so they make so much money from the user. Well, maybe the same thing happens, happens in chat-bots, I don't know exactly, but it looks like the advantage of OpenAI is that they got a crazy distribution, not just in the, not just in the, uh, the, uh, the, uh, the, the, uh, the, the, uh, the, the...
Models.
Well, that's the thing about the model and the difference, right? I'm just a few episodes back, well, not a couple of years ago, I think Anthropic is young and, in terms of, some programming tasks, I like them there more and so on. And I was wondering how to compare it. I want to compare it to something, you know, on some kind of baccalaureate. And, uh, Yandex started the Yandex Cup. Well, there's an Olympic for programming, and everyone who wants to be involved. I think I'm curious, and what happens if I feed the different, uh, chat-bots and see, and then what happens in the end. It's a little gray in here. Before I was put in tomatoes, I'll tell you that it was all out of, uh, participation, like, uh, sometime out of the contest. I mean, I've done all this, but whatever I need--
I did, but then you didn't win.
Yeah, yeah, yeah. I only made a selection phase. The selection phase consists of six tasks, uh, on the Olympics for programming. And, accordingly, I fed it first, Aic. Anthropic quickly produced the result in-
I've done it fast. Overall, the direction he chose was correct, but the task was not accomplished. Yeah, so, no tests go through. I've been a little bit of a pain, I think, well, maybe I can get you crazy and you're still doing this. It didn't work. Yeah, that's a lot of time for that, and it's been a little too much, and it's just that you had to get pretty deep into the task, using that tool, to solve it. Then I think it's okay to feed ChatGPT. I picked four degrees and, to my great surprise, the solution was one of the same as Anthropic. So Claude 3.5 and ChatGPT 4o, this particular task has been decided one in one, that is the letter in the letter. And it's both wrong. So they've been in the direction of the whole thing, they've got the code completely identical, and they're the same non-working code. So you copy it, you put it in tests, you don't pass tests, tests go out. Then I think, okay, I'll give you a chance of one. It's like that, "I don't waste your thirty requests a week that you have." Well, I don't even spend them. I think I'll drop the test for a reason. I'm putting the first task at the o1. He's making her come along. I'm just making a decision, putting all the baccalaureates in the Yandex, making a full decision, 100% scores for this task. I'll take the second task, I'll put in, start the code, and I'll go back to the task description. I'm sitting here reading. I think I'll at least read the task. By the time I was doing the job, there was something like 100-and-eighteen seconds thinking about, I copied the decision, set it up and then I got a 100% decision. That's what the five system has decided from six tasks. Literally, in two tasks, I must have just written a few times to clarify the proms. But look, this qualification phase is two hours and six tasks. Five tasks are real with GPT, this one is a one-time thing, just copying, injecting, not even into, that's, that's what's written in the task. So a man who totally doesn't understand, can copy the condition, insert it, then copy the decision. And if something has not passed, you write, "They've failed, think again." And you're actually putting a second decision. And the second decision is likely to pass. Or there's a solution, but it's not the best in the memory. Yeah, and there you see a message that you're using too much memory. You write, "Your decision uses too much memory." It rewrites it, you get up, and you're working. And here, of course, I've experienced some shock. So when Anthropic and GPT-4o failed, I think, well, as it would be clear, the Olympic programming probably didn't have those tasks on the learning dates. And actually, there, like, he doesn't know how to deal with it properly, and something's got generyl and something's not working. But when one of these things went through, and most importantly, I'm saying, I did it faster than I've read the job through. If you remember, a few episodes ago, I was a little bit on a one-stop, like, there's a man who's been accompanying 4o, and the same results can be achieved, yes, just there, not one prompt, but, uh, through, Twenty proms. But I couldn't handle it. I mean, there's a ten-twenty proms from four, that's not what I've decided. I'm not an expert in Olympic programming. I mean, I know something about it, but no matter how, I didn't get through it. And the one-on-one prompet did it all. Here. Anyway, this is some kind of shock--
The next winner is Yandex Olympiads, our listeners.
I, you know, I think, where's the guarantee that the other players don't do that? I mean, well, that's a perfectly trivial decision.
You're taking the task, you're getting up, you're getting a solution. I didn't really get in.
It's very, very, very interesting. How people do themselves. My nephew is planning to go to the top nuke in the US now. He's got his next month's exams, yeah. And he, himself, is planning to do everything. I understand that a lot of people will probably try this with machines, and the system will be-- people will actually be, at some point in time, at some point in time. If they get caught, and they're gonna-- someone's gonna be hard to learn, someone's gonna get kicked out, someone's gonna leave, something else's gonna happen. But we-- this zone is starting to spread-- well, how much the power of proliferation becomes very broad. So we talked about the resume, the job, and a lot of companies are testing, various online tests. You once, Ignar, gave examples of complex, there, in terms of olimpiad episodes. That's all, of course, that's what's causing very specific things. And I had a long conversation yesterday about what economic areas will be, very affected and not to be present without a cool model. I mean, when this AI is coming, and then you can't be completely, well, you can't be in this business if you don't have a model AI. Now, there we were talking about e-commerce and talking about e-commerce in the sense that some things would just be impossible to enter the market if you don't have a model, there's no tool. Or, for example, some instrument will appear in theory in the commercial. I'll be in New York next week, by the way. I'm just trying to get it out there. We are now starting to start testing for one new AI instrument that end-to-end advents. So when a man doesn't even participate, you know, end-to-end on two sides, like he'd gone. I got a man telling me about end-to-end in the Amazon procurement all oversale. End-to-end is so clean, no man at all. That's how you get to Tesla, the man's driving anyway. And Elon Musk showed that the man would not sit. That's when a man doesn't sit. And I, I think it's not very easy for these industries to enter because there are always the main beneficiaries in these industries. They're not interested in the computer on the other side doing everything.
For example, Amazon is not interested in selling the sellers of good quality. The Yandex is not interested in Google, and it's not like people are doing the right thing to make the commercial. Well, they're not interested in that. But the other moment is gonna be... and they're all algorithms that make people final, that's what Sasha said, that's a sauce, that's the final sauce that adds people to, right? Is the question that these people can overplay the new algorithms on that side? Or if the sauce starts creating algorithms, what will be the battle of these algorithms? Because he's telling me in e-commerce, I say, "Look, but if Amazon changes algorithms, the system will sink." He says, "We need three days to re-education, some kind of space-based parameter, just space, three days." He says, "and we're the coolest in three days, 'cause those who aren't, uh, those who worked without AI and Amazon changed algorithms, they can't do anything at all. Well, they're not gonna change anything at all." So you're changing the industry. And that, of course, is, uh, pros and minus in terms of different industries, yes. And again, people get to know and figure out how it works. Because some, uh, some professions, I'm not so supportive of saying that some professions are going to disappear or not, but many because there's a ChatGPT, uh, because you use it, you use it, you use it, even there. I am deeply convinced of this fact that we are less than 5 per cent of the chat subscribers. Here. Another point that if the OpenAI wanted, it would have had more than a billion subscribers, hundreds of millions of permanent users. Is that why they don't want to? Or maybe they can't? Or maybe it's not theirs, it's not theirs? That's another question, yes, why are they...
Interesting test, Aina. Interesting test.
Yeah, yeah.
But with programming like this threshold is coming. So a programmer with a model and a programmer without a model is just uncomparable people. I was wondering if I asked a friend about--
But it's a tricky one. But a good, good programmer with a model is good. When we say the man is an idiot, well, not a model, I don't think he'll be able to...
Why not deal with the model right now, I don't understand.
Well, Sas--
If I walk through this Olympics, I'll be right there.
I'm wrong, I'm not correct, I'm sorry for my wrong word.
In America, yes, be careful with those phrases. Look, uh, look, this is the test I passed. Tan, you'd have gone through this Olympic program--
I'm telling you, imagine if I win this Olympic.
But I won't say--
That's the news.
I'm not gonna tell you the real challenges that were on the Olympics. I've been deciding justice for the sake of qualification. But, Tanya, the qualification of five tasks out of six per cent would definitely be in the 20 minutes, just copying these conditions. So, Sasha, even here, you know, at some point, the monkey with O1 is getting comparable to a pretty good programmer in the way of a solution, uh, some kind of thing, right? And I was wondering. I've been talking to acquaintances. Here. And one of them, uh, a master developer, works in a German company. He says, "Look, he says, man, we're getting a scout at work, uh, Copilot, and they're urging everyone to use Copilot for routine purposes."
Of course.
Testing, documentation and other. And that's what it is, well, the industry standard is actually working. I mean, while we're playing with this, it's all on the market right now, it's all in the corner. There are companies that think it's--
There are companies that think it's standard and without, and they're almost impossible to work
without it.
Yeah, it's 100 percent. And again, again, you say, sinner, people are qualified. If a qualified man uses tools, Aaaa, AI, it goes to the sails, it's about editors, it's about designers, it's about a lot of people, right? The engineers are different, of course, that's what's making them work. It's just that people who don't know it, and for example, less qualified, are much more difficult to do. There are areas where all this is not a standard. And there will be a very strong outward and a winning man. But I can see, for example, that I'm very effective when I have such tools in the day. I mean, my speed is... I signed some corporate documents in the morning, there, buying one company in the US, and I know I'm just asking a professional question. I'm aware of that. I'm in a lot of trouble here. I'm in business for 20 years, but it's making me look great. I mean, I'd have checked through the lawyers, through the acquers, through some people who are in the business. I just don't need a lot of things right now, I don't need these rechecks. I mean, I'm gonna re-check myself in a series of certain actions, and it's really cool. In particular, in simple things. I was approached by a comrade, and I showed him how, uh, youTube could be built, where, quick, like, subtitles, or a different text to the way people talk, there, text, oh--- work. So you've been re--smashed, and, uh, audio to the text and that text is further refined. He was just shocked by the speed that could be. I mean, he-- he's a very smart man, but he didn't see that scenario, that's the way he works. He realizes that if, uh, he had staff working with that efficiency, that's how they would use tools, they'd just, well, they're-- that's, that's, that's, uh, that's crazy high--- that's just a high-end. A level of competitive advantage, yes. You just know how to do it or not. You use some side-soft or you use a normal one. What's the most important thing that's gonna lead to? So what's going to happen, uh, next progress, what's going to happen? And will there be any kind of a jump or no reset? I can see, uh, over the past 18 months, very serious changes. It's only been a year since ChatGPT became famous. Two years after ChatGPT released three zeros, if I'm not wrong, two years passed, right? Or-- yes, two years.
Three s-- three and a half, yes, two years
ago.
Yeah, two years, yeah. Three with a polo-- not even-- three and a half, right? No, well, there's three zeros, three and a half. And, and how much industry and opportunities have changed, different chips and themes, and characteristics, and how many cool softs have increased in terms of internal algorithms inside their AI presence. Well, we'll definitely see a lot of changes. The question is, what are the areas? I always say, "Who's gonna knock." Well, if the OpenAI goes on to HGI or they'll, for example, be in the consummation. By the way, yesterday, it was literally the news that the former technical director of OpenAI, uh, attracts, like, 100 or more million dollars of investment, uh, on his own start-up. And, uh, what are they gonna do with the start-up? And to be honest, I, I, sometimes-- it looks really weird for me. You work with a technical director at a company that's just, just, like, a top, a top, a top. You have a limitless resource, you can do anything, but people, um, and still psychology, right? People don't agree with each other, they don't discuss anything. Everyone has their ambitions, right? Although that team must have had, and perhaps there is still room to do some incredible breakthroughs very quickly, very quickly to create something that would bring about just fundamental changes. Here. But you and I will definitely see changes, there, and in education, and in, uh, education very serious, yes, and in reallocation, as work is done, work. Well, that is, especially from a country perspective and access, and access to models from different countries. Please write in the comments, what do you think about artificial intelligence in which areas of artificial intelligence are intact, and, uh, enters in the next six months, not in the next five to ten years? For six months. So what are we gonna see? That's how we were talking about the resume, about the vacancy. Or did Ilinar tell you about the programming, what about-- what do you see now? Or some examples, bring the case too. It's clear that examples, case files, theirs, their havoc, some inconceivable number, and you and I see a lot of systems. Thank you very much, everybody. Don't forget, please sign, slide the canal. You're very supportive of him. We're going out every Sunday. We're always here four people. Insides and news from the Silicon Valley. Absolutely our unique opinion. Bye, everybody.