For your own instance studio headshots, they got QR code on your fashion. So they pass that around or hold it. So maybe we could have add a vias come all the way from Guangzhou everything go shopping in Shenzhen but I
so, I, because I lived for many years ago, he had to tell you. So, you know, governments auction off bandwidth. Right? They go, right. I've got what is it? Something megahertz? Yeah. I want to sell it on a put an auction for billions of dollars. His company builds those platforms. Right. Great. Yeah. As you can see, he's very humble man. But he has a deputy in one job, which is made up of people from all over the world. And what I found fascinating was we bumped into each other in Bangkok, at a conference, Ruby, on programming languages. Now. He's like, Oh, we did something. So maybe you can tell you very quickly what he's doing, because that QR code that hopefully you stand done for now, but you do not know.
So thanks for being here. And allowing us to do a little, little bit of kind of a combined beta test and promotion. So yeah, as you said, we've been doing software business, consulting, mostly in the option space in what we call high stakes options. So very high value auctions for intangible assets, for a very long time. But recently, we've gotten to AI. And so we just launched this new product called instant studio. And it's all about image generation using Stable Diffusion. If you're technically minded, we build our Stable Diffusion cloud. So you can run image generation tasks in the cloud, for any data. And one of the things you can do with that is you can train an AI what you look like, and then generate pictures however you want. And so so as as sort of a demo as a as a as a way to show off the technology, we built this consumer app. It echoes what you've probably seen ads for on Instagram on YouTube. But for tonight, we made it free for you guys. And it has little logo, weapons, a logo on it. And we also added Christmas cheer photo. So please do try it. Please do let me know what you think. And I'm just I'm really here to hear what you guys think and to, you know, see what we can build together using this kind of technology.
Right? Cool. So try it out. You won't have to buy any Christmas cards, Isaac really doesn't use the way to draw his own Christmas cards. So the second thing I just thought was so great. And we've got a 13 year old. I remember that Reiki it really is just now. And begins. That's us. That's us before the AI gets stripped so brilliant January to anybody, like as follows in the generative AI world. Just remember, little kids are still artists to be. So to me, it's really interesting, because I have a daughter that seems wondering how she got a career. So it's very interesting to see this quite believable. Maybe we should get together. Yeah.
Just talk a little bit. You know, kids got the best, you know, yeah, test for today's AI, become the next generation. And they are AI native. And he's using air tools every day as well, when he's reading to the system,
and other schools don't really, you know, because schools are always going slow. So that's very amazing.
So we'll talk we'll talk about education in the moment. So make sure you go upload your photos. And then you've got Christmas cards. It's very ESG friendly, right? Yeah. But it burns in color. Or an issue with like burning carbon with MLMs. Yeah, I
mean, there was an article yesterday, but I'm just not gonna worry about it right
now, anyway, so to have a chat with babies. Smart. I just thank you so much. Excellent. Here I am. I don't know if you notice there's a movie coming out. Yes. How long I waited for that. I grew up eating pizza. Napoleon's nappy, therefore Pagani apart blown apart, nappy rash, the growing up in England called the photo you can imagine. So my moment is now so what added did very well, is I said, we have to jet AI, me as an opponent. So I am now Spanish spamming the internet would be as the problem. So there you have. So I think when you're doing technology, it's very important to be topical, because content is what makes technology interesting. So talking about content. Our next guest is Kevin Durant Pereira. So I saw Kevin, I met Kevin I think, like a year and a half ago, two years ago at a conference. And he had this boost. And it said blue AI or anything, say I just couldn't understand it said, artificial intelligence consultancy. Nobody was going to his group, right year and a half to two years. It was a marketing content,
marketing, conference, finance, sales and marketing.
So I went and spoke to him. And now two years later, this guy can't get enough of like he's everywhere, suddenly, people. So maybe you just tell us a little bit about what you're doing. And I want you to the reason I asked you out is because I wanted you to predict what's happening this year in AI, if you refuse to do that. Sure. So there's confident after they can predict Ivan, can you predict? Who can predict will
like make one prediction now, most predictions will be wrong.
That's all killed everybody else. So go ahead. Tell us what you do.
Yeah. So hi, everyone. Nice to be here. Thanks for the invite. So I work at a company called artificial intelligence. So we do AI consulting, and also AI education as well. And I think when we first started, you know, we thought, stick out on the shop window, people are going to come running, as we found out from that conference, and also just like, what happened. But I think when when people think of AI, they often feel fear and confusion, right? Part of the fear is Hollywood, you know, Terminator center, I think the other half of the fear is it's going to come and take away my job. And so we have those two things that's pretty difficult, right is actually proliferate. And so we found that one of the solutions for confusion and fear, especially education, we started doing a lot more work on the education side, both in terms of corporate education, that also we did a few talks at universities, and that ended up turning into a full fledged courses. So now, I'm an adjunct faculty at a few of the schools in Hong Kong. So he can you expect us to see you and others and teach courses on AI and big data. So we will be solving alongside education of the
AI finally, because 100 educate people about something that's changing every single day. And like retro, I was saying education is behind. So you could talk about AI three years ago, but how you have you keep it fresh?
So I think part of it is mindset, right? You I think can't teach this in the sense of this is what it is. I think part of that is, this is what it could be. And I think business shapes, right? Because where technology, or I should say where rubber really meets the road with AI is what is the right way to use it. Right Way there's an ethical component, which I think involves a lot of education. There's also business. So a lot of the classes we do actually in the business school, right? So the idea is how can the students think about this. And then I encourage them to think about AI as a tool. And I encourage all of you to think about like, as long as you have any tool you can use it for good as long as you and as a mentor. There are white faces and then scenarios use it. That also wrong as as us so it's really it's about showing all the tools, but it's more about the mindset of when to use the tool. How
do you think I've seen with the quicken the speed comparisons, it's not working? My battery's running right through those comparisons to when the web first started, right. And it's actually it's what you build on top of it. It's interesting, the sort of fundamental so do you think that's a fair comparison? So you know, when when I first started, there was a lot of people fearing it was multiple jobs to do and then it was Before educating how to use the please.
Yeah. So I think if it's a tool, you need to know how to use the tool. But I think as anything else, it really depends on what are the outcomes, right? So if you ask, for example, people, how's your phone book, most people have no idea if it works, but they become really good at music. So I think you need to appreciate what's actually happening on the sort of under the under the hood. But I think the more important part is already using the sport and the car takes in from placing to be that's really useful. Do I really need to know the insurance? Probably not?
Do you think there's a whole area of this is not working on ROI, ROI, but it's running out of battery? You need to get Elon Musk on this? Do you have an alarm? I've got an add enough voice. But the what I find fascinating is for some reason, the general population has learned to look under the foot, right? Everybody wants to know what's happening underneath, like you're talking about driving a car. Everybody wants to know what's in the engine. So what I find weird about AI when the web started, I was webpages, nobody wants to know how it worked. This is like put it online. Why do you think there's so much fascination about how does it work? What's under the hood? Is this engine got eight valves, 12 valves? Who owns the engine? Is the board about to fire him? Because he owns the engine? You know, what, why? Why is it so focused on that? So,
I mean, so from my perspective, on the education side, especially in the business area, I actually don't think that's the case. I think there's a lot of folks that are, you know, it's fine as a black box, I'm gonna trust the black box. There's a subset of people who really care about the black box. And I think, you know, if we think about business, regulators really care about points. So recently, we've been asked about regulators to do education. And they're asking, can you talk to us about Explainable AI? The truth is, if you look under the hood, for a lot of neural networks is very hard to explain why an AI is doing something. It's like saying, you look at a piece of art right over there. And then the question becomes, why do you like that piece of art? And can you give me a percentage breakdown and why you are emotions?
In the case of NFC, it's because nobody else is buying it. Because I wouldn't call that art. And then be cheaper, for sure.
But if you're talking about the picture of itself, and you ask people about that, it is very hard to attribute my positive feeling to that thing. And that's in effect, what you're asking when you're talking about neural networks. So I think many people, they're focused on the outcome, and as long as the outcome was for them,
then they're okay. So you're gonna change education, right? You're working your way and you're corporates. So you're talking to a lot of young people in college? Are they as concerned about the future as the media pretends they are? Or are they like, Okay, I'm going to appear on council doesn't matter. My dad wants, he's going to cancel. My mom wants me to be a doctor, I'm going to be a doctor can't do the surgery. Yeah,
so great question, right. So where are the students thinking about careers in the future. So my view on a lot of this stuff is I think a lot of jobs will actually change pretty drastically. And they'll change because they will have to use AI tools there. So take a doctor, for example, today, a doctor does a bunch of diagnosis, I think in the future, that's probably not going to happen. But there are still skills that are needed by the doctor. So if you're gonna die in six months, you want a text message telling you that or when you want to talk to her with the empathy and all those pieces, you're
not talking about doctors in Hong Kong
actually has some of the highest life expectancy, and some of the lowest happiness rankings in the world. I think that's actually a bad trade off, right. So really high, and there's a life expectancy where you're really miserable. I feel like I like it the other way around. Who live you know, it's actually a random tangent, but one of the biggest predictors of why the life expectancy is so high is because it is very easy in terms of time to get to an emergency room. In Hong Kong, I think it's under seven minutes on average from where you are
telemedicine doesn't work yet. So go back to that. So the in the education space, I'm just curious, because you're you've developed courses, right? So is there a huge demand are people are those courses being applied to like traditional jobs that people have in this era? You know, kind of make sense in engineering and computer science, but outside of that, like law, medicine, people who are attending, are they from those faculties?
So they are and I think what they're trying to think about is what are those industry or carrier specific tools? If they get rid of those tools that incite the future level of work for many of these things? Is the AI tool combined the human right and so, in the education that we do, I'm a big fan of the message that we are To be complimentary to the AI going forward. So if you think about our jobs are going to evolve. My view is every job consists of certain tasks, some are easy to automate, some are hard to find. So the real question is for the job that you're in, what is the ebb and come in and take out? And then secondly, how's your job going up? All right, so let's say we have a job with 10 tasks, seven of them are easy to automate. So then the question is the remaining three, because the human beings spend all their time on that? Or do we put in three new tasks intellectual. And my sense is the tasks that are hard to automate are the ones that have to do with human to human interaction. So I take a little bit of a story from my personal career, I used to be a private banker before AOL. So I'm probably a good move.
But it was great to have the credentials to do it. Now.
I also talked about because nine, this is still on my list, Altera. So
I think it's such a new field right now. And I would never say that I'm an expert at everything I do, because I think it's impossible for one person to do that. But I think that you're now seeing a new collar area, where companies and organizations are interested in people who know enough about technology, but also understand the business side effects. I think, previously, you had people on one side of the fence or the other, but very few who stood on both sides. And therefore the education that we're doing the business schools is trying to get those business skills a little bit more on the tech side. And I think if they can fulfill those roles, then the other interesting thing that's coming on the job angle is new jobs we never conceptualized before. Right. I think that is something that AI will start to open up mentioned. Very excited about that.
So training question that you're talking about, yeah,
I'm trying to help people, not only for how their jobs evolve, but can they actually become ready for those jobs, we haven't conceptualize?
That's interesting. So you're saying that these people they get their own education into? So it's funny, because the way I see it is that, you know, say, 20 years ago, people would advise CPS, you know, proficient in Excel, Microsoft Word, PowerPoint, and then the tech guys go, you know, you'd like to be like, G to E, your je two out of the whatever ready, put all the Java, C++ from reading. So is all of this can be replaced. So your credentials, when you use a rat sauce, if you can do XL is going to be like, do you know? Where? Because there's not going to be a member, there's gonna be some kind of credential, right? So if you have a credential, is it like a? Is there a credential? If I took a course with you? Am I going to get a certificate that says, genius AI? Or what's it gonna say? He's gonna say, he understands that language models, he understands prompt engineering, is it pride is always other acronyms are out there.
So the courses I teach at universities tend to be part of like an MBA program, or like a larger kind of master's in something. Right. So it's obviously a master's in AI. Specifically, it will be a master's in marketing, let's say, and this will be the AI course in the restroom. So in terms of the your question on the resume, and when people are going to have, I think it's going to be domain specific tools. But this is my view, the best way to use AI is when you have domain specific expertise. And you might
say, I was drawn to solving orange juice.
I think the best folks who use AI are those who have domain specific expertise, and can use a tool. And why I say that is because if you have elucidation, the person who has domain expertise can see that, right, the doctor looks at what the LM is saying, as a treatment for this disease. The doctor by just looking at it straight away, those investigations are not, whereas someone like you and me for my other health care experience, but we'll go ahead and be like, oh, yeah, that sounds about right. And so I think it's really important to have domain specific expertise. And I think in the future, there will be domain specific tools that I think people will sell, because that's the right
example you just give them isn't a battle between the doctor has to feel he or she has more knowledge about goats. Yes. So I think, I think I think that you have pushed people a lot. I know people who graduate from universities and doctors, I would not go anywhere near. Sure. But you're saying that this might actually push up their qualifications because they've got a competitor? A reference point? Yeah,
I wouldn't call it a competitor for sure. I think it is a tool that's going to enhance what they do. Right. So the doctor that uses the diagnosis piece, it becomes part of their arsenal, as opposed to I am fighting with that thing. Okay, now I go back to my initial Friends, which was if the AI can do all the tasks in your job, then you're in trouble. But if AI can do some tests, and you can do some tests very well, that's the right thing to think about. So
what would you say to Ricky that he's 13? Right? He obviously gets plenty advice from his dad. And that's very cool to bring him out. Occasionally, what would you say to your boy who's particularly good at drawing?
So this is what I would say. I would say, right now, some in the area of AI. One of the hardest things as a category is human to human interaction, right? So creativity, critical thinking, sales skills, trust building, there's a lot of concern. What I will also say is AI is getting better at an exponential rate, was easy to was hard to automate today, might become easy to automate tomorrow. So this is going to be this is constantly going to be changing. So my advice to the younger generation is, at certain points in your career in your life, reevaluate what is hard to automate, and get good at that stuff. And understand that that might change. So I think the larger mindset is continuous learning. Learning doesn't solve any finish at school might have worked in previous iterations, or right now, that's not going to be the case. And I think we need to all continuously learn, we don't do that, we are probably going to be in trouble. So they'll be my year to be complementary to AI that's getting better and selection. Things breaking.
What's the 10 year communication skills? Do you have any? Maybe you can ask? If you don't have anything, you can ask a question. Yeah. You can disagree with your dad, particularly.
I agree. I agree. Now
if your answer. Okay, so what's interesting, I let's take maybe one or two questions I've seen up here. Does anybody want to ask this video? Angela? is in the recruitment, HR business. She spends her days getting people to think about that future. And we'll be talking about this a lot. Because what you just said to me is I spent my whole life re educating myself. I love Chinese and Japanese and Italian and magnetite. Don't stop educating myself. As soon as you stop your brain cells, just go check out time. Right? So what you're saying is keep that burning? Question. Can I use a microphone?
I need to let people who maybe they've become a bit stuck in the job. And I think that's the cause of perhaps how companies have molded the lines and how they hire people. So what I find is sometimes, if I can think of a way of how I'll be able to actually get the job safely during on a day to day basis, not only interview better, but come up with better solutions than I'm thinking that's a very dangerous place for that individual to be. So how would you advise someone to advise people who are in that position because they probably go slightly under there of the of AI potentially taking your job, but at the same time, you don't want to Get them to lunch? Because they just might they just don't understand, like, potentially that they might use that as well. Or they may not.
So if you're asking like, what's the characteristic, right? I think part of it is cultural. Right. But I find that there's a lot of folks and let's say, Petia, I think the symptoms weren't all that bad. So I think we still, you know, hey, this is coming. I think that kind of makes them realize, oh, I need to actually go do this. Right. So for me, if I was advising someone to get stuck in their career, I would say, take your head on the side, firstly, look around at what is the AI that is being used in your industry. And I think that then starts in their mind to lay up, which tasks I'm doing my job could be automated, I think if they really start to see that using the framework we talked about before, and they see that that percentage is quite big, that's hopefully a red enough light or big enough red light in their face, to be like, You need to start using some of these tools. And actually, it's funny, because a lot of the workshops that we do on the corporate side, and I've done a few for a university alumni are some of the schools I went to the people that attended our let's say, 50, early, late 40s, early 50s, right. So they're that bucket, which you're talking about, which is I'm kind of stuck, but I'm senior enough during the interview
that there's nobody like that in this room. So
but what I find is once they start using what we call low code, and no code tools, meaning you don't need programming and coding knowledge, it's almost liberated. Right? Isn't it before, they're sitting there thinking, Oh, I can't do coding or programming, I can't even be involved. And I think sometimes I think the education that we're doing is not to make them experts have the tool, it's literally to force them to go into the swing. And I think once that they're actually forgotten how to swim. So in some of the chat groups that I'm in for those sessions, we do these people are not producing amazing kind of images are producing a lot of stuff. And they're showing everybody that has a great positive cycle where other people in the group are like, wow, that person did it. And I can definitely do it. And then you see the positive cycle strategies. Right. So I think part of it is getting your head in the sand, starting to use these local developer tools, and really understanding you I think that's really the big, that's
a very good segue to Steven. How sorry, token, economy?
I'll send you some bitcoin. Couple. What would be the examples of the most at risk jobs? What would be the new titles of future jobs?
In your programming question, right? So I think if you asked me, let's put in categories, the titles of our heart, those three categories. So one area I think, is eventually once companies start using a lot of AI, I think they're going to need AI emphasis. So people who specialize in the right use of AI, but not on the technical side, but more the social and ethical illustrations. Part of why I say that. I recently gave a talk about high school hypothesis, I've given thought to the philosophy students would you wouldn't really think about AI when I asked that this is a question, what do you guys think is gonna be the core job? And they said that we'll have philosophical frameworks and things that we're learning with AI, you can actually start to apply this stuff. And right now, those jobs don't exist, because the AI is not there. So you don't need them. But once companies start to use it, next question is going to be if you really use this, how can I go wrong, and what's the right way to use it? We certainly hypothesis seven would be an example of one of these newer jobs, never conceptualized before, but now that may be needed, because AI is actually being introduced. So I'm personally very excited about like, kind of what's coming there, from the perspective of new jobs. So you
think finally, like a student of psychology will get paid as much as the bank and
it'll be closer and closer. And I think there'll be things that bankers do that is less valuable. If you think about Microsoft full path, right? I mean, I interned in a bank when I was in college, and part of the job at the internship was, here's a PowerPoint presentation, move the line three points to the left, cuz I feel like it's better. Right. But now, if you could just have your hands and have a leg up, so that you don't may not, you know, in terms of the things that the bankers are doing, they will not do that stuff. But they might actually go and meet clients, which is a human to human interaction. So they might still get paid more, but I'm pretty confident what they're going to do in five to 10 years from now is going to be completed. And I think that's a good thing. Because a lot of grunt work that I had to do, I hope they don't have to. So if he actually goes in terms of bank, I hope it'll actually be a really, really fun experience. As opposed to this seems menial stuff that I'm really not getting much value out.
So you talked about banking. So we're in a transition environment where where do people find you too, they just go on perplexity. to buy and the revenue name Kevin Pereira Blue Dot our funding
stream so best way to do that although you're
on over LinkedIn Yeah, my feed is just Sam Altman you somebody else you
so we managed to get the AI working so so Anyway, happy to chat with any of you guys afterwards easiest way to find me WWW dot beyond you know Ltd that's the company website and so we do consulting education you guys have any questions about that stuff I
love that people say www it sounds so old like that's why when HTTP backslash
hypertext Hypertext Transfer Protocol
excellent right thank you very much is going to join us at a very critical education is really fascinating I mean that's huge So our main events Steven who's been very patient ain't never been so patient to do drink no I'm okay because really weird I feel hot
it's a screen
generates a lot of something going on right? You've got to like lock the icons psychologically.
So I reason I invited seen as I was at the Hong Kong FinTech week to do you go to that? Obviously, three years yeah, I've been of course, while I was got RFID NFC and his, as you will know, NFC he's got it is really of interest with insurance. Your NFC sounds really dirty. Stand up and show me your NFT flashes rental exit here.
So you can put your iPhone and goes to the website address for the owners mentioned. So from 69. So, only 69 of this to this producer. And he says like the 4941 wishes.
So I think humans, Pharrell Williams, or aging changes, Caleb, imagine if he did, collapses. So you're talking about forever billions, with Canada. Not for hours. So we were a FinTech week. And I saw he had a really interesting interviews through talk about what's happening in finance. And how I guess fits into what you were saying is the jobs are changing. And so maybe you could just tell us a little bit about what you're doing with us, Flora. As you firstly was Laura, like your first girlfriend. asked Laura. Learning
oriented risk algorithms, it's a lot more drier. I've got this asset
risk algorithms because
when I had an interview was the highest Wednesday CNBC invited me in and the first thing the journalist said to me was this Wednesday a dating agency for geeks says yeah, of course they'll be late so get back to Laura so you know I'm reset us you come from the your credits you were a mathematician might have signed I want to say Scientologists but that's wrong. But yeah, Scientology I'm very curious system. Tell us many how you got to where you are now is lightning in your in banking, has that whole mathematical thing evolved into red right now?
So I was kind of the positive aspect of banking, quant trading pot fun.
And then, so if you're not in banking, what does quant actually mean? It's
math. It's math. Statistics, the way
you write programs to look at math. So you're
you try to come up with models for previous behavior failures, price behavior, previous stock behavior, fundamental behavior, and you try to model
Okay, so you're already kind of creating the fundamentals of behavior, but no mathematical sense. Spending
models to hopefully predict or come up with portfolios diversify risk, get a price of a derivative options, stuff like that. And that's been kind of the central theme of my career. And then, basically, the company that I was working with, asked me to do due diligence on deep learning. So this is 2017. And this mirror, right, right. And in 2017, is I was in Korea. And this is when AlphaGo beat the Go champion in in Korean, and this has actually been something since everybody thought, even just kind of completely destroyed. And so
after Kasparov was in that's
way after Kasparov, Kasparov was probably decades ago. And that was actually a bug. The IBM has kind of come out with incident was a bug. And Kasparov kind of thought the machine was being creative, and really freaked him out. And he started getting mentally sucked out at last. But it was actually a bug in the machine and deep blue that caused it to create a weird move. And thus, this is probably a 20 or so years ago. But this time, it was a series of games. And the core champion said, you know, this is something that, you know, that we should definitely better. So that's one over and over again. Whereas, IBM after V Kasparov, and realizes, never show that machine again for a few years. But
nowadays, they just call it hallucinating. Well, it is to lose thinking about
hallucinating is a almost a feature. And
I'm gonna send it. So
basically, because AI is actually right now, mostly math. And we've been doing this for a while. And for all the things is math has made it to math this country bigger, and the math is scaled more. But it's still, it's not some magic. So I was looking at it at the time. And it's kind of going to Darren saying is can you use this to predict stocks and beat the market or let's go have fun going. And I did a real DD and I said that I don't think that's the case, you can probably help some services, give me a few years. Let's say you have a few months come up with the AI fund, you're going to market as quickly so kind of started a research project that became a company that research project that became a project that became a company can't just start up in Hong Kong.
So are you in that journey of maths? Right. Do you do you think the phase right now is you said it's only maths? So the maths is just telling you what's the next word? Or what's the next image? I mean, I'd be curious because that is to be produced there. They can planning images. But those are you saying that it's all mathematical equations, which drive in that
it's a mathematical equations that are driving how these things are working. But there's two things that are making hallucinations or this kind of creativity is that the first day is a little bit easier, as everything is probably the probabilistic models. So they kind of have to think about how we talk or how people draw. You never do the same way. Some people can try the same thing over and over again. Normally, artists do try the same thing over, you have things that are called very. So like the Mona Lisa, you look at it from this side, she's laughing on this side, she's crying. And pictures are depending on how you size it, how you, you know, you could change that axis you have about the power to change the axis a little bit, it just was completely different. Right? So pictures are variant, and you need to have some sort of creativity language. So the model is set to never do the same thing. Exactly. Again, and that's how you speak you don't ever say the same thing word for word intonation by intonation repeated exactly the same. So the machines are taught to take probabilistic probabilities and try to come up with more creative solutions. The other little bit more technical and employee base why some creativity is that our experts
way we have knowledge is we've got varying knowledge and we've got a very college meeting that things like math are American. We know one plus one is equal to two, no matter what happens in any kind of way, but language you can One plus one is three, there is no two. Or actually, when that becomes more complicated to accept theory, you have things that are paradox off works. So you can imagine if something just kept reading and reading and reading and explain every single thing on earth, and it's memorized everything on earth, then it's not very complete, narrow, or very shallow knowledge of everything, but in this variant kind of weapon. So it's a stone seen every single image on Earth, and it's read everything. Everything that it sees is, is some sort of fuzzy distribution. But for us, so not exactly sure about the Skynet, things that drew us we know certain things that affect certain things are affect and who will not change. It's invariant in every language, every culture, every place on Earth, certain things, and that's why it's not as good at math. Sometimes it messes up bonds, formulas, sometimes it elucidates. So, I mean, that's kind of why the legislation
got me thinking about little things is that as I've gone through my life, the most intelligent people I've met are usually not very good at communicating, some very stereotypical, because they consider everything. The outcomes are saying this. So because sounds like the kind of these engines are moving towards that they have such a depth of knowledge, that you learn the outcome of saying something, and I say lots of stupid things on purpose, because it's short and fast. But the intelligent machine might not want to say is in those might be outcomes, like three or four steps that
I don't think, I don't think we're there yet. It's actually very shallow, but it knows everything. It's like, if you took a genius, 18 year old, and he had to photograph, he or she would have photographic memory, and she just read every single book on Earth. And no one was there to tell her, Hey, this is what it is. This is what happens in real life. It's like that thing, and that will be Goodwill Hunting. If you ever see the famous scene where Robin Williams is like, you've never seen the Mona Lisa, but you've read about it. You've never seen the Sistine Chapel, but you've seen it in pictures, and you've never held nine person in your arms. And but you've read everything. And you've and that's exactly what you can kind of think what charges BTS. So let's actually take a talk. And it can be very interesting. And people like it, just that it doesn't know anything about hidden meaning, the deeper meaning. And if you go any beyond the first order, that's when things start hallucinating.
So how does that How would you? Because how would you apply that to the finance, right? Because we like I did, I had a very drunk conversation tonight o'clock, with a really smart, half Thai half Chinese programmer who's trying to hack trade. Right? So you said you were trying to do with quants and all that. And his his concept was like, I think a lot of people in your space. There are a lot of sources. I mean, a lot of traders rely on information from Bloomberg from whatever sources, right, they have their own research teams, etc. Do you think that can be because you're analyzing data from multiple sources and use so you have Bruce, the news, gentleman in the room, the news guy from Zillow. How's that? Doing? News? Right? So, Isaac is looking for news, right? And how AI can kind of digest, summarize, and create news. So how does a man in the finance world? How does that How do you so
this is my opinion, and I can be completely wrong. But I don't think AI is gonna prediction. I don't think AI is going to real critical diagnostics or real critical, critical, important things. Because it doesn't have that depth of knowledge. And even we're barely starting to give it ideas how to think in second order abelian light shade, it's called chain of thought, this barely working.
Sorry, excuse my channel.
So if you ask, When did Napoleon have his victories World War Two. You have to understand the point it did not fight me over to Napoleon fought in the chaotic wars and then this was six years before that, or whatever it is before that, and then it's it's two things at once, right? So it's actually two things that this person did not find it here. So that's for the dinamika victory during World War One World War happened in this. So even though America is very simple question, there's two things about this. And this is barely being handled right now. So I go we tried to do it's actually almost exactly what is close to even the stable diffusion is that our company has kind of from advisory or we're trying to go into the search in your country of research. And I was talking about some of the things that even like Goldman Sachs, just in case, I respect what they do, I think they're experts. But that's still supported content is expertise content, right. A research analyst that works at JPMorgan is not right 90% of the time, it's still content, and so is something on Wall Street, that's, that's commenting. And what we're trying to do is have original content, for finance that sound convenient, and make their own opinion and make their own decisions. I don't think that AI in any time, the next couple of years is going to be good enough to predict stocks, I don't think it's good enough to be able to replace a doctor, I think it's always kind of a first line of service or first line of contact that it happens. So like a doctor does not when you walk into Doctor emergency room, you get there in seven minutes, you don't see a doctor, first thing you see is triage. Yes. So the triage nurse is will not make a life or death. But she'll be able to know, you know, if you need to see the doctor right away or wait in the waiting room for three hours, right? So those kinds of diagnostics, were the lines of, you know, the differences are much easier to make are probably where it happens in finance. I have, you know, my portfolios done 25% Cut loss, but those kind of critical things will probably be not ready for it to happen for many years. Because
I remember what maybe 810 years ago, there was this concept of the robo investor. Right? If you're buying securities here in Hong Kong, and they were, it was kind of, like 2% PR, and 10% truth, right. So the idea was you had some kind of robot that would trade on your behalf. But when you went look behind the scenes, all that was happening, because they had some they set some rules. And that was the robot, but you get your money and the rules, it followed the rules favored by yourself,
we'd better for you in some interesting, I think one of the things that we realized, without score, and we were trying to do AI to get you more risk management people trading in the last five years, has turned into also some form of entertainment, I need to enjoy what I'm doing. And not just gonna
quote you on that train as a former head spin. It's,
it's all pleased. But I think the rise of Robin Moore is less interesting that people I mean, there is some element of this that they need to have, I need to enjoy what I'm doing on Robin Hood food too, or whatever, I don't want to do something. I mean, frankly, if you want to just go out 20 videos and just buy an ETF to total portfolio for 20 years, that's probably the best thing to do that no one in finance will ever tell you that, right. So
it's your right, because you don't know those like food, too, and Robin Hood. And basically, they all bring in like a gaming element, right? So there's in rankings, and they're trying to deduce community and then you can see so you're right introduces is like the game industry, suddenly, there's a whole community you pass off by spies, he or she or they're, you know, that's all very icon kind of personality of your avatar.
It's kind of Wall Street bets in the US, there's sites here and that YouTube, it's the people created content that is creating more interest and more trading and more volume in finance. So we're not in any way, using AI to predict stocks, we're not using AI to say this is gonna make money or not. We're using AI to do what a research analyst would do that stock forms analysis. And I say this is the risk, this is the this is the, this is the pros, this is why you should invest, this is why you should worry about it and be able to do this in a very, very conversational way. To kind of say this is someone that understands finance and you can talk to him anytime via VPN. You know, when you're taking a bathroom break or you're sitting at home really intense. So if you have someone like this that you want to talk, then it It's content at hand. That's original, that's accessible.
So there's a few pieces now I wanted to kind of break up the the research idea we've worked on called for three years in some phase in your life, even if you look at making a party party with a research analyst, right? Usually I'll jump somewhere and they've got all the hot people on. And they'll research on the scarcity degree into going and meeting companies and interviewing the, you know, the operational management team, the founders, whatever, right. So there's not just looking at their annual report. So now, maybe that's just the party part of it. I don't know. But they spend a lot of time talking to the management teams. Right? So is your look at Eros, is it really classy? But is your is your How do you if you're replacing the analyst part of it? How you worry about the judgment of the management team or judgment? Kind of sorry?
Okay. This is our generation of people used to do this and go and meet Yeah, well, it was a lot of hand and shaking a lot of whatever, all over. And a lot of this is now becoming saturated. So a lot of this is you, you're not allowed to talk about this before called insider trading. Because before I remember, when I was younger, we would go to WWE.
Title calm, we'd go
and talk to them, but they will be very, very methodical, this is our latest thing. It's not so you know, information anymore. And it's, and so it's even if you go to the company, the human element is a lot of more regulations and what they can discuss, and it will be much more especially larger
investor relations to take over. Yes. The second thing
is, any information that they can give out is often the rap. You can know all the shipping information, all the back channel information, all the
changes, obviously, right, not privatized.
No mean negotiating companies, if you've got shipping things, and you have shipping rates are all available, the fact that our Panama Canal is now backed up, and it's all internet, right. And if you think about 20 years ago, we probably wouldn't know until a few months after the fact. So if you can scrape all of this information, and you can connect some of these dots in a more invariant form, which is what we're working on our research under the hood, then you can kind of come up with some more analysis. So when
would you how would you be collecting this data you rely on, like existing technology existing? Or you build your own? Or do you for us, we,
if someone says they're making their l then that means that they've got at least $30 million, or 500 million, if they're making their own LLM. And their company doesn't have $50 million of at least of product credits. It's not even Cloud Credits, because like, we've got a few 100,000 of credits that we can use, because it just used up you got to have your own, like GPUs or something direct access. They're not making their own. We're trying to train existing OEMs a little bit further. And that's coming like a week something out much some project
because of the speed of those elements in the hospital as I wanted to show the user terminal so they throw
your GPUs are not available, the GPUs that can train these not available
to you to GPU on top of the LM CPUs, GPUs
like access to them. Even in Google Cloud. If you have $100,000 in Google Cloud, you cannot use enough GPUs. You can spend 100,000 Is that all you're finding? It's not we don't need, we don't need the same. It's the GPUs and use our train station. Okay, if you want to just use this simpler model, then you can actually do it off the small ones if you want to train something, and even chat to became 1% of champion champion team, you need all these special Nvidia cards and talk with each other and do the distribution and have the memory to be able to hold these models. That is totally untenable to do one percenter, so
why would you if it's that costly? Why would you do this as an independent agent? Why wouldn't you do this under the auspices of JP Morgan Goldman Sachs, HSBC agents.
First thing is large companies are focused on customer service, the customer service, it's actually really really, really expensive and also very mission critical relative to any other business. Because when you're going to customer service and a bank or consumer
facing goods Not not be it's not investment to
be neutral with the consumer. Because the consumer is calling you up. And they are talking about 1000s of dollars, it's not my Amazon order for this t shirt when my size is wrong. It's where is my money with many zeros, and you are a very state of urgency when you're calling this brokerage, you're back. So these mission critical, being able to automate this is a very, very key element of the business. The second part of it is if I go to JPMorgan, and I say, I can't do this, and I'll get rid of all of these people that you're paying million dollar bonuses to, I'm sure not many people are gonna be happy at JPMorgan. So it's kind of like a Kodak thing where a lot of these finance firms to be able to do this how to kind of live their most probably prized asset right now. So I need to get rid of our research and you can get it connect our people are our our people intellectual property, and be able to automate it, it's gonna be a lot longer process than someone did for the external have to come up with a first to have them do this now then do well, I mean, JPMorgan is hiring. You know, they're, they have 100 data scientists on it. But their first key thing is customer service. And being able to cannibalize their existing thing is not going to be a very interesting
because we don't know there is very little candor about customer service trends about shipping people or something. He's not snakehead, but he's like he was saying, obviously, a lot of value in his businesses, that customer service at the moment seems to be a lot of the a lot of the energy in business and kind of AI. Is the customer engagement, you know, kind of chatbots, I guess, right? Is how do I how do I remove or replace make my customer engagement more efficient? Right? And how do we start off with IVR? Press 123. And every human went to the call center? button, tap, right? Yes. Basically went downhill, because you can never talk to anyone. If
you think about it, they burn out because I'm sorry, I can't help you with this. And they can't, I'm sorry, this is the best I can do. I just I really sorry for your experience. And it actually makes other person angry too, because you don't seem very sorry, seem like it is Ctrl C Ctrl. V, I'm very sorry for your experience. And the people get burned out, the user gets frustrated. And when the stakes are higher, it becomes a bigger and bigger issue. So if you have thought about what I said is, it's very good at being creative and saying the same thing in many different ways. So I really apologize, this is shouldn't be happening, or I'm really sorry for your experience. If there's anything I could do except for this. Nigga keep saying the things in a different way. They're human attire human cat. So this is very attainable.
So it's got patients face. Yes, it's the
real emotion. It's it's robots. They have been in sci fi and all this stuff that I think. Yeah.
So you think that's one direction? Not a lot of energies happening in the finance world? Is everyone. Everyone's
cost customer service? If you ask me what's what's prediction for next year, customer service will start getting better with AI. This also happened in less critical domains. And then slowly I know for sure that brokerages This is not the the digital brokers, this is your number one, there's a number one, but
you're confident yourself. He said earlier, it doesn't have to speak right? I mean, I understand if it's like, you know, there's a delay, when you call a broker, you want to know immediately, like, how's my stock doing? Or can you track the speed? Are you saying that speed was going to be there that the customer service engine will be able to refer to existing at that point, swap prices check your money? See what should ship this is
what the rack This is my auto focus is on the site called rack, which is take my information feed into an LMS talk to a user.
What's the goal such fancy name right is like what is it reference free trial of
man degeneration, but because that's made it so I think metaphysical person who first kind of went to it, and then research in AI right now is just going in exponential speed. So it's kind of becoming like, if you can't call it the catchy acronym if you can't come up with a nice, whatever. Name your research, no one reads it.
Okay. So the right concept was I've got a pool of data that your machine hasn't learned yet. I want to put that into the Illumina
I mean, it's just it's maybe it's your private comment permission. I think maybe it's way that Qantas structure or whatever I noticed in that whenever shop, so it seemed that I need to know this. I'm feeling this to this LM and it should talk to the user in many, many, many, many different ways.
Okay, so you're saying that's where the customer service spaces, rag
is perfect for customer service? It feels like a kind of a small errors, it doesn't matter. Because I mean, it's not as critical. But you're trying to make sure that the LM knows this information and this information only, but
surely you have to train that allow them to respond and digest whatever you're throwing at it right? Or is it intelligent, nothing can throw anything at it?
This is this is where all the different flavors or brackets, you can you can you can train it more, you can actually haven't know what these things are, you can actually try to to further tune it. Or you can really find him if you
if you go on YouTube, there's all of this. It's a bit like the crypto days, there's so many people talk about rag and you know, what is it RH? Human? RL, hmm. Reinforcement Learning by human feedback, right? So YouTube has gotten it's quite interesting because it's moved away from crypto mania T's into like everything you've ever wanted. Right? So let's go back to the fancy so the customer service. So your your mission is when our customer service
is a real point above that. And the easiest way to say is that we're trying to create original content about stocks. So so we want to not summarize other people's content, we don't want to we want to be able to create original content with given all information about that stock be able to performance on the stage, but
don't you have to refer to Bloomberg and Reuters and NASDAQ some the you know, they've all got they're all spitting out content on a time crunch. This
is the viewer in finance. He says all the latest. Every very, very, very, very, very, kind of went through you're kind of like very out there. So many data, places that are cheap out there. Some of it's just free. Some of it is. Some of it is you know,
you're assimilating that
content. You're packaging. Yeah, one of the you're almost like becoming the next era.
Bloomberg is no word Bloomberg is very, very structured paid off. Yeah, we're trying to be is this person who's very long and can talk to you about? Hey, Napoleon, what do you you know, like you asked me about, you know, Tesla, while the Eevee market is kind of interesting, it's, it's, it's kind of out of flavor now in the US more much too much hype. So I'm not exactly sure. And adoption is tax laws. And if I told you like that, as opposed to you having to read back pages, yes. It's more I would, I
mean, this house full of financial advisors who
could be any time it could be any time whenever you want to talk about it?
Or how progress are you? Where are you
alright, to find the GPUs to be able to we're not fine tuning we're not doing these, we're actually trying to further tune something. So we have we get out of Google Now that we have money, we're doing something wrong, it just there's so much people trying to do this. So that it's it's very difficult. We're really two things under the box. So we're actually looking at that. We're trying to train and figure out what the train do benchmarks. Excuse me, Sophia, we write research papers and stuff like that. Can
you give him any help? We'll talk about talk. Alright, so anybody have a question? And then sign your nose, the topic and a Unison you can always move on to every question
thanks for sharing so much. So I guess my question is, you know, given the fact that it's so hard to find these GPUs I mean, what where does where it's what is the solution to that is are there going to be more GPUs available to people just have to raise more and more money? You know, how do you see that sort of that market playing out? And, and will it just become a battle of the giants because they're the only ones that have the cash to to make anything happen?
I think one, there's nothing else but it'd be a stock going up. So that's why
nobody's knocking the competitors. There
is AMD The thing is, just like anything else in voting, it's there's a network effect, the more people use it, just everybody just goes into it. So that's that. In terms of just general use case of these LLM, I think we've kind of hit a wall of how much smarter gets regardless of all this kind of talk. I don't see these MLMs going even better and smarter given this current way that it's shaped. It can't be more well, there's
rumors that a less smart why they can't coach like humans, the more you know, the more kind of like, well as smart in recovery,
I think there's a lot of tricks that open AI does, including that Hrlf thing that they're doing, it's not just reading to thing is, there's a limit finite limit of abilities to read without going to breaking privacy laws, right? So I can't go and read your whatsapp and, you know, read your own tweets when private stuff, right? So there's a limit on how much could be. And is definitely once you have that limit, it doesn't matter how many more nodes they call it, how much bigger your model is, it doesn't help if it doesn't have things to read, now be able to look at videos and pictures that there's still videos and things and more things to look at. But I think it's kind of getting to that convergence is a deep gap. When I actually went into this space in 2017, vision was all the vision. So it was all very technical. We weren't thinking about how to do these layers, what do you do the learning rate, all these different things, but then it kind of hit the wall, and then it became domain applications. Since time you have all these different things that are actually looking at exact pictures and being able to put that into use commercial use. So I think LM is the theory is actually kind of or the mathematics, it's kind of hitting some sort of wall. And it will be hopefully people like me, people in law, medical, any of these spaces that are using this as a first frontline kind of thing, like a first line of kind of how to interface, the triage, the content creation, the first things. So I don't think you're talking about jobs and stuff, I don't think anywhere in the near term that it's going to be able to change all of these things, because this is very non scientific, just off the wall theory of this is that people don't make TV shows about things that are boring and repetitive. I mean, there's a few shows, but you don't have too many shows about people coding, right, or accountants, you have all these shows about medical, I mean, I think it's a very important job. But it's very repetitive, right? You have these TV shows about doctors, so er, doctors, lawyers, and sometimes some finance people. But the thing about
Ernie is they do naughty things. Yes.
But the thing that actually, if you think of that kind of attracts these producers can these things is that a Doctor House says something and does something different from this list, A to B to C in a memorized medical book. So you want to go to a doctor that can actually think differently. So I don't think that's ever happening. But probably in the five years, when you go to a hospital, you're probably having someone who's just typing in your stuff. And it's telling you whether your shoe has one two or three. If you go to a back, you won't see a private, you won't see a bank teller, but maybe someone if not, you know, the interfaces, and they're all very well done someone just kind of typing in yourself and of telling you all the different things.
Right? That doesn't answer his question about that kind of the GPU dynamic, right? Because
it was so so excited. So what I was saying is that the GPU dynamic I think, immediate will go up, but I think the usage will kind of come down but be supplanted by people like us, who are more domain implements use it for fine tuning, weather tuning, modifications, and then inference. So this is where it's probably headed. But yeah, I don't I don't see this lasting for years.
But from an adoption layman. Kevin's speaking, obviously, from an adoption perspective, if Google and Microsoft and Mensa are all rolling out AI solutions into their platforms, the general public is just going to settle with those right? They're not going to want to work outside of those environments.
Well rather use tragedy before which you go and ask for your medical advice and you would adjust for taxes.
Or I mean, it's like so for example, in I've spent a bit of time in advertising people, you know, Metro is rolling out better ways of buying ads, using AI and that llama to As you know, I look at all the analytics and go, this is a better way of reaching. You know, that middle aged guy in Hong Kong who wants a beer or whatever. There's a.
So for marketing, yeah, for some of the soft skills that Google is not a tech company,
Mehta is a great thing.
It's an advertising company. Yeah. Industry, if you look at their operating deficits, their operating profits are composed 110% of our customers, they lose money on Cloud, they lose money on everything else. They even call their other stuff, Other Bets in the financial report. So meta is a advertising company is not a tech company. It's an advertising company. And these companies are just using this technology to do their core business. In Google, everything revolves around out.
So you're saying because they focus on that they're not driven to come up?
No, no, I mean, these things will be used in that space. But I'm saying that how is going to be used in other spaces, it's not going to be open AI is not going to be these kind of companies, they won't be, they don't care about these other industries, right. So it will slow if the general LLM 's are hitting a wall, how much better and better it will get, then it will be more, you know, people like us to be able to his case. And Google using it for advertising or Facebook is for advertising. Any more questions?
Say no for us.
Yeah, my name is Ed degree. Yeah, I'm gonna couple of observations that I've been on the other end side. It's interesting what you say. But there's a lot of open source LLM coming out now. So if you work on the basis, Alabama is going to get to some sort of equilibrium point where you know, maybe at the 50 trillion conductivity, we reach a point where there wouldn't be any more, then if there's open source models of those going around, it's all going to be customization. And the other sort of dgpu problem was goes away.
We actually use llama, and you're trying to further to llama. So that's why we need to pues we're not, we don't have to, you're not doing it. Yeah, we don't have 100 million spare. And we all have a exclusive contract with AWS, we're trying to use lava to and we're trying to further tune it as in we're trying to have shapes the inner workings of what, how to read more, as opposed to doing fine tuning. It's fine tuning is actually doing mock exams. It's like as if student, like a high school student went through high school without learning any classes, but spent four years just doing mock exams. And they would aced SATs and or any exam that they had, but they might not understand it. So we try to have it actually read. And hopefully it has gained some extra knowledge. So I think it's actually very true. Open AI is the shining, I think I think is HR enough to detail on all these tricks that they've done with a singular focus of trying to do this as opposed to Google who has to worry about their search, but llama reuse lava mistral is some French model that's coming out open sources. We're using all these open sources, which are used by the domain people. So instead of having open AI, Google and these researchers trying to create large models, general models, it's probably going to be people like us putting, you know, domain use case models, or so that's where I think the GPUs will be used. Okay,
yeah. Yeah. So the other thing that I find really interesting between your two presentations is that in a sense, you've contradicted each other because you've talked about the real focus on on human human interface or insistence to provide chat GPT type things or for customer service. So in other words, more and more connectivity with human beings have more and more perceived empathy. As they say, if you couldn't fake empathy, you can fake anything. And the the idea that you can, you could build these better and better response systems that can give someone a feeling they're talking to someone with empathy, and is interested and can give them a recent presentation they'll be interested in. But you know, Kevin was talking about how that would probably be the hardest bit to do in terms of in terms of being able to appear like you care. You know, he mentioned about, you know, doing all the diagnosis, but not being able to tell you in a very sensitive way that you've got 10 weeks to live. So your two positions a slightly different in that respect slightly,
but yeah, I would actually say that. So having worked there To try and track by, I know, when I was younger, I'd be like that salesman. Well, what is this guy doing and he just, he's just there partying and kicking people out for drinks, but the best ones can really just hate Napoleon and just melt that customer to do whatever that they want. And that's never gonna be a after the doctor that's really good and empathetic and the ones that you want to go to maybe not the Hong Kong doctors, but you really, you know, our care carry, or their strict with you as a doctor, as do you want to, you know, live as long as you want or just something like this, those kinds of people that that's not possible. But when we were talking about customer service, you're actually getting the worst side of humans, most overworked person getting every people stress, the the darkness just coming in. So we're getting an AI to beat the worst, I get frustrated when I get customer service. And then I hand out the phone, like my God, I'm such a, you know, aihole, like, this guy's just doing their job. But at that moment, you vent your frustration, and you know that that person is now in a but
also I mean, there's me using PI, right? Now, pi pi.ai. Okay, so it's one of the founders of DeepMind. So pi, and you go in and you can talk to it is unbelievable. The empathy that prompts you to go to war and talk about having a hard day at work. Do you want to talk about your relationship? Do you want to talk from the word go, they're not saying we're here to help you, like analyze the next to triage, that are going to vary. It's the guy behind it started his life off trying to make peace between people. So he's created idle AI and you go in is Godsmack. Or try it. And I don't know, how were they doing their data from, but I tried to test off.
This is where it's really important to know the history of AI. I mean, I've been in this space, like 30 years. So I could say never is a very dangerous word to use, we used to say that they've never reached the stage we've reached now. And we have, so we don't really know where we're going. And, you know, bear in mind that in the 70s, even in the 70s, and 80s, they were experimenting with, you know, a laser, you know, trying to help people with manic depression. We've been in this space for so long believing that we could get there and now we're starting to see some things we never thought
another voice of question
is getting, we still need to thank you for this, this is amazing. And he has a feeling that we are diverse. Isaac, the founder of here, that means we are using AI to generate news, you know, that is based on fact checking, you know, later to present users instead, information from all the groundwater. So yeah, we actually definitely have a vertical part. But like financial means, from the same open data, open information. We are focusing on trying to remove those biases from different information sciences, because they're mostly conflicting with each other. My question I get on the show questions, but your third party services, so language model, you're using llama, which we also compare different models. So I just wonder why you choose llama over the knee is TPP fault, which is I think it defaults very much, you know, the the crown, the tools today now, and what's your opinion on that? And why choose that? And would you worry that someday some other competitors would use GPT. to instantly to, for example, to surpass you? Yeah.
Yeah, so there's a lot of things LGBT or open AI and as you do, but one of the things that doesn't let you do it is further training or breaking points. So what do you think he is, doesn't let you do that? And it's such a massive model that you cannot do it. We can't do anything about it. And there's a lot of extra things that GBT has done to make it that much better. and law means a lot of human effort that went into it. That's why it's human reinforced learning for human. Yeah. So I think that's, that should be part of it so we can use it. But I think it's kind of like GBT is the best by far for general use, the way that we see a lot of architectures working is very small models doing small things. Everything's nano size. So that first triage, whatever it is, good bits, the second diagnostic was smaller. And we've actually had research. And I think, eventually USC professor, somewhere that even smaller models can handle most small exact tasks, you don't need to have GDT, for me to just, you know, pulled out PE for a stock, right? So we actually don't even sometimes use llama we use other things, small things that we can do in small GPUs use for GPUs to train it. So that when you use llama two, there's many flavors of it, sometimes we use the smallest one. So we need to be able to use it at certain places, so that it's gotta be small enough for us to modify to custom work on it, and then use it in
production. Okay, so
actually searched one thing that I had, I was talking about as interesting because using this, if somebody, probably straight is a pair of experts, you should never say never. But I'd actually also say that this also goes back to the why people are really interested in some types of under hood or whites. So the news is that for young people, I think they're already kind of tech savvy. So they're used to looking at things out of the hood, or always, especially engineering people, anyone that's 35 and under are used to looking at under the hood, I would say people that are our age 40s 50s and above, we grew up with sci fi that had all these things that promised in the 80s that there will be AI and it never happened. Even to this day, we drive the same kind of cars more million Tesla's we drive, you know, keys in the car, maybe we open locks and our restaurants are served by people. But throughout the 80s, we were taught in 1020 30 years without flying cars, and we'd have robots serving us everything would be done by AI, or you would talk to a robot or a computer and you wouldn't have to type it on the screen. The computers have gotten better but the way that we interface with the computer has not changed in 30 years. So I would say that we are all kind of we're very fascinated by this young and old because it's been our culture for the older people young people are just kind of you know, trained that way. And I would say that we certainly never it's never in 30 years was impossible but 34 years ago we thought this stuff would have happened 20 years ago so that's kind of like the
usual to the like there's a guy that I follow as emails as they might be has read this for for exponential view. And he says the all the pieces are in place the videos the power the habits, often it only happens when you've got multiple pieces in place right? It can't happen independently by the way one more question. Any more questions? No? Very good well thank you that was I'm I when I talk to people in a while I was my brain just lights up so thank you that was really helpful because we have another half an hour to hang out and go wherever you go. So if you want to say you've got some tokens left is that what tokens spies real cash. Thank you. Thanks.
Gin Tonics are good here
my brain spin. More on coloring. That answer your question about you asked about the stock stuff. If you want to know more gun drilling now. But he was interesting, as he told me originally, that he wanted to build a solution for consumers. Then he realized consumers was too much hand holding. So then he moved into a platform for b2b. To sell, to sell. To sell a research tool to consumer is a very difficult thing. There's a lot of trust and I'm just gonna go to the loop. research you can bet your bottom dollar more sophisticated already and that just helps us with sports it is not a personal I've just got a lovable science