Some Future Day

Marc Beckman: Milos, it's so great to see you today. Welcome to Some Future Day.
Milos Bujisic: Thank you, Marc. It's a pleasure to be here today, and I hope we're going to have a great conversation.
Marc Beckman: Milos, we always do. So, um, there's [00:02:00] no doubt in my mind, but I want to cover a topic which most people don't talk about from a macro level. Where is the upside to artificial intelligence commercially? Why aren't we seeing mass adoption of artificial intelligence at the enterprise level? Where are we going as it relates to generating new streams of revenue, new forms of income for individuals, for businesses?
When we look at AI. So that's the macro conversation I want to get into, but let's back up a little bit. Why don't you take a second to introduce everybody, uh, to who you are, your world, talk a little bit about your background, and then we could roll all the way up to that issue of whether or not AI is truly in the golden age.
Milos Bujisic: Thank you so much, Marc. So my name is Dr. Miloš Bojšić. I'm a professor, uh, associate professor at NYU and New York University, uh, [00:03:00] specifically an integrated marketing program, and I'm a co director for faculty research at School of Professional Studies at NYU. Apart from that, I've been a serial entrepreneur for most of my life currently.
I have two businesses. One is located here in the U. S. It's called MADIntel, but what we are really developing is a platform called BrandSocialValue and we have a company for pretty much development back in Serbia where we work on the development of that platform, but also provide some services for other potential clients.
What we really do is similar to what I actually teach in school. So in my classes, I actually teach digital marketing and I teach, uh, different marketing analytics courses, specifically, uh, statistical measurement analysis and the research, and we do a lot of branding analytics. So ultimately what I was trying to.
accomplish both professionally outside of academia and academia is to create [00:04:00] a use of AI that would simplify analytical process for branding for people that are not very versed in classical analytics using Excel, using SPSS or R or a state or whatever platform, but something that can simplify the process and allow people to generate, uh, Outputs, uh, that provide valuable information using AI and interpretation of, uh, those outputs.
And this is something where I think, uh, there's a lot of space where there's potential for growth, because as you've mentioned, we still haven't found different ways. When I say we, I think the entire industry is still struggling with ideas, how to monetize everything we are doing, especially in that corporate space, in the enterprise space.
Marc Beckman: So, so for the, um, people that are viewing and might not understand the terms of art that you're using here, can you just explain, [00:05:00] uh, for the layman? What this means exactly. Why is it important? And maybe you could give an example of how that might work as it relates to a brand than a consumer interacting.
Milos Bujisic: So you mean on the side of A. I. How? How does it work? So what we've seen is that there is now, since November 2022, there was a mass adoption of AI among regular consumers among, uh, end users, according to some statistics, and I'm questioning that one, uh, India has 40, percent of population is actually using generative AI on daily basis, much higher than US, which is kind of interesting.
That's the number one country in the world in terms of usage, but, uh, we're still seeing that there is a level of reluctance among corporations. And there are fewer ideas of how to actually monetize this. Where is the money, basically? [00:06:00] Is there something at the end of the rainbow? Yes, we can replace potential employees.
Yes, we can, uh, create some efficiencies. But are we generating anything new? Ultimately, it's kind of weird because the word is a generative AI. So we are supposed to generate things, but are we generating value? That's a big question that we have. And I mean, as we move through the conversation, we can actually try to respond to that one.
Marc Beckman: So in your opinion, are we generating value?
Milos Bujisic: Uh, I don't have, like, let me be perfectly honest. I cannot give a straight up answer to that one. Yes. And no, uh, Yes, it's generating value, but it's generating value through, uh, a process which is not necessarily what we want to accomplish the generating value currently mostly through cost reduction. So we're saving money and indirectly generating value in a sense that, uh, Truly no new value is being generated, but for companies, the perception is that their bottom line is going to be affected.
Now you cannot actually build your [00:07:00] growth strategy on cost reduction. Like no company became a trillion dollar company or extremely successful purely on cost reduction. You have to generate Completely new industries. You have to generate a new value for consumers, which is ultimately what the marketing is, and there we're still lagging behind.
There are some attempts, but it's very few and rarely within enterprise sector. I would say Probably the biggest value generation that you could actually see is among students that are using this to create their own assignments. And it's generating value because now they have assignments that are written in a better tone, that are more extensive, but not necessarily more creative.
Marc Beckman: Mm hmm.
Milos Bujisic: And this is where the problem is.
Marc Beckman: Yeah, let's let's let's focus on the enterprise piece of it. But because what you're saying, um, is interesting. It's inversed, uh, in a way, like most people are saying, artificial intelligence is great because it creates efficiencies, both with both with [00:08:00] time and cost. And it, yeah.
As a result, it eliminates expenses. It takes expenses off the books from corporations. Okay, we understand that. So you're saying cutting, saving, cost savings. However, your analysis is very interesting and, and most people aren't looking, and this is where I'm saying it's the inverse analysis of, of most.
You're saying people are not using, let's say, generative AI or, or, um, uh, other forms of AI to build new enterprise to capture new streams of income and new streams of revenue. Is that, is that your analysis?
Milos Bujisic: Uh, mostly it's especially for, that's especially true for generative AI. Other forms of AI are actually a bit better in, uh, value creation. And arguably, I mean, it's first, there's a whole notion of how it would be even defined and where does just algorithm stop and AI begins or, uh, but, uh, to give an example.
So, uh, for example, yesterday, I was [00:09:00] tracking a website of, uh, I believe it was a Lululemon, one of the, uh, uh, apparel companies, um, and they've, they've now, they're now using, uh, chatbot, which is AI powered. What does that chatbot, uh, ultimately do? It removes an employee that would sit somewhere in some, uh, Call center, probably abroad and it's saving costs.
It's going to get so good eventually, even though right now it's, it's pretty good. It really to, uh, replace a human being and it Cut costs for a company. So when you look at Lululemon's, uh, bottom line, uh, they had a labor cost of let's say 30%, uh, of their sales. Now, suddenly, because they're using Genai, uh, the labor cost drops to 20% of the sales.
And, and Genai, the cost of genai was maybe 2%. So they've saved 8%. So ultimately, bottom line is, uh, affected. But did it increase their revenue? That's a big [00:10:00] question.
Marc Beckman: Well, maybe it did though, because if we look at it as a situation where we remove that human being and automate it with the chat bot, it's a clean swipe, right? And then it just drops the additional, um, income to the bottom line. However, perhaps Lululemon. Is doing what many artificial intelligence pundits talk about, which is freeing up the monotonous tasks and allocating then bigger thinkers, maybe a stronger designer on the design team, maybe perhaps a stronger marketer on the marketing team so that they can develop new top line revenue, new top line growth.
Milos Bujisic: Yes, that could, that could be, uh, uh, one potential explanation if it really does free up, uh, labor hours of, uh, top developers, which arguably it's still tackling mostly entry level, uh, people within organizations, but eventually it will [00:11:00] save up time for people. I mean, uh, what we are trying to promote is the whole notion of symbiosis between.
between, uh, human and AI so that you as a person would be able to perform more tasks during the day and be more creative yourself, not necessarily they I would do it for you. And I think in that sense, it could potentially generate value, but it's still an indirect impact. Still, there is this mediator in between.
So it's saving somebody's time so they can actually focus on creating value themselves. But it's not AI that is creating value. It is still a person that is creating value because they're, uh, Time is, uh, saved. So this is where I think the big problem is currently, uh, A number one thing that we are thinking it could, uh, the AI where the AI could actually create new value is about personalization, that AI would ultimately understand the customer better and be able to provide personalized, uh, results for them or personalized advertisements, or [00:12:00] ultimately like suggest products or ultimately even directly buy products for them.
And this is another thing that is actually common in different retailing websites is I don't, I'm not sure if Lululemon actually use it, but it's like you would get AI requests. commanded products. That is not new. And that is not generative AI. And Google has been doing, uh, personalization pretty well for two decades now, almost.
Marc Beckman: Milos, I agree. I think that. The technology that you're talking about as far as like machine learning, customer purchasing habits is stale, outdated. It's boring. I'll go into Amazon and we'll say, Hey Marc, we see that you purchased this black t shirt and therefore maybe you like this black blanket or this very ugly black butterfly t shirt or whatever it might be.
I think it's antiquated. I believe that the personalization component in marketing. At least the piece of it that's going to be very exciting for me is going to be, um, [00:13:00] creating a little bit more seamless relationship between the brand and the consumer. For example, how many of us are bombarded with annoying email blasts from brands that are talking to us about things that might be interesting?
You know, to our, our wives, our loved ones, our children, but I, I don't really care to see another, you know, baby, uh, outfit or a toddler outfit in my, in my feed all day. So I think artificial intelligence on the personalization side can ramp it up as it relates to being smarter in conversation with the target customer.
It could limit the level of communications vis a vis social media, or let's say, let's say earned, uh, paid and own media. It could become more customized, more specialized. It will be more cognizant of my time. And I think the artificial intelligence agent will be an interesting part of that too. So, you know, talk to me a little bit about it, Milos.
When, when I, when I want to, when I'm taking care of my business, my day [00:14:00] to day business, but I know that I need to go out there and buy that black t shirt, how can I implement my own personal artificial intelligent agent? To go into the marketplace, identify what I'm looking for, find the best price. Like, how does that piece come into play in the, on the personalization side?
Milos Bujisic: So the, the big, the big question there is who is going to ultimately control, uh, the algorithm or the agent, uh, uh, themselves. And are we going into an, uh, open, uh, source, uh, model, or we are going to have a few big players or multiple players in the market, and this is going to influence a lot how this value creation is going to play into customer side and also onto enterprise side.
So purely on the customer side, let's imagine a couple of scenarios. Scenario number one, where a couple of big players, let's say it's OpenAI, maybe it's Google, maybe it's Microsoft, have created very advanced platforms [00:15:00] that we as users can create. can adopt and those platforms learn about us. Now, Google, actually, even though it's not, uh, ahead when it comes to the raw development of general AI, uh, would still have the best chance of developing, uh, this in because it has already existing data about each of us that no other company can match.
Other companies have a lot of data, but Google is still by far the, uh, the most. The one with the highest amount of individual data. So you could, Google could create a custom bot, like call it a personal shopper, personal agent, uh, an AI that would do all of the functions for you as an individual based on information that already has about you, but still ultimately.
The bot is under control of Google, and this is actually critical, uh, because currently what we are thinking about is that one of the features of, uh, generative AI could be that, uh, [00:16:00] Google's monopoly on search, at least in, uh, Western economies. might basically collapse in the next few years if people keep using search through, uh, generative AI instead of searching in, uh, basically Google's website.
This is where the revenue is. That company is ultimately living off their, uh, Advertising revenue. You Google could divert that revenue or other players like OpenAI or, uh, for that matter, like, uh, Meta is also developing their own platforms could divert some of that ad revenue to come from the people's usage of generative AI, where you are actually, uh, GPT for certain information.
Let's say you're looking for restaurant and then the recommended result is not only based on the most relevant results, but it's also based on those companies that have paid to open a I in order for their, uh, restaurant to be promoted in the response that [00:17:00] ChatGPT gives to a prompt of recommend the restaurant nearby that serves Italian food, for example, and this would be a great source of revenue for a company.
Uh, ultimately it's It's arguably not creating a completely new market, it's just taking away from, uh, search, that Google is, mostly Google, I mean there are other companies as well, within that space. Uh, the other scenario, Is if we have an open source, uh, solutions where it's not controlled by a company, but let's say you as an individual, which is going to always be somewhat difficult because of the technical problems, uh, that come with, uh, self installing something you as an individual install your own.
A generated AI bot that is going to you teach it, uh, what you want it to know about you, your shopping habits, your, uh, preferences for vacations, uh, your personal relationships and so on. It [00:18:00] learns from you, but ultimately there is no company behind it to make money, uh, through, uh, recommendations. And this is going to resemble more than anything ad blocker.
Like we are, Google is fighting, uh, our other companies are fighting against ad blockers within our browsers. But, uh, if you had a bot like this, it's ultimately serving the same purpose, like you were saying. It could filter out all of those unnecessary emails. It could filter out all of those offers, uh, to products that you really don't want.
That is providing a lot of value for a consumer. But there is no company that would actually benefit from that.
Marc Beckman: It's interesting because I often say that artificial intelligence and cryptocurrency are inexorably connected now. And I believe that this will be, you know, these two. Um, emerging technologies will represent the largest growth sector in our economy, in America's economy over the next four or five years.
I really believe this, um, what, what I'm taking away from [00:19:00] you is a concept that really isn't addressed a lot. We have self sovereignty as it relates to your ID in the Web3 space, in the crypto space. And really what you're trying to describe here, I think is self sovereign artificial intelligence. Am I right?
Milos Bujisic: So I I, in a way, uh, for that to actually work, it
Marc Beckman: It's kind of cool.
Milos Bujisic: it would have to be a hybrid of, uh, blockchain and uh, uh, ai where that sovereignty that, that you're implying would actually come from a combination of that your persona would probably exist within blockchain. And your persona is what is driving AI to make, uh, decisions for you.
And this is where the role of trust comes into play. Who would you trust more Google to provide that service? Or, uh, a bot that ha that is not affiliated with anybody else. It's purely yours. Trust might be critical, but there is another problem, and this is convenience. We know that Linux is superior operating system, uh, to both, [00:20:00] uh, iOS and Windows, but people don't use it because it's just annoying to use for most users,
Marc Beckman: Right. So centralized platforms, you're saying therefore the, the Google trained AI agent, individual AI agent of as much as we hate to say it may be more convenient and therefore it might equal mass adoption than us training our own AI agent and it being, um, controlled by the individual.
Milos Bujisic: potentially. Uh, the only scenario where I see this not happening is that AI actually. truly becomes smart enough that it becomes convenient on its own where the convenience does not stem from a bunch of programmers that are sitting at Google, but the A. I. Would, uh, these redesign itself to fit your specific needs.
Marc Beckman: But the initial still we need this initial step where we would have a community that would be dedicated to that. And that has to be a community of like minded individuals [00:21:00] from around the world that they're willing to share it.
So Milos, how far are we really, or how far are brands from? Interacting with individuals, AI agents where again, you, you wanna spend more time with your family, you wanna spend more time concentrating on your career. You don't wanna go over to the shoe store or even take the time to go on to Amazon to buy a pair of cool a, a cool pair of Nikes.
Marc Beckman: How far off is it until you can unleash the Milos AI to go buy that, that you know, that new pair of Jordans?
Milos Bujisic: So I think we are still, there's still a bit of time before we adopt that and when I say a bit of time, that could be anywhere from I year and a half to two years all the way to a longer time and it's not only due to technological factors. I think that the tech factors and, uh, sophistication of AI models is around the corner.
I think it's due to sociological factors where [00:22:00] a lot of technologies require time within a society. for people to accept them and to start to use them. It is something that we've seen with pretty much every technology out there, from railroad, uh, all the way to, uh, AI, uh, right now that it takes a bit of time for society to realize what its benefits are and to be willing to, uh, sacrifice some of its habits and, uh, uh, not necessarily convenience, but actually gaining convenience, but, uh, uh.
Venturing into unknown space where What we've, like I I, I've conducted several interviews, uh, two years ago with, uh, some of the leading, uh, uh, executives among others, the former CMO of, uh, Meta. Uh, and we actually wanted to explore which products, pro product types and categories people would be most willing to, uh, give out their shopping, uh, process to [00:23:00] AI that would make decisions for them.
pretty much boils down to for whatever is boring to shop whatever you're not enjoying the shopping process so let's say toilet paper i don't think there's anybody that enjoys the process of shopping for toilet paper why wouldn't ai make a decision for you but then when it comes to if you want to buy a luxury watch owning a watch is part of the experience buying that watch and going through the process is probably as interesting so this is something where ai might not make a but if you're buying a watch as an investment AI will make a better decision than you might make.
Marc Beckman: I see. I understand. I get it. Um, so, you know, one of the conversations that you and I have had historically, which I find fascinating, um, applies to where we are in the timeline of growth for artificial intelligence, um, from a macro level. Specifically, you, you took me through recently. You took me through the evolution of Ecom in Web 2 and, uh, made a point that [00:24:00] many of us forget where there was a, a bubble and it popped.
Can you, um, go through, uh, where, I guess that experience that we had, um, uh, as consumers with regards to e commerce, the early days and then, and then the, the, the, the bubble bursting and then going forward and analogize that to where we are with regards to the use of artificial intelligence at the consumer level.
Milos Bujisic: Uh, so the best chart that, uh, we could follow for this is actually, uh, the amount of, uh, investments that were going into, uh, tech companies. It's, uh, uh, pretty much, uh, looking into how much were different, uh, investment bankers and hedge funds and so on. How much money they were actually putting into, uh, uh, Uh, different companies and what we've seen.
So since the mid 90s or actually kind of since the early 90s, there was a slow rise all the way to 97. And then [00:25:00] suddenly, there was a huge increase in popularity of e commerce companies. Amazon that we still know of and it's still around, it's still here, uh, was one of them, but then you had companies like, uh, boo.
com, uh, and many, many, many others that pretty much, uh, ceased to exist very soon. So for a couple of years, you had this huge uptick and there was a lot of, uh, Uh, excitement and what was driving the investments. This was, I mean, of course, ultimately greed drives the investments, but, uh, it wasn't unfounded greed.
Uh, there was, uh, true belief that they're coming into a completely new industry that is going to generate new value, that is going to replace how we shop, which ultimately we've seen it did happen, uh, and that companies that are The ones that entered this space first are the ones that ultimately would create biggest returns for investors.
[00:26:00] What we've seen is that money started pouring into those industries because there was a promise they are going to make a lot of, uh, return in the future, but then it turns out that the infrastructure and the society, like there is infrastructure, there is society, there, there are two things when I say infrastructure, I mean, like, The number of, uh, people with an adequate credit card that they could use, uh, online, the ability to deliver products and to have a reliable shipping that it doesn't take two weeks.
Uh, the mentality that most people are not going to steal the package that gets delivered in front of your home. So these are not tech factors. These are social factors. There was still not enough maturity within the social factors and the technology was there. Pretty much. And suddenly all of those companies were really not making money.
They were not making a return. And then we entered the disappointment stage when we realized, yes, this is amazing. And then, uh, stock market ultimately, [00:27:00] uh, Overblows every emotion both the positives and the negatives. There was a disappointment. Oh, there is no money in this the few that survived 2002 to 2000 and 2001 period Started going in a steady growth.
And then if we follow the investments in the following Decade or so we could see a gradual growth Uh, to kind of return to the pre dot com, uh, levels, even though it actually never returned to the peaks of the pre dot com levels, the investments, but the revenues up absolutely dead. And then we see, uh, Amazon actually started turning first profit in 2001, but arguably this was, uh, Just a small profit.
It really started making money, uh, years, uh, later, but ultimately we cannot argue that e commerce became a big thing and it is a legitimately important industry, but it had a period of disappointment [00:28:00] and I'm worried that we are going to see a same trend in AI two years ago, or it's not even two years ago.
Now, I guess it's like, uh, uh, from November 2022, actually two years and a couple of months.
Marc Beckman: Yeah.
Milos Bujisic: After that, there was this huge optimism in a new type of AI. AI existed for a long time. Different types of predictive algorithms. This generative AI existed as well, but it became readily available. And investors are now obsessed with it.
But still, the companies are not making a lot of money. And I'm not saying the companies that are providing AI. I'm saying the companies that are supposedly using that AI. And we're probably going to see a part of disappointment and then slowly. As the society, uh, starts to understand and starts to adopt the new life with AI, we're going to see growth.
And I cannot see that this is going to disappear. I cannot [00:29:00] see that this is not going to ultimately succeed, but I could see that it's going to be a bumpy road with one huge bump that honestly might be just around the corner.
Marc Beckman: Well, Miloš, let's break it down a little bit. You, you reference, uh, companies that are currently leveraging AI, um, for growth and, you know, it's very public, right? Microsoft has just invested billions and billions of dollars, not just in open AI, but in Microsoft products too. Um, I'm curious, have you, um, personally used all of the Microsoft A.
I. Um, software within the suite of Microsoft, uh, products
Milos Bujisic: No, no, not all of them. Far from it. I mean, uh, beyond Copile. Far from it. Beyond co pilot and some of the features of co pilot. Uh, I did not use it too much. Uh, I mean, I did test most of the things, but I cannot say that I adopted it into everyday operations.
Marc Beckman: a lot because you're an early adopter. You're curious. You [00:30:00] embrace new technologies and even you where it's it's made and built in a convenient way are choosing to not implement.
Milos Bujisic: Uh, well, again, it's, uh, the, the question is, uh, what is the value that it, it's giving me and how convenient ultimately it is for usage. The problem with some of the Microsoft, uh, products is that they're not necessarily the most, uh, convenient, uh, products out there. And, uh, Microsoft struggled with pretty much every one of their divisions apart from office, uh, for many years.
And like you have something like Microsoft Teams that. It reached a high level of adoption, but you're rarely going to hear people, uh, uh, speaking in a positive way. Skype is almost but forgotten, uh, tool, uh, by now, uh, and similarly, uh, I don't see Microsoft being able to, uh, uh, utilize this unless it's truly incorporates it well into their, uh, office [00:31:00] products.
And it's trying to do that. It is doing that to a large extent, uh, but it's. Still in a phase of being somewhat gimmicky, uh, and not completely, uh, uh, useful that people that are used to doing things in a certain way would be willing to stop doing it that way and now use, uh, AI for that. Maybe this is going to change for younger people.
I'm used to using Excel in a way I've been using it for the past 20 plus years, but maybe somebody that's a brand new user of Excel would not use it the same way as I do.
Marc Beckman: Well, it's interesting because earlier you mentioned one of the other major tech companies Google and referenced how artificial intelligence on the search side will totally or could totally disrupt Google. I've been pretty public about it. I think the Google search experience is substandard. You end up with tons of advertising in your face and then Um, the result is a bunch of links that I need to then [00:32:00] go to and, um, click in and do some research now with chat GPT and even deep seek, which we could get into right now.
The text box is a lot bigger. It allows for you to get a lot more detailed with regards to not just, uh, the results of your, your query. Of your input, but also the reasoning behind it. It just seems a lot more intellectual. And frankly, it's when you talk about efficiencies, Milos, it's like this instantaneous, right?
So I think the idea of mass user adoption of AI will probably come first and foremost through search. It's in my opinion, it's already a better tool. And with deep seek, it's free. So, um, what do you What's your opinion on the search component of AI? If you want to start with DeepSeek, I know that's been top of mind for everybody now.
Like we can, we could break that down too.
Milos Bujisic: So just to mention something about Google. I mean, ultimately, the positives of Google search did not change for the past almost two [00:33:00] decades, the negatives that you did not have as much ads as, uh, as you have now. So it's it is definitely an outdated thing. And just imagine in the, uh, tech environment. We talk about two decades.
This is something that is absurd. Things change almost yearly and things haven't changed for two decades. And I do believe that probably the biggest disruption in terms of, uh, which companies might be affected is within the search space when it comes to the chat agents. Now, going back to the deep seek, the fact that it's free is a Big difference, uh, for now, but we don't know if it's going to stay free forever.
I think the ultimate, uh, difference is not necessarily how much it costs. Like 10, 20 doesn't make that much difference, but being open source is the[00:34:00]
So this is where it becomes different in a sense that, uh, Individuals themselves, users themselves, and companies can create their own versions using this platform and not necessarily pay anything for it. And this is where it becomes, uh, fundamentally different from what you would, uh, use with, uh, ChatGPT.
Uh, I'll just use one example, uh, still, for example, uh, in academia, we extensively use SPSS for, uh, data analytics. It's IBM's, uh, statistical analysis software, and it costs. Uh, not too much, but there is a cost involved. Majority of industry adopted to using R and Python. Not because they're superior, they're free.
And you can adapt them and there is a big community that is consistently creating new things for them. And this is where I [00:35:00] think, uh, DeepSeek might turn out to be a winner. Not in a sense that DeepSeek as a company might win from this, but that companies are going to ultimately create their own versions of DeepSeek, adopt their versions and start utilizing it.
As long as DeepSeek keeps developing and improving the algorithm and still sharing it for free, which we do not know. And I'm questioning if that is really going to happen in the future.
Marc Beckman: So let's, let's stay focused on the quality of DeepSeek's product before we get into last week's news surrounding, um, this, this upgrade of its newest, of its newest model, the cost thereof, et cetera. Um, when you talk about DeepSeek. I know you and I connected on this last week. There, there are certain limitations with the product, right?
I know that you were comparing, you were using the same text prompts for both, uh, GPT, chat, GPT and deep seek. Can you bring the audience through, uh, your experience and what you found as it [00:36:00] relates to, um, how deep seek provided an output, how it responded to your inquiry. And then the same thing with regards to ChatGPT?
Milos Bujisic: Well, uh, ultimately, uh,
Marc Beckman: it's really fascinating,
Milos Bujisic: both, both ChatGPT and DeepSeek, uh, are not completely open in a sense, uh, that there is a level of, uh, management and there are a level of internal limitations of what it can and what it cannot to do. Of course, those limitations are vastly different. So if you ask any question that is involving, uh, China.
you're going to get a very limited amount of answers. Like, for example, the simplest one that everybody has tried by now, what is Taiwan? And it pretty much does not provide, uh, answers so far. But, uh, on a funny side, when I use ChatGPT and ask it to create an image of, uh, Snow Princess, uh, uh, I would also not get the output because it violates [00:37:00] copyrights.
I never mentioned Elsa, by the way. There was no mention of Disney. But, like, for my daughter, this is an activity we have every night to generate images that my daughter describes. But, uh, it wouldn't do it. Yes, you could jailbreak it. You could go around it and ultimately ask it to reframe your own prompt so you would get the answer.
But what's interesting is that DeepSeek is actually easier to jailbreak. So you could get the answers you're looking for, but it requires a bit of effort. But eventually they'll become better in controlling the content that is being provided. Going back to the open source side, if you have your own instance of DeepSeek, You don't have to have those limitations right now.
They cannot be hardcoded.
Marc Beckman: does DeepSeek incorporate a certain type of bias as it relates to China in general? I think your experience with regards to, yeah.
Milos Bujisic: Yes. So, uh, I've used the same prompt [00:38:00] regarding like comparison of education in US and China and, uh, ultimately the tone changes and, uh, the filler content, uh, actually changes quite a bit. I would say that, uh, DeepSeek is probably a bit more knowledgeable of, uh, Chinese culture. So incorporated elements that they've never mentioned in a prompt like Confucian, uh, culture and so on.
are giving advantage to Chinese educational system, which were not properly, which ChatGPT never mentioned. So it does certainly have a level of bias, uh, that I think is expected if we would be able to actually figure out what are. All of the hard coded delimitations and hard coded information that might exist within it.
And I wouldn't be surprised, uh, that there is even a literally a slider somewhere in a system, similarly to how we use temperature, uh, to control how creative you want AI or how precise you want AI, generative AI to be, that there is a slider for the level of, uh, [00:39:00] bias or the level of, uh, control that you might get.
It's. Probably buried somewhere deep in the code, but, uh, it certainly could be programmed that way.
Marc Beckman: Is there concern though? Like, is there, like, is it a real issue? Should anybody care about that?
Milos Bujisic: Well, I think it depends on what you're doing. If you're doing political science research related to that region, uh, that is certainly a concern. If you are, uh, doing R related stuff, probably not so much. However, Uh, we still have to, uh, be aware that we are in a phase of a trade war and I'm emphasizing the word trade and tomorrow when you're, when we actually convert DeepSeek or ChatGPT into our own personal shopping agent, the same way as there is bias there, you might end up, uh, getting BYD and not, uh, necessarily Ford.
If you, if You let, if you let ChatGPT, or if you let DeepSeek make a purchase choice for you.
Marc Beckman: That's really interesting. So you're saying that the Chinese [00:40:00] generated AI tool DeepSeek may, in the future, use your personal AI to shop on Chinese commerce platforms versus American commerce platforms.
Milos Bujisic: Absolutely. Absolutely. And have a bias in that way. It could be literally hard coded, uh, that it's, uh, being suggested, uh, that it goes into, like, Teemu instead of shopping on Amazon.
Marc Beckman: That's interesting. That's so interesting. The other thing that I found really interesting with regards to your analysis of DeepSeek is, um, um, It starts with their assertion that for only six million dollars, they were able to upgrade this new model, this new reasoning model. I know that you, uh, find that you're skeptical about that, but your analysis is really compelling as you move on.
Can you share your insight with regards to that situation?
Milos Bujisic: Uh, so I, I think everything has to do with the timing of, uh, the events and, uh, how everything unfolded, uh, I can't remember that was, [00:41:00] uh, late January, the date was probably 26th or 27th, uh, when the world found out about this deep seek. And if you look at the timing of what was happening in US at the time and what was happening in China, there are certainly, uh, some coincidences that might be coincidence, but, uh, could explain, uh, uh, why the release sounded the way it sounded and why specifically, uh, Such a low amount was publicly shared 5.
6 million. I believe was what they said regarding training of a model. I'm deeply skeptical about that number and there are two levels of how this number could be somewhat fake. I mean, it could be an outright fabrication and I do not know that like I'll be perfectly straightforward. The second one which I'm more inclined to personally believe in is that it is just part of the it.
cost that was involved in the development. For example, this does almost certainly does not include the, uh, hardware cost for all of those [00:42:00] H100 NVIDIA graphic cards that were apparently used for the training or later on, apparently Huawei cards are used now for. DeepSeek's operations, they're not using NVIDIA's GPUs.
They're using locally produced GPUs, which apparently operate at 60 percent efficiency of NVIDIA's cards.
Marc Beckman: But good enough.
Milos Bujisic: Yeah, that's more than good enough. And that part, I'm actually, we can touch on that in a few minutes of whether that could be true or not. Uh, but, uh, specifically when you say that you trained a model for six million, you're creating a shock in the, in the industry because we all know that all of the other models spent tens of millions and sometimes even hundreds.
And if you would look into the years of development, uh, uh, And OpenAI spent billions of dollars into development.
Marc Beckman: public, into billions to train open AI.
Milos Bujisic: it took billions of dollars. Now, suddenly, the biggest shock was not [00:43:00] about necessarily about the quality of deep seek being much better than, uh, ChatGPT. We can see that immediately a week later, ChatGPT released their own reasoning models to show, Oh, we are actually developing the same thing.
We already have it. So I don't think the. But that was the key component. The key component is this is actually super cheap. This does not require billions of dollars and it does not require hundreds of thousands of NVIDIA latest graphic cards in order to work. What happened is Nvidia lost now, I think from that day to today is probably close to 20 percent of its market value, but there was like a big drop immediately.
And that's not the only company. There is a series of chip producers and then it basically moved into the rest of the tech sectors across U. S. And as we saw past two weeks, like stock exchanges, not in the best shape. And I think the critical [00:44:00] component is the developers of DeepSeq were not an AI company, they were a hedge fund, which by definition, uh, plays on the stock exchange.
If you've spent objectively, let's say 2 billion, but, uh, you were, uh, uh, behind some of the funds that made from, uh, my research 6. 6 billion. only on shorting NVIDIA stock plus additional several billions on, uh, on few other stocks that were also shorted around the same time. You've just made a return on your investment of let's say 2 billion for several billion return just on, uh, from shorting the stock.
I'm not saying this is true, but it certainly would be a feasible scenario and it would explain why such a shocking statement was released also around the time when NVIDIA was supposed to release their newest line of graphic cards which failed, uh, epically, uh, for their delivery [00:45:00] because they produced under three, four thousand cards for the entire, like of their highest model for the entire US for the launch date.
So it was a double failure for NVIDIA.
Marc Beckman: NVIDIA failed to deliver, uh, to satisfy the purchase orders?
Milos Bujisic: Yeah, not necessarily to satisfy purchase orders, uh, the launch was supposed to be on January 30th at 9 a. m. Eastern Time and nobody could actually purchase cards because apparently retailers like, uh, Micro Center, which is one of the largest physical store retailer had under 300 cards in total for their 1590 like their, uh, The most advanced, the latest graphic card, uh, and according to some estimates, there might've been under 3000 for the entire country.
And the demand was probably tens of thousands on a launch day. So they just did not have enough capacity because they do not have production capacity to create enough to satisfy. And they've been struggling with that for a while, but [00:46:00] this was probably the most shocking event, most shocking, uh, new generation of graphic cards launch events for the last few years.
Marc Beckman: So Milo, just going back a little bit because it's really super interesting. You're literally the only person that I've ever heard speak about this theory. You're speculating that perhaps the founder of DeepSeek, whose background, he started, um, HighFlyer hedge fund, which is backed by China, um, about, I think it was in 2016, so almost 10 years ago, um, and allocated most of those funds, as I understand, my research showed me, towards the entire AI ecosystem.
For simplicity purposes, we could talk about energy, hardware, and the software, the algorithm itself. You're saying that they basically played the market. They shorted the market. They shorted NVIDIA. They knew if they came out with this crazy statement, um, specifically at that moment in time that they've only invested, they were able to get a product as good as open AI, as good as [00:47:00] Sam Altman created for chat GPT for just 5.
6 million. Um, that all of these stocks would collapse and they shorted. This entity shorted NVIDIA, thus resulting in billions of dollars of new revenue to the hedge
Milos Bujisic: So, uh, that's,
Marc Beckman: Am I catching that right? Am I getting that right?
Milos Bujisic: that, that, that's certainly like, uh, a potential explanation of, uh, what happened again, uh, I, I cannot claim that they have any material proof for it, uh, but it, it could be a possibility that it worked, especially when you also think about the timing that this happened just a few days after the project Stargate was announced, which if anything was supposed to, uh, increase NVIDIA's price dramatically, because out of 500 billion, NVIDIA would actually, uh, be probably the company that would Take the biggest share of those 500 billion because they're the ones that produce the physical.
Marc Beckman: which really wasn't a lot of money. So just to back up for the audience, for those who don't know, um, Stargate was announced in Washington, D. C. in the White House. Um, [00:48:00] where, uh, Sam Altman from OpenAI, Maan from SoftBank and Ellison came in and announced a $500 billion, investment in ai, but over a four year period, right?
Milosh, so that's really not that much money. We know you and I are very well aware that the Chinese government has invested trillions of dollars on an alley, like regular basis
Milos Bujisic: yeah, especially into
into building up AI.
Yeah. And, and, and they have the strategy of, uh, developing not only AI, but, uh, completely, uh, rearranging their industry until 2030, 2035. So I'll just read one piece of information. Um, uh, traders betting against, uh, AI's darling NVIDIA earned profit of about, uh, 6.
Milos Bujisic: 6 billion. The biggest single day move ever on the stock, according to data analytics. Analytics firm Ortex. So this was a single biggest move. And on top of that, uh, short sellers [00:49:00] also, uh, uh, shorted the Broadcom and made 2 billion in profits on shorting that one. So, and there are few other, uh, examples of companies like Supermicro, uh, was another one, Vistra and so on.
So there was a series of companies that lost from it and that were shorted according to these, uh, reports by the largest amount.
Marc Beckman: you're talking about approaching perhaps 10 billion in revenue as a result of shorting stocks within that chip sector.
Milos Bujisic: Yeah. Yep.
Marc Beckman: that's, that's pretty wild. And you think it might've been deliberate. How would anybody find out?
Milos Bujisic: well, I mean, uh, the best way to find out is actually to follow the companies and the investors of companies that were shorting the stocks. And if you could actually, uh, Dig deep enough to realize if there was any money that was indirectly or directly connected with the owners of DeepSeek or for that matter, not [00:50:00] necessarily with them.
It could have been another entity within China that could have done that. So that's probably the best way to approach this. But even then the results might be somewhat inconclusive. Because we might not figure out who is really, uh, behind all of those funds.
Marc Beckman: So, Milos, just to shift gears for a second, I know that you're a tech optimist, you've been, you know, rooted with deep, uh, love for technology forever. Where do you think we'll see the most, um, advances in artificial intelligence? Which business sectors over the next, like, let's say 12 to 18 months?
Milos Bujisic: So I think that the biggest change that we're going to see, uh, uh, should come from it. What the first part that we've talked about is, uh, the whole disruption of search. Uh, and this is something that could happen in the next 12 months that ultimately people are going to embrace the search using, uh, uh, generative AI and slowly move away and start eating away, [00:51:00] uh, Google's, uh, ad revenue.
The second one, I think, are consulting services. So we know that companies like McKinsey and Boston Consulting Group, and even marketing agencies like WPP, are developing their own variations of, uh, uh, generative AI that is trained with their own proprietary data. One of the problems with current versions of generative AI, because it was trained on Pretty much terrible, uh, writing and terrible data and terrible information.
When I say terrible, the publicly available one. I always joke with my students, uh, that when you ask ChatGPT to write a student assignment, it actually sounds exactly like students because it was learning from students assignments. That was very readily available data for them to train with. However, OpenAI did not have access to proprietary, uh, reports, uh, made by, uh, McKinsey.
They couldn't train their models. So it cannot provide a similar, it cannot create a similar type of report because it basically [00:52:00] barely ever saw, uh, reports like that. Now McKinsey can actually utilize all of their things to train, uh, all of their existing data to train the models and start creating novel, uh, types of reports.
And this is where Potentially, as the models are becoming smarter and smarter, could not only replace the people, but actually create, uh, or original and creative and high quality content that surpasses the human ability.
Marc Beckman: That's really interesting. So I've been thinking, I've been a very, um, bullish in fact about specialized AI, right, by sector. If you are in law, you create a legal AI like Harvey and you stand it up. If you are in sports, you create a sports AI. What you're talking about is even more specialized. You're saying, These companies in the consulting space, and frankly, probably a great entrepreneur will step in and push out against McKinsey.
McKinsey doesn't always have the greatest reputation out there as it relates to brands that hire a [00:53:00] McKinsey, um, they could step in and disrupt and create specialized reports on the consulting side of businesses, whether it's marketing or supply chain or whatever it might be. That's really interesting.
And then you can Deep D staff essentially, you don't have to carry such a massive load of staff. You can save probably a ton of money with regards to in person meetings and travel and conference rooms and beyond. That's a real interesting efficiency that you're talking
Milos Bujisic: Uh, but it's not necessarily that you're only de staffing, you could actually change, uh, the model where ultimately, uh, instead of, uh, the work having a typical working relationship with a consultant where you're going to have, uh, uh, your account manager that you're going to work with, and then they are organizing a team in the back that is doing work for you, this could actually become a McKinsey's, uh, platform where a company subscribes to that platform and, uh, McKinsey creates a custom version of, McKinsey, it could be any of the big consulting companies, a custom version of their own software.
[00:54:00] Currently, uh, some of these agents are still reluctant to do that because it's such a big shift in the mentality of how they do business.
Marc Beckman: They're afraid.
Milos Bujisic: Yeah, I had some conversations within WPP. I mean, it's a different type of business, but still, uh, they don't necessarily want to monetize their stuff. They still want to use it internally for their own purposes to make their own existing products better instead of giving their own tool, uh, for subscribers to utilize, which could be instead of you having, uh, uh, Your own account manager and the copywriters and so on that are sitting in one of WPP's agencies.
It could be that you just access their tool and it's going to do the same work for you and you're paying a subscription.
Marc Beckman: So for the viewers who don't know, WPP is one of the largest holding companies in the advertising space, and it's really interesting, Milos, because what you're doing is looking at disruption in advertising. You have generative AI, which can totally disrupt as it relates to filmmaking, [00:55:00] content creation, visual stimulation, and You Copywriting, script writing, et cetera.
You have the data piece of it, which can, you know, really, really, um, be broken down, but what we haven't, what people really haven't analyzed yet is the media buying. So these massive advertising agencies really make a. Ton of their profit, their cleanest profit margin on the media buying side. And I would imagine that artificial intelligence will be superior to humans as it relates to identifying and securing the best place to stand up brands, messaging, consumer facing at the best prices,
Milos Bujisic: Yes, but ultimately in the long run that is going to become a zero sum game because As media buyers become more, or AI behind media buying becomes more sophisticated, the media selling is also going to be operated by AI. And you're going to have two extremely smart agents finding an equilibrium, and I think that equilibrium [00:56:00] is eventually going to Uh, level itself out
Marc Beckman: interesting.
Milos Bujisic: that it's not going to be a situation where one would be able to dominate the other.
There was probably a situation where you're going to have a type of let's, let's call it the cold war of media buying and selling where both sides, sides reach parity. And I think this is what we might see at the first, the media buying might become more optimized and start generating more revenue as the big agencies.
will have more resources. But ultimately, as the media companies also become sophisticated, it might cancel each other out.
Marc Beckman: Miloš, just for fun, uh, we talk a lot about, you know, when we look, when we look at hardware, we're localizing against chips and super chips and all, but just for fun, um, a lot of, of people have purchased Nest, which is creating amazing efficiencies, cost saving efficiencies because of artificial intelligence.
In your opinion, where do you think we'll see [00:57:00] hardware, new hardware? That consumers could purchase and put, perhaps, in their home, uh, to create new value, uh, life experiences.
Milos Bujisic: Well, for a while, uh, there have been conversations again, another component of web 3. 0 would be IOT. And, uh, I think the big component of, uh, Having your own personalized AI agent is going to be how much information that the agent knows about you and how much it's learning from you. So I think that a lot of stuff is going to be related to different house tech, but also wearable technology that is its purpose is not only to directly help you, but actually indirectly to help you through AI.
Uh, that it would be constantly connected. So, for example, you're going to have a water bottle that knows exactly how much water you're consuming in a single day if you're taking a water bottle to work. But this is going to be fed into your AI, and based on that, the AI is going to have an understanding of your hydration and might recommend [00:58:00] certain medical treatments or relaxation treatments.
Or purchasing decisions of what you should do. But this is where I think the connection between AI and IoT is critical is that ultimately all of the information from IoT needs to go into your personal AI agent. And that way it will make the best decisions for you.
Marc Beckman: So when you talk about the convergence earlier on, you were talking about the convergence of humans and technology. Um, and then you talk about IOT and AI. It's interesting because we know that Elon Musk, as well as other companies are preparing To put a chip into your brain, Neuralink, I believe it's called.
And I am not afraid of it personally. We already have devices and chips in our bodies. People put them into, you know, their heads for sight and for brain and into their hearts and beyond. So I guess my question to you is that, is that going to be like when things are on full throttle, the ultimate [00:59:00] level of AI and humanity coming together where you have a chip inserted by Elon That can figure out all of the different conditions of your body and your activity that then links back to a centralized personal AI, perhaps in a GPU, and then ties back to your home with an Internet of Things connection.
Milos Bujisic: Yes, I mean, uh, I can give a, for example, a perfect scenario that you are returning from work. Uh, your chip is reading your body temperature, heart rate and, uh, other biological, uh, functions and sending that information to AI. Uh, I mean, that is probably in some cloud, which is going to adjust the home temperature, lighting and prepare the music that would get you in the best mood based on the stuff that, uh, how your body is currently feeling.
Uh, and prepare the program of activities for you, depending on your physiological function. So this is a [01:00:00] technology that is honestly, uh, not around the corner. We already have most of the resources, not necessarily through direct connection to our brain. But if you're wearing a smartwatch, that's reading your temperature and your pulse and doing your EKG, you pretty much can predict some of these things.
And then you have smart home features. That could be adjusted. So this technology already exists. In a sense, it just needs more connections. And the question there is how much it's going to be centralized within one company, or are we going to develop standards that are going to allow those different pieces of technology to communicate seamlessly.
Marc Beckman: It's very cool. So does that scare you? Or are you excited about it?
Milos Bujisic: I personally am excited about it, uh, and I think that a lot of people are going to be scared, but we always forget that, uh, people are willing to sacrifice a lot of their concerns, uh, for convenience and ease of use. Uh, again, a joke that I do in my class, I ask students, are [01:01:00] they, how much they care about privacy or personalization as a binary outcomes. And for every student that says that they care more about privacy and less about personalization, I ask them if they own a smartphone.
Marc Beckman: There you
Milos Bujisic: then I tell them they're just being dishonest to themselves.
Marc Beckman: Exactly. Exactly. Milos, you have been so generous with your intelligence and your time today. Every guest that comes on my show ends the show, um, in the same way. Basically what I do is I integrate the name of the show, Some Future Day, into a sentence that I begin and I allow for my guests to finish it.
Are you up for this little game?
Milos Bujisic: Sure. No
Marc Beckman: Okay. So in some future day, artificial intelligence will change the way that people live
Milos Bujisic: Uh, so in some future day, uh, artificial intelligence, uh, will change the way that, uh, people live, that people shop, that people make, uh, decisions, [01:02:00] uh, the, the way that people educate, uh, themselves and ultimately, uh, the way that they reach their own, uh, personal and social happiness.
Marc Beckman: Milos, thank you so much for joining me today. It's really been, you know, great conversation as always is truly insightful and a pleasure. I really appreciate your time today.
Milos Bujisic: No problem. Thank you so much, Marc.