The Future of AI: Insights from Farhan Naqvi (Avalon, TenViz) at IVYFON 2024

Sayyed Farhan

Join a forward-looking panel discussion from IVYFON's Future of AI Conference, held in San Francisco, September 18–19, 2024, where leading voices explore how artificial intelligence is transforming industries, investment, and corporate strategy.  🎙️ Panelists: • Farhan Naqvi, Former CFO & Head of Corporate Development – iLearningEngines • Andreas Deutschmann, Managing Director – Avalon Securities • Konstantin Fominykh, CFA, CEO & CIO – TenViz  🎤 Moderator: Tess Hau, Founder – Tess Ventures  This high-level discussion covered AI's disruptive potential, investment frameworks, enterprise adoption, and how business leaders are positioning themselves in an AI-driven future.  🔹 Event: IVYFON – Future of AI Panel 📍 Location: San Francisco, USA 📅 Date: September 18–19, 2024  Note: Farhan Naqvi served as CFO at iLearningEngines at the time of this panel.  #SayyedFarhanNaqvi #SayyedFarhanNaqviiLearningEngines #FarhanNaqvi #FutureOfAI #IVYFON2024 #ArtificialIntelligence #EnterpriseAI #TechLeadership #iLearningEngines #TessVentures #AIInvestment #InnovationPanel

Transcript

00:00Yeah, sure. Hi, everybody from New York City. My name is Andreas Deutschman. I work as a

00:15manager director for Avalon Securities, a female-owned boutique investment bank here

00:20in New York City. In my past, I've led teams in due diligence, valuation, portfolio management,

00:26risk management, financial modeling with companies like JP Morgan, where I was a managing director,

00:31John Hancock, where I was a vice president, where I was a manager, and I carried out my duties in

00:38New York, Chicago, Boston, Luxembourg, Zurich, Vienna, Singapore, and Tokyo. And recently,

00:46I taught the venture capital course at the Central European University in Budapest, which was

00:51the end.

00:57All right.

00:59Farhan, please.

01:01Hi, everyone. I'm Farhan Nakhwe. I'm the CFO of iLearning Engines. We are a leading learning

01:07and workflow automation firm. We've been in business for almost 14 years now, and we went

01:14public through the DSPAC route earlier this year. I have been with the firm for almost

01:21five years, and prior to that, I was doing technology investment banking on the West Coast

01:25for almost 10 years.

01:29Great. So, Konstantin is online, but he is not available right now. So, and Darwin should

01:37be logging in any minute. So why don't we get started, and then we'll move over as the other

01:43people join. So, can you talk about the current trend in AI in terms of the general, you know,

01:49marketplace, and do you think things are getting a little bit frothy, or what are your thoughts

01:55in terms of opportunity?

01:58Konstantin, can you introduce yourself first?

02:01Yeah. My name is Konstantin. I run 10Vs. We build and offer the clients in the investment

02:08management community predictive tools, which use AI to forecast capital markets, raise bonds,

02:13commodities, equities, and so on.

02:15Right. Yeah. So talk about the general marketplace, and then talk about where you guys are in the

02:20market. And I think everybody uses ChatGPT as a point of reference. Can you talk about how

02:26you're different than ChatGPT, et cetera? Or do you use it?

02:30No, Andreas, why don't you start, and then we'll go over to Konstantin and then Farhan.

02:38So you want me to start?

02:39Yeah.

02:41Yes, we do use ChatGPT. We think it gives us some interesting insights. For example, we're

02:50looking at innovative investing, and we're looking at sectors like AI, cyber, block, gaming,

02:56crypto, web3, cloud, VR, metaverse, 3D, robots, and drones. We come up with a number of parameters

03:03to invest them by, and then we have it sorted out. And so, of course, there's always a risk of

03:10hallucinating and whatnot. So we overlay some of our financial experience, and we find that

03:17on a relative base weighting, AI rates, AI is the number one investment category, followed by cyber,

03:25block, gaming, crypto, web3, cloud, VR, metaverse, 3D, robots, and drones. We have plenty to talk

03:33about individually, but this is kind of a first pass on something we're working on for our investor.

03:40Great. And maybe later we'll talk about whether AI actually increases the value of some of these

03:45older, vintage companies. So, Konstantin?

03:50We actually build our own AI. We call it Zerzer. Just to build it with context. Generative AI,

03:59ILLM models, it's a very, very, very small subset of what AI could potentially do. It's really,

04:06really very tiny, like maybe it's very popularized right now, but that's a small fraction of what AI

04:14can do. Essentially, what generative AI does essentially looks mostly with an existing body

04:20of knowledge to extract either existing relationships or pretty close relationships,

04:25which are still not obvious. But there is no way in how generative AI will tell you something about the

04:31future. So, we've built predictive AI, completely different approach, which we use for guest investing.

04:40Okay. Farah?

04:42Yep. I think Konstantin nailed it. There is so much more to AI than just Gen A. And it's not as if AI just

04:52sprung on the scene with the chat GPD being put out for public. AI has been in the works for decades,

05:00if not longer. And so, we as a company, we've been doing, we were founded in 2010.

05:10The founder of our company has been doing AI much before than that. We've been doing AI,

05:15building AI software since 2010. And where we have landed up is, we are actually one of the few

05:25businesses who have been able to build enterprise use case using the underlying AI technology.

05:32There's a lot of talk about large language models being out there. What we have found to be useful is

05:39deploying small language models, if you can call them that, in different industries.

05:44So, just to take an example, we have 40 plus proprietary models that we built up on our own,

05:51that we deploy depending on what the use case is. So, going back to what Konstantin said,

05:58LLMs and Gen AI are just a very small part of the whole AI revolution. There's so many things that are

06:06being improved and being made better by applying the broader AI tech.

06:16So, outside of the area where you're in and outside of large language models,

06:21where do you think are the best applications of AI today?

06:24For me, I think if you look at the large players, OpenAI, Chetch, EDP, Gemini, Mistral,

06:38like they've kind of used the same varieties of AI across very large, broad data sets that may not

06:47exactly be differentiated. So, it's kind of like you got a situation where you've got some people working on

06:53ChatGDP or Copilot or whatnot and it's like you've given them all the same advantage. It's like taking the NBA

07:02Basketball League and giving every player three extra inches of height and then wondering if there'll be

07:09different outcomes in the games and the championships and whatnot. But I think what the gentlemen Farhan and

07:15Konstantin came up with were talking about some differentiated

07:19horsepower on the algorithm side and some differentiated data is exactly where we need to go.

07:32Konstantin, any thoughts?

07:35Outside of what you're doing, what's interesting parts of AI?

07:38Yeah, kind of building up on what Andreas has mentioned, and I think Farhan was actually heading

07:48that way too. It all comes to the use cases, right? As Farhan was saying, essentially, we were using AI

07:53all of us without even knowing it for almost decades, right? Basically, don't forget planes are 99%

07:59flown by computers. So, it's kind of like a rhetorical question, are you using your AI? You were doing that

08:05without even knowing it. So, I think it comes to understanding what the use cases are and what we're

08:12after. I will only speak for the financial services, but before I go there, generative and LLM-like

08:20AI probably should be very good in the law profession, right? Actually, to the best of my knowledge,

08:24when I talk to people in the related fields, I don't see the people innovating a lot using true

08:29generative AI in the field of law, even though it's almost perfectly suited for them.

08:33So, but let me speak about the field of finance and investing, specifically investing. That's what

08:39we do, right? That's our expertise. That's what we do pretty well. And the essence is we forecast markets,

08:46like stocks and bonds and commodities within here in the uncertainty. And the only goal in the other

08:54exercise is to perform the humans. So, nobody knows the future for sure, but what we found a way to,

09:01that machines can extract some interesting relationships within a time series. And based

09:07on that, we see these trends either breaking or evolving further. That's an idea. Understand what

09:14you're after and sort of like start building it backward. What is the use case? What exactly are we

09:19trying to do instead of like, oh, this is a magnificent opportunity, a magnificent tool. And like,

09:24um, um, um, so that's, that's our approach.

09:28Fantastic. Great. And Farhan, can you talk some more about what you, what you're doing? I know that

09:36you talked about a little bit, but can you talk about other areas outside of what you're doing right now?

09:40Well, uh, so in terms of, uh, where we are deploying our tech, uh, we have 12 verticals that, that span from

09:51oil and gas to insurance, to healthcare, to education. Uh, and depending on what the market dynamics are,

09:58different verticals use AI tech, uh, and our model very different. So that's why, uh, at a functional level,

10:08we are able to cover, uh, from learning automation to workflow automation, depending on what the use

10:13case is. I'll, I'll borrow from one of the points that Andreas mentioned, uh, which is about data. So

10:20a lot of, most people, uh, probably are, are unable to, uh, to appreciate that the value that data has

10:30in this whole AI ecosystem, right? AI systems, uh, get successful or they break down, not on the basis

10:41of the architecture that, that is used to build them up. They, they perform well, or they do not

10:46perform well, depending on, depending on the data that is being used to train them. So securing that data

10:54and, uh, sanitize, sanitizing that data is as important as anything else. If you're building,

11:00uh, an AI system and making and trying to make it work. So, uh, different people have tried to address

11:08that issue very differently. We've, uh, we've found a unique solution which works for us and which is,

11:15uh, procuring data from the source. Like, so if, if we are deploying a model in the oil and gas space,

11:22for example, we would want it to be trained on data produced by the oil and gas industry. So

11:30procure, uh, procuring that kind of data and making sure that we have that data pipeline set up,

11:36uh, I think is very, very critical to the success of our operation and any AI operation in general.

11:45So can you guys talk a little bit about this whole data aspect? Because it seems like in lots of areas,

11:51you know, uh, there's problems with data where obviously it's, um, it's human error and data.

11:57And also there is, uh, you know, limited access to all the data you're trying to,

12:02you know, so, so like, for example, if you have a insurance product, or if you have an insurance

12:07application, you have to get the, you either have to get insurers to go along with you so you can train

12:12the data for their application or, you know, you need to get source the information one way or the

12:19other. So, so, so how does, how, how do you take that step? Constantine, same with you. I mean,

12:24is most of the data that you work with, is that public knowledge, publicly available, or do you work

12:28with more specialized data? Maybe you guys can all talk about that. Constantine? Yeah, sure. Um,

12:37so yeah, we, we mostly use our public data either available for free or we, but we actually purchase

12:45a lot of data for, but essentially it's data available to the market, right? It just like,

12:49we go through the enormous, enormous number of times series every day. Like, so come, and that's

12:55actually tells us actually one of the bottlenecks. There is no, I think for Han was kind of leading to

13:00that. If you take as much everything in the world and try to, uh, try to boil the ocean,

13:07you would not get much, right? You would not get much. You need to go through the specific elements

13:11of data that, that you have a good fundamental, uh, or expertise, like you need expertise that should

13:17likely say, yes, there's gotta be some nuggets of gold in this ore before you start going too deeply.

13:24So that's what we do. Not only we look for the, we, we take millions of, like maybe half a

13:29million time series every 15 minutes or so. Uh, so it's, it's not enormous, but it's big enough

13:35given complexities of possible interactions. But the idea is actually, we know for sure

13:40these relationships are not fully understood and, uh, interpreted. What, what's it, what's the

13:47application? Is it used for trading or for investment decisions or?

13:53Marty, this is a great question. Um, and I appreciate you saying that, yes, it's used for investment

13:57decisions. And that's actually a very important qualifier because 99.999% of the AI used in the

14:05investment community right now is used mostly for trading and some sort of back-office automation.

14:12Very, very, very little of that is used for midterm or long-term investing. Um, even the biggest and the

14:19most established sharps, I would be talking and work with most of them, or at least talk to them. They

14:24actually still, again, mostly use it to automate existing processes like, or some sort of back-office

14:29operations. So in order when it comes to say, making decisions like, hey, you should be buying

14:35US treasuries or selling technology stocks or buying Brazilian stocks is still mostly done by humans very

14:42subjectively. So you're saying that the existing application is to take, uh, traditional or internal

14:56processes and automate them rather than to think about new ways to do the business.

15:01Correct. And that speaks to the core of most AI models, including, uh, uh, generative AI, which are

15:07essentially the technical term is classifier models. So they look at the variety of, uh,

15:13possible outcomes and they smartly classify them where the way you've classified classification is

15:18pretty established, right? You see, like it started many, many years ago. Like again,

15:23Farhan has referred to that with their fraud detection and so on, right? Like, and you can actually

15:27save as a high degree of certainty that was a portfolio in transaction. Yeah.

15:32Farhan, can you talk about the, you know, how do you get data? How do you acquire it? And

15:48is this something that requires, you know, you, your, uh, models to learn from internal corporate

15:53data or do you have other data sets that you use that you are able to acquire?

15:58So we buy data in a lot of instances, we buy it from the source, the people generating that data.

16:05So taking a hypothetical example, if we, if you're deploying a model that has been, uh, one of the 40,

16:1240 plus models that we've built, uh, if you're deploying one of those stay in the oil and gas space,

16:19right before it actually gets trained and deployed on the client site,

16:24that our model gets trained on industry specific data. And the industry specific data is the data

16:31that, that we have purchased from oil and gas industry, uh, players, which, which is then used

16:38to train the AI model before it gets deployed.

16:41And then do you track how the models change once they are deployed into a corporation?

16:53So, well, if, if you look at AI models, it's the processes, you take data, you train the, the model that you have,

17:02it produces an output. Now, if you put this into a recursive loop, wherein the output, uh, is being used

17:10by, by the people to generate more data, then, uh, it's, it's a self-reinforcing process.

17:19You're generating very relevant, updated data that can then be used to train the model again.

17:24So the output gets better and better.

17:28Right. Do most clients have that purpose?

17:31So if you think about it from the client side, uh, at the end of it, it's the client's data that is

17:38being used to, uh, to train the model that eventually produces the, the, the output or the product that

17:47the client themselves is using. Right. So it gets very, very customized to the client's use case.

17:55It's like using the client's institutional knowledge that they've built over years

17:59in putting that into an AI model and then using the, the output from that AI model for that particular

18:05client. That's, that's the, that's the way we are able to customize and automate the workflows

18:14within a specific client to their, to their specific needs.

18:23And Constance, do you use the same process?

18:26Roughly the same, right? Or roughly the same. So obviously feedback is always, uh,

18:30feedback, uh, incorporated and we do it in usually two layers.

18:33Uh, the, uh, the, uh, the element of sort of like self-reinforced learning, but that, that,

18:37that, that, that is kind of a hinting to your, we do that, but also like periodically we do model

18:42upgrades or, uh, uh, complete revisions. What else could be done? How are new approaches? Like,

18:48that's when we, uh, go to the new generation that wants in a, uh, two or three years.

18:56Andreas?

18:58Uh, yes, just, just on the data, I had been pitched, uh, by someone who had

19:03only AI experience about, uh, starting an AI hedge fund and, uh, didn't think much of

19:10unstructured data as being a component to it. Uh, just a little history. I used to have a client,

19:17uh, in the nineties called Renaissance technologies. And back then they were using neural nets,

19:22uh, and millions and millions upon data in stocks, bonds, equities, commodities, and currencies.

19:31Uh, and they had hired over a hundred PhDs, either computer science, physics, or math,

19:37and had done so well that they returned all the billions that they, they, uh, invested on behalf of

19:45clients because they're making too much money and they just kept it on their own. So, um,

19:52one of the interesting things that he did, not only are they coming up with models that would

19:56predict when these assets or asset classes would go up or down so they could buy or sell them.

20:01They also predict a second and tertiary depict when a model wasn't working, what was the next

20:09evolution of model that they should be using in that particular environment. So, uh,

20:18talking about that's data they collected mainly from public sources and equities

20:22and wasn't subject to these types of scrubbing and other issues that banks have in other areas.

20:30Many of them have, uh, merged over the years. JP Morgan is now one of 70 companies. Every time they

20:35emerge, they're putting a different, uh, amount of data and it's not matching one to one.

20:43And now that they've managed to melt it together, they can't really understand what they really had

20:47to begin with. Luckily, some of this stuff is like social security number, phone number, whatnot, but also

20:54credit ratings, uh, and the like. But anyway, um, data is key. The cleaner off you are.

21:05Yeah. Yeah. Can I add something to what Andreas was talking a little bit? If you don't know, if you

21:10don't mind. So, uh, Andreas mentioned the example of Renaissance, uh, technology. And it's a good

21:16example, again, a very short-term trading. I know for sure that they were mostly trading within a day

21:20or maybe two or three days max. They will never, so that's trading. It's pure tracking inefficiencies.

21:27They were not really invested. I know that for sure. And that's fine. Again, so again, like it still

21:33sits within, um, my initial assessment that are, it's less than, less than one-tenth of a percent

21:39of all investing is done by computers, by a truly invested, right? Truly invested, right? Um, by AI,

21:46at least. So again, in Renaissance, that's why they, they gave all the money back because you cannot

21:51scale it. Those inefficiencies are very compartmentalized. You cannot expand it. Those are

21:55small-term inefficiencies, which they became, they, these, these were brilliant guys and they are,

22:00but they just, it's not like you're investing in the video for the next five years. No,

22:05these were like very short-term trades.

22:10So can, can you please talk about the market opportunities that you represent or that you're

22:13looking at in terms of, uh, you know, the AI application? I think Andreas, you've got more of a

22:18portfolio point of view, but, uh, uh, sure. Right now I'm involved with, uh, raising capital for a

22:25robotics company, uh, called Mind Children out in Bashan Island in Washington state. Uh, they've come

22:35up or in the development process, I should say rather with a short-term robot about three feet tall,

22:42whose job is to, uh, measure the vital signs of handicapped children and alert the nurses when

22:49there's any issues. And also they come with a, uh, a screen on their chest, which they use to deliver,

22:57uh, educational content to these by and large bedridden, uh, children. That's one. Two is, uh,

23:03rejuve bio. That's a longevity AI is basically using its research capabilities and differentiated data

23:12to come up with supplements that have proven to reduce things like Alzheimer's, uh, and other, uh,

23:19long age maladies. And third, uh, another, another raise with earlier today, uh, AGI as a service, uh, on a

23:29decentralized basis. So not using the mass data out there that co-pilot Gemini and the others use,

23:36but basically only the data sets for its client base. And that's focused on, uh, financial services,

23:46uh, medicine and logistics.

23:51Yeah. When you look at these deals, do you, do you wonder about, you know, adoption pace or speed?

23:58I mean, that's always a question, right? Uh, well, the two first ones are, are smaller. Let's call

24:04that five to 10. Um, I'm not that worried about that. The other raises, uh, more in the lines of

24:15250 to 500. And then that's more, uh, concerning to size wise. However, we do have clients, uh, taking and

24:25testing, uh, their AI or AGI rather, uh, which might end up having those clients providing the capital

24:35for the raise. So we'll see, but in general, I mean, last year they, I read somewhere, I don't know,

24:42it was pitch book 37 billion raised in AI. A third of that was the 11 billion that, uh, Microsoft

24:49and Altman, uh, it's not exactly a glowing marketplace, but there are some signs of life.

24:58I think today Microsoft and BlackRock announced that they were raising a $1 billion fund to invest in

25:05AI, uh, earlier this year. And, uh, I mean, Maureen kind of talked about this earlier,

25:11just who's investing in what, but nobody's really broken out the AI component in BC from everything

25:17else. Right. Cause I, I think that tells a story. It's, it's, it's actually similar to the, the story

25:23of, uh, the big tech companies versus all the small fish, right. And every other company that's

25:28out there where they have this amazing, you know, tenfold performance versus, um, others. So,

25:34so I think, uh, you know, we needed, uh, drill deeper into, you know, the AI funding and, you know,

25:43what's the basis of stuff. So Constantine, can you address the same issue we were talking about,

25:48which was that, uh, you know, what's the application and, and do you, do you worry about, you know,

25:53speed of adoption? I mean, I, you've been doing this for a couple of years, I think. So,

25:57and you have a couple of clients on wall street, uh, or a few clients on wall street,

26:01I'm not sure how many, um, you know, what, you know, obviously it's probably better that you

26:05have been positioned, um, seven years, seven or eight years.

26:18Yeah. So talk about your, you know, your positioning and, you know, speed of, of adoption and, and,

26:25you know, how, what's your model and how you intend to, you know, grow this business.

26:30So essentially we have two businesses. First one is actually

26:34access to our predictive tools for people to use our forecasts live, right? For example,

26:40if you're a commodities trader, we'll give you a forecast for relevant commodities.

26:44If you're a buyer trader, we'll give you a forecast for relevant, uh, bonds and so on.

26:48And we think opportunity is huge because essentially to build something like that in house, you need

26:54to spend to hire at least several months and work very hard with them for several years. Again,

27:00it started approximately 2015. We have dozens of clients in New York and Canada and in Asia and so

27:08on. And it's very, very hard to build something like that in house, in house, very hard, very hard.

27:15In fact, we know that even the most reputable investment houses, they don't really have much

27:20like that in place. It's years of investment, which might never fail. You know, we think

27:28opportunity is immense because investment industry is a big part of our, of the financial services

27:34industry and the pace of adoption remains slow. Humans are inherently conservative and they are

27:43reluctant to be reluctant to embed into what they see as a core of their process.

27:54So, so can you talk more, is the opportunity with traders to sell it or, or what, is there a

27:59possibility of an off the shelf product or there's a heavy amount of tailoring that needs to be done with

28:04what you do? We offer mostly mildly, mildly customized products. Essentially we have the order of the

28:14product like which kind of like off the shelf and we do mild customization by client's needs. So it's

28:20actually pretty flexible toolkit that you can apply to your asset class of your geography. Basically we cover

28:27everything in the world, which is investable. Everything is a pretty big statement. And that's

28:35could be usable. And actually again, the biggest insight that we bring to the table is providing

28:43linkages from other areas into what people invest in. It might strange, sounds very strange, but even

28:50trades in something that people say like, Hey, buying Nvidia or something like, but moves in Nvidia could be

28:56correlated to the price of commodities or Australian dollar or Norwegian Crona or Brazilian reais.

29:02Something that is not obvious. Yep. Constantine, are you doing it for private market investing too,

29:10where there's not so much data on price for an asset or performance of an asset?

29:17Good question. We do work with these private investors a little bit, especially on the private credit

29:23side or private loans because they actually very well could be proxied by public interest, by public

29:30bonds, including credit. That's what we do because the linkage is actually pretty strong there. But

29:35well, let's assume private equity, like we don't really do that because again, it's highly specific

29:40and we need essentially some sort of like daily trading marks to understand

29:44whether it's getting. I mean, I assume that when you're doing private, you know,

29:51there's a lot less data, risk management really matters a lot as well. Like, so you've got your

29:55historical models for doing this sort of thing. And then, you know, being able to model out the future

30:02opportunities. I mean, if you're doing equities versus lending, I mean, lending, I think it's a risk,

30:07it's, it's far more risk management, but on the equity side, you want to see it, you know, upside,

30:12et cetera.

30:14Correct. I mean, lending private loans, they actually trade, I mean, like,

30:22they could be proxied by private credit by, I'm sorry, by public credit, i.e. private bonds pretty

30:27well. So we deal with that all the time. It's actually pretty good proxy. If you, if you're,

30:33especially if you're honest about this sort of like about there's not smoothing the noise,

30:37which is inevitable in trading.

30:41You know, somebody posited at our event last week that, you know, so I said, you know what,

30:46there's not a lot of innovation on wall street, you know, they kind of bang out products. So,

30:50you know, they get the same 6% return, whatever you want every, every year. And I was asking,

30:56you know, is there an, is there an actual, you know, greenfields opportunity for an AI powerhouse

31:01to grow in the finance world? You know, and, and, and who would that be, you know, et cetera. And,

31:08and then Andy Fish said that that's already happened and that these three quant funds pretty

31:13much dominate. They won't tell anybody anything about their success. And you can, what do you

31:19think about that? Has it already happened? And we'll never know, or?

31:23It's, it's, it's a great question. I hear all the time from friends and our, and, and the clients

31:29are, if you're so good, why don't you start your own fund? Okay. We actually, we do actually many,

31:34manage money for clients and because we do apply. So we essentially, we have two businesses. We're

31:40already doing that. We have buy side and sell side. However, we are not, there are plenty of people

31:47who are trying to produce philosophically selling ideas and some signals to buy side. Of course,

31:56it's our, it's, it's, it's a way of, the world will work. However, we have not seen, we haven't seen

32:03anyone. It's what we do. We haven't seen anyone. And, and I, and I wish it was any other way, because

32:11again, we're kind of like pioneering, but it's, that's why it's actually, it's easy to grow the market

32:15when you have a big company and people are rushing into that, right? So it's, I think it's an

32:21opportunity to, and yet it's, it's inevitability because I think the times when a human was able to

32:28think through all possible combinations of political risks and economic risk and business

32:33risk and valuations and currency moves and everything in his head are like long gone. That's

32:39impossible, right? Even the best of the best long-term investors, like even Warren Buffett, are not able

32:43to produce excess returns anymore. Right, but he wasn't using AI.

32:50No, to the best of my knowledge. He has a very different, I think Warren Buffett has a very

32:54different model than most people, right? Because he's become, he's the, he's the proverbial person

32:59who's become the market, right? So he does best when there's downtrend for these cash heavy.

33:05I mean, in many cases, I mean, he did great in 2009, eight, nine, yeah, he did. 2008 and 90,

33:11he did pretty well. You know, but I, I think that, you know, I guess that some of the, I mean,

33:18you tell me where the, where the money's being made in AI, is it done on volume, speed,

33:23speed, new opportunities? What do you think in terms of, you know, trading, hedge funds, finance,

33:32et cetera?

33:35Again, right now, most, most of the AI applications are on the most of the back office. So they

33:40automate trading, simple classification, preparing for the tax, extracting say earnings releases and

33:45news. For example, there was the Fed day today, right? And I'm probably two and a seconds before it was

33:51actually even published in official public sources. So like AI has already decided what to trade or not

33:57and, uh, short term. And, um, obviously, uh, this was short-term trades, which influences,

34:04added to the noise of the market. But again, we are yet to see that somebody will be offering

34:09systematically this product for service as a certified service, right? We do that. It's not an easy

34:14exercise convincing people that they, um, convincing people to reinvent themselves, their process.

34:21Farhan, can you talk about your application and why it's not just e-learning, but it's AI?

34:28Uh, yes. So, uh, we, our applications are primarily used, being used on, on two vectors. First is, uh,

34:39learning. And the second one is, uh, workflow automation. So learning automation and workflow automation.

34:45You implement our system in a learning environment and the learning actually gets better. And we have

34:52two points for that. Uh, that's what our clients pay for on the, on the learning industry side.

34:59The other one, which is workflow automation, which is you make existing processes more efficient,

35:06much better, uh, much more accurate. So we see that happening at, at the lowest rung,

35:13as well as all kinds of complex processes can be made better, more efficient, and much more

35:19accurate, accurate by implementing AI, uh, within those or inserting AI within those processes.

35:26So you guys have talked about, um, AI in the legal profession. I have a hard time with that. And

35:36I'll tell you, I'll tell you my analogous story. So for years, radiology has been done, uh, in India,

35:44right? So if you're a radiologist in New York, you know, 90, 100% of x-rays get sent to India

35:50and doctors in India make the judgments. And then they take out what they call the 10% toughest.

35:55And that goes back to the doctor, the radiologist in New York. And they say, okay, uh, the judgment

36:02was correct. Well, and they sign off on all of them because the New York doctor is, is licensed,

36:07right? Um, so I, I think that AI will probably help reduce the work of that person in India.

36:15Uh, uh, that'll get bypassed before the doctor who's licensed in New York gets bypassed because

36:21somebody still has to have liability for the judgment, right? Right. They're, they're the tool

36:27user, right? I don't think, I don't think a tool if they made a mistake would be liability free. So

36:35I have the same issue with legal. What are your thoughts?

36:38Right. Because, because lawyers are licensed in state by state and they make mistakes. There's

36:46liability. And I agree. I think the law firms are going to benefit. They're just going to reduce

36:50their cost base. Right. And I mean, I've already, there's a lot of, there's a lot of legal research

36:54being done in India. For example, there's a lot of, there's actually a lot of, um, American trained

36:59attorneys who have moved back to India and made, started these 200, 100 person staff.

37:07Legal process outsourcing is a real thing.

37:11The process outsourcing is I'm saying is AI going to get rid of the company that was outsourced too,

37:18or is it going to reduce the staffing at law firms in the USA?

37:22I think it's going to make everybody much more efficient. So jobs wouldn't, and this is my

37:29personal opinion, jobs would evolve. Uh, they won't get cut out. Like we have seen that with, uh, with

37:37all kinds of technological revenue. So if, if a certain kind of jobs, jobs get cut, the people get trained

37:48on doing other more effective jobs or much more specialized. Right. So I will give, I'll give you

37:56another healthcare example, which is that people who, I mean, a lot of doctors, there's a huge wave

38:01of doctors retiring. And I mean, if you work for one of these Medicare shops, you're seeing 30 patients

38:07a day and your notes are crap, right? So now that there's a little bit of AI taking notes for you,

38:12you don't have more time to spend looking at the quality of your notes and quality of your care.

38:16But I mean, just, there's going to be more work for you, right?

38:19Yes. And I have another comment from the audience here, Maureen.

38:24Yeah. So, um, having doctors in my family, um, that's actually the case. Uh, my brother

38:29wants to retire and he's like 50 and he's an AP cardiologist. So he's fed up of taking notes

38:35and he's in New Zealand right now. I have another sister who has an accent and, um, uh, they do use AI,

38:42but it can't figure out the accent. So she still ends up having to do all that work. So I, I think,

38:48I think it's, it's helping, but it's still not, I think they'll just put more work down on the

38:54doctors because that's where they get the money and it'll just become, I don't necessarily know

38:58if it's going to really help the doctors at least at this stage. The second thing I would say is that

39:02businesses who are trying to use AI now, it seems like many of them have hired a lot of the engineers

39:08or the coders, but I think what's really going to, uh, be pivotal is that you're going to have to

39:14either train or orientate people that can have the business case to work with the coders or the people

39:22that are developing the software to actually really make it work and be more effective.

39:28Cause I think what people are also not taking into account is that there's going to be a J curve

39:33to investing in AI. And, you know, you could be making these investments. How do you know at the

39:39stage you're making the right investments or that you're planning for the staff to be able to evaluate,

39:44like we built it. So now is it really working? And so there's, it's IBM has been gambling on AI for

39:50years and it hasn't worked out well for them. So, yeah. And so, so I think these are, uh,

39:55and I'm not necessarily, I don't, I don't assume that just because someone loses their job to AI,

40:00that they'll be trained to be able to do something more technical. I think that's a little idealistic,

40:04you know? Oh, no, no, no, no. I would say that, you know, AI is like all tech is destructive to jobs.

40:13I would say, um, I agree. So the back office functions will go,

40:18how they get re-skilled, re-tooled. That's a different story, but police here will be able to,

40:22um, you know, save a lot of money because that will be done by AI. And it's not without the humor

40:28error, right? So this main person here sitting here signing off still is liable because that's

40:33their license on the line. Exactly. Whether it's a lawyer or a doctor. So now they have to pay attention.

40:40Definitely. Because even previously with people doing it, it's no different, right? It's still on your head.

40:45The second thing I think is planes have lots of technology. If you make a mistake, you know,

40:50Delta plane crashes, Delta still gets sued. And so does Boeing. Exactly. Yeah. Keep on going.

40:55The second piece is, um, skill sets, right? So when we talk about, to your point, we are training a lot

41:02of professionals in their core skills, like doctors and engineers and lawyers. The other key skill set

41:07that's going to come up is how to use AI. Because you can make tools, but if nobody's using it, the change

41:13management aspect of it, they don't know technology. Having the best technical people,

41:17building the best tool in the world is not going to help at all if the person using it has no idea

41:22how to use it. So understanding that part as well and training all these people to say, hey,

41:28you'll have a co-pilot helping you in this five areas that this is how you can best use this tool

41:33to make your life easy. I think it's going to be key for no matter what professional

41:38we look at. And I still think there is a barrier because of licensing. Yeah. Right. And that's that

41:44I think that's going to be significant, whether you're a broker or real estate broker or whatever,

41:49you know, if something's missed, there has to be. Yeah. Yeah. And there'll be the compliance people

41:54who get involved too, because it'll be again, back to the liability. Um, but I do think that not to be

42:00underestimated having been a business person, trying to work with a bunch of fonts and engineers

42:05and coders to get what I wanted. Um, that's, that's a skill itself to, to have that translation.

42:12And also to even after you spent all the exhausting time describing what you want,

42:17you have to also budget in a lot of time to test it. Right. So it has to be like one,

42:22you have to make sure you have the user interface, right. So that people will use it.

42:25Then you also have to make sure, because it takes usually several iterations before it gets right.

42:29I think, I think a great example for Constantine is if I was teaching an AI to be a fiduciary,

42:33what would that mean? I think about it. Never, never, never happened. Uh, I was hoping,

42:40Andreas, are you still there that maybe you can kind of address, you know, this from a portfolio

42:45point of view, because you, you probably have more of a holistic viewpoint, um, and specialized.

42:50So I can give it a holistic one. Uh, 2 million years ago, we were homo sapiens,

42:56and we were hunting and foraging 10,000 years. We can started, uh, agriculture. Uh, and with

43:03monocultural farming, one person could feed a hundred. That meant 99 people had to go out and

43:10jobs outside of farming. And they did that for 8,000 years until the industrial revolution.

43:14And that kicked off in late 1800s. And the pace of change and job reductions is nowhere near what we're

43:26going to see here. If, uh, one large, uh, BC in California prediction that 85% of Americans will lose

43:37their jobs comes true is almost unconscionable because you can't retrain those people quick

43:44enough. And those of you who don't remember 1929, uh, the great panic and the worst depression.

43:51So you got 24% unemployment. Who's going to buy these guys, who's going to pay the subscriptions for AI

43:57when 85% of the workforce is out of a job. So I, I really, uh, we have data on this already,

44:03right?

44:04It's grandiose statements and really don't do anything to do, do anything about it.

44:10So, so, so, so, so, so, so some of the data that we have on this is that how many people who

44:16worked at map makers or Kodak or, you know, old school telephone makers, but wires,

44:23you know, got jobs in tech afterwards. Um, yeah, I'd say very few, right?

44:32Those, those, those people, I don't know what happened to those guys in Rochester, right?

44:38I would, I would agree with Andrea strongly. And actually,

44:41I think Farhan was also mentioning that the history of humanity, any innovation improves

44:45productivity, but the best majority of these gains go to the consumer, right?

44:50If we look at actually, um, even I think Andreas' analogy to farming is actually classic.

44:56It's phenomenal. It's like, because 90, like even like 300 years ago or 400 years ago, like 97,

45:0197% of population was doing basic farming, right? And with increasing productivity right now,

45:06we have maybe two or 3% of population in Western Europe or in the U.S. are involved in the farming

45:13because just they saw good enough, right? So that freed up the rest of the humanity to whatever they want,

45:18right? And who benefits the consumer, consumer benefits, right? And so we'll be with AI,

45:23there was a history, like jobs will be evolving and so on. And, uh, Ivan, can you share, like even

45:29taking the most, the most basic industries, like which have not really evolved much, like I'll share,

45:34I'll share for the fun of it, say, airlines tickets for the last 80 years. As much as people complain

45:42about high oil and weight prices for pilots essentially in real, in real prices adjusted

45:49for inflation, airlines tickets were dropping pretty much with ups and downs. So again, the most of the

45:56benefits of even the plane technology has not evolved a lot, but small incremental improvements

46:02all accrue to the consumer. And that will happen with AI as well, uh, in my opinion, right? There will be

46:07obviously some major gainers, um, on the way there, but history tells us it's very hard to sustain that

46:14edge and consumer and humanity in general will benefit the most. Well, if you guys have any idea

46:20who's going to be the AI supermarket, please let me know as soon as possible. Uh, and my final question

46:25is, so I kind of say that most of you, you guys all have roots in other countries. It seems like, uh,

46:32uh, can you talk about the international scene for AI and, you know, what's your viewpoint? Because

46:37I'm going to predict that Europe's going to be a slow adapter because they like to protect jobs there.

46:42And also because there's not a lot of energy for AI, right? Uh, but I would say that places like Asia

46:48might have, you know, South Korea might adapt up pretty quickly, right? So can you give some perspective

46:54on the international outlook for AI? Uh, sure. So, uh, I do a lot of sourcing of a lot of our, uh,

47:03investments in Europe. Uh, recently the European Technische Hochschule, which is one of the biggest,

47:11one of the larger engineering schools in Zurich has announced that it's becoming the AI hub

47:17for Europe, uh, at least on the continent, uh, as I believe there are some London universities that

47:25are doing the same for the UK. Uh, concurrently with that, if you go to Doha, Qatar, or the U-A-E in Dubai,

47:36uh, they're also contesting to be, uh, the AI hub for the Middle East region. The difference between the

47:43Swiss and the, um, U-A-E in Doha is that U-A-E in Doha are considerably well, better funded

47:53and are out to prove the other one wrong in their quest for the head of AI hub of the Middle East.

48:03I've got a, you know, highly favorable viewpoint of the Swiss when it comes to doing things,

48:08you know, they said they were going to become the European capital of crypto and they've done pretty well

48:12in Zurich, but you're right. I mean, they are underfunded and they tend to wear not very good,

48:17nice clothing either. You know what I'm talking about. Uh, I just don't get how they match up their

48:25outfits. Uh, Konstantin? Uh, we, ironically, let me say a couple of words, right?

48:35You're off, you're off mic. Yeah. So if you look at the actually, um, what was actually

48:41became a revolution in AI, at least in terms of mass adoption of LLM and Generative AI.

48:46This paper, which was written like a few years ago, uh, by the guys essentially working at Google.

48:51So essentially this guy, Ilyapolosukhin for the, uh, the kid from Ukraine wrote the whole thing.

48:56I think he conceptualized it with another guys, but even just naively, again, uh, looking at their

49:01last names and first names, you can realize guys that came from non-US, right?

49:06And again, it's just a reality of the world that we have lots of high quality brains who are hungry

49:13and they see innovation as a way to realize themselves and achieve something better. So

49:17this is actually, again, a good example that a greater deal of innovation, even though it's kind

49:23of like in the U S yeah. Um, the Genesis, some inspiration and some educational background

49:29presumably comes out of us. Um, and I think to be a statement, but there's a lot of high quality AI

49:36people in India, in China, in Eastern Europe and France is doing a remarkable success. Like

49:43you gotta look how much money French or, uh, LLM sort of like, like Mr. Al have raised. It's, it's

49:50phenomenal. France is a huge country, like maybe 80 million people, but the amount of contribution

49:56they are doing to development AI is way disproportionately, like in a good way.

50:00So I think France has always been a leader in tech and, uh, in software development in Europe,

50:05right. And historically many centuries ago in mathematics as well, right. So that's related,

50:12right. And, uh, do you, do you guys think access to energy will be differentiated? I mean,

50:17obviously Qatar and UAE have tremendous amounts of energy that they can spend on this versus Europe.

50:24Andreas, maybe, or Constantine? Uh, we're going to get to Asia a couple of seconds.

50:28Yeah. Uh, it's unconscious, but I mean, the, the king of, uh, or the prime prince of

50:35Saudi Arabia is worth a trillion dollars. Uh, the head of Qatar is $350 billion. These are numbers

50:41that dwarf our American billionaires. Uh, they're not short of, of cash. I do want to throw one,

50:47I think out there, uh, IIT and Hyderabad and India, I believe is also becoming a, uh, AI hub. So I didn't

50:54want to forget about them as well. Right. And I understand that India, uh, just got some cheap

51:01oil from Russia. So, uh, Farhan, your, your thoughts. Cheap oil is good.

51:10Well, I'll go back to the previous, uh, question.

51:12The international assumption of AI. So different countries and different jurisdictions are building

51:21on AI in very different ways. Uh, India, for example, there's this whole initiative about, uh,

51:27AI for public, where it is being, uh, pushed by the government, uh, as a public utility. So that's a very

51:36different way of developing this ecosystem as opposed to what you see in the US and probably most

51:42of us in Europe as well. And, uh, we're talking about AI at scale, a scale of 1.4 billion, uh, people.

51:51So the applications that get developed, the way they get developed, the way they get deployed,

51:56it's very, very different from what you would see here. Uh, and we are already seeing that divergence,

52:04uh, shaping up.

52:07Yes. And actually on the, since, uh, Marty, you had mentioned, uh,

52:12the energy component of it, and actually kind of like linking a little bit to the previous panel,

52:16when guys were talking about a lot, um, and ladies were talking a lot about energy.

52:20I mean, there is obviously, we will probably experience some sort of energy renaissance,

52:24at least power generation renaissance, you know, centers where AI will be deployed, right?

52:29Uh, we talk, people talking about nuclei being, being big, and we see a more global sentiment shifting

52:35towards more favorable towards the nuclear energy, including coincidentally gain in France,

52:40um, and so on, right? So, um, so yeah, I mean, it's kind of like essentially highlights the

52:48gaps and holes in our ecosystem, which needs to be upgraded to facilitate that process.

52:53Yeah. Uh, guys, it's a shame that Darwin wasn't able to make this call, uh, this, uh, conference,

53:02because, uh, he's been doing AI with China for 10 years. And, um, I think that there's a,

53:10there's an AI race that's ongoing between the United States and, uh, China, as well as, uh, you know, uh,

53:18uh, quantum computing, which I think will change everything, uh, especially, I think it's going

53:24to be a security issue where everything is from the internet going forward with when quantum computing,

53:29uh, uh, comes around. Uh, you know, I think, so I was talking to Dan Moore, who presented to our

53:36group many years ago on crypto, and we asked him, so what do you think about, uh, you know,

53:40quantum computing with regard to Bitcoin? And he said, well, we'll have to get quantum Bitcoin.

53:44So, um, I think it's, you know, it's, it's, it's, it's kind of a, it's like the space race. It's,

53:51it's, it's a real, it's, uh, the money that's going to be spent on this stuff will be out of control.

53:56So anyway, thanks so much. Very thought provoking panel and, uh, you know, uh, appreciate a lot,

54:03a lot. Thanks so much. Thank you.

Category

Transcript

Recommended