How Funds Use AI to Beat Rivals

LEARN THE CAPITAL RAISING STRATEGIES AND FRAMEWORKS used by alternative asset professionals: https://go.fundraisecapital.co/apply

Can AI actually find hedge fund trades a human would miss?

In this episode of Making Billions, Ryan Miller sits down with Jan Szilagyi, CEO and co-founder of Reflexivity, the AI investment analysis platform used by firms including Soros Fund Management and MUFG.

How do hedge funds use AI to find trade ideas? What is a knowledge graph in finance? How do you stop AI from hallucinating numbers when real capital is on the line? Can a small fund use AI to compete with a multi-manager giant?

Jan answers all of it from a two-decade macro career that included trading alongside Stan Druckenmiller at Duquesne and running a 15 billion dollar book.

Together, we discuss why the expression of a trade idea matters more than the idea itself and how a knowledge graph maps market ripple effects a spreadsheet cannot show.

Learn why AI is a leveler for fund managers and how to run scenario analysis before risking capital. 

[THE HOST]: Ryan Miller is a fund manager, capital strategist, and former CFO turned angel investor in technology and energy. He is the founder of Fund Raise Capital and Aequor Capital Partners, and has mentored over 1,000 fund managers across private equity, private credit, venture capital, real estate, and alternative assets globally.

[THE GUEST]: Jan Szilagyi is CEO and co-founder of Reflexivity, the AI investment platform used by Soros Fund Management and MUFG. At Reflexivity, he pairs large language models with a proprietary knowledge graph to surface trades and map risk in a zero-hallucination environment. Holds the world record for the fastest Harvard Economics PhD. 

Subscribe on YouTube:
https://www.youtube.com/channel/UCTOe79EXLDsROQ0z3YLnu1QQ

Connect with Ryan Miller:
Linkedin: https://www.linkedin.com/in/rcmiller1/
Instagram: https://www.instagram.com/ryanmilleroffical/
X: https://x.com/_MakingBillions
Website: https://making-billions.com/

Transcript

Ryan Miller

Hey, welcome to another episode of Making Billions. I'm your host, Ryan Miller, and today I have my dear friend Jan Szilagyi. Jan is the CEO and co-founder of Reflexivity, the AI investment analysis platform used by firms, including some of the largest funds like Soros Fund Management, MUFG, and a lot more just like it. He holds the fastest economics PhD in Harvard history, earned under Ken Rogoff, along with a BA and MA degrees in mathematics and economics from Yale. Over a two-decade Global Macro career, he traded alongside Stanley at Duquesne Capital and managed portfolios at Fortress Investment Group and served as the co-CIO of Global Macro at Lombard Odier, running a $15 billion book. Reflexivity has raised over $40 million with a $30 million Series B led by Greycroft and Interactive Brokers and personal backing by the man himself, Stan Druckenmiller, and Grey Coffee. 

Ryan Miller

So what does this mean? Well, it means that the man who builds the AI tools that the biggest funds in the world are using right now is about to show you how to use them too. So you can better find trades, protect your portfolio, and compete with funds a hundred times your size. So with that said, Jan, welcome to the show, man. 

Jan Szilagyi

Thanks for having me. Looking forward to it. 

Ryan Miller

Yeah, it's good to have you here, man. I've been a big fan. And I remember the first time that we met, you absolutely blew me away with what you've created at Reflexivity and how it helps a lot of hedge funds and traders just really understand what's going on in the market to a degree that it just synthesizes a lot of data. And I know I'm not doing it justice. You're probably the smartest guy I've ever talked to in my life. But I'm excited to get into this, man. So with that said, Jan, you ran a $15 billion macro book. You traded alongside the legends Stan Druckenmiller at Duquesne, and you hold the fastest economics PhD in Harvard history. And today, your platform, Reflexivity, powers research at firms that are some of the biggest in the world. So walk us through exactly how a fund manager uses AI to surface a trade idea that the human eye would easily miss.

Jan Szilagyi

Yeah, well, first of all, thank you. You're  super kind. The main keyword here that I would emphasize is the knowledge graph. So we at Reflexivity use a combination of the reasoning layer of large language models like Claude or, for example, ChatGPT. And then we combine that with a knowledge graph, which is really a very comprehensive mapping of all sorts of relationships in the investing world that matter. So giving the system some degree of understanding that if the yield curve is steepening, that could be good for banking stocks, that if there is a rise in oil prices, that might change the ethanol, gasoline blend, and so on. What portfolio managers will therefore do is use this capability to be able to better understand ripple effects from a number of market events. So something happens in the market, it could be a macro event, a micro event, a geopolitical event. And what everybody is racing to find out is what are the implications for my portfolio? Are there downside risks? Are there potentially opportunities that are arising from this? And a system like Reflexivity, and I think AI at large is extremely well placed to be able to very quickly ascertain some of these implications and give you an edge that way.

Ryan Miller

Brilliant. So when you and I first met, you, we talked about placing trades and really just how do you form thesis in hedge funds, and you're definitely experienced that. And, you know, we kind of landed on really comes down to the trade idea and that expression of that idea. And so I'm just curious, based on that, how does Reflexivity and the software that really helps these traders, how does it tie into both of those very fundamental things as far as building a hedge fund?

Jan Szilagyi

Yeah, that's a great question, actually, because I think a lot of focus often when it comes to investing is on idea itself. And the reality is that I think a lot of people will come onto the same idea, but where those who are extremely, extremely good at this, but usually Excel is the actual expression of that idea. And that I think is something completely separate insofar as you're now saying, how do I either structure a trade, how do I express it, and so on. And so if you think back on, I'll give you a  Global Macro example because that's where I come from. But when people talked about the very famous source pound trade, you know, they would often hear him say, well, you know, we basically bet the fund on this, and you think, well, that's crazy, right? That's a lot of capital to put at risk on a single trade. But when you look at the structure of the trade, you understand very quickly, and he had explained that in several interviews, the way the exchange rate mechanism worked at the time, if he was wrong, he was only going to lose maybe a few basis points. But if he was right, he was going to make several percent. And on a base of having the entire fund bet on that, that ended up being quite a lot of money. But the key here is he found an expression of the idea that offered by far the best risk reward and was therefore able to and ultimately monetize it in a really, really big way.

Ryan Miller

Brilliant. What a cool way and so he was able to not just understand the trade, he had a good idea, but the expression of the trade, one of those was he actually structured in a way where you could bet, make a lot of money, have that asymmetric risk to reward, and still cover your downside. And then the upside, as we know, one of the most famous trades in history. He did quite well on that. I love that. I'm just curious when working with Stan, what did you watch Stan Druckenmiller do? Just walk us through maybe of some steps that you were able to observe in the moment he decided to size up on a position.

Jan Szilagyi

What I noticed is he was extremely good at basically after he laid out a particular thesis and how he thought about why a trade made sense, he was very good at then observing whether or not the subsequent price action actually was consistent in response to incoming new information to what he would expect it to be. In other words, if you know you basically had a had a trade-on that was long dollar and that was contingent on higher interest rates and therefore higher interest rate differential and inflation numbers, and you ended up getting a higher than expected inflation number, but the dollar stops strengthening in response, that would immediately trigger some alarm bells. In other words, the trade wasn't, the price action wasn't really corresponding to the what you would envision as your hypothesis was. On the other hand, if it did, that's when you could see him validate the thesis and then really scale up the trade so that it could go from being a very small trade to a very large trade in a very short amount of time, and conversely, could go from being a very, very large trait to having no trade at all, also in a very short period of time. So this was there was this incredible nimbleness in response to intuiting and observing how the correlations between prices and different assets that he considered to be important to watch for a particular trade were   acting. 

Ryan Miller

Awesome. Were, so could you give an example of how that may have may show up in today's market? How would somebody maybe replicate the methodology on that?

Jan Szilagyi

Yeah, I mean, I think this is in particular where I think having some help from, in this case, we've obviously built Reflexivity particularly around that concept, but the idea would effectively be that you are starting to, or you assume that you know perhaps the equity market has a lot further to run, that that's basically based on the assumption that interest rates are not going to rise and that the Fed will continue to be accommodative. But you start to notice that actually as yields come down,  equities are no longer actually rallying that much. They seem to be kind of stuck. And you can do this by watching prices obsessively. So that's one way of doing it, or you can have, which I think is a far easier way of doing it, you can have the machine just in real time calculate these sensitivities, calculate these correlations and so on, and basically alert you to moments when it sort of feels like either the trading volume or the price action is really no longer as bullish as you would like to think it should be. Experienced traders will rely on any number of different market indicators, including trading volume and breadth, right? Like for example, within the index, how many stocks are actually rallying, how many are actually carrying the index forward. And any one of these could become a signal to the trader that the conviction is basically weakened by the responses you're seeing in the market from the asset that you are either long or short. And again, this is where I would say machines are just far better because they're really good at very conscientiously keeping track of the numbers, seeing any changes in correlation, changes in beta, and so on. 

Ryan Miller

Brilliant, brilliant. And Reflexivity also helps people to do that. Is that right? 

Jan Szilagyi

That's correct. Yeah, because we thought where you could make the workload a lot easier for a portfolio manager or an analyst, is to enable this kind of wholesale monitoring of what is ultimately just a gigantic correlation matrix in the market.

Ryan Miller

I love that. So Druckenmiller wrote you your first check, if I'm not  mistaken. Not bad. Yeah, not bad. So pitching a billionaire investor. How does a fund manager, if they want to be able to pull off something like that, how does a fund manager structure a pitch that gets a legendary investor to say yes? Like, how did you get that and maybe teach some methodologies that other people might be able to find useful?

Jan Szilagyi

You know, I wish that I could draw out some method that I could say, like, oh, you know, here is, here's the recipe, here's the formula. I think in this case, obviously it happened because he and I had worked together for a very long time. So I think he knew what I was interested in, what I was capable of, what I proposed to do. And I think there was a degree of trust that came from that. Now, that being said, I think as you and I discussed, I think the first time, there are ways in which you are able to build that trust if you establish some kind of a track record of what you say you will do and then ultimately do it, deliver it, and so on. I think people generally like to see that when you previously had said, look, I'm going to go out and I'm going to build this, that there is evidence that you did go out and build it. It may not have worked, but you did what you said you would do and you executed on that. And I think that's a huge part of that trust that ultimately will get somebody to say, like, you know what, I'll take a bet on you.

Ryan Miller

I love that. So he kind of hinted that maybe you had an unfair advantage, but I think there is something. And in that with that specific investor, yeah, I get it. But there is a method in a lesson I just don't want to skip past   people listening. And I say this all the time on the show that the three most valuable assets in your possession are your reputation, your relationships, and your results. And so you really did have a method that worked when raising capital is to say, I had a relationship with one of them. See, and then you had a track record, which is results, right? And so, and then you had a reputation with them and you earned it because you worked together. So there's many ways, but for sure that's one of them. But what this talks about is what it really underscores how you got there is less important for the context of this, is that the methodology still holds that reputation, relationships, and results is what gets people to invest. Like if you worked with them but you had a horrible track record, would he invest? Maybe not. Or maybe had a great track record that he's never heard of and he doesn't know you, could he trust you? Probably not. And so there's a lot of those things that need to be in place. And when they are, whether it's Stan or somebody else, now we're able to start raising capital in our own funds or our deals, whether you're building a hedge fund or otherwise. Having those three R's are absolutely critical. So even though you're like, well, it's not fair because I knew him. I was like, well, in that case, yes. But it underscores and it adds further credence to those three R's, those foundational things. Would you agree?

Jan Szilagyi

I do agree. I mean, I think even though in this case, as you say, I had sort of a captive audience in him, I think that generally that's absolutely true. That I track record, be it in hedge funds, be it in building a business and so on, it does matter. And so long as you're able to show something that people are able to evaluate somewhat objectively, it is going to kind of reduce that barrier to entry for them to then say, like, yeah, I'd I'd love to join you on this journey.

Ryan Miller

I love that. And he certainly did. So, you know, then that leads me to wonder how does a fund manager, let's say they're running a sub 100 million, right? And you and I are both like, ugh, ouch, sub hundred million dollar fund, which is really hard. You could do it. It's hard to run. So if someone's in that position, how do they use these same AI tools to go toe-to-toe with a multi-manager giant, like some of your clients?

Jan Szilagyi

Yeah, it's a great question. And I do think on some level, and I explained how I think that works,  AI is going to be a bit of a leveler here in the sense that when you think about a, let's take a fundamental long short equity fund,  a lot of work that they have to put in before they know if they have a trade idea is front-loaded, right? Getting access to all the data, doing the work, doing the analysis. And at the end of that, you still may not end up with anything that you actually include in your portfolio. You can only do that so many times. I'm obviously simplifying hugely, but you can only do that so many times in a given year because it takes experience to decide which ideas to pursue, which ones not to pursue. And even so, you're still constrained. If you're a smaller fan, you're particularly constrained because you may have one analyst, or maybe the PM is also the analyst. And so you're kind of doing both portfolio construction, idea generation, and so on. What a system like Reflexivity allows you to do is we like to say that we are massively reducing the cost of curiosity, right? You are now delegating this front-loaded effort to a system like Reflexivity and saying, can you find out for me whether or not this particular hypothesis that I have has any merit? Is there any way that this could be a trade or not? And if the system is then able to procure the data, run the analysis, and give you the answer in a matter of minutes as opposed to days, well, your throughput has just increased quite dramatically. So you have access to better data, you have access to more analytics, the productivity gain is absolutely real.

Ryan Miller

Brilliant. So this really helps to get you into testing, I guess, in a more safe environment of different ideas, trade ideas that we talked about before, the idea and the expression of the idea, and then Reflexivity. One of the features that it helps with fund managers on the hedge fund side is working through those ideas and maybe finding other correlations, just testing your theory without putting all your money on the line. And so I guess it takes away some of the gambling. I know hedge funds aren't gambling, but, it takes away that gut feeling to say, no, we are very confident moving into this trade because we've used certain tools that have really helped us to solidify what we're working on. Would you say that's an accurate review?

Jan Szilagyi

I would. And you know, maybe another analogy that I could draw here would be to say if you think about how triage works in, let's say, an in  the ER, right? You have a lot of people coming in, and a very experienced doctor or nurse is basically going to have to decide who needs immediate attention and who can wait for maybe a few minutes or an hour or so. And a lot of it is in-the-moment decisions. They don't have a, they don't have the time by definition to give everybody a full physical and then decide how to triage this. The role of Reflexivity in an example like this is that actually suddenly you do have the time to do a much more complete evaluation of every idea in a much shorter time span. So, like the time frame has constrained. And what I think ultimately that means is the maybe it's going to give an upstart money manager a little bit less fear in competing with the more experienced ones, because the experience was particularly valuable where you had to rely on some of these heuristics where you had to say, like, oh, you know what? I've seen this movie before. This isn't going to be a trade. I know how that plays out. And you're like, Well, you know what? I actually in a few minutes will know the answer because I can actually test it quantitatively. So it is also maybe shrinking that gap a little bit.

Ryan Miller

Brilliant, you know, that that kind of leads me, earlier, you talked about knowledge graphs when we first kind of opened our discussion. So, how does a fund manager use a knowledge graph to expose a portfolio risk that just a spreadsheet doesn't show? And before you answer that, maybe explain what a knowledge graph is, just in case this is new for some, and then maybe go into how does that show that it's far more superior than just data on a spreadsheet?

Jan Szilagyi

Yeah, no, I mean, obviously, we love knowledge graphs  at Reflexivity, but I think the best way to think about this is that you are effectively giving an otherwise very sophisticated and capable system a mapping, an ontology of what are the things that actually matter in the market, how they connect with each other. Right? Like instead of having to figure out from scratch how an economic system works, how a bank works, how it relates to, for example, interest rates, how commodities depend on the shape of the yield curve and so on, you give it that mapping, which then means that all of a sudden, whenever there is a particular change in any corner of the asset universe or the economy, it's able to traverse those relationships to really quickly figure out what the ripple effects of a particular event are going to be. And if you're now a portfolio manager that is in possession of a system that has this mapping, it means that you will be much, much faster able to figure out whether or not an event that maybe, on the face of it, doesn't seem like it has much to do with your portfolio at all, actually does expose it to a particular vulnerability that you would not have guessed just from a spreadsheet that probably only has these first-storm relationships. I think you and I talked about some examples last time when there's been obviously quite a lot of energy and effort spent on figuring out how is the crisis in the Middle East, or the war with Iran, how that's going to impact asset prices. And there is that first level of implications that I think is obvious. Okay, oil prices going up, inflation is going to be higher, yield curves can be steeper,  it's good for drillers, maybe good for oil companies, and those fields right there. But there's a lot more, right? There's a production of plastics. It impacts, for example, sugar growers because the ethanol, ethanol gasoline blend might be changing. It impacts, for   the amount of fertilizer that's available, and therefore the harvest in the Western Hemisphere might be less than it otherwise would have been. There are all these, and that's what makes financial markets ultimately so exciting, is that it is an incredibly complex system. And without a mapping of sort, it's really hard to organize the kind of the relative importance of these consequences and relationships and so on.

Ryan Miller

Brilliant. It reminds me almost like, we'll say a spiderweb. And geek out here. I don't. I don't have a bunch of PhDs, so  I'll use dumb guy talk, but I'm joking, I'm delighted. But they, when one strand is affected, it might be felt somewhere else. And so that knowledge graph does help to create a little bit of that ripple effect to say, yeah, if you shut down the strait of hormones, for example, where will that be felt as you were talking about? And then it can kind of show you a little bit. So that's, I think is that's what we're saying. Yeah.

Jan Szilagyi

I think that's, I think that's exactly the right analogy because if you pull on the spider web right in any corner, you can kind of see which other parts are moving. And that's exactly how a knowledge graph ultimately works, is that you have all of these edges that are connected and traversing them lets you figure out really quickly how and where the ripple effects are traveling.

Ryan Miller

Hey, if you're finding value out of this discussion, could you do me a huge favor? Could you just take a second and hit that like or subscribe button? It costs you nothing, but it tells the algorithm that this is valuable information and it helps us to get it to more people. Thank you. You're incredible. Now let's get back to the show.

Ryan Miller

I love that. And you know, especially when you have capital at risk, that's really which is almost every fund manager, but when you do that, especially in hedge funds. So I wonder if you can just walk me through how a fund manager should run an AI scenario simulation before putting real capital at risk when they're using Reflexivity.

Jan Szilagyi

Yeah, scenario analysis actually is a great example of how and where this is particularly beneficial because, as I think has become pretty clear from the discussion we've already had so far, financial markets, by their very nature, the the kind of the ripple effects and the connectedness of the system means that the implications will be far and wide. And what you ultimately are trying to figure out as a portfolio manager is under a certain set of assumptions, under a certain macro or microeconomic scenario, you really can only correctly guess the implications and the consequences for your portfolio if you have fully modeled all of the different connections, right? Like if you are assuming that, I don't know, there's going to be a scenario in '27 where the US economic growth slows considerably, it's extremely important that you know what the economic sensitivities are of your banking stocks or your tech stocks. What does that mean actually for the funding environment and so on? Those are very complex relationships. And what large language models really are very good at, and humans tend to be less good at, is this kind of multidimensional synthesis, right? It doesn't run just one question in parallel. It's basically doing all of them at the same time. So where you might, you and I might kind of go linearly from, oh, you know, if oil prices go higher, inflation is gonna be higher, then interest rates are gonna be higher. A large language model is doing all of these different possibilities all at the same time, which is particularly helpful when you're evaluating risk through kind of the lens of these different possible scenarios.

Ryan Miller

Brilliant. Yeah. So then a big part of that, and I'm gonna we're gonna talk about the H-word in the AI hallucination, but how does a fund manager make sure their AI never invents a number when there is that real capital on the line? Because that I mean, if you overly rely and just say, yeah, it's right, whatever it says I'll do, like how do you how do you deal with that?

Jan Szilagyi

Yeah, I mean, I think this has been a big obstacle to not only wider adoption of AI within investment management, because obviously these are high-stakes decisions and you need to know that you can rely on the analysis that you're seeing. Yeah,  but it's also made everybody ultimately use it for, I would say, kind of low-stakes decisions. It's almost like, okay, fine, it's okay as like maybe amplified Google search, but I'm not going to rely on it in order to actually determine the valuation of a company in which I might place 10% of my fund. What we specifically have set out to do is provide an environment where not only do you prevent a large language model from making up a number, but also to give you the kind of visibility and transparency and ultimately auditability that you're always able to check all the steps. Now, ideally, after some time, you're not doing that anymore because then the productivity gain is not quite as large if you're each time doing a complete  audit. But it gives you that, I would say, it gives you that confidence in the answer because you think, okay, well, if I wanted to, and previously I have checked it, I could, and I know that it will tell me I don't have the data if it doesn't, as opposed to be just saying, hey, here's some number. Oh yeah, and by the way, I made it up. Got it in. 

Ryan Miller

So  it sounds like really  just putting. It in my own words, you just you said it great, but just my understanding is it really just makes sure that you don't really put it in a place where that's possible, I think is what I'm hearing. And not always easy to do, but I think that's the recommendation as of right now is to say a lot of great value that comes from it. But if the closer you get to the final decision and relying on AI to do that, that is a real risk. But if you do it on other parts that can still add a lot of value to synthesizing data to establish the idea, and then um now we're in a position to say, yeah, hallucination, as long as you have AI, do it in where it's good at and don't try to push it in the areas where it does make more mistakes, then that's typically the recommendation. Is that, that's what I'm  picking up, what you're laying down?

Jan Szilagyi

Yeah, that's right. And I think you're also, and this is already happening, but I think within finance, just like we're doing with Reflexivity, and it's happening in other parts of finance as well, we are starting to use the power that LLMs bring, which is the ability to reason, the ability to run and procure data, run analysis, and so on, but then create constraints around it and say, okay, you're going to operate in a way that is much more akin to how we're used to working in investment analysis, which means you're going to check your results, you're going to only work with data that you trust. If you don't have the data, it's okay to say, I don't know the answer, and so on. So it changes, I guess it changes the utility function of the system a little bit compared to an LLM that is much more geared towards really wanting to effectively give an answer almost at all costs.

Ryan Miller

Yeah, yeah, brilliant. And, you know, speaking of the utility, speed is definitely one of those things that we can do things a lot faster. It's, therefore we can scale more with less. And so, what can a fund manager do on Reflexivity in 10 minutes that used to take an analyst, say a full week or more?

Jan Szilagyi

So I'll give you a funny anecdote which happened when  we were giving a demo to a very, very large fund. And just at the time, you may remember this, across the tape came a headline that basically suggested that President Trump had, you know,  tried Coke in Mexico, thought it tasted a lot better, and so suggested that Coca-Cola in the US should replace fructose with real sugar. The client looked at, you know, we were a team of three doing the demo and said, like, okay, look, we would not spend the next two, three days building supply-demand analysis of the sugar market on the off chance that there's a trade somewhere in this. But if you guys are as good as you say you are, why don't you show us  what Reflexivity can do in this case? And in a matter of minutes, it was actually incredible to watch. It really went into the depths of, you know, we connect to S&P Global, we connect into DataStream at LSEG and so on. It pulled up sugar harvest data for Hawaii, Texas, Louisiana, Florida, the four sugar producing states, and created and estimated the gap that would result from that kind of change, identified Brazil as being the country from which the gap would most likely be plugged, and then a couple of companies that in that scenario would be probably mildly undervalued. It didn't amount to like a fully fledged trade idea, but it was much closer than what you would have done if you were literally just basically spitballing with that headline and kind of guessing. And so I think it really showed how you were able to bring a lot more rigor, a lot faster to this kind of evaluation of market events to see if this is worth pursuing or not. And I think that's only going to become more and more obvious as more people start using this technology because they'll just be much faster and more rigorous at either dismissing or pursuing trade ideas. And I think there's ultimately tremendous benefit in that.

Ryan Miller

They loved it. It sounds like you were able to really showcase what the power this can do and just in a few minutes with and then finding Brazil and all that stuff. And so, right in real time, they put you to the test, and it sounds like you crushed it. So I love that. Now, the question a lot of people wonder is the impact of AI on alpha. So when every manager has the same AI tools, how does this fund manager still generate an edge?

Jan Szilagyi

This is a question that obviously we get repeatedly, and I have thought about quite a lot, but I always come back to the same answer, which is at least the way it's currently designed, the quality of answers that you will get when you are interfacing with Reflexivity and AI at large is really only as good as the quality of the questions that are being asked. And I do think that, yes, having access to Reflexivity will give everybody the ability to answer their questions much faster. But it also now suddenly gives a huge premium to the value of the question itself. And so what I mean by this is that I think people who are able to think creatively, who are able to think about questions that others hadn't thought of and so on, are actually the ones that are going to find it much easier to generate alpha in this market. And in particular, it's going to be true of those who maybe previously would have censored themselves because they thought, well, I have this question that I believe I know the answer to, but it involves a lot of quantitative analysis. I'm not a quant, and so I don't know how to even pursue it. You've taken that out of consideration because you now have, you can tell Reflexivity to effectively run almost any kind of analysis, it'll do the computation for you. And so the excitement becomes what are the types of things that I should be interested in, right? So the value of the value of curiosity, even as the cost of curiosity comes down, basically skyrockets.

Ryan Miller

Brilliant. So cost comes way down, but really at the root, first principles is to say, well, the edge really comes from the question more than the tool. The tool is neutral. So the value is unlocked through the questioning, not necessarily just by having an AI. Totally agree, man. I love that. So your company is named after Soros' theory of Reflexivity. So how does a fund manager actually trade the loop where prices start changing on those fundamentals?

Jan Szilagyi

So maybe I'll just give a quick refresher for those who may not know what Reflexivity basically is, but George Soros popularized this term that effectively referenced the fact that yes, prices reflect fundamentals, but actually fundamentals might also be impacted by prices. In other words, you may have situations where companies are actually taking action or making decisions on the basis of what the share price is doing. So you have this self-reinforcing mechanism. What I think happens in an environment that we're now seeing, and with what I was just telling you, I think is going to happen as everybody gets access to better AI tools, is that I think that loop is just going to be faster and faster. So if you have access to Reflexivity now, you're able to do that calculation much, much quicker. You basically can see, ah, okay,  here's what the share price has done. This is what I think it's going to lead the management to do in response to this. And so I may be able to trade in anticipation of that reaction. That's that gap between one action and the other, I think, is going to be shrunk because it's not, it's going to take just a lot less for the market to process any kind of relationship, any kind of implication that previously might have been, you know, one or two days of endless time.

Ryan Miller

Brilliant. So tons of value unlocked on that. Given that this is such an important development in high finances, using AI, and people are still finding a lot of use cases, and they're great. But what specific skill should every fund manager start building today so AI makes them more valuable instead of replaceable?

Jan Szilagyi

Yeah, I think it goes back to this idea of creativity, curiosity, and so on. I think it becomes very important to try and be very broad about what you think the implications of a particular event, how you might think about a particular trade structure, how you might think about an expression of a trade idea, what kind of instruments you would use to express it, and so on. I would imagine that we are entering a world where for the first time, the huge amount of data that we've been accumulating for a very, very long time is now going to be matched with the infrastructure that can actually use it. And in doing so, I think you're going to add incredibly rich layers to a particular view that not only will be rooted in much more rigorous analysis, but will actually be able to say, like, well, wait a second, there's this other relationship here that I hadn't even really thought about. And so instead of me just going long oil because of the conflict in the Middle East, maybe I'm going to realize much sooner than everybody else that there's going to be shortage of some kind of plastic component that is absolutely crucial in the production process of something that otherwise seems several, several worlds removed from what's happening in the Middle East. And the ability to start thinking in those terms, the ability to kind of anticipate and look beyond the obvious is what I think is going to make some fund managers be able to use this system incredibly, incredibly more productively than others. Because if yeah, it, it'll be obviously just as fast in answering boring questions, but where it's going to be extremely useful and generate a lot of alpha for managers is where you and some of the questions you might ask will be crazy, will lead to nothing. But the point is that you'll be able to find that out very quickly. You can churn through dead ends in your trade idea generation much, much faster, which means that the cost of that is going to be negligible.

Ryan Miller

Really brilliant. You know, my last question, this has been fun, man. I have a feeling this will be one of many more chats to come between you and I. 

Jan Szilagyi

I hope so. 

Ryan Miller

Yeah, this is a good time. So I think I know part of the answer to this question, which is, use Reflexivity. But maybe give us a playbook. What are the first couple of moves a fund manager makes this quarter to put AI into their investment process? And part of that. So I, and I'll say it,  use Reflexivity. I'll say it for you, buddy. But they  maybe walk through a little bit of, you know, you get in, what is that experience  using Reflexivity? How does that help you to really get that, as we say, the hockey stick moment where your fund really has lifted off using that tool? Maybe walk me through the playbook of a little bit of that.

Jan Szilagyi

Yeah, and I mean, obviously use Reflexivity, but let's talk more broadly AI. I think step one is, particularly for those who haven't done much with, is you could actually even just play around with AI to try to think through what are some of the biggest pain points in your current process, right? Is it data acquisition, is it analysis, is it summarizing things, is it processing investor presentations? Like just identify certain things that are either tedious or take a long time. You know they have to be done, but at the moment you just think if I could make one aspect of my process better, these would be the three things that are the top candidates. Once you've identified those, then I think the next step to do is experiment with AI technology to try and think about could I use this in order to make any of these better? And I am absolutely convinced that with this kind of experimentation, two things are going to happen. You will find that some aspects can be really resolved through the technology that already is available. And number two, I think you're gonna develop a very valuable intuition for how this actually works. What are the pitfalls, right? We already talked about hallucinations. Obviously, that's a negative, but I think that if you start to get better at spotting some of the answers, you will develop intuition for that. And therefore, I think it's going to become more useful to you. 

Jan Szilagyi

It's a lot like when you're working with, when you're working with an analyst, there are probably also some tells that you develop and you're like, you know what, I can tell that the way you're talking about this, you're not totally confident. I want you to go back and check the numbers because it's not, I get the sense that something here doesn't entirely add up. The third step then would be once you've kind of done this, just do it all over again, right? Like you now have a process, you have AI helping with some steps of this, do it again, do another overlay, try and see if there's now another aspect, another step where I think you'd be able to improve what you're doing. And the reason why I'm saying this is that if there's anything that we have really seen in the development of AI so far is that the time frame has shrunk incredibly. The speed with which we've seen all of this advance and the capabilities that six months didn't seem possible now seem to be effectively table stakes, is incredible. And so just because you did something three months ago doesn't mean that you wouldn't have an entirely new toolkit available to tackle those problems and do them even better.

Ryan Miller

Brilliant. So before we wrap things up, final remarks. How do people enter your world? Where can they go to learn more about Reflexivity and that tool, anything at all?

Jan Szilagyi

I mean, I  would just say that I think we're entering what I think is a change in how we do investment analysis, how we tackle financial data that I don't think we've seen in decades. Like it's hard to really compare to anything that we would have seen in our lifetimes. And so obviously, we're more than happy to talk about this and knowledge graphs and so on. And so you know, just if you even go to reflexivity.com, you'll see that we are extremely passionate about this. But I would say a much larger and overarching point to all of this is AI is here to stay. I think by now everybody knows it. I think the only decision that everybody needs to make is where and how are you going to start learning how to use it, how not to use it, when to avoid it, when to leverage it, and so on. But you can't ignore it.

Ryan Miller

Yeah, I love that. So just to summarize everything Jan and I spoke about number one, use knowledge graphs to help you identify impacts and sensitivities of investments. Number two, the value in AI comes less from the tool and more from the questions that you put inside of the tool. And then number three, just make sure you use AI in your fund where it makes sense to avoid gimmicks and use the right tools like Reflexivity that the megafunds also use. You do these things, and you too will be well on your way in your pursuit of Making Billions.