Odd Lots - How Hudson River Trading Actually Uses AI

Episode Date: October 31, 2025

Unfortunately, it doesn't seem as though you can get great stock picks just by going to ChatGPT and asking it to recommend some investments. And yet financial firms of all sorts — including trading firms — say they're increasingly using AI. But are the tools actually being deployed? And how do these tools differ from traditional machine learning or algorithmic approaches to trading, the likes of which have been used by quant firms for decades now. On this episode of the podcast, we speak with Iain Dunning, the head of AI research at Hudson River Trading, a major US market maker. We discuss the firm's attempts to use AI not just for more efficient trading, but also to make short-term predictions about price, which further gives its traders an edge. Dunning walks us through his work, his views on the main constraints facing the space (labor, power, chips, etc.) and how his work is both different and similar to what's happening at the major cutting edge research labs like ChatGPT.See omnystudio.com/listener for privacy information.

Bloomberg Audio Studios Podcasts Radio News Hello and welcome to another episode of the Odd Lots podcast. I'm Joe Wisenthal. And I'm Tracy Allaway. Tracy, I've always had this idea for the podcast or a thing that I've wanted to do conceptually with podcasts

is schedule every guest for two interviews. So you have the opening interview and you ask a bunch of questions and then it's oh God, I really wish I had followed up on that. I had more. I was just starting to sort of get my head are on this thing. Now I could have asked the good questions. And then, like, have the person come back next week. Also, the audience complains. I wish you would ask that.

And then fill in all those gaps that had been inspired by the previous conversation. I don't think it's a bad idea. I think it would double the number of episodes that we put out. Yeah. But sure, there are topics that come up, usually things that were just kind of new to. Yeah. And we're trying to learn about specifically technical things. And one of those has to be AI, right? AI. And also, you know, I really had a great time, I guess last month. We were in Chicago. Yeah. We talked to a bunch of different, it was like a trading related trip. We interviewed Don Wilson. We interviewed the head of the CME. We had some other chats. They're all about the world of trading. When it comes to trading, it's like, you know, we talk to long-term investors,

portfolio managers and Dowellas. We talked to some people on the hedge fund space who like maybe have a holding period of several weeks or whatever. I actually really want to learn more about the trading like these people who have like a holding time of one second or something like that. Because that's where a lot of the tech and a lot of the actual like action is. And how that world makes money and how they actually deploy technology is very interesting, but still something I don't have my handle on. Well, the practical application, right? And also the culture of AI on Wall Street.

I find that really interesting because I remember, I guess it was like more than a decade ago. But remember Lloyd Blankfine saying that Goldman Sachs is a technology company? Oh, yeah, yeah, yeah. And all these bank CEOs saying, we're going to install pink. ping-pong tables to get all the coders. And now I see ads at trading firms. And it's like, we have a data center full of B-200s or we have a data center full of G-300s. Come work for us. The only thing besides all their tech that I know is like every time you read a profile of any trading company, they're like, and they love to play backgammon. They love to play.

Like all the articles, the chess boards are out. They could be seen playing chess over lunch, etc. I get it. Okay, they like odds. They like games. They like whatever. Let's move the ball for it. Well, there's also the underlying theme of, is this all hype? Yeah, right. Because you do get the sense sometimes that companies are putting out press releases where they just mention AI to tick a box, to be seen to be doing something, and hope that their stock actually goes up. And because so much of this is proprietary and people kind of have an excuse not to go into detail about it,

sometimes you do get the feeling that people are just talking about it and not actually using it. Cynics, and I'm not saying this myself. I know you're not a cynic. Speaking of trading and technology, cynics would say that CMEs deal with Google to build a cloud to put trading on the cloud was hyped. That was a press release. People have said that. People have made that charge and they don't understand why.

You don't have to comment. You don't have to say anything further on that. I do have a comment, but I'll hold it for our guests. I'm just saying there is this world where people do press releases and cynics go. I don't really understand the point. Anyway, there's a very long wind up. Let's learn more about the world of trading. Let's learn more about AI and tech specifically.

What does it even mean to apply AI within the realm of trading? We are going to be speaking with Ian Dunning. He is the head of AI at Hudson River Trading. He was previously at DeepMind. So his trading and AI bona fides are about as good as it gets. You've established them. We've established that. Really the perfect guest to answer all our questions.

So, Ian, thank you so much for coming on the podcast. Yeah, I'm really happy to be here. I agree. is the mystique factor is kind of overblown, even if it's understandable why people embrace it sometimes. We're going to blow past the mystique. Let's start with some like really just like rudimentary questions. Just the first one is like Hudson River Trading as a company. Yeah.

How does it make money? Yeah. So we are a sort of quantitative, automated proprietary trading firm, which is a lot of words. But I guess the way I see it is we are a service provider to markets. Okay. The most clear example is. is market making. There is a sort of utility to the world

of being ready to buy yourself any product anytime, anywhere. And for us, that means stocks, futures, options, crypto, bonds. And if you could, say, build a magical machine to quote a price to buy or sell any instrument and you would want to be like the best possible price, like the tightest price.

People would trade with you. They would be happy because there's a counterparty for their trade and they get a kind of good price, like a low spread. And we're happy because we essentially pick up a penny in front of a steamroller. We are making sort of money from that spread, and we can pick up the pennies in front of a steamroller

if we have a really magical device, which tells us how everything should be worth. When the steamroller is coming. Yeah, it tells us when the steamroller is coming. And so I think that's kind of the very, very sophisticated sort of middleman in some sense. And the same way about Amazon is. Amazon doesn't make stuff,

but it's a very valuable, profitable company. It provides a service people who get value of. Same thing. We're moving stocks, bonds through time and space between different counterpodies and yeah. We will ask you about the steamroller in a few minutes. Sure. But before we do that, how does AI or the way you're using AI actually differ from the algorithmic

or quant trading of olds? Because I guess that one of the questions is, is this, you know, a sort of evolutionary change, you know, maybe a marginal improvement on what already exists, or is this something seismic and a step change, a big shift in the way trading actually works? Yeah, I mean, I don't want to over it. overstate ourselves in some sense because in the space, as you mentioned before, it's very opaque what sort of different firms of this class are doing. I can still speak to our own experience, which is we've been doing this type of trading for 20 plus years.

And much like everyone who was doing this, the way it kind of worked was you handcraft features. Based on human intuition, oh, I don't know, the order book looks imbalanced. There's more people wanting to buy themselves. The price is going to go up soon or something like that. And maybe you get a bunch of very smart people and they think very hard. It's almost like making a very fancy watch. You kind of artistically craft all these pieces. And then maybe you use relatively simple mathematical techniques like linear regression to combine those predictors.

And I've been going to conferences and things in recruiting for a long time. And even if today you just go on the internet, you'll people say things like, oh, that's all you can do in finance. For some reason, they'll say this. They'll say something like, oh, it's too noisy or markets are too non-stationary or things like this. And so that's all you can do. And I guess that belief isn't really backed up by anything, in my opinion, and like lived experience, I guess. And so we sort of viewed it more for a long time as well, there's everything that's happening in the world. And ideally you would put this into kind of like a machine that does not have any human biases.

I don't know how to trade stocks myself. Like I buy broad market ETF. What do I know? And so we, but if you could put all the machine data into a box and it kind of could turn all that data, it would find. things that you would never be able to do with this handcrafted thing. And we started doing that very early, relatively, and so 2014-2013 period. And over time, over the last of decade or so, much like in other contexts that are not finance, there has been sort of a hockey stick.

And you can measure it by the size of the models, the compute deployed. And over time, that way of modeling the markets initially was not like a hybrid with the traditional way, but actually kind of just like overtook it entirely. And so now our trading is entirely driven by this magical machine, it consumes over data. I kind of keep saying this magical machine that consumes over data for a reason, which is that this is how chat GPT is trained. It consumes all the data, all the internet. It's kind of scraped and connected into one place.

You train a model that kind of takes it all and something emergent comes from it. And that's why I'm kind of a bit leading, but that's why I'm talking about in the sense. And I think that is materially different from the like, I'm using my intuition of the markets to kind of construct a predictive model. So just to be clear, how much of usefulness of AI here is about execution and the fact that you can crunch a lot of data really quickly with hundreds or thousands of GPUs versus spotting sophisticated patterns or discrepancies that you can exploit? I think it's both.

I think one of the things that people sort of missed with a whole, like do a linear regression type thing, is when you really think about how much data there is in financial markets generated, And when I say data, I think it's important to think of it as every event that happens in Macs, not the sort of time series of prices, but like the actual low-level substrate. People are quoting, trading, retracting quotes. That, like, low-level stuff is Internet scale data set sizes. And one of a sort of bitter, lessen-y-type things of AI was like, you shouldn't think too hard about how to feature engineer this and pre-process that you should kind of throw it in to something, a form of computation that can kind of make use of Internet-scale data. In the 2010s, it was like computer vision. People used to make detectors for edges of images and things, and they would combine them.

And same thing, it's like, that was a good approach, but, you know, it's completely dominated by the idea of getting a very large number of GPUs and a kind of a pretty generic neural network form and powering through it. As for, like, how is it finding things that other methods could not? It's very hard to say. Our models are not very interpretable. And I think that's fine because, as Joe mentioned, our sort of trading style, and holding times, a bit of thought of as minutes, hours, maybe like a low single-digit days for the most part. And I guess in my mind, it's unreasonable to expect them to be interpretable because I don't

know, if I looked at the orderbook data for Tesla or something, am I really going to be better than random with the price of Tesla will be in a minute's time? And so I kind of think it like that. If you have something that's clearly superhuman already, what level of interpretability could you expect? It's very different right to normal AI, right? This gets into some areas that I'm very interested. But just to, like, establish what we're talking about.

Yeah. You're trading a stock, like a Tesla, like, a Nvidia, et cetera. With your magic machine. Magic machine. Now, we had another episode where we talked. Well, that was the money box. As a magic box. That's a different one.

With this AI machine, it is sort of arguably grown, right? It's sort of grown in a lab more than it is programmed. Much like a chatbot. I know it's a very different technology. Like, what is the price of? of Nvidia going to be tomorrow? Or what is the price of Nvidia going to be this afternoon? What you're saying is with your technology,

you have a better chance of getting that right. You actually might be able to make an informed prediction about the future in a way that you couldn't have done, say, 10 years ago. Yes. And that people who talked about this, they would come up with reasons, oh, the stock market, it's not like chess or go, and therefore you can't really do predictions the same way.

But what you're saying is that with these models, which are different than LLMs, there is some, at least on a short time scale, predictive capacity. Yes, I find this still, to mistake,

a little bit hard to believe. I think you get this kind of efficient market hypothesis stuff jumped into your head. It seems like someone's saying that it can predict their price of a stock in an hour. Your instinctual reaction is incredulity.

Like, just sounds like you're kind of bluffing or making it up. But no, these models can predict this. And I think it's the way to kind of reconcile the like, really, man,

like kind of instinct is that the predictions are very bad. in some sense. We don't normally talk about like accuracy, but I think the way to think about it is like the accuracy is like 50.1% type thing. Like they're only a little bit better than random. But I suppose an extra 1% like blows up your profits if you're doing it at scale. Doing it at scale, doing it enough times. And over time you kind of realize the biased coin flip. And as for why it might be possible to do this without kind of invoking magic. It's like markets are very beautiful interaction of like many different parties. all the different kind of utilities, risk preferences and things. And the only way you really see what people are doing is by the actions they take in markets. And you kind of, it's sucking up all that signal,

micro signal and extrapolating. The cynicism or the skepticism about the possibility of machines that could predict the price of stocks is a little strange, right? Because machines ingest data than whatever. Maybe they see a pattern more likely than not. This constellation of data means tomorrow will be green. Humans do this all the time.

what else do we have besides data, right? You have an analyst and they put out of Tesla or whatever. InVidio's going to go to $500 a share. How dare you insinuate? I'm not smarter than a computer, Joe. Like, all humans have this data and much less data. And yet humans are making predictions all the time. There's a whole industry of it.

So the idea that therefore was for some reason a computer couldn't do this with much more data analysts ever have. I understand why the cynicism comes off as, strange. I think some of the doubt stems from this idea that a lot of these models tend to be backward looking, right? And some of them occasionally are pretty bad at spotting or reacting to big regime breaks. And I guess the thinking, again, sometimes, is that maybe humans are more flexible, maybe more adaptive in their thinking, and they can kind of spot these big cultural shifts. How do you actually, I guess, prepare for those big pattern changes? Yeah, I was at HIT for

COVID and I thought that was kind of like the most... That was a big pattern break. That was a big pattern break. And things went totally fine. Actually, it was more of an engineering crisis in some ways. Stock market volumes exploded. And every system was just like screaming trying to keep up with a volume of activity. But in terms of the predictions, they stayed quite good. And I had to like reconcile us in my head as well. I guess it is a matter of like horizon and like how far in the future are we talking? Intraday, I think a lot of the price movement. is driven by just observing like the flows.

It's hard for us as humans to observe, but it's like the relative patterns of buyers and cells in the markets. And it's like, yes, during COVID, the volatility was massive and prices were moving up and down a lot. But they were growing up and down during, say, March 2020. And so these models, it was sort of out of domain for a human, but I don't think out of domain in some sense for the models. But I guess I also don't know how you would apply this thinking

if you were trying to make sort of month ahead predictions. I often get like people being like, oh, everyone knows hedge funds, which we're not a hedge fund, is like a flipping coins and it's some survivor bias thing. And, you know, I genuinely don't know about months out prediction stuff. That is not a data rich environment. Just by definition, there have been more days than months, right? So therefore, prediction on a day basis, you're offered a lot more data. Is that what you're saying? Yeah, rule of thumb is basically very useful.

It is a, and it extends all the way down to seconds. Yeah. And we see that empirically all the time. And so, yeah, I guess all the things I'm saying do have this caveat, but it does rely on the sudden of being a certain level of signal to noise. I definitely cannot make reasonable claims about the price of things in like a month, using the same kind of like AI hammer. I guess also to be specific,

I'm talking a lot about using market data to make these predictions. And that's because on the sort of intradate timescale, that is the most important thing. It's all about flows and things from back and forth. if you're thinking about things in a month's time scale, I think that's fundamentals. And can AI be used for that? I don't know, to be honest. And it's definitely outside my wheelhouse.

And I guess people have various opinions about that. And maybe some people very much would like to claim that they can. And others maybe don't. But it's definitely outside of my area of expertise. And I don't know. Wait, talk to us about the data that you're using or talk more, because this is another area where people tend to talk in PR speak sometimes. We have access to all this data, unusual data, alternative data,

and that's going to enable us to use AI better. What are you actually looking at? And what have you found, I guess, most useful? Well, I think the thing that I found most counterintuitive when I started was that when you're thinking about predicting the prices of anything, a minute, an hour out, by far the most useful thing is just market data. This is the market data feeds that you can buy from the exchanges for a pretty reasonable price. People often think this is some sort of like competitive moat. The data fees for these

exchanges are not particularly high. And in crypto, you know, where it's like a Wild West, but everyone can collect these feeds. And so that is the most useful raw ingredient. That is the most true expression of everyone's intense, right? They're going to the market. They're quoting the buying selling. That is the primary ingredient. People get kind of caught up on the whole like, I don't know, do you have a Twitter feed type of thing? And Bloomberg sells a Twitter feed through its data products. And yeah, buy that. By that. Every now and then, obviously something happens. news happens during market hours, it moves the price, disdicates the price. But if you really coldly rationalize that, that is a relatively infrequent thing compared to the overall

massive markets. So for thinking intraday, think these market data feeds. It's literally like a little event. Someone quoted that this price and this size. It's all anonymous. Market data feeds are anonymous. And so that is the roar stuff. And it is vast.

There are just millions and millions events per day per stock per future. when you get to the day, days, time scale, that's where the alternative data, quote-unquote, kind of really comes in, as in alternative to market data, the SEC filings, the news feeds, balance sheets, brokers, reports, things like this. That's where that comes in. And there's a vast sea of data offerings that people try and sell. That I think in that kind of situation, it's a very low shop environment you start getting into, and it can be hard to attribute the extra shop for each of these things. But in some sense, it's a lot of, also very democratized.

Maybe people collecting very secret datasets, but my inbox, and I'm not even the person in charge of buying these alternative datasets, is often full of people trying to sell me the latest alternative datasets. And I think a lot of them don't necessarily have much predictive value, but clearly it was a market burn.

What's the craziest one you've seen? Can you remember? I mean, people have definitely reacted very strongly to the Wall Street Betts era. And tried to kind of create a bunch of Reddit-extracted thing and go beyond just like raw captures of Reddit and trying to distill it into something. But, you know, I just, even just thinking about it, the meme stock thing is kind of was

talked about more after it happens than it happens before. And so like, I don't know. This is sort of a sideways question. You mentioned interpretability. And this gave me something I've been wondering about AI for a while, not even in the finance realm specifically. You're a deep mind, which, of course, produced a great Go player better than the greatest grandmaster in the world. I play chess. We know that chess and chess. are much better than any human. On the other hand, as far as I can tell, there is no good AI chess tutor.

So in other words, the chess crusher do, but I've never been able to get a thing where it's, okay, you did this move, but you know what? You're closing this rook file and down the line because it doesn't do that. The chess.com human talk is very rudimentary, et cetera. Can you talk a little bit about

why there are these problems where some version of AI or machine learning or whatever can do fantastically well, but then the actual explanation of what it's doing, which I think is kind of what interpretability is, can't articulate in a plain English why it's able to do what it does. I think it's just because these neural networks are, some sense, just like a big old blob of numbers. And what we're aiming to do when we're training these models is to almost like,

free ourselves from almost all structure. And they might learn things in a way that is nothing at all, like how we learn things. And so my best guess for like why it's hard is because they might be reasoning in some sense internally. And people use these words like reasoning. It kind of makes me wince. I've seen imagination and things used about neural networks. I don't know.

It's like kind of anthropomorphization of them. It's kind of dangerous because they are essentially processing things internally in this way that I think is inherently not like how we. we do, and that is my best sort of guess. There are some interesting counter-examples. One of my favorite sort of things, you know, Pasquilis, was Golden Gate Claude, which was Anthropic made that the model

basically get very interested in the Golden Gate Bridge. Every question they asked would come back to the Golden Gate Bridge. And so they're not completely impenetrable. Yeah. But it's clear that, like, it gets hard beyond a point to kind of map this back to how any way, like, we think. And it's very tempting to and exciting to, and especially

for like AI safety applications, which aren't really relevant to me so much, but I think it's very tempting to try. Yeah. No, it strikes me is that if you could solve that many jobs, you could actually make a lot of productivity gains, but I do think that's an important hurdle. When you're training your models,

so your models are different than large language models, et cetera, but what they have in common is this incredible amount of data, incredible amount of compute demand. How applicable if someone had worked on LLMs, would your training process be to them? How could they move from that? environment to yours, are there enough similarities in the basic notions and compute and requirements to train a model such as yours versus what people are doing at the major labs?

I would say now in 2025, absolutely. But I would not have said that in 2020. And this is something that kind of caught me by surprise, having done this for a while now, is that our problems are kind of defined by long, sequential strings of information in some sense and extrapolating from that. If I think back to the past of AI, it was like, is this a hot dog or not? It was kind of like the image classifier, you know, test. Then there was some stuff with audio and things. It was a little bit more familiar. Robotics.

But when we got to this sort of LLM error, it got very interesting because suddenly the problems were very similar in that you have, you want to think back over like long histories, long contexts. Okay, sounds good. You've got a lot of data and you want to turn through it in as efficient way as possible. you also have to serve this model. It's a more than has to run in a relatively reasonable speed, especially for the LLM places.

There are a million people typing into chatGBT.com and they want to hear a response in a relatively prompt manner. Of course, for us, also, the models have to make their predictions in a prompt manner of ways the predictions aren't useful. So all these things mean that our sort of way of thinking about it has become very similar to the frontier LLM things. It's just a very different modality.

We're operating on, I guess, primarily text, And we're operating on this fileless interpretable, but still sequential stream of tokens, except our tokens are market events. And so it's a lot of fun because, you know, in terms of like the research that is still published, you can kind of look at it for inspiration and draw comparisons. But it's also very much its own problem, which just kind of keeps me interested every day because it's like its own unique thing. But it's different. I want to go back to the point you made about data and I guess democratizing finance in many ways. Maybe this is a weird question, but I'm thinking back to the 2010s. And we used to talk about the big investment banks as flow monsters.

They see all these orders. They get all these orders. They see all the flow. And that allows them to optimize on funding costs and other expenses. Is the idea that data and AI can kind of replicate that advantage so that everyone, or not everyone, but Hudson at least, becomes its own little flow monster? Yeah. I think there's still some trends and markets that worry me a little bit in terms of, I guess

our platonic ideal market structure is probably like everyone trades on exchange in a centralized place. But that is not really how things seem to be going. And there's a huge amount of like off exchange, dark, quasi-dark volume. And I think there is still a lot of quantities of a trading world where like being in the room is kind of like this big advantage. And this is a very much anti-AI play in some sense. It's data is hidden, the flow data is hidden, and it's not something that you can feed into machines

because it's very sparse amounts of it. And so that's kind of an interesting trend. A lot of us did get sales to get reported in a centralized place later, but it's not prompt enough to be useful. And so to say that AI thrives on data, this is in some sense like an issue for a long run. You need to kind of be in the rooms where the sort of trading is happening. I'm glad you brought that up.

because that's specifically what I'm curious about from the sort of physical infrastructure side. Like if I have a query at Chad GPT, I don't care if the model is like trained in like Abilene, Texas or wherever it gets back to me and whatever. But I know that for high frequency trading, at least on the execution side, there are certain parts that you want to be literally co-located and you want to have the shortest possible wire. And however short it is, ideally you'd like it to be. shorter. Can you talk about the differences and similarities between essentially your physical hardware stack versus what would be required at a large language model frontier lab? Yeah. I think at a bulk level, it was actually some pretty similar things. So I often think about it as like latency and throughput. Latency being the time to react and then throughput

kind of like how much thinking you can do in a certain period of time. And so you're right that like this space demands like low latency. Early in the 2010, so it was a sort of Flashboy's book and perception where it was like really kind of about arbitraging latency. I'm happy to report that in some sense all the latency has been arbitraged for the most part. There's no more edge in shortening the wire. This is probably like a little bit, but it's relatively small. And like I think if you look at the big quant trading firms, the need to like really make the wires as short as they possibly can is done or are no longer relevant, which is great because I find out stuff pretty

boring personally. I think about it more as like for a given kind of like speed of response, you should be the smartest person. So it's like this curve. If you're going to take a second to come up of your trading decision, it'd be a really, really good decision. And then it doesn't kind of matter that it took a second. And if you're going to take a micro second, well, A, you probably can't do too much in a microsecond. But you know, it better still be the best response than a microsecond. And so you could be a little worse. You could be a little worse. And in the second. Yeah, for sure. And so essentially for our training, we use the cloud. We have our own training data centers that we've built ourselves.

That is basically the same, although much, much smaller scale. The scale of Googles and things, I don't know, it blows my mind the spending on stuff like this. We are, I think, big, if you're not comparing us to Google or meta, but that's not like billions of dollars. So training is kind of the same. Inference, we need to put the devices close to the exchanges, and we need to think very hard about the power. power usage and the latency, but we have hardware teams. We make our own FPGAs. We make our own chips and we use off-the-shelf GPUs.

And what we try and do is we try and make sure that for any given sort of speed or response, we're making the smartest possible decision you can. So you can kind of... Field program mobile gate array. Oh, there you go. There's a, sorry. Catching. A.

Katsy. Yeah. Basically, all these different devices have different latencies and throughput. GPUs have very high throughput. They are, that's what they're useful for, right? And so, but the problem with markets is they're kind of like narrow. The amount of traffic flowing into these, like, LMs, from everyone typing into their web browsers is massive. And they do all sorts of clever things to kind of batch up requests and process some things.

We don't really have that luxury, really. Like, the markets are going to happen. At the speed they happen, we can't kind of like duck out for a while and catch up. We kind of need to stay in the game. So we have always sort of interesting design challenges around how do we use GPUs, which are relatively high latency. They take a while to give back a result, but they can process the whole stock market on one GPU type of thing versus the fast response. And so we have whole teams dedicated to thinking about, okay, I've got this, like, I've got this, like, intelligent blob.

How do I get ounces out of it in different ways at different speeds? And that, I think, is where a lot of us smarts are going in this world these days, rather than the, like, how do I make sure my microwave towers are, like, slightly better align somewhere in, like, rural Pennsylvania, which is a cool challenge in its own right, but it's done, I think. I think people have found the straightest line from New Jersey to Chicago. Joe brought up some of the cynicism around CMU's cloud deal with Google. And this came up.

Speaking of a specific cynic who went on the record in one of our episodes, Don Wilson basically made the argument that matching on a cloud doesn't necessarily make sense because you might put in two orders and you're not really sure which order gets filled first. I guess you're kind of back in that black box environment or maybe it's a latency issue. I don't know. Is that a problem that you're seeing? It's something that I worry about. Our general philosophy is market should be very transparent and as fair as possible.

So like equalizing access is a good thing in terms of participants shouldn't be at all like basically pull weird tricks to be faster. On the other hand, I think you want reliability. So like this concept of like orders arriving at different times and being filled in different orders just doesn't seem like a very sensible way to run a market. It's something that requires a lot of effort to engineer around and it's just a good market design to have. It is a very widespread though in existing exchanges across the world. We've traded in a vast number of countries. And some of the exchanges have such amazing hardware that like if two orders are sent within like a nanosecond of each other, this exchange will never process them in the wrong order. Even though it's 100 different network ports and they're all connected, they have this amazing time stamping stuff.

On the other hand, you might have like a crypto exchange where it kind of feels like a kid learned JavaScript and set up a website and you're kind of like. You send an order and you may not be confirmed that they even received it. And then you kind of have to refresh your account balance page like five minutes later to see if there's money in it or not. And we kind of will take, we'll deal with it as it is. But certainly we have a preference for kind of equalized access, but sort of predictable outcomes. And I think that kind of leads to like people spending effort. I think it's not a necessarily very great thing for society for people to be like stressing very hard about Y-Land. Yeah, no, probably. I'm glad.

I'm glad that you report that we've moved on a little bit since then. Where are your constraints? You know, when you talk to LLM people, there's debates about, right? Is it electricity? Is that the big constraint? Is it there just aren't enough GPUs? Is it talent? Is it whatever?

When you think about where you are now versus the optimal version of where, or is it, I mean, data is the other big one because there's all this concern that LLMs are going to run out of training data, et cetera. Where is the big constraint for you that you feel like you're solving for right now? I think in terms of like really long-term strategic planning, electricity is like quite clearly a very binding consideration. When we think about spitting up new like GPU-based training data centers, it really feels like, is there electricity, like finding a piece of land to put a building in. There's a lot of land. Yeah. The electricity negotiations.

And that's an issue at HRT. Even for us, you know, because we have a sort of hybrid mix of using cloud providers and building our own data centers. And, yeah, the negotiations and thinking about power constraints, we have an existing data center in a very cold place. And we want to make it bigger. And the data center people are fantastic to work with. But they're saying, like, well, we need to go talk to, like, the power grid and negotiate this next trunch and so on. And it's just, it often feels like that is the bottleneck. And on the terms of a GPU availability, it definitely was a crunch at some point in the past.

But I don't feel like that is. It's better now. It's not as competitive. stock market is writing on. Say a little bit more about how you perceive the GPUs. I think, I think if we ask for GPUs, we will get them delivered in a prompt manner. Not necessarily like next day, but I don't feel like that is the thing that we have a long pole and spinning up more. When was the, when was the worst of the crunch? I guess 2023, late 2023 felt pretty bad. I was, I guess that was like the Nvidia Hopper generation. And I saw some number in Bloomberg yesterday that

I think there was like in video conference yesterday, and I said something like, it was like one million hopper class GPUs have been made, but already like four million Blackwell class GPUs being made. So I think there's been a ramp up of supply, but I don't think they're also sitting on unsold inventory either. I think it is being consumed. But yeah, in terms of like, what is the hard thing? I think electricity, and I'm this insane, as a very millennial person, I guess climate change was a big thing growing up in college, a lot of discussion about climate change. to see people spinning up data centers very fast by basically buying as many gas turbines as they can and putting them outside. I'm like, whoa, like, what are we doing?

It's wild, but that's like the only way to get electricity promptly. You just have to throw gas turbines outside the building and turn them on. It's pretty radical stuff. And I don't know how all the numbers that people are talking about for future data center expansion kind of math out because you just back of the envelope the power usage and things. And I know that Sam Boltman's in the world I've thought about this, I've talked about this. Oh, we need to be generating this much

new power generation per unit time. But there's such daunting numbers. I just don't know how that is all going to work out. But yeah, even for us, in the grand scheme of things, like a much smaller player in terms of power consumption, we think in terms of like tens of megawatts and not gigawatts, which is more than most towns and cities and things, but still.

And but we find it like a challenge to find electricity at a reasonable price. On this note, can you talk to us a little bit more about where competitive advantage actually comes from in this space? Because if the GPU crunch is somewhat solved and if latency isn't as big an issue as it used to be, where are people actually getting their edge from? Right. I mean, people talent is one of your ever things.

You asked a set a constraint. It is a very competitive people market. We're essentially asking for people to know a lot of things, be both good researchers and good engineers because, I don't know, in this AI era of a distinction, it's pretty blurry. It's not something you can just whiteboard and then the coding is a little bit afterwards.

Any kind of research idea you have is intimately connected to how you implement it. So that's already like a tough ask. So people are constrained, people that we like I want to find and we pay well for those people as a result and it is competitive. But I think the more subtle edge

is almost like putting it all together. do you have people who can, like an engineering team that can collect all the data, record it, make it available to the GPU training data center. This is like many, I guess it's petabyte scale data sort of sets. And just storing that much data, streaming it from wherever it stored to wherever in the world the training data center is reliably, these training runs are very expensive. And then once you've got that model serving it, so it kind of sounds to do everything. And maybe that's kind of like a lame answer.

It really is. I think you need to be just optimizing the whole stack. And so my team is like the AI team. And so what it really means in practice is we're focused on training the models, which is an important but not sufficient part of a whole stack because we would be kind of dead in the water without the teams at HIT who think about how to actually kind of get the data and things to be systems and then the decisions out to the markets and keep up when things get busy, all these things.

So when I think about our competitors, I think, think there is a benefit to scale. I can't imagine how you would start a new company like H.R.T. in the year 2025 because of the huge initial lift to kind of build enough engineering scale to achieve this sort of thing. And so I think our sort of peer companies also have invested very heavily in engineering and will continue to do so. And there was an article in the FT like a little like a week or two ago about how firms like HIT are kind of extending themselves more into slower trading. And there are firms that are kind of, you know, those slower firms, it's trying to kind of go faster.

Yeah, I was just going to ask about, just like on the prediction standpoint, okay, maybe you could predict what's, with some reasonable confidence, what's going to happen in the next hour, sometimes if you're lucky, maybe a day, like maybe a month that's just ridiculous. But do you, in your work, is that horizon, has it broadened? It is. Yeah, I think one of the things, for people who are aware of HIT, even at all, I think the sole perception is sort of a pre-2020,

perception of we are purely high-frequency trading firm, but we would say we are both high-frequency and medium-frequency trading firm. And it's like a big part of our business. One way to think about it, I think is that if I really have a view on what a stock should be in like five days time, let's say I want to buy that stock. I'm going to acquire that stock over time. And maybe it's what's the best time to buy that stock over the five-day period? Well, I have a model that tells me that the best price is an hour. So maybe the shorter-term model should inform the longer-term trade and cascading all the way down. When you're doing this sort of slightly longer term or slightly slower frequency trading,

is the fundamental job still the same, which is you're in the liquidity provision service business just of or longer, you want to hold that warehousing? Or does it some, because when I think of a fund, when I think of a hedge fund, I certainly don't think of, maybe to some extent some of their strategies might be sort of liquidity provision, but it's more directional? Is it still that? Or is the fundamental reason why you make money, the service you provide, does it? it change, by definition, change over that horizon? I think the market making service provision does

break down. I think it stretches the analogy too far. I think you have to think of it as like liquidity taking, which somehow seems more like aggressive or something. Yeah. But the, we're trading against orders resting on the book. Someone was like, I want to sell this stock and we're like, we will buy it from you because we think that in the long run it'll be worth doing it. And so we do cross the spread and we do pay this transaction costs. Sometimes, you know, you can also kind of acquire position by market making, but with a tilt. So really at the longer horizons, I think the sort of market-making service analogy does break down. But in some sense, there's always a counterparty and they wanted to trade for a reason.

And I think a mental model that, I don't know, you tell me if this sounds like too, too wishy-wushy, but... I love a mental model. Yeah. You mentioned Go. Yeah, yeah. So the thing about those is that there are zero-sum games. There's only one winner. It's truly like a no...

Like someone's unhappy, someone who's maybe equally unhappy. Yeah. Plus one minus one. I think the reason that trading works is because it is in some sense positive sum. You know, money is conserved, and I guess a little fee goes to the exchange. So in some sense, money is at that moment of a trade is actually negative a little. But utility, people's general happiness. I don't know, my paycheck goes into my 401k provider and it buys some ETFs.

I'm relatively like insensitive to how exactly that happens. I just, I'm not going to look at it for another 40 years, right? Don't lie. I try not to look at it. especially lately. But yeah, like the utility, my utility is a very long horizon. And so someone sells it to me like at one cent different. I don't really care.

But like the person who made the sense happy and I'm happy because I got good liquidity, it didn't cross a huge spread. So that is kind of why I think it all kind of makes sense and why people are trading together. But it's also why like thinking about markets like an alpha go sense doesn't make sense because it kind of doesn't really apply. If you thought of markets as HIT and all our competitors all kind of in some sort of like deathmatch, who's the smartest, who's trying to pick each other up, then, well, markets would be kind of like this giant standoff, or no one would be trading, everyone would be kind of be like waiting.

But obviously markets are very vibrant. I think it's because even when we were crossing the spread, it's because we're crossing the spread against somebody who wanted to sell for whatever reason. If we were right, I guess on five days' time, they might be like less happy. But maybe they weren't actually. Maybe they were just like hedging a position. They don't care what the stock's prices in five days. They just wanted to like hedge their position and we traded with them. So that's the way I tell Rick and Silas in my head. They can still be like a sort of service provision.

We make money only because someone else wants to trade. If no one was trading, we wouldn't exist. Right. And different market participants with different motivations and goals and aims. I want to go back to the talent question for a second. And I get the sense that engineers like open source and they like contributing to the research ecosystem on AI. And then I get the sense that trading firms probably do not like open source. And they're much more into protecting their proprietary.

models or data or whatever, how does a company like HRT, how do you actually balance that tension? Yeah, I mean, this is also like a sort of really honest answer. Annette, many years ago, this was a relative comparative disadvantage for us for recruiting. Some, we often have conversations with maybe, especially PhDs who are graduating, and they would say, like, well, I can go to Google and I can still publish my research. And that kind of gives me optionality. People will know who I am. If I go into an HRT or actually like firm, I essentially go behind this veil.

and I never emerge. And people just have to kind of take it on faith. I did smart things for many years. And I would have basically no strong counter argument, apart from the fact that actually writing papers is kind of overrated. I've been there, I've done that. When you get older, you will not care.

Now, though, there's this interesting situation where this golden era may be of, like, being out of work at a big tech company and be paid for public research is very much over. The papers that do come out of the big AI labs are essentially kind of either very stale or not important. And if you're working on the most important

cutting-edge things, you can't share what you're doing and it's very secretive. So in some sense, the problem solved itself a little bit for me and people now recognize that IP should be protected. I've even seen some of the sort of AI lab

people think out a lot about non-competes in public, thinking, tweeting about non-competes and things, which is like an amazing ton of events because I feel like... That was very anathetical to all. Right. I mean, they're like literally,

effectively banned in the state of California. And I think people, were almost like proud of this fact and which also kind of hold it against a new york sort of trading world being like oh look at these people with their non-confeiting things and then someone comes along and pays a hundred million dollars or whatever for like your researchers and a lot of that money is being paid for talent but it's also in some sense paying for intellectual property yeah and like those people know how the soup is made and they are not writing it down and not committing any just like explicit sort of IP theft, but if you hire five people who've been making the soup.

Process knowledge. They know a lot of process knowledge. And you might suddenly feel a little differently about protecting that we spend a lot of time training our employees. It takes a long time for them to be productive. In some sense, it would be a shame if people could just take that knowledge and immediately leave. And so, yeah. Just going back to the steamroller. I promised.

I promised we would. When I hear AI in trading or I know people are very excited about agent-based AI nowadays, part of me thinks back to one of the more amusing events in financial history, which is, Joe, I'm sure you remember the time that one of night capitals, algos went rogue. Yeah, many people would not find that to be an amusing event at all, the worst nightmare possible. Yeah, for them. The peanut gallery. Right, right. Schadenfreude. So this Algo went rogue and bought like $7 billion worth of stocks.

Yeah, exactly. Exactly. What are the guardrails that you put in place to avoid the destiny of night capital? So every training cycle, we have a talk about the nightmare worth of a K, and we have multiple X, the night employees at H.R.T., as you might expect, just from a lineage of a successful trading firm that ended in a kind of unhappy way. And we have many people who are at night. The story is crazy. A successful training firm that ended in about 15 minutes. Yeah. Yeah. So it's fair to say that that stuff haunts us. We try and take as many lessons away from that as possible, defense and layers.

So I think one of the things that I like to emphasize with the AI stuff in particular is that it is not like there is some neural network directly sending orders to NISI. It is, in some sense, providing a plan, and then traditional human, heavily audited, risk-checked layers take the actions. And that's just kind of how it has to be. And so for us, we are kind of on an operational day-to-day basis. It's just many, many layers of sanity checking throughout the day. And then at a sort of high level, it's a very careful process, including processes to

specifically avoid the KCG type scenario of how are you even releasing new versions and what pre-release checks do you run and audits? And we even during the day, we have some, I guess you could call them like sanity checks of the neural networks to make sure that they are producing the values that we expected they would be producing. And those sort of checking processes are kind of a little bit behind because they kind of can't keep up with the like flow but like for enough to kind of just again like every think of a numeric stability of the model sane and things it's not it's not about losing money

or making money in jrdavi's not like oh like risk in the kind of financial sense it's like operational risk but paranoia is deep and that's probably something that's still very different i think from this market from the sort of other AI world which i guess anything goes and like failure rates to kind of just priced in yeah but yeah you could you could imagine just ruining everything. And I guess we worry about losing money, but I think we worry more about taking an action that a regulator would not want us to do. Because if you lose that trust of regulators, you lose it for a very long time. We trade in a lot of markets, and we pay very close attention and have deep respect to the regulators and their decisions in all those markets. And the rules

are sometimes very complex. And man, do we watch that stuff like a hawk? Because you don't be kicked out of a country for making an operational error. And this is a very low tolerance culture. from regulators in terms of making mistakes. So we stress it a lot. And I think we should because it's, you know, like the profit you make in 10 years by still being in the game versus move fast and break things. It's not move fast and break things.

But you still want to move fast. I have like a million more questions. But for the sake of time, I'll just ask one more. And I don't know even know whether it's something you're in a position, great position to answer about. It's something I actually wanted to do an entire episode about it at some point. But as you would characterize it, what happens in the second after a jobs report is released?

And what I'm talking about specifically is numbers either flash on a screen or a piece of a text appears on a website and markets move around a lot, all that. And there's people then suddenly it's actually the jobs report was good. And if you actually look at the wage number and then the stocks. But in that instant, in that first micro second after the release, markets are already moving. I know. Certainly before any human has had a chance to read the thing or form of view. So what I assume is that there's training on here is the text and here are the things and whatever. But as you would put it, or from the perspective of HRT, what happens in the millisecond after an event?

Yeah. So we have like a Bloomberg headlines feed that is like pretty low latency. And if it's like an important article, it's like a star in the feed, things like this, right? But you can do everything from having kind of a handcrafted logic to look for keywords through to putting it through like an AI model. One of the things that I like still can't kind of wrap my head around is I guess without saying specific company names, there are options trading firms that have thousands of people that are essentially cyborg trading options. They have maybe 10 people trading like the options for a single big stock like Nvidia say. And they are humans staring at the feeds for these things and

clicking buttons, and they have user interfaces that have set up for them to hit the green button or the red button, essentially, very fast. It's weird. We actually want a monkey on a computer. For a hackathon, we got a PlayStation controller and gave people a chance to try and practice reacting to events very fast. It's really tough,

but it's a learnable skill. I think, in an efficient market sense, this should be AIable. It is challenging, though, because if you're imagined to kind of plumbing it into chat, GBT, it would be too slow. Like, the latency would probably be sufficiently high. I mean, it's not that fast, right? It's fast for any normal day-to-day thing,

but for markets, it's kind of slow. Also, and this is like a very interesting research challenge, is like you can't literally use chat GBT to back-test anything. It knows every Jerome Powell speech and knows what happened afterwards because it's trained on the whole internet. So how do you really get confidence that for the next Federal Reserve speech is going to do the right thing? Traditionally in finance, you back-test things to see how it had done in the past. But if it's, there's a But in this case, it's all kind of in sample. Like, it's seen it all before. Yeah.

And I've seen academic finance papers if they try and, like, grapple with this. And they say it still works and they try and account for this. But I know, this stuff is really that smart. Yeah. The whole kind of thesis is that it's memorized everything that's being trained on. So why would it be reliable? And so whenever you see someone's being like, oh, I ran every Federal Reserve speech through Jat GBT and it got it right like nine out of ten times, it's like only nine out of ten times.

Like, why not 100 percent? So I do find that I do think it is interesting. There are how many humans are still involved on relatively high-speed trading. There are a lot of people still doing this instead of niche products, and it's presumably because it's very hard to integrate all the information. It's AGI 20, I don't know, 2028, 2030. I don't know. There's still a lot of humans trading stock and options.

And so like I don't know how to reconcile that, but I think about that when I read. Ian Dunning. That was fantastic. There really are like hours more of conversation. Sure. Are we going to have you back on next week? We'll have you back next week's episode. But no, that was great.

Thank you for having me. Really appreciate it. Yeah, a pleasure. Thank you. Tracy, I thought that was really great. I like this idea of this sort of anti-cinicism because you do hear a lot of people say, oh, no, like AI could solve things like chess or whatever.

But the stock market is fundamentally different. And I've never been totally satisfied with some of the theories for why. And like, I get, stocks are not like, necessarily a solvable problem in quite the same way. But humans make money on the market by matching patterns. Why can't smart silicon brains do the same thing? Well, there's also history now. We have many years of HFT trading and algorithmically driven trading where people have made a lot of money. So it seems to be working. The light bulb moment for me was where Ian talked about the

time frame and the importance of the time frame. And I think that's really the key in many ways. It's adapting what you're doing with AI to the data that's available. And the data on markets, most of it is going to be very short term and more seconds than minutes, more minutes than days, et cetera, et cetera. And a lot of the data is also biased to immediacy versus past analysis, which he spoke about as well. It is always funny in finance people. They're like, oh, 17 out of 19 times there's been this death cross of the S&P 500 stocks went down. It's like any serious data scientists would spit at that sample.

It's like beyond a joke level to talk about a sample size of 19. Yeah, but putting death cross in a headline is so tempting. That's true. You cannot advise a journalist never pass up a chance to put a death cross. I was glad to hear. A few things are interesting. One is, I was glad to hear that the wire length problem is no longer a thing. It's not just this race to get closer to the exchange.

That was kind of boring when people were talking about the Cold War and HFT and all of that. It's interesting that the GPU market is eased versus where it may have been a couple years ago. And it's interesting that even at a scale of a trading shop, that electricity is proving to be a main constraint, which does raise questions about, are we just going to hit up against a wall given some of the AI plans just that so many people are banking on for the chatbots. Yeah. I thought also, I guess, the cultural shift in some of the labs. Yeah.

It was really interesting, this idea that they've become more proprietary and perhaps more mysterious in some ways rather than the trading firms becoming more open. Yeah. Lots of great conversation, answered some questions. Yeah. Plenty more to go out. That was helpful. Yeah.

And I'm sure we'll talk to him again. Maybe not next week, but soonish. Maybe next year. All right. Shall we leave it there? Let's leave it there. This has been another episode of the Odd Lots podcast. I'm Tracy Alloway.

You can follow me at Tracy Allaway. And I'm Joe Wisenthal. You can follow me at the stalwart. Follow our guest, Ian Dunning. He's at Ian Dunning. Follow our producers, Carmen Rodriguez, at Carmen Armin. Dashel Bennett at Dashbot and Kel Brooks at Kel Brooks. For more Oddlots content, go to Bloomberg.com slash Oddlots for the daily newsletter and all of our episodes.

And you can chat about all of these topics 24-7 in our Discord. Discord.g. and if you enjoy odd lots if you like it when we dive into how companies are actually using AI then please leave us a positive review on your favorite podcast platform and remember if you are a Bloomberg subscriber you can listen to all of our episodes absolutely ad-free all you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there thanks for listening