How AI Gives You a Massive Edge on Prediction Markets

0:01 All right. Inside prediction markets we're starting with the three big stories in prediction markets this week. I'll do my lead off which is the new Kalshi Pro terminal that came out. It's pretty cool. Better UI for a sophisticated trader I would say. Still short of what we do internally. If I'm being honest. Probably. It's kind of interesting to see something that sophisticated come out.

0:33 It's so hard as a consumer business to maintain multiple UI, UX, and everybody's already started splitting because they already have the like, oh, here's the base. Any PM experience, but do you want a more sports specific experience? What's your odds format, etc.? It really is a nightmare to maintain these things and also open API's You would think that you could just kind of offload it to a third party to make the best souped up trading platform, like what happened with Betfair?

1:03 I don't know. Did you take a look at it yet? I saw a little bit. It looks like they're trying to replicate your old stomping grounds in Bloomberg. Yeah, it's the Bloomberg kind of look. Yeah. And I, I wonder if they actually care about this or just like trying to go with, hey, look, we are not just traders or like, we're not just like, this is not gambling, this is trading, you know, kind of thing. The second biggest thing I saw was the Bryce Harper VIP when there was a clear the real story.

1:31 This is a clear hit piece on Bryce. and Bryce Harper. Okay. Do you mean that? Yes. Okay. I genuinely like I saw the headline and it was so misleading. If they actually see it. Like, I think what happened was Bryce Harper got a cameo request. He was charging 900 bucks a cameo, which is like a I think it's like a minute. So for $900, $900 for like a minute of his time, maybe takes him three minutes to record or something like that, and he just read a message.

2:05 And what had happened is that FanDuel host used that message to give it to a problem gambler, and that who ended up losing almost a million and a half or $2 million. I think they're lost 2 million total. And they were down a million and a half the time they got the message. Yeah, yeah. And for some reason, people painted Bryce Harper as a bad guy and not the FanDuel VIP host as the bad guy, which is just like my mind host.

2:33 Definitely. That's like a sleazy job. Yeah, but but it's like, why are you attacking price Harper? Like, what did he like? You can criticize him for, like, being super rich and needing to do cameo fair criticism. But like, he read the message, like, what do you want him to do? He can't vet everything of where it's going. I was going to be used for. Yeah, it said FanDuel in the message, but it could be a prank on one of his friends. Like, you don't know what people are.

3:00 So I did some boots on the ground research on Bryce Harper's page. First of all, you can still go get a cameo from Bryce Harper. I mean, I don't know if he's actually fulfilling that. I'll give you one for love. I kind of wanted to get one. But there is a there's a button for corporate cameos, which I suspect is in part for this. I bet a they're charging businesses more, and b it's like you're not supposed to be using these videos for commercial purposes, right?

3:28 It's supposed to be a thing you get for your dad on his birthday. And, and so my just scrolling through the UI and for, for this kind of reason. Right. So what I could see looking through it is if you're actually getting a video that you're using to like, promote a product or do something like this, you're supposed to go through a different work stream on cameo, which if FanDuel did, and it sounds like FanDuel did have some kind of corporate relationship with cameo.

3:56 But if if there is this relationship, then it is certainly the case that the league knew what this is about and the agency knew what this about. Now, look, I think all this might have been inked in the early days of passport repeal. We're kind of nobody. This is 2024. People were just. Yeah, but I mean, when when these contracts get done right, I'm making excuses for them. I think people certainly he doesn't want to record that video today.

4:21 Or but the alternative possibility is that some rep just like I think $900, like a regular user, wrote what he wanted and hoped he would read it and got away with it. I think that's exactly what happened. Wow. And I mean, trust me, I know a lot about these VIP hosts, their animals, terrible, terrible people. And they do some very, very shady stuff and does not shock me at all. That should be the focus of how evil the VIP in the industry really is in terms of gambling and stuff like that.

4:54 And I don't think Bryce Harper deserves much. I mean, the only blame me deserves is, why are you doing cannabis for 900 bucks? Don't you have enough money, buddy? Like, if you do cameos like 1 in 10 cameos, is you saying something you're really not supposed to say? That's part of the whole project I did. It's funny, I did get a cameo from my my dad first, but it was like Wayne Bennett, Jack's legend, and it was like, okay, it was it was Louis cameo before I won't say who, but I have it on good authority that a not insignificant amount of time and effort for the legal departments of OSB is spent getting VIP hosts out of jail, because they're constantly brushing up against the law.

5:32 Yeah, that that tracks. All right. Story three yeah. So she's getting into another legal battle with New York. You know, I saw the governor and the AG were, like, going and tweeting at how bad couch is. This is clearly what it's going on. They're trying to protect the corporate interests and FanDuel and DraftKings and whatnot. And the same thing is happening in all of the states. I'm not really shocked to see it. I think all it really confirms is this is just going to go to the Supreme Court, and eventually all states are going to win or all states are going to lose.

6:09 And these individual state battles are kind of irrelevant. It would be annoying if we lost New York in the meantime. Oh, certainly. Certainly. But that hasn't happened yet. Michigan is down for Kelsey. Nevada is down. Yeah, yeah. And when Michigan's down algos 12th I mean they have to pay 500 K a day. Maybe it's worth it to pay 500 K a day. I thought people said that you already couldn't trade in Michigan that had been implemented. No.

6:35 It's obvious 12 I'm pretty sure I saw it. Maybe there were fake screenshots. I don't know. It doesn't matter. Yeah. What? Would you go to? Michigan? Yeah. All right, I'm going to put you in the hot seat because you had a fun week. I did. So I'm in the object odds hot seat. I was at AP summit. Advantage player summit is what that stands for, which is a a retreat this year in Canada. I went last year with Rufus and Rufus Peabody and we talked about it.

7:03 Is there another Rufus? That is it. AP there's a guy named Rufus Peabody. Dog name? And so this is my second year. There is a contingent of sports and prediction market people there now. There was a roundtable on prediction markets. Every year. They closed with a series of presentations and roundtables, although it's, I think, gotten harder for them to staff them because nobody wants to leak edges. Classic.

7:34 Yeah. Yeah, I know, and I learned a lot. It it's always eye opening to see people who are applying a similar way of thinking to different problems. That being said, I think the first thing I wanted to bring up was AI and how AI applies to what we do, because I talked to John Reader, who hosts Risk of Ruin, probably, in my opinion, the best podcast out there. Very good podcast in general about edges.

8:06 I would say about like gambling, but also kind of novel ways of gaining an edge that he does really high production value approach on. And he's, it's funny, he's doing something with real estate now, like I think he talks about on the show, but he's he's built a model to value parcels of land relative to how quickly and efficiently he can get the permits to build on them. And then he does the permitting and he sells them to developers.

8:34 And and he's describing it like he was handicapping baseball or something. And I'm like, John, this is just legitimate business like. And there were a few people who had these things going on, like, like, oh, you buy a car, but you, you know, you lease it out to somebody to Uber and you can claim the depreciation like, yeah, this is just people. AP gamblers discover businesses have profit margins. Yeah. Right, right. And yeah. Yeah. So I won't even get into a lot of them.

9:02 Had never even heard of some of the big fraud scandals that I was describing, like, you know, and with the disability and taking care of your relatives and whatnot in New York, I'm like, oh, you guys are peanuts compared to what these people are doing. But anyways, okay, back to AI. So I talked to John about what we're doing. I think he found it interesting and he tweeted about it and there was some conversation, so I thought I'd discuss it for a few minutes here and maybe get some of what you're doing as well on on your side of the aisle.

9:31 In order in our company, every single engineer is on a Claude max plan. Of course, we have guys who like ChatGPT and have asked to use it. And I think even because currently today. And this changes day by day. The latest anthropic model and the latest chat model feel kind of neck and neck. But the anthropic model is kind of handicapped because they've put these boundaries around it, because they can't stop putting their foot in their mouth with the US government.

10:06 So I feel like there's some desire from our guys to maybe be able to switch back and forth. But I also feel like maintaining a single ecosystem is really helpful. Yeah, I'm more of a fan than you. Claude for code ChatGPT if you need to edit like an email or you need to edit like something you're writing and then I for research for like easy, quick research projects, I like to use perplexity. But but I think Claude is just fine for research as well.

10:37 It depends on how you build them out. Like, to me, Claude, once you start dumping context into it. Right? Like and really I'm talking about. So for us, what I think becomes a big edge is if you have, you have one big shared repo of code and all your Claude docs live there, and you have a policy that anytime somebody build something or accomplishes something, like you have to update with documentation around how that works and skills and actions that whatever level of your choice can run, right.

11:06 So it's not just like, oh, I did this thing, but there's a single line you can run and your LLM is going to explain to you how it works and how to take over it, and how to improve it and everything. And I think that building those ecosystems is half the war with AI stuff currently in small to medium size works, and I think large orgs, it's impossible to have that shared context, which is one of the reasons big enterprise companies are struggling to adapt to all this.

11:35 So that's kind of how we think about this stuff. And then we also, I think about half of us do most of our prompting with voice now. Really? Yeah. Okay. I'm still typing. I'm not gonna rat on it, but I certainly all day long am just holding down a record key and talking and brain dumping for like ten minutes before ripping something off. You have you have a ten minute long prompt. Yeah. Wow. I think they're I think they're way better.

12:01 And I think there's a signal that's captured in the hesitation, backtracking, explaining how you got somewhere, you know, like if I say, okay, I need to improve the speed of this thing. So I'm going to switch to a fire and forget message queue system, and I'm going to branch along all these partitions because I think all this is faster to if I just tell Claude to do that, there's a lot less signal there than if I say, like, I think this is going to be faster.

12:32 And I think this is a big speedup and this is a big speedup. But I could be wrong. And this is another thing I excluded and give it the chance to give me all that feedback. And so I feel like I get way better results doing that. Because when I like to prompt, I like to give examples like, I like I this is what I want. Here's an example of what I'm looking for. Please replicate this process like multiple times.

12:55 That's kind of how I like to prompt it. For some reason, I think people have gotten confused. It's like they were taught in the Google search era, like, oh, shorter and concise is better and but you should be very long descriptive with the prompts. I really try to go hard ten minute long. That's extreme. I haven't heard that, but they also I think they mirror effort level. Like I think doing that just triggers your agents to be like, okay, this person expects me to do a lot of stuff right now.

13:24 Like there's a there's a reflection to it. We haven't gotten to this point yet, but we're close to just recording all the meetings. Yesterday, as yesterday, we wanted to do a meeting where we recorded it and dumped it, and I just don't have the mic set up properly for it. But I think we're close to there. And then you start getting in the problem of like, does that workflow become so efficient that nobody's talking to each other anymore?

13:49 Or like you're almost performative talking to each other and and what happens there? Like, I don't I don't have a great strategy for how we will continuously onboard and upskill junior engineers straight out of college. I just don't know what to do with them. You kind of just need like the top 5 or 10%. Like if you're really elite out of college, we'll get you. But otherwise, like, yeah, maybe that's all it is. But yeah, I kind of expect when we hire an engineer for me to have to do almost no onboarding and for them to get incredibly familiar with everything super fast, because you can just have Claude say, okay, read the system, explain to me how it works.

14:28 And then it's just like, okay, we had a we hired a guy who'd, like, drove a bunch of his stuff down from Canada and just had a voice chat with Claude going the whole time and just talk to him about the code base, the whole really, by the time he got here, it was like he'd work here for five years. Yes. Like I didn't know that. Okay, I'm going to say his name, but it was like, okay, good job.

14:49 Everybody should do that. But yeah, like, that's that's hard to institutionalize. All right. So then let's ask more direct question. What exactly are you using Claude for? Oh, my side is mostly writing code and writing like system level. Like like like multi service, wide ranging infrastructure that powers all of our trading stuff. Definitely. We do, you know, research and document generation and, and I have basic integrations built out to every PM so that I can just kind of chat with them.

15:28 Right. Like just provide environment variables to and documentation on API's so that I can kind of naturally, you know, see what's going on on Cal sheer Poly Market. But most of it is cogeneration. What about you? For me, code a little bit. I mean, there's some things where I'm like, hey, Henry, can I do this thing? It's basic. And if I fuck it up, which is highly likely, it doesn't, doesn't matter. And I'll have cloud ripped for that.

15:54 But the most important thing for me is a lot of research projects, which is like, okay, I want to know, are we getting crushed in this one market? Why I have I just go into the database, take all the all our fills and everything like that, dump it into Claude, say like, give me your thoughts. And I love doing that also. Just like if I want to do some basic research projects. Hey, how do I think there's been a line fake here?

16:23 Look it up. Basic other things. If I'm looking for an edge and I can just, you know, take this all star game, I can just say like, hey, MLB tonight, I can be like, hey, what does the what are the scores of the last like 50 average and real quick can type that in. And they can do that in like a minute. It's really easy. I don't have to do I'm like quite confident. It's right. I don't have to do all the work myself.

16:46 That's kind of what I'm using it for. Not a lot of code, more like research projects. One thing we've both used it for and you've gotten great results. So this is on the the business side of of what we do. Oh PowerPoint and stuff. Yeah. Not also like so for instance, we we finance parts of the business. Right. And so you have all kinds of partners, a lot of whom don't come from this and might have questions that don't really reflect the way we think about trading. Right.

17:14 Who will say things like, you know, what's your max exposure, your 99% or how much could you reduce it if you changed A, B and C? Like what? How could we tune these knobs to manage risk? And you go, oh well okay. And like let me do some you know looking analysis and and you're able to just like generate reports so fast to give the real example is just like we have a lot of RFQ and parlays that where it's like, okay, well what is our max?

17:42 What happens if the worst result happened. Cloud run 50,000 sims of this of this. And it's like bam. This is this is it. And just understanding, you know, terms and how to structure a business in ways that we maybe are not familiar with. Yeah. Yeah. So, yeah, that's kind of what we're doing. And I feel like I do see the, the gap widening. It's crazy talking to people who just aren't doing any of this at all.

18:11 How are you even existing? I said, like, I think we're in a little bit of a bubble. I mean, but it's crazy. Just like I was building a basic scheduler, okay? Just because we had we hired a bunch of new traders hiring a scheduler for them, would it probably taken two to 2 to 4 weeks to build? Yeah, I ripped it with Claude. Did it in one day. I did like 20 minutes of work myself, just prompting it and just like.

18:35 And Claude was ripping for a couple hours and was perfect. Just incredible. So I'll jump back into some AP summit. So the most fun thing happened was we had a mutual friend who tried to talk us into finding an edge on a particular slot machine a month ago, and and we gave him a hell for it. Like, there's my own personal purgatory would be trying to find and grind edges on slot machines. And the funny thing is, when we shot back at him, he's like, oh, we're haters.

19:07 Like, I think he rage quit the chat. It was a it was a calamity. And so I go on this trip and I'm like, oh, I got it. Because a lot of these guys play slots, you know, advantage play slots. I said, I gotta ask about this machine, like, what's the deal with it? And a few people there, it was like, oh, that was the biggest edge I've ever found in my entire life. Like life changing money.

19:28 I'm like, oh man, I really know this guy. In apology, they had like all this merchandise for the slot machine. One of them has a bedspread with the logo of the slot machine on it that his wife now is to sleep under. So you discover things like that which are really amusing. And I think this goes to show that, like, at the end of the day, it doesn't matter how stupid the edge sounds, if you can get it for a lot, it's good.

19:54 Like, yeah, it's I mean, I think we both had the thing the experience of like, you know, you come up in poker and poker is pretty sophisticated. Being good at poker is cool, you know, not to like glaze, but you gotta, like, do some stuff and then you can find sports edges that are so mind numbing comparatively. But it's like you got to go where the money is. I know it's great. I mean, you just think back to that like England Mexico game.

20:21 It's like anyone with a brain cell could just be like, okay, I'm going to fade Mexico and back England. It's 10 to 15% EV and you can't get those edges in in poker regularly. Like you got to grind for them. And it's like, oh, I did like three years of studying, another year of like practice, play and I'm ready. And I ran I ran hundreds of thousands of sims and generated aggregate reports and memorized all of them for no reason.

20:46 Yeah. So yeah, a lot of these guys come up like learning blackjack stuff that is relatively difficult. I played some, like collaborative card game one time called crew and then remembered like, oh, all these guys can literally memorize a deck of cards. So, like, I'm just dragging this group down so much. They're so good at card game. It's like, I just got to leave. But then, yeah, slots is whatever. I can't imagine grinding slots. If you could make that much money, you'd grind slots.

21:17 Do you know what a flea is in in slots? I think I do, because you told me. But the audience didn't. Just a gross guy that you pay to wander around Las Vegas looking for a good slot machine for you. Does it have to be a gross guy to be any guy? I think once you're called to flee, it's hard to overcome the fleas. No, they're called fleas. That I don't know, I hope not. So it's like beards for slot machines.

21:40 And it's like you're just to find a good slot machine. Then you ping me, and. And these guys will, like, literally fly to Vegas if they're nearby and there's something worth playing. That's that's how it works. The interesting piece is these guys kind of live at the edge of, surveillance and the risks of being backed off and and evading, you know, detection. And I think that's something that, you know, for people who bet on sports books, it's obviously been a concern for a long time.

22:13 Right. How do I avoid getting limited? And the first response to Ms. has been to say like, oh, you don't have to worry about that anymore. But I think as time goes on, like you see this in traditional finance, right. Like identifying order flow is a thing, right? And so it does become like timing, sizing, venue choice. You know, asset type. Right. Like you can play the same edge a hundred different ways. And so it's kind of cool to think about, like put yourself in the shoes of somebody who's really in charge of finding you or figuring out what you're doing at all costs and what are they looking for and then trying to approximate.

22:56 Is somebody really doing that? How would I know if they are? How would I react? Right. Yeah. Well it's funny, one of my former partners was a AP blackjack player, and he was an aggressive team in a very complicated team. Not not the basic counterparty. Like, he was like memorizing cards and whatnot like that. And he said part of their strategy was they would try to know when they'd get backed off and when just under that and leave and never officially get backed off.

23:28 But their fear was if they got backed off, they would look in and see what what were they doing. And then they might patch the exploit they had. So they would they never officially got backed off from any casinos, but they weren't they weren't allowed to play anymore and they had to get other people to play for them, etc., etc., which is interesting. I love the idea of the big player in AP stuff, which is the guy who it's his job is to look like a doofus who really wants to lose a lot of money.

23:57 And I was asking him about the slot thing and which they, you know, BP is applied to, and somebody said, oh, you would be a good BP for this. And I took it as a compliment. At the moment I'm like, I love that I would be good for this. And then later, like, what does that really mean it did also just make me think, you know, a lot of these guys are aware of prediction markets dabbling etc..

24:19 And we've already seen what you might call kind of AP like behavior. You know, penny jumpers, things like that. People who maybe they're more into the metagame of how these things work. And, and so it just made me aware that, like, definitely over the next year, we're going to see more and more inflow of people trying kind of naive strategies or exploitative strategies in these spaces, which can make make it harder to win, but it also increases surface area to attack.

24:52 Right, because ultimately they're not the most sophisticated players. And I remember this me back to it throws me back to a tweet. Like a couple months ago, I saw what someone was doing. Technical analysis on a sports market, price history. And I was like, and he was serious about it. He was like, I think this is a goodbye because of this technical. And I was like, oh boy. I mean, there's some form of that that works, right?

25:17 It was very shitty. Yeah. Like, look, it was not going to problems when you're applying. Yeah. Like the chart looks like this. So it's going to go this way. Yeah. And it was like not how it not how it works. Yeah. you know, you talked about England, Mexico, how you can just collect Mexico and flow or you could on this game because there was so much appetite for it. Conor McGregor. So much appetite for Conor McGregor I think his yes no market did ten x the volume on cash that Max Holloway did right.

25:45 Yeah something like that. That's a function of missing capital on the supply side. And so the more you have people who identify like oh this is capital constrained. And I have the capital right that that's going to fill in. That's that space will fill in with their money. When that happens. How does the ecosystem change. Right. Because if all of a sudden it's like, oh, there actually is like a complete there's a floor of 100 million in, in liquidity at kind of the opening number on a UFC fight.

26:19 Like then you get into a very weird state. And, and so I think these are the kind of trends that everybody wants to watch as they evolve this year. And it's this is why it's useful to go to things like this. You know, we have Bet bash coming up next month, right, to get a get a sense of who is entering the market and why and what they're thinking about. And I think a lot of people are open to talking about things.

26:41 Nobody more so than ourselves. Right. Like we talked about this stuff all the time. And, and so I think surface areas are going to reveal themselves for attack over the remainder of the year. Agree. Yeah. Okay. Anything else you want to talk about this? No. That's it. All right. See you next week. Yeah.