Prediction Markets hub - Open Discussion On Existing Challenges

Transcript

[music] >> Yeah, so hi guys. Thanks. It's like the end of the day so I'm going to make it like easy to understand. Also, Abhijit was the one that we both start like initially got the idea for like proposing the prediction markets hub. I'll quickly let Abhijit just talk about the story. Like I will introduce ourselves as well. So I'm Devansh. I worked for about a year and a half at the EF on funding mechanisms. And a lot of that focus was on how we can use the infrastructure of prediction markets to allocate money to projects. So we had a project called deep funding to fund uh 5,000 open source repos using prediction markets to scale up impact evaluation. So that's kind of my

background and where I came from. And now maybe I'll let Abhijit introduce and say where the idea for this hub started. >> Hi everyone. I'm Abhijit. I uh Previously was working on a trading product and uh throughout 2025 we spent a lot of time building strategies on Polymarket. Uh and a lot of that kind of turned into what the social experiments could turn out and how you could have micro prediction markets that could be run among the friends. That led me to build this project that I'm working on right now is banter.fun. It's um sort of like you could turn existing prediction markets into different formats that you can play with your friends or uh smaller communities.

And um the story around prediction markets in general if you see it, it's like gaining a lot of boom, but still like I think so the education around it is very limited, but um when you we uh both met in Bangalore in August, we were doing this a small session. It was supposed to be 1-hour session uh in Defioli's office around what are prediction markets. Um quite surprisingly, we missed our dinner, everything, and we were there till 11:00 and that room had 15 people and everybody was so engaged. Nobody left like throughout that period. It was a reminiscence of like D5 summer.

The last time I was in Bangalore was in D5 summer. I could sense the same energy that you have so much of interest in prediction markets and about different models, right? So far we know widely about Kalshi and Polymarket and binary markets. There's so much more to be done and so many new problems that are solved like use cases. So as Devanshu and I got together and like hey, we really need to put together at least like a forum where people come in. It is a great learning experience for us to see what people think. And example of that today is we learned about DCR. We didn't know about what is directional markets. And that is something that this whole Devcon is about, right? You come in, you might be working in a

space, you might still discover new things and that's what we have been learning and trying to kind of say share what we have and bring everyone on a forum and that's what whole prediction market hub idea came about. Devanshu can talk more about it. >> Yeah, so maybe like one thing which is maybe surprising to people here is like prediction markets are like gambling. Which is kind of like somewhat true today. Like information is a nice byproduct but the main purpose is people being able to put money behind events. Like sports games and politics primarily. But I think the wider ambition we see with prediction markets is that it's a two-sided marketplace where someone wants some information and they create a market and then traders are

are betting on what is going to happen. Like it could be something like even employing someone. Start a market saying who am I going to employ and then you can see the market odds and then use that to actually interview candidates and make a decision. So I think like similar to how Uber has passengers and drivers, we can create we can use prediction markets for uh people who need information and people who can provide information. So that was kind of the broad insight of why this got approved on Def Con because we are not interested in the gambling use case. We're interested in actually getting useful information and like being able to pay for information. Um so how did we actually structure the hub? Now we'll come to the part about how our hub is structured and other things. So we

wanted to make it an open learning space for everyone. So because we have so many different teams in prediction markets, we didn't want it to just be like something you could sit at home and watch as a YouTube talk where people are just coming and giving presentations. But we actually just organized roundtables. So we only have roundtables every day from 10:30 a.m. to 12:00 and from 2:00 to 4:00. So that's the only time we have programming and we'll talk about like different challenges with prediction markets. So for example today like I be mentioned we spoke about market design and creation like directional markets and other things we learned here.

I'm just writing a tweet about it as well. So I like to tweet out all the learnings. We it was like overflowing our hub because we were actually asking all the people attending to share their thoughts. They were taking polls over there. Some of them were users. Some of them were traders. Some of them were market creators. So we actually got to like conduct market feedback at this hub for everyone. So it was quite like a vibrant experience. Tomorrow we are doing on oracles and market resolution because so many markets fail because the way you resolve the market is flawed. Then a lot of AI trading and trading strategies, new use cases for prediction markets, how do we give liquidity to them? So we are trying to go deep rather than wide. So when you have a talk, you just like usually scratch the surface.

But our whole purpose of the hub is that we just talk about a few issues, but we'll all go really deep into them and try to advance the state of the art. Um so for example like that this is the the quick area of how roundtables work. Maybe Abhijit can quickly just talk about like what the roundtable was like today. Yeah. >> Uh largely it was a surprise to us as well so because we predicted it to be like 15 20 people come up and we will have enough time for like one-on-one interactions. At some point we had around like 40 people show up and it was like a lot of takes to come and we had to have do some heavy moderation.

But the idea is so if the whole idea was first today we were learning but the goal is that we have a question what people feel okay certain things that they know today and then what could be basically are things that not so obvious to a lot of people. Like especially about today was what market designs exist. So binary markets are very popular but not a lot of people know scalar markets and there are not people know about conditional markets and also discussing about why certain protocols didn't work in like certain setups. If they chose like a certain model and that is a very like I think so two-way street where you get a lot of

feedback even if you're building our product why it would work and why it wouldn't work and why it would work and what other users are thinking about it. Right? For users it becomes interesting as what they can see as like okay if I'm trading on binary market say BTC price versus if I'm trading on scalar market what is the benefit for me? It's like that's like I think so first touchpoint we wanted to establish today and then it just like it turned into like a discussion where like we had a lot of viewpoints. We had new kind of like ideas also which we can create into markets which will lead into our tomorrow's session which is new use cases on 5th actually. So this is also like a

great starting point. Today's session was this um and um Do you want to talk about what our thought processes in like structuring the round table? >> Yeah um So like Like like I mean if you take a simple theme like this one new use cases and privacy like market feedback which we've done shows that privacy is needed in two ways for prediction markets to really evolve into new use cases. One is that the outcomes in in market itself are private. Like if I set up a market about who is going to perform well at a new job opening that you have in your company. Now, you obviously can't make the outcomes public because if you apply for a job when you already have a job and your employer sees your name in that market, you'll land up in trouble. So,

privacy is needed to really take prediction markets to more than just sports and politics and crypto prices in the next 20 minutes or whatever is popular right now. Um and the second form of privacy is when the odds itself are private. Like if I set up a market and I get odds, but it's public for everyone to see, then I'm essentially like I'm the one paying the bill, but the entire world can see the odds and get the benefit. So, these are kind of like the very niche kind of things we we like we go very deep into. So, if you're actually interested in the field of information finance, prediction markets and you want to go into a deep dive, I think this is like a great way to do it where we'll go deep into each of the issues. Like another one which is

starting tomorrow is on oracle and market resolution, which is like all of you know the Zelensky story of how he was wearing a suit and then there was a big fight over whether it was a suit or not. So, how do you actually like start addressing these problems? If like users are able to create a market, like if you want to go create a market tomorrow, how will the market be resolved in a way that you can't be evil and corrupt and try to like steal all that money. So, that's where like oracles and market resolution, some of the thorny problems are. Um like another very cool part is like because a lot of these markets don't have established use cases, like if you start what is the price of BTC going to be in the next 30 minutes, there are already people at Jane Street and

traders who do this full time. But if you start a market on like what is the value of this open source repo as judged by a maintainer, what is the likelihood of this person getting hired at this company compared to other applicants? No one actually trades on that today. There's no established market. So, you have to build up that community of traders and that's where like AI trading can be useful. But when we tried doing it, we ran into challenges like Claude just refused to trade and it said, "I'm saving to memory that you're a malicious person who's asking me to create a unsafe autonomous agent. That's the exact safety check it flagged. So, some of the issues we faced in like AI trading in prediction markets and like trading strategies. Like who who are the traders actually earning money in prediction markets? Like we spoke to

some traders and they're saying, "Yeah, this is my strategy for earning money in Polymarket." So, these people are going to come to the hub and share some of their learnings, some of the people building AI bots trading in it and how we hacked around Claude's safety controls. So, that's the sort of discussions and it's all Chatham House Rules, so you can share everything, but you can't attribute who said it. Um this is like maybe the like amongst the concluding ones, which is liquidity provisioning. So, even Polymarket subsidizes liquidity to up to 5 million a year. So, it's very hard to actually get liquidity in these markets.

Um and that's kind of the open problem. So, that's what we're trying to find out from all the different projects which are there. How do they actually solve the liquidity problem? Um so, this is kind of a quick introduction to our thought process, uh smart contract security. Any auditors in the room? Uh yeah, okay, I see one auditor, yeah. So, like we'd love to have you come in the round table and talk about the process of like smart contract security in prediction markets. And most of these presentations are more as a conversation opener. So, you like open up a conversation rather than it being like this is like a pitch or anything like that. Like we have a very strict no shill policy. No one talks about their projects here. It's more like you introduce it in the context of the theme that we care about. Um so, yeah, this is

the program. Anytime that there's no round table, it's just a space for the community to hang out and like meet each other and talk. So, it's like it it can be a big place, Devcon. So, this is a place where you can just come, recharge, sit with your laptop, do work. Someone comes and talks to you about prediction markets. So, it's really meant to be like a comfortable space. Yeah. >> Wow, interesting. Um I think I we don't have a question yet, but I have you one for uh you. Uh do you think that we can use the AI to get the result of a prediction market?

>> You can use AI to get? >> Yes, to to solve the prediction market, to get the result. Is it possible? >> Resolution, like basically. >> Resolution, yes. >> Um So, I think so what AI enables to you people to do is build custom like harnesses for that they could create a lot of like different sources public information and feed that into like solving problems or like solving markets much faster, but uh still sort of like the challenge with there is how you still need attestation to uh these agents and AI because this itself is like consuming information that's being fed to it, right? So, you still have like a lot of teams say one team popular

team could be pushing like one result into AEO and the others wouldn't be. So, it's not like really uh accounting for all the sources. So, still like I think so there is a um a bias can be generated like whoever is feeding in the sources that AI takes in all the information from this. So, there's a um bias at the source which will lead into like it creating um problems and resolution. Uh but I think so it's like for a lot of things that are publicly inform- uh information is available, AI is starting to prove like really good as like a initial benchmark where you can collect a lot of information with it very fast. So, like what like a manual process is or API

accumulation will take a lot of time. Uh AI agents can do it much faster, but right now you need kind of a human in loop or a council in loop that kind of says, "Okay, how um how reliable this information is for resolving a market." So, I think so we're getting there, uh not just yet there. So, >> It's interesting. I don't know if you we have a question, but uh I I a history a funny history related to this. Imagine that it was June and the basketball in US, the final game.

And I was I work with Oracle. So, I create my prediction market and with Oracle asking AI if the Knicks win the game will win the game. But because of the date time it cut of AI, it was saying that Knicks didn't win the game because like the game happened the day before. So, this is something that we could think about the challenges related to use AI or at least to grab the information in a place that is like in this case, you must go to the official site to get the

result. Not use AI to ask if they they win the game. And that was my fun history. >> Now, I think like another aspect is like when we try to use like resolution for subjective events, AI becomes much tougher. Like for example, like one cool application we wanted to try in DevCon is a prediction market for how many people will attend each talk. I mean the numbers in that market can actually influence the room allocation. So, for example if if the prediction market is saying that fewer people will attend this talk as compared to another one, then we can be allocated in a smaller room. And then we can actually go and check how many people actually attended the talk and

then we can resolve that market. So, I think markets can be used for a lot of things. Humans are like a committee-based system or a market. We have two methods for making decisions. And really I think the longer-term promise which are all trying to work at in the hub is how can we use prediction market infrastructure but for decision-making rather than just the kind of sports stuff and the politics stuff that goes on right now. >> I I think so one very interesting discussion around today was how we could make the existing systems more efficient. So, insurance market was very deeply discussed and it's like right now there's no open data of how an insurance company prices premiums or for example

what factors that it it it accounts for when it is building a policy. But now if you have open markets for it and this this acts as like an information source. Uh it goes both ways. For an insurance company it gives like where the demand is and how much people are willing to sort of like hedge. At the same time for users it starts to become like they could package some very innovative like say insurance policies. I'm not sure if you heard of this but there was a bar in New York that wanted to really rapidly grow to prepare themselves if Nick wins Knicks win. Because this is like happening in like once in a while kind of things. Right? In this century the

Knicks haven't won any NBA championship. So, if you scale your stock immensely and then nobody shows up because Knicks lose in the last game then basically you are left with a lot of stock and you're paying of course the more cost on the stock and the storage. So, they built out and nobody's going to insure this, right? What they built out was a very interesting hedging strategy or kind of insurance was that they bought Knicks will not win as a bet like as a position on Kalshi and they went on to give this promo.

If Knicks don't win then we will pay you the drinks for free. That's kind of like a gift to the city. But if they win then you have to pay for the drinks. Which is like if nobody shows up they're not going to sell anyway so they might just reduce the cost of storage and get make some of their existing customers really happy so you have loyalty points there, but if they win, then they're going to have a lot of like users like a lot of people come in party in the bar. Anyways, they're going to like be selling a lot.

So, this is kind of like things where people can get really innovative of how they ensure different scenarios, which are not right now um fit in the traditional insurance uh structures. So, just like the use cases which uh Demange uh is mentioning is like we know prediction markets in a very limited scope today. And those are like the markets that we see popular markets, but a lot of interesting use cases are happening in the background, and some we are learning ourselves at this prediction hub. So, hope to see you guys at the hub.

Thank you. >> Yes, no, don't don't go away. Don't go away. No, you can clap. You can clap because they are so good. Yes, but don't go away because now we have questions. And yes, uh before the questions, it was so interesting like I talked about Nick's and you gave an amazing strategy related to the game. So, this is really cool. Yeah. And how good do you think AI is at predicting predicting real-world events in prediction markets today? And how much better do you expect it to become over the next 5 years? Wow, this is a long time, especially in our space. Do you think AI will eventually outperform?

>> Yeah, I can take this question because I worked on exactly this problem. Um so, one part is what we call superforecaster AIs, which are AIs which make predictions and which they usually turn out to be quite correct. But, I think where prediction markets shine and why they won't become defunct in such a world is that they become an ensemble method. Ensembling is a way of combining different AIs outputs. So, if you ask Claude for an estimate on an event, and you ask Open AI for an estimate on an event, now which of these models do you choose? And that's where uh prediction markets can help, where you can take the estimates of different AIs and use prediction markets to combine them into one master estimate. Um how accurate are

they today? I think the record is a bit mixed. Like, if you look at the Brazil elections, it predicted over 80% for Bolsonaro to win, but then he ended up losing. We don't know if it was legitimate or illegitimate, but like the markets were wrong and that's kind of caused a bit of a concern around accuracy and all of that. Um so, that's maybe a short answer to that question, yeah. >> Okay. And uh yeah, the questions are changing the Yeah, but we can do that. Uh what needs to happen for organizations outside the crypto to start adopting prediction markets?

>> So, I think so, I'll just go back to the same example I just now gave, right? A bar in New York uh trying to adopt it. I think so, as we expand the horizons of markets itself, what kinds of markets can be created, we'll see a lot more participants that are not traditional participants. Could be like not even traditional traders um come into the markets and utilize those. Uh one of the other example that I gave is insurance companies using it or research firms using it as a cheaper method to derive like certain like information, which if they do all by themselves, they have to not only fund the research, but hire actuaries to find

the right percentages and to get those. But, if you have incentivize um like the end users themselves, which you are eventually going to research, um you can find like much better results, which are accumulated by users itself. So, the whole process becomes much cheaper. >> the one like caveat I would add there is the privacy element that currently insurance markets treat the uh the insurance companies treat the research as highly proprietary, but we We not have the tech currently to make the odds private only to the liquidity provider or the insurance company and the outcomes are also often publicly traded on. So, I think until we crack the privacy question, I don't see like that it it's a it's a limited use

case to whatever can be public, which is actually quite small for the outside world. Crypto is unique like those are unique, everything is public, but that's not how a lot of companies outside our space work. >> Yeah, and I think the first question and the the fourth is related uh to the government part. So, will prediction markets be restricted by governments or they will become more friendly? What's happened? What do you think about it? >> Yeah, I think it's very funny that like Polymarket is banned in India and we are over here. I think that's like hilarious. If you try logging on Polymarket, it will like block the site for you. Um will they be restricted? Will they become more friendly? I can see it going

like two ways. Like at one level, prediction markets are truth machines and some governments don't like the truth and they would prefer to be able to control the narrative. So, in that scenario, I see the government becoming way more hostile to prediction markets. Um and on the other hand, I feel like it can help bring much-needed transparency into governments like say a RFP process of government filled with like rife insider trading and corruption. But now you can run a prediction market saying who is the government going to award this contract to. And if there's a disparity between the market estimates and who they eventually award the contract to, then that's something which investigative journalists like I used to be will question them on. Um so, I think I can

see it both ways, um but I hope that it starts getting used for better use cases around government contracting and decision-making. >> Yeah, and we didn't do the first one now. So, what needs to happen for organizations outside crypto to start adopting prediction markets? >> I I think so he uh covered that. reading. don't you see too many markets for financial risk? >> Yeah, let's go to that one. Yeah. Why you don't see too many markets for financial risk? For example, will BTC volatility exceeds 6.5% this month?

Uh it will have implications for markets, options, etc. How is this? >> I think so this is one of the topics we discussed today. Um I think so that the model itself right now is largely binary because they're most adopted market structures. Um and so as we see more sophisticated participants come into the system, which is already happening. We have market makers, you have proprietary uh funds or prop funds come in. Uh what will happen is we will have structures like scalar, which is possible to do this thing. Um things like will like what will be the volatility of Bitcoin say, what range would it fall into? And you could create much more concentrated, much more

capital efficient systems, which then will bring in more uh capital into these markets. And then um I think so largely we are also seeing uh adoption of more simplified options uh starting to happen. And it is also kind of like progression of options largely because not everything can be fit into the current structures. You not need a lot of liquidity to even like bootstrap those markets. But I think so with prediction markets we are already seeing a very simple example is I think so compute uh is starting to become like a big use case and it was been like uh it's been very hard to price it on get participants in like private markets and private options, but on prediction

markets is starting to take off. So I think so those are new um um use cases we are seeing for financial risk uh and new systems that are also coming in uh more around like what can be possible and we will see more market types that will now be say focused on like certain kinds of markets. So yeah. >> Yeah, I think someone said it well that like these Bitcoin type markets, the price of it is like inferior options. So, there are much superior products for trading on those questions than the prediction market architecture.

>> Yes, and the we have more questions now, but I would like to you to answer the other questions in the prediction market hub because >> [music]