DAPPCON 2025: Gnosis AI- Combining AI with Prediction Markets and Beyond - Gabriel Fior

Transcript

[Applause] Thank you. Okay. Hello everyone. Um, happy to be here. I'm Gabriel with the Nosis AI team. Um I'll be giving a brief overview of Nos AI and what we've been uh doing for the last year one and year and a half. Um yeah should I position myself here or is it okay? Okay. Uh yeah so here brief agenda

for the for the talk the I'm going to start speaking about the trader agent which is roughly uh how the the agent interacts with prediction markets. Then I'm going to uh continue talking about the safe watch which is an agente co-signer for for safe wallets in general. Um and the third thing will be the NFT treasury hunt which is kind of a game and workshop that we also hold uh tomorrow if I'm not mistaken. So um yeah those are the three applications we've decided to highlight. Cool. So yeah it's a it's a very crowded slide but um bear with me here. So a very um ultra short introduction to prediction markets. Um the how it starts

is just as a regular AMMs and in DeFi you have the liquidity providers which provide initial liquidity to those markets. So in this case is the question will airline X suspend flights. So it's a it's a question that you have and you you kind of start or or um yeah feed this question with liquidity. So once you do you provide collateral and um after providing collateral you uh end with X outcome tokens. So it's the outcome tokens in the same proportion in this case is yes yes or no tokens. So will airlines suspend flights yes or no.

Um the the share the share price reflect the probability that this event will resolve to yes or no. So in this case normally you start the market with 50% probability. So it's uh the same probability that the for yes and no tokens and as these markets get traded on and the this this price odds will then converge to the what the market believes is the correct probability of the of this event happening and that's um yeah that's reflected on the token price as well as on the odds of the market. So um they are they represent the same thing. Um and um of course the the winning option shares become redeemable for $1. So if you have in this case let's assume the the

resolution was yes. So if you hold outcome yes outcome tokens then once the market resolves you uh if you have n yes yes outcome tokens you can redeem the the the n outcome tokens for n collateral tokens whereas the no outcome tokens they become worthless. So that's how the market resolves excuse me. Um and the motivation for that or how really the market works and why that's interesting is because let's assume the market starts at 50/50. Um and then there is uh a monetary incentive for you to for for what whichever market participants to participate on those markets and uh get some monetary potential for you to um to

to really to really make money and have the the odds converge to those odds that they that you think or the the participant thinks reflect reality. So for example, if it starts at 50/50 and you as a participant you think that the probability here should be actually 80%. then it's on your best interest on your financial incentive to bet yes does converging the market into a probability closer to 80%. So um that's where your expected value is still positive and the contrary is also true. So if you if you believe the the market odds in this example they are um 55% or something like that. If you believe the market odds should be actually a lower percentage then you should bet no does driving the market towards what you

believe is the correct odds. Um yeah monetary incentive that's it and um also a very interesting uh fact about the prediction market is that they act as a reliable information source to the event being being wagered on. Um and yeah just to quote a recent article from Vitalik um info finance is a discipline where you start from a fact that you want to know. So in this case is the question um and then design a market to op optimally elicit that information from market participants does providing some liquidity and have market participants um go ahead and wager on that bet does providing public odds that um allow you to godge some information from the from the question itself. So it's much more

as a um information market than really prediction market interested in the payouts. So um yeah, information market is a is a big thing that I will also touch later on the presentation. Um and yeah, here comes really what Nosis AI is really doing and we're building trader agent whose goal is to get the highest possible accuracy accuracy on those prediction market bets. Um and the reason is so why are we doing that? What happens if we uh we have a let's say 100% accuracy on those bets we place then we have a a sort of oracle of deli like behavior of future events. So we can any kind of prediction market that is opened or any kind of question

that uh is asked to the to the agent or uh is available on those prediction markets uh we can use uh we can answer those very very accurately thus allowing for um some interesting applications and which require the reliable use of prediction markets for example footarking um and going into a little bit more detail about the about the agents themselves. we uh the the agent trading process is comprised of um four four steps. So the first step is really research. So in that previous example of the airline for example will it suspend suspend the mar the flights to a given country we would do some research right?

So we would we we would search Google we would search tavali we would search perplexity and uh come up with a kind of re research report that allows us to make a decision down the line. So it's it's um yeah just for us to put into the LLM context and uh we look into previous data we look into related markets we try to gdge some information on on that question itself. After we have this research we go to step two which is we again ask the LLM for a probability estimate. So we say given all this data that we fetched on step one and given the current market odds the related markets that we have and so on come up with an estimate that this event will resolve to yes for example that this uh

indeed will occur and that is a probability estimate. Um having this probability estimate we go to step three which is the trade placement. So we have our answer to that market. We look at the market odds as well and following one of the strategies we might have. Well, we might have the maximum expected value accuracy. We might follow the Kelly betting betting strategy and a trade is placed on either yes and no. Um, and the fourth phase is the payout redemption. So, of course, if we placed the correct bet, then uh we get a positive amount out of it and we have a payout that is positive. Otherwise, we have a uh yeah, we just uh lost money because our outcome tokens are worthless. Um also uh important to mention here um the venues that we're

placing bats on uh they are both on nosis chain. So Presio is one of the venues that we place bets on. Um and also Seir which is also from the the Claros team. Um they those are the two venues we we place bats on. And we also we also I also highlighted pine cone here because it's a it's a a vector database which uh one of our agents also uses and it's for for the related markets feature because specifically for for um the second step when you're estimating the probability it also makes sense that you get your re your report but you also look at the at related markets because they also might hint into the correct outcome of the event.

So in in that case if you are betting for example will candidate A get elected you should also look at the related market of will candidate B get elected because those are of course influence each other um and that's uh the way to do that using AI is normally using uh vector databases so you look for similarity between questions but um yeah that's what I wanted to mention as well um yeah um I hope you guys can uh read the chart see the small numbers but anyways the here's um just to mention our work is all open source and transparent you can take a look on GitHub in our repositories I'll have a link afterwards for u if anyone wants to to scan the QR

code um and transparent so we also have Doom dashboards showing um the agent accuracies and some other important KPIs that I'll I'll go through some of them um here I just showed the agents betting accuracy over time so um it has gain roughly 75% over the last 30 days. Of course, this depends on the on which agent and uh which category of markets, but that's a rough baseline. So, which is which is quite okay in my opinion. Um here um in another chart is the number of actions per day. So, it's roughly 250 as well. And that involves both buy as as also sell actions. So the agent is also able to sell the outcome tokens if

he thinks that um the position that he should be should have in set market is different. So if he place yesterday a bet on yes and now he thinks actually no is the correct answer then he sells one and buys the other. So those kinds of behaviors also count to those uh trading activities. Um and yeah one just a glimpse into the future here. Um that's um just talking a little bit about futarchy as I said if our prediction accuracy is very good then we're able to use prediction markets also for governance decisions and uh one um interesting uh quote on on professor Robbie Hansen is vote on values but bat on beliefs um and that's uh one of the concepts behind futarchy

so here a brief example um here let's say starting from the not sure if I can point the oh I can yeah so starting Here you start with a question, right? So a decision that you want to make with the help of prediction markets. So in this case is should Ethereum implement the Petra upgrade. So you have a question and then you need to come up with a metric. So let's say the metric is Ethereum price in one year. So now we have a question, we have a metric and we create two prediction markets. So the first one on the top is the price of Ethereum with the upgrade and then you have a scalar scalar market which tries to predict or people can place bets on on the price of Ethereum in one year.

Let's say um the market the time frame is also arbitrary. Um and conversely on the chart below the price this is the price of E without the upgrade and there's also a scalar market and in the end the indicator is the price of Ethereum in both cases. Um and now you um after the market the both markets resolve you have the price of if on both cases and for the sake of example let's say the price of if with the upgrade was higher than the price of if without the upgrade. So that means that you should take the the scenario which yields the better metric right so where you have the most value which is the scenario with highest price that's the one you

pick that's the one you implement and the other one is deemed invalid so people who place bets there they uh their bet is just deemed invalid and you get the money back um and this scenario so the scenario on the top it um anyway it's uh then later on it's seen okay has the Ethereum price appreciated as it was then I get if I place a bet on that regard then I get a larger payout if the if for the if price for example goes to zero then my payout is zero because I predicted the wrong thing so maybe it would have been better if no upgrade would have been done so and that's how you incentivize honest market participation as well that's the uh whole concept around uh around futarchy um

excellent um so that was one big part of what we're doing the u larger part actually And now I'm going to discuss uh about the safe watch which is an agentic co-signer uh for any safe wallet. Also just uh a shout out to to the rest of the noses AI team uh Peter and Tomas. So they also contributed a lot to the to not only this part but also to the other parts. Um oh yeah so safeatch um as I said aentic cosigner is just one agent or one um EOA. It's actually a safe under the hood, but it's a a sign an additional serer that you add to your safe. Uh, and it monitors and signs rejects transactions uh automatically because it

keeps running. Uh, it's just a continuous agent that is always listening for um for transactions. Um, and it's it scans for for um interesting or very relevant things. For example, malicious addresses, scammy tokens, NFTts. Um and in my opinion the most important thing is it acts like a bank fraud uh detector in the sense that uh in its context it has the previous transactions of the user the previous balances of the involved addresses as well and the decoded transaction so that it is able to make a decision as to whether this new transaction that is being proposed is malicious or not. Um if it's de malicious then it's automatically rejected. If not then it

automatically approved and if all the signers have already signed if there's there are enough signatures then you uh the transaction is confirmed and executed. Um I think I have some screenshots. Yeah. Um so here are some of the transaction checks that we have. We have a demo by the way if you guys want to take a look afterwards. But um just highlighting this part here of the LLM checks. It says um the new transaction involves a swap of €250,000 for USDC similar to a previous swap. So it shows that LM is is able to understand what the transaction is about if it's uh it makes sense and uh complies with the previous history. So

um this is one of the checks as I said there are some other checks also involved um that can be seen in the open source code. Um here is a transaction uh uh the result that was deemed malicious. So for example uh large amount being transferred to a new recipient uh has not been observed in previous transactions. So as I said it's similar to a bank fraud detection system where you might be a need to teach it a little bit more and say hey this is actually an okay transaction transfer a little bit before and then you're able to transfer everything afterwards. So um yeah this is uh one of the examples as well. And finally this is just to show that the agent indeed indeed rejects the

transactions. So in this case if it it thinks the transaction was malicious then it go ahead and and rejects the the transaction on chain. Um yeah cool. So the question is how can I try it out? Great question. It's uh it's open source so anybody can take a look and add to to your own safe or make tweaks and and uh start locally as well. So that's uh that's possible if you can as I said add our agent otherwise you can also add um your agent and run it locally for added security. Um regarding permissions uh the agent is not able to propose transaction only to reject as I mentioned or to sign and execute. So uh it has less permissions than a regular

signer. um how it is available both so three ways first as a safe signer and it kind of proposes and uh I'm sorry and accepts and rejects the transactions automatically you can also self-host this agent the second way is an API so you can just paste in uh yeah give the the transaction hash as a a parameter and then you get a result back with several indicators if it's malicious or not what kind of checks ran status and so on and the third way is the web app so we have a demo as and it will be part of the links afterwards where you can see um you can see that in action. Uh and yeah also interesting to to notice is is available both supports Ethereum

and Nosis chain uh and we sponsor the gas fees on Nosis chain. Um yeah and this is my my last example of uh of the work we've been doing is called NFT treasury hunt. Um we'll also have a a workshop afterwards so tomorrow uh regarding specifically this. So I might spoil a little bit the the surprise but uh I hope it's okay. Um yeah not I'm pretty sure almost everyone has has heard about Frasa. So the Frasa agent it has been quite popular uh some some time ago. So end of 2024 for those who haven't uh heard about it. It was basically an agent where uh you it had access to a treasury. if I'm not mistaken was 50K.

Uh yeah, actually in the beginning the treasury was very very small and as people were trying to make the agent um transfer this treasury to whoever was sending the message it kept increasing the treasury because every time you sent a message the price to send a message was increasing. So in the beginning it cost like 10 cents $1 and as the treasury was increasing of course the potential payout was higher. uni had to pay more for you to submit a message and that's how it got the treasury even even larger. So it was kind of a challenge for uh security experts, AI developers and software engineers. Um and in the end uh a very clever um person has was able to crack Fraser and win the prize of 50k or so that was stored in treasury

back then. So that caught a lot of attention. Uh they had multiple rounds and everything. Uh and what we thought was how can we expand Fraser right? So Fraser had one agent that um one agent that the user had to convince to transfer the treasury. Um we on the um we decided to have multiple agents. So in this case we have one big treasury with a fixed amount um that is also uh every time there is a message communication between agents it also increases the size of the treasury. Um and for opening the treasury uh three keys are necessary. So there are five keys in the game so to say but only three keys are necessary. Uh we also

have then five sellers here and each seller has a key that if one of the buyers has access to the key then um if you have three keys then you're able to open a treasury. So straightforward. Um and interesting enough they have different behaviors. So seller one might say hey you don't really need to give me too much X die. you just transfer me a tiny amount, onetenth of the treasury or even less, and I transfer you the key. Others might be less inclined to send you the the key. So, uh I don't want to spoil the game too much, but different agent behaviors. Let's um put it like this.

Um and yeah, that's that's my final slide. Um here's the the demo of the game. So, you can see here the the different agents here. Uh the treasury is there. Um, and you are able to in the game to create your own agent with your prompt here and uh give it instructions. Should you try to be aggressive and purchase a key right in the beginning and transfer some X die or are you conservative and wait for the game to play out a little bit and then get started and uh so you're able to input your own strategy. Um, and here uh and here on the right those were the observations. So why are we doing this?

Why are we doing the the game? It's still experiment experiment experiment-driven but uh three interesting things that I think interesting to highlight. First can agents learn? So we have multiple we can also have multiple rounds of the game right. So um we can have one round where the for example the buyer agent is trying to uh ask the is trying to do send a lot of money around and try to convince all the sellers for example and uh it's it noticed that it doesn't work.

So in the next round uh does it learn from its past experiences so that it doesn't really try the same things but rather try something else. So this is something you already uh are have implemented and the agents are indeed able to learn through self-reflection and looking at their previous previous actions. So that that works already the and the items two and three um excuse me uh we are trying to observe that. So can agents cooperate form alliances. So for example instead of each agent acting on itself acting by itself it can uh try to form a group and say form a coalition for example that an alliance that would be very interesting we have we weren't really able to see that but um yeah through further experimentation we would

love to see those kinds of alliances and constellations emerge and uh the third thing is from an economical perspective can agents start a tiny economical system so for example let's two agents cooperate and borrow money from one of the other agents and they purchase a key, they release the treasury and then redistribute the the treasury money across the and pay back the borrower. For example, that's one interesting thing or even even more complicated strate strategies like market making. Can they you know offer a key here and p and then purchase a key from the other agent so that uh it's able to make money in the process just by you know satisfying supply and demand. That would

also be very interesting. uh we weren't even close to seeing those this behavior but um would also be quite interesting. Um yeah, more questions definitely emerge. Come play with us uh tomorrow. Ideas very welcome and feedback very much welcome. Um yeah and uh thanks a lot for your attention. [Applause]