Leo Zhang - The Next TikTok is a Prediction Market - TOKEN2049 Singapore 2025

Thanks for uh taking the time to come here today. Um I will be talking about the next Tik Tok is a prediction market. Um my name is Leo. I'm the founder of Alchemia. We're working on a project called 42.

Space. So the key intuition here is that for any kind of engaging informationational systems they turn informationational entropy into something that's single continuous update signals that's very easy to interpret and very easy to act upon. Uh for the remainder of the presentation I'll talk about uh what are the implications what are the learnings to design and distribute such kind of system. So it is very obvious that media has evolved a lot since we were uh we were were kids and uh well some of you are still kids and um we went from very passive curation reading editorial notes newspaper to something that's interactive what your friends see is what you see and your interest graph every tiny little glitch of your mouse becomes uh information loop that helps people uh push what you uh see next to uh generative which I guess we're still as a civilization figuring out what what's the implication of that. Um so social and video are now primary news rail for a global audience.

Uh weekly use of YouTube for news is now 30% worldwide and for Tik Tok it's 17%. So 34.5% of the active social media users say that uh the number one reason they will use social media is to find out what's going on. Um so social has become the news destination by design and this is not this is definitely not a new realization but the problem is now feed is mediated instead of proper news discovery. Um and and this is this is a pretty widely discussed phenomena but at its core something like Tik Tok and to right there are sensors these are sensors that yield very dense feedback that's decoupled from social graph.

So everything that you do, every tiny little micro behavior becomes inputs into this this cold learning beast that uh outputs the right kind of content for you to consume. So this is a very high high frequency uh marketplace that allows content creator to constantly bid for your attention and very uh precisely matches the right kind of a longtail niche content with the kind of weirdos that just happen to love that kind of content. Um so as a result we have a very uh a custom printed uh a news app that feels like this is this is it knows me so well and and which drives session depth and it it the return frequency as well. So the dark side of that which also has been explored pretty widely is that this accelerating information cocoons that form around your reality and everything you see you see more of it and the more you believe this to be the objective truth of the universe. And now overlay that substrate with LLM, we get something far worse.

And uh it is very clear that um this is going to be a one-way street. Right now 70% of people are using AI chatbots as the primary source of uh news. And this is uh increasing at at uh at an insane speed. Um now 53% of the population worldwide says that AI has profoundly changed their life for better or worse. And this is only going to get get more and more so from from here.

So as LMS becomes the de facto middleware for us to read and consume information on average our echo chamber will become worse and worse it'll become narrower it will become more confirmatory because LM will say anything to get your approval right and more polarized so um how do we break out of this problem and this is a very very comp complicated problem so it's very multi-dimensional and it involves many many aspects of how we interact with the world but in this in this world of of infinite murals that's built around us. So finding social consensus is becoming harder and harder and um markets markets this good old invention is still the most effective consensus finding instrument and prediction markets is very very special right they turn information they turn private knowledge into into uh common common information and and in turn turn into your information cocoon into something of a decision support matrix and markets often be posed and this is something that we're learning more and more uh as for in a classic five cycle study that with close to a thousand people it shows that for election the market prediction market actually beats the actual poll by uh roughly 74% of the time. So markets are special because they can still cut through noise. They can cut through the echo chamber and because they turn these these these noise into priced adversarial bets which is still have to resolve um ground truth. So the key the key thing here is that social media they reward you for for being loud but prediction markets they only reward you for being right.

Now uh prediction market is obviously something that has gotten a lot of attention. There's a lot more conversation around this topic or at least on my own Twitter feed that could also just be my information cocoon. Uh so but it's very clear that there's a lot more volume compared to just a few years ago when this is still a high concept. Um and the top players are doing insanely well especially in the past few months. Uh the user behavior they start they they start as very scatter but now concentrate around several major headlines US elections sports macro prints uh major uh tech milestones.

Um overall the topics are are still very very US- ccentric because the the background of the first movers. Um and most of the markets outside these headlines they're very small. they only have uh less than 100k in total volume. So what this profile means is that with the right design we can actually unlock a lot more headroom for growth uh especially in areas that haven't been touched by uh the top movers. So uh in our surveys uh we asked what are the most important reasons that's stopping you uh from using prediction markets today and the common the common recurring themes are one simplicity because most people they they care about uh outcomes right whether I saying it's going to happen or not who's going to win rather than pricing probability but the probability of a certain event happening whether it's 47% versus 43% that really doesn't matter to average people and secondly liquidity there's a lot of refragmentation across different topics fix uh most of the the spreads are pretty terrible.

This is a well-known problem. Distribution, how exactly do we match the people who have the expertise to come and to to contribute to a specific topic uh versus people who are uh just randomly like looking for opportunities. So distribution matching the the the right market with the right consumer is still a largely unresolved problems. And lastly, resolution accuracy. This is also a very well-known problem.

This is quite complex and there are several ways that people have been trying to address this from uh using external third parties that's stakebased to complete centralized companies to today using a um a consortium of LRM which end up every one of them hallucinate. Um this is going to continue going to be a challenge for quite some time. Uh but it's it's very very interesting. Um anyways, so we do think that prediction market have the highest chance of going mainstream because simply because the the topic is much more easy to explain than you know 100 100x per. So but in order to get there personalization is very very important because you're only interested in in in in going there when the initial hook actually is something that you care about.

So um the kind of northstar moments that we expect for prediction markets of the future is that the the moment that you enter into this space you see that oh this really understand my niche and it doesn't take very long for you to realize that oh this this is this is uh something that's very very uh custom made for for me. So it doesn't matter if your niche is, you know, sneakers or or US politics. Uh it should be able to the right topic should be able to find you within less than 10 swipes and you can act within a few seconds the moment that you see something that you have expertise on instead of dragging this on and lastly you can learn about the outcome very very fast instead of waiting for a year or two before finding this out. So um in order to achieve deep personalization the first that we have to do localization and this is something that's this kind of playbook has been tried very very frequently with uh popular social media engines. So at the end of the day prediction market is a hybrid of trading both trading and a media business and a media business means that other than aside from P&L it needs to provide either deep insights on the data or huge interf value and both of which requires a lot of localization to to in order for the audience to actually resonate and care about these topics.

So right now most of the uh topics on the prediction markets today are very very US- ccentric and this is for uh you know vata historical reasons but the largest markets or addressable markets for prediction markets are outside of the US and especially looking at this part of the world there's high density of young people they're mobile native they're social media native they grow up in social media and they're very very engrossed in PVP games that uh crypto has has onboarded them to so um how do How do we open this up? How do we how do we how do we create this market for for these users? Uh the very first step is to recognize that in order to create to create the right topic to surface the right topic to this kind of audience, a lot of these markets are going to look long tail. But longtail markets, they're not niche. This is where the true local knowledge, the the tested knowledge really resides.

So liquidity bootstrapping has historically been very difficult for prediction markets because the conditional token framework which has been around for eight n almost nine years and uh this is the mechanism where you take $1 and split that into equal amount of yes and no and you put it into give to market makers or put them into uh terrible performing AMMs. Um however highly viral topics highly viral topics that are real time they tend to be long tail and the inherent liquidity bootstrapping problem is going to incur if you want to serve these markets. So if you want viral topics then you can't have liquidity in today's setup. So we have to change the mechanism to to really unleash that potential we have to make sure that a market that has a thousand dollar in volume feel just as engaging and fun as a market that has a hundred million. Otherwise, this whole thing just doesn't work.

Now, so there has been a lot of existing behavior uh in in the memecoin land that that uh show similarity to prediction markets. So like people were trading these Bowden trim coins as proxy to real world events and there are a lot more there are a lot more in in the in the trenches. And so this kind of behavior leap to go from using tokens as proxy to real world uh predictions to more structured more uh organized predition market activity is not very large. So if if we can just convert a fraction of the memecoin traders these retail traders to break markets that will material lift the overall user base uh of the of the market today. Now, so uh this is a a very big part that informs of both our design, our product and our go to market and distribution strategies.

Uh the product is called 42 and it focuses on delivering the right topic for you for for you. And this deep personalization is something that's very important uh for for the content on day one. And we focus on enhancing the trading experience to make sure that you're not thinking about precise probability cal calibration where you're thinking about whether this outcome is going to happen or not. And there is a very innovative liquidity bootstrapping mechanism that allows a small niche longtail market to feel just as fun as those markets that have billion dollar in trading volume. So um we just made the soft announcement yesterday.

Uh the site is called 42.space. If you uh have time uh check it out.