Stephen Grugett (Manifold Markets Founder) - Predictions Markets & Revolutionizing Governance

Stephen Grugett is a cofounder of Manifold Markets, where anyone can create a prediction market. We discuss how prediction markets can change how countries and companies make important decisions. Manifold Markets https://manifold.markets/: https://manifold.markets/ Watch on YouTube https://youtu.be/XoRXrnHg084. Listen on Apple Podcasts https://podcasts.apple.com/us/podcast/27-stephen-grugett-manifold-markets-founder-predictions/id1516093381?i=1000559667223, Spotify https://open.spotify.com/episode/0PIp8YHRRMGsYtOSO9yC3Y?si=ZyQQd2SaQ0y5h4kZrfOAnQ, or any other podcast platform. Episode website here https://www.dwarkeshpatel.com/p/stephen-grugett. Follow me on Twitter https://twitter.com/dwarkesh_sp for updates on future episodes. Timestamps: (0:00:00) - Introduction (0:02:29) - Predicting the future (0:05:16) - Getting Accurate Information (0:06:20) - Potentials (0:09:29) - Not using internal prediction markets (0:11:04) - Doing the painful thing (0:13:31) - Decision Making Process (0:14:52) - Grugett’s opinion about insider trading (0:16:23) - The Role of prediction market (0:18:17) - Dealing with the Speculators (0:20:33) - Criticism of Prediction Markets (0:22:24) - The world when people cared about prediction markets (0:26:10) - Grugett’s Profile Background/Experience (0:28:49) - User Result Market (0:30:17) - The most important mechanism (0:32:59) - The 1000 manifold dollars (0:40:30) - Efficient financial markets (0:46:28) - Manifold Markets Job/Career Openings (0:48:02) - Objectives of Manifold Markets This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.dwarkesh.com https://www.dwarkesh.com?utm_medium=podcast&utm_campaign=CTA_1

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Dwarkesh Patel?

  Now here's a question I've had for a while. Why don't companies who have direct incentive to get the best possible information on themselves, on the factors affecting their business, why aren't they using internal prediction markets? If this is the best way to aggregate information?

• 0:23Dwarkesh Patel?

  All right, this is gonna be fun. Today I have the pleasure of speaking with my friend Stephen Grugett. He's the founder of Manifold Markets, which has received a grant from Scott Alexander. Scott has written thousands of words about these guys. And they've raised a $2 million seed round. And you know, it's an incredibly exciting project. So Stephen, why don't you tell us a little bit about Manifold Markets?

• 0:46Speaker 01

  Great, thanks so much for having me on this podcast, Orkesh. It's great to be on here after seeing other people like Tyler Cowen and David Deutsch. I guess first thing, I'm one of three co-founders of Manifold Markets.

• 0:59Speaker 01

  But what we're doing is building a platform for user-created prediction markets. So the idea is that anyone can come onto our site and create a question about anything that they care about. And then they can have their friends and other people on our site come bet on that. And the betting process through the magic of our market mechanism will help, you know, get the best and most calibrated probabilities that you can find.

• 1:24Dwarkesh Patel?

  So let's talk about the mechanism here. So the, correct me if I'm wrong, but the idea is you use real money to buy manifold dollars. What is your reason for expecting that people will care a lot about how many manifold dollars they have? So what is your hypothesis about human nature and reputation? That makes you think this is something people are gonna invest a lot of effort and time into calibrating?

• 1:46Speaker 01

  Yes, so I guess to take a step back for people listening to this. Manifold uses a play money currency. If you sign up right now, we'll give you 1,000 manifold dollars, which is our in-platform currency for you to use and bet as you see fit. But the reason why we think that play money can work is that people are driven more by status

• 2:10Speaker 01

  Competitiveness than greed. You know, we, we kind of see ourselves as a social game where people can come on to hone their skills at predicting. And then demonstrate to others that they really do know more about what they, what they're talking about. And they can prove it with an, um, an objective track record from their betting history.

• 2:30Dwarkesh Patel?

  Yeah, that, that, that's a, that's a very interesting point, which makes me wonder. Do you expect that in the future? Wall Street firms will be enticed to get the top people on the leaderboards on manifold markets to come work for them. One of the objections that Tyler Cowen has to prediction markets is, you know, he, he's kind of tongue in cheek about this. But I remember at one point he saying, you know, if you guys are so good at predicting stuff. You would expect all these hedge funds to be trying to constantly hire you guys. The fact that they're not makes me think that, you know, this is kind of just a hobby. This is not, uh, so I, I'm curious about your reaction. Do you expect, uh, people to be coveting these, uh, these top predictors for their, uh, you know, for, for their prowess at predicting the future?

• 3:13Speaker 01

  Yeah, I'm obviously very biased, but I think something like that is possible. Um, I think the, we have a leaderboard right now. Our top predictors are actually quite good. The, the person who, who's currently occupying the top slot. Uh, one big on the Russian in, uh, on one of our markets on whether Russia would invade Ukraine in February. In the, so in the forecasting community in general, there actually is like, as you would expect, um, a fair amount of overlap between the people who you, who are using these services and financial professionals who actually do manage money or trade, uh, trading as, as their profession. So I'm not, don't entirely agree with the, with the premise of the question or Tyler, Tyler Cowen's earlier point.

• 3:55Speaker 01

  Um, although obviously our type of, you know, as a play money platform, we do attract a fair number of hobbyists who despite not having a financial background, still are quite good. I do believe that there is a lot of like untapped potential, um, in the, in these top predictors. And that sites like ours actually are able to identify them.

• 4:13Dwarkesh Patel?

  In your experience, what has made people who are the top predictor so good at, uh, so good at the art? Um, is it just their domain knowledge? Uh, you know, I, I know there's different theories about how people can get good at predicting. Um, you know, based on your experience, what, what kind of traits do you identify here?

• 4:30Speaker 01

  So probably a big one is just like experience. A lot of our, a lot of our top predictors have actually come from other sites and have spent a good deal of time thinking about forecasting and prediction in general. Probably the first hurdle that most people have to overcome in, in order to start improving as a predictor is getting used to clarifying and crystallizing, crystallizing your views about the world. You know, if you're thinking about some political event, you can, you can very easily incorrectly, you know, misremember that you got some event right or not. But if you're not like actively putting out like numerical predictions and recording those to keep yourself, to keep yourself honest, it's very difficult to improve. So if you, you have to have, that's, that's a good first step.

• 5:12Speaker 01

  And a lot of our users already cleared that before, before joining our site.

• 5:16Dwarkesh Patel?

  You know, we, we talked about whether finance firms would want to hire these predictors, but it would be incredibly exciting if you lived in a world where, you know, news firms would want to hire the people who are very good at predicting. Scott Alexander actually has a very interesting post where he, where he, where he speculates, what if you had investigative reporters who could short, like, let's say they're, they discovered that some government has really fucked up. They could, they could like, I don't know, short the market on the government's GDP in a year and then release the information and then benefit from having that, from their investigative journalism. I don't know if you have what your reaction is to that, those kinds of ideas.

• 5:54Speaker 01

  Oh, I think that's great. I think that's great. I think that's a very obvious case of pro-social benefit of markets using, using the, the market mechanism to, or in some ways it's kind of like providing a public good or that wouldn't, you know, society otherwise wouldn't be able to produce. Getting accurate information about, you know, corporate malpractices or whatever is, is very valuable to society. Anything that we can do to encourage that is good.

• 6:20Dwarkesh Patel?

  This is the question a lot of people will be interested in, and maybe you can't comment, but is there any potential that eventually through crypto or offshoring or some other option, that the, somebody's ability to predict will somehow be able to get reflected back on them in some sort of monetary gain outside of the, you know, play.

• 6:40Speaker 01

  Oh, on our site. I think that's, I think that's, well, first of all, we are planning to do things like host tournaments with real cash prizes. You know, that's something that we are able to offer despite being a play money product. You know, we originally started as a crypto, a crypto idea before pivoting, pivoting back to web two. You know, that's funny. We're, we're one of the first firms to pivot from web three back to web two for like usability, usability reasons. But, you know, real money crypto, you know, creating a real money crypto offering is something that's on our back burner and something that's worth thinking about. That we're thinking actively thinking about. It would probably be in the form of a separate product. Though, rather than as an addition to our, our current offering.

• 7:24Dwarkesh Patel?

  I'm curious what the usability concern was there? Is it like connecting the MetaMask wallet to your account? Your, your current process is so easy. You can basically make a bet within, you know, 30 minutes or 30 seconds of visiting the site. I'm curious what the concern was there with crypto?

• 7:40Speaker 01

  I think there, there are several things. The first is just like onboarding, you know, most, you know, only a tiny chunk of, you know, the global population actually has, you know, created a MetaMask wallet or a Solana wallet or any of these things. And the process of doing that and transferring tokens to the wallet is actually pretty cumbersome and not yet very easy. So that's kind of a huge hurdle just to start with. Then you have to add in all of the other frictions associated with being a crypto platform. It takes a while for transactions to be confirmed. You have to, you know, if you're doing everything in a purely decentralized way, you will have to sign each transaction with your wallet, stuff like that.

• 8:18Speaker 01

  Each of those things is just like an additional hurdle, you know, that will prevent, prevent users from, you know, being able to bet and enjoy their experience on the platform.

• 8:26Dwarkesh Patel?

  Yeah, not to mention transaction costs, which would reduce liquidity.

• 8:29Speaker 01

  Yeah, yeah, that's, that's a big one, too.

• 8:31Dwarkesh Patel?

  Yeah. So, you know, one of the things, speaking of reputation, one of the things that seems interesting to me. My prediction over the long term is that the reputation of the market maker or market creator rather will matter a lot. Because in your website, they're the ones who get to resolve questions or resolve markets. So what is equilibrium there? Because is it just that there's going to be a few trusted market creators who people trust to adjudicate? Adjudicate what happened? Or, you know, will anybody be able to have a, create a big trusted market?

• 9:06Speaker 01

  I think it'll be kind of a bi-modal distribution where you'll see a few huge market creators that are high trust and have you know, huge volumes across a lot of their markets. And then there'll be a very long tail of smaller individuals with less of a track record. And their markets will be participated on by people, you know, closer to them on the social graph, their friends or friends of friends and stuff, stuff like that.

• 9:30Dwarkesh Patel?

  Now, here's a question I've had for a while. Why don't companies who have, you know, a direct incentive to get the best possible information on their, you know, on themselves, on the factors affecting their business? Why aren't they using internal prediction markets if this is the best way to aggregate information?

• 9:48Speaker 01

  Yeah, this, this is a great question. So first of all, a lot of companies have, have tried using prediction markets internal, including Google, GM,

• 9:59Speaker 01

  the CIA and a bunch of other firms as well. But for the most part, you find that they, they will use it, people will talk about it for a bit. They will even praise the benefits of prediction markets, but ultimately they'll abandon them. And I, I think the, the main reason for that, and the, you know, a very Hansonian point is that people literally don't want to know the answers to a lot of questions. So in a, in a corporate context, that mostly manifests, manifests itself in the manager, the manager not wanting to, you know, the manager choosing a course or vision for their company and then not wanting to get negative feedback about that.

• 10:39Speaker 01

  Even if the feedback is mostly positive, you know, the fact or like it still can like seed doubt on the part of employees. And you know, the introduction of a prediction market by itself might just like lower, lower the odds that that mission will be successful. Yeah, you know, even though one wouldn't think that just the addition of more information would actually change the outcome. But I would say that that's the biggest point. It steps on management's toes and they don't like that.

• 11:05Dwarkesh Patel?

  Interesting. It's almost the opposite of his point about consultants. So his point about consultants is that they basically put a pretty face from Harvard that they allow you to say, oh, this you know, this highly credentialed person who talks well supports what I'm doing. So that's why we should go ahead and do it. Now, one question I've always said about this Smithsonian view is even if it's true, you would expect some firms to do the painful thing or some managers to do the painful thing. I mean, you're a startup founder. You know that many of the most successful startup founders will, you know, will go ahead and do something that is, you know, quite difficult that they don't want to do. And, you know, we would expect over the long run for the market to be dominated by players who have done the thing that helps their company, right?

• 11:51Dwarkesh Patel?

  Out of all the thousands of companies that are out there. So why? Do you think in like a decade or two, all the best companies in the world will have internal prediction markets? Or is the force from managers against us so strong that that's not going to happen?

• 12:04Speaker 01

  So I would say there are two parts. I would say the first is a rational concern on the part of managers to not use prediction markets because it undercuts their mission. Part of it, it's not merely an irrational quirk of human nature. Although perhaps perhaps it is in general, but it's not an irrational quirk of the manager's part in choosing not to use them for certain contexts. The second part is they're actually, I think they're the the areas where prediction markets more clearly add value without distracting from the mission or hurting management are where they're used for for their informational aspects or to do like market research, you know, questions where you need to like survey, a broad survey of like what consumer behavior will be, how our competitors will behave, that sort of thing.

• 12:48Speaker 01

  I think that's a much clearer example of where prediction markets can provide value in a corporate setting, which, you know, you know, doesn't fall prey to most of these downsides. So and then the question becomes like, why, why hasn't that happened yet? And I would say a part of that is just like a part of that is actually just a usability usability concerns. It's actually very hard to create a product that's extremely simple and easy for like all of your employees to use that doesn't require too much thought, but can still yield like reasonable results. And I would say part, you know, part of our goal with manifold really is to like bring the bar down lower such that more and more people can participate in prediction markets and it be like a pleasant, fun and easy to use experience.

• 13:33Dwarkesh Patel?

  Yeah, yeah, that it definitely is. You know, I was about to suggest, you know, maybe it would help Zuckerberg to know, maybe it's like subsidize a market to find out how many VR devices there will be in the world by 2030 to help make, help him make his, you know, metaverse plans. But maybe, you know, maybe he doesn't want to know that, or maybe at least he doesn't want that publicly known. So, you know, one question I have is, in what situations is it best to have a single person compound all the available knowledge? And is it, in what cases is the best to kind of decentralize the decision-making process by having a market?

• 14:13Speaker 01

  I would say that in general, like if you're making a decision in the real world, it's often very difficult to operationalize what the success criterion is down into one thing that the market can then measure and measure and be used to act upon definitively. I would say so that that's kind of the biggest, the biggest case for, you know, like personal discretion or like human involvement in decision-making. Or in general, I think like prediction markets work best when people are creating, creating markets on a variety of metrics and then using human intuition to figure out what the best course of action is to proceed using that information.

• 14:53Dwarkesh Patel?

  I'm curious, what is your opinion about insider trading and Congress, for example? I know this is like a big debate. Do you think it serves like a useful price discovery function or too much of a hazard of adverse selection? Kind of a tangential point, I'm just curious about your opinion.

• 15:09Speaker 01

  I would say in general, my my view of insider trading is, I basically have Matt Levine's view is that it should be viewed as a like taking from shareholders rather than a like fairness or equity type of discussion. I I do think in general, like insider trading obviously does benefit price efficiency, but there there are other things that we care about. We also don't want in general, our corporate shareholders to be profiting from, you know, from certain types of information and all those. In the case of like, you know, Kong or like, you know, congressional betting or insider trading by elected representatives.

• 15:49Speaker 01

  I think that's for the most part not not cool. We would prefer for them to be compensated directly through salary or other things which are more open and transparent to the public rather than, you know, you know, this is kind of a way of profiting or, you know, like a roundabout way of of corruption in some sense. You know, you can make a deal with some company and then trade on that. And instead of the money that never directly passes into your hands, but you're privy to this insider information. And the net effect is is, you know, equivalent to this practice, which we would not normally condone.

• 16:25Dwarkesh Patel?

  Now, one place where I'm skeptical of prediction markets is when we're talking about questions that resolve over a long period of time and which the which there's no good base rate for, so the market hasn't really been trained on that kind of question. One example would be what are the odds of catastrophic AI by like 2030, which I'm sure is a question on your website. And I'm not sure how much credence I put into the prediction on this question. I mean, one, because I'm not sure why I would expect that much

• 16:58Dwarkesh Patel?

  good information to be aggregated here in the first place. Would it just be, I mean, what is the information they're going off of? And and second, I guess more important is like, you know, why would somebody bet? Yes, because the best case scenario, if they bet S is they die anyways, right? They have nothing to gain if catastrophic AI happens. So in these kinds of scenarios, what do you think is the role of prediction markets? Do you have these kinds of concerns? What are your thoughts?

• 17:25Speaker 01

  Oh, I think, yeah, both of those concerns are very valid. I think a part of this is just like a fundamental like human constraint is if things are very far in the future, we care about them less. It doesn't really, you know, that's true in prediction markets, but it's also true in every other domain as well. And every other forecasting tech technique as well, you know, you know, the the interest rate is like a fact about human nature. Or like discounting of the future is like a fundamental facet of human nature. And you can twiddle with the mechanism here on the margins here or there, but I don't think it'll fundamentally change that. For your for your second point, though, your other point is is very well well taken, too, that you can't you can't like bet bet on an apocalypse.

• 18:10Speaker 01

  You know, you're not going to be there to to collect your winnings. The money is worth less in that universe, so you should rationally bet against it.

• 18:19Dwarkesh Patel?

  Actually, so Brian Kaplan and Eliezer Rudowski actually do have a bet on an AI apocalypse. The bet is about whether there will be like, you know, catastrophic AI, I think by 2030 is the exact bet. And so Brian pays Eliezer now, and if by 2030, there hasn't been catastrophic AI, then Eliezer pays Brian. You know, the the sum that Brian paid plus more based on like the ratio of their bet, which I thought it was a which I thought was an interesting idea. I was just going off this topic a little more. You know, one of the things that Scott Alexander brought up in one of his blog posts is when you have these kinds of bets that resolve over the long term. It doesn't have to be as like as far fetched as, you know, AGI. It could just be something like who's going to be president in 2024.

• 19:04Dwarkesh Patel?

  He pointed out that, at least at the time that he wrote it, Dwayne Johnson had like a 9% chance of being elected president, which seems high. Except why would somebody get their manifold dollars held up in a question that won't resolve for four years? Or I mean, two years now. The ability of speculators to correct prices is diminished because they don't want their money to be held up in questions that won't resolve for a long time. How are you guys dealing with that kind of problem?

• 19:34Speaker 01

  I would say like, so in general, the best way to tackle these long-term, more speculative questions is to try to break them down and address the proxy variables that are most relevant to, you know, to predicting their success in the future, you know. So instead of asking like, will the AI apocalypse happen in 2030 or like in 2050, you can ask like, will OpenAI's GPT-4 like exceed, you know, this number on this benchmark, etc. Or similarly for like the Rock, like, instead of asking like, will the Rock be the president in like 12 years or whatever, 20, whatever, you could ask like, what are his polling numbers today? Like, what are his favorability ratings? Has he like, will he indicate interest in

• 20:18Speaker 01

  running, show interest in running as a politician, etc., etc. And those are much more short-term questions which you can use to, you know, get a sense of, you know, of the longer, the longer-term, more speculative questions that, you know, that interest you.

• 20:34Dwarkesh Patel?

  Okay, so, you know, Tyler Cowen has this other criticism of prediction markets that, listen, these prediction markets are tied to financial markets that you actually could bet on, right? So I don't know if you think somebody's going to be, has a higher or lower odds of being president, maybe that has effect on, you know, GDP or something, and that has effect on, you know, stock prices, right? And so then he says, well, if you're so good at doing, so good at betting in prediction markets, why don't you just bet in the financial instrument that are linked to whatever question you're interested in? And Brian Kaplan's response to this is, there's so many different things that can affect any given financial instrument that it's not, it's, you know, it's not easy to explain, okay, because I

• 21:19Dwarkesh Patel?

  think A about, you know, you know, Dwayne Johnson becoming president, I can invest in asset B. I feel like that defense of prediction markets is a criticism of your response that you just gave, where you're saying, short-term prediction markets can substitute for your interest in long-term prediction markets.

• 21:40Speaker 01

  Uh, I would say yes and no. So like, uh, asset, you know, asset prices of, uh, you know, of major, major financial commodities and equities and other instruments, you know, are affected by way more factors than another prediction market on a more isolated question on some proxy variable that you care about. You know, the, the nice feature of prediction markets is that you really can isolate the particular risk that you care about, you know, and even if it's just discussing like one proxy variable, um, that's related to the overall picture, you can still get a much better understanding of the thing that you care about by picking multiple well targeted like proxy variables and creating markets on those rather than like oil prices, you know, which tell you many things, too many things.

• 22:25Dwarkesh Patel?

  Okay. So lay out your vision for me of like 10, 20 years, you know, we have, uh, prediction markets are not only, you know, much more liquid, more people participate in them. And it's kind of, uh, everybody kind of knows what a prediction market is. So like people understand what it means when you say, um, you know, this prediction mark, uh, you know, Biden, I guess it wouldn't be Biden then, but whoever the president is passing this bill had, uh, huh?

• 22:51Dwarkesh Patel?

  Well, one of the many things that would need to happen other than, you know, a constitutional change would be, um, major advances to longevity research. Um, oh yeah. So what, what would the world look like? Um, if people actually cared about, uh, you know, prediction markets and if it was like common knowledge, uh, what the market thought about any given situation that was happening?

• 23:13Speaker 01

  Well, I, I would say, first of all, I would say that would probably be a much better world, you know, um, you know, across a wide variety of domains. I think it's, it's, it's better, better and nearly, but not all cases when people are grounded in terms of, of, of facts and like well calibrated predictions instead of like pure speculation or like, uh, you know, ideologically biased thinking. Um, but you know, more, more concretely, I can imagine a world where you, you know, the average like news article, blog posts, CNN station, if that's still around in 10 or 20 years, has like an embedded. Prediction market on whichever topic that they're discussing. And it'll just like allow people to, you know, immediately get a sense of what like the most grounded.

• 23:59Speaker 01

  Or like it'll it'll immediately ground the opinions that people are hearing in fact. And make it easier to have more productive conversations. And ultimately allow them to better understand the world.

• 24:11Dwarkesh Patel?

  I I I I hope that's the case. You know, there's a pessimistic take that people are not consuming politics to understand what's happening. Or and that they they'll almost resent you for presenting them with this kind of information.

• 24:26Speaker 01

  That's true, or certainly I'm I'm not implying that. A large part or even like a majority of the consumption of political news is based on a desire for accurate information. You know, I think that would be a very, very naive view of how, you know, humans operate. But but at least a small part of that is, you know, at least some portion. Of people's desire to consume news actually is to legitimately understand the world. You know, and insofar as people are attempting to do that, you know, prediction markets can can help.

• 24:58Dwarkesh Patel?

  I I really hope so, and I I think that actually could be true, which is very exciting. So one question I have is. These trades that people do on your market, you know, four percent of the four percent of the what is traded. Or the traders winnings, they go to the market creator and one percent are just burned. Correct me if that's wrong. But that seems like a negative sum. Bet, which means that are you concerned that that would reduce liquidity? Because people people have to get over a higher threshold of confidence before they're willing to bet on any given market.

• 25:39Speaker 01

  Yes, yeah, like your fee fees impede do impede the efficiency of the market in some sense. You know, there are people who on the margins would trade if there were no fees who don't when there are fees. That that certainly is true. But part of the reason why we have fees is like to encourage things that we like. You know, we have a creator fee, the creator earns a commission on trades that incentivizes them to create and resolve more markets. We currently have a liquidity fee as well, which feeds into the liquidity pool, which you can think of as subsidizing the entire market.

• 26:12Dwarkesh Patel?

  I'm curious, what is your background? So one of your co-founders is your brother. Do you guys have some sort of financial background or mathematical background? Because I was looking at your technical documentation. And I'm not claiming to have understood most of it, but I saw the formulas that were there. I visually skimmed it. And how are you guys able to get into this field?

• 26:39Speaker 01

  So I guess I studied computer science in school. Then I went to work for SIG or Susquehanna. It's an options trading firm for a little bit. Then I left to go work for my friends like RoboAdvisor startup, where I wrote their portfolio optimizing software. Yeah, so I have some financial experience as well. And all three of us are technical.

• 27:07Speaker 01

  And also have a previous background doing other entrepreneurial things as well. So we're all, we're kind of like full stack entrepreneurs, I guess you could say.

• 27:16Dwarkesh Patel?

  Can you talk a little bit about your previous entrepreneurial experience? I think listeners might be interested.

• 27:21Speaker 01

  Sure, so right before this, my brother and co-founder, James and I, were working on our app called Throne. It's a subscription group chat app for online creators. So people like you, or like Instagram creators, or DJs, or anyone who's built up an audience online. The idea was, they can set up a private group chat where their audience, where they're paying a subscription fee to join. And basically, the premise of our app is that we provide a much better chatting experience for larger groups. So it's designed to accommodate the creator's entire audience and make that a seamless experience.

• 28:06Dwarkesh Patel?

  Yeah, I have a friend who has a popular fantasy football channel and he has, you know, he has a very profitable Patreon where basically from the point of the Patreon is it'll give you the link to the invite to the Discord, which is, which seems convoluted. And this product, we're just basically combining those two services. That seemed very useful. So what happened with that?

• 28:30Speaker 01

  It's still going slightly. Basically, we couldn't get the growth that we were hoping for. The ultimate cause is it's kind of too late. Like, the creator market is already saturated with all different sorts of tools. Most creators are already monetizing one way or another and aren't keen on moving, monetizing using a new platform.

• 28:51Dwarkesh Patel?

  So when it comes to the manifold markets, I mean, the idea for having prediction markets, even the idea of having prediction markets to play money, I believe, has been around for a long time, right? What took so long for somebody to make a user experience that was so comfortable as yours? Or is there something else that, you know, prevented somebody from making a manifold market much earlier?

• 29:13Speaker 01

  Yeah, I would say the other key piece of the puzzle that other people were missing is the idea of user resolved markets, which on the surface sounds a little bit crazy, that you just allow anyone to come and create a market on any question, where they are also the judge of that question and can resolve it in any way that they want. You know, that obviously opens the door to fraud and abuse and people taking advantage of the system. And the fact that there is a possibility of fraud basically has prevented other people from like even exploring this option. Instead, they're opting for oracles or centralized authorities deciding the outcome of all markets. But of course, that like severely limits the scalability and reach of, you know, of a prediction market platform if users can't

• 29:57Speaker 01

  go on and create their own thing. So it's really realizing that user resolved markets actually can work that led us, led us down this path. You know, there is a small amount of fraud, but it's actually quite small and manageable. If you, you know, allow users to choose which markets to participate in, they for the most part are making pretty good decisions about where, where to allocate their, their time and money.

• 30:19Dwarkesh Patel?

  There's this great essay called Unix is Worse is Better. I don't know if you've gotten a chance to look at that. The basic point the author makes is, you know, if you ask like, why has Unix been so successful? You can say, oh, the, some of the features and some of the design decisions are arbitrary. They're somewhat inoptimal in certain situations. And then, but then you, it seems that maybe it was those bugs that actually allow Unix to be the kind of thing that can actually run on real computers. Maybe it's not, maybe it's not optimal from the, the perspective of somebody who wants something super elegant and beautiful. And, you know, Gurren has a really good blog post about this, Bitcoin is Worse is Better,

• 31:02Dwarkesh Patel?

  where he makes the same point about, you know, Bitcoin, like some of the, I'm not remembering the details right now, but some of the, you know, some of the actual constants involving like, you know, block size or whatever. They're pretty arbitrary, but the fact that just somebody just put out the arbitrary numbers meant that innovations that could have been around years before in terms of cryptographic tools actually got instantiated in like a, in a product like Bitcoin. And it kind of seems like you're taking a similar attitude towards prediction markets. Even if it's somewhat arbitrary to let users resolve markets. You know, that's better than having some sort of oracle structure, which is super convoluted and hard to use. And it may sound good in theory, but is kind of impractical as something that people want to use.

• 31:47Speaker 01

  Yeah, that's, that's exactly correct. I think the, the most important thing for, you know, not just for us and not just for prediction markets, but for creating like a usable platform is making sure that the key mechanism is something simple and easy to understand, more so than handling every possible edge case perfectly. You know, as long as it's, the core experience is extremely simple and easy to use and user friendly. You know, that, that matters much more than like the 1% of cases, you know, where, you know.

• 32:20Dwarkesh Patel?

  I'm curious if this is something you learned at your previous startup. Did you guys make this mistake and now you've learned to avoid it? Or curious how you came to that realization?

• 32:32Speaker 01

  You know, I don't know. I don't know that I can think of a good example that's directly relevant. Although we did like simplicity of, I guess, like both the business model and our user interface for our previous chat app were big considerations. You know, the idea of having or like a subscription model is kind of like the minimum viable business for like a creator to support themselves in a real way over time. That might be one manifestation of this principle in action, but I don't know.

• 33:01Dwarkesh Patel?

  Okay, so one concern somebody could have about manifold markets is, you know, you said you were giving out 1,000 manifold dollars. Somebody could, you know, just maybe even make a bot that, I'm not giving anybody ideas, but make a, just like create a thousand different accounts, make different trades on all of them. And or make a sequence of different trades on all of them. And in one of them, they just make, you know, a series of the best trades that anybody has ever made on the platform. And now they're on the top of the leaderboard.

• 33:33Speaker 01

  It's certainly true that if you're giving away free money, you know, that's not a, you know, economically sound mechanism. We currently do do some things to prevent abuse and like bot behavior. I'm not going to tell you what those things are.

• 33:47Speaker 01

  If we end up open sourcing our code, though, the cat will be out of the bag. Because you can, you know, see for yourself what abuse practices we have in place. I would say that we're not necessarily committed to maintaining our, you know, free giveaway or like free signup bonus indefinitely. You know, part of that, that's definitely partially a, you know, a growth and like just starting off thing. In the beginning, most of our users are, have been very well behaved and haven't abused, abused this system. You know, if more and more people, if manifold becomes like a, a key nexus for people trying to weigh in on the future, we may have to be more strict with our policies.

• 34:26Dwarkesh Patel?

  It kind of reminds me of the PayPal story where they had to, they were giving like a 10, $20 bonuses for people for signing up. And it basically meant they got exponential growth, but they were losing money faster. I guess in this case, you're not like losing real money, but I don't know, maybe you're diluting, the diluting the reputational value that the best people in the leaderboard have, at least for now, maybe.

• 34:48Speaker 01

  Yeah, I would say one of the other things that we're interested in, or like the, the concept of a global leaderboard becomes less important over time, the larger our site gets. You know, a lot of the markets that people bet on are personal markets or markets between friends. You know, what you as an individual care about um in terms of, you know, ranking who the top traders or predictors are. Really is not all of the markets that we have in our platform, but the subset of markets that interest uh that are of interest to you. We, we actually just remove or like we uh yeah, used to have this feature called communities where you could um. Create a community centered around a collection of different markets and we would show the leaderboard just for those markets.

• 35:29Speaker 01

  Um, you know, that's a way of judging uh, you know, or it's a way of assessing the ability, the ability and skill of predictors within some particular domain that you care about. Rather than globally, we'll probably bring that back in some capacity in the future.

• 35:44Dwarkesh Patel?

  Uh, can you, can you give an example of that? Like, I guess something like fantasy.

• 35:47Speaker 01

  Yeah, so one, one thing you might care about is like, who is the best, like Russia, Ukraine? Who is like the best, like, uh, you know, you like geopolitical and war predictor? You know, and what the one way to assess that would to just be to like handpick a few markets. Um, about the invade, about the invasion and you know, the, you know, subsequent, you know, course of events during the war. And then um, just see how much profit people have made within those markets. And I think that that'll actually give you a reasonable estimate of um, you know, who, who the top predictors are there. Versus just like a global thing, you know, uh, for a while, one of our like top uh markets on our site was.

• 36:25Speaker 01

  About whether this guy uh would allow like, whether this stray cat would uh allow it's like this random human to pet it. You know, totally, totally random stuff like that. You know, if you're, if you're trying to assess who the best like geopolitical thinkers are, who the best economic thinkers are, you may not care about the stray cat. Um, but a lot of people really do care about the stray cat uh in, in other contexts, so. It's a, it's a fine balance to strike.

• 36:52Dwarkesh Patel?

  Another interesting part of your platform is so people can buy more play money. And I guess the concern is even within these, even within these subdomains where people are being compared, that, you know, the actual, the actual proficiency of a person could be distorted by the fact that somebody else could just buy more manifold dollars. And as a result, maybe able to have a higher profit. I guess you can measure it maybe by, you know, percentage gain on the total amount of money or something. But yeah, I'm curious on your thoughts about this.

• 37:32Speaker 01

  So in, in general, I'm a pretty pro efficient markets. You know, if you think about what's happening when a whale comes in and buys up a huge amount of money and like bets it on the wrong side, what they're essentially doing is just like providing a huge subsidy and bonus to the people who are actually correct. You know, market, markets are a mechanism which is like the most efficient way of solving the whale problem. If you are too much money and you are not very smart and are very confident in your views of the world, like the markets have a very fast and easy solution for you to solve, to, you know, to alleviate that problem where you will no longer have so much capital, you know, to, to, you know, throw about.

• 38:14Dwarkesh Patel?

  There's a direct analogy to this, actually, as I'm sure we're in, you know, like real markets where like a pension fund or some other sort of big fund needs to like rebalance its books or something. So it'll, it'll need to just like dump a bunch of stock and, you know, liquidity providers can make money off the big trades that the, you know, that these big funds need to make. So I guess then you wouldn't be, you wouldn't be interested in schemes like, I guess quadratic, like it's some decaying, some, some decaying value of the manifold dollars that you can buy in one account. So like the net, the marginal, the marginal manifold dollar costs more than the one before it.

• 38:58Dwarkesh Patel?

  You, you think this would inhibit price discovery, right?

• 39:01Speaker 01

  Yeah, although I have I have toyed with some other interesting like monetary schemes in the past. Um, so one one idea I had is introducing demurrage or like, um, basically negative interest rates on cash balances. So like you could imagine a world where like, uh, where the your your purchasing power kind of erodes over time. You know, if you made a bunch of like really good predictions this year, you know, it's not necessarily the case that you should be like rewarded by, you know, having all of that cash forever, you know, maybe some portion of that uninvested, those uninvested cash balances erodes by, you know, 20% a year or something like that. Um, I think it's a really cool idea in principle, but in practice, like users absolutely hate,

• 39:45Speaker 01

  um, you know, losing money for any reason at all. So it's kind of like a nonstarter.

• 39:52Speaker 01

  But it's it's interesting to think about.

• 39:54Dwarkesh Patel?

  Sorry, I I'm not sure I understood. If the reexplanation is unnecessary, then we could just cut it out at the final, but just for my benefit, can you can you clarify what your, uh?

• 40:04Speaker 01

  Sure. So so the idea is like, um, uh, at certain times you'll have just like uninvested cash, you know, that's sitting in your account.

• 40:12Dwarkesh Patel?

  Yeah, yeah.

• 40:12Speaker 01

  Um, the the idea is to just like charge you, charge you for that. So charge a charge a sales tax at the rate of like 20% a year. But like maybe like once per once per day at midnight, you lose, you know, whatever whatever amount that is. A small fraction, a a small fraction of your your uninvested cash.

• 40:33Dwarkesh Patel?

  Um, okay, so what what you know, one thing that makes, um, financial markets really efficient is you have these big firms that are putting up in large amounts of capital to recruit the top talent in the world. Um, because it's worth it for them, you know, they're they're running like these supercomputer simulations. Um, and, um, you know, they like it's expensive to do all this stuff, right? Um, yeah, but it's, you know, it's worth it if you're getting like point zero zero one percent of a trillion dollar market. In the case where you have play money, what is the incentive for? I I get that people who are hobbyists or are in other ways motivated to do this might choose to do it and like choose to engage in it, um, themselves.

• 41:15Dwarkesh Patel?

  What what would incentivize somebody to put in the kind of effort it takes to uh, have efficient, uh, financial markets and, you know, cost and time and so on?

• 41:26Speaker 01

  Yeah, so I would say that in general, people are willing to put a lot of time and effort into virtual economies irrespective of financial gains. You know, if you look at things like World of Warcraft or earlier like Second Life, which has like virtual real estate. People are willing to, you know, invest the equivalent of hundreds of thousands or millions of dollars in in in this game game world. I feel people's, you know, people's ability to drill in and focus and work on things that are of interest to them. You know, is just like a very, very powerful force, even if it's not directly tied to a financial payout.

• 42:03Dwarkesh Patel?

  So what is the point by which manifold's own internal decision making will be informed by prediction markets involving, I don't know, the firms, predictions about the firm?

• 42:15Speaker 01

  Oh, it is already. We, you know, we actively create markets on all sorts of things which are of interest to us, you know, most. So most recently, we created a market on whether we would be able to complete our fundraising round by the end of April. You know, that was a cool market. The market, the market believed in us for the most part. I don't think we like fell lower than like eighty five percent, somewhere around there. For, but we, we, you know, internal, we, we, we actually do try to dog food, dog food our markets as much as possible. We have a new market now and whether we'll be able to onboard three employees, I think before the end of June, something like that. You know, that that's a number that we like try to keep our eyes on.

• 42:57Speaker 01

  I guess probably the first, the first big case where like the market actually, you know, substantively informed our, our decision about how, how to act as a company was an early market we created on whether, whether we should try to monetize by selling the fake, by selling our, the play money. You know, we initially, we weren't sure whether that would be a good thing to do at all, whether users would hate it or it would seem scammy or whatever. But we, we created a prediction, a prediction market on this subject. I think free, phrased in conditional terms that if we did introduce this, would we, if we introduce the feature, would we keep it for, you know, some period of time? And the, the market, the market seemed to think that we would.

• 43:39Speaker 01

  So that actually was a factor which played in, played into our decision.

• 43:44Dwarkesh Patel?

  Yeah, I'm asking these questions in probably the wrong order. So this is probably one of the initial questions I should ask you. But talk me through the timeline of developing manifold market. So how long ago was it launched? Um, and like, what were you? You said in the beginning, it was supposed to be, um, based on crypto. When did you guys change your mind? But what is the timeline here?

• 44:04Speaker 01

  Yeah, so this company began long, long ago, way back in December of 2021. Yeah, yeah, it's a very, very new company. Basically decided to ditch crypto after like a few, maybe like a week, a week's worth um, of, you know, speculation and research. Mostly because we, or part of it is that we just, uh, thought we could build a play money prototype very quickly and we could just like test that and see, see what the experience is like. And if we wanted to, we could continue further on into crypto. Um, and partially is just based off of, uh, you know, re- reading about the, you know, regulatory nightmare that is, uh, prediction markets and, and crypto.

• 44:48Speaker 01

  Um, and partially is the, the usability concerns with crypto. Um, but basically we, uh, we all like in the month of December, we came up with the original idea. We applied and received, uh, the ACX grant from Scott Alexander. Um, and we timed it such that right, right when the, uh, the grant announcement came out, we had a working prototype. Um, of our prediction market system ready to go. And then we were able to on, onboard a bunch of users from the ACX community, um, immediately, which formed the, the base of our platform. You know, and, and since then we've been growing and, you know, respond, you know, uh, you know, responding to user concerns and improving,

• 45:32Speaker 01

  uh, improving our site to be ever more usable.

• 45:35Dwarkesh Patel?

  I know you guys use something called Dynamic Paramutual Betting System. You know, I looked into the paper. I think I understood like maybe a quarter of it. So is, was it like a experience of learning as you went? Or is it something you guys are already familiar with?

• 45:51Speaker 01

  No, definitely the former. You know, we had heard, I had heard of Uniswap and some other mechanisms. I had heard of like Hanson's log market scoring rule and some other things, but I didn't actually delve into the details until quite recently. You know, originally, we actually came up with the idea for Dynamic Paramutual, by ourselves, thinking about it from first principles. And then we later, like, I went back through the literature and read a bunch of papers and realized that it was called Dynamic Paramutual and found a more elegant implementation of a Dynamic Paramutual system based, you know, based on a quadratic cost function

• 46:29Speaker 01

  that I hadn't considered previously.

• 46:31Dwarkesh Patel?

  Yeah, you'll have to talk me through it over dinner the next time. Yeah, so you mentioned that you are hiring. Maybe, maybe there are some people in my audience who might be interested. So you want to talk about what kinds of roles you're interested in hiring for? Other kinds of things that might be involved in recruiting?

• 46:49Speaker 01

  Sure, yeah, yeah. So we, right now, we're looking to onboard a few people to our team. We're looking for full-stack developers who preferably have front-end, a lot of front-end and potentially React experience. We're looking for a community manager. It would be someone to help us manage our Discord, help write blog posts, help with our Substack, help reach out to people on social media and organize online events. You know, someone with no previous background in doing this at a startup would be great, but we're open to people from other backgrounds as well. And then we're, then we're looking for a head of growth or someone who has, you know, experience scaling up, scaling up an early to mid-stage startup.

• 47:34Speaker 01

  You know, we're looking for, you know, someone who, you know, who's, who's helped, you know, take the company from, you know, like... Uh, to like one, like one to 10 million dollars annual recurring revenue or around like 50,000 monthly active users, something somewhere in that neighborhood. But yeah, if you, if you are interested in prediction markets, um, and manifold in particular, I encourage you to reach out to us. You can either email me at jo- or e- email us at jobs at manifold dot markets or me personally, Steven at manifold dot markets.

• 48:05Dwarkesh Patel?

  Excellent, excellent. Um, and any other topics involved with prediction markets or manifold that we have not discussed yet that you, you would like, you, you know, you want to touch on?

• 48:15Speaker 01

  Hmm. I guess there, um, one of the things that we've seen from launching this site is that users have come up with all sorts of like exotic ways to use prediction markets. That we hadn't previously considered. So part, part of that or like prediction markets obviously can be used on, you know, predicting the future, basic yes or no questions. But some exotic things you can do are like, uh, getting users to help like research topics for you. Um, of in, uh, you know, do, you can create a market on phrase these types of things in a market based way. And then people can propose certain answers and other users can bet on them. That's an interesting feature of our site.

• 48:51Speaker 01

  People have hacked our platform to do things like create lotteries, um, to play games. We did manifold plays Wordle where we created a series of prediction markets on which word we should guess next. Um, you know, people, yeah, there, there are just like all sorts of other things, non-prediction type things you can do with a, with a, this market mechanism, which are really cool.

• 49:11Dwarkesh Patel?

  That is an interesting idea. Um, I'm not sure how th- that was conducted, but like I could imagine, uh, for, you know, researchers saying, um, will, will my conclusion on a paper of this question be, um, X or Y? And, um, and then to move the market, somebody would not only have to bet, but also like maybe put in the comment section, the data that makes them think it's X or Y, which can be the basis of the paper. Uh, so you're basically getting research assistance. Um, that, yeah, that is very interesting.

• 49:37Speaker 01

  Oh, yeah, yeah, I would say relatedly, one of the the new types of markets we've introduced is the this concept of a free response market. You know, so in addition to a yes or no market where you're just betting on something will happen, we have this idea of a free response market where you ask a broad open-ended question and people can submit an answer. And when they're submitting an answer, they're also placing a bet on that answer and other users can bid, you know, bid up, you know, different answers that the the market creator will then will then will then choose, you know, and that really opens up the, you know, the space of possibilities towards allowing, you know, much wider wider array of questions.

• 50:17Dwarkesh Patel?

  Stephen, thanks so much for coming on the podcast. And yeah, so that's manifold dot markets, right? Any other places or links that people should be aware of?

• 50:29Speaker 01

  I think I think that's the big one. Manifold dot markets, you can join right now and we'll give you a thousand manifold dollars to bed on any market you like or create your own.

• 50:39Dwarkesh Patel?

  Excellent. Thanks so much, Stephen.

• 50:40Speaker 01

  All right. Thank you.