Playback speed
Share post
Share post at current time

#16: Stephen Grugett and Austin Chen

Manifold, Manifund, Manifest, prediction markets, and EA

Stephen Grugett and Austin Chen are co-founders of Manifold Markets, an online play-money prediction market and competitive forecasting platform. Stephen currently serves on the company’s management, while Austin recently stepped down to start Manifund, a unique, open-source grant program. This video is not sponsored in any way by Manifold, Manifund, or Manifest - I just think they’re cool.


0:00 - Intro

Stephen Grugett

1:20 - Are prediction markets actually bad?

4:11 - Would Manifold use real money if allowed?

5:24 - How Manifold would use real money if allowed

6:08 - Would Manifold use crypto if allowed?

7:17 - Can you ever get long-term returns from prediction markets?

10:01 - Would subsidies ruin markets?

11:23 - Why Manifold beat real money on predicting the 2022 elections

16:00 - Would Stephen implement futarchy?

19:54 - Manifold Love

23:22 - Bet on Love

26:21 - Why Manifold is miscalibrated

29:06 - Insider trading and market manipulation

31:42 - Is it easier to make money on prediction markets or normal markets?

32:37 - Good prediction market UI

34:35 - Why should people trust market creators?

35:34 - Derivatives on prediction markets

37:20 - Stephen’s ginseng adventures

40:55 - Audience Q: why don’t Americans consume American ginseng?

41:35 - Audience Q: cancel culture and Richard Hanania

45:50 - Audience Q: why aren’t there more institutional investors in prediction markets?

47:33 - Audience Q: can journalists help resolve markets?

49:45 - Audience Q: is there any role for sweepstakes other than regulatory arbitrage?

Austin Chen

51:14 - Are prediction markets insufficiently powerful?

54:22 - What prediction markets can do if not futarchy

55:36 - How Manifund was designed

59:35 - How Manifund chooses regrantors

1:00:49 - Why donate to Manifund?

1:03:09 - Does Dustin Moskovitz have too much power over EA?

1:04:29 - What Manifund would do differently with more money

1:05:52 - How Manifest gets so many interesting people

1:09:10 - How much did SBF’s fall damage EA?

1:10:04 - OpenAI

1:11:54 - Is this decade more important than other decades?

1:13:01 - Why aren’t more philanthropic organizations open?

1:15:35 - Manifund’s best projects

1:17:25 - How short AGI timelines would affect Manifund

1:19:21 - Audience Q: how Manifold ships fast

1:22:11 - Outro





Manifold’s Twitter:

Manifund’s Twitter:

Austin’s Twitter:


More Episodes



Apple Podcasts:

My Twitter:

My Substack:

Theo: Welcome back to episode 16 of the Theo Jaffee podcast. Today, I have the pleasure of speaking with Stephen Grugett and Austin Chen, two of the co-founders of Manifold Markets, a play money prediction market company. Prediction markets are like financial markets, except instead of betting on stock prices, you bet on the outcomes of future events. On Manifold, you can bet on markets created by other people or create your own on any topic you want. Manifold has all kinds of markets from who will win the 2024 presidential election to will AI destroy the world by 2030 to what will happen next in the manga One Piece.

This is a very special episode of the podcast, my first in-person interviews done live at Manifest 2024, Manifold's annual conference. The first interview with Stephen goes in-depth on Manifold itself, the theory and practice of prediction markets, Manifold love, and Stephen's background as a ginseng merchant. The second interview is with Austin. Austin recently left Manifold to start Manifund, a unique, fully transparent grant program. In our interview, we talk about Manifund, effective altruism, and the EA funding ecosystem. I had a great time at Manifest, and these interviews were some of the highlights for me. This is the Theo Jaffee podcast. Thank you for watching. And now here's Stephen Grugett and Austin Chen.

Part 1: Stephen Grugett

Theo: Welcome back to episode 16 of the Theo Jaffee podcast, part one. This is my first ever live recording. We're here live at Manifest 2024, and I'm interviewing Stephen Grugett, the co-founder of Manifold with a live audience.

Stephen: Thank you. Thanks for having me on. I'm super excited to be on your podcast.

Theo: Awesome. Thanks.

So for the first question, Works in Progress just wrote an article called Why Prediction Markets Aren't Popular, which argues that, contrary to the traditional view that prediction markets aren't popular just because they're regulated, prediction markets are actually quite legal in the U.S., and Calshi and others are able to do them. And the reason they don't work is that they just aren't very good. So aside from being zero-sum, they're usually quite small and quite illiquid, and that it would take expensive subsidies to make them large and liquid. And also, one of the reasons that Austin Chen laid out in his leaving Manifold document is because he thinks prediction markets feel insufficiently powerful. So what do you think about that?

Stephen: The first thing is I think the premise is not true. One of my favorite prediction market facts from Robin Hanson is that the turn of last century, prior to the 20s, there was more trading on prediction markets on U.S. presidential elections than there was on the stock market. Average Americans were speculating on these sorts of political contracts, and it was hugely popular. So I think that certainly, within the U.S., we would see huge volumes on at least election markets by themselves if they were legal. That's totally a regulatory issue.

I think there is the other question, though, of other use cases besides election speculation. There are, right now in the U.S., some limited regulated markets on things that don't touch on these subjects, and the volumes on these contracts haven't been that high right now. I think part of the reason for this is not necessarily an inherent lack of interest on the part of the public, but the fact that there hasn't been a platform that makes it really easy and simple and engaging enough for the public to consume. So that's one of the things that Manifold is trying to address.

And I think this just takes time. The regulatory barriers for prediction markets have prevented adoption in the past. I would guess that in a counterfactual world without any regulation, you would have seen a platform like Manifold arising much earlier with real money with very large liquid markets. That's a much larger part of public discourse.

Theo: So if prediction markets were fully deregulated, like, tomorrow, would you leave Manifold entirely based on mana, or would you make it real money, or would you make a separate real money prediction market?

Stephen: I think we would have both. So I think one of the things people don't get about play money is that it's not just an inferior version of real money, but its own thing entirely, and that it comes with a number of advantages. So the benefit of play money is that it's just way more casual and frictionless for people to consume. If you want to get someone to sign up for a real betting platform, that can be difficult. People have all sorts of psychological barriers. They don't want to invest their money. But when it's simple and a game and doesn't come with any financial commitments, it's much easier for people to participate.

There's that, and then there's also just the freedom aspect. You can do anything you want with play money. The moment you introduce real cash into the mix, then all sorts of regulations and know your customer and anti-money laundering regulations come into play that make life very difficult. So I think even a world where real money is fully legal, you probably, you would still see a large play money platform catering to this other source of consumer demand.

Theo: Have you thought about extensively what Manifold would do with real money if you could?

Stephen: I've obviously thought a little bit about this. I think we would spin up a separate USD-denominated version of many of our markets for people to trade on. I think even in a world where it is legal, you would expect pretty substantial regulations. So I would imagine Manifold USD-denominated markets would be much more severely limited and on fewer topics than our play money markets, but we would definitely have been creating as many as we could.

Theo: But what about crypto? If there were no regulatory barriers, would you make crypto prediction markets or is there just too much speculation in crypto?

Stephen: Crypto, in addition to having the same regulatory issues that all real money markets would be, has the additional burden of being much harder to use. One of the reasons for crypto in the first place is this kind of regulatory arbitrage thing where people turn to these decentralized mechanisms precisely because certain types of contracts cannot be enforced in a court of law in the U.S. But I'm more skeptical on this fuller Web3 vision where everything would have its own token and everyday Americans would be actively engaging on the blockchain. I think that's less likely due to how cumbersome and difficult it is to use these sort of products. So I actually think in a world with more liberalization and fewer regulations, you would just see way fewer people using crypto, both in prediction markets and in general.

Theo: Do you think prediction markets are fundamentally by their nature zero-sum permanently? Or do you think there will be an equivalent to an index fund, something that traders can put their money into to expect some kind of return over the long run? Is there anything traders can do to do that?

Stephen: Prediction markets on a mechanistic level are zero-sum in that the most common way to structure a prediction market is to have contracts on whether an event either does happen or doesn't happen, yes or no. That's inherently zero-sum. For a lot of our markets on Manifold, the environment isn't zero-sum because there is this third party which is typically but not always the market creator who's actively going into the market to subsidize. So I think subsidization is actually very important in a prediction market context.

The basic idea is that if you want to have your question answered and it's on a pretty narrow niche topic, you may not get as much liquidity on that from purely profit-seeking traders. A lot of questions that you may want an answer to have massive adverse selection where one party naturally knows much more about this topic than the other and the price would move very rapidly in response to trades. So to cut back against this a little bit, in order to work well, you have to pump your markets full of subsidies in order to entice traders to predict in the market. A subsidy is basically just cash that you allocate, that you put into the market. You can think of it as making the, adding more friction to price movements. The more subsidies in the market, the less the price will move in response to trades. But I think in that sense, it's not zero-sum.

But I guess the other part of your question is trying to use prediction markets as an instrument to gain equity-like returns. I mean, I think that doesn't really make sense to me. Even with the subsidy, it may not be zero-sum in the sense that there's a bunch of dumb, intentional dumb money in the form of subsidies being added to the market. There still isn't really anything like stock market beta or sensitivity to broad economic growth. But you can't, for instance, if you select 100 random prediction markets and invest $10 into each, you would expect that to return $0 and not to increase with the size of the economy.

Theo: But wouldn't the subsidies that you would need to make prediction markets work in the way that you're describing be tremendously burdensome to the process? Incredibly expensive?

Stephen: Not necessarily. This is one of the things that we found with Manifold. We're a play money platform. If the user experience is sufficiently compelling and game-like, you can get a huge crowd of people, such as the people in this live audience today and attending Manifest this weekend, who are interested in prediction just for the sake of it outside of the monetary rewards. And when you have this system set up, that means you can get by with a much, much lower subsidy than you would if you were actively going out and commissioning the traders to give you your forecasting estimate. So I think this is one of the nice things about Manifold, is that you can purchase information much more cheaply. The nature of the platform itself kind of elicits information out of traders at a pretty low cost, much cheaper than you would be able to otherwise. So each subsidy dollar in turn can then way more efficiently get you information than whatever the alternative is.

Theo: Why did Manifold predict the 2022 midterm elections better than real money prediction markets like Polymarket and PredictIt?

Stephen: This is interesting, because I was on the other side of this trade. I thought during the midterms that Polymarket's numbers would be more accurate. So I bet a lot on the other side, and I lost a huge amount of mana because I was wrong. And now I've learned my lesson, that Manifold's numbers are more accurate. I think, honestly, this is kind of an n equals one thing. I think people should be very wary in general of trying to judge the accuracy of any pundit forecasting platform tool or anything on just one election cycle. So one very simple story that you can tell about this is that Polymarket had more Republicans and Manifold had more Democrats, and the Democrats won. So really, we need to repeat this over several election cycles with different parties winning in each in order to get a better sense of each platform's accuracy.Form's true calibration. Is there any existing data on which platforms have performed the best on different elections, or is it just too recent, there haven't been enough elections? It's not too recent, and there are also other, even older academic prediction markets which have a track record behind them. One of the first big prediction market experiments in the 20th century, after the progressive era in which they outlawed all of this stuff, was the experiment conducted by the Iowa electronic markets in the 80s and beyond. At that time, they found that their markets were more accurate than both individual pundits and a bunch of different aggregates of pundits. There's a similar track record from the more recent prediction market attempts.

I got my start on prediction markets with Intrade, which is defunct. They were an Irish prediction market platform. I remember trading on the 2012 midterms, and I believe that their numbers were more accurate than pundits at the time. But there are a bunch of studies on this. You can find actual answers to these questions. I don't have them off the top of my head.

More recently, though, what we found is that Nate Silver and FiveThirtyEight have performed basically on par with prediction market and other forecasting platforms. I think that will change as prediction markets become more liquid and more people are trading on them. I do think in the limit case, with tons of money being actively traded on these things, that prediction markets will be the very best mechanism and will have the best track records. But they're already pretty large and pretty liquid. I don't know about that. There's millions of dollars that are being traded.

Theo: So you're telling me that these thousands and thousands of traders, many of whom are pretty smart in aggregate, can't beat Nate Silver, even though they have financial interests in doing so?

Stephen: Yeah, I think a lot of it is due to the big thing you need to guarantee that prediction markets can live up to their full potential: institutional liquidity. You need Goldman Sachs and hedge funds to be able to be counterparties to all of these bets done by retail traders on platforms like Polymarket. And that does not appear to be in the works anytime soon, mostly because of regulation. I think it is true, though, that having a hundred million dollars on the line should be very enticing to people. That is a lot of money, even for very talented, wealthy individual traders. But there are still these structural barriers that prevent a lot of individual traders from participating in real money markets.

Theo: Accredited investor requirements and stuff?

Stephen: Well, or the fact that US citizens legally can't participate in Polymarket. Many do. Many use VPNs to access these markets offshore. But the regulatory issue and the usability issues with crypto are a major barrier.

Theo: So for my podcast audience, I'm sure everyone in the Manifold audience knows this, but futarchy is a political system where you would base policies on prediction markets. And so if you had the option to do so, would you replace our current political system with futarchy?

Stephen: Ah, that's a great question. Maybe this is a little bit heretical, but I've never actually been that on board with futarki as a concept. So, firstly, I think the first thing is, my view is that prediction markets are a tool, or it's kind of a category error to talk about it as a form of government. Governments are not just decision-making mechanisms. They're people who have particular values, who implement decisions. Even Robin Hanson often formulates this as a bet on what will happen. Bet on values, not beliefs.

Theo: Yeah, vote on values, bet on beliefs.

Stephen: Yes. So even in this formulation, part of the governance formula has to include other stuff that isn't just the mechanism. So there's that aspect. But in terms of using prediction markets to totally replace all existing decision-making bodies, I'm more skeptical. I certainly think on the margin that governance quality would improve a lot if people actively were creating, subsidizing, all sorts of questions on different policy impacts of various proposals. That would be a great thing. People have talked about using NGDP futures to help central banks determine their monetary policy. I think all of those are great things that we should be doing.

In theory, a sufficiently liquid decision market on topics where decisions can be enumerated exactly in some domain should be good. There shouldn't be any problems with that. If the market price is predicting what the outcome of various policy interventions would be are out of whack, then rational profit-seeking traders will come in and correct them, and their probability should be accurate. Policymakers make a lot of decisions. A lot of decisions are about things in smaller markets where people don't really care about but in which there are very strong vested interests. If you're making some micro-policy decision about shrimping rights off the coast of Maine, maybe the shrimpers will be willing to collude and place bets that other rational profit-seeking individuals wouldn't be quite motivated enough to do. That's one issue with futarchy.

I think the other big issue, that's a problem with the mechanism, I think the other problem with futarchy is that it doesn't address the fundamental concept of the political. The real political question is who gets to create the markets? Which are the importantValues that people actually care about determine how we allocate the liquidity to subsidize the prediction markets to get the answer on. Even if we do move into a much more futarchical world, which I support, that won't solve that problem.

Theo: Let me frame the question differently. Do you think if the Bay Area governments were replaced entirely with futarchy, would it lead to better outcomes?

Stephen: I think replacing the Bay Area government with anything would lead to better outcomes, so yes.

Theo: Clearly not anything, right? Replacing it with Stalin wouldn't.

Stephen: I don't know.

Theo: For my podcast audience, Stephen's brother, James Grugett, is one of the other co-founders of Manifold. Why do you have so much more mana than James? He has like 200,000, you have over a million.

Stephen: A lot of my mana comes from betting against James, which is interesting. One of us was guaranteed to win and have more money than the other.

Theo: On what markets?

Stephen: I think our biggest source of disagreement, and one of my biggest sources of profits versus James, is on the success of Manifold Love, which is our dating platform. I guess for the benefit of Theo's audience who may not have heard of this, the basic premise of Manifold Love is that, you know, it's in part an OkCupid clone where you can create your own public dating profile, and then the twist is that we have prediction markets on each of the people in this ecosystem for people to bet on who would be a good match with each other. The thinking is that your friends, relatives, or other random strangers who scour through your profile would be interested and motivated in matching people off based on this, and that would be reflected in the market prices.

So this, obviously, this is an insane sounding idea. This is a thing that people outside of the Bay Area would not do and would probably roll their eyes or laugh, or some combination of all of these things. I first want to say that even though I never believed in this as a large venture scale business, it actually has been successful in producing multiple long-term relationships which are still going to this day. Who knows, maybe they'll result in marriage or something like that. So I think it's too easy for people to cavalierly dismiss crazy Bay Area ideas involving prediction markets. And even if they don't live up to the full hype, they're still capable. I feel like the premise of the Manifold Love actually was vindicated, but on a smaller scale. I think it can work in this community, at Manifest, in the Bay Area, for like-minded individuals. I still have my questions about how well it would be able to scale to the rest of the world.

Theo: What are the fundamental limits? Just that not enough people know enough information about the couple to be able to make good decisions?

Stephen: I think... Like, they'd be very small markets, necessarily, right? Well, I think a lot of people are just put off by the concept of public profiles. This is actually a huge barrier. I think it's not necessary for everyone to be on board with the premise of the app for the app to still succeed. Many people really despise and hate dating apps, and yet those are a big thing. When dating apps were first introduced, they were seen as really weird and gross and disgusting, and only the worst part of society would use them. But since they were so useful, adoption has gradually increased, and the bull case for Manifold Love is something like this story, that even though it sounds really weird, some people have told me it's repulsive, that over time, that would fade, and the benefits would become more apparent. I'm just not convinced, though. I think too large a chunk of society just really doesn't want to have public profiles with people betting on them.

Theo: Speaking of manifold love, you did a related... I don't even know what to call it. Part game show, part live musical, called Bet on Love. How did you get the idea for this? How did this come about? What's the backstory? What was the idea behind it?

Stephen: Yeah. I think it's interesting. Both manifold love, our dating site, and the idea for Bet on Love essentially grew out of the last Manifest, our first conference here. In particular, we noticed that a lot of the markets that people seem to have the most fun betting on were relationship or romance-related things, many of which involved Aella, and you can look those markets up yourself on Manifold. We were trying to think about how we could capture that energy and use it to drive more engagement. Obviously, the natural thing to do is to have a surrealist prediction market dating show musical with Aella as the star bachelorette. The show actually... My original vision was much more limited. Originally, I was planning on just doing this really small-scale, very low-budget indie event where it might even be at the same venue that Manifest is happening, out in a courtyard, and we just stream it on one webcam. After I explained my idea for a prediction market dating show featuring Aella to one of my friends, they told me that, in fact, Vibecamp had actually done a prediction market dating show featuring Aella, and that I could watch the footage of this video, or watch the footage of the recording. I did, and I was super impressed by the theater company that put it on. I knew immediately after I watched this that we needed to hire them and get Manifold involved in some capacity, and that tying their theatrical and musical genius to betting on markets could be a product which is super compelling to people. I really like Bet on Love. It was very entertaining. Very interesting. I guess I do have to say this is pretty polarizing as well. I think you, the audience, will enjoy Bet on Love if you like musical theater, if you are really into niche nerd humor, and you like dating shows. If you love all three of those things, you're absolutely going to love this. If you love one of these things a lot, you'll probably love it. If you love none of these things, you probably will not love it.

Theo: I don't particularly love that. I don't love musical theater except for Hamilton, and I definitely don't like dating shows. They're boring, but there was something about Manifold Love. Maybe it was just the specific type of guy who was in it. I don't think it would work with most normal people. It wouldn't have the same charm.

Manifold has a calibration chart at that shows whether events happened as often as they predicted. If you go to that chart, you'll see a bunch of dots and a diagonal line. All of the dots are below the diagonal line, which suggests that events happen less commonly than they were predicted to at all data points. Why? Are the traders just overconfident?

Stephen: Yes.

The interesting thing about this is that you might naively think you could just write a bot to bid the contracts up and that you would make money. The reason why things like this can persist is that that's harder to do than you think. The moment you introduce YesBot that bets yes on everything, people will see that your bot always bets yes on things and will bet against you or will exploit you. They'll bid the price up higher than what the true price should be, and then you'll be stuck holding the bag with your worthless yes shares.

Theo: Has anyone tried making YesBot?

Stephen: Yeah, they have. I think it is interesting. One of the first things about our calibration chart is that it's just a firehose of all of our markets. It includes even pretty low-quality markets and markets that don't have that many traders. One of our users actually has created this website called Calibration City that allows you to create calibration charts that are more granular and targeted towards markets with whatever attributes that you want that have 1,000 traders or that are on particular topics. I suspect that if you added more filters to filter for higher-quality markets that a lot of this effect would go away. But it still remains to be seen.

I don't know. I think the brute fact, even for our lower-quality markets, that they have this pattern is surprising to me. A priori, I think I wouldn't even have been able to predict the sign of whether our markets would be over- or under-confident. I don't really know why this effect exists and if or how long it will persist. But in general, when you find things, yes, bot is not going to work as a strategy. But if you do see consistent wrong patterns in markets, you can do more sophisticated things to try and correct those. This is a sign that there are possible trading strategies that you could use to profit from this since it does appear to be pretty systematic.

Theo: When you were on the Dwarkesh podcast a couple years ago, you told them basically that you don't like insider trading, even though a lot of prediction market people do because they think it makes prices more efficient.

Stephen: No, I love insider trading.

Theo: On real financial markets? Or insider trading laws.

Stephen: Yeah. The classic libertarian story is that insider trading laws are bad because markets are about information and giving good prices to the public, first and foremost, and that when you remove restrictions on who can trade, it makes the prices more reflective of reality and more efficient. So I think that's a pretty good argument. The counter argument is more of a fairness argument. It's not fair for corporate officers to be able to make so much money doing things which are relatively dumb, of having access to earnings reports before the general public, or more maliciously, it's bad that they have an incentive to try and sabotage the company or other things, et cetera. I think those are very real concerns and probably the ideal legislation would do something to limit that in some fashion. Maybe the absolute chaos would work. That would result in a society which is functional. It may be better in some ways than a more restrictive legal climate, but it probably also isn't the absolute best regulatory regime.

Theo: So what do you think about other forms of suspicious market activity that isn't exactly insider trading or fraud? Like, for example, what Roaring Kitty is doing right now with GameStop, where he's somehow memeing the stock up multiple billions of dollars in market cap. Should the SEC do anything about that?

Stephen: I think probably not. In general, whenever there's ambiguity about the harms of particular actions, as a good general principle, it's good to not have litigation or regulations there. The world is very chaotic. If the outcome is not certain, it doesn't really make sense to get lawyers involved. Or really, when you do add regulation on this, the only parties who actually win are lawyers because then there's increased litigation. Society doesn't really benefit anyway because it's ambiguous. There are benefits on both sides. It just doesn't matter from a societal perspective. I think the government, you know, financial regulation should be limited to more severe, severe harms, which everyone can recognize and which are dealt with in an easier fashion.

Theo: Do you think it's easier or harder to, in the long term, do well on manifold versus actual financial markets? Because you might think it would be easier because they're less efficient, but you might also think it would be harder because they're more zero-sum and you can't just buy the S&P 500.

Stephen: That's a good question. So again, we have the subsidizer dynamic where people are putting up huge amounts of cash because they want to have their question answered. So as long as subsidizers are an important part of the ecosystem, or insofar as that's true, that makes it easier for people to earn money because the subsidizer is just paying you to do that. They're not paying you in the same way to trade GameStop stock. There isn't someone naturally tossing a bunch of money into that outside of other retail investors. They are paying you lots of money to trade.

Theo: At the beginning of the interview, you talked about how one of the reasons prediction markets aren't more popular is because a lot of them are hard to use. So what do you think are the good elements of a prediction market user interface that will make people want to use it?

Stephen: Simplicity is key everywhere. A big mistake other platforms and other forecasting platforms have made is just making it too complicated, having too many different market types, having too many order types, showing too much information on the screen, etc. The simplest consumer apps are things like Robinhood where they strip away all of the extraneous content and just have you focus on a few key numbers and make it super obvious which user flows you want to go down. In the case of Manifold, one of the flows that we try to optimize for is market creation.

Making that really easy is part of it. That includes having it all fit onto one screen. We don't have a multi-page setup. We try to keep that pretty minimal. The other aspect of that is we've tried to standardize market terms. When we launched Manifold, when you created a market, we allowed you to set the initial probability and choose the exact amount of subsidy to provide in the marketplace among other things. The model we've moved towards now is where the market automatically starts at 50% and we standardize on certain liquidity tiers. That's just to make it much easier so you don't have to think about what you want to do when you create a market. The lowest tier markets on Manifold all cost the same thing. You don't need to think about that. If you want to subsidize them more, we've recently introduced a market tiers feature which have liquidity at different levels and you can just choose among these discrete options. That eliminates a lot of the paralysis that comes from having too many different options available.

Theo: So why should people trust market makers? What if they resolve markets incorrectly on purpose?

Stephen: The big thing is reputation. One of the nice things about our platform is not only do traders accrue a reputation for trading well on the platform, but market creators do as well. The better market creators not only resolve markets fairly and quickly, but they also do a better job of anticipating edge cases and having really well thought out resolution criteria, which is a skill. So it's not just not being a scammer. There's also an art in crafting markets such that the entire process is smooth and unambiguous. Our view is that over time, the market itself will select for creators who are better at doing that. We internally at Manifold will promote their markets more versus other markets with worse criteria.

Theo: What do you think about derivatives on prediction markets? Is that a thing that needs to exist?

Stephen: Prediction markets themselves are a kind of derivative contract on information or other real world financial assets.

Theo: This is interesting.

Stephen: Actually, one of the things I feel like Manifold's user base now is pretty high caliber. Immediately after we launched Manifold, we kind of blitzed through all different sorts of random derivatives on Manifold, which weren't really that useful directly, but were really cool demonstrations of different things that you could do. So we had immediately users created leverage prediction markets where you would do things like resolve NA and return money most of the time. But in some world, you choose like 1% of the time the market will resolve 100 times more or something or give you 100 times the payout, something like that. We experimented with volatility using other prediction markets as volatility swaps on other prediction markets where you can do that in a few ways by saying, will this prediction market trade outside of this range within this particular date? That's where you can extract volatility as a separate signal. There are a bunch of other stuff as well. I feel like eventually those will be useful for the biggest prediction markets on things where people are putting up huge amounts of money and want to hedge their risk. If you created a five dollar market with your friends, betting on who's going to win the next game of pickleball, maybe it's not so useful.

Theo: On your LinkedIn, it says you used to be the founder of Rareroot, an online ginseng marketplace. Can you tell us a little more about that, like how you got the idea, why ginseng, why you moved on?

Stephen: I was not expecting to be grilled about my past as a former humble ginseng merchant. This is a very long backstory. The first commercial vessel to ever set sail from America to China was loaded with several tons of American ginseng, and American ginseng is a separate species from Asian ginseng, indigenous to Appalachia, and closely associated historically with the fur trapping trade. Fur trappers like Daniel Boone would collect ginseng and sell it to these ginseng merchants who would then ship it overseas to China during the off-season for the fur trade. So there's this very long history of trade. China is flowing in the opposite direction of what you might think. The key facts about American ginseng today are that it wholesales for about $1,500 a pound for the simplest type of roots. Many Chinese people value roots that have very interesting or exotic shapes, which can be worth a significant multiple over the base wholesale price. The most expensive individual ginseng roots have sold at auction for $500,000 to a million dollars. Ginseng occupies the same cultural position that a really fancy bottle of wine would in the West. It's a thing you would give your boss if you don't know what else to give, and there are different gradations of fanciness that you can calibrate your gift to.

My random business idea was to try and become the Alibaba of American ginseng. I noticed that there were several layers of middlemen between the growers of American ginseng roots in Appalachia and the ultimate consumers in China. Ginseng is typically exported to Hong Kong and then smuggled over the border to mainland China to avoid taxes. It's then shipped out to the rest of mainland China from a small town in southern China where a bunch of Chinese medicinal products are located. I was trying to think about ways to disintermediate these layers of middlemen through a website. However, I realized that no one in the Chinese traditional medicine world operates at startup speeds and they're much more set in their ways than people in Silicon Valley. I ultimately realized it would probably take a decade to build a serious business in this domain and that there were a lot of other interesting things I could do instead. I did sell a little bit of ginseng, but I only had two or three sales total, so it wasn't a huge success.

Theo: Now, let's take some questions from the audience.

Audience Member: Why don't Americans consume American ginseng?

Stephen: Well, they actually do. People in Appalachia do consume American ginseng. I've also heard that truckers in the south will sell ginseng at truck stops. The most common way that Americans would consume ginseng is in Arizona iced tea, although that's mostly Chinese ginseng, not American.

Audience Member: The next question is about my views on cancel culture and prediction markets, and specifically my views on the Richard Hanania controversy.

Stephen: Cancel culture is bad. If you want to help people, you should try to help them improve their views. Prediction markets can play an important role in getting people who believe incorrect things to believe better things. They provide a better calibrated picture of how the world works, which can help people improve and hold better beliefs. However, prediction markets won't tell you whether things are right or wrong. They will tell you whether people believe things are right or wrong or will believe them at some future date, but they won't address those questions directly.

As for Manifold's moderation policy, we have tremendous faith in random internet strangers to mostly do the right thing. We want Manifold to be culturally neutral and not enforce particular political sides or stances on issues. We prefer to allow as much ideological diversity as possible. We believe it's bad for social media platforms to impose any particular narrative. We're trying to operate as close to a free environment for anyone of any political persuasion as we can, within the limits set out by the law and other structural factors that we face as a business. Regarding the specific case of Richard Hanania, I think it's bad he was cancelled.

Theo: What specifically was this controversy, for the audience?

Stephen: The original thing that set off his cancellation was when it came to light that a decade previously, when he was a college student, he wrote a bunch of dumb articles under a pseudonym. Some journalists discovered that the pseudonym was him and released these in the future. He released some statements saying that he disavowed the dumb things that he used to believe and doesn't believe them. For most people, many believe really dumb things in college or as teenagers. I think it's important as a society to understand that people should not be held accountable or publicly punished as an adult for things that they believed as a teenager. I think it would be very bad for platforms like Manifold to take a strong stance against content like that.

Theo: Do we have any more audience questions?

Audience Member: Why aren't there more institutional types in the market? You mentioned before you think that would improve the market.

Stephen: This is a great question. A lot of it is actually just regulation. If you're an investor and you invest in a regulated exchange and you lose money, that's understandable. If you're investing your limited partner's money in some exotic financial instrument that's unregulated or is offshore etc., if you lose money you're going to get sued. This basic factor prevents a lot of institutional capital from moving into unregulated domains. If there's enough money in this space then eventually that demand will emerge. Crypto is a good example of this. Crypto even right now is still not legally kosher everywhere or even in the U.S., but there's beginning to be more and more institutional capital pouring in just because the opportunities are there.

The other reason why there isn't more institutional money in prediction markets is just that there's not that much money in general. I think similarly to crypto, the trajectory that prediction markets and Manifold in particular will follow is that we're starting with the consumer use case. Once we get more consumers and retail trading volume on our platform, eventually over time, institutional capital will follow especially if that's accompanied by deregulation.

Theo: Anyone else?

Audience Member: I guess I have a question. Is there a role that journalists and media publications can have and maybe being incentivized to help resolve certain difficult questions or participate in that process?

Stephen: Sure, even today on a lot of markets on Manifold, you'll see that a common type of resolution criteria that people will employ is deferring to mainstream media to decide the outcomes of markets, particularly in cases where outcomes are ambiguous and you need some independent neutralish source to make some sort of judgment call.

For instance, we had a market recently on whether in the Israel-Palestinian conflict there would be an invasion of Rafah. Invasion is actually a totally ambiguous term. There's no strict legal definition of invasion. If you created your own personal market on whether it was an invasion in your heart, people may not bet on that because they don't trust your ability to have a reasonable understanding of what that means. We've had several markets on whether the New York Times will call it an invasion. That's a good way to operationalize this really difficult fuzzy claim.

A lot of the work is doing things like that. Another type of pattern that people look for is for a general media consensus on something, which is usually an indication of fact in cases where, like U.S. presidential elections, typically are not disputed but the last one kind of was and perhaps other ones will be in the future. In a politically tumultuous time, being able to enumerate a list of different journalistic bodies and say if most of them say this then we're going to resolve according to that, provides a reasonable standard and baseline.

Theo: I think we have time for one more. Yes?

Audience Member: Is there any role for sweepstakes other than regulatory arbitrage?

Stephen: Yeah, so the concept in American law that makes something a sweepstakes is this concept of alternative method of entry, which means you have to be able to enter into the sweepstakes without paying. If you have to pay to participate in the contest to win a prize then it's not a sweepstakes.

The key thing that makes sweepstakes good and fun relative to other types of mechanisms is that it allows free play. As I mentioned earlier, even in a world where there are totally deregulated real money prediction markets, I think we do want this space of play money prediction markets where anyone can participate. Insofar as sweepstakes are a way of achieving this, I think they're good and will continue and persist into the future.

Theo: All right, well I think that's all the time we have, so thank you so much Stephen Grugett for coming here and doing this live interview with me at Manifest. Everyone go check out Manifold Markets at and yeah, I think this was great.

Stephen: Yeah, thank you so much for having me.

Part 2: Austin Chen

Theo: Welcome back to episode 16 of the Theo Jaffee podcast, part two, again live on day two of Manifest. Today I'm interviewing Manifold co-founder Austin Chen. First question...67 days ago on April 2nd, you officially left Manifold and in your farewell post, you gave four reasons for doing so. Manifold is stable and doesn't have much left to iterate on, you're not excited for the next steps including the pivot, prediction markets are insufficiently powerful, and short AI timelines muddle everything up. So far, the Manifold market has predicted an 8% chance you'll regret it in two years. I'm assuming you don't yet regret it, but do you have any more details to offer, especially on the prediction markets being insufficiently powerful?

Austin: The prediction markets being insufficiently powerful is a point I've thought about many times throughout my tenure at Manifold. It was pitched to us as a revolutionary mechanism that would help us figure out what the future will hold and how to navigate the many decisions you have in the world. One thing I noticed pretty early on is that a prediction market can only tell you very few bits of information. It will tell you how likely the thing is from 0 to 100 percent, that's the main source of information that a prediction market by itself gives you. But you need a lot of bits of information to navigate the world. When you're making a decision like what policy, what feature should I implement for Manifold, it became very hard for us to use our own markets to figure out what we should do.

James, my co-founder, has thought of some pretty interesting mechanisms to try to get around this. If you look at a prediction market, most of the bits of information are in the question itself. So James thought to invert the traditional market structure and let people submit the questions and crowdsource the question creation part as well. That could hypothetically generate a lot more bits of information. But that mechanism hasn't proven out to generate really good policies, really good paths, really good plans for navigating the future. So I still think we're kind of at the drawing board with regards to how do we use these predictive mechanisms to make better decisions.

I was a true believer in the beginning that prediction markets can really help us act in the world. Now I still think that there's a good business in prediction markets, they provide fun, they provide a game that people enjoy betting on, but I'm less sure that these are the things that will help us navigate.

Theo: If they can't provide foundational governance value to society, what areas do you think that prediction markets would actually be better than the alternatives for predicting?

Austin: They're a pretty good aggregation mechanism, one that doesn't really exist in other areas. They can cohere all of that into a single point, like you can get a much better question of what does the world believe about whether Biden or Trump will win by having them bet on a prediction market than you can with a variety of other mechanisms. So I think the aggregation function of prediction markets is probably the most valuable one. Besides just aggregating all the data into a single percentage estimate, you also have people ask comments and bet back and forth, which are additional add-ons. They're not core related to prediction markets, but as you extend this functionality and people are all in the same location, you get additional benefits.

Theo: So now you're working full-time or mostly full-time at Manifund, which is a very unique charity organization that has a whole bunch of unique features like, for example, re-granters. So you have people who you entrust a budget to, to let them donate money. How did you make some of the design decisions behind Manifund?

Austin: A lot of them were based on my own experience as a grantee in the ecosystem. I've received some grants from the Long-Term Future Fund, for example, who can give a pretty small grant, and a Survival and Flourishing Fund, which can give a pretty big grant. I noticed a bunch of shortcomings. For example, they tend to not give you very much other feedback, rather than did you get the grant, did you not get the grant. There's not a lot of other data points to look at. You don't have a sense of what kind of grants they're actually looking for. Most of these have what are called open grant databases, but it's really just a single sentence, or maybe just the place where the grant went, and how much money it was. It doesn't tell you what was in the application, or what are the decision processes behind why the grantor decided to pay out to the grantee. So those are things that I wanted to fix with Manifund.

On re-granting specifically, re-granting was a mechanism that was really popularized by the FTX Future Fund.

Theo: Now the FTX name has been tarnished, I would say. It was good for its time. I remember during the FTX glory days, when Sam Bankman-Fried was on Nas Daily and the Dwarkesh Podcast.

Austin: I was perhaps the last SBF fanboy and still a die-hard. I guess your idols die very slowly.

Theo: SBF did nothing wrong!

Austin: However, I wanted to separate out the Future Fund from FTX itself, because one massive unfortunate shortcoming was that they tried to tie these two things together. Future Fund, I was pretty close with a lot of people running it, like Leopold Aschenbrenner, Avital Balwit, and I've spoken to some of the other people involved as well. They were just really good people. Good both in the competent sense, but also in the virtuous, trying to do good things for the world sense. For instance, on their re-granting program, they made the decision to not announce who their re-granters were in public, because they didn't want this thing becoming a weird status badge that would change the dynamics of the EA ecosystem. They didn't want to be seen as the ones awarding status. I thought that was one small example of a decision they made very thoughtfully.

The Future Fund did a lot of really cool things. One cool thing they did was the re-granting program, where people were just empowered to have individual budgets of something ranging from hundreds of thousands to millions of dollars, where they could more or less make a decision on a grant without having to get external approval from committees or things like that. I think Future Fund would just do a safety check, but then the re-granter basically had full discretion of how to spend the funds. This is actually a kind of thing that's very rare in the entire grant-making ecosystem. Most of the time, people think that if you have to give out money, you have to do it with a process. You have to do it very carefully. You have to have written-up concrete justifications for why this grant is being given to be accountable. Future Fund was like, no, let's throw this out the window. Let's just let people give out money. Let's do it really quickly. We're going to try to put an emphasis on getting money out the door very quickly. I think these were all really great things. Things that I had suffered a lot when I was a grantee and I really wanted to promote.

So Future Fund collapsed, but then at some point later, one of the people involved in Future Fund put me in contact with one of the donors and was like, hey, we think the re-granting program is still really good. Even though FTX isn't around, we might still open when the fund is. So then this anonymous donor gave us 1.5 million dollars last year and 1.5 million dollars this year to distribute to ASX re-granters. And they are making a lot of the grants in Manifund right now.

Theo: So how do you choose re-granters and how similar are good re-granters or philanthropists to good investors?

Austin: We actually didn't choose the re-granters in this case. Manifund views our role as more of a platform, a neutral platform. The grant maker, the person who provided funds, had about five or six people in the ASX space. We validated their picks. We looked over them, made sure they looked like they were going to be able to give grants in the ASX space. But we did not make the decision on who the re-granters were.

Theo: How similar are good investors to good philanthropists?

Austin: None of our re-granters are investors. All of them work in ASX basically full-time. You can see the list, but there are people like Leopold.

Theo: Very impressive list.

Austin: I think we worked pretty hard to find good people, but again it was up to this anonymous person whose identity is still not known to the rest of y'all. I think they already had some connections to these people and as a result that's how we got the list of re-granters in the first place.

Theo: So why should someone donate to Manifund over something like GiveWell or Open Philanthropy?

Austin: You can't even donate to Open Philanthropy, so that's one reason you need to give to Manifund. If you want to give away your money, OpenPhil won't take it, as far as I can tell. Unless you're Dustin Moskowitz, I guess. Then OpenPhil will take your money. GiveWell does take your funding. GiveWell only donates basically to projects in the global health and development space. That's mostly things in projects in Africa or other ways that they can find to help out humans very cheaply. So depending on what your worldviews are, if you believe that humans alive today are important but maybe less important than the welfare of animals because of the amount of funding in these spaces, you might want to give to a different animal welfare related fund. Manifund doesn't have too much of that. What Manifund does have a lot of is AI safety research. So insofar as you think that the future of humanity, humans living in the future are still a pretty neglected cause, you might think that giving to projects on Manifund would be good.

Right now it's not really the case that Manifund even accepts that many direct donations. Most of the time when you go to Manifund, you are screening the projects yourself. Manifund is kind of like a Kickstarter where you can just look at the project proposals and yourself decide, I think this is promising. I think this is a shot. I think I want to donate to this. This is actually I think closer to the roots of EA than GiveWell is today. Because today when you go to GiveWell, you kind of think like GiveWell is this one trusted institution. You can just give money to them and they will distribute it wisely. But back in the day when EA was just getting off theWhen GiveWell was just getting off the ground, there was no other trusted source they could look at. They had to make all their decisions themselves. So I would say that if you are in a position of trying to give some money, it's a good thing to make that decision yourself a little bit. Try to put yourselves in the shoes of a grant maker and try to evaluate whether a project that is about to go out will work. Are the founders good? Is the plan of impact good? This is the kind of thing that the people at GiveWell, way back when it was getting started, had to think a lot about.

Nowadays, EA has become a lot more institutionalized in a way that I don't quite like. In that you just try to guess who these people are who are smart. It's a little bit more political, a little bit more affiliation-based rather than doing your own research. So Manifund lets you do your own research and make your own decisions about what to fund.

Theo: Earlier you were talking about Dustin Moskovitz. So how much power do people like Dustin Moskovitz and Cari Tuna and Jaan Tallinn have over the EA funding ecosystem? Is there a centralization risk there?

Austin: This is a thing that lots of people in EA discuss. I don't think as a practical matter Dustin or Cari have that much direct influence because they don't make day-to-day governing decisions at OpenPhil. If they want to make a change they'll probably communicate it out to this 200 person organization. And that message then has to trickle down to all the different grant makers and people who support the grant makers at the OpenPhil institution.

OpenPhil plus Dustin and Cari who are maybe the largest voices at OpenPhil but still, I think less than 50% of the stuff that gets done by OpenPhil you would causally attribute to coming from the heads of Dustin or Cari. OpenPhil as a whole is a big influence in the EA ecosystem. Yawn I think does more direct thinking about what to invest in and does make those decisions. They're both big players. Manifund is trying to be the place where all the other smaller players can find all the other grantees and kind of set up the marketplace, the clearinghouse for that.

Theo: How would Manifund's priorities change if its annual budget were a billion dollars or ten or a hundred?

Austin: I like to think that I often think of Manifund just as Future Fund running the same-

Theo: Future Fund 2.

Austin: Yeah Future Fund 2, the Future Future Fund. I think their playbook looked pretty good. Their explicit first year was trying to run a bunch of experiments on different ways you could distribute funding in large amounts. That's why they were excited about the re-grantor program. It differs from the OpenPhil classic. You have program officers at OpenPhil who are just in charge of large budgets and make decisions relatively slowly. So future fund wanted to try a different approach with maybe a hundred different re-grantors with small budgets who can just make decisions very quickly. That's the kind of testing future fund did. That's the kind of testing I would do.

I'm pretty excited by something like impact certificates which is trying to set up a venture ecosystem for charity or effective altruism grants. So that's the first thing that I would try out but I'm not so committed to it. I think the meta strategy is try lots of things and see which ones work and scale them up and the first obvious thing I would try would be impact certificates.

Theo: So how did Manifest manage to get such a high density of smart and interesting people and internet celebrities? You think it would just be selection plus being in the Bay Area but very few other places and gatherings are like this. So why Manifest specifically?

Austin: I think what you might not see is the amount of time that our team, mostly me and Saul, have put into just sending out invites. We have this CRM of maybe 300 different people who we thought would be really awesome speakers and we spend a bunch of time just writing individual emails to all of them.

Some of it is taste in that basically Manifest is a gathering of a lot of the people who I think are the best writers the most interesting thinkers in the world and I've tried to invite them and try to position myself in their shoes and think what would make a good event for them well why would they be excited to come and pitch it to them. That often involves calling out a few of the other names. Name-dropping I guess a little bit which I don't feel that great about necessarily but I think it's a core human being. The first thing you think about when you're going to a party is who else will come to this party who else do I know at this party. So I try to highlight that for the speakers who I invite.

So that's a big part of it just the manual work of spending 15 minutes per person sitting down to write an email from scratch to invite them to come to this event that we're putting on. I think two other things that are working in our favor. One is that we ran Manifest last year and it was a really good event. And that just leads to word of mouth growth, more or less. People tell their friends, hey, Manifest was a great event. And that is so valuable. The virality of having created a good product. They say if you build a better mousetrap, people will beat a path to your house. And I've seen that with Manifest, I think. So many people have said, oh, my friend told me about Manifest. That was so great last year.

Theo: I couldn't miss it this year. I experienced massive FOMO last year when it was happening, so I had to make sure to be here all summer so that I could make it to Manifest.

Austin: I'm glad that you work here and I'm grateful to have you here.

Theo: It was well worth the $200.

Austin: $200? Oh, because you got a student discount. There's a separate digression about how we use pricing discrimination a lot at Manifest. For the speakers, there's basically a negative price. We will pay for some of their flights and housing. For students, we try to make it relatively cheap. I try to charge people who have a lot of money a lot more. That goes into the economics of making something viable.

Just to finish answering your question, I think the last part about how a lot of really cool people come to Manifest is that we kind of lucked out with prediction markets as a topic. It turns out that many of the really smart people in the world just think that prediction markets could be cool, at least. They're kind of open to weird mechanisms. And this is pretty differentiated. A lot of the rest of the world has not heard about these things. So it turns out that running a conference just on prediction markets will draw out the right crowd. I don't know if this will sustain, especially if Manifold actually succeeds in growing. I don't think you could do such a good conference on blogging, for example. But we'll see.

Theo: Because it's just not differentiated enough?

Austin: I think so, yeah. Or especially like podcasting, or TikTok, or something. And I don't think those would be nearly as highly selected for interesting people.

Theo: Earlier, we were talking about Sam Bankman-Fried and how SBF did nothing wrong. Stan SBF. So how much do you think the status of EA has been damaged by him?

Austin: Quite a bit. I don't know if I'm the best person to answer this kind of question. I'm relatively new to EA. I only got into the space a couple years ago. My sense is that it's just much harder to be unapologetically EA. You always have to caveat with, oh, but the SBF thing, et cetera, et cetera. And I do know that I feel less intellectually excited by EA, either its ideas or its participants nowadays, than I did two years ago. And I'm not sure if that's because of the SBF thing, or they were just co-coinciding for other reasons.

Theo: So a lot of what you do at Manifund revolves around future of humanity kind of stuff, including AI and AI existential risk and safety. So how have your views on open AI and AI in general, especially because you just did a podcast on OpenAI, but how have they changed since the Leopold Aschenbrenner piece the other day, if at all?

Austin: Unfortunately, the Leopold piece dropped during Manifest, so I haven't read most of it. I don't think my views have shifted that much. But yeah, again, I just haven't really read it in depth and only skimmed it. Hard to answer.

Theo: What about all the OpenAI drama over the last couple of weeks? You wrote on the podcast page a specific note about, oh, this was written before, like this, and this, and this, and this, this.

Austin: I am maybe an apologist for Sams everywhere, not just Sam Bacon for you, but also Sam Altman in this case. I kind of, maybe because I've spent some time in the role of somebody similar to SBF or Sam Altman as an executive of a startup that was growing and had to make decisions, I kind of see reasons why things that look bad in hindsight, such as the massive fraud of Sam Bankman-Fried or the NDA thing with Sam Altman, weren't really that attributable to the leaders. With the fraud thing, I think I was probably wrong at the time. I now put a lot more weight on the fact that SBF knew what was going on, and that was bad. So I made a couple there. But with the Sam Altman thing, I'd say with the NDAs, as an executive, there's so many things that you're trying to do all at the same time. You often don't have that much time to go into the details for each one. You don't know in advance that, oh, this NDA thing is going to be bad or good, and it's going to blow up. You make 100 of these small decisions every single day. So it's not that surprising to me that this kind of thing would slip past Sam's radar.

Theo: So in EA a lot, there's this concept of hinginess, which relates to whether it's better to donate money now or invest money for the future. So do you think this decade has greater hinginess than other decades? So will money donated now, if it's a pivotal moment in AI or something, have more of an impact than other times?

Austin: I tend to think so. But I also have kind of direct financial incentives that would lead me to think so, which is roughly that on Manifold we take a 5% transaction fee of donations that happen. So it would be better for the Manifold budget if a lot more people donated now as opposed to wait a few years, something like that. So yeah, it is not a topic that I thought that deeply on. As far as we can tell, we get funding and we're pretty much given the mandate to spend this funding within a year or so. So the question of higher level portfolio allocation, should you try and save up more to donate in the future versus not, is not a thing I've spent a lot of time on.

Theo: Why aren't more philanthropic organizations open? Especially the ones that have "open" in their name, like Open Philanthropy. Is it a naming curse, like Open AI?

Austin: I think being open isn't a huge pressure on philanthropic organizations. There is some pressure, but it's mostly to be open to your donors, not as much to the general public. It's common for a philanthropic organization to host a dinner event where they talk about what they've been up to, but it's not as important to publish a blog post or a YouTube video to get the same message across. This is because the lifeblood of philanthropic organizations is donations, so a lot of it is optimized for the donor flow.

Effective Altruism (EA) probably does somewhat better at this than most other philanthropic organizations. They try not to consider the donor the end user, but rather, the recipient of the good stuff, the person whose utility is being maximized. So philanthropy is very difficult in some sense, much more difficult than regular capitalism, because you have to deal with three competing parties: your donors, the people who are doing the work, the grantees, and then the recipients of the good stuff. With typical capitalism, the people who are receiving the good stuff and the people who are paying you, the donors, are the same people. So you have more of a tight feedback loop. You know that whether or not the good stuff is actually happening, because they will keep paying you money if it is and stop paying you money if it's not.

There's a bunch of things in philanthropy, and it does lead to all kinds of weird things, such as the incentive to, not necessarily incentive, but just lack of incentive to talk more about what's going on. I mean, it is also the case that many capitalistic institutions are just not that open. Like, most companies are closed source, for example. They don't publish most of what goes on internally. They view that as a differentiating advantage.

Theo: What do you personally think are some of the best projects that Manifund has funded, and why those specifically?

Austin: I'm biased, because as a re-grantor myself, I usually pick out the ones that I put money into. One that comes up is Lumina Probiotics, the tooth bacteria thing. That one, we were very early on. It was before even Aaron had secured the sequencing for this. He came on to Manifund and was like, hey, I think there might be an opportunity to get this bacteria and then give it to a bunch of people. I think this could be a good charity. I think this could be a good business. And then we actually invested in that very early on. I think the fact that it is so well-known and widespread, at least within the rationalist EA community today, is kind of like a success in re-granting stock picking, I guess.

I think most of the grants that are on Manifund I think are pretty good. But here is another issue with philanthropy. I don't actually have that much expertise in AI safety. My expertise is in startups and building websites and technology. We are trying to run this grant-making program on behalf of people in AI safety. So my sense of whether our grants that we've been giving out have been good is mostly just based on second-hand reports. Do the people who I respect think that the grants are good? And they mostly do. Our donor thought they were good enough to want to renew the program for a different year. That is, I think, the strongest signal I have that we're doing something worthwhile. Otherwise, it is hard to say. Especially with AI safety specifically, it is such a field of really long feedback loops. In some sense, did the world explode or not? We won't know for another five years. And the projects that we're working on in the meantime, did they constantly affect that or not? Very hard to say.

Theo: Would you fund Lumina if you knew that AGI, benevolent AGI was coming in five years that would be able to cure mouth diseases itself?

Austin: That's a great question. I think yes, because I think concrete wins are really important of the kind that Lumina has basically already delivered. It's still a little bit hard to say how effective the bacteria is because it hasn't gone into a lot of people's mouths for a trial. But I think winning is just super important. And insofar as Aaron builds up the skills set of being able to market a thing and promote it and share with a bunch of people, I think that will be robustly useful in the coming AI future more or less. So I think helping him accomplish this goal will also mean that in two years, if AI stuff is going nuts, he will have a lot of the capacity, resources, network, talent to be able to help out with that. I think he actually wrote in his management application that he would prefer to be doing something, something AI safety, because he thinks that's more important. But this is just such an obvious low-hanging, dumb thing that society is dropping, the fact that we should not have cavities at all, that someone should go do this. And he was the one who thought about that.

Theo: I wonder what other low-hanging fruits are just sitting there like that.

Austin: Yeah, I think chasing that is actually probably much better for the next for a smart EA person who's trying to figure out what to do, trying to upskill in AI safety could be a good option.

Theo: Of course, if everyone is doing AI safety and there's no one left doing anything else of value, then that would be a problem.

Are there any questions from the audience?

Audience Member: I would say having observed Manifold and Manifund, a lot of your success seems to stem from the fact that you guys execute really fast and move quickly. What have you learned? I guess it's a two-part question. How do you think you guys got off the ground and were able to execute so quickly? And then, what have you learned in the process that you've iterated just taking action as an entrepreneur?

Austin: It's interesting because I don't even really think of Manifold as moving fast. I just think of most other software organizations as moving slowly for some hard-to-describe reason.

We picked some really good winners with regards to the technology early on. We started with Next.js, we started with Firebase. These were both tools that helped us iterate very quickly. We happened to have a fair amount of skill in these before starting Manifold. I had been working on an online board game for a long time before this. James and Stephen had been working on another React site before this. So we came into this with experience launching websites and startups. I think that helped us maintain a very high development velocity.

I do think that software is just not that fundamentally difficult. It is a field where you can iterate very quickly, and we leveraged that a lot. So that's on the building side. Then there's also the feedback loop side. We took the YC advice of talking to users to heart. We have a Discord channel and a Discord server where a lot of our power users hang out. They talk to us. Whenever something goes wrong, we find out about it very quickly, or we can ask them about things all the time.

The Manifold site itself is also another way where people can talk to us because this is a particular nature of building a social network. We're just on it all the time, and just by being on the Manifold site, people can create prediction markets about Manifold. A lot of this was how we got to very fast iteration early on. Just knowing that it's possible, I guess doing things that you have familiar clarity with, that's all helped with the execution.

Theo: Just ship, that's the key. Well, thank you so much for coming on the show.

Austin: Thank you so much, Theo.

Theo: Thanks for listening to this episode with Stephen Grugett and Austin Chen. If you liked this episode, be sure to subscribe to the Theo Jaffee Podcast on YouTube, Spotify, and Apple Podcasts. Follow me on Twitter at Theo Jaffee, and subscribe to my sub stack at

Some of my biggest takeaways from this episode include, prediction markets work better the more large and liquid they are. It's fundamentally hard to apply them to certain areas like dating. And there's a lot of room for innovation in philanthropy, like what Manifund does.

Be sure to check out Manifold Markets at, Manifest at, Manifund at, Manifold's Twitter at Manifold Markets, Manifund's Twitter at Manifund, and Austin's Twitter at akrolsmir. All of these will be linked in the description.

I had a great time at Manifest, and really enjoyed doing a live, in-person interview with an audience. I hope to do more soon. Thank you again, and I'll see you in the next episode.

Theo's Substack
Theo Jaffee Podcast
Deep conversations with brilliant people.
Listen on
Substack App
RSS Feed
Appears in episode
Theo Jaffee