Unchained - Numerai's Richard Craib on How to Crowdsource Good Predictions - Ep.89

Episode Date: October 23, 2018

While many companies like to use buzzwords like crypto, blockchain, AI and encryption, Numerai actually combines all these technologies in a novel way. Founder Richard Craib discusses how his hedge fund started off crowdsourcing good financial models from data scientists to optimize its trading strategies in global equities. He then explains how it airdropped a token, the Numeraire, to those data scientists, giving them the option to stake some Numeraire for models they were especially confident in, and how staked models have higher returns. He also describes Numerai's new Erasure protocol, which helps build a marketplace for predictions for sellers who could profit more from selling their predictions than trading on them, and for buyers who can now use the signal of staking to filter between good and bad data. We dive into the protocol's "griefing" process, in which buyers can punish sellers, as well as the thorny issue of whether selling predictions based on proprietary information could constitute insider trading. Plus, we explore whether or not other hedge funds would really want to buy a token created by another hedge fund, and what Numerai's plans are to become more decentralized. Thank you to our sponsor! Onramp: http://www.thinkonramp.com Episode links: Numerai: https://numer.ai Richard Craib: https://twitter.com/richardcraib My Forbes article on the Numeraire: https://www.forbes.com/sites/laurashin/2017/02/21/this-is-the-worlds-first-cryptocurrency-issued-by-a-hedge-fund/#3cda45f360b6 Erasure: https://erasure.xxx Announcement about Erasure: https://medium.com/numerai/numerai-reveals-erasure-unstoppable-peer-to-peer-data-feeds-4fbb8d92820a Unchained episode with Olaf Carlson-Wee of Polychain Capital: http://unchainedpodcast.co/why-the-first-employee-of-coinbase-launched-a-hedge-fund Unchained episode with Joey Krug of Augur: http://unchainedpodcast.co/joey-krug-on-how-augur-is-like-any-other-tool-ep79 Learn more about your ad choices. Visit megaphone.fm/adchoices

Hi everyone. Welcome to Unchained, your no-hyped resource for all things crypto. I'm your host, Laura Shin. If you've been enjoying Unchained, pop into iTunes to give us a top reading or review. That helps other listeners find the show. Your branding and website are the first things your audience will see. In the ever-expanding world of ICOs and blockchain startup, you need to stand out from the pack. OnRam is a full-service, creative and design agency that will help you amplify your brand with a perfect website, logo, collateral, or custom design project. Get big results in no time by visiting think onramp.com. My guest today is Richard Crape, founder of Numeri. Welcome, Richard. Good to be here. Your trajectory in crypto is a bit atypical, and it goes in stages. I'd like to start at the very beginning, but also move somewhat quickly so we can get to your new, your most recent announcement about your new data marketplace erasure. But to start, describe what you were doing before crypto and how you got into crypto. Before crypto, I was studying pure mathematics. And I then went to work for a asset management firm, and I was a quant.

And my job was to basically try to apply machine learning algorithms to the datasets that they had to try to make a new fund product for them. And it worked very well. But around this time, I was also reading about Ethereum and investing in Auger and Ethereum. And I was sort of thinking about blockchain as well, and thinking that maybe machine learning plus blockchain could kind of allow you to make a new kind of hedge fund altogether. Oh, interesting. So that I think leads us to Numeri. Why don't you describe what Numeri is, and especially and also how it worked at the time you launched it? Yeah. Numeri is the first hedge fund to give away all of its data and allow anybody around the world to model it with machine learning and submit predictions back to us, which we then train. in our fund. It's not a crypto fund. It's a global equity fund, but the way that the alpha is created

is unusual. So usually it's a really bad idea to give away your data because it's like you're giving away your edge. And so the way that Numeri, the trick with Numeri was that we gave away our data, but in this totally obfuscated form. So people could do machine learning on it, but they had no idea what it represented. So if you download the numerai data, it's just 50 features. This is feature one, feature two, feature three. You have no idea what feature one means or feature two means.

And you're just modeling a binary target, one or a zero. So it's this huge grid of numbers between one and zero. And it turns out if you can still do machine learning on that, you can still find patterns even though you don't know what you're modeling. And so this kind of aligns incentives, like the point. problem is with sharing data is that people could run off and start their own hedge funds. By doing it this way, you're not only making it impossible to steal the data, but you're making this incentive to once you build a model, you should submit the predictions to numerai.

Interesting. And just so I understand, essentially, they don't really understand what they're looking at. It's really more just this sort of abstract. It's like when you look at one of those quizzes where it's just like the numbers and you're looking for a pattern in the numbers and you want to get. ask what the next number in the pattern is. Is that kind of what they're doing? Exactly. And one way to think about it is if you have a scatter plot of points, you can easily find a line that's the line of best fit across those points. Even if you don't know what the X and Y axis is, you can still

find this line. And Numeri, it's like you're finding this line, but in 50 dimensions. And so the patterns are all there. And a machine learning algorithm can pick up on them, even though you don't know what the data is. Interesting. And I know that there are other hedge funds that have turned to crowdsourcing information. So how does Numeri differ from some of those? Well, the big difference is the way that we share data. And what's important about doing it the way Numeride does it is we're not trying to get our users to build an institutional grade, long, short, equity strategy from the ground up using raw data. I think that's a task that's too, to, to, big to crowdsource. So I don't really think you can crowdsource a whole bunch of trading strategies,

but you can crowdsource the kind of like machine learning optimization of a particular strategy. So to give you an example, like if we were to let users do anything, they could say make a model that's really good at Japanese tech stocks. But maybe we don't even trade those stocks. And maybe we don't like very short-term trading. By giving away the way, the data and the way we do it, we're deciding what the universe of stocks is, which stocks we want to include in the universe, which features we want to allow, and the time horizon that we want to hold for. So in a way, we're really shaping the strategy, and our users are just coming up with the best possible model for that strategy. Oh, interesting. But then do you shape it via the type of data

that you're giving them? Yeah. So, yeah, we only give them certain data that, that's sort of like features we want to use. So it limits them to model only our dataset, which has its advantages in that they're not going to produce anything we don't like. And also, it's like a little bit more scientific, because if we just said, send us any strategy, it's not as good as saying, here's a data set. You have no idea what it is.

It's completely blind. It's like a blind test of whether you're good. So you can make much better claims about whether the models are going to generalize if the people haven't overfit. Okay. And what kind of data were you giving the data scientists and where were you obtaining this data? Well, we don't say what it is, and that's why it's obfuscated. But it is quant-style data. So it's the kind of data that other hedge funds might use. It's nothing too obscure. It's not like image data. or satellite images or anything strange like that. It's really like very high quality multiple decades of equity data.

Oh, interesting. Okay. And just for people who aren't familiar with the hedge fund world, just to find the word quant. Quant, it's sort of short for quantitative financial analysts, maybe. And it just means like someone who's looking at finance really without a, business mindset and more looking at data and statistics. Okay, interesting. So then let's talk about the way you were encrypting the data. How were you able to encrypt it in a way that they were able to make use of it? So one of the first things that really inspired me was reading about fully homomorphic encryption

back in 2015. And since then, you've seen a lot of people talking about it. It's basically like a neat way of encrypting data where you can still do operations. on the data, but you have no idea what it is. But it's very, very slow. So it's basically impractical. And it's not like two times too slow. It's like a million times too slow. And it's not really getting much faster.

So the first idea for Numeri was that we might use fully homomorphic encryption, but then we ended up releasing a dataset that was much more manageable for the average data scientists because you have this thing with Numeri where you might have data scientists that are very, very good at machine learning, but have no idea about crypto. And we wanted to be accessible for people like that. So the data set we released uses different kinds of obfuscation techniques to basically make it very difficult to know what the data is. And then at that time, you know, before you launched your token and everything,

how are you paying the data scientists for their contributions? The very first week of Numurai, I think, we paid people with PayPal. And we only had three people at the company, and we were all pretty busy. And I was just like, we should just use Bitcoin. And at that time, I think Bitcoin is on $400. It was 2015.

It wasn't particularly good year for Bitcoin. And many of our users maybe hadn't even heard of it. Like I said, they're data scientists. They're not necessarily into crypto. But it just was much easier for us to use Bitcoin. And it had this weird effect. Like the minute we switched over to Bitcoin, suddenly we were a much cooler company.

And suddenly it was like, oh, you know what? It's much more interesting to win $1 in Bitcoin than to win $1 sent over PayPal and delayed by two weeks and blah, blah, blah. So it also allowed our data scientists to be anonymous, which also was a very compelling feature. And Numeri kind of had this early momentum of being, it sort of like didn't look like a hedge fund, It looked a lot more like a crypto cypherpunk kind of thing.

And that ended up being kind of part of the early story of Numerai. So essentially, just walked through the whole process. You essentially were giving them this data. You were releasing it what? Like on a Monday? And then by Friday they were submitting models. And then who would get paid? Yeah.

Yeah, we gave away the data. And we had these, at that time, it was monthly tournaments. And people would download the data. run all kinds of different machine learning algorithms and submit predictions back to us. And we paid them based on how well they did on the out-of-sample test data. So it's this pretty rigorous way of checking how good they would be. What happened, though, is people ended up making lots of accounts. And so you'd make like a thousand accounts and just hope that one of them got lucky.

And that's, in a way, the broadest problem with crowdsourcing. it's sort of it's sort of not immune to civil attacks and you might have someone make a thousand accounts, try to get lucky and not really be submitting the model they really believe in

and that's when we started thinking about creating our own cryptocurrency because if you can make people stake even if they have a thousand accounts they're going to want to stake the one that's the best and it ends up being this awesome filter to make people to make people stake cryptocurrency and enables you to create a crowdsourced hedge fund where in the past you would always have these problems of people overfitting and not having models that generalize to real data.

So talk about that. You launched the Numerary in 2017, in June 2017. What was it and how did it work? Yeah, we announced Numerary on February 20th. I think Ethereum then was on about $5. You wrote about us, actually. World's first hedge fund to create a cryptocurrency. And we were really, we were trying to make an application. We were trying to make a Ethereum application that people actually used and that actually solved a very specific business problem, which was not knowing which users to trust. And then in June 2017, that was when we launched. And we decided not to do an ICO.

We gave away the tokens to our users. We gave a million of them away. and within the very first day, people were using it to stake. And since then it's been staked almost, I think it's over 18,000 times different stakes. So it's making it one of the most used Ethereum application tokens by far. Any other utility token might have a lot of transfers, which is sort of just to do with trading of the token. But very few tokens have actual use where, the functions that are not transfer are being called in those contracts. So it's pretty cool that it became

this really well-used token, even though it's quite a niche user base. And so describe that staking mechanism, they were essentially, instead of doing the Sibble attacks, now they actually wanted all their models to be identified with them? Is that what happened? By asking them to stake, even a small amount, the staking, the deal is, if you stake your predictions, then we can destroy your stake if your predictions turn out badly or if they turn out well, we'll return your stake and pay you more, eath, pay more money. So it had this effect of, you know,

no one's going to risk a stake on something they think might get burned unless they really believed in it. And so suddenly all the people who were staking had much higher shop ratios on their back test, much higher returns than the people who weren't staking. So it's just kind of amazing to just like use a new technology and you just like flip a switch and suddenly the whole company made sense. It was like now you really can crowdsource predictions.

Oh, interesting. So I want to learn more about that. Like what were all the behavior changes that you saw? Like did you see a reduction in the Sibyl accounts and stuff? Yeah. So we, about half of the users. ended up staking, I think of the people who participated, the participation went down because you now had to stake.

So the only way to make money was with staking. And then of the participants, the ones who did stake had better performance. So it was really quite instantly exactly as we'd hoped for. And a lot of that is to do with the fact that we gave it to our users and we gave it to the users who had earned the most Bitcoin in the previous kind of version of the website. So we were giving it to the people who were most likely to use it and the most likely to perform well. Oh, interesting. Yeah. And just to give us a picture over time, so initially you started with PayPal, then you switched to Bitcoin. How much were you paying out to people in US dollars based on?

We were paying very little at first, something like $10,000 a month. And that would be split amongst how many users? I think we're paying the top 20 or something like that. Okay. And then by the time you launched Numerare, how much were you paying people? Well, Numerare ended up being put on exchanges. So, like, Bitrex put us on the exchange and started trading at these very high prices. So for a short time, you know, we'd given out a million tokens as an airdrop.

and in the early days of trading of NMR, it was trading at $160. So it was like an air drop of $160 million to a very small user base. So that created some distortions that we couldn't really have predicted. Like meaning they staked only very little because it was worth so much in dollars? Well, just, yeah, we didn't really know, didn't know how much the token was going to be worth or whether it would be traded at all on an exchange. But it was, yeah, trading at these very high prices, which made some of our users like overnight millionaires.

It made Numeri by far the most well-paying data science tournament ever. At one point, we're paying like $10 million per week. Oh, my God. It's just sort of like really, really crazy. But that only lasted a short period, right? Yeah, it's sort of like cooled down a bit. And we also... Quite a lot.

We adjusted the number of NMR we paid out. So it's sort of strange. Like it did make me think. Like on the one hand, it's a really good idea to not do an ICO. But on the other hand, it sort of sets a price. So if we, you know, other projects had done, if you do an ICO at $1, then maybe the token would only be worth $3 or something. And people would have some kind of rational expectation of what the price should be.

But with Numerare, it was like, we're only giving away. There's no possible way you can buy it. It's only for our users. and then there are just so many people who are interested in the project that they also decided to buy it on the markets and make a market for it. And so essentially what happened was you did thisirdrop and there was no value. It was just used this in our tournament. But then because BitRex opened a market and all these non-data scientists that couldn't

access the token wanted it, that's what drove the price up. Is that? I think so. It's always hard to say. And so it was at that point when it went on Bitrex that it got a dollar value? Yes. Oh, so you didn't send a dollar value. No, because we never sold it.

So we didn't know what it would be worth. Oh, wow. Fascinating. And so after you introduced Numerer, did that affect the rate at which data scientists were joining the site or no? Yeah, we definitely got a lot of new people joining. A lot of crypto speculators joined our Slack. A lot of like scammers joined our Slack and we had to shut down our Slack. That's surprising.

It was very like, felt like, wow, we've been like overrun by a different. a different group of people. Well, I think the whole crypto community thinks that. But anyway. But I think, yeah, they also, again, it was like we're just making this very simple, simple application that we wanted people to use. And we didn't have any, the sort of marketing of it wasn't big. So a lot of people like, why don't you, why don't you, you have a telegram group?

We didn't even have a telegram group. Why don't you do this? why don't you do all these different kinds of activities to get to get more people to buy the token or something like that. And so that was kind of peculiar to have this new group of people that felt they had sort of try to exert pressure on the overall project. And I think what the choices we have made have had much more like durability and long-term focus, which is now paying off as the other crypto projects are, you know, either gets

or get zero usage. And so I think we're looking relatively good now. And how is the performance of your hedge fund changed from before numerare to after numerator? Are you like getting better performance? Well, we don't talk about the performance of a hedge fund. But can you just say performance is up after introducing the numerator or not even that? We can't say.

But what we can talk about is the back-tested performance. And so on the three years of holdout, which the users can't see the answers to, we can show that the accuracy and the returns are better on the back test. And is that only with the staked models? Because there are still people submitting, but they're not staking. Exactly. So there's about, yeah, about 1,400 submit, 700 stake, something like that.

And the stakers are much better. And are you getting more models submitted now with the numerals? rare or no? More since... Like, just the sheer number of models that are being submitted? Well, we've definitely taken away out a lot of the bots because they were like, you know, so it's hard to know. But they're definitely a lot more genuine users who are much better than ever. So it's all been really good. And what about assets under management? How much do you have now and how much did you have before?

We also don't talk about that, but there's a way to check whether we have more than 150 million. and if you were to check, you'd see that we haven't filed for that, so we have less than $150 million. Okay. The scorebed app here with trusted stats and real-time sports news. Yeah, hey, who should I take in the Boston game? Well, statistically speaking. Nah, no more statistically speaking.

I want hot takes. I want knee-jerk reactions. That's not really what I do. Is that because you don't have any knees? Or... The score bet. Trusted sports content, seamless sports betting. Download today.

19 plus, Ontario only. If you have questions or concerns about your gambling or the gambling of someone close to you, please go to conicsonterio.ca. Local news is in decline across Canada. And this is bad news for all of us. With less local news, noise, rumors, and misinformation fill the void. And it gets harder to separate truth from fiction. That's why CBC News is putting more journalists in more places across Canada, reporting on the ground from where you live, telling the stories that matter to all of us. Because local, local news is putting more journalists in more places across Canada. Reporting on the ground from where you live. Reporting on the ground from where you live. Telling the stories that matter to all of us. Because local. Because local, because local, because local Local news is big news. Choose news, not noise. CBC News. I should just think of more questions where you're going to be like, we can't talk about that. But instead, all of them. AUM and returns are the only things we don't talk about, I think. Okay. So I'll ask you, I know that Howard Morgan of Renaissance Technologies was an early investor in Numerime. And, you know, they're a big pioneer in quantitative and algorithmic trading. So how was that influenced? you know, the development of Numeri. Well, Howard Morgan is an incredible investor to have.

He's not only extremely knowledgeable about quant data finance, but also VC broadly. He's a seed investor in Uber as well and partner at first round. So he was a really amazing person to meet. And then he also has, he used to be a professor in computer science, so he kind of knows about crypto, has very, like, very, very, very, high raw intelligence and so was very flexible in his thinking about, you know, like learning about Ethereum. And so he was an amazing investor to have. All right. So we're going to discuss

their new decentralized data marketplace erasure. But first, I'd like to take a quick break for our fabulous sponsors. OnRamp is a full service, creative and design agency that has helped numerous companies, including many in the blockchain and crypto space, maximize their brand awareness, gain traction and accelerate growth. Whether you're a startup company launching a new brand or an established brand exploring a new campaign, EnRamp has you covered. OnRamp has a passion for boosting business results

and can help with everything from logo and website design to full creative execution. Focus on your core technology and leave the rest to OnRamp. To learn more and see how they've helped passionate entrepreneurs achieve their dreams, go to thinkonRamp.com. I'm speaking with Richard Crape of Numeri. You recently announced a new protocol called Eurasia. Describe it.

Yeah. Well, the thing about Numeri is that all of the modeling is happening on the data set that we give out. And so the question is, you know, Numeri has really been able to solve the problem of crowdsourcing machine learning models on a given data set. But what if we could crowdsource data as well? And what if people could use their own data? to generate predictions and submit those. And I have been thinking along those lines for a while. And, you know, they are really two pieces of having a good hedge fund.

And number one is having good data. Number two is having good talent. And Numeri has, you know, 100 times more data scientists who have signed up to it than any real-world hedge fund in the world. But there's a – but maybe we have way less data. than Renaissance have today and way less than 2 Sigma. So could you do something creative on the crowdsourcing of data? And so I wrote this white paper actually over a year ago about erasure.

And it's like it's acknowledging the fact the same problem is if you try to crowdsource data, you'd get a bunch of really bad data. If we had a part of our website that said upload data and we'll pay you, people would upload pictures of Pepe and Trump and it wouldn't be financial data at all, or even if they did upload financial data, it would be really badly formatted. So you need to have some kind of stick, carrot, and a stick. So you get paid for submitting data, but you also could get your steak burned. And so the idea is could you have a decentralized data marketplace where anybody could

upload data and stake it as a way to put skin in the game and to show that you really believe in the quality of that data. And if you did it in a decentralized way, then anyone could buy it. So right now, Numeride doesn't do any crypto trading. But what if someone had predictions on crypto? They staked a whole bunch of predictions on Bitcoin and Ethereum. And then Polychain came to Erasia and bought those predictions to trade in their crypto. fund. And there's no reason why we should limit that, why it should just be a data marketplace for Numeri. So we started thinking about it as being a whole new protocol that anybody could build on and anybody can submit data to and anybody can buy data from. So I think it'll be like,

it'll become like this place where all hedge funds buy some portion of their data and get connected with amazing data sources of all kinds. So it's definitely a whole new future for Numeri because the Numerare token is the token that ERA is going to be, its native token will be Numerare. So something I didn't understand was in your Medium Post where you announced Aracier, you wrote, you're a hedge fund manager. You get an email that says, over the last two years, I have predicted Google's daily stock price with 70% accuracy. And I'm willing to sell you my next prediction for $10,000. You calculate that this next prediction, if it's right, can make your fund in an expected

$1 million in one month. So, first of all, do people really send emails like that? In a way, yeah, sometimes I'll get emails like that where it's like, but it's a little bit more elaborate, where it's like we have the special data, other hedge funds buy it, it works really well for them, you should buy it as well. It's very high quality. If you integrate it with your systems, you'll make more money. And we're like, yeah, but how do we know?

it's going to really work. And if it really works, why are you even selling it to us? And so there are all these questions when someone is pitching you to buy data. You always have to be very doubtful. Oh, right. Like why wouldn't they just use that data to make out trade? Or aren't you being, or are you part of a civil attack where they're only selling you the best one that they have and all the other ones that they, that haven't worked,

they're not selling you. So they have nothing at stake. Okay. So that was the first thing that I wanted to know. And then the second thing is, so in this marketplace, is all the data labeled? Because, so numeri, the way that it's been running so far, is like, people don't know what the data is. But in this case, people are saying this is what this data is? Yes, exactly.

So what they're doing, it's typically going to be used for prediction feeds. So people are going to submit data that'll, you know, it'll be like a CSV file. And it'll say Bitcoin is going to go up with a 55% probability. Ethereum is going to go up with a 0.49 probability. And so everything will be formatted in that way, almost like as a prediction. But there's nothing to stop you from making a data feed of anything else as well. But I think the main use case will be financial predictions. Okay.

And so what are all the different types of things that you can imagine people using ERAC for? Well, I think there are people who are very good. at making models, but they're not really in a position to trade them. So you might be like a student at MIT or something, and you've downloaded the blockchain data, and you've connected that data set with Twitter data. And now you have like a sentiment indicator that also uses blockchain data and actually can make really good predictions on Ethereum. It's price. But, but, But the problem is you only have, say, $5,000 of savings. And so you could get on Coinbase and trade that $5,000 up to $5,500,

but it's sort of like not worth it. So the case for Easure is that there are a number of people in this position where they actually are very good at making models for things that are quite good, but not good enough to want to start their own hedge fund. And instead of starting their own hedge fund, the much wiser thing for them to do would be to sell the data. The problem is if they come to us and say, we want to sell you that data,

or they go to polychain, we want to say, no one will even respond to them. But if they say, instead of trading the $5,000 they have, they put the $5,000 by NMR, they stake it on their predictions. Then suddenly, I might buy those predictions because, look, the guy who's created them is saying, I don't mind if you burn the stake if it doesn't work out. So you have much more, yeah, much more sense of like the fact that these predictions are

actually going to work if you have someone staking on them. And so what format does the prediction come in? It's literally saying the price will hit this number by this date or what? Yeah, it's just this like a CSV file that says BTC 0.51 and maybe some metadata on the fact that that's a weekly prediction or a monthly prediction. Like by January 1st or, okay. Okay. And then to go back to, so just you keep referencing this example of like crypto prices, but what are some other, you know, is it really crypto funds that you think will use this or? Crypto funds will definitely be early users. we've talked to some that are very interested

but Numeri would be the other major user for equity data and again there are a number of ways you might think of making a model to predict the price of Netflix or the price of any U.S. equity based on any data sets and if you create an erasure feed

out of your predictions numeri in particular will be one of the first funds to look at them and consider buying them. And so it's pretty much all financial markets data, whether it's crypto markets or equities. It's definitely the focus, and that's what Numeri is good at and our users are good at doing. But I do think it will be more widely used than people think. I think there's a lot of examples where you have this position, that a seller of information is in, if you're trying to sell information,

the person's going to be like, what is the information before I buy it? You'd be like, well, I can't tell you because you won't buy it if I tell you. And so instead, if you stake it, then it's like you're giving some evidence that some economic, putting some economic risk to say that the information's good.

So a good example of this could be with journalism or with like whistleblowing or something. Be like, I have this very valuable data something to do with some kind of secret or something, and you encrypt the data, stick a big stake on it, and only the buyer with the key can look at what the data is. Oh, interesting. Wait. In that scenario, so journalism, it's extremely competitive. Is there something where, like,

it's a one-time purchase? Like, I buy it and no one else can get it, or how does that work? Yeah. Most of the feeds on ERAO will be, like, exclusive. So only the buyer can look at it, but that'll also be extended. So it could be, it could be other use cases as well. I mean, you also might not want to submit a prediction. You might just want to create the price feed. Like what's the price of ether? And you could just create a, you say, hey, I'm hosting a feed, which is the price.

And if the price is ever wrong, you're welcome to slash my stake. and now you have maybe a more reliable price fee than you otherwise would have. Oh, interesting. Yeah. And so, okay, so you're saying most of the data that's sold will only be sold to one buyer. Because when I read about it, I thought to myself, oh, if we have a lot of hedge funds that are making trades based on certain predictions, then won't those trades affect the price, which would then keep the prediction from being as accurate as it would have otherwise? Yeah.

Well, you definitely, if you are buying a prediction and then you subsequently trade in the direction of that prediction, then you are going to move the price toward where that prediction is saying. But you're profiting all the way along. So it's definitely good to have the edge where you're the only one who can see it. Right. Okay. Okay. Yeah. And so who decides whether to sell it only to one? Because why wouldn't somebody who's selling it just say, I'm not going to limit it to one? In the case of four hedge funds in particular, they're going to prefer if it's exclusive. So on Erasier, both, this isn't discussed as much in the white paper, but both the buyer and the seller put up a stake. And there's a, yeah, the one thing that people don't understand is it's kind of like how blockchains don't have internet connections. So you can't say something like, the stake will get destroyed.

if the predictions turn out badly. It's like, how do you know whether the predictions turn out badly? What data source are you going to check? So everything on ERAA is discretionary. So if you create a stake and I buy your prediction feed, I get the right to slash your stake

for whatever discretionary reason I want. So if I'm unhappy for whatever reason, maybe I did make money on the prediction. but I didn't make enough, or maybe there was some other piece of what I expected that I didn't get. And giving the buyer the right to slash the stake for discretionary reasons actually gets around all of the problems of having validator nodes and decentralized oracles and all these things that don't really exist today and may never quite work. So having a discretionary relationship with the buyer is much better.

Well, yeah, so this is that concept griefing that you described. Like, I don't understand. Why would somebody pay you to destroy their own money? Like, even if it means that they can do this, you know, sort of revenge move against somebody else. I don't get it. Yeah. Well, it's not really rational to do in a setting where you're not in a repeated game. So if you've ever heard of the ultimatum game, it's like this game where imagine I get $100.

I have to decide how much to share with you. And if maybe I'll say, I only going to give you $10. And if you reject it, then neither of us get any money. So I'm like thinking, I wonder how much I should give you. Okay, maybe I'll give you 30 and I'll keep 70, just to make sure that you, you know, don't get angry. But really rationally, if I give you $1, you should accept because you get one more than zero. Right, right. but in practice when you have these games repeated people tend to grief people burn both people's money

until it gets to a level that somehow seems fair and it tends to kind of converge on like 70-30 or something like that and so erasure is a bit like that you're not just saying oh i'm upset i'm going to grief i'm going to grief the person by destroying some of my tokens to destroy a lot more of theirs you're actually, you're, and then the game's over, it's actually, I'm going to destroy a little bit now to make them notice that I'm watching them and I'm upset and that in the repeated game, it'll tend to, it'll tend to, you know, keep everything in line. Another way to think about this is it's similar to like a lawsuit or something. Typically in a lawsuit, if you don't kind of go to court, you end up settling. in some way, or even you're just doing it to kind of scare the other party to know that you really want their compliance to the contract. So it's some kind of like stick in the relationship.

So even say with Bloomberg, like Numeri buys data from Bloomberg to, for some of our hedge fund operations, we need Bloomberg. Now, the one threat we always have with Bloomberg is if we're upset with the service, we can just stop paying for it. But the other threat is maybe we'll sue them if the service is really bad. and it might cost us a lot to sue them up front. We might never settle for more than it costs us. But it's the fact that that option is there. It's not that that option gets used.

The fact that that option is there keeps the alignment much nicer than it would ordinarily be if there were no legal system. And I see the griefing on erasure as like the legal system. It's like, not how many people are going to do it, but the fact that it's there, it's like a little bit of a, incentive to do the right thing. Okay. And so are these relationships between buyers and sellers? Do they extend over time? Because if you're saying it only works in a repeated game, like if I buy data from you for one month, but, you know, and I'm unhappy and then I grief you and then it's over. Like, did you really

learn a lesson? Yeah. Well, definitely, in the prediction setting, it's definitely a repeated game. you're not really going to be buying one-off things because for say a quantitative strategy if you build a model on the price of ether you can keep running that model and keep generating a new prediction every single day so every single day you're going to have this prediction and if you're engaged with a buyer and they decide to start trading that signal in their fund they're going to want to keep getting that same signal because they're going to build systems to that rely on that signal. So you're always engaged in a kind of feed relationship.

And that's what's also interesting about erasure versus other decentralized data marketplaces because if you do have any situation where, imagine I'm buying like a whole book from you, the minute you give me the book, I have it. I'm not looking for the next page. I've got the completed book. Or if you give me a photo, I have the photo.

There's something about feeds where, all of the value is in the future. I don't care about your predictions from last week. I care about what you're predicting now. So you always want to maintain the commercial relationship with the person you're buying from. Okay. Okay. So now I think I have a better understanding. So basically when I enter into a relationship with a seller, it's like over some period of time. And during that time, I can grieve them multiple times within that time period. Is that? Yes. Yes. Yes. Yes. Okay. And you can choose any amount to grief them by.

Okay. But actually the griefing, I thought the amounts were set by the seller. The griefing factor is. So it could be like $1, could destroy $10 of the seller. Okay. But you could choose, you know, I only want to grief, I'm going to grief $100 to destroy a thousand of yours or much less. All right. So let's walk through the whole thing from end to end. Okay. And just so I fully understand, so from the beginning, both the buyer and the seller need to own Numerary in order to do this, to do anything. Is that correct? Yes, because both the stakes are in that.

Okay. And then what? The seller posts something on the Eurasia marketplace. What does that look like? Yeah. So I'll explain the first kind of problem. If you, if you, the first piece that the protocol solves is the idea that you can't trust someone's track record. If I say, oh yeah, I've predicted the future perfectly over the last two years. There's a sense which you just don't trust me. So the first thing is if you're a seller of predictions, you submit all of them to the blockchain. So every time you make a submission of a prediction, you submit it to the blockchain so you get a timestamp.

And it's a timestamp that's completely reliable and that everybody will trust for the rest of time. So at least when you start building a reputation of being good at predict, something, everybody agrees that you are because everybody can go back and see that what your track record is. So that's the first very important point about Eurasia. All right. And then keep walking through the steps. Okay. And then imagine you see someone that they have got this totally verifiable track record of being good at predicting something. And they've also put a stake on it. So you know that they care enough to at least do that. And they're saying that they're going to charge you

100 NMR per week to buy their feed. So the buyer comes in and says, okay, I want to buy this feed. What the buyer gets as a special right is, everybody can see the historical prediction performance, but no one can see the most recent prediction unless you're the buyer. So the buyer comes in and says,

I like this guy's performance, I like how much he's got at stake, I like that he doesn't mind if I grief him, this seems like a decent thing to buy, sends the first payment for the first set of predictions, gets a key, can decrypt the most recent predictions so that only he can see those.

And then from then on, is kind of engaged with the seller to continuously buy every prediction that's subsequently submitted. And with those predictions, takes them to do whatever he wants with them, including trade them in his own hedge fund. Okay. And then every single day

or at any given moment, the purchaser can grief the seller? Yes. Okay. And then it's based on the ratio that the seller sets. Yes. And that's important to me. People say, well, how do you model this?

Or like, how do you know for sure that the griefing factor is right or that it's going to work? And the important point is that people have freedom to choose whatever griefing factor they want. You might create a fee to say, I don't want anybody to grieve this. I just want to, I don't want it. it that way. But then you might not get as much buyers of your feed or as much interest. So it'll be up to you to decide that. And I think market forces will get the griefing factors to be something that seems sensible to both the buyers and sellers. Yeah, I imagine it will like converge on some sort of standardized. Yeah. Exactly. So if you've got a really good feed, you probably stake a lot and allow

a high griefing factor. Right, right. This is interesting. And then, like, let's say that I enter in a relationship with a purchaser. I'm a seller. And I've set my griefing factor, you know it. I don't know, like one to ten or whatever. But then I realized, like, do you ever read, like, reviews?

And then you just sort of, like, I notice sometimes you see a really scaping one-star review. And then you click through and realize that that person gives only one-star reviews everywhere. Like, so what if you're a self? and then you suddenly realize, oh, this person I'm in a relationship with or in this marketplace is like one of those types of people. Then can you change the issue? Yeah. You're going to be, I think you should be able to not engage with certain people, maybe. And I do think it's going to be quite nice to have your reputation follow you around. So it's pretty clear that, oh, this person is griefed, those other people, I better be careful with him.

and then the same for sellers, if they have some prediction feeds on crypto that are very good, and now they've started a feed using the same Ethereum address for global equities, maybe you're going to trust them a little bit more. So there are a lot of these sort of softer factors that will come into play where people will start to develop a kind of trust. And maybe it'll be, it'll certainly be known that this is Numeri who's buying your feed from you. Or this is 2 Sigma who's buying your feed from you.

So I don't think people will mind trusting that 2 Sigma isn't going to be just totally malicious and aggressively griefing for no reason. And, you know, right now people trust numerai with a griefing factor of zero to infinity. Because on numerai, people are staking and we can destroy their whole stakes without destroying anything of our own. And they're happy with that relationship. So it's kind of like a matter of preference. and over time, the right price, people will get paid full price, and the griefing factors will kind of stabilize.

Okay. But also, once I'm in the relationship, can I change the griefing factor? Or is it set right at the beginning and cannot change? All of these are kind of like parameters that it's the question of maybe on day one, we don't, we don't allow that. But yeah, everything will be, I'm sure there will be many more settings. The way I see it is if you make a, if you make a, if you. make a stake and you're engaged with a buyer, they kind of have 30 days to grief you.

And that will be the first, like, setting. Okay. Yeah, you probably don't know all the details yet. How does this differ from other prediction markets like Auger? Yeah, it's kind of interesting. Like, ERAA is a predictions marketplace, and Auger's like prediction markets. And it's like, how are they related? The thing about Auger is it's similar to a market.

So, like, if you had predictions on, if you had some insight into Apple and there was an Apple market on Olga, you would buy the Apple market. And thereby have your private information kind of be expressed in the price of that market on Olga. So both of them, in a way, get information into the market. but erasure is a little bit more direct for people who maybe have some information, but they don't really want to trade on it themselves. So they'd much prefer to sell the information, led a hedge fund who's much more experience at the other parts of trading to do with, say, risk neutralization or diversification and other things.

Yeah, like you can imagine having some very keen insight into, say, like Apple's hardware, knowing Apple's trends and hardware, and you kind of want to sell that information directly. And you don't want to, if you just bought Apple on Auger, you're kind of exposing yourselves to all kinds of market risks and things going wrong with the software, things going wrong with the currency, things going wrong in China,

and you're kind of like exposing yourself to all those risks. But if you were just selling the information and the information was good, then the hedge fund who bought that data would kind of, be able to manage those risks themselves. That's what they're good at. So I think in many cases, you're going to want to directly sell your information, not trade it.

And what are the sorts of prices you're imagining this kind of data will sell for? Well, what's good about it is that these historical track records of the erasure feeds will all be publicly verifiable. So I kind of like an idea of a future where it's like, say it's 2002, okay? and there's one feed on erasure, which is predicting the S&P 500.

And it's just got this incredible track record, a sharp of three, it makes 30% a year. And 2Sigma comes to, finds out about erasure, comes to the website and sees that there's this feed that's just incredibly good. Everyone in 2 Sigma could hate Bitcoin. They could all be skeptical about ICOs. They could all hate the blockchain, but there is no way that they do. don't believe that the track record of that, that prediction, uh, is real. And that's what I really like about it. It's like they'll have no choice but to start participating in the marketplace if over a period of, of a few years, people actually submit really good predictions.

Is there a problem with people putting proprietary information on erasure? Like, could somebody get in trouble for kind of like insider trading via the erasure marketplace? Yeah, that's, And that's a similar, you can raise a similar concern with Auger or even any technology where, you know, probably the most used technology for insider trading is probably like text messaging or email or something. So because this razor will be decentralized, there's none of the data that's being put onto it will ever be on, like, numerized servers. It'll be on IPFS and Ethereum. So it could be used for anything. Tend to think people will be using it much more for quant strategies

where you're predicting on 5,000 stocks versus, you know, you're actually like, you actually work for Apple and you've stolen some company secrets and you're selling them on Eresier. Well, you did mention that thing about knowing, what was it, something about Apple's. Apple's hardware or something. Yeah. Yeah, I mean, that's a, I think that is an example. that could be legitimate, but I think most likely people are going to be using it for quant predictions.

Just because if you, imagine you did submit predictions to Eurasia and you were right, about one stock in particular, it's actually hard to know if you're going to be good next time or the time after that. But with quant, what's pretty cool is you can make 5,000 simultaneous predictions. And if you're right on 2,700 of them, that's actually pretty good statistical evidence. Right. So in a shorter time period, you can develop as good reputation.

And what's your sense of whether or not hedge funds will actually want to buy this data? You mentioned some crypto funds. Is that pretty much it? Or like, you know, you keep talking about two sigma two. Do you know if the more traditional hedge funds are interested? They always, they always kind of have to respond to the new, new things. So in the past, very skeptical on new, kinds of alternative data sources like sentiment analysis data or or even skeptical about new tools

like machine learning but once other hedge funds start doing it there's like a huge pressure for everybody to catch up so to start with numerize is going to be a huge buyer of these feeds and we're going to be on the buy side of this marketplace giving our users time to develop their track records but at any time another hedge fund could come in and outbid us and um Over time, as it gets better, I think it'll be like the best place to buy to buy high-quality data because no one else is giving you the same kind of guarantees. First, that all of the historical predictions are accurate. And second, that if anything goes wrong and the data isn't act good, you have something to do about it, which is like destroy the guy's stake. And so you have not talked to anyone other than some.

No, we just announced Eurasia very recently. and we've spoken to crypto funds who are interested, but we haven't spoken to other hedge funds. And frankly, we prefer that it be a lot of crypto interest and allow us to be the major buyer for the equity stuff. So I think that's how it will play out initially. Because you don't want the competition. Yeah, for now.

So how willing do you think hedge funds will be to transact into cryptocurrency that was launched by a different hedge fund? Yeah, it's kind of a strange proposition in a way. It's like you can use this market. anybody can use it, but it's sort of, you sort of need to use our cryptocurrency. Like, how much Numerare do you guys hold? Well, we can mint Numerare. So we, that's something that is a little bit problematic.

Like, right now, Numerare, the way it's used on our website, it's very centralized. It's like, it's a token for Numeri. It's a token for our users, and we can mint it, and we're the ones who can destroy it. But as we shift to ERAO being a protocol that uses Numerare, you don't really want any central power over the system. So Numeri will have no special rights over any other hedge fund entering the marketplace. Except you can mint Numerair. Yeah, even that, we're going to have to stop. And so when ERAI is not going to be able to have any special powers, including minting.

Oh, then who will be minting the Numerary? we will have a finite supply like rep. Oh, okay. Okay. Yeah. And so do you guys, how will Numeri make money from Eurasia or will you not? Well, we definitely will benefit from giving more use to our already well-used utility token and more things for our users to do. And then, again, it's a data marketplace we wish existed.

I wish all the data was time stamped on Ethereum so we didn't have to trust that Bloomberg's keeping. the right time stamped on their data. So I can definitely see it having really large benefits for us as a user. And then it would be pretty cool to have the whole hedge fund industry basically using our protocol. Cool for you. So I did see, I mean, you keep talking about the volume in Numerare. And I saw in the Medium Post announcing Eurasia that you said that it was, I think, the most, it had more stakes in dollar volume and transactions in June than any other ERC20 token.

but I was looking on DAP radar and right now it's like ranked in 97th and over the last seven days there's only been like 2,500 transactions with 11 users over the last 24 hours. So why is it dropped so much? Yeah, tell me about DAP radar.

Basically they don't calculate the numbers right because the way that people interact with Numerare is in a very unique way and so because we gave it away to our users all of the stakes they're making are coming from our websites. And there are very few people using like custom Ethereum addresses to make the stakes. So in a way of thinking about it, there's very few stakes being made from Metamask.

There are many stakes being made from Numerize website itself. So it's not on chain. So it is on chain, but it's all coming from one address, which is Numerize address, even though the stakes are coming from individual user Ethereum accounts. So if they calculated it correctly, you'd be able to see this. But the DAP radar and state of DAPs, I think state of DAPs says we're number two, the number two token for storage. I was like, I don't know what that means.

I don't know why we're in the storage category. Maybe it's a store value or something. But I think over time, when ERAESR launches, more of the interactions with Numerare will be with MetaMOSC, and these numbers will start to be much more clear. But right now, you'll have to be a little bit more sophisticated and actually look at the blockchain yourself to see all the usage of NAMR. Yeah. So essentially, when I said there's only 11 users over the last 24 hours, one of them is like multiple data scientists on numerize. Yeah.

They are like, yeah, they were just this last, like yesterday last week or so, there's been 750 stakes of, and. some of those of quite large stakes of NMR. Okay. Well, wait, but what I said was that DAPRator over the last, somebody says there's been 2,500 transactions and you're saying 750? They're saying $2,500 in the last 20... In the last seven days. Maybe they're calculating it correctly or differently, but I don't know.

Yeah. Every time I check them, it's like, it's just completely off. Okay. Because, again, it's coming from one address, most of the stakes. Oh, interesting. Okay. apparently I guess on DAP radar it showed that in February you guys had 419 users but then now it's showing 10 to 15 so why was it such a big number? I really don't know. I don't have no idea. I think it's yeah there they have been a couple of users who've like again a lot of those users could also just be traders like they could be people trading NMR on decentralized exchanges or something like that where it's coming from a special address.

but that won't even, that has nothing to do with staking, which is the main, which is the actual use. So try to discount anything that's to do with just the transfer function. Right. So staking is not on chain. It is on chain. But it's not represented, um, on sites like DAP radar. Oh, okay. And same with zero X. Like you might see, oh, there's people using zero X, but they're not really using the token and they're not using it except to transfer it to each other. So there's a question about, you know, whether it's, being used at all. So the price of Numerary has fallen over time, as we mentioned.

But even after Eurasia was announced, I noticed it just, I mean, it doubled, but, you know, from a little over $3 to about $6. And then now it's back down to like below four. So how do you expect ERAE will affect the price of Numerare? I've never understood cryptocurrency valuations. There's so many strange distortions in the market. There's certain projects that are doing very. very sketchy things, like wash trading and trading against yourself.

So we don't really pay much attention to the price. And I think that's the right thing to do. And I think a lot of people, it's strange how kind of well-known it is that there's certain projects that are completely overvalued and just sort of, continue to stay overvalued. BitConnect is near universal acceptance that it was a fraud, and it took a very long time somehow for it to reflect in the price. And then in the crypto community, near universal acceptance that what Numerize doing is really, really cool, and it's a premium project with top backers and it's really being used and it just sort of has no impact on the price.

So it's always hard to know. Yeah, well, I agree. And I also don't pay too much attention to price, but just happened to notice. And then I wanted to ask about this previous prediction market in trade. It was, I guess, based in Ireland, but eventually U.S. customers were banned from using it, and that led to its demise. So is this legal in the U.S.? What you're trying to do with Eurasia? Yeah, definitely selling data is legal.

And, you know, in the same way that numerized buying from multiple data providers and every hedge fund buys from multiple data providers, and there's multiple centralized data marketplaces like Kwandle, where you can buy all sorts of different data. So there's a lot of precedent for that, and that's not a regulated industry. But there's definitely other regulations to do with being an investment advisor. And so one of the ways I kind of see it developing is you might have people

using ERAO for in ways that are not legal just like you can have people use email or HDP or anything for illegal means but then I think all the big hedge funds or all the big data buyers will not use things that aren't kind of like

stamped in some way so that's sort of how I've been thinking about it lately like you might you might have some kind of concept in the distant future of like a verified seller where they're really you know they're really honest people and they're a real business and they maybe provide some proof or metadata that they that they're that they're actually Bloomberg and they should be trusted something like that. But for the time being, it's, you know, it's very early days.

And you previously said that your master plan is to decentralize at least parts of numerai. What does that look like and how do you get there? Well, definitely ERA is a part of that plan. We want to decentralize in a very like, pragmatic way and have everything that we create be extremely non-speculative. Like, it's kind of cool that I can explain the whole protocol to you in a podcast and there's no math or things that kind of make it difficult or no speculative technology like a decentralized Oracle that you have to rely on before it could even be created.

So with ERAA, it's really something that none of the technology is speculative. Everything can be done today. we've already done a private beta, which had hundreds of users. So we've already got the smart contracts. So I think it's going to almost like, yeah, it'll almost certainly work. We know it's already working for numerai. But in terms of decentralization, this again, by us first being, okay, let's make a kind of centralized thing with numerai and numerari. Let's get our users to really use it.

Let's give it to them. And then let's really build a proof that this can work. And then, okay, maybe now let's throw away our key to be able to mint and let's let's take, let's decentralize further. With the hedge fund itself, you know, all of our trades are equities. So all of our trades are in a centralized hedge fund structure with, you know, real prime brokers like UBS and Goldman Sachs. And, you know, there's no, there's no way that that part of the business can be decentralized right now. But everything to do with the data and the modeling and I think it's going in that direction very nicely. All right. Well, this has been a great discussion. Where can people learn

more about you, numerai, numerar, and erasure. Go to our new website, erasure.xxx, to read more about erasure. And the medium post is there and the film is there. And our website for numerize, numero.com. Okay, perfect. Well, thanks for coming on Unchained. Yeah. Thanks so much for joining us today. To learn more about Richard, numeride, Numerary, and Erasure, check out the show notes inside your podcast episode. New episodes of Unchene Come up every Tuesday.

If you haven't already, rate review and subscribe on Apple Podcasts. If you like this episode, share with your friends on Facebook, Twitter, or LinkedIn. And if you're not yet subscribed to my other podcast, Unconfirmed, I highly recommend you check it out and subscribe now. Munchin is produced by me, Laura Shin, with help from Raylene Gallipali, fractal recording, Jenny Josephson, and Daniel Nuss. Thanks for listening.