Hashing It Out
Hashing It Out

Episode 85 · 2 years ago

Hashing It Out #85- Gauntlet.Network Tarun Chitra

ABOUT THIS EPISODE

In this episode, we explore the incentives of both consensus systems and defi with Tarun Chitra is the Founder & CEO of Gauntlet. Tarun is applying his experience of AI and financial simulation to blockchains. Using ‘agent-based simulation’ he helps protocol designers to discover unexpected strategies to extract more than their share of value.

Links:Gauntlet.Network

The Bitcoin Podcast Network Social Media

Donate to Hashing It Out

Discuss

Now into a welcome to hashing it out, a podcast where we talked to the tech innovators behind blocked in infrastructure and decentralized networks. We dive into the weeds to get at why and how people build this technology, the problems they face along the way. Come listen and learn from the best in the business so you can join their ranks. Welcome back, everybody, to hashing it out. I will be your host today along with Dean. Hey, everyone, say what's up, everybody, Dean? What's up everybody? Dean? I know we're going to do that every time, but we're going to borrow that from or in calling from now. So today we have with US tearoom Chitra of Gauntlet Network. Heay, so please. Hey, I'm Trin and the thunder of Gauntlet and I have kind of been spending time trying to make the economics and kind of crypto currency land a little more interpretable. But I've spent a long time where on simulation based research and hoping to apply it to Crypto as people develop crazier and crazier systems. CRAZIER and crazier is the way Brittant. So the I think the the trademark question we asked the being. This is usually what's your back story? How did you get into Crypto? How you get into the space? Spells about that please. Yeah, definitely. So ten years ago I was working at a company called dshaw research and we built a six for doing physics research. So a six in like two thousand and ten, I'd say. The majority of the people who are working in a sex either building it, building them for telecom use cases. So a lot of like customs for FT, like hype throughput FFT devices for you know, different sorts of like encoding and encryption mechanisms, as well as, you know, defense use case. But there wasn't this current frenzy we're right now. There's everyone and their mom trying to build an a sick for self driving cars, from machine learning, for Crypto, etc. But we we basically we worked on, you know, building this crazy a sic for doing physics research with the idea of applying it to drug discovery. So we were were building the simulation machine that simulated sort of the protein protein physics. And Yeah, so it was was kind of just like crazy billionaires research lab thing and he was spending all his not his fortune, on building these a sings. And you know, in two thousand and ten and eleven, when I was working there, there were a bunch of times where we would buy chip space, and the way that the chip space selection works is, you know, there's a big is big physical wafer and unless you are spending a billion dollars or buying a huge amount of chips, basically TSMC and Sam Sawing and people like that will just kind of tell you to help, probably even less...

...politely than this fuck off. They will instead be like go talk to someone who can take a hundred different orders, put them together onto one wafer and like make it economical for them at like a billion dollars, because, like they they really can't run their fabs economically until it's billion dollars. So the way that a middleman designer that integrates all these late little people who are important enough with their time, it goes to TSMC. Yeah, so the way the way this works is basically you take the way for they you cut it into blocks, so let's say one millimeter squared blocks, and then you option those blocks. are like you make a design in Ourtl, you synthaesize your RTL, you say like okay, this chip is going to be like ten cent and ten, ten meters by ten centimeters, or the total surfer, and then you go and buy. You tell the producer, hey, we need this much space. The producer takes your design, adds in a bunch of insulation and like other like physical hardware logic to make sure that your design doesn't leak into someone else's, so that, when it's sad, so most fabs work by laser etching, and so you don't want the laser to accidentally someone else's design of a gate to etch into yours. And so they hold the risk of making sure that your design zone meet and they also are oftentimes, whenever you hear US politicians complain about, like China stealing our Ip it's usually this step where this happens. Often because because the these these integrators who put together bunch of orders, they actually have to have more detail about, like what each circuit is doing, to make sure that there's like enough power in a certain place or like there's enough insulation so that there's not, you know, tunneling event stuff like that. Maybe this is too deep on on the really mixed because this is interesting and crazy, but this is like yeah, so you guys were making a six of a six, essentially to bring down the cost of a single a sick well will, someone was doing this for them, doing this part. Yeah, yeah, so like like until Bitman got to a certain size, they also had to buy from the same person. And so what happened in two thousand and eleven was one time our supplier just who's this person? Integrate season? They we gave them like twenty million dollars and then they just like ghosted us and disappeared. And they disappeared for like two to three months. And you know, when you give someone twenty million dollars, because still an escrows. At least it wasn't like you gave them money and they completely ran away. They they kind of ghosted us, didn't tell us what what there, what's going to happen, etc. And they they basically you know, three months later, Akay will give you a ten percent discount. Now, when you're you have a hundred people working on a chip and you get a three month delay because someone ghost to you, I don't think a ten percent discount is exactly what is going to cover that. So we we like we I guess my work threatened to sue them and all the stuff. And in that time that's how we found out about this bitcoin minor which at that time later became the people who made Avalon, which is one the earlier combiners. Long Story Short, I was like, people are willing to pay twenty million dollars to buy up chip space. I should go mind this thing because this seems crazy. So they jump like a couple line. Like the delay was because this provider was like, oh no, these people are like this bigcoin mining producer was willing to outbid you so that they would bump you off and screw you over just so they could get a few extra like some more lead time on the Hash War. Yeah,...

...exactly. So you know, like I said, you know right now, if you go to this similar supplier, it's extremely competitive and you're waiting for you four months. Because of self driving car Li Dar, because of like TPU M L chips, like people building special purpose and edded chips for doing machine learning, plus crypto like money devices. But back then there was no one. So we weren't used to this level of like aggressive competition, and so when that, when we saw that, I was just like Florida was like, I need to go mind a bunch bitcoin because I need to understand what this thing is if there's like all these people in China going crazy about this who we work with. And Yeah, I mind a bunch. And then two thousand and thirteen there's like a big crash and I just sold everything and I was like I'm never gonna never go to be interested in this stuff again. And then two thousand and fifteen, the paper by some Polinski and Zohar on ghost greatest heavy, heaviest greedy, heaviest observe subtree came out and that was the first time I was like, okay, this thing makes sense theoretically, because, having worked in distributed systems, I was like, I didn't really understand why this works. A lot of the probability theory that is used as wrong, like intraprobability of like understanding how put on distributions work was wrong, and like the early papers and foreign posts and stuff like that, and so I was like, okay, this this is just a bunch of crack pots. And then that paper convinced me that there's actually this safety and liveness guarantees that are probabilistic and real and like the proofs made a lot of sense and it seems like it was rigorous, and that was when I started really paying attention to academic literature. Fast forward a little bit. In Two thousand and sixteen, I went worked in high frequency trading and we were doing simulations of, you know, trading strategies against models of the market and and at that time proof mistake was first coming out, and that's when I was reading a lot of these like algorithm, not perfect, was first coming out. Right, of course there's pure point and xt and stuff, but like credible terms. Again, yeah, maybe I I because I was coming from like a more academic place, as this research place where, like I was like okay, like I needed to see the stamp of approval of Sylvia macully. I mean now I feel the opposite, kind of like you know, oftentimes the the luster of the famous person is not is a negative. But I saw the algorithm paper and I spent a lot of time reading it. The first version was way too long. It was like a hundred thirty pages and then they finally cut it down to like their twenty page submission. But that was when I was really like, okay, this the cryptography is amazing, but people are making derivatives and they're not pricing that in into their fifty one percent model. And Yeah, I just started working on writing a bunch of of code to to value figure how to value the derivatives that were implicitly being created and profstake. And the reason I say it's a derivative is is you know instake, you bond an asset, you lock up an asset and you get payments probabilistically right. If you get slashed, view as a payment. If you get a block, you in a payment. You also get transaction to these which are publicic. But then you also only get your principle at the end if you can UN bond. But there's a sense in which if everyone UN bonds at the same time and the network holds some liabilities, you might not actually get back all of your principle because of slashing events or dilution events or things like that. And so it looks a lot like a slop in some St's where you put principle up up front, you get these recurring payments and at the end, if if everything matches up, you get your and toil back. And I just like hadn't I was looking at this...

...and as I can prove a work. You don't have that unless you have derivatives on your ethics and cash power rotives, and those are so a liquid that it doesn't factor and how people value these thing. So you kind of have this leg hybrid, like, like sounds like a very unique hybrid of experiences that and you come across these ideas being spoken about in one language, you know, and which is distribute computing. You also have this experience working in high frequency trading and finance, and so you. I mean, was it what do they talk about drivers in this paper, or was this sort of a mapped in the model in your mind from improved state to and you sort intuitively thought about it as sort of financial mechanisms? Yeah, sorry, that's, I know, my own very long winding tangent, but yet it's story. Really. That's a good question. I think it. I think it's like I didn't totally realize it when I was reading these papers. I was reading them out of like, you know, I was reading them on the subway on the way to work, because like that it's just like, oh, they're interesting distorted systems things. I think when I first realize that they're there was this derivative aspect with some former co workers of mine went and raise a lot of money to do a stable Colim and I was I was like this is nuts. And at that time they were like we're going to make a layer one chain, we're going to have this proof sake thing, and I read their paper. They like sent it to a bunch of people before, and that was the first time, as like they they I think, actually really were the first people who I had seen have writing about, writing anything about these assets of being derivative. But I think that was when it really hit me. It was like this, the idea that the layer one chain would have a stable coin attached to it is like the only way that works is kind of adding in these these synthetics. And Yeah, just kind of so. So where does it get like what is it that you think? That like kind of clarifies where you're coming from the backstory. What are you doing now with it? Yeah, what is it that you're using that in safe for working on these days? Yeah, so it in in algorithmic finance. You know, a lot of the problems you work on are don't have sort of closed form analytic math solutions and you're really you have to resort to numerical methods, mainly simulation, to try to get an idea of what is the expected outcome with the average outcome? What's kind of the ninety five percent wriskcase outcome? How much risk do I hold? Stuff like that. You can't just say like, Oh, I'm going to hold five dollars for risk. At the end of the day, you can say, however, I'm holding, I probably will hold five dollars a risk, but there's a two percent probability that I owe someone ten dollars. And those methods, I think, are very important as a compliment to, you know, Smart Contract Auditing and security auditing to find out if there are incentive issues and bugs and these in contracts that come from how people use them rather than like statistically, like behaviorally. Then they then like one plus one equals five, and if I re enter this particular function call, I can take all the fund and basically I worked on those tools for more or less a decade and in different ways, and I thought that, oh, people were building the systems but they're not stress testing them against the stuff. And so, yeah, I we gantlet we've kind of built a bunch of tools for building financial models...

...directly against these smart contracts. So we have a custom version of like an theorem client that we made that's like optimized to words stress testing different things, and we run, you know, millions of simulations against the protocols and say like here is the different set of outcomes, here's a probability of certain bad things happening, good things happening, and if you tune these parameters, you you here's how you change the probability of certain outcomes coming. So we really work on stressessing this week. Initially we're working on a lot of proof to stake type of stuff, but I think nowadays, you know, proofistic in practice seems to be a dinosaur that ossifies very quickly and no one is willing to change the network, you know, like everyone's like, oh, we're going to add all these complicated governance things, but I guess what, we're never going to vote on anything because we're too afraid to change, make any parameter changes. Once it's running versus like, you know, it's much as aligned as it is. I think like the defy space is doing, inventing the exact same types of assets, except they're much wilder. Their change they're willing to change parameters through all these crazy things of them, but that experimentation is way more interesting and it's also the place where I think you need more safety guarantees. So I just I just want to synthesize so far brillant like what what it is? My understanding is like basically, you take smart contracts. Somebody might come to you with a set of smart contracts, probably defy stuff, and they're not looking for you to see if there's any bugs in it in the the sense that like an auditor like myself would look for so much as are there like strategies that exist in in this like as a result of this contract existing that are not easy to inticipate but need to undesirable. Comes said, maybe, do you have that? Like Goodna. No, that's that's a great way. Have you have you seen the movie office space? Yes, of course, so an office space. You know, there's these people who figure out this bug in their employer system where they can like pay themselves a sent a penny more every day and it compounds until they take up all the companies. Like that's exactly the type of attack that I think you can have in a smart contractor where a certain type of user can figure out an advantage that's not perfect, like it's a probablistic thing like fifty. It's fifty. One percent of the time they can take more of a blocker word than they're supposed to. Then their STAK allows them to, but they can, they can, you know, even though there's a bunch of variants and that strategy like one out of five times of fails, they can still keep executing it until they slowly take more than their fair share or more than kind of the rules. That like the intentions of the person who wrote the contract, where that revenue is supposed to get allocated in a certain way. But someone found a strategy in which they can slowly take more than they're supposed to, and that's the office space what I called office space attack, and that's the type of thing we stressed us because those are trading strategies, right. A good trading strategy finds that loophole slowly takes all the money and that's the type of thing you may not want in your decentrazed exchange. So I guess that's a that's a good differentiator from you guys, from like more traditional auditing companies. But are you familiar with the block science team and what they do? Yes, more or less. What would you say makes you guys unique in comparison to them? Yeah, for sure. I think you know, we come at these types of things from the more of like a infrastructure and finance view of this rather than kind of a more like high level game theoretical version of the world where you try to make this like really complicated model...

...of the system that doesn't interact with the real code and is but you try to like solve it in a more fancy way. I think we spend a lot of times time on building these things as trading strategies and making sure that you get the execution correct in trading. Is Eighty percent of your piano, like really understanding, for instance, the distribution of latencies, of packet arrival times. Is The difference between you losing a hundred million dollars and making a hundred million dollars? Yeah, and so we spend a lot of time on on making sure we execute the code exactly as it's run on nodes, making sure that the models themselves are aware of low level block chain details and, like, you know, things like to things like how fast they're getting packets, understanding that them pool, understanding gas bidding, and so we focus more on these like low level details than we do. I'm like, Hey, here's a differential game that may or may not have a stable Nash, and do you converge to it or not? You can argue that the the methodology that we choose can get stuck in local minimum more easily, but at the same time I would say that it is more practical because it can provide you more direct seedback and you know, it's completely modeled off how people do this in trading, where you know you have these really the stimulation environments that are exactly the same as like the live trading environment and you constantly running simulations as you're trading. Strategu is running live to say, like predict whether you should turn things off or whether you should like a size up or stuff like that. And it's not perfect. It's not theoretically perfect, like I can I can give you lots of examples where this type of methodology doesn't get you to the global optimum, but from an engineer standpoint it's much more practical, like you can know when, like a certain malformed transaction causes a catastrophic loss, versus, like I know that in this high level model that doesn't ever interact with the real blockchain, there's some potential bug that that could exist. So I think, you know, we just try to focus on being like really more in engineering side than like infrastructor side, versus kind of here is a differential game that you're trying to solve. So I don't think you said like the term that I usually associate with gonelet is agent simulation. Yeah, actually were yet. So maybe maybe, like because that's what's happening right now. You're scoring these smart contract systems. So tell us a bit about that please. Yeah, for sure. So an agent base simulation? Yeah, and think that that's a very good point. I think agent Basse simulation is basically taking different models of the different types of users. And you know, if we are, if we're going to take the simplest version of this, each user has two types of parameters. Number one is sort of a profit or utility function, which measures how much how they value certain things in the system like it made. There might be one type of user who is super risk averse and they value lending and compound much more than they value borrowing because they don't like taking leverage. They like their utility comes from earning one percent slowly. And then there're the Gen gambler type of people who all their utility comes from just like borrowing as much as possible and like betting it on Pomo d or something, I don't know, and so and gamblers. They're not even really rational, but they might stumble across something. Is that correct? Yeah, so, so that's a good point. So the idea is, you know, you model things into as one does this utility or profit, and...

...one is the decision function. and the decision function in classical econometrics, for instance, is the thing that says, Hey, pull the slot machine lever, or go spend it on poker or go invest it in a you know, a bond. It's a thing that represents the actions that you actually take. And in our case, and this is something again I think we differ quite a bit from block science, but there's our actions are the actual smart contract transactions and interacting with the real code, interacting with the blockchain itself, versus like interacting with the fake version that's like idealized, and you now have many of these agents interacting with the same blockchain or the same smart contract, and they each have different utility functions and decision fuckings. So one of them is like Hey, I'm going to I'm going to be happiest if I lend to compound. Another ones like, Hey, I'm going to be happy at if I borrow a lot and default in compound because I'll I have this shitcoin and I would rather have dollars so I can deposit a compound, borrow stable cooins against it and say goodbye shitcoin without having to do ky se. And you run many of these simulations of like these different types of users and and try to get some statistics of, Oh, does this type of user take most the profit? Does this type of user have a lot of variance in their profit? But a really key thing to this is that in distributed systems and Cyptography, I'd say the majority of papers spend a lot of time working on this model of people are either perfectly honest or their Byzantine. And this gets to your point of the what you're saying in terms of degen gamblers. Are the digen Gamblers Really Byzantine? And in reality they're probably not, because most of the assumption of Byzantine in distuted systems and and Bysante in general. Problem and stuff like that really assumes that Byzantine means any action that someone can take in a memory list way. They're not. There's no like there's no assumption that at each round of stay like a vote in the Byzantine system, someone doesn't like use the how effective their previous strategy was to influence their future strategy. If you read these proofs, people always prove these things assuming kind of this memorylessness. They don't actually assume that they're I mean, now people do care a little bit about these adaptive strategies, but the the traditional cannon interestruted systems, and I think this is one of the reasons why bitcoin and cryptocurrencies took so long to be invented, even though they're kind of simpler in some sense. Like it is because people like drilled it into their head that there's this model of purely honest and purely doesn't these the Gen gamblers. While they are Busyantine, they still learn from their previous mistakes. Right, they're still trying to optimize their wealth somehow and they need to have memory of their previous actions. And so rational agents. Unfortunately, it's a much bigger space than honest in business, you know, it's to a very particular theoretical sense. It's infinite dimensional space, but it it's a large set of strategies and you can you only have a finite amount of time to stress test, though, so you want to cover as many of those as possible while also being able to be computationally feasible. And that's that's that's the art, I'd say, of these agent base models is really knowing how to cover the state space and parametrize it effectively, even though you know it's this infant dimensional thing with sharp p complete problems everywhere. So how do you go about doing these simulations? What tools do you use? Have you guys written your own custom stuff for yeah, so we have a little we basically, you know, for to get and parody and like wrote a bunch of custom stuff for eath.

Most of the reasons for for really having a build your own or fork a lot of the stuff in the client is when you're running these closed simulation environments, you want to remove a lot of stuff that you don't necessarily need. Excuse me, for instance, like you actually don't want to verify signatures. Why? You're controlling the entire environment, and you are. You're defining all the users interacting with the system and you're going to spend most of your wall clock time in CPU running your signature generation. So why not just give people unique ideas and pretend it's trusted, because you're defining kind of all the users in the system. Another reason is that you may actually want to modify gas costs and see how under different gas models are under something like the one thousand five hundred and fifty nine, and like how how that changes behavior in in gas auctions and stuff like that, and so so a lot of the economics is actually embedded quite low in the in the clients. It's not the submergent type of thing, and so you want to you want to control that. Other optimizations the way you have io and communication between the agents. They might send side channel messages and stuff like that and you want to make sure you offer of that. And the last thing is you need to model the external market. So these agents are interacting with the blockchain, but they also are trading on coin days. They're also interacting potentially with other chains, but in our current environments not like that. So we've built our own environment. You can think of this like open ai gym or any of these like sort of reinforcement learning playgrounds. But Customs amized around how you interactive block chains and Sore tooling is based on that stack. You know I when I w because I guess I spent ten years in trading and the SASIC stuff. You know, I wrote everything see puls plus. Of course my cofounder was like let's rewrite everything rust. So most of our stuff now is like kind of rusting go. But we also have this sort of domain spehemic language that's in Python, because data scientists only one use python. I you're never going to find someone who's like I love to reliberate rust as much as that. You know. Yeah, Rust Fan Boise, are out there saying that, like you can do all your new Americs and rust. Find me a data scientist WHO's not going to e basic. And so so we have python bindings that we compile for both the C puls plus rust and go side. So they data scientists see something that looks like pyetorch and they can basically script the the the strategies that there and then from there, you know, it runs on like a cluster, just some normal like combrainity side of thing, and you can run, you know, thousands of jobs and and get the results. So the ideas, like you the simulations are a lot of it is centered around abstracting a certain things about the client away, optimizing the client and then making a DSL for interacting with that. Yeah, so I gather you have pretty much software which is similar to block science, Cat Cat d but, as you previously mentioned, the difference of what you guys are doing is you're testing it with the actual blockchain. You like not abstracting away this entire environment. Is this stuff open source? Can I play around with this if I want to? It is not. We we were working on open sourcing some of it. It's just okay, I think it's it's most of our I think we're at the current, current state. We're just we've just been like really doing focusing on more doing these sort of audit like getting seven really I'm time. Yeah, I guess a lot of the time when you end up writing tooling for yourself, the code is terrible and then you have to open source it and have to actually clean shit up. Yeah, exactly, I get that,...

...but but you guys are aiming at open sourcing some of these things? Yeah, yeah, definitely. Yeah, I think it's it's also like we're we've we do want to support another virtual machine at some point. So we're also trying to abstract some of the eight particular each specific stuff out. Yeah, but yeah, it's not quite at that. I think at that point that's like a good point of open source that we'll have to clean up a lot of stuff. Yeah, in order to just like what one really specific question is like does it you know, we need to. You're like looking at the application layer and smart contracts. There are you know, you take into account like front running in and even like a chain reord involving front running. It don't know if you saw, and it's been hard to explain an idea, but like this really cool dice to w in exactly happen a little while ago, like using the that your agents would have stumbled across that. Yeah, so that's a really good question. So we do model some aspects of consensus. So we separate but the usually the consensus model is not the exact code for a bunch of reasons. One is the wall clock time of running the real consensus code is forever because you're there's just tons of like timeouts and network delays and like the actual way that these clients implement like peer to peer networking, means that if you try to run the real consensus code, you'll end up spending almost the real wall clock time. And your goal and simulation is always to make simulate, the ratio of simulation time to wall clock time as high as possible because you want to get the most samples as possible. The more samples you have, the more statistical confidence you have in the more you trust these numbers and the the some of the consensus components. We we do simulate, I would say on Eth we don't usually similar consensus. I think for her protocols that have more of their incentives connected to consensus. We do so like we've done a lot of sumulations of cello and cello isn't eat is an east fork. But they have basically they've taken gath and they've stripped out the consensus pieces of death and put in a proof of stake consensus implemented as a smart contract. So it's similar to the move and we brow that way and in that situation we actually do we we do model kind of like the the forking behavior and also the the interaction of the application layer with staking. So I don't know if we would have exactly got the dice to and thing. We do model the men pool and then we also have models for how if I send like we you know, I think we have a model for like the networking of if if I have a hundred agents, what's the distribution of latencies that they have until they hit the first minor or like in the peer to peer gossip network? How many hops do they do they go through? So we have some high level models of that and those do capture a little bit of this forking behavior. But I don't think we can have these kind of gigantic reorg situations of art sumulation. But one reason I bring up the the incentive alignment is, you know, I think we wrote this paper last few late last year that was presented at the Stanford blockchain conference that shows that these applications can actually interfere quite a bit with consensus in proof...

...of stake in a way that's much harder than to kind of proof to work where if you have a a lending protocol, say like compound on on a state network. If the yields on compound are way higher than the staking yields, then people will be incentivized on bond and move all their stake, dassets into the lending vehicle. And the question is, is it possible for there to be more than fifty percent of stake in the lending vehicle so that fifty one percent attacks become cheaper than attacking the lending smart contract? And so the paper kind of shows that there are conditions under which that will happen. We're basically depending on how you structure the monetary policy. For certain monetary policies it's actually you're inevitably going to have this thing where, let's say it's a big point, montery policy gets super deflationary, eventually validate or say look, I'm getting no rewards, no new rewards, I'm just going to move all my money into lending, and then the network becomes easy to have. So we studied those very carefully because those are I think those are the cases that are extremely important and in this is I think, fundamentally, we're proof of staken pro work differ is a proof stick really is first a first class financial object, which means that it's competing with the things running on it. And so yeah, we if that answers your passion, we we do. We do care a lot about that when that happens, because that that that that can be of very catastrophic type of I think. So let's talk about everyone's favorite discussion again, a bit of Defi. What's your current take on it? We talked a bit briefly about it, but do you think the things which are happening right now are being done responsibly enough? Do you think most of the protocols that are out there should be doing agent based simulations with you guys? What's your thought on that? Yeah, so I think in defy it definitely matters more. As I said, proof of stake so far has been this weird exchange game of like how much steak can the single exchange get? And you know, it ossifies and sometimes, like once there's an exchange that has like thirty percent of a stake. They basically never vote on any of the governance things because all of them are adverse to them. They're like, Oh, this is you need to think more and like you don't get people don't change them very much. So I think the modeling doesn't matter if it's basically centralized. You know, as much of people want to say, Oh, defies not decentralized, it has way more users than any of these proof of stake networks and they're it's not even close. In some sense, I think the cool thing about defy is that you have a lot more transparency into when everything is blowing up, and so I'll give you an analogy. In the financial crisis, you know everyone's favorite villa and it's mortgage back securities. But mortgage back securities, one of the reasons that the blow up was even more catastrophic than it should have been is that there's not much transparency into when mortgage back securities are miss price. And the reason for this is, let's say you own a home, you're you're paying, you're paying your mortgage and all the sudden you're like use your job and you say, okay, I'm not going to make my interest payments, I'm going to make the principal payments because I can. You know the law less, you pay your interests later with extra fees, and then you still in a bad time and you're like, okay, I can't pay my principle payments and then you say, okay, I, I, you know, I default. Then the bank that underwrote that or the mortgage issuer has to say, okay, this, this mortgage defaulted, and then legally they have to tell all the people they fill the securities to, Hey, this mortgage defaulted, and then only then can...

...you be like, oh, the price of the security should reflect the fact that these defaults have that whole procedure takes six months and you have no clue when the default actually impacts the price and there's tons of principle agent problems that that are not you can't all participants can't observe. Defy is basically still making more each back scurities. Let's not let's not kid ourselves about the yeah, it's basically the same thing, but the price reflects the fact that the information propagates almost immediately and I think that is a big source of loss. That's an improvement. It may be a nuance improvement, but I think the the transparency actually does help you ensure that the price, the price, reflects the true risk faster and and so that to me is a fundamental innovation. It compared to normal finands. Like you do in normal finance, there's a lot of reasons that people can try to hide information from the market at that doesn't get price and I think that that that, to me, is like we're what keeps me really interested. This stuff is like it's just it's so different. I get that part, part about the whole transparency and how that makes a fundamental difference to these products, but I often feel like that is kind of an excuse by the protocol builders to build shittier products, because they then say, oh, it's transparent if people like if shit blows up, everyone can see. But it also assumes that the actors participating are able to interpret the information correctly, which I think there's like a big gap between the amount of people using defy and the people who actually understand the protocols enough to be able to make like the nuanced decision on whether something is priced in or not for sure. So that's where I think simulations super useful. Is You know you can you can give people who may not totally understand all of these details an idea of what the Roi looks like and how different strategies perform with the smart contract and under what conditions is profitable not profitable? Under what conditions you're holding a lot of risk and I think simulation is the way to bridge kind of the gap between I wrote the contract and I know what transactions it's admitting to Hey, I'm a I'm a user who's like, I want some yield and I'm willing to take some risk, but I'm not willing to take this muttress and here's here's my preferences, and Code my preferences into an agent and run the simulations and say like that's what, that's this is how operates. I definitely think that's that's the direction that thinks will go and that's the direction that normal finance already is. So when you, you know, you go to your K in the US, you'll see these like money Carlo outcomes of like Oh, if you choose this mutual funds, this is like the average performance, this is like what you might the best case worst case, and they'll give you all and people have already internalized all these simulations that hide a lot of detail of like how the arbitrage loops and etfs work or how the funding rate are works at the end of the day, every day, or like why their mutual fund. You know how the NAV rebalances itself daily, stuff like that, and I think crypto feels like it's going to go that direction. It just isn't. It's like the one thousand nine hundred and eighty great. Yeah, so do you think for like be fight to get more mature? It would make sense for protocols to not only have all these simulations done but open source them and provide like step by step processes for people using these products and how to interpret all the information that is taken into account? Yeah, definitely, I think. I think they're starting to do much more of that. I think like, for instance, balancers UX for Balancer for...

...a little back on, is sort of a Una swap your automated market maker, but it has a different curve and it it lets you have portfolios of assets. So instead of trading Eith for die you can hold a portfolio kind of like metf of Eith die maker some other Shitcoin, I don't know, I can't think of enough names in my head. But some portfolio. And they actually have a tiny little pnl simulator. Now, is it a very accurate like detail thing? That will make no, but it's the first time I've seen something that looks like these the things that you see in normal finance when you go to a k or two better ment or well front or any of these SYNTAC products that you people Investin, where it gives you some prediction, some kind of high level thing of like Oh, if you make this action, this is what outcomes you might have. And I think part of it is open sourcing, the simulation. The part of it is also U X. I actually think you have to find a way to present those things so that people don't roll their don't glaze their eyes over and say, like this is too much content. Yeah, but but you're starting to see that. In the end, I think it's just like you know, I I tend to really think Kyppt overminds me of like how markets looked like in the kind of s and s where we're basically people, especially especially when it if you especially in the sense of, you know, in the s there was there are a lot of people running exchanges. So so once the Internet existed, people just started building exchanges in their house. You know, it sounds crazy, but there were. There's there's a guy in West Virginia who had like thirty percent of equities volume in one thousand nine hundred and ninety five from some exchange he built sitting out of sitting in his basement. It's reminiscent of like two thousand and fifteen and Bitcoin, where like people were running exchanges out of their basement in some sense right there were. There's Quadriga was literally as a bad example in some ways, but Quadrigez was really just a guy who his basement right for a while. But as a Canadian, no one hurts. And basically what happened is over time, like the incumbent started like compete, that you have this thing where you have thousands of exchanges and then that fragmented liquidity so much that people started consolidating. And once they consolidated, the consolidation led to kind of these this oligopoly which looks like what the Coin Base Finance Bitnex by bit, you know whatever, like you know the twenty twenty people who have most volume. And then there's kind of eventually a reaction to that and you have all these syntech products that hide that from the end userer like Robin Hood, and user doesn't even know what exchange they're buying stock on anymore. And this huge distributed systems problem of there's thousands of exchanges people need to figure out. There's a law called Reagan Ms. This is all the US story, so I should copy out this. That says that if someone wants to buy a share of apple, you, as the broker, have to go to every exchange and say find the best price possible and sell it to them at the best price. If you don't, you were fined. But that SEC and this national best bid and offer guarantee basically destroyed all human brokers. It it it. It basically said, hey, there's hundreds of exchanges and if you incorrectly report the best price, you get fined. So all it did was create this kind of higher quincy trading industry to basically optimize...

...price discovery and sending prices everywhere. And that led to the Robins, because that basically killed all the banks from from market making and it made it such that buying data was more important than having relationships. And I cryptos kind of has a similar thing, exept. That's the buying data. It's like, you know, people don't trust their government and they'd rather, you know, go down this this rabbit hole. But I this consolidation and then kind of alternative and user interface type of thing, as what I see thee fine Defi, where it's like they're a bunch of exchanges, they consolidate their few exchanges and then there's this network of like market makers and market participants who are inventing a user interface such the end user has no clue, doesn't doesn't actually know what they're interacting and I see a lot of similarities in crypto. Anne. I think there's there's going to be. It's going to be so abstracted to the band user that there's no difference to them between buying a share of apple and buying bitcoin. Right Robin has trying to do it, but they're not really doing that with their crypto product. makes a sense. So I guess you mean one of the things. I think he was in deans first question. They defy is I think he sees is kind of a shit show, while best do you see it professionalizing and like sort of becoming more stable, reliable, listen the money? I think. I think maybe it's because I was burned in the early days of Dick Colin in two thousand and eleven and thirteen and interacted with really shitty exchanges in which I lost. I've lost at least well and yet at that time. So this is just embarrassing, but like I lost at least like ten to twenty big Cooin at that time too, to basically exchanges that just got shut down or like disappeared and stole my money and exit. SCAMP. I. I think that I just count right. You just it's gotta happen everybody at some point when they're getting anyway games. It's your I. Yeah, yeah, well, it was just, yeah, I was just painful. But I think it's like a little better than that era at least. Like yes, it's harder to reason about, but man, the scam artist in two thousand and eleven and thirteen, like to thirteen, we're just like unreal, like everyone was brazenly just trying to take take your your coins and like doing everything possible to to take your keys away from you. I think while defensive should show, there are clearly things that seem to be an improvement, and I think your you'll see consolidation around those. The question that I don't think. I think in the same way that there was like hundreds of people saying, like I'm running a bitcoin exchange, to now there's like twenty you're going to see the same thing where there's going to be thousands of there's going to be a two thousand and seventeen of defy stuff. I don't know what shape that comes in, whether it means there's a lot of capital that goes in or whether it means they're just like millions of projects, most of which are useless. But there's going to be something like that and then whatever sticks from that is going to become the point based fiance, and I think it's going to end up just consolidating, because people don't want to trust their capital into many places. And you know, I think that's why compounded maker kind of have these liquidity Moses that people kind of trust their contracts, for better or words, even in spite of black Thursday, people just trust them a lot more than then the next second best option. It's really a like lock in effect, and to some extent I think that that that's a little bit of what happened is happening for Athereum.

I remember I got into this space being light or like full time working in this being like Oh, there's like all these famous professors and there's all these like really smart people working on building these competitors and or whatever, but it's pretty clear, I think, to me that none of the other lawyer ones really have serious chance. I think they're all going to end up being layer two is four dreum. That's almost level. That's a whole huge other topic. I feel like keep it on drag. Well, lad so you say, yeah, you say that there's going to be this another two thousand and seventeen, but just for defy to meet. It feels like we're already there, because in two thousand and seventeen what we saw as a bunch of people selling ICO's saying that they're changing the world, whereas now we have defy where everyone's like all, we're giving all these financial products to people who need them, whereas to me it looks like essentially just a bunch of white dudes who are traders getting richer off of instruments that they're pumping themselves. Yeah, I think the the difference was there with no, you didn't have to, you didn't have to prove anything. In two thousand and seventeen, like the brock pierces of the world were able to sell stuff without by selling their lifestyle and not like selling something that a piece of code that was able to manage capital. Hey, dude, he's a greedy ducks, he's legit. I do distinctly remember this party out of that in New York in two thousand and fifteen or sixteen, and he was like selling es to a bunch of people who were on psychedelics and it was like color. It was like, I don't see that, that sounds like Brooke Pierce. It was definitely not the is definitely not the type of thing I think that like the defied Bros will be able to do. They don't have the charisma of the mighty ducks. They don't have the marketers of that two thousand and seventeen there. It's like to annoying for people to understand. I do agree with you that it's kind of people getting richer. There's no there's no things, so getting around that. Yeah, I always laughed whenever people are like we're solving finance, like people are going to take a defiling to to buy a house, and I'm like no other not. This is just traders trading. The people who take defy loans to buy their house already have enough crypto that is worth more than their house to borrow against. Yes, said they should have just sold their crypto. Actually, yeah, I guess maybe it's definitely. It's definitely. It's definitely not like this. Thank the undanked. Yeah, a lot of the things just aren't really capital efficient either. If that's the right turn to use, for sure, I think that that's improving, though. That's the thing where I'm a little more hopeful that we will do better than the normal finance industry, because the lack of capital efficiency is actually the quite quite a bit of the source of inequality. Yeah, in that if you if it takes more capital to extract a certain profit, but but like only ten percent of people have enough capital to do it, then those tens of people will be taking that block, or word, so to speak, from that that instrument, and no one else can can join. I like to say the arbitrage is the proof of work, of defy right, like you can only kind of like exact a defy profit if you can like actually prove to the protocol that you did arbitrage. And Yeah, and I think that's a fundamental difference than I seeos, because I see us arbitrage was do you trust hooded man giving you psychedelics? That tells you to buy put a million dollars into ears.

You know I mean, it's like it's a little bit harder now. Will there be stands? Of course, but you know, it's it's it's someone from I think the hardest part for defies. There's no way to do credit right. Yeah, no identity. There's no. There's no identity and there's no. There's no. There's no way to do unsecured or under secured, at least blending right. And until until someone solve identity, which is a much. That is the banking, the unbank that crypto needs more than more than like being able to give people learns is like finding a good, decent, frized identity notion that still preserves privacy. Somebody will just put a token eye, though. They'll just be like here's your here's your social credit to gains, here's your ID coinlateral. Yeah, well, I mean, I guess this is why I added the privacy codia. Like somehow you have to find a way to give, you know, an anonymous identification that can generate zero knowledge proofs of certain qualities of the owner without having to leak things over time. The problem is, just like you can't really privacy coins have to be built assuming that the transaction graph can be identified. The any digital identification system has to make sure it it doesn't leak information based on the number of queries that it's made, because, like, if I make a bunch of queries and someone can identify that transaction graph, that I'm screwed. But transaction graph mixing is probably the most difficult technical problem I think that exists in Crypto. Made me that you guys might disagree on. I don't know how you feel about that, but I think that's the biggest hurdle to ever doing these decentrized ID that is also private. Do you think that you can have the you talked about the benefits of defy, whether things being transparency, so being publicly visible that you know things are being repriced, for example. Can you get some kind of like that sort of public auditability in some way in you know, balance with some form of privacy confidential transactions, or is introducing disintroducing that through something like an Aztec or like? I know that there's actually like more of a spectrum of privacy and confidentiality with various technologies, but these there's some mix of that that also allows you to have the right level body ability and public visibility the activities. Yeah, I think that, again, all of these things boil down to somehow forcing the transaction graph. Like you can do all those zero knowledge grow stuff and make a single transaction and office, like I'm going to say, perfect. You. You've you've figured out how to optimize fri you figured out how to, you know, compress your circuit really fast. You figured out how to do all the optimizations and bells and whistles to make yourselage proof work. The problem is that's just a single transaction and mixing the transaction graph in such a way that you can still identify that something is true without leaking information over time. I just don't know exactly how you solved that. Like there's there's there's some fundamental limit in nature to how there's tradeoff between like number of interactions I make with a system versus how private each of those interactions has and like the more interactions I make, even if they're private, each time I have to be revealing some information into the system, and I think the only way to really do that is to somehow like throwaway ideas a lot and do this mixing type of thing on the transaction graph itself. And but the problem with that is like the competational complexity that grows way too quickly. But I do think that there is, there is...

...some way of, there must be some way of doing a zero knowledge proof on the transaction graph that itself the whole transaction graph for all people. That gives you a that like can do that. I just think that we're very far technically from that universe where that happens and people are I feel like people are really working towards that. It's just like that feels to me like a twenty year goal of like true transaction graph blinding. I think the other thing is in the privacy coin space there's just like a lot of claims that people make that are, I would say, under substantiated. Right, like we've seen so many of these empirical attacks on the cash and then era that it is. It is very clear that like it's really good to get single step transaction privacy working. But Yeah, you somehow have to deal with this this temporal aspect if you want to get all of the things you were you were saying to work. And Yeah, I just think there is some fundamental lower bound of like you can't have more than a certain number of interactions before you start leaking some information. There's some information theoretic bound right, like that says, and once we understand that, then we can design systems that try to like keep yourself saturated as like, so that's as private as possible. But I'd really right to like me perhaps guide you, like your your personal assistant ever say hey, it's time to use some new addresses or something like rotate your your addresses. Yeah, exactly, like there needs to be some type of thing that, like, you can yourself monitor that you have leaked x, like some bits of information and you say, like, I'm comfortable leaking this many bits and if I'm not, then I have to go get you right at ID like. I think somehow that has to be built into the UX and how people use these systems if you want them to be truly private. Yeah, that's a that's a really fascinating question. Be Interesting to see where things go with that. So I think I think we're at time and this is where we search to wrap up and say thank you very much and maybe tell people where they can follow you and learn more about you and what's your work as you bols. Cool. Thanks. Yeah, my twitter is is my name, Turin Chitra. And Yeah, you can rewrite a lot of papers and blog posts at Dotlook doct work. And Yeah, just send me messages whenever. I'm happy to good talk any of this stuff. But yeah, thanks, thanks for having me. Is Very Great.

In-Stream Audio Search

NEW

Search across all episodes within this podcast

Episodes (127)