Ending the Modular vs Monolithic Debate | Kyle Samani

Transcript

Although our guests this week are managing partners of a registered investment advisor nothing in this podcast should be considered an offer of multi-coins investment advisory services or should otherwise be confused for investment tax legal or other Financial advice [Music] what's up everyone welcome to another episode of Lightspeed today we have Kyle samani from multi-coin Kyle's one of the most independent and clearest thinkers in crypto Kyle welcome to the show Garrett mert thanks for having me on excited to be uh early early guest of the new LightSpeed show yeah we're uh we're lucky to have you I want to start with one of the primary debates in crypto right now which is modular versus monolithic I know you kind of have a spicy take on this um what are your thoughts on the topic

and why do you say monolithic is a misnomer the correct term is integrated um if you have any kind of substantial understanding of the history of Technology like this debate has been had many many times um starting as far back as like the IBM Mainframe and like the reason that IBM was like here Microsoft like go forward make our dumb operating system is because like they thought the value of the system wasn't being integrated um and you've seen this pendulum swing back and forth the different layers of the stack uh obviously in terms of PC components you had windtail kind of rise in the 90s and then obviously iPhone kind of led to reintegration in terms of mobile context in software you've seen something similar um AWS started as a fairly um modular platform and it's slowly been

integrating up the stack with with pass and with all of their AI tools and whatever and now they have a [ __ ] competitor and whatever call center tool you know like so you you see this playing out in in Silicon you see this playing out in in Hardware configurations you see this playing out in OEM she was playing out on operating systems and and everything else kind of moving up the stack um the important thing to take away is not that there is a universal answer or a universal framework on how to think about which pieces of the stack should be integrated versus modular um the the there isn't an answer because these things vary as a function of uh like different layers and how those layers interact with each other and then they also vary over time

um the most important example here being like apple and integrating all the pieces in the iPhone together and like swinging the pendulum away from uh modular components and towards integrated components especially with systems on a chip and the full integration they're in so um there isn't an answer to the question of modular versus integrated um it's just important to recognize that like this debate's been had many times and actually typically both configurations coexist uh Microsoft did not kill Apple um the existence of iPhone does not preclude Samsung from having a very large ship business the existence of uh whatever Celestial ethereum doesn't preclude Solana from existing Etc um these things actually you typically

see kind of both configurations coexist it's pretty rare that like one kills the other you have rare exceptions like CIS versus risk but like basically speaking and that's not really integration versus modularity that's just more of a question of complexity and Chip design but like uh these debates have been had um and uh so that's kind of common a is like there isn't a universal framework comment B and this is probably actually the more important one is um like you need to understand like what is the objective you are trying to achieve um uh in my opinion like the customer of a blockchain is the developers and

developers want to build as good of applications as possible while doing as little work and thinking as much as possible both of those statements are actually those variables are gray and have are multifaceted in terms of the inputs that that comprise them so you can't make kind of universal statements but like one thing you you can say pretty definitively is all of these modular things op stack ZK stack supernets you know what whatever pick your flavor of choice um they're all increasing the complexity that developers have to think about um across many vectors uh modular people kind of assumed that that's isolated to

like this Vector of the security assumption between the bridge of the L2 and the L1 and that's like very myopic view of that but but it kind of fails to capture the more the broader social reality which is like okay well if your users have assets on another chain are they getting over what if you need to bridge out uh are there on-ramps for your new chain or they're not uh are exchanges onboarding directly there like there's a limited amount of UI on the withdraw function of coinbase um like there's all of these weird things that emerge address form I mean apparently I learned this recently on even on different evml2s the address formats change across these things so like the token contracts or whatever like Ave is not the same on optimism it is an arbitrum and like that's a new

form of social complexity that every developer and every wallet has to you know like you got to deal with all this [ __ ] um and and so you just keep increasing this this complexity um in dealing with all of these problems uh and uh I think that that complexity is is being pretty dramatically under appreciated by most most developers in that in that world so you put it pretty well there and you painted a very balanced picture I would say the sentiment in CT is very in CT equipped with Twitter it's very modular dominated people seem to be like 80 modular bullish let's say and they

seem monolithic as a bad word which integrated thing it should yeah it shouldn't be it shouldn't even be model I think it should be integrated why is that why why is the popular discourse so dominated by this modularity talk and how can we what are your thoughts on how to fix that or maybe make it more balanced I mean just look at the market cap of eth versus Soul like that's that kind of the simple answer uh uh I I hate being I guess almost like tribal about it but I think that's basically what it is um most people are just going along with whatever their bags say which look you can say the same thing about me so maybe I'm the one who's not thinking this is the thing that crypto has that does not exist in other technology markets which is like you have public bag holders in

very large quantities um even when like the basis for which the kind of Core Business winnings uh the the traditional vectors on top of which businesses would win or lose uh opportunities uh have not yet been been defined typically people become public backholders once the major questions have been answered um and in the case of crypto that's just not not the case uh and so back holding effects are very real um in terms of how to change the discourse yeah I don't know I mean Justin Drake popularized the term monolith which which is pretty frustrating to me

um I mean he's very smart guy he generally does have a good sense of History um and you know I I was I was a little annoyed when he chose to use the pejorative term instead of the historically correct framing um really the answer is to have a conversation with him and like okay like let's let's really dig into this and like okay you coined the term monolith and versus modular so like okay let's unpack it and let's have fun with it um I've had to be fun I don't get the sense you would opt into it but uh would be pretty fun to do so um while we're on that topic actually so the the tribalism that you kind of you know the bag holder syndrome must call

it to to label it it clearly plays a pretty big role here uh but it could also be used to bootstrap communities right uh token incentivized communities which which is something you're very big on so in in that regard it's like a double-edged sword like what do you think it is is it a future or a bug how can we best leverage it uh it's definitely a superpower it's a superpower you need to know how to wield as a developer um most developers don't know how to build the superpower well um folks actually like the ethereum foundation I think they don't weld it they don't wield it particularly well or poorly they're they're kind of just treat it neutrally um uh

generally I think the term Community is um used as an excuse for lack of thinking and also used as a euphemism for or excuse for product Market fit or lack thereof depending on the context in which are presented I generally don't think Community matters as a construct um uh Financial incentives matter um and like you can be very clear about how to think about financial incentives I think deepen being like a very good case study of where you can reason for first principles why those financial incentives should or should not work um it gets a lot more nebulous with l1s and other types of products though um so sticking on the modularity and the

financial incentives you see this with Roll-Ups today with l2's arbitrary optimism and doing airdrops and that's how they're getting users and high tdls I know you've had a lot of comments on roll ups in the past what what is your bear case on Roll-Ups or or has it changed um I mean I'm not opposed to the concept of them they I mean look they can exist kind of therefore they will exist like these are permissionless systems uh I think the right question is like what is the right configuration of them uh and I think it's kind of like I say the ethereum community is like Hammer everything is there for a nail um is like a pretty high conviction that statement that that's correct

um and it turns out maybe like five percent of the things that people think are nails are actually Nails is like probably correct maybe it's even under five percent uh so they'll be out there uh I mean look like at the limit if you really push the system like Solana or like fire dancer um do you what really is the bottleneck at scale is it data or is it compute uh I've had different I've gone back and forth with this on different people and depending on which benchmarks you use you get the different answers my conclusion is I don't know um I actually I don't particularly care um because the correct answer is uh Moore's Law and Nielsen's law uh and don't think so hard

um uh but assuming the answer is compute then you could get gains in total system efficiency by doing except by separating execution from uh from D.A um so like we'll see if you know that that matters for scaling total system throughput I'm I'm pretty skeptical that that that Delta matters but potentially is a thing uh and then maybe there are some developers like maybe people come up with cool use cases that are still unclear um you know the most interesting use case for being in your own chain is uh a Doritos Dex

um people the idea of app chains is not new obviously Cosmos and polka dot were the kind of two first people to push it hard and I've spoken you know kind of every 12 months or so for the last six years I've checked in with the cosmos and the polka dot people and been like hey like what are the cool app chains like what what are you you're telling me it's customizable so like what are people customizing right uh and the only answer I've I've ever received that I think is even plausibly good is derivatives Dex uh and and that's because Dex has a very specific set of parameters or properties rather that are unique whereas obviously they are money I mean that's where all the money is being made and lost

obviously you have hft and latency and all that other [ __ ] um and then specifically in the context of derivatives it's particularly sensitive um you know if people pay 150 bips on coinbase um it's hard to argue that retail cares about spreads uh and like you can make a like a DEX work with 50 basis point spreads it's just like it's just not a problem derivatives are a little different though because the people who trade derivatives are a lot more sensitive to those things um they're also sensitive to the Oracle into the collateral management like it's just a you know like 100x higher bar in terms of like median amount of caring about quality of execution um in a derivative is X versus a spot

Dex and so there is a case to be made that by doing latency optimizations primarily by centralizing the sequencer um or pseudo centralizing the sequencer you can you know deliver a Oracle plus collateral management plus latency set of optimizations that are impossible to do in a permission list uh open system uh and therefore that is the right way to unlock you know derivatives decks um that is like a fundamentally viable Theory uh it has not yet been proven or disproven in either direction uh I am of the belief that you can do it on a permissionless system I think the without question you can get optimizations on a system like dydx or say that are not plausible on Solana no

questions asked you can get to better performance I am not particularly convinced that the marginal basically performance gains you get and spread reductions you get uh are like what are going to matter but there's a chance I'm wrong there beyond that I haven't seen any apps that like I've been pitched and I'm like actually derived explained from first principles like why you need to be in a new place um I I think a large part of the challenge is this course is that like the point of blockchains is shared State uh and like shared States should be kind of interpreted as broadly as possible shared between applications shared between users

um and this it kind of derives from like the core Atomic unit of what do you do on a blockchain which is like I have five coins and I send you five coins and so my balance goes down by five and your balance goes up by five and like the fact that that Delta of five is enforced uh is the point of the chain uh whether the denomination is an nft of n of one or whether the domination is n of a billion doesn't really matter like that that's the core thing that these systems are enforcing um and and so the idea of app chains always struck me as wrong I was like not that there aren't apps and not that those apps can't be fairly isolated in terms of the types of States changes that they are doing but it's like the point of these systems is that they're

about shared States and as long as you have parallelism per unit of State then like what are you getting by going to a different a different you know asset Ledger like I just I don't understand what you're gaining um derivatives Dex being the very notable interesting uh thing that is is TBD and look maybe there are other things that are going to emerge that I'm I'm you know uh not creative enough or just haven't seen yet because it's too early but uh I I it's not obvious what those classes of apps are like that's for sure it's definitely not obvious oh I want to tie two of your ideas together which are really about composability I tweeted like December 2021 I think that 2022 was going to be the year of

composability and then you um later tweeted talking about sequencers and how ethereum the consensus has kind of shifted from we're going to decentralize our sequencers so we're going to have shared sequencers and with doing that they're saying okay we're actually looking for Atomic composability instead of censorship resistance um I'm curious do you think one how important do you think Atomic composability is um sounds like uh you're pretty bullish on that you think it's a big deal and then two do you think that's the right trade-off that uh L2 should be taking right now oh I think Atomic composability is not that interesting um even though my name is composability Kyle on Twitter and has been for two and a half years uh I I think Atomic composability is uh topic was really is useful for uh

actually by definition the only class of user is useful for are Arbitrage Bots um and they kind of Define that broadly whether it's Dex Arbitrage or some flash loan whatever whatever uh but that's not who you should design the system around um the system should be designed for retail maximizing developer ease and maximizing retail access and minimizing retail gas fees and latency uh Atomic composability is intellectually cool though uh when I I talk about composibility I think of it really more broadly in the context of like hey uh one of the problems in all of these

crypto systems is that like they are mostly heterogeneous above the asset ledger so like okay you've got Solana or you've got ethyl one and like that is a logically consistent Global homogeneous construct and like everyone understands what the rules of those systems are and what the code is and right like it's the magical asset letter in the sky and it updates uh but then everything else above that is heterogeneous um the exchanges the wallets the indexing and query providers like the Fiat on-ramps the currencies that you trade to get in the geographic rules and regulations like everything above Bell one is heterogeneous um and making all of those things work

together uh across any arbitrary configuration of software end users uh is pretty comp pretty complicated it turns out that if you everyone's using the same L1 there's just a lot of those problems like get solved kind of naturally-ish um and basically the L1 becomes a standard to which all of those other people can form themselves uh and when you start breaking up into lots of more pieces you have to get all of those players to reconform themselves around these different configurations of things and the more configurations of things there are the more total system complexity there is um and and so when I talk about composability it's it's not so much about Atomic transactions and Flash loans it's about just like as a

developer knowing you can like build a thing throw it on the standard which you know is ethyl one or Solana for the purposes of this discussion and just know that everything all the wallets will read and write from it correctly all the exchanges will interface with it correctly uh whatever else like you just know is conforming to that standard um and so you get the maximum iteration of of uh compounding in terms of developers and users instead of thinking about bridges and liquidity and on-ramps and like all this other all this other stuff so two questions there one is you you said that in your view at least the point of these systems essentially

single globally shared state I think probably I agree with that but I'm not sure if everybody would so I'm curious to see like for people who just aren't familiar with this concept why is that the point of these systems I mean uh there are two kinds of of software in the world uh abundance software and scarcity software um the vast majority of software you use in the world is abundant software uh like there's an infinite number of video streams there's an infinite number of tweets an infinite number of telegram messages and Gmail messages and whatever you can always create more content and just keep adding more and more and more and more and more and I think the point of those systems is they just everything

grows towards Infinity uh and then there is scarcity Centric software which is basically just the realm of money and finance um which is like I have five dollars and I hand you five dollars and so now my balance of dollars is down by five and your balance is up by five and something somewhere in the world needs to encount the fact that I am down by five and you are up by five uh and obviously this applies for money for ass just asset I mean this just applies for assets um whether they're Commodities bonds equities whatever doesn't matter uh and and like block like blockchains are the kind of generic term I use for them across l1s

and l2s is asset ledgers because they're just ledgers keeping track of who owns which coins like that's really all they're doing whether those coins are nfts or whether they're fungible it doesn't matter that's that's what they do um and so like if you kind of take that to his logical extreme the kind of obvious endpoint would be like well wouldn't it be nice if in the world there was just one asset Ledger and like kept track of every you know like what everyone owned like that's kind of The Logical endpoint um now there are political reasons why that won't be the case and then uh perhaps technical reasons um as well but like that's kind of clearly like the the cleanest construction of the world is just one asset Ledger that enforces those rules

I I think you can get a lot more scale out of a single asset Ledger Than People appreciate uh while still enforcing these social properties of blockchains in terms of Nakamoto coefficients as well as number of replicas and validations um I don't believe you need to validate these systems on a 200 computer at home that's a very different thing to optimize for which in my opinion is Superfluous uh but like it's just I mean what's nice about physics in the real world is like we have a common set of rules in physics which is like I have a green piece of paper and I hand you the piece of paper and physics enforces that that green piece of paper has not replicated into two pieces of paper right and like that set of rules governs everything in meat space

um wouldn't it be nice if we could have the digital equivalent one set of rules or in this case one asset Ledger that enforces all Notions of asset ownership around the world I just wanted to give a quick shout out to permissionless this is the biggest and best conference in all of D5 it's the one that we do with bankless who's a great partner for us last year we had almost 7 000 people there in West Palm Beach we are moving this year to Austin Texas so I'm September 11th through the 13th and if you're a listener of bell curve any of these last five seasons this conference is basically custom made for you we're going to be talking about liquid staking the theme of this season we've got a bunch of great panels on Mev if you listen to the app chain thesis we've got

a bunch of Cosmos folks out there in full force we're talking about the converging architecture of Solana the roll-up space in eth and Cosmos so I would love to see all of you there and to reward you for being such great listeners to bell curve you get a special 30 off code it's bell curve 30 that'll get you 30 off tickets click the link in the show notes and then head over to the permissionless site and make sure that you get your ticket today again that is bell curve 30 click the link in the show notes what is Solana's mode against other highly performing blockchains like sweet modad other l2s that I just said you know that are experimenting with the llvms like what is emote in crypto when it comes to these l1s because at some point you would think that the tech might become a

commodity uh the state is the moat uh well sorry two things it's the state and it's the Integrations to the state um obviously if the code's open source that's not a moat uh so what is the state uh well this data is who has how many coins and so the more people who have more coins like that estate um and then there's not all forms of state are fungible for the purpose of the question you're asking for example an LP position in uh in a a m pool is different than just like passively owning coins because that LP pool position is providing liquidity to other people to do stuff um but but like the the broad answer to your question is State uh and then the

second part of that which is important is how many places are reading and writing from that state um and so that kind of means like on-ramps applications that are you know like uh Hive mapper or whatever helium like those things are theoretical demand drivers to create a lot more State um on on Solana uh I think most people in crypto uh over value the total amount of State on all of these systems today uh by several orders of magnitude um and it's very hard for them to comprehend that like there is a world in which a

asset Ledger is like actually running at 50 000 TPS of demand and like all of that stuff is happening around the world uh and that is like if you compare that to what exists in crypto today like that is actually a thousand times like a hundred to a thousand times larger than what we have now and it's very hard for people to grasp uh what that state of the world looks like because I think that state of the world has very little to do with the current state of the world other than maybe some weird path dependency I mean there's some path dependency obviously but like that state of the world is very far away from the current one and so I wouldn't

overly uh project the status quo to that future State yeah and when you talk about State how big of a deal do you think real world assets are like you see Solana making a push with um The Saga phone like that's almost to me it looks like Amazon who started off they're just e-commerce but they're almost building modes by having physical infrastructure that's actually hard to compete against like do you think I guess one I'm curious what you think about the Solana like mobile push and how important that is but two just real world assets in general because to me that that might seem like a another part of state that would be a big moat um uh uh I'm I'm very excited about Solana mobile um I I own a saga phone nice uh here

um uh I don't think Saga will sell in very large numbers and I never expected it to um what's interesting about Saga is SMS or the Solana mobile stack um which is free at open source and available to any Android OEM uh and the Hope always has been from day one is get other phone manufacturers to implement SMS um if it doesn't cost them anything then the bar for implementation is not that high uh why might those people Implement SMS uh that's the question that that needs to be answered but like it like roughly speaking uh should be those phone manufacturers I think they can

generate some Revenue so if you have some sort of App Store rev share uh from the from the DAP store back to the oems um that's a revenue driver for them and so the question from there is well are there apps from those app stores that are generating any Revenue that can be shared back to the oems uh so far the answer to that question is there are clearly who plays dexes and nft exchanges that obviously have trading fees that you could imagine flow back none of that's happening today but I don't think it's very hard to get to that world I think if you went to Magic Eden and tensor and said will you take X percent of your trading fees and send them back to Samsung or whoever if you know that they're sending you that that traffic I think they would all say yes

in a heartbeat so I I'm not too worried about the meat space negotiations there um so that the bet is you know you can figure out a way to help these guys drive revenue from from the apps Apple and Google have choke holds on their respective app stores and the Android oems hate that and have tried many times to get around that and have you know alertsly failed and so if you can have a new class of apps that by definition will not exist and work in the old model and they can give some sort of rupture back that's pretty interesting um I'd say I'm fairly optimistic that you know within a few years you'll see some major Android oems uh adopt SMS uh it doesn't cost them much to do it and

and if it's basically a close to free call option um and so that that's kind of the big opportunity there is distribution via oems so while we're on the topic of real world assets or or the real world or as you call it meet space um I want to talk about deepen decentralized physical infrastructure Networks you're obviously one of the leading thinkers on this entire vertical I would say so actually before we get into that I think a lot of people when I talk about this on Twitter don't actually even know what it is so can you just briefly explain what dpin is in your own words and why it's important yeah stands for decentralized physical infrastructure networks the the term was

coined by by the folks over at massari uh they're like broadly speaking two classes of dpin networks there are ones that are uh GPS dependent uh and then there are like server networks which are kind of GPS dependent but not really um so the simple way to contrast that would be like filecoin versus Hive mapper I'm assuming if you're listening to this you know what filecoin is it's Airbnb for your hard drive in very simple terms uh and like you know latency to get around the planet Earth uh is roughly 100 milliseconds 130 milliseconds to do one lap around the earth at the diameter um so for like most applications like you know whether you're in New York or

Chicago it doesn't matter um there are some class of application in which that does but for the most part it doesn't uh and uh because human perception just can't perceive five millisecond kind of increments again for the most part um when you talk about uh High mapper let's contrast let's say file coin with Hive mapper uh like Google has several thousand cars today that drive around the world and capture pictures of the roads um and they use that to update their maps not only the shape of the roads but more importantly where all of the speed the stop signs are and the red lights and the speed limit signs and the yield signs and the lane markings and all that stuff um and it's having the shape of the road

alone is not really enough to have a useful app you kind of need to have all these other things to have useful map um wouldn't it make sense if everyone driving around the roads was just capturing that information and sharing it back and creating a global database of a map uh and so that's what Hive mapper is in high map where you buy a dash cam you put it literally under your review mirror in your car you drive around it captures you know images and stuff uh sends those up to their servers and then they process it and they do AI to detect the speed limit and the yield signs and Lane markings all that stuff uh and then uh create a global map that's uh you know Bottoms Up from from the people there's like roughly 15 000 people who've um set up a dash cam in

their car now they've mapped roughly four million Road kilometers uh they're about 60 million Road kilometers in the world so call it seven eight percent seven percent or so of Road kilometers have been macked so far um in about eight months right um so a pretty impressive overall Google for context is estimated to have about 20 million kilometers uh so they're 20 you know one-fifth the size of Google um on obviously one one millionth the budget uh um so like what does depend sorry about the original question uh there's two classes there's the server tile class of app which is the file coin render kind of a thing uh and then there is the

helium Hive map kind of thing which is very GPS dependent obviously having a picture of Madison Avenue New York is not equivalent to having a picture of uh Congress in in Austin um so you need to you know get those things differently um we spend a lot of multi-coin is very likely the largest investor in both categories of apps as a function of dollars um and perhaps also as a function of number of names as well um so yeah we've been doing it for a while Kyle can you touch on um why you think it's important that maybe these behaviors that these projects are locking into are passive versus active and that's talking about for example drivers that are already driving

yeah um the only way to make these D5 these d-pin systems work for for the uh is like you you need to have some sort of structural cost Advantage versus Google or versus uh in the case of let's say AWS Amazon um or in the case of helium versus Verizon and a t and T-Mobile and like by far the easiest source of structural cost Advantage is if a consumer was going to buy that hardware for some sort of personal consumption use case and then they can repurpose that Hardware to make it an income generating asset um the opportunity cost for those people like they were those consumers were otherwise going to assume zero dollars of Revenue generation from whatever the

the device was or service was uh or time investment was and now that number can be larger than zero if Amazon is buying a GPU like they are underwriting that saying we're buying this GPU for whatever a thousand dollars we expect to make whatever ten dollars an hour I'm just making these numbers up right there for after you know five months we'll break even um uh consumers are not running that math and are not that cost of capital sensitive so there's a real structural source of cost advantage in order to take advantage of that you basically need to be able to have consumers passively offer or uh whatever the thing is so in the case of Hive mapper it's like someone's paying people to sit in a Google car and drive that Google car

around um but if you're driving from home to work like obviously that's that's free uh in the case of helium it's you're paying for your your internet line at your house um and so if you want to be able to resell that bandwidth um you can put up a hotspot in your window and then resell that bandwidth by the way I actually um just coincidentally I actually have a high mapper camera right here with me so just in case people are interested it's just I have one right here love it sweet device um and the cool thing is like as a developer you can actually plug into the apis and kind of you know build whatever you want with the maps and and contribute to the ecosystem as well um I do want to so I asked this question to you I believe on Twitter before but I

want to ask you here's your story but he's aware and at first I was a little skeptical because I figured okay why do you actually need blockchains or crypto for this right can't you just pay the consumers or whatever the passive uh agents of the system let's say with with non-crypto means and I think you had an answer to this I believe it might um involve like the cold start problem so can you touch on why crypto uniquely enables this yeah um actually I wrote a blog post about this in December of last year uh it's on the multiplayer blog post I think it's called crypto payment flows or something like that if you also want to throw that in the show notes um

but uh to kind of recap that here um one you generally need micro payments for these systems to work um when you're talking about people permissionlessly putting up their stuff when they're spending their own electricity um for these things uh they need to be comped in not real time but like something approximating in real time uh in the most generous insert that's kind of 24 hours and other answers it's like every hour or something um and there's kind of different technical imaginations to make that work in a trust minimized way but you need to have lots of micropayments and obviously if you expect the supply side of these networks to be coming from hundreds of thousands or millions of people all over the world

in very small denominations across many currencies and countries and geographies and banking systems than like the only answer is blockchain um so that that's quite important and we we see this already today with both helium And Hive mapper that they're the probably two most at scale um people who will tell you like from first principles and experience why that that's Paramount for their respective use cases um and the other one and this one's actually less obvious but but arguably more important um is about incentives uh um if you are an Uber driver uh like people talk about decentralized Uber and like what if the drivers had a stock and Uber for driving

um there's that kind of classic commentary is not wrong but it misses a more important subtle thing that that is very distinct in helium And Hive mapper um so in Uber if like you're a driver and like let's say you're the only driver just for Simplicity and like our personalities the app and opens the app and like calls you you know you could pick them up like well you you were getting paid dollars for performing that job right um uh and like and therefore the the demand side of the network meaning the rider is happy because they got the ride and the driver is happy because they got paid for providing the ride right and like that's pretty straightforward um but other classes of services require a

minimum level of scale in order to function so um let's take helium or Hive mapper these are perfect examples um if you have a map of 500 Road kilometers on hive mapper and let's assume you have like the most perfect map conceivable uh whatever that means just imagine it's absolutely perfect um what is that worth and the answer is is not uh Google Maps 5 500 divided by 20 20 million which is what Google Maps has or 560 million which is the number of roads in the world um uh which would be kind of the like very crude way to try and approximate the values so to speak um the value of those 500 Road

kilometers is zero uh and it's zero because there is no customer that cares about 500 kilometers it's simply too small of a number for anyone to bother to get out of bed and care um it is unclear obviously what the minimum threshold is by which any given customer cares I can tell you definitively 500 is too small for anyone to care about anything um maybe the answer is at a million kilometers people start caring maybe the answer is five million kilometers I mean whatever there's obviously different thresholds for different kinds of apps but the point is is you need a there are some pretty high minimum thresholds that's required for the system to work at all uh the same is true for Telecom networks um a telecom a helium network with 500 nodes is worth zero because it's just not large enough for anyone to care I'm

a helium a telecom network with a million nodes is now worth something more than zero uh because you've achieved that minimum level of scale what does this have to do with tokens and blockchains um you need a way to incentivize people for being early the guy who installed the first helium hotspot took on risk that the helium Network as a whole was never going to achieve a level of scale to where anyone would bother to use it at all that guy who that guy took more risk hotspot owner number one then hotspot owner number 300 000.

because if you're at number 300 000 you you may not know hey what's the Threshold at which the network is useful or not but clearly it's a much higher probability that it will achieve that Threshold at number 300 000 than at number one right and like that is the the purest definition of risk uh and therefore the earliest contributors to the network should be compensated for taking that risk um and like the simple way to compensate them is with more tokens and there's different obviously uh mechanisms you can use for Distributing tokens uh miles driven hours providing service cover whatever those machinations vary by network type um uh but like you need to reward the people who took the most risk uh and

tokens do that very elegantly uh in the case of networks like uh render um which is about gpus for uh machine learning workloads um there's another element to this that that's pretty poorly misunderstood which is that uh machine learning work machine learning workloads and tended to be fairly lumpy uh it's like someone shows up and is like I have a job like I need whatever a thousand you know GPU hours to do the job uh but then they finish drop and they go away maybe they'll come back in a month with another thousand job but like they're not telling you I need you know 10 10 GPU hours per hour for the next three months um they tend to come in these very lumpy

spikes now as n becomes sufficiently large Network wide that always evens itself out and it's fine it doesn't matter and you can just you know you can model these things as being nice Booth curves but in the early days of a network those those spikes are a huge problem I mean you can't model them as as being a smooth curve uh and so you need to have way more Supply than there is demand in order to get the network off the ground because no demand is going to show up if there's no Supply uh and so how do you how do you intend how do you compensate the supply for showing up even when there is no demand um and again tokens are a very elegant answer to the question um is you could actually compensate the early suppliers uh in that in that way

and I believe that is like fundamentally justified in terms of building out the long-term Network value uh of the system what do you think about the arguments sticking with render and GPU um I've heard from some people say you know gpus are not fungible and also a lot of the compute is going to be done at the edge so like on your actual local device and people point to things like Facebook and open sourcing their models and you could actually run that device potentially in the future what do you think about that um yeah I've pretty high conviction views here uh that this is like as perfect middle of the IQ curve as is possible um of a take um when demand for a service goes up 100x thousand X

over call it a period of a handful of years which is pretty clearly what's happening right now for just flops uh which is like Loosely defined uh it's very stupid to be like well this configuration of flops is not going to be able to service any of it but this other configuration of flops will be will only be able to service it uh and that's just obviously very stupid um the uh amount of time it takes to build a new Fab is like five six years with God knows how much billion dollars of capital Investments um but like demand is up 10x in the last six months and Fab capacity is up zero X in the last six months and so like what that means is all available software is

gonna slowly reconfigure itself to all available Hardware um and the hardware itself will become more heterogeneous and is distributed in whatever ways it's distributed and the software will configure itself around that so that you can attempt to try and get the markets to be as efficient as possible um the kind of classic debate actually this this reminds me of and I put a tweet about this a while ago uh was in the early 2000s uh as kind of businesses started to to meaningfully uh adopt uh computers at for kind of White Collar office workers and like intranets were a big thing there was a big discussion of uh thin clients versus thick clients rollouts and the The Thin Client view was that like all you need

at the desk is a keyboard and mouse and a monitor and everything else can be living on the server uh and the thick line view was no like performance matters and people care about the experience and you're going to have enough applications open that you're not going to be able to have a server big enough and whatever and it turns out both of those views were stupid because the answer was demand for total computational resources was it was effectively infinity and continues to grow towards infinity and you there just need as much chips and as much storage there's as much RAM as possible in like all configurations and it was just like a stupid debate to begin with um and I look at this in the exact same way all configurations of machine Learning Systems uh will exist uh there

will be stuff running locally in a privacy answer manner there will be stuff running on render there will be stuff running on whatever other render competitor and it'll be stuff running in Amazon uh and all none of these will kill the others they have different trade-offs and different people will be optimizing for different things so while we're on the topic of hardware and meet space a lot of people actually don't know that you have a pretty comprehensive background in AR and VR and when the Apple Vision Pro came out you had a really good thread which I enjoyed and I want to take this moment to maybe explore your thoughts on this new paradigm spatial Computing and if you think it has any weight whatsoever

or tie in with crypto in the metaverse the theory of crypto metaverse being tied to each other doesn't have to do strictly with with glasses or goggles um but the theory is you know if you spend more time in some sort of 3D virtual space uh with people and presumably NPCs running around doing things those systems will have economies and if this isn't economies then the question is what rails do they run on and like Loosely speaking there's kind of three answers to that one is uh inside of a central company two is through the banking system uh and three is crypto rails um and of those the one that I find to be most elegant and palatable is crypto rails but I'm a fairly biased party so

we'll see if that actually plays out um obviously there are some people who tend to believe that will be the case today look at the kind of MMOs like Eve online and star Atlas and some others that are kind of moving in this direction but still super early and you can't really say anything with conviction um the relevance of all of that to Apple and um to Oculus is like you know is that 3D virtual world experience so much more immersive and compelling in the uh goggles form form factor such that you induce people to spend more time there which therefore basically helps grow that economy because if particularly there's more time spent than presumably more dollars will flow through those rails

I'm talking about VR I think you know the two competitors right now that are out there are Apple and and Facebook or meta um Apple Vision Pro focuses more on like the personal Computing experience like productivity and home entertainment and then you have quests which is more social and gaming um how do you feel about those two because there's that quote that says like come for the come for the tool stay for the network curious which one do you think is the right way to go about it here with form factors it's a very hard problem to uh build adoption uh and it's a and it's the thing apple has a lot of experience with both the desktop layer as well as the phone layer uh and then and then the watch layer and now um

uh uh you know the goggles Lair or goggles form factor I should say but the most destructive example for the purposes of goggles is desktops so we got to go back to 1980 here uh and in 1980 no one was like I mean yeah maybe Bill Gates and Steve Jobs were like it's going to be so cool for games and like I don't know they got a pong demo and they were like oh my God this is amazing uh but like I think they've all recognized that like the first applications were going to be business applications and they realized that because once they understood like this could be a lot of white-collar office jobs could be made dramatically more productive those people those Come Those companies would be willing to pay for the hardware and network

infrastructure to get them set up and to train their people um and obviously a lot of those early employees were skeptical they were like what the [ __ ] is this and this is dumb and I don't look at the screen and my eyes hurt and I don't know how to type and I mean like there was a lot of friction that like had to be overcome you know in 1980 for all this stuff to happen that we all take for granted now uh and uh but they did it because the productivity games were incredible uh I mean in like imagine building Excel models without Excel like I've literally never done that in my life because I'm young I'm 33 I'm like young enough that I didn't ever do that but I actually cannot comprehend how to like do Financial modeling without Excel it doesn't it doesn't make sense to me

or like how to write documents without word like I don't I don't I don't know what to do um uh and and so like those are incredible productivity gains uh and uh so I think that's the kind of important takeaway is like in order to bootstrap the form factor of desktop you need to identify the two and like the two use cases were documents and Excel and then kind of email which is kind of both of those things PowerPoint is like actually 100x less important um but like the first app was visi calc and the second app was word processing those are the things that mattered uh and everything else was a rounding error um

and for phones actually Apple funny enough didn't experience this problem at all because everyone universally understood that the most important application of phones was first phone calls and then later SMS but like getting call getting in touch with your family and your friends and that was universally understood and there was no question about that right um and the criticism of the iPhone was not it doesn't make phone I mean there was a lot of jokes that like the phone calls aren't great but like everyone understood it made phone calls and like the question was are apps useful and the price point and all that those other things but it wasn't hard to get adoption because everyone understood like call your mom um and then if you look at the watch Apple had a bunch of theories of uh if you if you go watch the first um watch keynote

uh they lean hard into apps and if you then look at the subsequent three Keynotes you just see them slowly positioning away from third-party apps to basically health and fitness um and to a much lesser extent like fashion design statement um and they realize that like this thing has one function which is health and fitness and apple is now in in sourcing all of that into into the watch and they've just been iterating on that now for for many years um but it took them a while to figure that out um so now getting to the goggles or division Pro uh the question is like what's the what's the use case that justifies the new form factor which like and again this form like no one was like

man iPhone bad like it's bigger than my Nokia brick it's not going to throw my bot like no one really said that no one had a problem with the watch form factor when the watch came out but the perform actor people complained about was desktops in 1980 and like there's so many great articles of being people being like workers aren't going to stare at this green ugly screen all day and I mean you know [ __ ] like that and obviously now the equivalent article is No One's Gonna wear these damn goggles all day um and obviously there's lots of people who who kind of think that so Apple understands in order to overcome those uh criticisms you need to have a productivity gain that is so large that like all of those criticisms are irrelevant uh and that's what Vision Pro seems to be optimized for in the

workplace and that's why they're leaning into their strength of saying uh we're it's backwards compatibility with basically all Mac OS apps and it's basically just the world's greatest monitor like is is basically the right way to think about it um and obviously certain you know designers and Architects and I don't know whatever graphic design I don't know like clearly that class of person is going to be the first adopter of this product who's going to wear it for many hours per day uh and then like the question from there is how do you expand you know from there into others that's uh quite unclear at this stage but certainly step one at this point is fairly obvious um the other bet that Apple's making is just like again it's the best TV screen you've ever had and so they're going to try and you know create some very

compelling content experiences at home with 3D entertainment from like the Courtside shots at the NBA and you know getting Disney on board to make some some stuff we'll see if any of that is any good uh or or you know uh what's it called entices consumers but like that those are the best they're making is just like the hyper premium experience for the person who demands the best uh and then try and come down Market from there I have more conviction on the workplace side of that at least for step one than I do have the consumer side of the stat but I haven't worn the things and seen the consumer experiences so who knows Facebook is taking the opposite approach saying like hey can we deliver something that you want to play a fun game and then we'll try and make it social kind

of roughly seems to be the the Facebook theory um I don't have particularly views of like how well it's working or not there's a fair bit of negativity in kind of the mainstream press that like it's not working um I think it's I assume it's unfair because the Press tends to uh hate on things they don't understand uh but anecdotally my views I guess aligned with what the Press says but I don't actually know what the data data says so unclear so I'm gonna pivot topics here while we're short on time you are obviously a very big proponent an advocate for first principles thinking and just being a contrary and being right crypto Twitter is a very interesting

place dominated by discourse of groupthink and let's say peer pressure but really it's just group think how do you so that's your questions one is how do you cultivate such a contrarian and first principles thinking approach to crypto especially you know Twitter is very Infamous for being very critical of contrary in folks and two if you could use that thinking that first principles approach what is the one product or service you would build in crypto that nobody's building uh I disagree with the framing of contrarian uh I think it is not a useful way to think about people or the world um what is useful is first principles

thinking um and I do my best to be first principles uh in my thinking uh I think the I don't know how to cultivate that other than try and do it uh I love it I mean like like that and then like you know ask anyone who's been on our investment yeah Spencer ask Tryon as for shawl they've been on the investment team for for quite some time ask even the people who listen into our investment team calls who aren't on the team like our CFO on our GC and others who you know listen in and what they'll tell you is like enough people have argued with me enough about enough [ __ ] that they have come to realize that they were not thinking for

first principles and like that has rewired their brain so One path to getting there is argue with Kyle enough um and Tushar and others um uh so that that seems to work on very small scale uh I don't know how to engender it on a larger scale that seems kind of sociologically impossible given my understanding of like broad information flows and human biases uh via low Fidelity communication channels uh so I don't know I don't know what the answer is unfortunately that was part of the question what was the second part of the question the second is if you could build one product

that nobody else is building what would it be what is everybody missing uh [Music] um I mean this isn't this is kind of a non-answer but uh figure figure out how to imbue some notion of financialization into a mainstream consumer Behavior uh or like to go kind of the uh crude way of saying it would be decentralized social uh which I think is most a counterproductive term uh but like uh I want I think it's very important for

crypto as a whole to have at least one app that is like fundamentally social in nature that like my mom will understand what it is and why people use it even if she does not want to use it on her own and that it is like you know universally accessible and understandable very quickly um and that embeds some notion of financialization uh that that to me is the most important thing obviously many people today are running experiments on how to achieve that from forecaster to drip to dialect too much of other people uh and we we talk with a lot of these guys we're investors in a handful of them uh and are continuing to try and refine our thinking uh on the paths to

get there but like that to me at the ecosystem level it's not a question is like the single most important thing that could happen because once you have a senator and you have a congressman and they're like their kids use it like the discourse completely changes but you have to take on ZK that a lot of people don't talk about and CK is used for privacy and proof of computational integrity and I'm curious if you think something with that's going to be one of the next big apps or is it going to be leverage for the next big app and and maybe you can go into the fhe uh introduce that to people as well uh uh I'm fairly certain that any large-scale consumer app in the near

term the first handful of them to the extent they exist will almost certainly not use EK uh just because the technical complexities of introducing ZK are just so large and you just like you can't iterate quickly and do product experimentation and just like ZK is given to your technical complexity basically antithetical to like consumer product building um now that that's like mathematically true um it's just practically true because of like how building consumer products Works um so I think it's very unlikely ZK is relevant at all to anything consumery um uh that's question one question two I guess is like how are you thinking about

CK and privacy and proofs of Integrity um I'll comment more on the Privacy side than the proof and Tech proofs of computational Integrity AKA scalability side uh I'm fairly confident ZK is not the correct answer for um privacy on chain um I want to caveat that with a couple of things uh if you're if you are trying to say I can prove to you that I am 21 so that you can tell me alcohol but I'm not going to show you my my address on my driver's license uh that maps to ZK extraordinarily well I mean that's like literally what his urinal proof does is is that um so if the goal is to preserve privacy is it like proving something about

yourself that works very well if the goal is privacy in the context of defy so you now now there's some sort of notion of shared State and there's an LP pool and uh or a limit order doesn't matter and you have people crossing the spread and you have people doing interactions and there's no math happening between people uh I don't think that maps to ZK for privacy um the fundamental reason um is in the world in which people are submitting ZK things to a blockchain to do these kinds of of financial transactions there is no notion of global State uh and and therefore if there's no notion of global State you cannot reason about global State

um that was a fairly abstract statement so let me unpack that um the simplest way to understand that is to just think about zcash um in zcash like what is the Z cash proof say whenever like if you submit a uh I think they're called Z transactions or whatever you can Z cash uh and and the proofs roughly says um there are some series of utxos that have been sent to my series of private addresses in this encrypted blob such that the total number of utxos I've received is less than the number of utxos I have sent out including the current transaction basic leaders are saying my balance is larger than zero is really all I just said but in for the context of you know

defining a zero knowledge proof that's basically what the proof says um why is it important why like the reason I I spoke about it in that way is because if all of those proofs are in fact valid then you know that there will not be more than 21 million Z cash um the cash for Bitcoin so it's 21 million Supply schedule uh but there's actually no way to audit the supply of zcash and say are there more than 21 million tokens um this has been a known fundamental problem I'm not going to call it a problem I'm going to call it a property of this this proof design system uh from day one uh and in fact there was a bug Circa 2019 or 2020 in which they

discovered that there was a bug and someone could have been taking unlimited number of zcash in in the in the encrypted pool uh no one believed that it was taken advantage of but it was discovered it was patched and then disclosed afterwards by the electric Coin Company um which further highlights the fact that there was no way to audit the system from the top down um and and so coming back to D5 why does this matter well if you can't reason about the system from the top down then like defy kind of at least D5 as we know it right now doesn't work um if you can't reason about the system from the top down then there is no xyk you know what K is you never know what X and Y are you don't know how those things move along the curve if if

there's no way to reason about it from top down then what is how does collateral work in office a like it just it doesn't and then title management and Ru solvent and your health factor and all this stuff like those Concepts don't work when everyone is submitting a bunch of private proofs to the chain everyone knows their own view of what they think is valid which is fine but for defy you need to also have a top down view to work and that like fundamentally doesn't map to a bunch of encrypted ZK transactions now that doesn't mean it's impossible and obviously there's a bunch of people alio and Aztec and dark Phi and that's probably have a lot of teams I'm forgetting that are really trying to enable these these uh you know ZK sdks

for private defy but like they're all dealing with this like very basic logic problem of like how do you reason about the system and like I I cannot describe to you in finance terms how to do that let alone then translate that into computer science and code um and so I I think that that's kind of fundamentally not gonna work um to me the right way to get to zkd5 is just fhe or for homo for encryption um which literally means just imagine all of the contracts as they exist today on the Chain but they're encrypted and you can just run all apply all of the state transitions to encrypted stuff on chain and no one could and the validators don't need to actually know

what any of the balances are to apply the Transitions and run the comparative if statements and such um that that to me is like and the beauty of that system is you just the core logic of the system is preserved and in fact assuming you have an fhe to VM f-heed either evm or svm uh you would likely be able to reuse the substantial majority of code that's already been written um for that new for that new world and that strikes me as like the right way to solve the problem yeah that's super interesting I think I think we can probably go into the rapid fire round as we close up I do want to ask Kyle one one quick question I I love your communication style I'm just

curious has it always been that way like you're very direct um authentic more so than most people are and I'm curious like do you intentionally do that has it always been that way and like do you encourage people at your firm to do the same uh definitely encourage people at multi-coin to become more uh direct and concise and I think it happens kind of naturally over time uh uh I'd say I've mostly become this way uh primarily because of writing I was a substantially less clear Communicator pre-writing than post writing um certainly and probably Ben Thompson was probably the most important uh figure who helped me refine my own writing I mean I've been I've read every

single Decker article ever written without exception from inception to today I've never missed one article um and he's had a big impact on me uh and then going through your own blog posts and refining them and refining them one thing I find that it's very fun to do is obviously fun it's very painful to do is to go read your own [ __ ] from six years ago uh and see how terrible it is um uh but but a better way to do that and that's more productive and less intellectually painful uh is write something that you want to publish you think is good and then don't publish it and then come back to it in six months and or three months um and you're like wow like my grammar construction and was very poorly was not

concise and such and after you've done that a few hundred times uh with and be real hard on yourself that typically helps improve but it's a pretty intellectually painful process I also edit this podcast and I have to listen to myself afterwards so uh yeah that sucks um I know it's a good learning experience so I'll look back in six months and probably be like wow what were you doing back then but yeah it's a good learning experience yep I'm not gonna I'm not gonna do rapid fire uh I just want to ask two final questions um for generic cliche questions which I'm sure Kyle won't like but uh number one is you obviously talk to a lot of Founders um especially in the crypto space what does one piece of advice you'd give to

crypto founders don't listen to crypto Twitter do do what you think is best and what is one piece of writing advice you would give hi this depends on what your objectives are as it pertains to writing uh if your goal is to if if your goal is to improve the own quality of your thinking uh then you need Cadence um or you need you need wraps so you need to set a defined schedule of writing X number of things per week let's just say x is one uh because you can't you don't have the time to do more than one uh it doesn't have to be public it doesn't have to be face publish it

doesn't need to be publicly facing but it needs to be something that like passes whatever you think the the intellectual bar is for like intellectually interesting um and obviously that that's going to have some context dependence depending on the nature of your business uh and try and produce one of those per week and it could just be internal like do you keep to yourself or it could be something you share with your team um or it could be a public-facing document uh but force yourself to go through that process of write refine edit your own stuff and you just need to do that a lot of times I haven't been doing this a lot now but when I was trying to write one article

per week one thing you noticed too is you actually start noticing things and and what do I mean by that is like you know you have to produce content that week so throughout your day you're actually going to notice events like hey I should write about that that's interesting or I should think about that differently um but Kyle thank you so much for coming on uh this is an amazing conversation I recommend everybody to read all of your blogs are on multi-coin.com I'll put all the links in the show notes um but yeah thanks again this is a really great conversation uh Garrett Martin thanks for having me on also our URL is multi coin dot Capital uh multipoint.com is I think they keep trying to sell it to us and I keep not you know bidding their stupid numbers but anyways thank you all for having me

on awesome to be here excited to be an early member of Lightspeed and excited to see this thing go up into the right [Music] foreign [Music] [Music]