State of web3 research

Transcript

[Music] all right so good morning everyone uh so this week of CSX is going to be uh sort of the research uh Team week and I think I speak for all of us when I say this is absolutely a highlight of our year for the research team so let me just sort of open it up with a little bit uh maybe five minutes or so just to kind about what we do on a day-to-day basis and then we'll have time to hear from each of the team members uh in some depth then we'll a good half of the session at least 45 minutes we'll open it up for Q&A from everyone everyone here the needs of the space the needs of sort of people building in crypto are spam many different disciplines and so the research team is accordingly built to give us coverage across all of those

disciplines and so the areas of expertise of the team members right so they range from you know market and mechanism design to cryptography you know to consensus and distributed systems to governance all of these things are crucial for the success of the space and so we want to make sure we have coverage of them uh and then again you'll see more of that sort of later later today um and as far as what it's like to to be on the research team was funny I gave I gave a sort of keynote earlier this year at the financial crypto conference and I was being introduced as the keynote speaker and I was introduced as the person with the with the best job in crypto and I was kind of like you know what that's not it's not totally wrong actually um and so so why is it so fun for all of us to

be doing this well we kind of lead I'd say two lives which at first blush maybe sound like very parallel but there's actually unbelievable Synergy between them so one of the lives that we lead is just helping our portfolio companies and you know companies in CSX uh solve basically hard technical problems across all the disciplines uh I just mentioned and that's the sort of Life of ours that we're going to be emphasizing uh today and tomorrow right so you'll get sort of an overview in the morning you'll sort of see us in action in real time this afternoon in the live workshopping sessions and then of course we'll have uh one-on ones um tomorrow uh and so you know the a16z crypto portfolio you know it's over a 100 sort of startups that's not even counting all the CSX projects and we don't work with every single one

but we work with you know a lot of them dozens and dozens um of the of the companies in the portfolio it's just because the sort of space is so early and because it's just sort of has such a technical nature um a lot of our the startups in our portfolio they're really blocked by what are in effect sort of research problems and so it's just sort of a perfect fit um sort of for our expertise um so that's that's the first life uh and then the second life is really a little bit more taking a step back and being just really you know trying to do the best sort of research uh in crypto that anyone is doing right so all of the team members for example regularly publish you know sort of rigorous serious research papers and sort of top venues that's sort of an

expectation um for everyone on the team uh you know so like you know you know one example um sort of led by sort of Justin Taylor who you'll hear from in a second right would be like the jolt um proof system sort of a new zvm that's really like 102 pages of actually like sort of pretty sort of really difficult theoretical computer science research um which then was sort of built out into sort of into a real system by um two members of our of our engineering team uh you know and that's the kind of thing where you know all the research we do is open source right so those are just those are papers in sort of you know eurocrypt to top academic conference um anyone can use those Innovations you know we would love it if some of our portfolio companies sort of built on it

and if they do we of course give them a lot of assistance we would also love it if other companies so build on it as well right because we ultimately what we really care about is seeing the space uh succeed and so a big part of so this other life is really just sort of again doing the most high impact research to advance the space as quickly uh as possible and so I think One Direction is probably pretty intuitive right so having this sort of really cutting edge knowledge of what's going on in research you could imagine that comes up all the time when we're giving advice to portfolio companies about sort of how to solve their how to solve their hard problems but what you might not know is that the um sort of impact in the other direction is massive right so if if you

work in a sort of really mature area you know like my original training is in algorithms for example right and if you want to sort of figure out what would be a great algorithms paper you can kind of like read the great algorithms papers from the past couple of years and just try to sort of improve upon them right it's sort of very mature so the research tends to be more of an incremental nature than a breakthrough nature right whereas in the areas that we're working in right so in web 3 right there aren't even you know for the most part textbooks yet right so even just sort of like the most what's going to be viewed as the most fundamental work 10 years from now a lot of that simply hasn't been done you're not going to be able to just read some research papers from a

year ago and know what's the really important research paper to write so how would you ever know what the really important research paper to write is you guys basically tell us right so we're meeting with projects like yours and startups in our portfolio all the time asking about what they're building where are they stuck right and something that happens to all of us all the time it's kind of like but be maybe like a twoe span you know where we meet with three different projects and we kind of realize they're all blocked on exactly the same thing right and that's like the Universe telling us this is going to be a great sort of classic research paper to write no one else is at the opportunity to even see that this paper needs to be written right because no one gets the bird's eyee view that we do by

talking to so many different projects so that's why it's sort of just as a just from a purely research perspective that's why it's such an amazing opportunity and it's so exciting for us to have this opportunity to talk to to talk to all of you about about what you're building so with that uh I now want to go through the team members one by one and so I want want to ask all of you to on the one hand um do a brief bio now everyone in the audience right has had a chance to kind of see you know sort of you know the sort of the basics of your bio so feel free to see a sentence about that or two um but maybe also I'd like you to augment the bio with kind of a personal story so I talked about how a lot of what we do is

helping portfolio companies uh maybe you don't pick out an anecdote right and just you know maybe it's an anecdote that summarizes a couple different things that you've did how how how does that sort of happen for you specifically in your areas of expertise to give the crowd a better sense of of what it is we do on a on a day-to-day basis so Justin let's start with you great so I'm I'm Justin Thor I'm an associate professor at Georgetown and a research partner here in the research group at a16z I work on cryptography um algorithms theoretical computer science I guess I spend most of my time these days on zero knowledge proofs one of my main focuses um for the last year or so has been uh building a proof system called jolt um with a couple of the engineers here um

although one has uh just left to start his own own Venture as a Founder now just like most of the people in this room today um and so that that's been a wonderful experience uh this sort of came about uh where uh through our research program summer internship program uh what I was just talking to to one of the interns um about how we thought the right way to design what are called ZK VMS are um and the these are uh extremely programmable like usable ways to um develop Zer knowledge proofs and at the time they they were just horrendously slow um so we you know we were discussing the approaches people were taking and Alternatives uh that we

we thought would be better and uh just a very clean um uh approach to this grew grew out of those discussions and uh you we sort of were telling the engineers how how excited we were about um this this system and they they wanted to help us um so it actually took a little bit of convincing um to let the engineers help us it turns out that uh there was no way we would have ever built the thing without about two years of uh engineering time the system is still maturing and we're we're hoping that in the coming years uh it will kind of move the entire space forward um and the the entire blockchain ecosystem will uh just be able to accomplish a lot more uh as a

result of of this project uh which I really don't think could have happened really anywhere else because of the combination of of research and Engineering um so really been a wonderful experience being here um and the the work has been informed by portfolio companies and you know the people trying to use this technology um but also just the combination of expertise inhouse so in addition to the intros I'm going to as we do them ask each team member kind of a question to try to get them to hold forth on something their areas of expertise so Justin let's talk a little bit more about zvm um I think probably most people in this room are kind of aware you know of the you know many people believe that blockchains need Z zkv Ms for scaling right and so my question for you is you know do zkm sort of need

blockchains or is it somehow a more fundamental advance from a computer science perspective right and so I have in mind you know are there applications in you know in the web 3 world other than just sort of scaling up L ones that need more throughput and also feel free to sort of ask answer a broader question you know sort of zkm where they might they be used outside of web 3 as well right right so I think that um blockchains are a fantastic application domain for zero knowledge proofs because they have two really unique properties um so so one is uh you really really need adversarial guarantees right like everyone in the world has an incentive to convince the blockchain that they have you know thousands of eth when they don't furthermore the the statements

people want to prove in the blockchain context are often very simple um in the scheme of things like I know a private key controlling this account um and therefore I'm authorized to um send this transaction or something like that this was like the one domain that could have sparked like the development of this technology to reach the state um it it that is actually usable um because uh yeah it gives you the very strong adversarial guarantees it's also very slow today uh but it's okay that it's so slow in some context because the statements being proven are are so simple um so I you know I I think that uh blockchain zkm will make them much more scalable

and will bring privacy to the blockchain where today there might not be privacy um and then I think the the spillover effects outside of blockchains will also be substantial um so in in the blockchain setting like there's a a statistic that has been going around the last few weeks um that you could prove everything that happens on ethereum Main net in a year with a million dollars of compute costs right which um sounds great proving can't be that expensive if you can prove everything on ethereum for a million dollars you know all activity in a year my take is is more negative however I tend to have very negative views on things we call them our house kogin just so you know so um so these proof systems are like a million times

more expensive to prove a computation was run correctly than to just run the computation so what this really means is that the total amount of computation H happening on ethereum today if you just ran it on like a cloud machine is a dollar's worth right it's like one machine for under a day of time okay ethereum you know I would like it to be like a world computer where you know everybody in the world can use it to do sort of whatever they want and if the total amount of work the world computer does in a year is equivalent to one box for a day that doesn't feel like a very powerful machine to me right so um you know it's you're viewing the same fact in two different ways one is oh look at

it's sounds like these proof systems are so fast it's only a million dollars to run them uh the other view is the blockchains are hardly doing anything today you know once blockchains scale up you know who knows what applications they're going to be used for but then things will spill over to um non-blockchain settings you know I was uh Joe is an expert on network security and uh you know he tells me that like the INF internet infrastructure today just does a lot of really silly things um that don't have to be so silly if you have scalable zero knowledge you know people download like huge amounts of certificates from certificate authorities and verify their certificates themselves when instead you know the entity that has all the certificates could just prove they have the certificates um and so these things

can happen in the background of existing infrastructure and make a huge difference to like the cost of running the internet and it'll just be hidden from everybody um so I think we'll see all of this and probably it never would have happened if all the investment from blockchains hadn't spurred the the Innovations great so let's let's go to one of our two virtual guests let's go to Prav next pav you want to go ahead and introduce yourself and maybe an anecdote about sort of the collaborations you do with portfolio companies oh yeah hi I'm Prav um I graduated from Columbia um about a year ago now um and I've been at a16z since um and kind of my my background is more on the algorithmic game theory side mechanism design side and in particular kind of problems more with

infrastructure layer and so I study kind of different l1s l twos and kind of lots of problems around how how do transactions kind of go from a user to their wallet to potential block Builders or Searchers and to eventually landing on chain and so in particular people have been talking about kind of the transaction supply chain over the past year or two and so I kind of studied a lot of problems in this area kind of trying to understand the incentives behind um who like who kind of wants to look at transactions and possibly extract M from them who wants to bundle transactions in certain ways kind of how much should transactions be willing to pay for some type of block Builder or minor to include that transaction in a given block um and so yeah so I kind of

work on trying to understand this metag game between all these different participants that go into the block construction process on various different L ones and I I think a good anecdote of kind of how this comes to practice in terms of me collaborating um with some of the portfolio is I'm particular with some of the old ones kind of like salana and Aptos andu and that they're kind of facing problems that haven't really been studied before and kind of we kind of at this point of a long history in ethereum of understanding um how to scale things there's a lot of people looking at the incentives there and looking at the kind of the the market micr structure that's emerged between the various Builders and things like PBS and understanding thefe mechanism like EIP 1559 and so Theory

might say at this point there's still lots of work to do and it's still kind of a a fastly evolving space but there's at least kind of lots of people paying attention and asking kind of various questions around how should that transaction be structured and how do things like designing TBS change the incentives um for different kind of parties in the ecosystem and these kind of questions have been just much less studied in other um l1s and so one thing I've been kind of very interested in is trying to understand um even one simple question is how should you design the mechanisms on the S ones right so in particular for kind of for Solana they have the notion of a a paralyzed virtual machine which means the resource usage of transactions looks pretty different

um from ethereum where things are processed in a more single- thired way and this is the same for kind of aptus and so who also kind of use this um parallel execution but even kind of like Su aptus have a different version than Solana where kind of salana has a kind of a pessimistic approach where they kind you know ask kind of users to declare all kinds of possible conflicts of the transactions up front and kind of build blocks accordingly and soon after go the opposite route and kind of have an optimistic approach they try to kind of handle conflicts on the fly during execution time and so this actually turns out um kind of changes how you want to design fees to kind of address these two different kind of areas and so I've kind of been actively working on how like how should you formally think

about this problem what kind of the uh exact objectives you might want to to achieve um and like kind of how should you kind of redesign certain things uh to basically provide you know good ux for users while still being efficient and you know making the system Sol isn't the spam um and still kind of Performing efficiently and yes I kind of work on problems like that kind ofl2 and kind of various different teams that kind of need um that have questions around this type of a mechanism Des sign work that actually largely answers the question I was going to oppose to you to drill down on so let me let me pivot and ask a different one again so there's a lot of research papers we could write how do we choose which ones kind of to write and pava I feel like in your case right

because what what you work on is so quickly moveing right I mean so like you said I mean Once Upon a Time right all there was was ethereum for smart contract platforms and you know I think we all know that model well there's kind of you know transactions just have a gas which is sort of the amount of computation you just sort of charge per unit a gas you have some sort of relatively simple auction for deciding sort of the price per unit of gas and now between you know l2s Innovations the consensus layer Innovations in execution the rise of me right all of these things have happened you know by research standards a very very short time frame so how do you just like what how do you think about like what what are the things that you know you know you save to just tell our

portfolio companies one-on-one what are the things that are worth writing a blog post about what are the things worth actually trying to really you know formalize and writing a sort of proper academic research paper about how do you think about that yeah I think all this stuff is pretty fluid I'd say um I think it takes me a while before kind of even starting I kind of I guess um I might kind of first have an idea by even being on Twitter or maybe like going to a conference chatting with some people and you kind of like see the Sparks of some kind of common threads of problems people are working on like d mention issues are kind of and this happens all the time people like oh like the networking stack is broken oh for some reason transactions aren't landing and you know they'll have some Engineers

they like ship some fix and it like kind of work for a couple months and then a couple months later something else will break and you kind to see this pattern throughout lots of different parts of the space um I think for me kind of like the benefit we have as researchers just being able to see okay what's like how can we abstract these problems and say you know from an engineering point of view of course you have to fix these things um and there's certain kind of shortcut or Solutions you might apply at the moment but like what really is underlying problem um that's kind of springing up ecosystems are also not always super in tune with each other and so I think kind of as you mentioning before one of the nice things is you have like a very Bird's side view so for

example you get the chance to talk with Outlet people and Salon of people and so kind of when we're tring with two different engineering teams I kind of have their own language describing the same problems I I think kind of when I realized okay they're actually talking about the exact same thing um and really they're working on the same problem but they're just coming out from different angles have kind of a different there's certain path dependency and kind of the way different infrastructures kind of spring up and so the different teams have priorities and like what exactly aspects of this SP problem they should work on um and so I think something you know there's always like blog posts to right of just kind of even surfacing these problems so I think often times um I might not like enough time might not even know kind of what's the best way to

like fully abstract this yet but even just kind of kind of like look like here's this thing that's happening um let me try to just like put some more proper words to it and kind of a bit a bit more formalism but not like a full kind of paper um and so often times I think that can be really useful for a blog post just to kind of get people aware of this these problems that are happening um you know kind of with like shorter portfolio engagements often times you know might have some like intuitions about how to answer or some of these problems um from kind of seeing some of things crop up but like it's still not a point where it's kind of like a full I have a fully formal Theory or I wouldn't I claim to have an answer myself of kind of like a rigorous answer

of what I would do but just kind of from what I've seen are best practices what might be most reasonable for the team to do and kind of what are the main kind of you know like foot guns they should avoid that like kind of other teams have run into um and I think kind of after chatting with teams for a while and kind of like mulling over kind of a problem for a while I think eventually certain of these problems get to the point where they actually do end up being kind of more full research papers I think it's um definitely helpful to kind of let these things be run for a bit it's often if I like Tred to write a paper immediately after learn about a problem then like the the whole thing would shift andever you know that four or five

months it takes to get for idea to actually finishing a paper in which case you know they'd be wasted effort so I think it is kind of good to see how do things play out um over the course of a year say and then kind of once it feels like there's a really substance to fundamental problem there um be appropriate to WR a full paper all right thanks pav and if you need to if you want to go to bed please feel free so pav is beaming he's he's a total champ he's beaming in from Singapore so um feel free to get some rest if you need to so next up uh is Joe hi everybody uh I'm Joe um I'm a associate professor at NYU and a member of the research team here and my main focus area cryptography I'm kind of most interested

in the full space of cryp crypto and problems that cryptographers think they can solve and the fact that actually very few of them are being done in practice not just in the blockchain space but more generally the if you spend enough time working in academic cryptography there's really just incredible solutions for so many different problems the vast majority of which when you look in practice like people aren't encrypting anything they're kind of doing very simple uh Solutions so I've always been pretty fascinated with uh with why and sort of the limits of what we we don't actually do technically maybe a research anecdote project that I worked on a lot in the last year um I really started last uh last summer so like summer 2023 but we

talked to multiple teams who are interested in you know data availability storing things and uh you know either blob data for on ethereum with you know dank sharding or other Solutions and we kind of noticed there was always this like well yeah like we there's no incentives for actual data download we're just sort of hoping the data is able so you know somebody will be able to download it at some point if they really need it but there was always this kind of I notic that everyone sort of recognized that there was this kind of hole that there were never really incentives for actually um making data available um for download so you know all the the crypto that's actually been deployed and there have been multiple

systems that do this is proof that you actually have the data proof of data retrievability or possession um and there are no proofs anywhere that you actually let people download the data which is kind of this like tree falling in the woods thing if the data exists but you know nobody can ever download it does it really does it really matter if someone is storing it so yeah I mean this was a a big project with me uh Lara who uh former member of the research team on a on a sailboat now and I think five different uh interns that we had so it was a really fun project cuz we had to uh invent some new cryptography to actually do this proof that you uh let a client download data there's some interesting incentives to

think about the the only way to solve this is to use a um to use a blockchain basically as a as a trusted third party there's a famous uh cryptographic result that you can't do Fair Exchange in general for data or anything else without a trusted third party we also had to do some pretty careful um like specific engineering to get the performance to look at all reasonable for uh you know the way that uh blobs are committed on ethereum you know using kg commitments we had to sort of build it in and make a custom Uh custom snark protocol around that so a really fun project I think it showed off kind of the strength of the research team taking on a project that had you know multiple different things that needed to be

engineered and also this was uh I mean I think really like there's a lot of these still but one of these Corners where everyone's like yeah we're not really using crypto for that problem yet we're just sort of like hoping for the best so hopefully in the future that's you know something that we'll actually have uh you know strong incentives for people to actually serve data for download great so let me let me sort of uh you know double click on one of the things you said I like that you mentioned this there's lots of problems where there are in principal Solutions on the academic side you don't really see them in the wild and often I don't know if the rest of you feel this way I'm often sort of unsure is it because the academic Solutions are purely academic or is it just the practitioners really should be

more aware of the academic Solutions and they do great in practice so and I think there's probably lots of cases in both directions so just to sort of tie this back to sort of the web 3 world a bunch of the challenges that a lot of projects face are really around sort of information leakage right so like me being like one huge example right with validator sort of knowing transaction data sort of before they sort of have to commit to a block you know you can imagine other things too like if you try to run an auction over a blockchain right you worry about people sort of learning people's bids in advance so you know but we have like we've had encryption for what almost 50 years now right so what's your take on sort of how much is it that like you know we can

just you should build better web three projects taking better advantage of for example encryption to for example solve me or is it sort of more to the story than that yeah so this is the you know really interesting question I've been thinking a lot about recently um yeah I actually think with uh so I'll I'll talk about encrypted mol specifically so this idea that if we just encrypt every transaction and then we commit to some ordering so you usually people call it blind ordering and then you reveal all the transactions then mvv should go away because there's an ordering before anybody's seen what the what the transactions actually are yeah I think that's one um it sounds really simple it sounds like kind of the elegant natural way to solve the problem um when you really drill down

into it though cryptography it doesn't prevent a lot of the leakage that actually matters in practice so it can't prevent people from learning easily at least what time a transaction was posted to the network uh you know encryption doesn't limit the size unless you Pat up and there's kind of a big performance hit there um and then there's this specific spe ific problem in web 3 that doesn't exist very much in the crypto literature which is that you somehow uh you have to to prove that a transaction is actually funded that the transaction fees are paid for before it's sequenced um other well if you don't do that then you face a big spam problem because you'll end up you know uh sequencing tons of transactions that aren't funded and just go away and waste

block space but you know if you have to to actually show that a transaction is funded it's very very difficult to do that while not revealing like who posted the transaction um or any link to the uh to how much money the the creator of the transaction has so I think that's actually a case where kind of crypto veterans uh you know who think who look at this problem recognize like actually getting rid of all the leakage is going to be extremely hard um requires you know a lot of engineering and we'll have actually like a non-trivial performance hit which I think is why it's probably not like the idea hasn't gone all that far um cuz these are actually fundamental problems that you know C the cryptography Community kind of

understands that there's no way to solve uh like time and size side channels without a big performance hit auctions is interesting too like we also worked on auctions a bunch on the research team and you know have some papers on it um and that's another one where the academic literature is really deep on uh encrypted auctions where all the bids are you know a sealed bid auction all the bids are encrypted then all the bids are revealed um and when you try and do it do them in web 3 you get into kind of a similar problem where you you want everybody to commit to their bids but you have to have a proof that they actually have enough money to support the bid um and that sort of requires that you have like a a hidden source of funds and you get into a lot of a lot of

issues so basically when we looked at it we realized like 95% of the papers in the crypto literature on auctions just don't don't really work uh in the web fre space um so yeah uh I maybe I drifted instead of actually uh answering your question but I mean I guess you said is is there something more going on and yeah I think like there is a lot more going on and uh you know that's why I like it but um you know uh cryptography is very very hard to actually do in practice so and I think it's interesting because both academic cryptographers sometimes would say like why can't we just throw cryptography at the problem and also people in the space who don't work in cryptography think like and it's kind of only you know I

like working kind of in the intersection I guess where you know recognize the the complexity excellent uh Scott great to see you all um I'm Scott commers I teach at Harvard Business School and the Harvard economics department and I'm a research partner here at a16z crypto I work on market and incentive design so I spent a lot of time thinking about you know the design of systems whether they're sort of like Micro Systems like you know sort of toonomic plans um or or incentives in a primary sale mechanism like in auction or or some other sort of you know uh sales mechanism or where people are going to interact and and discover price or discover an allocation uh to macro questions like um you know sort of platform incentives balancing supply and

demand um how you scale and grow your platform sort of how you think about platform strategy relative you know platform and protocol strategy relative to competitors you know as such I spent a lot of time helping different teams if you will you know break things early right as a lot of time commenting on you know sort of plans for for token incentive schemes uh trying to figure out all the different ways that people might you know game them or or misuse the uh the system uh and and spotting sort of like subtle flaws like you know you have two different objectives like your token is supposed to feed two different systems right one is your like you know main protocol but you're also hoping to use it as a gas token and like these two things can can contradict each

other in various ways like the macroeconomic implications of you know having a token that driving the operations on your uh you know on your infrastructure layer but is also supposed to be used in an application layer you know can can sometimes conflict um alternatively uh you know I'll help teams think about as they sort of think about potential partners and and the layers of their Stacks right you you have a protocol that maybe has like a like a token issuance component and a processing component and then some like application layer built on top of it I'll help them think about where value capture in this system is likely to work you know where you can have a sustainable competitive advantage and

where you want sort of lots of other players to sort of build on top and actually like maximize competition in a way that uh you know sort of drives up the total value of the ecosystem I can't quote specific I i' sort of hinted at some abstracted a lot of these I can't quote a very very specific instance because a lot of these token Protocols are still in development and and still subject to adjustment um but you know I'll often find myself you know sort of helping teams realize you know versions of you know if you're you've introduced a form of staking in order to you know sort of in order to pull tokens out of your ecosystem that can contravene the ability to use tokens as the utility to function in your ecosystem right it's

like you've opened an arcade uh and you have a rule that says I want people to store their arcade tokens as Savings in a bank but then no one can play in the arcade these sorts of things can can sort of introduce contradictions um other instances uh lots of times teams will spend a lot of time like optimizing uh you know sort of like optimizing a very specific and fixed monetary policy um right that they've you know sort of decided here's exactly how the tokens will be issued here's how the tokens will be structured uh but then actually like sort of the use of their system demands like sort of a flexible monetary policy over time and so I'll help them think about you know okay you just launched this thing it's in its sort of first stage or you

launched your Proto token right an offchain you know sort of reputation or or medium of exchange system how does you know how do you use the information you've learned from watching how people use the token to you know sort of calibrate how you want to use it going forward how you want to evolve it going forward so Scott in teaching at HBS for the past sort of decade plus right I mean you've been in this sort of amazing position where sort of the evolution of the tech you know you're taking a lead in sort of evolving sort of the education at the same time right and so for example my go right you got there and uh you need start teaching sort of web 2 business strategy among other things right and now you're really kind of you know amongst the you know amongst

the you know very small set of leaders defining what web 3 business strategy even means and so you know I know you have a lot to say about sort of what's different between web two and web 3 you already sort of touched on sort of the token aspects you know I'm sure we'll talk a lot more about that in the workshopping session um this afternoon and I'm sure in other weeks we'll touch on that too so let me actually drill down on something separate which is um kind of the cold start problem right which is something that certainly sort of you know Web Two startups have had to deal with web 3 startups often have to deal with so the idea of sort of network effects you know how do you get them going how do you sustain them you know pitfalls you know playbooks how has that

changed you know as we sort of gone into this web three era awesome great question um and yeah I should say I've been for the last couple of years I've been working with uh one of my colleagues Shai Bernstein and we've and and also leang wo who's uh who's our you know sort of basically like chief of staff at our Harvard crypto lab uh and we've been building hbs's first ever web 3 curriculum and so it's been really exciting to like you know bring in you know sort of both players of The Cutting Edge including a bunch of portfolio companies here and feel the Evol the evolution of the Strategic questions in real time right like this is you know Tim was talking earlier about the feedback between research and practice and practice and research and like you know all of that feeds together into

pedagogy and we're sort of trying to train people on how to you know build web 3 right uh correctly sorry I should I should use proper grammar um so the question is about how like what can we learn about platform launch and growth and scaling and and what can we sort of what's still true from the web 2 era and what's really changed so first of all the some of the core principles of platform launch are still true so like one of the foundational qu you know one of the most common questions I get in fact is you know sort of towards this cold start like do you go after Supply or demand first and the fact of the matter is there is not a uniform answer to this question right like some people will tell you oh always go after Supply

Supply is always hardest like that's not true right what really matters is the impact getting sort of Agents on one side of the market has on other sides by the way this is this is Being Framed as if it's a two-sided platform but but lots of platforms are one-sided especially protocols often just interactions between big pools of users some of them are many-sided the Insight is the same what matters is you know how much does one you know bringing in one player increase the gravity you know to of your of your platform or protocol to bring in others right so you know you subsidize the agents who drive the most value and gravity for for other agents to come um and why is that by the way that's just like simple you know sort of

most effective marginal use of your subsidy right if you're subsidizing a player uh that's going to bring in lots more players that's way more efficient than if you try and subsidize all of those other players individually so you might that might mean you bring in a marquee you know a marquee development partner or a marquee brand if like you have some sort of media sort of media platform or protocol it might mean conversely that you actually subsidize Demand right you know if you think about um a lot of consumer marketplaces a lot of consumer marketplaces started by subsidizing demand as a way of driving more suppliers to like want to shift more of their supply presence to the plat but okay so lot that's still true but lots of things are different so if you think about the competitive

framework right the competitive structure Web Two platforms were about or primarily about agglomerating inside of you know what we call a Walled Garden right sort of an enclosed ecosystem as many users as as you know and user data and sort of user transaction flow as possible Right the idea is sort of a network effect based around holding everybody in and making it very hard for them to to move out well if you look at web 3 where users control their data and digital assets and can take them with them seamlessly across the web you can already start to see that it's like very very different right you know sort of if your model is built on you know if

you're a social media platform and and in web 2 you would have kept all of your users you know sort of enclosed in your platform made it impossible for them to access sort of others web 3 very different right your user has their identity their content their posts and now they can move you know sort of seamlessly across clients um protocols generate Network effects through embeddedness and sort of through integration so they're being included in many other applications sort of you know have many clients pointing to them uh and also through uh what one of my co-authors and I refer to as as social cohesion right the idea that or Community cohesion the idea that users want to use your system rather than some

competitor and all of that points towards a much more virtuous you know sort of competition cycle right it means that Protocols are actually competing for users to want to use them it also means you have a a surprising form of cross sort of like cross Network complimentarity um if you think about it you know I was maybe you're a you're a platform that's going to interface with lots of different payment providers of various forms well you're a sorry you're a protocol that's going to interface with lots of different payment providers it's actually to all those payment providers advantage to have each other on the protocol right that's

really different right there's actually it sort of induces a bit of collaboration between otherwise competing firms and the reason for that is that if you're one payment provider and you're going to integrate a protocol you're going to do all the work to create sort of the the ability to like run payments through a given protocol then you really really want that protocol to win right it's actually optimal for you if that protocol that you have invested in using as part of your infrastructure as part of your rails becomes the winning protocol same thing if you're a client you know sort of bringing out you know sort of building on a a given decentralized social media Pro protocol like all of your development effort matters if and only if that protocol becomes a winner

and that means you actually want more development of you know sort of competing clients that means if you're a you're the payment processor you want all the other payment processors on the same protocol because that raises the probability that that protocol wins that all the users are going to be there and that's going to you sort of reinforce the value of your connection to it and that's a really different competitive strategy framework right it's it's just not the way we're used to thinking web too right like normally you you payment processors don't want to sort of like interoperate with each other here the protocols actually become these interoperability layers and they succeed by encouraging sort of global interoperability across the system more on that I'm sure in the uh in the event

this afternoon and and happy to talk you know in in previous csx's I've spent a lot of time talking with teams again you know either trying to trying to break things or or prefix things as you know as strategizing you know sort of how do you get the incentives at every layer of of the stack right um really excited to talking to you more Q last but not least Yim hey everybody uh my name is Yim uh I am the most uh recent addition to the research team uh uh at a16z crypto I joined actually over the summer this year uh before that I was a PhD student uh at Stanford um and um you know my sort of area of expertise in the area of like distributed systems and and

protocols um to give you a little bit kind of a better understanding uh sort of what that entails right so so when I was at Stanford um some of the work I've done is related to uh ethereum's proof of stake consensus protocol specifically um you know whether this protocol is secure um like uh you know working uh with people from the ethereum foundation to kind of uh look at earlier versions of the protocol uh finding bugs and then uh going back and for kind of making the the protocol more robust and sort of at the same time um while this kind of practical sort of you know improving the protocol uh is going on sort of developing kind of the um our understanding of um you know what is the

security properties that we want from this protocol um you know if we wanted to attempt like a rigorous security proof um uh you know what what would we want to show and and how would we go about this given this little bit of of like context um sort of what I do um the way I help portfolio companies and and you know people in in Industry more broadly uh is uh I would say you know primarily maybe we can say it falls into two categories um one has to do with kind of uh distributed systems kind of consensus blockchain protocols in the more narrow sense right so this this is about the following so you know traditionally if you think of like a blockchain right what what does the

blockchain give you well it's this kind of uh you know ordering service essentially you know sometimes we like to think of it as a world computer there is this sort of abstraction of a computer in the sky where everybody can send transaction and then somehow the validators that sort of run this protocol uh agree in what order they all want to um want to execute these transactions and then uh you know once this ordering is determined um sort of they they go and and and do the actual actual work and then we see all the State updates right and so um a lot goes into kind of Designing this ordering protocols um and you've heard uh you know earlier from Prav about like transaction fee mechanisms and kind of all the economic questions that go into

this but there is also kind of a more kind of distributed system/ security aspect to this which is okay suppose uh you know not too many of these parties that run the protocol um are malicious and and don't follow the protocol as we want them to can we sort of still guarantee that this protocol uh you know cannot be stalled basically and that everybody actually agree on kind of uh what this uh computer in the sky should be should be executing right and this is what we you know call the consensus problem or state machine replication problem kind of in a more kind of narrow sense right um and that's basically the foundation of of many many um like blockchains and l1s that we see in the space and um many companies kind of need

a primitive of this kind or something similar and so uh you know when I work with them uh this is about like helping them sort of understand okay what are their requirements um you know how many parties do they foresee run this thing um how do they determine who gets to run this thing uh how much throughput do do they expect from this thing uh what latency do they need from this service those kind of things and then and then based on that sort of uh make a decision kind of what's the right protocol for them uh are there like uh open source implementations that they can sort of tap into uh are there systems that they can uh sort of build on top of that's one part and then uh the other part is sort of a little bit kind of further up on the stack which has to do with like

uh for example rollups or um sort of you know infrastructure I would say that sort of sits like slightly uh on top of like l1's in the traditional sense right like um or even yeah like you know questions of like L1 design that have to do more with execution right if you have like execution uh logic or execution semantics that kind of in involves tees The Trusted execution environments for example um or uh you know snarks ex sank proofs uh or fraud proofs or some sort of optimistic schemes then how do you kind of integrate this uh kind of with the with the ordering protocol in the kind of more classical view um and you know how do you enable for example uh

you know uh trustless bridging and like all these uh kind of interactions you know with with other chains interoperability with with other uh systems uh um in the ecosystem so that is uh you know just to give you like some some rough idea sort of the the kind of questions that um you know I can hopefully be helpful with to you so I'm sure everyone in the room is sort of you know comes across crypto haters now and again maybe they in their friends or family or just at a bar randomly or whatever but I don't know if people in the room know this but like some of the uh most vocal crypto haters are actually computer science academics right um and there's a variety of reasons for that um one of them is that they think all of

the problems that we work on are solved problems uh and it is true I honestly like in the in the early days of crypto I think there was a little bit more of Reinventing the wheel there could have been a little bit more knowledge of for example you know consensus protocols in the early days now I think actually I think the crypto space is well aware of kind of the academic developments and I think Academia has always underestimated um the sort of depth of the challenges that we face and trying to build out this Tech so let me try to simulate such a person and see what answer you get which is blockchain's a madeup problem I've been running paxos in my company for two decades uh good uh I hope not as a blockchain protocol um no yeah that that that's a very good point so um yeah so

when we interact with like academics um uh in particular ones that aren't following the space as closely right uh sometimes we like encounter this kind of uh uh kind of person I don't think it's true right I mean uh obviously yes like academics have been doing this for for some decades um and they have uh you know developed like Rich Theory and like uh you know very kind of canonical kind of constructions and canonical techniques that are you know very like valuable for us to kind of not reinvent but to to have like an understanding what's already out there um but all of this is like sort of developed with like a quite a different kind of mindset about what sort of the application right like paxos for example right if you're

running like a replicated kind of database you know among like five data centers all of them in North America or something like this and you kind of control the environment you control the network you know you can sort of like reasonably expect that sometimes you may be facing a power outage and like one of these data centers goes down um but uh you know you're probably not like uh building these systems for like actual Byzantine Behavior right you're not building them at all for like uh rational players right like players that have their own selfish interest in mind and sort of you know one data center kind of looking at your database being like oh very nice but you know I'd really like to kind of influence the transaction ordering because I'm also you know maybe a seller on this

Marketplace or something like this um and so I think there are some uh you know very kind of genuine kind of uh aspects of the problem that that are really new uh uh with blockchains right um or with web 3 and this kind of Internet scale kind of setting right like if you think for example uh permissionless right arguably you know Bitcoin is the first of this kind of an ordering service where um you know you didn't have to like enlist into some sort of a service or like go and and and you know be allowed to kind of be part of it right everybody could just show up they had to kind of uh you know invest compute in order to show that uh they're not like making up a whole bunch of fake

identities uh but then you know one CPU one vote right um and today we have something similar in ethereum uh with like proof of stake right it's like one coin one vote um and that's not not really something that that people have uh studied previously in general like I think uh you know Byzantine consensus Protocols are sort of a an academic um how shall we say maybe hobby uh up until like uh blockchains come about right only now uh you know previously people are like okay nice protocols but like they have a huge overhead compared to like crash fall tolerant protocol calls so uh you know we're going to think about Byzantine and then we're going to implement crash fold uh but you know in

in in web 3 on on the internet that's that's not really an option similar question right if you think of typical kind of replicated State machines uh you know they're within one organization they're catering to like uh one particular kind of type of user right like they run within Google um their customer or other Google services or something like this for example now you know we're running ethereum and people use ethereum for all sorts of things right everything from like uh you know games uh on top of like rollups maybe in order to get transaction fees down uh to like uh you know large like whale payments that like are going into uh into exchanges and um you know there's

like a huge kind of heterogeneity in the user base and so uh you know it's really hard to design these kind of L1 protocols that are like kind of a one siiz fits-all kind of uh kind of protocol and that's why this hasn't been done before and and only like in recent years uh people have been making uh progress on on those kind of questions and so yeah I think it's really uh you know it's really exciting there's uh there's a lot of uh kind of new questions um where uh you know people come with this kind of genuine kind of blockchain inspired questions to designing these protocols and of course we're building you know we're standing on the shoulders of giants we're building on the protocols and the techniques that already exist but um you know there's it it's good for academics

to pay attention to kind of these new new questions hey thank you guys very much Mo the concept of ZK and what we do just our businesses we're essentially a way for AI agents to to pay for services on demand and one thing that I've seen um in the market are these non-blockchain applications of essentially zero knowledge coming to us the concept being there seems to be a problem where folks want to be able to sell these valuable data sets into these AI um in these AI businesses but they're concerned if they just deliver full data sets they're kind of done they'll um these other businesses will use that data and they won't give them any attribution Etc and so the concept that keeps coming up is like using ZK as a way of like obus skating the data but

able to deliver verifiable analytics to the service that is um trying to consume that so I guess broadly speaking like ZK outside of blockchain like do you see value there or is this um is this Reinventing a wheel that you know there are other solutions for that you think would be better suited I'm not sure like it definitely I could see using crypto to offis gate data and there's a pretty uh deep literature about different ways to do that and different types of solutions maybe this is a I guess everyone knows the you know the term ZK has expanded greatly from the original academic meaning of the term and maybe maybe ZK will just come to represent

like all of uh all of cryptography at some point but um yeah I hope I didn't just speak that into existence [Laughter] but but yeah no I mean uh I definitely there's like potential I mean you see this problem again and again that like party a has the data party B wants to do something you know and you want to give them limited access to the data to sort of just do what they want to do and not necessarily learn anything more and there's a bunch of different architectures you could think about doing that with MPC some people would say like FHA you could do like a limited release with differential privacy so yeah I mean I think there's a ton of value I haven't really seen like a very credible solution it would this is kind of what I was talking about earlier it

feels like we're still in the regime of people solving this by like having some trusted contract where they run it on some trusted server somewhere maybe they use sgx or some trusted Hardware maybe not but like as far as I know like nobody's really doing the full cryptographic solution to that yet so you know I I hope we get there obviously broadly stated I think that's a really valuable problem to solve yeah so in in general I think zero knowledge like the actual original meaning of zero knowledge can be uh the the guarantees it gives you can be a little bit of an awkward fit in ml applications despite a lot of excitement over it ZK uh protects the privacy of of the prover but in in a lot of ml

applications we would think of the pro as being you know some some powerful entity some beefy machine or something and um that's not the party whose privacy really care about protecting so it's actually other uh cryptographic tools that might be a better fit like MPC or FH or something zero knowledge proofs what you could do is like the the data owner could uh you know run some learning algorithm on its data right and if that data was committed in advance right the the data owner could prove it actually ran like the specified algorithm on the committed data right um that could be useful in some context but not in others so I think there there

definitely is going to be like you know meaningful applications of Zer knowledge proofs to ml but I think they'll be much more limited and targeted than sort of a lot of you know hype would suggest I mean a lot of things that you guys discuss there's so many advancements in in the recent cryptographic Primitives and uh we see there's a lot that's going to come what do you think about the tradeoff between using these New Primitives which are I mean very recently proven and they might have some uh Edge case or some assumption that might break them and we deploying it's kind of scary because we're deploying millions and billions of dollars of of assets on them uh so like what's your personal take on when should one choose the most recent uh cryptographic primitive and how should you

progressively kind of uh go about deploying money on it yeah it's a great question I mean I think this this comes up a lot talking to uh you know people in the space I mean cryptographers traditionally are extremely conservative so when they've developed kind of the great algorithms in cryptography like RSA or AES there was like traditionally something like 10 to 20 years between you know when it was first written down on paper and when it was actually used to protect high value systems in practice you know it's great that the the web 3 space moves a lot faster and people want to deploy something that was first posted on ePrint 6 months ago more traditional cryptographers that really panics um but I do think that we've I

mean so so generally the advice is like yeah you need to be very careful and think about what could break you know a lot of these systems like the mathematical assumptions might break or we might come up with a better attack more likely there's probably implementation bugs and most of this stuff it's very very difficult to implement cryptography correctly and that's usually where like 99% of the bugs are not mathematical bugs those you know those show up too but I mean I think we've seen a lot of the better projects in the space I think realize this and they usually build some kind of system to mitigate the damage you know if if an attack happens so you know even I guess maybe zcash was like the one of the first big deployments of ZK in the

space and they built a lot of things like kind of velocity limits in and out of the shielded pool so that if the proof system totally collaps you know you couldn't steal all of the the zcash in existence yeah and I'd say I mean the some of the cryptography that especially zero knowledge proofs that people are using today are vastly more complicated than these things like RSA that were rolled out much more slowly I think the community still can prioritize uh security and uh much more than performance um today I think the priorities are are out of whack just a small followup on that like it seems that the engineers and cryptographers don't speak the same language so like usually an engineer reading the paper and implementing it probably makes a lot of mistakes because there's details that

are not even in the paper do you guys have any thoughts on like how the paper should be written or structured or how there should be some supplementary there that could make it more clear have you seen similar problems well I mean this gets discussed in cryptography conferences all the time where everyone recognizes like details go wrong in practice but people don't get hired as professors or promoted based on the details just the concepts so it's there's been this incentive problem for a long time that academics are encouraged like as soon as they've sort of finished the proof they need to move on to the next project and not leave the details for somebody else so I don't know I feel like we've never come up with a good way to I I do think that the same issues keep coming up over and over

again um in in snark land um people mess up Fiat Shamir all the time in the same way for decades um there's another much more subtle issue about combining something called parallel repetition and Fiat Shamir which I keep seeing coming up and keep saying you know look out for this foot gun you're going to blow your foot off and then people blow their foot off and um so I think a lot of this can be addressed just by compiling like this is like a perfect topic for a blog post or you know there have been research papers that you know um categorize the the bugs that we've seen um and so that's probably the the right solution versus trying to get academics to like

not have any bugs in their paper as written I would say this one of the best things about the research team here is that like you know we have Justin who's a theoretical expert on ZK and has spent two years you know working shoulder-to-shoulder with Engineers building it so um you know hopefully that's like a step in the right direction should say that's very it's very rare to find that anywhere in the the whole broader crypto space hi uh I really was amazed by the conversation Tim and Scott had about Network effect and how web is different uh so actually we are building more like trying to attack one of the big Network Factor say like the Apple ecosystem with the new Apple intelligence right so like big players like Apple and Google they have

they own the device and then they own lot of their own data and they build a very high Gardens over it and uh like to use web three approach and say okay users you can just be like a bit torrent just download your own data and then create your own like Ai and uh we give you tokens to like incentivize you to download your own data and build your own personal system but uh when you look at reality you realize oh just the fact of delivering that message is the customer acquisition cost is so high you need to pay PR agency to broadcast and you need to go to all this Google Facebook transition media to let people know that you are even challenging them so it's like a in reality the word is so complex I was wondering like how do you

guys as top academics see uh for this kind of Market design problem on internet like uh what what research is to be done I just feel like we're challenging it but we don't even know like say if I'm solving optimization problem what is my objective what are my action space what are my state function so were you asking the question specifically so you opened up by talking about um the challenge of a specific Market where users kind of harness their own data we you mean challenges there did we asking a broader question question um like uh I guess uh if you want to challenge a web to network like a big players that already built up Network effect they want to use web three approach it seems uh that's a new

problem that different from what Bitcoin is solving right the monetary system problem good so yeah so Scott maybe I would frame the question as as a web 3 startup with these sort of uh big scary Web Two incumbents yep what's the strategy do you beat them at your own game do you play a different game how should how should Founders think about it I mean the first part of the answer is yes right you know the uh the hope is to do a little bit of both right to sort of play and and this actually goes back I mean if you think about the term disruption is thrown around like has taken out like a lot of meanings but the classical clay Christianson theory of disruption um is that you know you have you know an existing incumbent business and you have a new entrant that starts

by providing services to sort of you know that that are not as good as is what the incumbent provides but are good enough for some other use case right to bring in new users into the ecosystem and then gradually that becomes good enough also to sort of compete with the incumbent and and the the idea is that the incumbent is sort of never incentivized to directly compete with this new entrant because they're in a different business space right the incumbent you know sort of is is incentive to you know Su you know sustain its existing position with its existing customers disruption in that sense sort of comes from leveraging what you can do especially um often in a way that is like you know sort of different or opening up like a new market space but then you know sort of gradually coming to compete with with the main

space um in the particular context you're talking about actually a lot of I've been thinking a lot recently with a with a large group of collaborators led by one of my uh you know in large part by one of my former students and frequent collaborators Chad esper um about sort of incentives for people to exfiltrate their data right like sort of like why would people want to bring their data out of an existing platform where you like you have everything sort of already stored um and on the one hand that sounds sort of like you know no you know on the one hand it sort of sounds like nobody should ever want to do that right like you know your your ecosystem inside of Facebook or or um you know or Instagram or whatever just sort of exists and on the other hand um you know

there's a lot of potential right if you if you're a Creator or you're a business that like sells products on the internet and Instagram is one of your primary uh you know sort of sources of of customer leads you're tremendously at you know sort of the you know sort of at the you know algorithm risk sort of like you know and at the behest of the platform and your identity may be disjointed in many different places right so um you know for example I a lot of lot of the the YouTube videos i' I've watched recently have like a you know go to our patreon to like fund our you know continued production of YouTube campaigns they have two separate like disjointed identities in different places and so you know there's some value proposition to exfiltrate your

data to you know sort of move into this new ecosystem where you hold the data yourself and and can bring in from place to place but on the other hand that's not a very strong value proposition for the average consumer right like you know sort of maybe a business has a reason to like go and build a core identity on a new platform a consumer doesn't necessarily and you have to give them some incentive that's not just about owning your data for your your own sense of control because we've seen repeatedly that that isn't something that the average consumer seems to have a lot of demand for but if you build it into some other functionality um you know like you know take out your data so you can you know so so you can load into this new social media platform or take out your data so

that you can like run new analytics on it and so and so forth and make that processing as easy as possible now maybe you can bring consumers over for that purpose meanwhile on the like you know lean to what's special about um you know what's special about web 3 you know a big problem with the web 2 platform architectures is that a third-party developer is sort of always a platform threat right you think about like what the early days of like a lot of like web 2 social media looked like they had all these like third party platforms on top of you know built on top of them you know various games Think Like Zinga or like you know cool sticker creators or whatever but then at some point the platforms like closed off the ecosystem

and and even you know sort of to the extent that the ecosystems are still open the API very ex you know directly limits what you can do right and and also is subject to change all the time think of Twitter recently like closing off its you know s or or pay Walling its developer API in web 3 you can build a platform that is committed to open access right to sort of like being permanently accessible that is a strong you know sort of value proposition for a developer that wants to build on top of you and you can use those developers to like build value in ways that are web three unique right like for example um you know there are lots of like mini games on farcaster right now right these

are games that use the foraster protocol as a backend for uh you know playing a questing game or or battling you know there like far castles where it's like North and South Castle they're very as simple games but they're a really good feedback loop for drawing in new users that sort of activity right where a developer comes and can feel confident bu building on the protocol that like they won't lose their you know sort of game you know sort of back end is web3 specific creates a new consumer value proposition right like I have friends who like this is the main reason they walk on to farcaster in the morning right like they they open up the platform so they could play a game and then they see the feed and that enables you to compete with the likes of Twitter because Twitter doesn't have a way to

offer that right they can't commit you know if you build a game on Twitter they don't have the the structure to build a game on Twitter but even if they did right even if Twitter had a version of frames they can't necessarily commit that that won't change that's sort of like the the user you know like the Creator um and developer experience won't sort of like vanish tomorrow and so I think of is there's there's like you know push and pull or you know sort of tug of war with these Web Two platforms like you want to use the web 3 functionality for everything it's really good at right individual user like individually owned user identity um you know commitment to developer empowerment commitment to open apis you know commitment to data portability but the underlying reason

for people to build have to be true value propositions right the value proposition is not consumers now control their data it's when you have this interoperable multi-identity you can you know you can instantaneously generate a character to play in a miname that will be your principal social media engagement is that helpful that's super helpful I want to add a small followup I guess in w three there is a another uh advantage to do customer acquisition instead of like uber you need to give real money to do customer acquisition right but it's in the will boost your customer acquisition and you attract a lot of eyeballs that just come to earn maybe they put the data off just to earn the tokens but on the other hand it also uh forms a probably not very good code

start your community is like initially coming to earn and how do you shift that to people that actually want to like use the value and in meantime you also need to leverage the initial data to build up a value that is like good enough to attract the rich user that come for the utility giving people uh sort of like an established stake in the future of your ecosystem which could be many things right earlier I talked about like tokens as arcade tokens right if you're going to be a multi-game ecosystem just giving people a bunch of arcade tokens when there's only one game in the arcade they still might you know have you know use value that expands and becomes more you know complex and sophisticated as people bu build more games and you know put more into the arcade so like lots of

different ways you can do this but your point your point about attracting the right users and the concern that if you use a form of token or other digital good as as marketing um that you might get the wrong user category it's a huge issue we see with lots and lots of projects these days the key is what you really want to incentivize is the type of activity that grows and anchors your ecosystem right you want you know like and and and Mining behavior is usually like token farming type behavior is usually not that right like you know what you and so you need to make sure that whatever your reward mechanism is is as tightly linked as possible to the actual objective you have not some sort

of like approximate proxy of that right like you know total number of transactions on our infrastructure layer might be a really bad proxy for total number of valuable transactions you know sort of with with established users um and you can adjust the design of the system to try and tilt further towards what you really care about right you can you know there are simple things like you know you know requiring like very very long-term unlocks and so forth or you know initially targeting in a very like you know permission way sort of recruiting initial community members um or even just simply giving the token only the functionality within the ecosystem until that is fully established right so not transferable you know can't be sort of you know

pulled out of the ecosystem and and transferred to to others only you for a long time only it like functions as arcade tokens and then once people would rather play in the AR the average user would rather play in the arcade there are enough games there's enough to do maybe only then does it become transferable and exchangeable so totally it's a it's a huge issue but you're right there's also a huge opportunity here to like use the potential of future engagement in the ecosystem as you know marketing direct to Consumer right like if you take out the marketing intermediate by giving consumers something that that wants to that makes them sticky hi I'd like to ask question about client side proving uh so what we're doing uh a lot of other people do is like a execute like a game P2P in

offline andov outcome set on chain so one like a problem we encounter pretty frequently is that now the gas fee is Sav but we have proving fee I think the proving Cloud won't let people use it for free forever so and the natural direction is it possible to prove on the client uh but there's like a memory constraint all of that just one to uh get an idea like how far away from like a realistically proving some substantial on the like a browser all client side yeah I don't think we're that far away um this is one of the benefits of uh our initial version of jolt is uh using elliptic curve commitments um which are amenable to really space efficient techniques um so I'm hoping this uh is

fully incorporated into jolt in the next four to six months um and initially you know I don't know that like version one will be um you could run the Prov in a browser but yeah I don't think it'll be that long um so ballpark I'd say you know a year from now um I I think you'll be able to prove really meaningful stuff client side and if you expand the the space usage beyond what you can run in a browser you can probably speed things up up to another 10x or something but still it'll be fast enough um small space in in the browser for so do you mean like both the proving side and the

aggregation of the proof yeah I mean you know proof aggregation is um in some sense it's just another computation to be proved um and yeah so I think I think both could be done by very weak devices I'm quite optimistic about about this yeah oh thank you so just a clarification to this last question when you say reasonable amount of utation done in the browser is that like will I be able to calculate five cats in 10 seconds let's say reasonable expectation so I I tend to think these days in terms of um like proving overhead compared to just running the computation with no proof and um I think that um in a browser that

overhead will be um below 100,000 right so you know if your computer can do it in you know one 100th one 100,000th of the seconds your prover will run in one second um and it's very GPU friendly so you know if you have a GPU you can take another 10x off that or something and that starts to cross a threshold uh you know 10,000 fold slowdown for the prover where I think a lot of things that uh you know become make sense that wouldn't have made sense previously so that's that's how I think like 10,000 times slower for proving than just running and the multiplier would be different outside of browser yeah so I think with

if you allow extra space so I um it's the difference between uh elliptic curve commitments and hashing based commitments and um my my guess is that with the hashing based commitments um you they will be faster by somewhere between 5 and 10x for the prover uh but that they won't be able to match the the space usage um so I'm we might be able to get it to a browser but um it I'm not so confident about it all right thanks again so much everyone and look forward to talking more today and tomorrow [Music]