zkOps: The Future of Zero Knowledge Deployment Tools Panel by Aaron Greenblatt

foreign hi everyone thanks for joining us today uh we're here to talk about ZK Ops the future of zero knowledge deployment tools so since zkops is literally a brand new term that we just defined I wanted to go ahead and Define it for you here today um you know fundamentally we've had a lot of really exciting advances in theoretical zero knowledge tools there are new proof systems we're really excited to see the progress that's been made over the last few years in this area and there are a lot of very exciting blockchain applications for this new technology uh ZK Ops is about how do you deploy zero knowledge and actually use it at scale and so this is an engineering problem this isn't Theory and Engineering in theory have always gone hand in hand and so we figured we need a new term for it just because it's such an operational Focus to what we do uh sends zero knowledge proofs rely on a lot of heavy duty computation we have custom designed hardware and software to make zero knowledge effective at scale and then there's also a set of integrated software tooling to make this easy to build on now because there's a lot of compute I want to talk a little bit about web2 at this conference you know the discussion is that web 2 is all about centralization and centralized control over systems and that's true but web2 actually turns out to be a lot of distributed architecture so when you go to a Facebook you don't just talk to their server in Menlo Park you're talking to several different caching layers that are geographically closer to you and so you can imagine there's a Content delivery network with static content close to you then you'll have a local data center where there's a database that you interact with and then that database needs to get replicated back to Menlo Park eventually and so really this is just a distributed system but it's controlled by one party one thing that I think is underappreciated is that the database replication traffic for Facebook is intensive enough that they actually couldn't route it over the internet backbone they had to lay their own fiber to go between their different data centers because there's that much traffic going between them and so this is really a distributed system but it's one where you're being very conscious of where the data is stored who accesses it when and where you locate your compute so moving to web 3 we have similar problems in ZK where you have a lot of compute you'd like it to be near you and there's lots of data flying around it's the same type of problem and so at today's panel this is what we'll be talking about excited to introduce the panel here so uh sevi is a senior cryptographic a senior cryptographer at semiotic Labs semiotic is a core developer of the graph and has received a 60 million dollar Grant from them at semiotic Savvy specializes in using zero knowledge proofs to upgrade the graph protocol prior to semiotic he was building secure systems for the US government at Sandia and then at Northrop Grumman drills is also a ZK engineer at Elio he worked on augmenting Marlin with lookup arguments and at scroll he worked on sequencer infrastructure Jules has also been deeply involved in the Supernova recursive snark algorithm and has created an open source implementation that's available on GitHub um chares I'm sorry if I'm killing the pronunciation he is a smart contract developer and back-end developer and he has experience building efficient and secure algorithms protocols and smart contracts for several D5 projects He previously worked at paraswap where he was responsible for developing a significant part of the core aggregation algorithm finally Mikko is a managing partner with Gumi crypto Capital 100 million dollar blockchain Fund in Silicon Valley Mika was also the chief developer evangelist for Java at Sun Microsystems and brings deep expertise in developer tooling to our conversation so with that let's just start the panel discussion um I've come with some questions today if you have questions towards the end happy to take them let me just start off kind of basic because this is probably a mixed crowd does somebody want to give like an overview of ZK tooling today and sort of um yeah just what tools you use and when and why so are you turned on take that that's the song check check there we go hello all right I guess I can take that yeah um yeah I've worked with a lot of different kinds of tooling over my uh over my career I worked on the Elio cryptography Suite I worked with artworks I worked with like Bellman the Z cash crypto stuff um there's a lot of there's a lot of different there's a lot of different kinds of tools out there uh and the one thing done closer the the one thing that you really notice when you work with a lot of these different crates is that they're very fragmented right they're very like vertically integrated so the uh all of the infrastructure all of the curve arithmetic all that stuff is very like tied to uh the different crit you're using so there's no interchangeability anywhere um and that makes it so that the crate that you choose is very context dependent right so when I worked on the implementation for the synchronova protocol uh originally because I enjoy working with it I wanted to use the zcash cryptography Suite because it's pretty easy to work with it's really like easy to reason with the code's pretty readable very quick but I quickly ran into kind of an issue where there's not a lot of extra peripheral gadgets there's not a lot of infrastructure I was kind of missing that so I had to Port over to artworks because artworks is one of the most built out libraries out there right now which meant I had to rewrite almost the entire repository right that took me like two days um so there's a lot of fragmentation of this ecosystem and deciding on one thing or the other is pretty difficult um because it's such a context dependent Choice um yeah all right I run it I run into exactly the same problems like right yeah it was part of the dark Works they got all the curves you can use this like another just a little Point like you know with this vertical integration it's like a lot of these proving systems it's hard to figure out the uh the actual efficiency and like the actual performance of this proving system right because there's vertical integration and there's probably a reason for it you know like this this preview system is very efficient over this particular curve with this particular field um and so it gets really hard to compare proving system so what are you going to use for your particular applications and stuff like that too so yeah hello yeah so I mean for example the circle right that's such a bad debugging tools like they don't have any like if you want to like ride any circuits you just have to write log statements right so it's um yeah we don't really have a lot of tooling to actually develop big systems like yeah for what we have now hello hello yeah sorry okay it works uh yeah I mean I think we're still super early in tooling right and one of the things uh so we're actually investors in Iron Mill and our thesis around this is is that there's always in uh developer tools I've been in developer tools since Java there's always been a Dev and an Ops right and they're not actually the same devops isn't a thing right and when you want to say I'm building an application and I'm scaling it to one billion users right like it's not just like okay here's the here's the algorithm then run right like it's much much more involved right so you know I think we're really excited to see emergent tools like uh circum like it's super easy to kind of like for non-zk researchers to start building circuits so like that's great but like that's the surface right like when you actually get to kind of like hyper scale prover infrastructure that's when you need kind of like big Iron and you need things under the hood like fpga a basic like you just need deep deep kind of like optimized compute and special purpose compute facility hello okay um okay thank you um I guess next question and since Miko I know you have another panel that you need to run this is mostly targeted at you how do you think about driving adoption of ZK at scale I mean what type of tooling do you think needs to be built in the near future so that everybody can start using ZK in their projects yeah so I think some of this becomes really really in for deep infrastructure right so if you look at some core applications you know core applications include things like Oracle right so there's a bunch of complex Oracle things that are cross-chain verifiers right so like the infrastructure developers are actually kind of okay right like you're seeing guys on stage here who are basically like well we can just rewrite stuff or we can just go low level and just like freaking do all of it right so it's sort of like the tools that make this easy are actually you know for the infrastructure side when you look at applications like ZK roll up right so ZK evm when you look at things like ZK Oracle or when you look at Bridget like the developers are super sophisticated so the tooling actually doesn't need to be that sophisticated right so I would say that like for the other applications you know like things like privacy and identity like we're definitely going to see more things like circum right which actually gives you like its rust based but it gives you sort of a JavaScript interface to sort of build circuits right so I feel like this is kind of what we're going to see I think next that will kind of create ease of use cool awesome um yeah so thank you um I want to ask the rest of the panel it feels like there's a lot of progress in recursive proof algorithms especially over the last six nine months even and um what do you think the future there looks like because on one hand you have recursive proofs that allow tremendous new efficiency gains on the other hand it's a recursive proof so you're generating more proofs how do you think about like where this is headed and sort of what do you see as the future end state of recursive proving systems yeah I'm definitely more of the opinion that we're going more closely towards transparent snorts so something that doesn't require a setup I think is uh just Paramount there um I did a lot of work on this at Elio actually I like I helped run the trusted setup ceremony I wrote a lot of infrastructure to work with the powers that we generated and on paper these things are pretty simple like the security is basically the same you know you're not really thinking about it but it's a difficult engineering challenge to work with you're working with a file of powers that's like 25 gigabytes long you can naively load that into RAM and work with it but that's obviously not going to be efficient that's going to be like terrible for your for your resource usage um and you're trying to you know you're trying to always be more decentralized so you're thinking about resource constraints right so you have to build a lot of extra infrastructure working with the file system this tends to be unsafe um it's all hassle to deal with and there's a trade-off obviously between that and using a transparent snark where proofs are like a lot bigger and you have a lot more State load but there are ways to amortize this batch improve sort of like that kind of thing um yeah I feel like that is definitely the future in that direction so yeah I agree I think it's really interesting I mean another kind of like I feel like uh I don't know standards popping into my head but it's not a standard but like there's a a formula that people are using right where you can kind of aggregate proofs now right so you have a like a one-step recursion where you have like one privy system which can improves your big statement but that's got a big proof but you want to get that proof up on the evm so you want like a small compact group that can be efficiently proved so then you hit that so then you have like a one-step recursion that uses plonk uh to basically prove the verification of your of your big proof and make it efficiently verifiable so I mean it's a really it's a cool concept right there's a lot of interesting things that I think people can use with this recursion um hello yeah so one thing I want to add about like recursive approaches like these days a lot of papers are claiming to use recursive groups to like have really high efficiency gains but it's important that they actually Benchmark these like how much recursions you need actually because sometimes they just like say okay you can have thousands of like compute machines but then they don't have any benchmarks in the cost itself and yeah like I mean I think recursive proofs are very good especially in blockchain setting because you have this like incremental model all the places right so I think it's going to be very good but I think we need more concrete benchmarks of like new systems which claim to use recursive proofs or like really high scalability so yeah I think yeah and also it also relates to iron Mill right when you have like fpgas and stuff right you can really paralyze these things and have these efficient games so yeah cool yeah thank you um I think on the topic of sort of recursive proofs and I think the topic of memories come up a few times you know one of the interesting things on web 2 is that we have a lot of um resource constrained nodes okay yeah and yet even very efficient proving systems you'd like to have 32 gig of RAM on your node how do you think about deploying out to smartphones to iot to Really the world of the internet given some of the hardware requirements and especially the memory requirements of proof generation yeah I I think one thing so I think what snakes come out here you're okay we've got a solution it can just prove you know NP statements this is great we can just throw this in the circuit we can kind of prove everything in ZK and I think what we're learning right and especially when we start getting this resource constrained so we have to start get we have to get really clever about what specifically do we want to prove a CK right because we want to minimize circuit size we want to do we even want to use this maybe we want to use a proving system that doesn't require a circuit right maybe we just want to put this one little thing if that could be ZK the rest of our system is fine so I think I think a lot of the features is getting really specific about what exactly do we want to prove a CK versus like just throw the whole thing in a circuit and be done it's like thinking about smart contracts right like you don't want to put things on Smart contact it's very expensive you want to really find what you exactly want to put on the like trust like what you want to give out the trust from right and then you just put that particular logic in ZK I guess to like really yeah uh I think from the perspective of uh like really really light clients I think it's pretty interesting I ran across a project called Kevlar that's like a Ultra ultra light client and it's basically almost like an RPC proxy and so you're definitely going to see things like that at the edge of the network anyhow Circle um I I can add about Kevlar so I created Kevlar right so kevler uses something called as previous stake in sublinear complexity it's not using ZK it's one of the things where we could build a very efficient like kind without even using ZK so when one of the arguments about zika is to not use ZK in the first place and come up with something better without ZK um yeah cool thank you um I'm saying we have about two minutes left so I want to open this to an audience question does anybody have something they want to ask yes there we go hi I have a question for Michael uh gummy crypto what yeah uh what do you find that the most exciting and the for the Futures development yeah so uh as far as the Venture investment Community like the most exciting applications in the ZK area really are cross-chain verification like cross-changing verification for things like performance obviously L2 scalability right and then of course the other applications are things like Oracle like for example hyper Oracle type of things right obviously price Discovery oracles maybe like a LaGrange or like things things that kind of really focus on that obviously we're super excited about iron Mill ZK Ops and actually like special purpose compute and then of course finally bridging right like I'd love to see and again that's also cross chain strong proof so like we're super strong we're super happy with that whole class of applications because you know we're getting like strong on-chain cross-chain proofs horse smart contract developer who wants to get his fee went right now on zero knowledge proofs what tools would you recommend that have on-chain verifiers circum for sure come as yeah circum has definitely if you write any circuit on circum you can get a smart contract verifier for that um and I think Halo also has it yeah yeah yeah yeah and then also there's uh so Punk so you see like plunk and it's like it's variance it's got a very efficient verifier on chain so I think that's why you see such like there's a lot of tools available for that proving system um and so I think like espresso systems has a pretty good uh like plunk ecosystem and I think they actually have a smart contract implementation too that you can just kind of like take and use that verifier all right um I want to thank all of you I see that we're out of time um I think we'll all be around after the talk if you'd like to ask more questions thank you all for coming today and enjoy the rest of the conference [Music]