Why 2025 Is the Year for ZK Proofs

Transcript

no I would not take that b i I tend to agree with Jeremy on that one I think really the question is can you prove an entire block and less time than it takes to produce a block if that's the case why why were we using prime fields to begin with why didn't we just use binary Fields right out the gate that was one of the first questions we asked when we started looking into binary Fields well yall have given me a lot to think about welcome back to the rollup your home for technical talks and open conversations that shape the modular future nothing on today's show is to be construed as Financial advice you can see our full disclosures list at the roll up.co Ventures sit back relax and enjoy today's episode after a quick word from some of our awesome Partners imagine if the world's GDP

moved on chain guys say GM to safet the newest protocol by safe isn't a blockchain and it's not a new layer to it's a transaction processing Network built to send money anywhere and swap anything on thousands of layer twos from a single Universal balance in sub 500 milliseconds it's fast it's smooth and it's simple and it's built with safe multisig level security in mind safet unlocks smart accounts across multiple chains allowing users to supercharge security scalability and speed so get started today at net. safe. Global and remember to follow at safe onx GM welcome to the show Jim your first time

on Jeremy welcome back um two guys that are very very in the mixed and experienc to to speak on the state of ZK um not going to put you on the spot but there's just been so much acceleration so much progress on multiple fronts here um we'll talk we'll get into a little bit of exactly what you guys are doing in a in a in a bit um but generally I wanted to start off with kind of just wrapping our heads around the latest in ZK maybe you guys can can give a comment generally um about what you're seeing out there and then you know we'll we'll talk a little bit about that before getting into to uh the partnership between risk zero and irreducible yeah absolutely I I mean I think this is a very exciting time for ZK in general uh I think we've seen

things move from the theoretical to the Practical and now to the it very usable even from an economics perspective we've moved from a world in which ZK was slow expensive and complicated to a world where it's really easy to use and at this point we cross the critical threshold where it's actually less expensive to prove something in ZK than it is to pay for the cost of doing verifiable Computing through consensus right so we've entered I think into a new realm and what's very exciting to me is that the cost of ZKA is continuing to drop precipitously as you know everyone in the industry is working hard to uh

make things faster cheaper and of course you know the work that we're doing together with irreducible will hopefully bring things things down even more but I I think it's very exciting to see where we're at yeah absolutely agree and I think we're approaching a point where there's a little more stability in terms of the techniques there's always new research but we really have very practical systems now uh I think about ZK in terms of uh as analogies to to Computing where like when zcash was around they were building these custom circuits which would be you know the early days of computing you were making custom circuits and now with with risk zero and the ability to just run arbitrary code this is like the CPU era of computing you just run write programs and you run

them and it opens up uh you know a whole bunch of use cases and I think the the next generation is when you have Hardware acceleration for ZK and that brings us to you know the equivalent of gpus for the gaming Revolution it it's incredible and I think it really requires like a paradigm shift in the way that people are thinking about blockchain and and their mental models here because you know right now submit a transaction you know it'll get broadcast it'll get picked up and then broadcasted to the validator set on ethereum and then it'll get re-executed a bajillion times that's a mathematical you know technical term and and that is all incurring cost all of

that re-execution is incurring cost um and ultimately we bear as as users we bear those costs now ZK flips this on its head and it and it says we don't need to go and reexecute all of this this this is like the original like satoshi's Vision maybe he was the one to say like the don't trust verify right like one proof and then the verification is so much more concise that we can we can you know still hold those same ethical like practical values without going and spending a ridiculous amount um and I I I I think that's incredible and I think more and more people will wrap their heads around it more protocols and applications will come to

the realization that these things are not prohibitively expensive as as they once were and ultimately we're going to see this it'll it's slowly and then and then all at once um and we're already starting to see it yeah and one thing I'd like to just throw in I think in the world of consensus you have a block and the block has a fixed size right there's only so much room within a block for computing so as the demand for execution goes up the costs go up because you have a fixed Supply right and with ZK what's really interesting is that you can verify an arbitrarily large amount for the no matter how much you're doing it takes the same cost to verify and so what what you can see is that you move from this like scarcity mindset of block

space to this sort of abundance where no matter how much compute you want you can just bring more compute online in an elastic Manner and the cost per compute doesn't go up right you don't see this sort of like you know gas costs Drive rapidly up when the blocks get full and in the case of ZK it takes you know you need to bring some more machines online but those machines don't cost any more than they did before and so you have a much more stable cost for computing As you move to sort of ZK verifiable Computing versus consensus based um and I think you also you know you can reduce things like the fragmentation right because one of the ways that scaling is solved now is you just go do it somewhere else the different consensus remain at another L2 and a roll up and I think that will

start see you know the impact of ZK in terms of this not only reducing cost but increasing sort of the scale of things people can build right like the complexity of their applications which I think is gonna be very exciting absolutely and and Jeremy you know you were you were on recently but I I don't think we got your take on this snark ification of ethereum um is that something you feel comfortable sharing you know some thoughts and ideas maybe spicy take if you have one I I I don't I think I'm pretty much in alignment with with where things are going by and large I I mean I do think that uh you know we will continue I think at where we're at now it's kind of inevitable already that most things that are being done with

consensus will at least be done with ZK as a as an extra bit of security um and I do think that long term we're going to see uh the idea of most State transition functions for most every train will be verified with DK um that's that's sort of my general belief yeah I think it's inevitable as Jeremy said it's blockchains fundamentally are platforms for verifiable computing on shared data and the way they implemented that before was through re-execution and so now it's it's pretty incredible that we were able to take this primitive from from the academic literature from the theoretical cryptography world and make it

concretely useful uh and able to fulfill it its purpose for for these systems yeah it it is I mean we're turning you know the the theory is accelerating so quickly which is which is really fun and then we're also applying this in a very practical way which is SC scientists and and theoretical scientists mathematicians like that's the dream is like to actually solve a problem and not just solve it but actually see the solution put into practice in the real world and help people is is a magical feeling um and and so yeah I mean I I'm also here because of the ZK movement and the acceleration I think it it is only inevitable uh that we see this uh integrated into mostly every chain and

and so it becomes a question of how exactly will we integrate it and Jim I think you guys are doing some really really cool things with Bas like you know this is something that's maybe a little bit still under the radar and not as um not as well known in in the space so maybe you could give us an introduction into irreducible and bineas and talk about why you're building what you're building happy to um I guess starting with like ZK it's been recognized that this the potential of this technology is exactly what blockchains need and the thing that was holding it back is fundamentally performance I think historically there have also been issues with developer

experience but I also Trace that back to a performance issue where if you can uh if it's really cheap and fast to run then you can uh sacrifice some performance for usability and still have a good experience at uh at the end so our mission at irreducible is to essentially eliminate performance as the barrier for ZK adoption we believe that Hardware software Cod design is fundamentally the way you make slow things fast and so we have a really strong Hardware team led by our CEO Rody and you know I have a more of a background in blockchain and and cryptography and so we were looking at what are the hardware friendly ways to build ZK snarks and we identified binary

Fields as this really overlooked um primitive that it was sort of now now was the time to to bring it back out and binary fields are a little esoteric to people who aren't deep into math and cryptography they're basically a mathematical object that's extremely efficient in Hardware that has some really remarkable properties that's already used widely in deployed uh cryptography so it's used in the advanced encryption standard standard they're used in QR codes they're used in uh AES uh authentication mechanisms so they're well known within the cryptography community and we kind of did the research to identify the way to

make the most practical for the the current ZK movement and uh and so we're really excited to be able to take the next step and productionize this with with risk zero yeah I I think if you know when I first saw the Bas work it was immediately clear to me that the impact of this as you move toward Hardware acceleration was going to be huge because I think it's very much night and day in terms of the sort of things you can uh do with binary fields in terms of Hardware implementation and so uh it I think it's a pretty impressive work and I I want to try to put this into terms that our audience can understand

and you know one of the most I I guess one of the most attainable or easy to understand simp simplest uh concepts of of the ZK proof life cycle how you know you have supply and demand and and you have a um a chain perhaps that is requesting a proof of maybe the state of another chain because it wants to interoperate with it and so it requests the proof of the state of that chain um and approver comes in delivers you know this is the the state of that chain and I have a ZK proof to prove it and then you know that that can get verified um and so I think one of the pretty

reachable in you know Concepts is this proof Marketplace idea where I have a Marketplace of provs um and and I have uh this demand side of chains that are requesting these proofs maybe chains are applications and then I have you know this network of provs that are that are all kind of like looking at the bids and looking at the proofs that are out there and I can submit a proof to to earn uh one of these bids that that uh someone will pay for that proof um now is that still the model here with risk zero and with Bas is it still like you know we're we're kind of like organizing these provs into a nwor and then these these networks can kind of

pick up the bids and Supply proofs to their liking um we just have a more optimized way of providing those proofs these provs can get even better at doing what they're doing Do Do it For Less cost and ultimately have a more affordable faster experience for for the demand user yeah I think you know our goal with boundless is really to make that whole process very seamless transparent for everyone in every chain you should just be able to request a proof and then it should just get done and you should you know pay for it but like to to minimize the complexity of that process but under the hood there is in fact this sort of Market there's this

bidding process and then the really the source of most of the cost when you consider a user who wants a prove to be delivered the source of most of that cost is related to The Proving and how expensive or inexpensive the proving is and with you know uh with irreducibles work the really the goal is to reduce that cost of that proving I mean that would be true regardless of how it was deployed right if you ran your own um you know provs or if uh a protocol like builds their own proving Network however however it gets done if that cost reduction still exists uh but you know of course our biggest interest is of course to make sure that these kinds of cost reductions are

available on boundless so that anyone can you know get the savings basically and another thing that's really powerful about boundless and that that Marketplace model is that you can use the same proving technology across a lot of different protocols uh and all these protocols that may have different purposes can all leverage verifiable Computing and leverage the same underlying technology uh and and proving and uh you know proving suppliers to do it yeah and I mean there may I if anyone's still listening they're not convinced on like the the purpose or power of ZK like just take a look at you know the things that are out there that

are using compute and Nvidia is like you know one of the most valuable companies out there you have all of these compute providers that are you know that represent big Tech like all of this compute is being used in in various use cases AI right is is a major major one of them um and the the verifiability of this is so so critical um and we haven't been verifying a lot of this comp compute because you know it has been it has been prohibitively expensive but with the work that's been going on and with the Innovations in in mathematics and Hardware we're bringing those costs way way down so Jeremy could you just kind of sound off and maybe just fill in

any gaps here because I do want to dig into a little bit more on the binary fields and the Bas acceleration that that's at play here um but before we get into that could you just give us like a high level reinforcement on why verifiable compute is so important yeah well I mean I think if you look at the entire blockchain industry really the really key features that make blockchain valuable is the sort of fact that there you're operating over sort of public uh agreed upon data sets and that the compute is in fact verifiable and I think that that moving that verifiability to something that has a much better scaling properties fundamentally both in terms of the the base cost is lower but also in terms of this ability to arbitrarily increase the

capacity without having to sort of you know go to a new chain or deal with all these bridging complexities that we we currently live with U I think is going to be super huge but I I actually think that to your point you're going to see a lot of use cases that are not would wouldn't be done traditionally in blockchain because of limitations start to make more sense to verify right um you know think uh you know like you know working with hash for example who's building sort of a it's effectively a model that's a centralized exchange that's very high performance very easy to use but instead of just trusting the uh exchange you actually have the exchange be verifiable and uh you that's

done using ZK and it allows you to have a lot of the same you know advantages you would get from centralization but without having to trust the exchange in the in the way that you do and I think that a lot of things that are being done now using centralized trusted services will move toward verifiable Computing and we'll really see the like com like we'll see we'll see as we see we'll see the web 3 blockchain world have increasing capabilities so the capabilities that it has are much closer to what you see in the web 2 world and we'll see I think a more emphasis in sort of the web 2 world on verifiability and so I think they'll sort of come together over the long course of time and really as the cost of verifying

compute goes down there'll be lots of applications where you're like what you're not verifying the compu what are you doing right so yeah and ultimately that makes a much safer system like we're gonna have less FTX like like debacles because those things are verifiable I mean heck I I'm based in the US I think you know you guys might be as well um you know Elon said the other day he's thinking about putting Doge on chain like the idea of the entire accounting for the US government in all of its and all of its debt all of its you know allocations to different programs and departments to to actually have that in a verifiable

environment to to know that you know I'm spending taxes I'm paying the US government um where is my money going right now I have zero clue I don't know if it's going to some military funding program or or going to some you know Garden that's gonna get built on a on a public playground right like I I don't know so to to have like some some actual insight into where my money is being spent not just mine but how all of these things are adding up you know we it it's the open source movement all over again will have more eyeballs and more verification on where the resources are going and hopefully optimistically we can get the

wisdom of the crowd into that and and we can have a little bit more accountability as far as where things are going um and it's not it's not just you know the resources of of a nation state or a government but it's all of you know things that are considered valuable whether you want to consider you know resources and money valuable at at the state level or you want to take like an AI project AI model and the weights and the tokens um and make those verifiable as well because they're valuable all of that can be done in a verifiable way if the costs of proving come down such that you know it it it is still profitable uh to make to verify the compute um and so again you know Jim what you guys are doing at irreducible

is is taking on a new type of ZK process um you're going from and this is where I'm going to start talking like a little bit over my head but I'm gonna do my best and please please correct me like previously we were doing finite Fields if I'm correct and we were Computing over finite fields to generate these proofs and that gave way to some of these proving systems like Starks and snarks okay now we're going from finite fields to Binary Fields please tell us what that means and why it's important sure so to be clear uh F binary fields are a type of finite field but they're they're pretty special so previously all of the systems used another type of

finite field which are prime fields and they're maybe a little under easier to understand um but binary fields are are really powerful for a few reasons as I said before they are already kind of a standard in cryptography because they have noteworthy Pro properties they map really well to uh binary systems which are you know computers are binary systems so Jeremy and and the risk zero team have built these systems that emulate a regular CPU architecture which is risk 5 risk five is a it's a computer system that's meant for create for instantiating with a binary circuit that has zeros and ones in it and all of the

the registers and the memory addresses have 32bit uh Banks and so when you work with prime Fields there's a pretty fundamental mismatch where you have this kind of weird mathematical object that doesn't quite fit into bits but you need to use that to emulate a digital circuit that's fundamentally created for the world of of bits and and Boolean circuits and we think that binary fields map way better to this uh for a few reasons you have these binary Fields can be a perfect number of bits right we have some that are like eight bits which is like a bite so you can take a bite of information and perfectly encode it into a binary field or a 32-bit integer which is your kind of fundamental register

type in Risk 5 all the way down to a single bit the BAS system can efficiently operate with single bits uh and that was one of the properties of our research work that uh really made this you know the benefits of this approach shine so we think that basically the translation the overhead of translating a system like Risk 5 into the binary field world is a whole lot lower than what it took to to do it with prime fields and that's why we're excited about the potential speed ups that we'll get from doing this yeah I mean I think another way to State it is that Prime Fields while they're mathematically uh you know very pretty thing they um they don't really it's

it's hard to translate things built for the digital binary world into sort of the traditional Prime field arithmetic that's used in these proof systems and similarly that Prime field arithmetic is also not super easy to build into digital iic under the hood right so you've got you've got something that looks like a digital circuit which is the risk five circuit over here and then you're putting it into this strange you know Prime field world and then you're converting back from that into the actual digital logic that runs it we're running on a GPU or a CPU and so if you can make things in binary it's both you get less overhead it's much more natural for both you know taking something

that's meant to be a binary system like a risk five and then also for how do you implement that resulting thing in Hardware so I think it just gets rid of a bunch of the impedance in this match a bunch of the places where the the parts take a little bit of extra work to put together and so there's a big efficiency gain you can get by by by doing it that way um which we're very excited about if that's the case why why were we using prime fields to begin with why didn't we just use binary Fields right out the gate it's a really good question um that was one of the first question questions we asked when we started looking into binary Fields um because there were some early papers that were very you know relevant and influential to the systems that

everyone's been building that uh did use binary Fields back in in 2018 and 2019 um and we you know on my team uh myself and and our researchers looked at this and it sort of seemed like the trail went cold like there was this mystery when people were talking about binary fields and then and then everyone stopped talking about binary fields and switched to Prime Fields um and this is for a few reasons um one the benefits of binary fields are extremely apparent when you're building custom Hardware but when you're running things on a computer system that has you know accelerated instructions for integer arithmetic the benefits aren't as immediately apparent

to like a regular software developer um so that's one reason another is there are some mathematical challenges that you need to over come in order to use binary Fields frankly they're they're weirder like if you haven't studied math they're not as Familiar of a concept as you know the integers modulo prime number so those are you know some of the things we had to work through um but when you when you get past that you see the substantial benefits I mean one one data point that we investigated on on our team is we looked at the efficiency in creating an Asic for binary Fields versus

equivalently sized Prime fields and found that when you take into account the clock frequency that you could run a chip at and the number of gates required it's around a seven times uh efficiency gain to use a 32-bit binary field as compared to a 32-bit Prime field so those are the things that were kind of fundamentally motiv motivating us to say okay let's figure out how to make this work does that translate into a 7x performance Improvement is that a one one comparison it would I think potentially it's more than that because there's like outsized advantages to fitting more logic onto a single chip if you need to do more proving work and you don't need to communicate between two different

devices um you can have like an outsized performance gain so but yeah like at a first approximation that does essentially translate directly to to proving performance and efficiency love it and and me from the RIS zero or RIS you know ZK VM side of things like what is the integration process here is it just a matter of is there an upgrade process is it integrating to Hardware like generally how does you once you see this kind of performance in a theoretical like simulated environment how do you go and you implement that practically yeah so I mean I think with what we're doing right now is so we have built the sort of circuit the set of constraints that emulate the risk five and we've built

that um on top of existing proof systems um and we need to sort of go through those set of constraints and adjust them to make them work in sort of the binary uh field version of the world um and so largely it's really sort of taking and doing a revision of our risk five circuit to make it work in that new world I think what's interesting is actually we're actually taking this opportunity as well to sort of take some other really great ideas that come from the irreducible team they have this new arithmetization M3 and it's sort of a different way of looking at the problem of how do you build uh a circuit in a stark B way that sort of decomposes the problem in a what

we believe is like a nicer uh more it makes it easy circuits the circuits more flexible more powerful and so we're actually I think going to support both our existing Pro system and able using this sort of new way of looking at how to build a circuit so I think the next rev of our risk five circuit is going to um sort of use this new approach which actually even on the even on the prime field side gives us a number of performance benefits because it allows you can reuse certain computations more easily so rather than decoding now I'm getting a little overly technical here but rather than take taking a decoding an instruction every time you run the instruction you can

actually decode it once and then reuse that coding over and over again um sort of mostly for free based on this sort of new sort of arithmetization model that comes from the irreducible team so we're excited about it across a couple of different boundaries but effectively the short version to answer your original question is is we just really need to sort of re-enter the risk five circuit somewhat luckily we have a lot of experience we're actually already on our sort of third generation of that circuit we've learned how to like you know build uh risk five circuits really efficiently at this point um so we're uh we're definitely uh excited about this next one Redstone is def's modular Oracle supporting over 100 clients on over 50

chains offering 1,000 assets Redstone is the leader for Bitcoin staking and yield bearing assets they offer the most robust customizable price feeds available in any model join the Redstone exhibition [Music] today yep I I saw something the other day it was like uh I think it was a it was an experiment that someone had done um it was uh Yuki from Cella Labs that posted this and I think it came from a from a report it looks like by pi squared where they went and they they

analyzed all these different provs AC Ross some some different uh proof types it looks like um or excuse me some different some different experiments in terms of logic so they they went from like basic logic very small to to a very large data set uh of logic um and then also they used ERC 20s from small to large and it it looks like Risk zero came out on came out on top on for uh Yuki say six out of eight of those tasks um and then there's also a chart below that kind of kind of lays out like you know a comparison when um I I think it's the size of the data set when the size

is relatively small um risk zero does quite well and then you know there's maybe some different optimizations that that could be done or things that are done a little bit differently as the data set gets larger I guess I'm curious like you know when you guys are are thinking about like the way to optimize and maybe there's tradeoffs and ultimately there's tradeoffs how how do you guys balance that what what is the reconciliation process or you know when you guys are looking at a set of tradeoffs how do you guys make a a decision on on which way to optimize so so I would say that um the the short version is that it's a

little bit of intuition but generally uh if I have to pick one thing or another I tend to ask myself which one would make proving ethereum run faster and that tends to be sort of my general nth star uh in the sense that it's a good representation of your sort of What kinds of problems are being done in the where the near-term used cases for ZK are um that said I think one thing to note is it's very rarely about uh tradeoffs it actually tends to be more about the we've seen this like I think almost every quarter there's a you know at least a three or 4X speed up of ZK systems across the board and that's not

just with our system that's you know industrywide right we're seeing everyone's making their stuff faster all the time and I think that one of the really things that's interesting is um you know where things are now it's almost hard to keep up with the benchmarks right like uh like it's like oh wait they just upgraded to the next version ver oh wow this thing now runs this system now runs two times faster or that system runs 30% faster right like and so I I think that you're going to see a lot of sort of back and forth in terms of performance numbers um as everyone's driving down toward where the bottom is and I think from our perspective we're really focusing on you know two different fronts right one

front is what do we do to take the existing system that we have and to like speed it up here speed it up there and there's like 20% bump on this and a 50 % bump from that but also sort of the bigger question of okay over the course of like a year or two years where are things going right and I think that so in that regard like I think you know that the project is irreducible is more of a significant bit of engineering um that's going to take a little bit longer to get done but also will be yet even bigger step function in terms of performance right I think eventually we'll probably hit the bottom but still got a long ways as an industry because

there's not only the improvements to the risk five system there's the improvements to the proof systems themselves there's improvements to the hardware acceleration and building them closer to Hardware more modern process right um I mean at some point it's got to not be a 2X every couple weeks but you know as it sits there's just there's a lot there's a lot of things to improve there's there things that are on our list that are like oh well that's a 10% performance Improvement we'll get to that when we're out of all the easier ones right right's a yeah or the ones that are 50% improvements those are going to take priority we're looking at Mor's law in the rearview mayor yeah um it's incredible um and yeah I mean you know you say this is a race to the

bottom in the best way possible Right like like yeah industry wins from it right like right yeah so I mean is there a even you know it seems like the the bottom is and we're talking about bottom in terms of cost right to proof which is great for consumers so is there a bottom in sight if not is there a theoretical bottom like you know mathematically I mean I think I mean clearly one can't ver make uh ZK proofs faster than you can run execution right you know there's a there's a it can't be ZK is slower than running it without ZK and and uh clearly there's going to

always be some multiple of of performance versus not verifying it but how close that can get to one to one I don't know that like we could probably make some back of the envelopes based on our current theoretical understanding and I think to Jim's point the theoretical breakthroughs have been slowing down the amount of engineering to catch up with the current latest Theory though is still quite vast um and and it's unclear if the theory is done or if there'll be more more more on the horizon I don't know Jim probably more able to talk to that than I am he's a little yeah I'm not like um I think that the research is coming a long way I still think that having true dedicated custom compute and custom hardware for

this stuff is going to be you know a step function it will be a new world essentially when we're actually running verifiable compute on chips that are designed to do it and you know why do I think this among other things we can look at analogies to other Industries like the Bitcoin mining industry where you know you if you were asking yourself where's the bottom when we were still in the GPU mining era of of Bitcoin mining you kind of wouldn't have been asking the right question at the right time um and then you know when you're looking at Graphics they made custom computers for doing Graphics processing those are you know video cards and graphic and gpus so that's you know I think after we get to the point where we're actually running

the entire proving protocol or the entire proving process on a device that is custom designed for verifiable compute then then we can start asking the question like where's the bottom of of uh you know cost efficiency yeah and we're going to get there like uh these these things the hardware is going to come out in theory you know like you said we we are getting as close to real time proving as it gets like is that the and when when we're thinking about like how fast can these things go like obviously the the cost is almost a little bit easier to measure it

would be you know like you know the cost of execution essentially and and then just ju in a in a one node system right because there's no re-execution so when when we're thinking about real time proving like Jeremy I think this is something that you you guys like have really ran with what what is the philosophy around real time proving yeah I mean I think what's interesting is is that we will probably get to real time proving before we get to the maximally performant ZK systems in part because you know not only are we improving in terms of the the cost and the speed of all of these individual proofs but we're also improving in terms of the ability to to split this problem up across

parallel machines right and so the first thing that will happen is that you'll be using a fairly you know decently large amount of Hardware but you'll get the proof done extremely fast by properly paralyzing the work which we already do you know today um and and and support in bondless and the the work that we're doing but I think that um and then what will happen is you're just going to keep that same you know latency and speed and just start needing less and less hardware and it'll become more and more inexpensive right so I think the there's a couple of different axes uh on on which this you know happens but yeah I think I think we're I mean I think that real time proving

is at least at the sort of level of um requirements for ethereum the a this year problem to be honest like could you remind us like what what are kind of like the benchmarks when it comes to real-time proving is it a certain amount of like latency and cost or what orc me really just latency right I think really the question is can you prove an entire block less time than it takes to produce a block right um and with some of the work uh you know in terms of the the the current sort of road mapap for um there's really there really two different sort of levels there's the um there's the question of can you produce a block in time for the next block which works for some models and there's can

you actually produce it within the same block right which is a little bit harder um but I think that you know we're I think many blocks were in the sort of 3 42 range you know already um and so I think we're getting quite close I don't I think only a few more we don't even need there's not even need requirement for a factor of 10 you know let it's like a factor of two or four or whatever like um the numbers are going down so fast it's hard to almost track it but it's it's it's it's I if I I would be very happy to bet someone that before January of next year that real time proving will be a

practical reality Jim would you take that bet sounds like a pretty good uh no I would not take that bet I I tend to agree with Jeremy on that one okay um and that's why we we think that I mean latency is really important for for a lot of use cases and for protocol design and for user experience but it does seem solvable with enough parallelization as Jeremy said and which is why we fundamentally care about cost and when you use cost it opens up new use cases like you can you can push transaction volume but you might also be able to get to a lot more use cases with client side proving and actual privacy preserving zero knowledge proofs um which you know uh an end user is not going to have a

rack of GPU equipped servers or something available to them but if you can get the entire proof running you know real time with with less Hardware then that's really powerful yeah I want to chat about that just for a little bit kind of in in these few minutes here because I I was on a a live stream just the other day with a founder of a ZK proof Marketplace and I was kind of laying out this Vision that I've heard from some of the local consensus maximalists like the guys at AA uh if you've talked to them like you know they're excuse me and and one balance too Stefan over there is really good and and Adrian an anoma they kind of talk

about this idea and you know we are the rollup like you know you know modular rollups million thousand rollups million rollups rollups for everyone and taken to the logical extreme it's like every user has its own state that it operates on and it conducts transactions on and we are the you know each user has that state represents that rollup and runs I don't know exactly what the right terminology here is but runs the logic to coordinate transactions for that state so you know the the I I was kind of laying out that vision and and this founder said you know that you know that's great maybe we'll get there you

know one day but it still makes sense to delegate the proving we shouldn't rely on client side proving but you know for that for that reality the client side proving is what kind of brings about you know the the local consensus and everyone being able to you know verify and Pro prove their own the the state of their own state if you will so you know when it comes to client side proving what is the what are the blockers like even if it could be something that even if it's something that we could do is it something that we should do is maybe the first question and then and then you know what if so how do we how do we do it yeah I mean we're actually working on our sort of plans and our road maps for client side

proving we have actually done some experimentation on client side proving our Pro actually currently runs on the iPhone it's not really production ready yet but um we actually our stuff supports metal acceleration and the iPhone has metal uh GPU acceleration right I I think though what's interesting is the way that we will likely see this come about it is harder and more expensive or sorry not more expensive but it's it's just a harder technical problem to make clients improving sufficiently fast um and so what I think we'll see is sort of a the long-term state may look a lot like a lot of the work does get done client side and clients are meaningfully able to participate and be participants and

this network of provs overall I think what we're going to see as a as a starting point though is this idea that some CR particularly privacy critical components of the transactions will move client side whereas the general less privacy you know critical things will still be done through sort of um you know a Marketplace of of provs where people have like high efficiency dedicated hardware for those provs I mean I mean at some point you know maybe maybe Apple phones will come with an you know a irreducible hardware accelerator built into them at which point it will be a no-brainer but I think you know in the in the meantime there are places where you can do most of the

protocol um you know in an open public network and only portions of it that are that are actually uh privacy sensitive like you know doing finding doing check of private coin transfer or whatever can be done client side and that amount work is much smaller and more constrained and is much more reasonable to run the client side approver um so yeah and there's a there's some interesting technical points around this so one thing that's unique about client side proving they might have metal you know uh gpus in Apple devices and you might have a GPU but we typically think of uh clients as being more memory constrained so you

need to create proof systems that consume less memory in order to produce a proof there's also ways to shift the uh the tradeoffs of how to prove something where you might find ways to create a zero knowledge privacy preserving proof that has a really large proof size and then delegate the compression of it uh trustless to a you know approver Network at that point um and I think that as we can explore models like that that would become really interesting and give users privacy guarantees but also get the efficiencies of like having economies of scale in terms of uh Data Center deployments and custom hardware for doing like the really heavy lifting and

I would say even if CLI side improving does become easier there will always be a place for these larger provs uh available of the network because there is just fundamental eies of scale that are you know we see the there's a lot of value in cloud computing today and and I think at the end of the day you will it will always be less expensive per you know unit electricity or you know dollar cost to do something on bigger iron than it is to do in your phone uh but I think so I think this you'll end up with a model where there there are both but I think this idea of decentralizing even further is is a

really very plausible direction for the the sort of almost like everyone has their own rollup is actually I actually think if you look at it there's place a lot of places where um you don't actually need full consensus you just need non-repudiation and verifiable Computing that you can actually build sort of a much stronger connection between things that would otherwise be in different consensus domains so I I I I actually think we'll see this type of changes but exactly how it's going to play out is unclear at the moment still so well y'all have given me a lot to think about and uh I I really do appreciate the conversation I always you

know walk away from these ZK conversations it's with people smarter than myself I I learned so much and I I really appreciate you guys taking the time to uh to have this conversation and and talk about the work that you guys are doing and also about the the state of ZK in general I hope everyone listening also learned as much as I did well it was wonderful talking to you always always a really interesting uh discussion for sure so yeah thanks so much this was really interesting thank you