Muthu of Ligero @ ZK Hack Meetup in Denver, 24 Feb 2025

Transcript

[Music] how's everybody um I'm Mutu I'm CEO and co-founder of Le hero I want to talk about what we've been building over the last year I want to say that this is going to hopefully redefine the way we interact with zero knowledge proofs how many of you here have proved a zero knowledge proof okay quite a few at the end of this talk every one of you if you have a cell phone you can prove it straight from your cell phone so if you are interested please keep your phones out

laptops are okay too it'll run faster so let's talk about like today what can we really prove in zero knowledge on a browser this is just a scale of application s with increasing complexity think of it the number represents the number of constraints that you need to verify I'll say that most people have reached probably the the 1 million uh Benchmark and in fact when you talk to teams that have deployed this on the browser they don't want to go beyond a million because either your key size is too huge like for Groth 16 or you just take too much time to to do it on the client's uh uh device so as is customary usually in these talks

there is an apt vitalic tweet to talk about so late last year vitalik says that you know ironically privacy use cases haven't reached client side proving and this is because client side proving hasn't been able to do anything beyond Merkel trees we want to challenge this status quo so let me quickly um let me say this I think we would be able to do real time ethereum block proving from the browser I'd say even in a year time this is my prediction so let me quickly um tell you take the LI Hero Zero knowledge proof

system for a test drive so this is a simple speed test that we um created I will share the link later you can all try on your device how fast it goes so this speed basically says that we can do 4 million constraints per second on this laptop granted my laptop is an M4 that I bought yesterday just to you know uh show how well are system scales but even on an M1 this is going to be pretty fast so if you have any laptops you know don't trust me verify go check uh I'll share you the link you can actually test how fast the liero system will per form on

your device okay that's about about predictions now maybe just to give you a little context of what you just saw in the speed test on the speed test we were just running a program that would prove posidon hashes uh on the browser and this is a slide that I you know the part of the table I borrowed from Ellie bason talk last year where he was talking about the stone and the St provs the stone Prov back when it was started they were able to prove 500 hashes per second on like a quad core Intel i7 machine and then they switched of course they made you know you know amazing improvements and switched to a small Prime and they

were able to go to 500,000 now I want to say back in the original sort of 254 bit or even 250 bit regime on the browser we can do about 27,000 posidon per second and you know we haven't yet optimized stuff this is just like sort of preliminary numbers that you can do from the browser and you know just to give you a sense you want like 500,000 or roughly on that order to uh prove an ethereum block but our vision I want to say is something different yes we have built zero knowledge that can be scalable that can run fast on the browser but really our motivation comes from building useful and usable privacy tools and this

is what I want to talk to you about so today privacy comes with a lot of tradeoffs and just thinking about the applications that zero knowledge can enable today Enterprise that want to use ZK they have to make a tradeoff either they get privacy without compliance or compliance without privacy privacy comes at a cost you want to employ prer networks you got to pay and you shouldn't share your data to approv a network so you know um privacy also comes at the cost of usability the the ones that are trying to bring privacy tools require you to learn new tool chains and this reduces the amount of development that's going to happen in

the system and finally again often you have to trade off your privacy with trusting someone with your sensitive information what we want to do is we want to build useful privacy and in our mind what useful privacy is applications that people would really want private and I want to say there are a lot of of people and an amazing you know set of people especially even in the ethereum community that is trying to push this privacy applications from ZK email ZK TLS as well as keyless login by Aptos and M labs and also on the Enterprise side you know there are so many applications where they don't want to reveal investor credentials the Europe is moving to digital ID there are

numerous applications that want zero knowledge today and these I classify as privacy applications for for individuals because you want to run them client side Now privacy also matters for organizations and for us this is captured by ZK validium you can think of Enterprises and organizations opening their own chains and they can control the privacy of their chains and they can hook on to ethereum any of these blockchains as a Valia usable in our mind is that you want the ZK to be usable yes you know there is a lot of I want to say amazing work done in zero knowledge I want to

say I'm a professor at Geor University I've been teaching zero knowledge now for you know close to a decade and the progress that we have seen in zero knowledge is incredible right but still to get to the point of being usable we want standard tool chains and today this is evolving this way people are developing more and more zkm affordable Hardware I mean you don't if you want client side proving you shouldn't have to go and buy a uh an ndia 4090 it would be great if you could just run it from the device you already have in your hand or in your bag portable implementation if developers want to build zero knowledge

they want to deploy it people want to build would want to run these things on their phones on their devices you cannot ask them buy an Nvidia uh uh GPU or buy a specific uh laptop and finally you want a privacy default approach and again this is a lot of the ZK today is focused on applications that don't demand privacy so they are not sort of focusing and also refining their systems all systems can be made zero knowledge it's just that it's not default today today so I kind of covered this that the modern ZK kind of falls short and here are just some of the reasons just

technical reasons ZK snarks often some of them need huge memory Stark based ZK again they cannot run client side yet like if you want to use recursive composition shards this is not going to scale on the client side and most ZK are built for verifiability and the Privacy ones require their own tool chain you have to learn Noir or you'll have to learn circom and in fact this is what people are doing today there's so much of application built in circom and Noir for these privacy applications so today in E Denver we want to unveil our zero knowledge platform that's going to redefine the way you are going to build zero

knowledge applications so let me quickly run through a demo uh on how we can do this so here is our platform and I'll I'll share the link at the end of the top anyone can go here there are already a few zero knowledge applications on the the marketplace for people who don't want to develop them you can already go and run them I'm going to just pick one of them this is something about text redution I want to prove that I'm a whistleblower I have some email I just want to reveal some of these contents and prove that I have the original message you can go here you can change the inputs over here to whatever you want and hit Pro and it's going to run straight from your device so this once you build it in our platform you can

deploy it directly and anyone just with a browser even on a mobile phone can can change inputs and can start using your application and let me quickly walk you through how easy it will be to even build these applications so uh not the text reduction let me go back so here is a simple application that measures edit distance between strings that's useful for genomic research I've written this code in cc++ all you will have to do in our system is write this and compile and once you compile you can run that's it so you can do an entire application from the browser you can write your application in C C++ and this week you can also we have also integrated rust so

you can also put rust programs in our system compile it and you can deploy it and once you hit deploy it'll go to the marketplace and anyone with just a cell phone browser can run your zero knowledge application imagine that so let me get back here is a quick Benchmark please take the benchmarks with a grain of salt we've tried our best to be Apples to Apples but this never happens in the ZK land so we tried the novanet and thanks vatt for helping us uh with with their system uh sorry if the performance doesn't really match what you have we try to do it with on our uh our system but I'm sure you know major improvements

are are coming in these systems so so no one net is slightly under a Kilz we we tried Noir system that's around like 44 khz the one you just saw like on my laptop is the fourth one which is about half a megahertz I want to say that if you want to run like more there's a difference between running structured application and unstructured structured is about hey I want to verify many many things many many signatures at the same time and unstructured is like edit distance and there is a difference in performance and I want to say that when you want to think of real-time ethereum block proving it's about like tons and tons of hashes tons and tons of signature verification you can look at the last line you can we can go beyond 1

mahz and if you have a better laptop you can go up to 6 mahz and I want to say these are preliminary numbers we have focused on a 252 bit Prime because we want usable applications usable applications today need signature verifications on big primes but if we reduce into a small Prime which we can just like Starks do then we are going to get a 20x Improvement straight uh from here so let me tell you two slides these are the only technical slides I'm going to tell you and if you want to zone out please do and you know uh I'll I'll get back after that so the the zero knowledge behind our system uh is is called ligron which is based on the

liero proof system that was published back in 2017 ligron is liero but memory efficient it is also a ZK VM this is what allows you to code in C C++ and rust and ZK by default and finally we used web gpus I know that web GPU is still maturing but we've managed to get a lot of coverage with a lot of devices and this is only going to improve because AI needs gpus so web GPU is definitely going to mature and we focused on doing it on web GPU as opposed to Cuda because we wanted portability we want everyone to be able to run zero knowledge what is the secret in

ingredient well there isn't any secret ingredient and that picture is really for my four-year-old daughter who's watching Kum Fu Panda every day but um our system is Hash based just like ZK Starks but the main difference is we use a different way to do constraints and we do code interleaving and if you don't know what code interleaving it's fine if you have heard of breakdown breakdown uses liero commitment scheme which is code inter leaving and that allows you like a lot of parallelism you don't don't need uh too much memory and this is what we can leverage and finally I can tell you about memory efficiency but if you know a little bit about the lih hero system it's based on something called MPC in the head if your underlying MPC is memory efficient we

can borrow that straight into the leero system but don't discount that I mean a lot of effort has gone into engineering this reality uh my engineer keeps telling me that hey like you can tell theoretically a lot of stuff but you know until you get to deploy it it's not real so uh even though I can argue theoretically why it's memory efficient still a lot of engineering efforts gone to uh making that a little bit about just the last couple of slides here is our road map today we can integrate C C++ rust I want to say pretty soon you're going to get circom and once you get circom all the amazing work Iden 3 has done and everyone has done you can just copy paste that circom code onto our browser and now voila you can you

can do all of those applications from ZK email to ZK login Etc I'll also say that what we are focusing Beyond this is a plug-and-play solution for identity we want a no code version of our we shouldn't have to code once I have a suite of ways of proving my identity it should just be check boxes and I should be able to deploy an identity tool on my web 3 or even web 2 and we are also looking into ZK validium we think truly this is a way for organizations to get scalability and privacy and also our ZK VM I've not talked about it but we're going to get sort of offchain compute of databases AI Etc we can equip ZK rollups and validium

with this additional features that people will want and that's in phase one but in phase two we also want to incorporate Noir and Koo because a lot of work is there we we don't want these efforts to be duplicated on our system if you can just borrow these and cross link then you can code and C pick a function from Koo and pick another function from circom and be able to compile it down and I want to say just a small uh thing here web assembly helps as the intermediate uh representation because it has this module Concept in Risk F this is going to get a little more tricky to do it so we chose web assembly I mean for other reasons but this benefit is something that we are going to get from that and finally I you

know back to my prediction I think we'd be able to I predict that we'll be able to do real block proving and even llm inference from uh the browser if you want to test how fast your device can prove just scan the QR code if you have a laptop you can just go to this site Devy platform. legatron ZK anywhere everywhere thank you [Applause]

[Music]