Lisa Akselrod - DevRel and Tech. Writer at Aztec Labs | Privacy Academy Interviews

[Music] Uh hi, my name is Lisa Axelro. I'm part of Devil team in AdSt. How I ended up where I am now? I'm currently at Etham in Amsterdam. Um crypto crypto event dedicated to privacy.

I'm here for two reasons. I work in crypto and I work in privacy. Um before that uh I worked for another crypto company that does not related to privacy and it's obviously more fun to work for crypto company with privacy rather than for crypto company without privacy. how I ended up in crypto. Um, I was fished by ZK, zero knowledge cryptography, because it's a lot of fun and there are not too many things in the world that looks like true fun.

I would maybe call three uh surfing, great people, and awesome math. And the sad truth is that most math that isn't incorporated into the world as we know it like in smartphones and computers is not very fancy like I don't know if someone knows how cash function looks it's not very exciting but in uh ZK in that we have currently in blockchain we have something called elliptic cryptography and we have something um called parents that is a part of elliptic cryptography If you ask someone like what is pairing, they will tell you oh just blackbox it and this is what all people do at some point um several people from um from PSC was like okay let's study parents and the first 16 months we spend on preparation just to be ready to study parents and I'm uh incredibly addicted to that stuff. It's like just feels like a lot of fun. Even though for most people it doesn't feel like a lot of fun. If you ever showed up in uh like there is a project called ZK hack and they do study clubs and uh they often run study club on moon math manual.

This is like the introductory book into ZK that combines a lot of like math and a lot of a bit of programming and uh the study club starts and there are like maybe 50 people and then after like second session they they say okay okay this math why do we need it like why we can't just write code and it assumes that like maybe 20 math lessons and then by 10th lesson you have like three people who are interested in math. So the whole reason why I'm today in blockchain is because in blockchain there is zk and zk is a lot of fun from the mathematical perspective. The repor of privacy project is orders of magnitude larger than the repo of like any other project because you just need to build so many things that do not exist like in privacy and blockchain today. You need to build things that you do not know how to build and no one in the world know how to build. It sounds incredible.

Uh I think it sounds less incredible if you are VC who invested in this company because like you don't really you can't really predict how long it will take you like ASAC is 8 years old and it's not that it planned to be 8 years old by the launch of test net. It just happened this way because for example we need to okay like some basics we're like we want to have privacy we need a definition what is privacy after some like experimenting and like being familiar with products like uh Zcash or like something monofunctional you're like okay we in fact want privacy to make it a tool for developers we want people to be able to build different products with privacy then you need privacy to be flexible then you think Okay, what is privacy? To be flexible, it means that you can make choice when you build an application. What you make private, what you make public because monolithical privacy is very constrained.

You think okay how we can do it like Aztec came up with using client side proving. Client side proving means that you want to process private data on like smartphone or laptop browser. And what it means it means that your RAM constraint is something like 1 GB and this is incredibly low for any proven system that is usually run on like 64 GPUs or something. So what ASAC has to do is has to do redesign a number of crypto primitives and to optimize them for client side proving and then it's not it's not like that okay we build that then we need uh we do client side proving we need to decide to design client side execution then we look what is happening in the area um we think about sequencing we think about the fact that for this system to be longlasting

we have to do it with centralized sequencer Even if maybe we would like to do it centralized, we just by design can't do it. So we start designing decentralized sequencer and it's very tricky matter like it deals a lot with mechanism design where your kind of like experiments what you can do without like launching the network. It's very minimalistic because we do we have test net. Test net is a magic money. it does not reflect how the mechanism will work with real money.

So it's also like very very very tricky. Then you also like when you do mechanism design you need to think about your counterparty. You need to try to come up with all types of mechanism hacks that the counterparty might came up with and uh it's also like very tricky but it's I mean so much fun. It's like you are solving the problems that no one knows how to solve. Okay.

I don't know what can be more fun than that today. Like for for super for people who are a bit less geeky than super edgy geeks, today's a great moment to start uh ZK journey or privacy journey because you have today you have spectrum. If you want to be super heavy on mathematics and cryptography, you can go for it. uh you can uh take videos on YouTube, you can take papers and just go for looking at uh what zero knowledge uh proven systems we have. We mostly have today two types.

We have elliptic based elliptic curve based cryptography or hashbased cryptography but also some like more experimental things like latis based cryptography and you can just uh like go for it. You just take a paper you read the paper you say oh I do not agree this is wrong. you go on Twitter and argue with everyone or do something more reasonable maybe start a company I don't know but then um as the next uh step uh you might learn some mathematics like some number theory about like fields about groups and then you can go for like more engineering side uh but uh more low-level development like uh using something like circum language uh for people who are like kind of want to be a bit abstracted from cryptography in terms of um having a programming language but still want to be hands-on because when you're writing a circuit uh when you're writing writing a ziki program you pretty much uh you want to prove that you have a program it was executed on given inputs and it was executed correctly. If you do it in low-level language, you have to write each particular constraint. It's like for for people with some mindset who like low-level development or like have feeling that they control what is happening.

And then for people who want to work with Zikke but they want to be like abstracted from all the cryptography they can go with something like noir the ziki language uh developed by adtec where it's high level language it's rust based language and people pretty much do not need understanding of cryptography and they still are writing the programs and they can go more into like product world or like real world application world uh without um thinking about technical aspects and um I think there are quite not too much but quite few materials already on that today and uh just a lot of a lot of people are learning these matters so it's very fun to team up with other people who are learning and learning together it's it's like hard to say but to talk about it objectively but I think that most developers maybe 80% of developers in the world prefer high level languages and they do not want to like deal with was level languages where like everything is very cryptic. So noir is just quite simple and native in how you use it like you write a program you do not write circuit and then this program is executed and the proof is generated. So I think it's quite na like it's quite easy to pick up at least like your first steps and like your beginning and then the question is how you can um how you can use it. Noir is chain agnostic in general meaning that you can write noir you can even uh write it outside of web 3 because what you do you execute the program you generate a proof then someone uh verifies the proof. I can just send you the proof in WhatsApp theoretically um and you will verify it and if we have like if we assume that WhatsApp is a trusted channel we are good.

Uh but in blockchain we also we can go for we have today noir verifier on stark net and on EVM chains. um we can go for that but in newer um like if we do it on EVM chain we have to build something like monolithically private we do not have the flexibility of states of like private and public state we can do something that like we put inputs on one side then everything is private then we get the output and then if you want this flexibility and very often in applications we want flexibility For example, if we are building a crowdfunding platform and we want donations to be private, but we want the accountabilities of how they were spent, we want to have both private and public aspects. In this case, um, one can go with noir on adtech. And I think that here it becomes a bit tricky because um there are like in general in newer there are some features that people did not get used to such as um you can't have dynamic size uh arrays you need predefined um predefined uh all sizes and people for them who came from rust they feel like very uncomfortable with that in the beginning or like Um we have a thing called unconstrained function meaning that uh okay in general when nor program is executed um it's zk proven the correct execution is zk proven but if you have some part of the program some function where you do not care about verify verifying the execution you can kind of like put it aside and only verify the result. It works for like heavy fail computation when taking a root is heavy.

Uh checking that this number is the root of this number is easy. So you would like to take uh root as unconstrained function and then just verify the result. So it just takes people um it's easy to pick up the basics to start building your first app to fill the syntaxes to uh see how it works in your hands to prove something to verify it and then it takes some time to learn the concepts but those concepts give you like completely new things that you can't do with any other language with any other tool and uh if people are fascinated ated by building something ZK, building something private is just in terms of flexibility and uh being explicit is the coolest way to go. Thinking about uh how much do we need ZK for real uh and not just because it's beautiful formulas. Um I think for blockchain privacy, ZK is the only way to go.

If we're talking about trustless privacy, not like TE's, we do not take TE's here. They are trusted. We want if you want to go for trustless privacy, um I would say ZK is the only way to go. If uh something is happening in the world and you want to ensure that it was done correctly and at the same time privately, you have only two options. You can either go there and check what was happening there.

So you like personally observe the correction. Um, is it still privacy?

I'm not sure, but I think it's kind of how web we can think about web two where you kind of have here a user and a server of the company and the user kind of trust the server of the company that it won't tell anyone and the server usually doesn't tell anyone because it makes business on that. But it's like this this type of privacy that is not really privacy but can be called also privacy and or like it should be cryptographically verifiable. Uh and in this case we pretty much have only only ZK. And the very unique thing about ZK in general is that for example if you have um if you do symmetric encryption or like some other encryption you can have a back door and like we know that if you can have a back door you will have a back door because why not if you can

and in ZK you can't have a back door by by design it's just this just like it's physically impossible and I think it's incredibly beautiful because it provides a type of objective truth

meaning that like the the prover like in Zika you have two parties the prover who generates the proof that something that the program was executed correctly on given inputs and you have a verifier that verifies that the proof is valid. So by mathematical construction of ZK proof the proof the prover cannot generate that something is correct if it's not correct it's just it's impossible and the same verifier can't say that the proof is valid if the proof is invalid because it relies on mathematics and it seems to be perfect u perfect fit for verifiability just uh especially for private verifiability when you want to kind of attest that something has happened somewhere without knowing what. So the first thing I think that for ZK for privacy zk is the best fit and we won't come up with anything else. we will just improve the game privacy and proceed with that. And looking at uh we are just wrapping wrapping up today the newer hack uh and look looking at a number of applications and the quality of applications.

I became like one bit more confident that ZK is the fit for privacy. But also looking at ZK outside of blockchain, I think we like every like we have the world like have today two parts. We have something physical like we are here at physical event and we have something digital like we all chat on telegram or maybe on signal if we are a bit more advanced. Um and we have this blended physical plus digital world and even physical world we are somehow like managed to develop the humanity managed to develop institutions to attest the truth. Like we have courts that like if something feels wrong we come to the court we say to the judge this is wrong please investigate and the judge tried to do his best or her best to somehow find the truth.

In digital it's way more like obfuscated. It's uh like we have tons of information everywhere. We have tons of LLMs processing this information and like shaping the humans within the like internet of how they should behave and uh I think that there is huge feed there for also verifiability and for private verifiability especially if we think about like something like proving proving AI models proving LLMs because LLM is like LLM is huge. Several dozens of people worked on this model. This model tells you do this uh gives you some recommendation.

you have no idea if it's good recommendation, bad recommendation. You can't recheck how the model was trained. Like comparing to Ethereum for example, um if we have like we have the same model on adtec like one uh very validator executes the block and then several hundred validators checks that it was executed correctly and so they test it. If it's like computationally heavy like LLM, training LLM is so hard that you will not attest it. You will never recomputee how the LLM was trained.

But you want to be able to question it because like the world as we know it today, I don't know how many LLMs shape my behavior in a day. maybe several dozens maybe already several hundreds and I'm afraid we can end up it's a bit like doomer paradigm but I'm afraid we can end up in the world where all the behavior of humans is shaped by LLMs humans are not aware and have no idea why LLMs are do are doing these actions because when you train the model you have so many parameters you have so many things to consider you have such a high possibility of like some flipflop mistake or something without being aware of that. So like in general merging training models or in general any automation like um self-piloting vehicles or something that is happening on its own with verifiability seems to be a very good fit because you want we want to have machines accountable. We want to have AI accountable. We do not want it all to control us.

It's very hard thing. It's not that like oh we can just take LLM put ZAR on top and it works. It requires like a lot of technical work. It requires a lot of like let's call it educational and ideological work. Uh because uh maybe not everyone will like the idea of uh LLM accountability.

Um but it seems that it's worth working on um at least trying and also ZK proven LLM is incredibly interesting technical task. uh if you are mathematician or cryptographer or fancy engineer or curious engineer or fun engineer I think today is incredible moment to start exploring the kay because today it's not obvious it's like there are I I I do not know how many but I think that maybe less than definitely less than 10,000 people in the world who understand like uh on a good level what ZK is and maybe I would say in my own like estimation less than 100 people who are literally good ZK cryptographers meaning that they can construct proven systems we know that like in common sense it can mean two things it can mean that we are doing something that no one really no one really needs and that's why it's so small Or it can mean that we are early and this is the right moment to start. And if you think about analogy with AI like most large AI companies started in like 2003 2004. Um in 2004 the penetration of like smartphone and internet was so low that most people like didn't believe that people will have computers like that computer will be consumer device

and thinking about like okay you want to run models on data on like which data like where will will you take from this data the idea of large uh LLM in like 2003 was like almost purely theoretical

and didn't make sense for most people if not all people but there were people who like this will be the thing people who were able to like feel the world feel the world needs feel the world sentiment where it's all going with early internet times with early computer times and I can't promise you but I personally feel like that very strongly that this is the same model for ZK for zero knowledge cryptography because this is incredible verifiability tool and the world is going deeper and deeper into more uh digital rather than physical. And whenever you have something happening digitally, you want to be able to verify it, maybe not because you want to have uh objective truth, maybe because you just want to be able to navigate through the digital world uh in some like manageable way. So I I I think that everyone should work on what they feel good about. But if you have never touched ZK, uh just give it a try. Like take several hours.

Uh watch some videos on YouTube, maybe read some paper, just give it a try and look at how it feels. And uh maybe you will find there's something for you that touches your heart that talks to your feeling and understanding of the world and then uh next year when we will be in Amsterdam and maybe you will be here too. Uh this is a bit of a sad thing about ZK. It's not the most uh well doumented area. Um I would think if you are more into engineering um try new try circum if you prefer more highle languages circum if you prefer more low-level languages for noir go to um um newer documents this is the best place to to start for sorceram maybe try the book that is called moonmath manual It also introduces some basics of mathematics.

Um for mathematics

uh that's a good question. Maybe also try a bit of moon math manual even though for people who like mathematics it's quite high level. Uh check the playlist of the project called zig hack. If you just open their playlist, it just list the number of uh crypto primitives and you will have no idea if you are coming from outside of the world. Pick anyone whose name you like the most and just got some feeling of what that and then maybe you can't stop or maybe you escape it forever.

And maybe another thing uh look at other Zika projects try to read their code try to uh understand how they built. I think this is the good point to start and uh another important thing is if you feel that like you are trying to understand a paper or code or something and you don't feel like you really understand don't blame yourself. blame the one who wrote this paper or wrote this code. Um, it doesn't mean that they're really guilty, but it will make you feel better because Zika is incredibly beautiful, incredibly powerful, but it's also quite complex and it takes time and it takes efforts. And if you like ZK, the next step you need to bring your friends into it.

And next time, next year, we all will meet in Amsterdam. [Music]