Exploring the Future and Potential of Zero-Knowledge Technology

Transcript

foreign [Music] hopefully you can hear me okay yeah good all right thanks everyone for coming out this morning we've got a very interesting uh set of people uh extremely uh interesting uh topic to discuss so I think it's basically just first of all if you just give your names uh where you're from and I'll introduce you to the audience so and currently I'm a doctoral student at etiad Zurich and the currently I'm a member of the developed team of hyper Oracle we're working on zika oracles and generating automatically approved instance thank you hi my name is I'm at

geometry and I'm on the research team as a research associate so previously I was at EF working on applied zkp so thanks for having me hey everyone my name is Ignacio I'm a cryptography researcher at nethermine and currently I'm focusing on Azure knowledge proofs and their security mainly hey everyone my name is artem I'm part of Oregon research team focused on ZK currently we're building some private boating Solutions but we also look into private dials and related topics so yeah I'm also a bit interested in zkml that's my little side hobby right now I'm I'm here talking from dogfight which

is an L1 for anonymous engineering we and we have our test net released now which is has generalizable programmable Anonymous smart contracts all right thank you thank you panel all right so just like to start off uh if you could just think about you know the last six months a year think of any new application in zero knowledge or new piece of research has really excited you the personally the the people wouldn't know about so you can go in any order so anybody who has a I mean it's a big news not only for our community it's actually for the order

what people are paying more attention on let's say artificial general intentions and large language models so so how to remain the confidentiality of the user data when during the inference time I mean it's really important right now because like due to the gdpr protection or such kind of violations always happens so I mean particularly we can talk something about maybe ckml so how do we use dks uh utilization to protect our user data and what is the situation of it I'm quite interested in that yeah

I'll certainly come back to see it came out in a second yes so well I guess for the past six months and even past one or two years we've seen a lot of focus and development on I think scaling Solutions and I think the technology and the fundamental approaches that teams have been doing have been around some for example recursion or around ZK VMS using different techniques and all this has been focused on trying to scale blockchains and train scale ethereum on L2 and I think a lot of a lot of interesting teams have started to use these building blocks and build things Beyond scaling such as identity

and even ckml so I think that's very interesting and I think there's there's so much has happened in the last six months you know the last one year uh it's very hard to keep up um the um I think yeah so there has definitely been a fluffy of activity in these areas all right thank you yeah actually I it's kind of funny I was kind of a mix of both previous answers in the sense that um very interesting application of zkml kind of at the intersection of um the validity proof of like a certain computation the Privacy preservation of the data and also the heuristic optimization of of machine learning and one application I was particularly

interested in as well is in identity management actually and things like ZK credentials and and okay it looks like zkml is the Hot Topic right now I guess the reason for that is we just learned about what are the capabilities of big language models about half a year ago so it's perfect timing for the question and actually I already know that well for me part of the problem is that we also need to prove that with the influence correctly to put the model on chain for example and right now the big businesses are making big decisions based on the outputs of these models so proving that the model was run correctly is quite important and actually no there is an effort for the doing this for his gpt2 by Daniel Khan so yeah I'm very very

interested where this goes in the future thank you there's a common Trope in sci-fi which is contact with an alien civilization for example in the expanse when they first discovered the ring and all of this crypto Tech which is coming out now is honed powerful techniques that is opened up an expanse which we're only just beginning to peek through the door or scratch on the surface and there are so many things that we want to make that is that is really crazy like obviously there's stuff around markets and building

Anonymous digital marketplaces but uh we also made uh the world's first fully Anonymous style Anonymous on-chained treasury Anonymous voting all the participants Anonymous and that was from our experience with Assange style where we raised 55 million to free Assange by all means necessary but then it was like a public Unchained treasury where all the participants were public for a political prisoner so it really just didn't make sense in a transparent context and then second of all we were like bidding in a in a auction and everybody could see our Treasury and the fact that everybody

could see it meant that nobody would bid against us and it just completely broke the game theory of that auction so anonymity really expands the design space of financial engineering and you know there's even other stuff like uh online marketplaces which to completely Anonymous where you can buy and sell anything where you can exchange crypto we could like where you can like like local Bitcoins completely Anonymous like all of the crypto infrastructure we should be building peer-to-peer and an anonymous also there's load of privacy projects going oh we're building privacy Tech but they're on Discord and they go

oh link your social media to like authenticate and it's like what like how are we gonna like really deliver the true promise of crypto when we're not actually building the self-sovereign infrastructure and consuming our own dog food so we made our own Anonymous chat it's the most Anonymous chat on the market there's nothing that exists like it has there's all the messages are unlinkable there's no identity like not even signal or anything else comes close we made an anonymous task management system and we're going to migrate away from GitHub so we move into a fully Anonymous peer-to-peer Paradigm and also another really core topic is information marketplaces like you can make a

marketplace where people can buy and sell information like you know like oh you have Hollywood movie or or secret document and somebody says oh this documents this thing and go like okay prove it and then you can have a protocol where certain random portion of the file gets decrypted and then it authenticates the claim of what amfo is and then everyone bids in an auction and the person that bids on it gets the uh decryption key and then essentially you've just destroyed the business model around secrecy for corporations and Trade Secrets because so it's just so exciting like there's so much to explore like and we need as crypto to like organize and mobilize and work towards that goal because this is a major historic invention like on par with the

printing press or gunpowder it literally sorry can we can move on I think we've sort of covered that uh quite well now so um I'd like to go into the the zkml side as things because you know if you want to get all your buzzwords lined up now you're going to do zero knowledge and machine learning now obviously there's there are problems with this so uh doing the sort of the the inference part you know the CK can can do this but things like you know back propagation that's you know as far as I know this is not not really feasible but I saw a recent uh thing at the CK Summit where they they were doing uh weightless neural networks so this this was a uh this is simply promising but I don't know enough

about this to judge this so uh so Adam can you give us some insight onto that I guess I'll give a big bit of background so I was part of the zero gravity team on ZK hack uh the winning solution and what we did we tried to reimagine how neural networks work so instead of um trying to fit in new networks into ZK by doing some truncations of the weights because right now so that you understand neural networks work using floating point arthmatics and what we use in ZK is more field arithmetics we just don't really work together there are ways around it but they lose efficiency and they lose Precision so essentially you get worse models after doing that and it's very hard to do it so we try to imagine the solution by

essentially going away from floating parismatic all together and using one of the previously preposed Solutions of weightless neural networks where you use lookups to optimize how they work that's that's what we built it worked on nist um but we had about two days to make it happen and right now actually there might be future work with the ethereum foundation on that but I cannot say much on how this will develop but uh also in general what people don't understand is that it's not enough to prove inference actually you need to prove the whole training process because it has been shown that an adversary can always corrupt the machine Learning Network from inside when they train it so that they can always control what it outputs

so it's actually a very big problem for the space that essentially you need to prove that GPT was trained correctly and it's already hard to train GPT like that imagine how hard is it to prove it so yeah okay so I also wanted to put some course remarks here so everything you mentioned here is like uh proving something is like training correctly like it is like a trade-off sacrifice your utilities for some preservation of the confidentiality or correctness and and most often and in the other occasions with when we talk about ckml we care about the influence right so

imagine you're talking with the chat gpts you want him to compose some some letters for you but actually you don't want to really review what you want to to him so actually what we get here is like okay inference this correctly but do not reveal my information so actually this is a little bit different from proving okay this influence is correct because if in that case you can like employ this you see correct it output as your witness something information notes to you but actually and this is the case in most of the gkml applications like easy care and such

kind of stuff but the question is like first if you really want to influence something and and outputs Space is really huge and you cannot do any heuristic augmentation inside it and it will be a question that which statement you want to prove right and another question is that I need to mention is about the visibilities I do not know too much about the training time but for influence time if we come to uh prove for more than 10 million parameters and the same level of probes actually it's not a big model right in our in today's view but it

would take like maybe more than 10 minutes to do so and which is not affordable really in practice so that's we're gonna figure out maybe in the Futures posts from with joint designs both from the lower level architectures operators and the algorithms are made design with as hair layer so that's all and maybe just to add a bit more like this is another way to think and the Third Way is to add fully harmonic encryption and and say that what about we want to prove my um like do inference on my private data without the other person knowing what I'm actually doing inference on but they have the model so that's like a third level of complexity and I think this is one of the hardest problems to solve right now yeah

so uh the bridge is kind of to a more General Point perhaps in that uh what what are the uh potential bottlenecks at the moment is there any is there any particular things uh industry or knowledge that you think well only if we could you know this particular hash function was quicker or something like that if only we could solve that then the whole area would open up what do you think Ignacio would you like to is there anything you can maybe one remark regarding the the last point I think something that was quite surprising to me is that many times people because of this LGBT kind of concerns about like using your information when they're doing the inference people focus on the fact that you can render your input data private I think what's really interesting is that sometimes in some applications you need your model to be

private and you don't care so much what the input is going to be but you do care that people cannot know the kind of models or the weights that you're using um for example one example I have in mind is you want to detect sybils in some Network for example and you've figured out some some heuristic way to do it but if you tell people that's the model we're using then they will go out their way and kind of evade that model so that's that's a a cool application that I had that I want to talk about something that I'm personally quite excited about and hopeful for is for developments in Hardware especially if Hardware that accelerates ZK proof generation can be on personal devices as

well as like algorithms that make that can improve performance I think generally speaking if you want to have privacy for users you want to own your own Hardware you want to own your own data and therefore you want to be able to generate proofs locally and if you Outsource that there's always a risk of of information leakage and loss of privacy so I think I'm I'll be ready I'm personally excited to see developments in this area so just let's go on to that point I mean Hardware um you know to get uh having memory on hardware and this kind of thing and caching it's it's as far as understand it's very expensive are we at a situation where you know the the the proof sizes and things like that are

small enough that uh it's not going to be too expensive to do all this or what is what is the current state of this does anyone have an idea uh I I don't have much information about Hardwares but I I don't want to talk about some some road maps for for our reasons like generating a really zero knowledge proof like I think like typically we have like uh if we want we have a organism some some kind of uh accepting some language and we want to generate uh proof circuits for that and uh a general way to do that is that okay so first we translate it into uh arithmetic circuits and based on this uh very basic pieces of gates of the

circuits of what what is we called this the witness and instance we kind of generate uh very basic system for that this is only one typical roadmap and for the other we can uh employ the terminology called lexic virtual machines like in this sense we we're gonna propose a portable maybe virtual machines or compatibles programs that we can understand what we are coding about it's not true to understand but uh actually it's just take our natural programming language let's assess let's say platforms or rust that kind of stuff

as the input and to generating the proof circuits but the question is that the drawbacks here is that the bypass the procedure of okay we really designed what is a basic arithmetic circuit so it will generate many maybe redundant or like very in terms of nested uh indexings requires more computational powers and more larger size of circuits to do so so I mean the questions here is that first should we figure out uh maybe a new

we see end-to-end ways to generate some proof circuits because like uh a very interesting efforts recently I see is from the Ezekiel the they do generate the proof leakage not from the program language but from the uh really basic services from this operators in Onyx sorry you mentioned someone who who yeah which is a cure Ezekiel Ezekiel yeah e-c-kl okay yeah yeah yeah they generate the proof deck is from the Onyx which is uh I mean some some learning architecture like uh cafe or like

pythons and so that's a good idea maybe we can not only do own the very basic level we can maybe let's say in learning we do everything on the tensor level right can we do something on this kind of Concepts instead of like focusing on the very fundamental uh gauges what do people think is it um well I think yeah like we're seeing a lot of different crypto techniques so I think the destination that we're going towards is rather than like a general for example ZK proving system we're going to have very specialized pieces and part of that is um as well related to two things which is one is the optimization where right

now a lot of that's done very manually by hand you like pre-define or you have like certain gadgets or like for example like lookup table was a fixed number of rows and it's very difficult um so yeah it's very difficult because you have to like rearrange all this data and and because it's not Dynamic it's like hard-coded you can't have it where where the entire tooling is intelligent enough to like generate those tables for example that are specific for that proof so it's optimized so very we're very early in that sense like if for example you look at GCC when you use GCC the reason why it's very good optimizing compilers there are very many different modules you can load for different

architectures so it'll be specialized for different applications another thing is abstract operators right now we use like mainly addition and multiplication in circuits and you can pre-define Gates but for example we saw Nova now or which comes from like I think Spartan and higher expert they added a fold operation so and kind of it would be very interesting when we see because I right now everybody's learned the wrong lesson from history which is that c is better than lisp but that's not the lesson from that history the lesson is that people prefer to see because it more closely matched the underlying systems they are working with and the systems we're working with are algebraic so there's like a different

Paradigm we have to construct rather than trying to create a c-like system and I see a lot of confusion with people where they're like oh so it's Computing the value and I think that's a mistake because we're mixing computation and proving together and they should be kept separate um and yeah and also very important thing is yeah membership proofs because a lot of applications when you build when you start to build them you see oh you're doing some kind of statement which is there is this object there's in the set of all objects and in this is a commitment to these values and then there's this value which is in this set blah blah and you have a lot of statements like that so set membership proofs yeah right now

like for the amount that we're doing they're not like well they are quite expensive um but they're like kind of manageable but you know if we could abstract the way that cost potentially the usage of set memberships will become really big and so we will get these very Advanced and interesting circuits like for example they mentioned the the machine learning stuff which right now there's this there's these big centralized models that are owned by um you know like these big people and people are like okay well I'm worried about my privacy by feeding this model so you know like we would like to have like our own models but where we like

pull our power and we compute it collectively um and a big part of that is that's really what you you've been talking about you did the uh you just done your talk about the the gas efficient thing could you want to comment on on where we're going with that I think I'd like I'm here to continue because you're on the thread that's very interesting yeah the the other part is the set non-membership proofs which I don't know if this is a solvable problem it probably maybe isn't because there's a there's a a result from computer science about uh like uh Bloom filters and I think it's called approximate membership query filters or something but it's it's uh basically is it's you can create a set membership

proof which is constant but is it possible to create that proof uh without having the entire set of data so if and and there's a lot of um like I don't know isomorphic uh problems like related to this domain like for example like if you get two graphs can you prove that this graph is contained in another graph like is the com is the union of these graphs have a perfect well-defined ordering and if you could do that you can create like very Advanced consensus algorithms they're completely Anonymous so yeah so there's a lot to kind of explore so I guess one

simple application of uh set membership proofs is uh or one simple development upon set membership proofs is that when you have a private application and you want to be able to comply with being excluded from a certain list of of entities that you do not want to be associated with we want one application or one development is that you prove that you're in the in the set but that also that you are in you're not in them in that set of of entities that are that you do not want to be to be part of so uh there are a couple approaches to this one approach which has been popularized recently is that you you use mocha trees another approach that I like to have C

hackers built on is to use case G commitments because if kgt commitments you could you could simply do elliptic curve subtraction and then you've you've obtained the the commitment to a set of entities that exclude what you do not want to be part of so um I would like to see I think as as a I guess as a developer I would like to see more research and um and development in this field um especially because I see a lot of talent coming into into ZK and they're looking for interesting things to build on and I think so take a look at what uh that Andre and idea and summer um where we use the kg complements as

the as a accumulator for the set instead of Omega trees um yeah so to speak up on this point about the the operations we have what what would you like to see you know as well as additional multiplication what what do we want I mean personally like first I'll pick up about the membership groups because actually we're building a voting vertical and basically that's what you need there like you need to approve membership but there is um in our particular use case because we are based on air C20 and nfts like for we need to prove that the person had nft at some point of time in ethereum blockchain and there I would personally would like to see connection between storage proofs that then converts to membership proofs where you can then

efficiently prove that you're a member of a set like this is a cool trick we learned where you don't need to do all the storage proofs all the time just convert to a friendly format and then do Boost for that and I think kcg or vertical trees have the new popular term would be quite efficient there as for operations I find that like there are two levels of abstractions like ones is when you for example program something in how to or like very low level languages it's like circum where essentially all you get is multiplication and addition and some lookups if you're lucky but then right now languages they're highly languages we try to abstract it and I think they're doing good enough job of just providing you almost all the tools you want something like Noir uh um and Cairo maybe and other ones so I

think it's getting better like for developers to jump into all right so do you want to if you have anything you would like to see in that in that side [Music] um actually I think um that's the direction we're heading to anyway when you see for example how an arithmetization system like turbo Blanc allows essentially for wild custom Gates um I think people have identified that this is what we need in order to to solve a range of different applications and ideally I would like to see any possible gate I'd like to see everything being built

in any case I think we're going into in that direction what about um what about modularity it seems that now we're getting different types of proving systems are you know picking piece of technology as well as you can you can pull fry into a Blog and things sort of thing is that is that the direction we're going in should we should we be looking at that or should we should we stay more insulate what's the advantages for that uh can we talk about like this and one pages I guess uh potential like risks because like uh I mean voting is a really good mechanism but in practice it has a very variations like Implement in different ways like you may vote by your quantities but you may vote by your let's say steak inside a block of a

blockchain and actually that's how like a proof of stake works right uh the moderators assign different uh verification tasks based on what's your assessed with the stakes and here and and comes the problems that Okay so what about that militias uh like users militias blogs Publishers knows that have acquired open PAD a certain portion of this block so that's because like when we talk about votings because we are always

believe that okay so there are some malicious ones but they are a few so actually we can believe that the majority vote out the good choice but question just in practice it is not like we can see that if a malicious users has manager to occupy more than uh one third of this block there's no blog can be or that new attestation can be verified it can prevent everything and if this number is far beyond uh two of two officers he can decide everything because he himself it is the majority and actually even if it doesn't reach the threshold

of like one third it can still like attacks or split still the tokens we need that with vary as more a slash risk so that is the potential risk and actually we have seen uh recent such kind of incidents in in Mev but mostly it is because of some programming bugs but yeah yeah but yes other people say well let's say what disadvantages they say in in these areas which yeah what limitations then do you see in uh in zero knowledge what's a

I guess for my sides uh I'm more of a like developer using uh different protocols but all of them have slightly different implementations there are too many implementations of the same thing and sometimes someone forgets something and it's very hard to audit all the tools to make sure that what you're using is actually secure because imagine there's at least five different uh proven systems and then there are also different commitment schemes and you can as you said do model combinations of this in different ways and everyone decides to build like most of people decide to build their own stuff because they can do it better but then actually uh it's not like I'm I'm not sure if everyone did the audits of that like very thorough audits so it's a bit uh create like as a developer I'm a bit I'm

confident about the tools right now right what I'm using I'll put this way this way for example there was a perfect talk about simulation extractability and that's for example some tools just forgot to add a few lines and then that's what happens but something that's really encouraging is that if you look at events like CK hack and ZK Summit that happened two weeks ago um a personal observation I have is that there's a lot of new uh developers coming to the space from different backgrounds so some just fresh out of college and some with like more than 10 years of experience uh jumping into ZK because they think that this is a very valuable field and a lot of developers

are not afraid of the of the learning curve and I think that's fantastic um there is a lot of um chaos I would say chaos in terms of like documentation lack of good tutorials and stuff like that I think an organization that's really helpful in this space is 0x Park which organizes workshops Summits um and online online um sort of like guided classes they've done a few of those um definitely check them out um yeah I I I want to say a few things I mean precisely I think this field is extremely exciting there's a quantity of things to learn and explore that that is

maybe unheard of the thing is as more people like come into the field we also need to make sure that they understand what the assumptions are on some of these proof systems on some of these commitment systems so that they can use them in a secure way because many times as you said they come from different fields maybe they don't have a mathematics background or they don't have a computer science background and if we do not State clearly here we are making this assumption about the security of this protocol when we use it in combination with this and so on and so forth maybe people start like taking pieces of of of of proof systems from here and here and then it turns out that the resulting protocol is completely

insecure and and and we need to make sure that the tools we are building are just secure for everybody to use and everybody understands what the security assumptions are and on that point there is there's akproof.org that try to build standards around zero knowledge do you have experience do you follow these do you have experience of these I don't know how widely use these are I think I think it's it's really early and standard civilization is difficult and lengthy but they do a fantastic job of bringing people together I've seen the output of those conferences it's very interesting I um just a general Point um is that I think we are precisely

because we are early the narrative in the public hasn't been set yet so as an example a narrative for crypto today is not so good because you had a lot of things happen last year we read it is now associated with negative things like um in the past it was like a black market in today's episode it's also the scams but we have not seen this with ZK we have not seen a narrative Trend emerged for ZK in the general public which is a good opportunity because if you just talk to any person who's not a technologist about what is your knowledge it's very intuitive that they want they see that this technology can protect privacy and thing that's fantastic because everybody knows that

the traditional system has failed in protecting people's privacy if you go to the US like your social security number is is already on on the dark web or like you can type in your name and find your address and the five addresses you live in previously so people know that their privacy is is not uh protected in traditional systems and this we need to use we need to leverage this uh moment to to establish a strong positive narrative around CK yes you said um about the areas that are important I think right now already with what we have we have like 80 percent of what we can use to do a lot of the first like

tier of ecosystem expansion uh but going forwards from that a very big important topic is shared computation like or shared State Etc so for example and okay um and you know you've got stuff like homomorphic encryption and NPC Etc as techniques and how do these techniques fit together so one thing I have a design for is anonymous liquidation engine so if you're trading derivatives with people and you have a position then when your position gets liquidated you the the platform needs to be able to know that info so that it can be able to claim

back the uh your position so that it can build this Insurance Fund which actually is a protection against going insolvent um so you can't do that with just purely ZK right now you need to add in some other techniques like one way you could do it is you could say oh you know every position has a bond and there's an incentive when you get liquidated to release your bond but then that's capital inefficient and so then that's when you start talking thinking about things like NPC or shared execution and a lot of now there's an interest in delegated proving so that you could have for example mobile phones and the way that it does that is basically with NPC like editions

of free but multiplications are expensive so all it does is it tries it uses these algorithms which minimizes all the multiplications to like one step or a minimal number of steps and it performs the addition on the servers and then it sends back the result to the wallet and the wallet does some multiplications and sends that back and the service finished the computation um in terms of the other sort of technologies that we could link with what what uh what do people uh want to see what would you like to I mean you mentioned you mentioned someday what other things do people and would people like to see um sort of or multi-party computation or I mean if if she becomes usable that would be amazing

but um wait was it sunscreen or uh they're built in this so hopefully uh they'll be quite successful but again their biggest Downs uh our hurdle right now is zero knowledge proofs so again getting more efficient there would be very important and I think this still gets it proven is actually one of the key ones because it's hard to imagine that everyone will have a hardware specific for ZK proven so it'll be a lot better if special servers somewhere have that and then your phone just sends it to these servers with the special hardware and does the job uh I would like not on technology side but in general I would really like to see a lot more collaboration during the project because again the space is very new and it's very important that it stays safe and secure and there are no big

vulnerabilities discovered after the protocols are deployed so there was a collaboration between scroll and Pauline Hermes for example trying to educate Auditors and that was a great idea what they did because like it's projects joined together to make sure that what they're building and the grant vision is going to be secure yeah yeah for me as well maybe not on the technology side but maybe more what I'd like to see in the future I think everything is kind of a trade-off you're essentially always in the Dilemma between the efficiency of your protocol and the security of the proof system and the parameters you're choosing for transparent proof systems for example this conflict is essentially between the security of your of your protocol and

the verifier time and I think what I'd like to see in the future is this conflict being resolved in favor of security actually yeah okay I think we've come closer then July opens up to questions now if people have any questions for the panel we've had you uh sorry lame question uh I think so try to put yourself in the view of uh both the user Who currently uses uh um ethereum or or any of the l2s um and who cares about you know their privacy and uh in um uh you know protocol developer

um who's deployed something on one of the one of the dl2s and cares about the privacy of their users which currently is obviously horrible um at ethereum admissibly uh so what would you recommend you know such kind of people to to look at like the the emergent um you know Technologies whether that's l2s or development toolkits for the you know for for the protocol author or you know for the user um any you know anything that kind of catches your attention that's worth exploring and feel free to pick just one of these positions Amir obviously you would uh you know uh specifically but I'm saying uh crypto is

very scammy and corrupted a lot of these VC funds they uh are beneficiaries of the cantillion effect where the Central Bank prints money and they LPA into these big funds and they have a check box like you know when you're in school and there were kids that got A's on all the exams but they weren't particularly smart and that's how that's they go like shiny website check big team check uh uh web share electron GUI check but then then you dig the fundamentals of the tech and there's nothing there and a lot of these funds they invested are over inflated valuations at the Peak at the peak of hype hype Mania last cycle and then got burnt like when these projects

went nowhere and their Founders have like 24 million mansion in Lisbon and a private jet so don't look at the Mirage of where everyone's looking look for teams they have a strong culture the the the they have an idea or a vision of what they're doing because small teams are like one good Dev is better than 20 devs so that's what you have to look for is find those Corners where where they exist in crypto but they're not the people that are competing to get the money who know how to trick uh people who know trick the big copper and also too I guess yeah for me

um if as a developer if you just want to integrate some with something already working and that's probably the best approach for you just again Google what you want and uh or ask one of the experts and we'll suggest you exactly what team for the use case because there is seven o'clock and other good Solutions which you can just plug and play and you don't need to do any ZK and that's actually good for you because you can if you mess up that's very bad if you're more hardcore get like find languages they're actually about 12 languages the good to start is usually circum because it's old and it's been there for a while and you'll understand how it works but then it depends how hardcore you can get you can go to Hello 2 or you can go to full-fledged Solutions like

again Noir and all these fancy languages that just try to abstract it as much as they can okay pretty good time for just a quick question if oh no one more question so I think that one of the reasons why there was such an enormous productivity in the crypto space nearly a decade ago that the tool the cryptographic toolbox have just boiled down to these two Primitives the hash function and the digital signature that you need to trust and everything else can be built on top of those and we got like marvelous things like ethereum and this ZK technology seems to be

flowering with Primitives like we have range proofs we have membership proofs and so on and there's always this problem that even though we have some Universal uh Universal Primitives but they're not really efficient and somebody clever always comes up with something specialized that does it much cheaper and better and do you think that there's a way out of this forest and we're going to see some crystallization into some small set of Primitives that people can trust and then people who don't even understand those Primitives in that but know what they actually achieved we should go in the opposite direction

this is actually a good thing it's the architectural level where we need to put the focus not on the cryptographic algorithmic level because they have trade-offs and it's not it's not it's not good like for example even now there's like a functional encryption there's a lot of really interesting things with that it's for example the you can you can create these encrypted filled elements and you can only perform one function on those encrypted elements to get a result that is decrypted to a key so but then and and that uses a number of different like I mean there's different constructions but one I was looking at uses a number of mathematical techniques but even those techniques are

modular and so sorry so you can use them in different I'm sorry we're going to have finished but please everybody thank the thank the panel [Applause] [Music]