 my experience is technical and today we have kind of things on session and I'll share what it'll take, what kind of lessons we learn from the hyperledger, what kind of like, you know, what we can use based on society and religions we've developed like for a while. And like, you know, this presentation is not related to the hyperledger in general, but it's very small aspect of it, like, you know, regarding like encrypted computation and the regarding zero knowledge proof basically regarding the data privacy and not necessarily only for it's a political point like average application. So let's start. Yeah, let's go. So the presentation was divided like a section, so like we're going to discuss actually like, you know, three main aspects. So the full homomorphic encryption and homomorphic encryption in general, that's an encryption that allows you to encrypt data first, and then run run the computation on that already encrypted data. So meaning like, you know, they would allow to like, you know, execute like various computational machine learning. So we have data belonging to one party and like, you know, the model coefficient of the models belong into another party and you don't want to share anything. So basically the idea here is just like, you know, that's a technology that would allow you like, you know, to compromise and then you privacy to share the data. So very easy example of homomorphic encryption, like let's say like, you know, you guessed the number and like, you know, I guess the number and like, you know, you encrypt this number, like, you know, independent there on our sides. And like, you know, we can add those numbers together and when you decrypt it, you can write it. So we actually see like a final sign. So for like, obviously, like, you know, one person is quite easy to predict. But if you have like, this is three participants, that's you already have a privacy, you know, meaning like, you know, you don't know nothing. Okay. And smooth. And like, you know, homomorphic encryption, like, you know, it's a very urgent market right now. And there are quite a bit of work done. And like, you know, obviously, this is very interesting to it's a technology to look to privately look in someone else's data without compromising security. So some work, like, you know, we did like related the fraud detection, like multiple banks can like, basically, let's say a TD web, like, you know, things, this transaction may be this credit card transaction problem. But I'm not quite sure, like, I would like to get a feedback from like another part of business, let's say RBC. And basically, and just like, you know, can privately encrypt the data and just like, you know, ask model to compute the final result. So basically, like, you know, for the fraud detection, for example, the steps like, you know, even here. So basically, again, like, when you encrypt the data, you have to use the public key, public key that it's everyone has a public key to secure. And the private key is a private key. It belongs only to the person, the organization that's, that's, and it's not shared. It's not shared. So, and like, you know, in the fraud detection mechanism, like a, let's say, we see this transaction from the client and then the, the, the bank and certain that this transaction may be a problem. And we input transaction and send it to bank B and then he encrypts the model, the model they have. And like, you know, and then apply this model to the encrypted data. So the results, like, you know, no one is seeing anyone's data. And at the end, you can see a final result. And the bank A can only see yes and no, convenient transactions, all of them are not convenient, like, you know, no information gets released for like, you know, for bank A or bank B only. So basically, they're going to produce it, like, you know, the visibility over it. Okay. So same thing, like, you know, you can apply for, like, not the credit check. So I don't want to go into this, but it is similar because like, you know, when you're doing like an off credit check. So we have this organization, this like, let's say, like, you know, Rogers, you just, you purchase them when you serve and send, like, you know, you're going to purchase a plan. And the part of it, like, you know, you have to provide like a credit history. And like, you know, we do have all this, the mechanism for this credit history, but it would be independent. Like, you know, you can independently ask like another banks, and they can, like, you know, look into the transaction history. And they need an Everson is encrypted, like, you know, all the data is encrypted. And since the data is encrypted, like, you know, you can simply calculate about it. And then if you're probability, like, you know, that's what you said, the right side, like, you know, is your credit score is lower, but the school are not basically to approve yours now. So basically, whatever the multiple party can like, multiple parties, like, you know, I know, like, we always can calculate, we always can interpret the data first time and ask about to do any form of calculation. Okay, so to generalize this idea, we can like, in our business, because all data is encrypted, not compromising any privacy, we negotiate it in the final result. And like, in a final result, in the market basis, it's just a single number, or like, even full and buy, let's see this and know. So the user, for a more different course organization, who would actually merge, like in the market, AI markets, where like, people can trade their AI models without showing that, because they will get second encrypted data. And like, you know, people can test those models, like by putting, like, you know, the data encrypted for these models. And like, you know, look at the results, just like an encrypted data. And so this is like an old liberation opportunity to like, you know, create the like, ranking market like, you know, and technologies like a blockchain, and later, and like, you know, that in particular, will be here for information exchange, prior information exchange, that will spend problem, meaning we're not cheating anyone, as well as the constraint system. So I mean, like, for instance, I'm looking through how we should like, you know, some rate, for like, let's say, 70% of the time. So and in order for, in order to go from harmonic encryption into the AI market, like AI model market, you actually need to have an ability to contribute and compute, and also you need to have a constraint system. So meaning like, you know, the system should follow certain rules. That's what we call smart contracts for like, you know, for blockchains. And since everything is encrypted here, so the smart current contract seems to be smart enough to deal with that encrypted data. And for that, we actually like, you know, we actually can use another innovation, zero knowledge proof. And I believe it's the next slide. Oh, yeah, that was too fast. But it was like, you know, people get, and if you look into, like, two examples, to use cases like, you know, where we can like, apply harmonic encryption for and like, you know, how to get money out of markets. And now they actually look into like algorithms. So in a fully harmonic encryption, basically, everything is encrypted. But like, you want to make it practical, meaning like, and we should be able to search it. So meaning like, for instance, like, if we have like, no number, let's say five, we encrypted, like, you know, and so then we had another number, let's say five again, like an encrypted, and then we subtract encrypted five from encrypted five, we have zero. We have zero. So when we decrypt it, like, you know, we know, basically, it's, it's basically zero meaning the numbers are equal. But let's say we have like, you know, one that the numbers are not equal. So basically, in this case, like, you know, if we actually like an algorithm to extend this technology for like, you know, string search, for instance, then we like, and we need to compare, let's say, letter by letter, or word by word, if that's how you encrypt, like, basically, you can encrypt like another string by numbers, by letters, by individual letters, or characters, or you can take like, the back of words, let's say, like, and encrypt like, you know, one, two, three, like, and then basically, you like, you know, start subtract and start comparing them. And meaning like, you know, you basically, you have to compare every single element in it, like, you know, to get a final result. So, and as you can see, the complexity of this operation is, it's a linear complexity meaning, so it's quite, quite, it's like, you know, when you have like longer text, it's just like, you know, quite costly. So what we can do about it. So in the computer science, everything is graph, or trees, like, you know, mostly graph. And like, you know, when we think about like, you know, optimizing like any sort of like a computer science problem, we actually, like, you know, the first thing we think about, we think about the graph. And in our particular case, like, you know, we actually can use a prefix tree. It's, we just call it a trice, I'm sorry for the spell. And the trice is spelled here. No, no, man. So, but like, in a century, like, you know, you see like, you know, on the left side, you have like, you know, like the words, but they all represent it as a tree. And like, you know, we say overlapping, like an elements in it. And like, you know, this number of leaves becomes to be much less in here. And also we see like, you know, when we see trees, complexity, like algorithm complexity comes from O and into, into, into the logarithmic complexity. So in this particular case, like, you know, it's going to be like, you know, look at, and that's one way, like, you know, how to speed up computations and you can hit next. So this one, like, you know, it's another way of how you can speed up like, you know, computation over encrypted data, because to recap, like, you know, when data is encrypted, you have no idea what's inside, but you still need to use, like, you know, some form of data structures to look into it. Because this is like, you know, algorithm, it's actually used on firewalls and a network of logical bloom filters. So the bloom filters, like, basically tells you if your number or like, string collection, some kind of data element is within is basically exist or not. Meaning it's more like a probabilistic structure. But like, you know, if the element is not existing in a set, in this kind of like a bloom filter string, it always will be negative. If the element, if the, and in order to, to compute like, you know, this filter is just basically we just like, you know, simply the hash, the, the, the hash of the, let's say, string call, like, you know, some kind of data element, and keep adding them using XOR like binary operation. And the bloom filters becomes to be quite big, but like, you know, but they represent like, you know, data compression. And in a sense, so they could be like, you know, used to like, you know, build some sort of database. And like, look into it. So you can see that data structures here is again, three, meaning like, you know, we go into logarithmic complexity, meaning like, you know, we decrease in like, you know, computation speed like a lot. And like exponential. Next. Yeah. So this one is kind of like, when we deal with voice data, or like, you know, kind of vision data, or like, you know, any sort of like wave data. So we're actually dealing with waves. And like, and like speech, for instance, processing like, you know, speech detection, like a voice detection, like a mechanism. And basically, we deal with the signal processing. And the signal processing is essentially like, you know, any signal, like, you know, whatever, like, for instance, a vision with a computer vision, or just like a human vision, or hear it. So it's kind of like, you know, could be, could be transferred, could like any signal could be transformed into a sign, or sign and cosine that like, like, you know, would be a number of sinuses and cosinuses. And like, you know, when they add them up, like, you know, we'll get like, you know, our, our like, you know, initial signal. So we're doing it by actually by applying the Fourier transform for it. But in keeping mind, it's just like, you know, any signal is nothing more than the sum of sine and cosine. And the full homomorphic works, works only with a multiplication and addition. And like, you know, the signal processing becomes to be a very good fit. Like, you know, in homomorphic encryption, because actually, like, you know, you can before encrypted it, you can present it as a frequency domain. And then like, you know, and then apply all sort of computation. Okay, let's move forward. So like, you know, when you're dealing with the language models, like, any language models, the language models, they, like, charge it is like, you know, very popular. Like, you know, everyone get excited about it. But the idea about any language models, most of language models is to use something called engrams. So I mean, you know, someone asked questions briefly, they asked, if you're someone, if you could explain just a little bit more about that, about what someone sort of seems to be previous. So like, basically, the idea of the Fourier transform is actually going from the time domain frequency. And like, going to be doing it by decomposing like, you know, any sort of like, any sort of signal, whatever is just auto signal, like any sort of signal, into the sum of sine and cosines. And like, you know, for instance, like, you know, when we're dealing with like, you know, the audio filters, right? So the idea is just like, you know, to do a cut of, of certain frequency. So meaning like, you know, when we actually transform in everything from time domain, like, you know, the signal into the frequency domain. So in the frequency domain, we'll have like, you know, the magnitude of the frequencies. And basically, but they will add up in the main signal. So what we can do, we can subtract them. And then the harm and we can subtract them in the harmonic space. So meaning like, you know, all this kind of sort of like, you know, filtering, like, you know, we can apply in a different domain, in a different domain, in the frequency domain. So like, you know, we can do like, you know, let's say, like, you know, filtration in a, in a, in a, like, remove like a certain frequency, or we can analyze voices, because analyzing the voice is also kind of like, you know, frequency related thing. It's called capsule transform. But it's just one more level of Fourier transfer. But essentially it's decomposing like a signal in the sum of sine plus cosines. And like, you know, and the, the source of that, like what you encrypt, you actually encrypt not a time domain signal, but you encrypt like, you know, the, like those kind of like sine and cosines, you encrypt frequencies. Once it's encrypted, like, you know, you can do whatever you want. You can just calculate, because keep in mind, it's just like, you know, signal is nothing more than just a sum of like, you know, harmonics. I know it's just like, there's quite a bit here to cover. I try to explain as much as I said, but like, you know, the idea here is here, it's basically, it's one of the, it's one of the use like, you know, you actually can apply like, you know, this kind of homomorphic encryption, not only in like, you know, so looking something in the database, but like, you know, create a homomorphic encryption filter for like audio processing for like visual processing later, we'll cover quantum computing as well. So going back into the language models. So in the language models, like, you know, all the language models comes into something called ngrams. ngram, like in this one, it's basically a language model is trying to predict the next world, the next word based on a previous word. And it's called one gram. So two grams, meaning like, you know, based on two words, we predict them like, you know, the next word. And that's basically, that's how child DPD works. That's how modern black neural network work. And then like, basically, and what we do basically, we just like, you know, create like, you know, the neural network, and the neural network is a symmetrical neural network with a bottleneck in between. And in the input would be, let's say, two words and output would be like the next word we're trying to predict. And like, you know, and the neural network actually like, you know, squeezed into like, you know, something called like in the middle, middle, middle, middle, there are many layers here. And this middle layer, it's called like laden in space. It's a bottleneck is like, you know, and also called embeddings. And the neural network is all about the set of coefficients and meaning like this kind of like laden in space, it's just nothing more than just vector. And like, you know, when and charge EPT gives you like laden in space to like, you know, as many of like, you know, other language models, like, you know, at least using like neural networks. So since it's a vector, actually, like, you know, start dealing with, with initial language processing as a vectors, for instance, like, we can you can spot that men and women are sitting very close in the left upper corner. And like, you know, King and Queen, they sit in very close too, because they actually have very, very similar in meaning. And like, it's, it's basically like the one of the most interesting thing about those vectors that like, you know, you can basically you can add vectors and you can subtract vectors, you can do like an operation on the vectors. And meaning like, you know, you can do like, let's say, like, for instance, man plus, man plus woman minus King will be equal to Queen. So if you just like imagine those kind of vector computation. So and that's a bad, that's why the way it's a great power of a charge EPT that I don't see much information, like, you know, people are talking like lots of charge EPT in form of like, you know, language models, like generation, some advice, generation with some kind of like no text, or give me the advice, but the really, really, the true value of charge EPT is in this kind of embedding space. You also can through API is accessible. So you actually can like, you know, do the same thing. And like, you know, you actually like, you can see like, you know, how many things are related one to each other. And like, you know, on top of this one, you actually can do like, some kind of new predictions, like, you know, by calculating those kind of vectors, the one they just sent, like man, plus woman, minus King will be equal to Queen. And, you know, that's a great power of like a charge EPT. So the reason this slide is here, all this stuff could be done in encrypted space in a homomorphic encryption. And the computation I just mentioned, could be done in encrypted space as well, like, you know, basically before doing anything, you just encrypt and then you calculate. So it would be encrypted man, plus encrypted woman, minus encrypted King, it would be equal to encrypted twin. So I know it's just like, you know, here's very a lot to cover, because it's a machine, it's a deep learning and like, you know, charge EPT and like those kind of language models. But the general idea, like, you know, using like, we still kind of get excited with charge EPT, but like, you know, only here, like, you know, very few sounds like, you know, about the data privacy. Yeah, we'll, we already have data. We already have a good models for like language models. But like, you know, we don't have an ability to share them confidentially. Can move forward, please. So this is a quantum blockchain safely, safety. And before we get into the quantum problems, so like, you know, and the quantum things like, you know, let's like, for some reason, like, you know, all the explanation about quantum computer is like, you know, really tough. It's like, like, when I read into it, it's just like, you know, supposed to be that difficult. So because a quantum computer was developed like, like 50 years ago, even more. And the idea is like very simple. So we have like, you know, we have a wave. So we have a wave, and not one way, we have a multiple waves, all we do like, you know, we just shift the wave a little bit here and there, we call phase shift. And we just add them. And if the way if the shift like, you know, let's say here is like, you know, on a lower and the high level. So then like, you know, we have so that we have a cancellation effect. So if that basically if they say in, in, in not in phase, if they actually can opposite phases, then like, you know, you have a cancellation effect, because you can see those harmonics would actually like, you know, would be canceled by the asset harmonics, because they just shifted. In the same time, if the signal is identical, and we just simply add them together. So there would be amplification effect. And whatever the quantum computer is doing, it doing nothing else, just these two operations. So get the harmonics and shift it and add it. And that's it. That's all that's only thing this quantum computer here for. So the physics behind it, like, you know, the implementation is very difficult. But the idea is like, you know, very simple. It's just based on the Fourier transform I mentioned before. It's just like, you know, really some of harmonics and some of the harmonics that the shifted. And if you can move next. And like, you know, and the here we actually talking about like, you know, one of the algorithm Paul Schnoer, a short algorithm. So this algorithm is actually promising algorithm that breaks elliptic curves, and that's supposed to break the hyperrelagia privacy and the theorem how like, you know, breaks Bitcoin, then like, you know, all other stuff. So the idea behind like, you know, this mechanism is very simple. So we like, you know, the the elliptic curve, the encryption behind the let's say theorem of Bitcoin is just like, you know, simply like, you know, you have two prime numbers. And one prime numbers acts as a private key. Another one acts as a, like a message. And basically, we just do an exponential operation, like, you know, and the, and just to revert it back is very difficult. Like, you know, it's, it's not an easy problem. But like, you know, it's called kind of discrete logarithm problem. So and like, you know, how to break it, you're actually like, you know, the Schnoer, the person, like the scientists, they came up with an idea. So basically, you have to, you can present those kind of schemas by through this, like, you know, number of mathematical formulas or transformations, you present this problem as finding the period within the signal. And like, you know, and as I mentioned before, finding the period is just simply like, you know, finding an oscillations. And that's what essentially, like, you know, basically, the idea, like, you know, once we have it in that form, you see on the left side, we just need to like, you know, we need to do a face shift. And like, you know, we need to measure, basically, we need to amplify those things we are looking for. And we actually need to cancel, like, you know, the other stuff we are actually looking for. And that's it. Like, you know, once you have this kind of like period, you know, from this period, you actually can say like, you know, what, what, what, what is my private key. So this kind of idea here is like, again, very simple, but to get into it is just like, you know, about 10 pages in a paper. Yeah, if you can move next. So like, you know, the kind of summary, what makes a data privacy important in hyperrelagia fabric? So hyperrelagia fabric is not a single blockchain. Indeed, is a network of blockchain. It's actually the most advanced blockchain, like, you know, ever been developed, because it doesn't represent a single blockchain. And it actually, it's a blockchain of blockchains. And the privacy in the, in the, in the hyperrelagia fabric is implemented through that like membership service and membership service, meaning like, you know, the, like, you just define like through ACL access control list, who is actually like, you know, participant of this network is not participant, this network. And it partially like, you know, using like, you know, the permission and partially using like, you know, cryptographical membership, meaning CA certificate authorities. And it really becomes messy, like, you know, when we actually need to exchange information for the parties that not a part of that, that that's not included in this current current channel membership. And for this kind of like, you know, people, the hyperrelagia fabric, for instance, is offering something called side channel side databases, like, you know, private databases. And it's really kind of like messy how to set it up. And it's like, you know, it increases complexity of solution a lot. It's supposed to simply like, you know, just get the homomorphic encryption, or zero knowledge proof, the one we actually will talk a little bit later. And like, you know, and with that, you actually can use any blockchain, you can use a theorem, like, you know, unless there's a quantum safe, like, you know, you can use any form of a blockchain, you know, the cost of development becomes to be much less. Because like, you know, you have to invest into the initial cryptography, but then the mean of communication doesn't matter. So zero knowledge proof is a one we're going to be talking in a bit, like, you know, it's slightly different from homomorphic encryption, it's more like a constraint system, as supposed to be a computational system, less homomorphic encryption really is. So and AI is a blockchain, like, you know, really, like, you know, zero knowledge proof and homomorphic encryption can leverage you and AI market and AI market is like, is someplace like, as I mentioned before, where you can actually say, and here's my model, and here's my constraint. And like, you know, on that data, it gives certain results. Yeah, and the last one, quantum computer and blockchain safe, he will talk about in a bit as well. Can you go next, please? So here's a slide about quantum computer again. We kind of like all scared that like, you know, that once quantum computer is developed, all our Bitcoin wallet becomes empty. It's not necessarily true, like, you know, because like, you know, you can protect this one in the numerous ways. Like, you know, first of all, like, you know, keep in mind that like, you know, blockchain, sorry, quantum computer has like, you know, really, it's a linear system. The quantum computer can only add and like, do nothing else. Like, you know, substitution becomes to be as like, you know, like phase shift. Eventually, it's the same as like an addition. And like, you know, there are about if you just check Wikipedia, like, you know, there are about only 16, like known algorithms are known for quantum computer. It's like, like the message here, quantum computer, when it comes out, it will be like a video card is something extra, something nice to have. It won't ever replace it like, you know, the conventional computation things. So basically, just like, you know, something that like, you know, doing like an external algorithm. And again, this number of algorithms are only about 16. And like, you know, they're very limited. And they all linear. So like, most of our solution is not linear, like, even like, you know, AI is linear, like most of AI stuff is not linear. So and like, you know, for blockchains, like, you know, basically, once blockchain has a smart contract, you can implement very simple thing, we can actually like, you know, like, you know, put in a smart contract, some hash of the random number you're going to hide. And when you're going to spend your money out from your blockchain, you have to provide this random number. And like, you know, that's the known by you. But at this point, like the smart contract could verify that like, you know, that see the committed values equal to this hash. And that's it. And that's it. That's a fix. Because like, you know, hash is known to be like, you know, post quantum thing. And like, you know, this, this kind of little bit tricks here and there. And like, you know, we still live in a more less like, you know, save world. Even like, you know, when the quantum quantum algorithm becomes practical. Okay, can we go next? Yeah, this is like everything. And then encourage you have like, you know, a few publications on the medium, talking about different things. So like, you know, I covered zero knowledge proof like very briefly. But like, zero knowledge proof is kind of like, you know, what we have. Basically, it's a private constraint system. And like, you know, you can think about like, you know, zero knowledge proof of homomorphic encryption. But with homomorphic encryption, you can do a computation versus zero knowledge proof, like, you know, that will really gives you like, you know, constraint system where you can verify like, you know, certain constraint based on an existing data. And like, you know, if you need more, like, you know, I have an article about it, like, you know, hands on of proof of ownership. So an example here is just like, you know, we have a mutual fund. And this mutual fund, like, you know, has a list of assets. And a list of assets are known, what is not known, that allocation for this assets in a mutual fund itself. And like, you know, you have a like, you know, practical example, like, you know, in Python, to, to basically to see, like, you know, how this stuff works. And there are a few methods I'm not going to cover them. So like, you know, and then I have like, you know, a couple of articles on like, you know, on the basics of encrypted computations. So the main thing that on encrypted computations are like a computer science problem, like graph theory, I would say what was presented because by itself, it's like, sounds very easy, like an encrypt calculator. So but like, you know, to make it practical, because it's not fast, it's slow, because it's just dealing with just like, you know, a certain type of cryptography called lattice cryptography. So and you have to like, you know, move it from the linear space into like a logarithmic space to make it faster. And yeah, like, you know, I have quite a bit of, I spent like, you know, many years, we're trying to get like, you know, that blockchain, like in different of blockchain working for like an enterprise. And like, you know, we have a couple of successful projects, I put my kind of like, you know, as a developer, I put my kind of lessons in this running, like lessons and like, you know, failures, like mistakes I've made. What basically what not to do in this article, like running that blockchain in production, what it takes. Yeah. And I think like, you know, we can go into the, the questions at this point. Okay. Okay. So we are in the moment for questions. If there are any, you can just please unmute yourself and then ask the question in the room. Any questions? It's really communication practical, right? So basically, basically, it could be built as a, as a part of the smart contract. So because it's just a library, right? So you just compile it in, you just like, you know, pass a public key and you do encryption and basically, and in computation and like, you know, when you get data out, you just use your private key to decrypt it. So in, in other way, like, you know, you basically, you can just like, you know, publish a row data, like, you know, on a blockchain and get cloud, like, you know, some form of API to do a computation because once data, like data ownership is important. So basically, once you put in a hyperledger, you, the data ownership is guaranteed, meaning like it's a double spend problem, obviously. But then, like, you know, you add the cloud or like, you know, any other system, like, you know, can take data from like, from the proof, proof source and do a computation and let's say you can submit the data back. It's like, it's like the same as dealing with the raw data. True. So, in a real life example, like, you know, something has been in blockchain in quite a few years ago. So to understand how the security works correctly in the blockchain systems, in a real life example, can you give me something, you know, something you have worked on or something you have seen where a security breach has happened or where the encryption that you spoke of has been put to you some, some kind of such situations? So, like, dealing with identity is like one of the case. So, like, meaning like, you know, you have a social insurance number, social security number. So, and you, like, you see, like, when you think about the driver license, like a driver license has your picture that could be homomorphically transformed through that signal processing, the one I mentioned, and it has an attributes like, you know, validity date and blah, blah, blah. So basically, the representation of, like, a driver license is really a graph of data. And once it's a graph, like, you know, you can do a computation on it. Once you can do a computation, basically, once it's a graph, now you can make an encrypted graph. Now, the government, let's say, like, you know, the owner of this license can, like, you know, encrypt this data. And now actually, you can do, like, you know, certain computations. For instance, like, check that you go to a bar, like, you know, you didn't need to ask 18 years old, you just have QR code and this QR code represents the thing. And you just, like, you know, do a computation and gives you yes and no. Same with just like, let's say, last name. And same even with your picture, because there are like, you know, algorithm of like, like, you know, face similarity, you can do some more for you too. But the beauty of it, like, when you start thinking, like, you know, when you're dealing with that encrypted data, you actually like, you know, you know, dealing with the real data. So you have to come up with the data structure. So and then basically, you have to come up, how are you going to compare, because at this point, like, you know, your input is encrypted, and your output is encrypted, and then as a data, you send this encrypted to. So that's, that's one more point on that. So, so let's say we have two sets of, I mean, there's the source encryption, and there's the call as the result of encryption. So this, the decryption, or what you call as the check of the data in a validation. There should be some kind of a key structure, right? I am supposing there's a public key, private keys kind of. Yeah, yeah, yeah. In common morphic encryption, there is a public key, the one. Yeah. So in that case, how, I mean, in your opinion, how would you say, how secure that that process would be? Because considering the non blockchain or the non, you know, the traditional systems that we have been using like 10 years ago, comparing to what we are having right now, like you said, you know, when you're implementing something like this. All right. What do you, what would you say, how secure, like, if you give a scale of security, what do you say? So you basically, you don't, in security, you never trust anyone. So don't trust me. So when you have any security concern, so there is this committee called NIST and this committee has exactly what you, the answer is like, you know, they have like, you know, metrics for like different level of security. So main concern right now for blockchain is a quantum privacy, which I kind of like would not agree 100% on this kind of threat. There is a threat for sure. But there is like, you know, different ways how to solve this problem as well. But in the same sense, like, you know, everything, whatever you do has to comply and like comply with what comply with the standards and the standards are published. Okay. And like, you know, they have like latest cryptography is not yet NIST certified, not yet, but it's very, very close. So and the committee gets are like, you know, I think they're like every six months. And like, you know, there's a certification committee. But you see, like, from the business opportunity, think about like stock often, right? So when like, you know, when you say like, you know, two years from the maturity date, the stock option costs very little, because like, you know, people actually don't see quite a bit of business because it's like lots of risk. And like, you know, but when a staff gets mature, right, like, you know, you have to office, you have to hold it, or you have to give it to someone else. And like, you know, with this one, like, you know, there is a market cap for everything. And granted, like, you know, it would be really hard to execute like, you know, kind of like the fraud detection in a bank, in a way, because like, you know, the cybersecurity people like the security people would ask show me the certification. Oh, like, yeah, we will be like tomorrow, but we'll come tomorrow. So the latest cryptography is basically, but in the same time, it's kind of not many people doing it many times, like, you know, we hear just like, you know, to explore like, you know, what could our life be in the future, and how we can make a profit about how we can generate the profit out of future, because the future is the most expensive asset, right? One question online says that many of these problems can at least be kind of more efficiently with trust executed environments. Yeah, yeah, there is a like basically Intel puts a lots of effort on it. CGX. Yeah, Intel SGX. Yeah. And we have started to see what was the question is, what would you consider at least? Basically, like, there's a solution like, you know, hyperledger fabric, hyperledger fabric is based on like, you know, the using the zero knowledge proof as a security add-on. This is like an accordion, which is another flavor for blockchain. They actually use like an SGX technology to do encrypted computations. And like Intel SGX technology is like really a kind of a separate isolated processor with just like, you know, the master key baked in it, it's not extractable, and it would allow it to do any form of computation inside. But you see this one, like, you know, tend to have like, you know, many vulnerabilities in the past. I was like, you know, so, I was so happy, like, you know, when I saw it first time, because like technically you can apply any sort of homomorphic encryption, like it's kind of very slow. Because first of all, like, you know, the latest cryptography is dealing with matrices. So that was a form of homomorphic encryption, just like an expanded one number is another number. So the SGX gives you like, you know, everything at the speed of like, you know, more than CPUs. But like with this kind of so many leaks, so many hacking, and like, you know, so many stuff, like that happened with this kind of, and basically once it's compromised, you there's no patch. You have to throw the hardware away. Just like, and with this kind of thing, at least you can re-encrypt, you can do something. But if your master is compromised for this processor, that's it. Like, you know, you have to invest in like, you know, a new processor and your motherboard and like, kind of, we'll see. But so far it's kind of, to me, it's not reputable yet. And like, and the hacks going not from that month that behind it, the hacks going from like, you know, inner process communication, from kind of some circuit, like vulnerabilities, from kind of some side effects, not from the elliptic curve mathematics, not from the encryption, they come from, from like in the manufacturing process. And I would rather stay on academic side, as it's just like, and as mixing, like putting too much technology in it. And, and like, you know, once you have this kind of becomes to be like IoT problem. So this kind of bulb, like, you know, gets hot, everyone gets hot through this bulb. And the only way to fix is just like, take it out and throw it away. That's my view. I've done like at least dozen POCs on it. It was a really good thing, like to use in the Azure. Because that would make everything like an encrypted but like, sorry, I lost my confidence level in this, JX questions. And you mentioned smart contracts. So have you in the process of building smart contracts? Have you given it to third party to audit the smart contracts or not? Technically, you have to just like, you know, basically, especially like, you know, when dealing with ICOs and like, you know, when dealing with all the stuff. We don't have like, you know, basically, when we're doing like, you know, like a regular development, like, so we rely on the pipeline, part of the pipeline is just all different kind of automatic scans. And unfortunately, for smart contracts, there's not such a thing. There is a SonarCube like enough, but it doesn't really, yeah. So basically, you basically, you would, basically, you would publish it and you would like pay about this like someone to hack it, like, you absolutely need to have a review because this technology is still immature, even with the Ethereum blockchain Bitcoin, Bitcoin, you can make smart contracts too. So I would not let smart contract going until it publicly kind of seen, not even privately by independent company, but publicly, because smart contract is here for a public to see the set of rules, right? So the constraints. So one of the things that happened was before Chad GPT came, we used to give the public contract for a third party company who specialized on auditing. And well, after different phases, we had different pipeline features to release, and we updated according to that. And funny thing what happened was we asked the Chad GPT to audit it, it did pretty well. Did it? Yeah. So it showed the mistakes, human errors regarding the verbs and the number of items also. So we were in dilemma, like, should we just go with Chad GPT audit or should we give the auditor? Because the auditor will charge a huge amount to just audit a single of days only. So before getting in, like, you know, we see humans, like, you know, they can do stuff, like, you know, basically in Chad GPT, there is a parameter called temperature. Temperature is a measure of entropy, and entropy is a measure of randomness. So with this kind of, like, you know, temperature, like, you know, you can control, like, you know, how good, like, you know, how generic the Chad GPT answer is. But in your case, like, you know, what I would do first before getting to anyone else. So there are lots of, like, you know, smart contracts out there. And remember, when I was talking about Chad GPT, you actually can access embeddings, or, like, laden and spacer embeddings. This is a vector. And, like, the way we, and basically what that means, you can take all the contracts from, like, whatever you see the contracts from, like, Internet, like, anyone you can find, you can go through a Chad GPT, you can get a vector, and you can run the Chad GPT through a smart contract, you get a vector, and measure two vectors as an angle between them. Yeah. So, like, you know, meaning, like, you know, if you have, like, you know, big angle, so that it's because in Chad GPT, those embeddings, it's actually great. Like, you know, it's a big consciousness, like, you know, behind the stuff. And basically, in this case, like, you know, when you measure an angle, you will see which smart contract from the public, it's similar to yours. So the widest, the better, the more real. Yeah, a sign is just like zero, like, one, it's like the exact match, zero, there's no match, right? It's opposite from sign, right? So basically, and you just compare everything, everything, and get this kind of, like, matrix, remove, like, you know, what is just, like, less than 80% and, like, you know, that would allow it to, to, like, you know, reduce, like, you know, number of, and what is a matrix, matrix is a graph, those kind of things. And then you look into a graph and then you look in the clusters, you know, like in a membership membership algorithm or cluster algorithms. And then from there, you actually can see which group of smart contracts you contract belongs to. And also, you should be able to see which volunteer, sorry, vulnerabilities, this contract may expose because similar contracts, they have, like, you know, this vulnerability is to, and with all this information, you just go into researchers and present it to them. And they just, like, you know, put their own thing, because, like, you know, you never, like, human is human, right? Especially, like, you know, people who actually doing this, when you're reading some papers or like, man, you just, like, the papers five years, five pages, five pages of paper, but you can see, like, it's covered here and there from this dimension. You can see, you can, you can really people's life. You can, here's one here, here is another two years, and the third page is another five years. So, they kind of still, like, basically, my understanding, like, you know, charge GPT is good for something that generic for like something for ordinary, something you like vulnerability, they should be ordinary, but they would never probably notice something that they charge GPT network never seen of. Yeah. So, a simple example is a linear regression. So, like, linear regression is like, you know, really, it's a basically, it's like, you know, coefficient multiply or weight coefficient, weight multiplied by the number and you sum it equal dot product, right? So, in this sense, like, you know, basically, you basically, you get like models of weights, and like, and you just encrypt the weights first, and then basically, you take like data, and you just like multiply one to each other, and you sum everything, and you got like, you know, final result. If the question is like, you know, if it's usable for training, because like, you know, training the model is not like a really a question, it's a question of like regularization, because you have to count like, you know, entropy. And like, I don't think homomorphic encryption, like, you know, there are some papers and there are some projects, like, you know, they let you use neural networks in an encrypted space. But like, for instance, I work with like linear stuff. And basically, you train your model on the plain data. And then you just use it on on on another party data. And you have like, you know, result and you just send it out. So the decision trees could be implemented that way too, because decision trees is just like one zero. Yes, no, basically. And neural networks is like nothing more than like a bunch of submitters. So, but it's another topic for like, for for quite a bit of discussions. So the github is the best friend in that. Thank you so much. So thank you guys. Thank you for everyone coming. Everyone online. Obviously, it's been recorded all to go to hyper ledger on YouTube, and get more information. So if you have any questions for Alex in the chat, we've put his is a LinkedIn account so you can contact him and have any questions. Is anything else but also ask David and also thank you again, David. You always put in the the extra work in. And thank you for this event, David. And thank you everyone online and in person for coming. Hope to see you guys at our next event. And just to stay noticed, it's on the meat event. Sorry, the meat website just just keep on checking in. And when we have the next event, make sure to come in. All right. Thank you. Thanks, Alex. Thanks, everyone. All right. Take it. Thank you guys.