WEBVTT
00:00:00.160 --> 00:00:05.919
But uh at the time of the deployment, we find like the the proof generation is like too small.
00:00:06.160 --> 00:00:15.039
Even we use a very large AWS uh instance, it's still like very slow, which like inspired me to design some special hardware for that.
00:00:15.279 --> 00:00:27.760
For instance, like I want to say like transfer probably like 10 10 bucks online, but I need to prove to the channel or like to some like uh authority that my bank balance is like more than$10.
00:00:28.480 --> 00:00:35.679
So like I can prove that part, but without releasing the exact uh balance number in my bank account.
00:00:36.000 --> 00:00:38.079
What are the implications of this technology?
00:00:38.159 --> 00:00:49.600
Is the ability to selectively disclose specific bits of information about yourself over a digital space, which is you know historically really hard to do because you you produce so much data when you're uh over the internet.
00:00:50.000 --> 00:00:52.479
Seems like in vacancy, you don't need to trust anyone.
00:00:52.560 --> 00:00:57.039
Like you just uh trust the math, and the math like you can do it by yourself.
00:00:57.359 --> 00:01:02.719
And we are not creating like a honey pod like for uh for the hackers to hack.
00:01:03.280 --> 00:01:06.239
Hi everyone, you're listening to the Blockchain Socials podcast.
00:01:06.400 --> 00:01:14.799
I'm Josh, and I am here today actually in potentially my new studio that I'm going to be using uh going on into the future.
00:01:14.879 --> 00:01:18.640
So more news on that that I will share uh soon.
00:01:19.120 --> 00:01:24.560
But before I go into that, um I wanted to introduce my guest for today.
00:01:24.799 --> 00:01:34.959
His name is Leo Fan, and he is the uh founder of Cysic, which is a super interesting project um building called an on-chain compute network.
00:01:35.120 --> 00:01:42.560
But before I butcher trying to explain it, um maybe Leo, if you want to give a quick introduction to who you are and what CySic is.
00:01:43.040 --> 00:01:43.599
Yeah, sure.
00:01:43.840 --> 00:01:45.840
Um my name is Leo Fan.
00:01:46.000 --> 00:01:49.760
Like I uh I got my PhD from Cornell in cryptography.
00:01:49.840 --> 00:01:59.920
After Cornell, like I did some uh research engineering job at NIST and Agrand, then started as an assistant professor at Rutgers.
00:02:00.400 --> 00:02:16.400
Um yeah, so when I was at Argoland, like uh we we delivered the first uh decade bridge became Agrand AC and Agorland Semana, but uh at the time of the deployment, we found like the the proof generation is like too snow.
00:02:16.560 --> 00:02:22.319
Um even we use a very large AWS uh instance, it's still like very slow.
00:02:22.560 --> 00:02:29.360
So which like inspired me to do something to like uh design some specialized hardware for that.
00:02:29.520 --> 00:02:33.599
Um yeah, so the the word like CISIC is a combination of two words.
00:02:33.680 --> 00:02:38.879
The first is like cipher, uh and the second part, the SIC is like coming from like uh ASIC.
00:02:39.280 --> 00:02:44.240
So it's a combination of like building some specialized uh hardware for cryptography.
00:02:44.400 --> 00:02:46.639
So that's like how we like started Cysic.
00:02:46.960 --> 00:02:59.759
Um and like while doing Cystic, like we we do a lot of things, like we do like the GPU and FPG acceleration, and we also did some like work on ASIC design for um for the K.
00:03:00.240 --> 00:03:09.919
So maybe just to help listeners, just in case they don't know what some of these terms mean, ASIC is an application-specific integrated circuit, if I'm not mistaken.
00:03:10.000 --> 00:03:11.599
It's kind of like specialized hardware.
00:03:11.759 --> 00:03:14.240
Often you hear about it in the context of mining Bitcoin.
00:03:14.319 --> 00:03:26.319
You uh you you get ASICs uh hardware in order to uh mine Bitcoin really, really well because it's a specialized hardware that's really good at specifically the thing for um mining Bitcoin, for example.
00:03:26.560 --> 00:03:31.919
Um and then uh if I'm not mistaken, cipher is also a cryptography term.
00:03:32.080 --> 00:03:32.319
Yes.
00:03:32.639 --> 00:03:36.800
Um that I believe is like, I if I'm not mistaken, is it it's the key?
00:03:36.960 --> 00:03:41.520
Is the cipher the key in like a cryptographic secret kind of way to explain it?
00:03:41.840 --> 00:03:48.240
So like in the encryption part, like the encryption world turned in this like painted into ciphertext.
00:03:48.400 --> 00:03:51.280
So ciphertext is uh encrypt kids uh plaintext.
00:03:51.840 --> 00:03:54.240
So ciphertext is kind of like the secret itself, I guess.
00:03:54.639 --> 00:03:54.800
Yeah.
00:03:55.039 --> 00:04:21.040
Um so jumping off a couple of levels from you know uh assuming you guys understand a little bit about cryptography, which is just the math around uh sharing secrets and uh saving data and memory for computers, um zero knowledge cryptography or ZK cryptography is a different type of cryptography that's a little bit more advanced and is kind of on the on the bleeding edge of things.
00:04:21.439 --> 00:04:30.000
Um maybe you can explain a bit what is zero knowledge cryptography and why is it you know, why is it expensive to generate and and use right now?
00:04:30.399 --> 00:04:41.199
So we can like treat the ciphertext we just talk about as encryption or like some representation of the plaintext, but in the in the ciphertext way.
00:04:41.279 --> 00:04:49.600
Uh so like for ZK, you can think about like we so they k to computation is what ciphertext is to uh plaintext.
00:04:49.920 --> 00:04:57.360
So like we basically compress a lot of like a computation into a like very tiny uh B string.
00:04:57.519 --> 00:05:14.160
And this B string will represent this like computation, but this like B string is not like the harsh stream, they still have some like structure inside which will enable people to trust you if they verify that the the B string you handle to them uh satisfies some requirement.
00:05:14.560 --> 00:05:20.639
So that's like how you can see uh the the K the knowledge part.
00:05:20.800 --> 00:05:26.399
So it's like uh a representation of the of the computation you just uh carry out.
00:05:26.720 --> 00:05:27.040
Yeah.
00:05:27.199 --> 00:05:31.279
And and why it is like useful, it has like two nice properties.
00:05:31.519 --> 00:05:37.519
The first one is uh their knowledge less, and the second second one is the succinctness.
00:05:37.759 --> 00:05:47.199
So their knowledge less means like it will preserve the like the privacy uh and without releasing the privacy, the private part like to you.
00:05:47.360 --> 00:06:03.759
Like for instance, like I want to say like transfer probably like 10 10 bucks online, um, but I need to prove to the say to uh to the channel, like to some like uh authority that my bank balance is like more than$10.
00:06:04.399 --> 00:06:15.040
So like I can prove that part, but without releasing like how uh like what's the exact uh balance number like in my eye in my bank account.
00:06:15.199 --> 00:06:27.120
So like I handle you this like uh this like zK proof, and then you can verify and then trust that my bank balance is like is larger than like 10 bucks, and then you can do the transfer.
00:06:27.279 --> 00:06:30.319
So that's so uh the knowledge part.
00:06:30.480 --> 00:06:32.560
The second part is succinct.
00:06:32.800 --> 00:06:50.480
So the succinctness means like the proof is like very short, it's probably like several hundred bytes or several hundred kilobytes, so it's a very small uh proof, which um which like you can use a proof to encode and like a very large net computation.
00:06:51.040 --> 00:06:53.920
So that's the first part of the succinctness.
00:06:54.160 --> 00:07:01.839
The second part is although the proof generation is like takes a lot of time, but the proof verification is like super, super fast.
00:07:02.079 --> 00:07:13.920
You can verify the key proof on your cell phone like in less than 15 milliseconds, and with a very very little uh computing resource, then you can do the verification.
00:07:14.000 --> 00:07:24.160
So next as I say we have this like MySIC app, which you can download it and then to verify the proof we generated like in part of our network on your cell phone.
00:07:24.319 --> 00:07:24.720
Yeah.
00:07:24.959 --> 00:07:30.160
So that's like two nice properties that their knowledge has and succinctness of like the key.
00:07:30.720 --> 00:07:30.959
Right.
00:07:31.040 --> 00:07:31.199
Yeah.
00:07:31.360 --> 00:07:44.240
I think one of the ways that I've also heard people try to explain Zero Knowledge Proofs is the example of like in the United States, for example, where if you want to go inside a bar, you have to show your ID, uh, your driver's license.
00:07:44.399 --> 00:08:02.879
And you know, you give your driver's license to the bouncer, and he doesn't he not only sees that maybe you are or are not old enough to enter your above 21, they know exactly your birthday, they know your full name, they see your address, they see a lot more information than maybe they necessarily need to know.
00:08:03.120 --> 00:08:13.680
So this kind of it's it allows you to say to the bouncer, I'm over 21 without giving any other information, without getting any any leakage of unnecessary information.
00:08:13.839 --> 00:08:22.240
So it's kind of like a way of doing selective disclosure, I guess is this this is the the zero knowledge part at least aspect of zero knowledge proofs.
00:08:22.319 --> 00:08:48.000
And I think that's why it's interesting from uh you know, what are the implications of this technology is the ability to selectively disclose specific bits of information about yourself over a digital space, which is you know historically really hard to do because you you produce so much data when you're uh over the internet and everybody, you know, if you if you own the platforms or if you own a lot of the the piping between everything, then you know uh maybe a bit more or you've given out more information than you intended to give out.
00:08:48.320 --> 00:08:48.399
Right.
00:08:48.559 --> 00:08:48.799
Yeah.
00:08:49.519 --> 00:08:54.240
Um and then succinctness, it's also good for compressing data.
00:08:54.720 --> 00:08:54.960
I guess.
00:08:55.440 --> 00:08:55.919
Yes, yes, yeah.
00:08:56.159 --> 00:08:57.120
It's kind of how I think about it.
00:08:57.360 --> 00:08:57.600
Yeah.
00:08:58.320 --> 00:09:10.480
Um are there any like um maybe specific real-world examples of zero-knowledge proofs and cryptography kind of um in use that like you know the average person may be able to relate to for you?
00:09:10.799 --> 00:09:11.120
Right, yeah.
00:09:11.200 --> 00:09:28.320
So like if so, like we all know like the uh I uh like both uh Bitcoin and Ethereum, they're like public ledger where like all people can check like the the balance or like how how many coins they have like in one uh wallet address.
00:09:28.480 --> 00:09:34.559
But there are some like some privacy-oriented uh blockchains such as like Zcash or Alio.
00:09:34.879 --> 00:09:59.679
They like if you do the they can basically shield the number of like tokens like you have in your wallet so that people cannot check the exact number on all the transfer, probably I would say like 80% of the transfer also like happen privately and which preserves some privacy of like uh the traders or like some other guy like who use the uh the blockchain.
00:09:59.919 --> 00:10:02.720
So that's like on the the knowledge nice part.
00:10:02.879 --> 00:10:05.759
And the second part is on the uh succinctness.
00:10:06.080 --> 00:10:09.519
We can use the succinctness like to scale up another blockchain.
00:10:09.600 --> 00:10:24.480
So like for take Ethereum, Ethereum for uh for instance, like if like how Ethereum like uh validates the transactions that they re-execute all these like transactions in Ethereum.
00:10:24.720 --> 00:10:31.120
So that's like which is like why they put a guess limit for for the like for a block.
00:10:31.600 --> 00:10:38.080
But in instead of like re-execution, you can generate a zK proof.
00:10:38.320 --> 00:10:51.360
Where like since like a zeker proof is like one proof, you will generate and everyone can verify it, and then if it pass the verification, then uh then they will like trust like the the statement in this proof is true.
00:10:51.600 --> 00:11:01.519
So instead of like re-execution, you can just verify some zK proof, which proves uh that all the transactions uh they are like valid.
00:11:01.679 --> 00:11:06.639
And the verification of like the zek proof takes a very, very small uh guess.
00:11:06.799 --> 00:11:14.960
So which which means it will increase the guess limits, which you can like have like more transactions in in one block like for Easily.
00:11:15.120 --> 00:11:21.600
So that's like two uh nice applications of uh they can in blockchain.
00:11:22.799 --> 00:11:23.120
Right.
00:11:23.360 --> 00:11:24.559
Privacy and scaling.
00:11:24.720 --> 00:11:40.960
I think kind of I think what's what's important to understand for a lot of for some users, at least when I've tried to explain cryptography with them is that like a lot of cryptography can be used for both privacy and scaling, and that they're kind of like um the same property.
00:11:41.360 --> 00:11:54.559
Yeah, like the same property of a th of of of this math that you're that you're using in these systems can be used for like things that you don't think are related, but uh they end up being related in in these types of systems, which I think is really, really interesting.
00:11:54.639 --> 00:12:01.360
It's kind of like a when I realized that I was like, oh, this is why it's like you know, quite expansive what what you can think about doing with this.
00:12:01.600 --> 00:12:01.919
Yeah, yeah.
00:12:02.320 --> 00:12:09.840
So you mentioned that like proof ZK proof generation is um fairly expensive or it's clo costly, it's kind of slow.
00:12:10.000 --> 00:12:17.279
Um it's still you know a type of technology that people are figuring out how to do um better, faster, et cetera.
00:12:17.440 --> 00:12:22.799
Um but I'm curious why, you know, where where does that like cost come from?
00:12:23.039 --> 00:12:31.600
Uh why is ZK proof generation so much slower than like say you know other forms of of cryptography that we're already using today?
00:12:31.919 --> 00:12:43.759
Yeah, so like I I guess uh other crypto tools like we are you using today is like uh encryption, digital signing, and also probably like some hash functions.
00:12:43.919 --> 00:12:50.799
So like compare against the these tools, the uh the data handled by uh by the cable is like we much longer.
00:12:50.960 --> 00:12:59.600
So like a typical ZK statement will handle this like degree of like very large uh uh degree polynomials.
00:12:59.840 --> 00:13:04.000
And the the degree is unit, like two to the 24 or something.
00:13:04.159 --> 00:13:06.399
So it's a very large uh polynomial.
00:13:06.559 --> 00:13:13.279
And for and the coefficient for this polynomial is also uh 256-bit none.
00:13:13.519 --> 00:13:17.840
So it's a large polynomial with a lot of like uh data in there.
00:13:18.000 --> 00:13:23.919
And handle this kind of polynomial will require enormous uh computing resources to do that.
00:13:24.080 --> 00:13:39.840
And also the computing resources, like when they do the uh process, they also need to communicate with each other, which means like you also need to have very high bandwidth to support uh the transmission of the data between this like processing unit.
00:13:40.000 --> 00:13:53.039
So that's the like two most important parts in the competition, like very large uh computing resource and then and a very high uh bandwidth to support the processing.
00:13:53.279 --> 00:13:58.320
So which also explains like why it's quite expensive to generate a data proof.
00:13:58.480 --> 00:14:03.840
You need to have a machine that's like powerful enough to do both things like very efficiently.
00:14:04.159 --> 00:14:04.399
Right.
00:14:04.559 --> 00:14:09.279
So it's just uh basically it's more mash, yeah, is uh is the simple answer.
00:14:09.600 --> 00:14:09.759
Yeah.
00:14:10.000 --> 00:14:11.759
Bigger numbers, yeah, bigger numbers.
00:14:12.720 --> 00:14:18.879
Yeah, bigger numbers uh takes longer time to both like computes and also to to transfer the thing.
00:14:19.360 --> 00:14:19.679
Right.
00:14:19.919 --> 00:14:33.840
So I guess uh so this is kind of like my understanding, at least, is that then um you know, like your average computer hardware is not necessarily built at the moment to be generating zero knowledge um proofs.
00:14:34.159 --> 00:14:34.559
Yes.
00:14:34.720 --> 00:14:47.679
Um and so I guess this is kind of why, if I understand correctly from what you guys are doing, uh that you guys also provide hardware, like a a network of hardware for people to use to generate these proofs.
00:14:48.000 --> 00:14:51.759
Um yeah, do you want do you want to talk a little bit more about the hardware side of things?
00:14:52.000 --> 00:14:52.240
Yeah, sure.
00:14:52.399 --> 00:14:54.240
CySink is a compute-file network.
00:14:54.320 --> 00:15:00.000
So by compute file, we mean like we turn this like computer resources into some early machines for you.
00:15:00.240 --> 00:15:08.559
You can plug in your like servers, your laptop, or even your cell phone into our network to do something something like in the computation.
00:15:09.120 --> 00:15:19.200
And of course, like we categorize like uh each we categorize this hardware into the different groups, which is like responsible to do different things.
00:15:19.360 --> 00:15:26.879
Like for instance, if you plug in your like server, public is powerful enough to generate some DK proof in a short period of time.
00:15:27.120 --> 00:15:32.799
And we also use this like uh servers to do some like AI inference for us.
00:15:33.039 --> 00:15:35.840
So that's like the server side.
00:15:35.919 --> 00:15:57.279
If you plug in your like laptop or even your cell phone, which is like not powerful enough to do the the two things I just mentioned, like then we assign this like um they key verification, AI verification job to you, which like you're not very powerful uh computing how uh hardware still can handle that, and also some other things like uh yield.
00:15:57.600 --> 00:16:06.480
And all the things were wrong in the in the back end, so you won't be able to notice any difference like in using your hardware.
00:16:06.639 --> 00:16:06.879
Yeah.
00:16:07.360 --> 00:16:07.600
Right.
00:16:07.679 --> 00:16:07.919
Yeah.
00:16:08.000 --> 00:16:40.720
So there's uh there's generation, the proof generation, which is taking in all the inputs that make up your your zero knowledge proof, and you know, depending on how you how you're uh um like setting it up, you create you generate a proof which is like kind of a reference to you know all these inputs and gives you a kind of answer, but then that answer has to be verified by some by you know the person that receives the answer to double check to know that this proof is saying what you it's claiming to say.
00:16:41.600 --> 00:16:47.679
And if I understand correctly, the generation is much more uh compute intensive than verification.
00:16:48.000 --> 00:16:49.120
Yes, yes, that's right.
00:16:49.840 --> 00:16:53.600
So there's these these these two types of compute, uh verification.
00:16:53.679 --> 00:16:59.279
You know, if you got uh probably any any probably laptop or or phone can do verification.
00:16:59.440 --> 00:17:04.720
Uh the generation maybe you can do, but it will take a lot longer than like you know, a beefier machine.
00:17:05.039 --> 00:17:21.119
Um but you guys are basically, if I understand correctly now then, you guys are creating a kind of network where anybody can can bring whatever hardware they got and they could potentially make a little bit of money off of it because other people are paying for the verification and the generation of ZK proofs uh on the network.
00:17:23.200 --> 00:17:24.079
Nice, that's super cool.
00:17:24.160 --> 00:17:33.759
It reminds me of I don't know if you remember one of these early, I think I think that was uh Golem maybe was like one of the first kind of attempts to do something like this, if I'm not mistaken.
00:17:34.079 --> 00:17:34.799
I think there are other ones.
00:17:34.960 --> 00:17:41.599
It reminds me of like these proof of useful work blockchains that you used to have back in like 20 uh 2017.
00:17:42.559 --> 00:17:43.519
Yeah, maybe yes.
00:17:43.920 --> 00:17:52.960
Yeah, so so that's like yeah, so at that time, like people were saying like uh all this like Bitcoin mining is like a waste of uh computer resources.
00:17:53.039 --> 00:18:01.839
It's like you produce uh some like specialized hardware which will also cost a lot of like electricity in computing the hash.
00:18:02.000 --> 00:18:14.400
So people think like he probably instead of doing that, like we can use use uh like the resources, whether it's like in electricity or computing power to do some like real jobs for human beings.
00:18:14.640 --> 00:18:20.400
Yeah, so that's uh time like in 2014 till 2017, I think.
00:18:20.559 --> 00:18:33.680
Yeah, so since this since like my like before uh coming to Cornell for PhD, and like I did some like Bitcoin mining uh back then, yeah, for for uh for several years.
00:18:33.839 --> 00:18:43.119
So like and the idea of like using some specialized hardware to accelerate the case, also as you mentioned, like this idea was also inspired by Bitcoin mining.
00:18:44.079 --> 00:18:44.960
Okay, nice.
00:18:45.359 --> 00:18:52.319
Um is uh actually um which uh network does Sysek uh connect to?
00:18:52.480 --> 00:18:53.440
Which blockchain networks?
00:18:53.680 --> 00:19:00.000
Uh yeah, so right now we are like proving um so it also depends on the market.
00:19:00.559 --> 00:19:16.240
So so problem like last year, like SySek is a biggest prover for for like several uh large like nail tools, including uh scores, equinq, yeah, things like the these nail tools, they their job is to scale up Ethereum.
00:19:16.400 --> 00:19:26.319
So they generate a lot of like transactions and they need awesome, which means they need a lot of like computing resources like to generate the key proof for them.
00:19:26.640 --> 00:19:42.480
And like so, for instance, like we successfully like reduce the proof generation time from like 1.5 hours to about like 20 20 minutes on a two-car uh 249k uh GPU card like machine.
00:19:42.640 --> 00:19:52.720
Yeah, so so that's why like we so this like layer tools, they they want to work with us to reduce the burden of like generating vaccine proofs.
00:19:52.880 --> 00:19:58.400
But like uh so that's like something uh that's a situation for us uh for next year.
00:19:58.480 --> 00:20:03.119
But as you can see, like all these layer tools, they are not doing very well right now.
00:20:03.359 --> 00:20:03.839
Yeah.
00:20:04.160 --> 00:20:13.839
So so like we shift like uh some of our computing resources to support uh Ethereum and also to support some perp uh perp decks.
00:20:13.920 --> 00:20:21.119
Um yeah, this perp decks, they also need to generate a ZK proof for each uh for each transaction on their decks.
00:20:21.599 --> 00:20:28.079
So like so that's also like a very large number to not large number of like ZK proof to generate.
00:20:29.359 --> 00:20:35.680
Okay, are these perp do they verify those ZK proofs on chain or are they just generating a proof?
00:20:36.000 --> 00:20:46.400
Um so like like our process is like we we like thin, we we can talk about the lifespan of like a ZK proof in inside network.
00:20:46.559 --> 00:20:57.440
So so like for for each like task we receive from say like this like perp decks, we pick like several like random uh proofers uh to generate the proof.
00:20:58.079 --> 00:21:06.720
And then like we pick a large number of like verifiers, which like on your cell phone or on your laptop to verify the proof.
00:21:06.960 --> 00:21:23.039
After a proof is like verified, say by uh eight out of like 10 verifiers, we call it this proof is like settled, then we then we put it on uh sizing network and uh and also additionally we send the verifier proof to to the perp decks.
00:21:23.200 --> 00:21:44.720
So the proof is already like settled in our chain, and some of the problem, like I I guess uh the the perp decks or like they don't have the energy to verify like each proof, they just do some like sampling on the proof uh already settled by us, like to do to check the uh validity of the proof.
00:21:45.119 --> 00:21:45.759
Yeah.
00:21:46.960 --> 00:21:47.599
I see.
00:21:48.240 --> 00:21:48.720
Interesting.
00:21:48.799 --> 00:21:59.759
So you have like uh the network also has a level of kind of redundancy, I guess, where you you want to you want to be able to have trust in the in the proof, and so you have.
00:22:00.799 --> 00:22:06.720
A few at least people run the same proof or verify the same proof uh to make sure that it's valid.
00:22:07.039 --> 00:22:07.200
Yes.
00:22:08.480 --> 00:22:09.200
Yes, I'm sorry.
00:22:09.680 --> 00:22:10.079
Yeah.
00:22:13.359 --> 00:22:17.039
So maybe we could talk a little bit more about compute fi.
00:22:17.519 --> 00:22:18.480
Uh I think it's interesting.
00:22:18.640 --> 00:22:44.640
I think there's there is like this uh uh general I mean I've just I I've I've never really in engaged with any of this, but I've seen some stuff of like um you know people trying to financialize a bit of their their like the computes on their on their computers to sell that as like future compute used in AI inference or something like that.
00:22:44.880 --> 00:22:53.359
Um I I'm wondering if you can explain a bit like how how do you see ComputeFi, what is it, and like um what does this future hold for us?
00:22:53.680 --> 00:23:03.279
Yeah, so uh as you can see, probably like there are like several authors similar platforms such as like iodanets or some some some others.
00:23:03.519 --> 00:23:16.960
So but but the like the iota nets or like some other, they are just creating a platform where like people plug in their uh servers there and and uh wait to be like rented or like used by others.
00:23:17.039 --> 00:23:28.319
So it's basically like AWS or like vast uhai, this kind of platform, but they are in in the blockchain or crypto worlds.
00:23:28.400 --> 00:23:35.279
So like SciSec differs uh from these uh competitors in a very significant way.
00:23:35.440 --> 00:23:48.720
So like in addition to creating uh creating this kind of like platform, we also bring some like real jobs uh to be computed uh by the by the servers in size network.
00:23:48.960 --> 00:24:06.319
Say, like for instance, the the key pool generation is like one thing, and also we have this like uh a lot of like AI products on SISE network, which also consume the under uh consume the tokens uh it's like generated by the underlying machines.
00:24:06.559 --> 00:24:18.240
So like in addition to just uh uh creating a platform, we are we are bringing some like real jobs for the for the servers on SISE network to process.
00:24:18.400 --> 00:24:20.640
So that's a very big, big difference.
00:24:20.799 --> 00:24:21.440
Yeah.
00:24:23.359 --> 00:24:57.599
Do you think like in the future, you know, one one potential well, I guess the thing that kind of you're describing, just to step back a bit, is almost like a you know, this this uh it's almost like a meme to me, but like this meme of the decentralized cloud, so to say, of like computes or computers being able to connect from anywhere to take part in this network and to provide that infrastructure in a way that is resilient uh and redundanced, at least at par with you know other types of cloud providers.
00:24:57.839 --> 00:25:07.359
Um and that people would be able to kind of like you know uh have almost like have have like make money off of the data center you have at home.
00:25:07.680 --> 00:25:08.319
Yes, yeah, yeah.
00:25:08.720 --> 00:25:17.920
You know, rather than necessarily the mass build-out of giant data centers that's maybe going on right now.
00:25:18.160 --> 00:25:36.160
Maybe one alternative is we could use the excess computes that we already have available to us with everyone's consumer hardware rather than needing to overbuild you know all these amounts of data centers, this could be one kind of alternative uh to that.
00:25:36.559 --> 00:25:36.799
Yeah.
00:25:37.039 --> 00:25:37.920
Yes, that's right.
00:25:38.400 --> 00:25:38.720
Yeah.
00:25:39.359 --> 00:25:54.160
So yeah, are there any other uh I guess the um yeah, are there any other non blockchain related ZK um examples that you're you're particularly interested in?
00:25:54.559 --> 00:25:56.000
Um Right, yeah.
00:25:56.160 --> 00:26:02.559
So like as you you can uh as you mentioned before, like the ID thing is like one uh very good example.
00:26:02.720 --> 00:26:07.039
Like in addition to showing up in like your ID, you can just uh generate a digit push.
00:26:07.359 --> 00:26:14.000
Uh saying like you are like uh older than 21, then you can enter the uh enter the uh the bar.
00:26:14.240 --> 00:26:27.680
And another example is like there are some um events like um say like probably like last year or the year before before last year, there is a there is a death count in in Thailand.
00:26:28.079 --> 00:26:40.240
And when like if you are like a local like in in Thailand, then you can apply for some uh apply for a special discount on the ticket for for the event.
00:26:40.480 --> 00:26:50.400
And how how you like proof you are like a local like is like previously like you just upload your your passport and then you are showing that you are local.
00:26:50.640 --> 00:26:55.119
But like that will reveal a lot of information you don't want to reveal.
00:26:55.279 --> 00:27:21.440
Uh so like very nice way of like using ZK to wrap around it, like you download a software and then you generate the ZK proof uh of your passport showing that you are uh from like Highland, and then you upload only the ZK proof to uh to the TK website, and then then after they verify it, then you can go the like special discount.
00:27:21.599 --> 00:27:30.960
Yeah, so that's yeah, it's basically digital ID, and you can use a digital ID like to do a lot of like things, yeah.
00:27:32.079 --> 00:27:32.400
Right.
00:27:32.559 --> 00:27:48.160
I and this is like I mean, this is super interesting to think about just in the current context with the amount of legislation that's being passed right now, um, requiring people to prove that they're you know over 18 or whatever to use certain platforms over the internet.
00:27:48.400 --> 00:27:51.759
I mean, it is something that um is being pushed.
00:27:52.000 --> 00:28:04.240
Different people have different different views on like whether or not it's like the right approach to kind of actually solve the problem that they that they think they're trying to solve around uh protecting kids.
00:28:04.480 --> 00:28:22.480
Um and so yeah, I mean ZK it it what what is interesting to me is that like at least as far as I could tell, a lot of the the technical uh specs of a lot of these identity solutions from uh I think with particularly like the UK and Australia, I know, had some, and I think the US is working on something.
00:28:22.720 --> 00:28:28.400
But um these these identity solutions don't include ZK, as far as I know.
00:28:28.559 --> 00:28:54.480
Um which is interesting, because you would you would think that, you know, um if if you actually wanted to solve the actual problem that they say that they're trying to solve while also not causing any externalities, aka creating a giant honeypot of people's data in someone's databases, like then zero zero knowledge proof should be like your first that should be like one of the first things you should be considering, you know, right?
00:28:54.720 --> 00:28:54.880
Right?
00:28:55.440 --> 00:29:07.759
Yeah, so I yeah, so like we are pushing like the vacay public to the traditional world or like the the mainstream to make people aware of that and then like uh to use to use it.
00:29:08.000 --> 00:29:15.279
Since like in vaccine, like you don't need to trust anyone, like you just uh trust the math, and the math like you can do it by yourself.
00:29:15.519 --> 00:29:20.960
And we are not creating like a honey part like for uh for the hackers to hack.
00:29:21.200 --> 00:29:31.359
They are just the vicate proofs, and they keep proofs are like due to the nice properties of vaccine, you basically cannot get anything out of like of the kid proofs there.
00:29:31.440 --> 00:29:39.200
Yeah, it's it's like more reliable, more transparent, I would say, like compared with the current solution.
00:29:39.359 --> 00:29:39.839
Yeah.
00:29:40.240 --> 00:29:41.200
Yeah, right.
00:29:41.839 --> 00:29:44.240
Um so yeah, I don't know.
00:29:44.319 --> 00:29:56.240
Call I I think it's like we're at we're at a really funny stage in in politics almost where like you know, it's like call call your senator and teach them about zero knowledge proofs.
00:29:56.640 --> 00:30:02.480
Because what they're proposing right now in this legislation like is not is not privacy preserving at all.
00:30:02.559 --> 00:30:10.799
It like really highly highly depends on uh you know someone to build a database that is like just going to get hacked.
00:30:11.039 --> 00:30:13.759
It's not a matter of if, it's a matter of when, you know.
00:30:14.000 --> 00:30:14.160
Yeah.
00:30:14.319 --> 00:30:18.640
Yeah, it's just a matter of time to to hack this kind of database.
00:30:18.799 --> 00:30:19.440
Yeah.
00:30:20.720 --> 00:30:21.359
Yeah.
00:30:21.839 --> 00:30:28.720
Go go to your capital and teach your uh government officials a little bit of math, a little bit of uh zero knowledge cryptography.
00:30:28.960 --> 00:30:30.880
Um yeah, thanks, Leo.
00:30:30.960 --> 00:30:35.759
Is there any are there any other uh last things that you wanted to mention before we close it off?
00:30:36.079 --> 00:30:49.119
Something like we uh we also want to mention is like we open source like the DKVM uh uh code repository, like you can try it out, and we have a lot of like products on on SciSec.
00:30:49.279 --> 00:31:05.039
Uh you can just check out like SciSec uh XYZ for like uh for products and then you can follow us on Twitter, like it's uh CYSIC-uh XYZ to for the latest updates like from our team.
00:31:05.200 --> 00:31:09.680
Yeah, so we have a lot of like interesting products ongoing.
00:31:10.079 --> 00:31:10.480
Cool.
00:31:10.640 --> 00:31:18.319
And how how can can people right now today be able to join the network to be able to make a bit of money off of their hardware?
00:31:18.559 --> 00:31:21.359
Yes, yes, it's like total uh permissionless.
00:31:21.599 --> 00:31:36.400
Like you can use your like uh even if you don't have a very powerful server, you can always use your laptop or even your cell phone to join, you can download like my Sec app on your whether it's like Apple or like Android phone, you can always try it out.
00:31:36.799 --> 00:31:37.759
Cool, great.
00:31:38.000 --> 00:31:44.640
Well, thanks, Leo, so much for coming on and appreciate you explaining Zero Knowledge Proofs to us and sharing about Seic.
00:31:44.880 --> 00:31:45.440
Thanks a lot.
00:31:45.599 --> 00:31:47.200
Thanks enough for having me here.
00:31:47.279 --> 00:31:49.440
Yeah, I had a very nice conversation.
00:31:49.519 --> 00:31:49.759
Yeah.
00:31:50.079 --> 00:31:53.279
If you like what I'm doing here, consider supporting the show on Patreon.
00:31:53.440 --> 00:31:58.640
Your contributions help me keep doing this work and dive deeper into the politics of decentralized technologies.
00:31:58.799 --> 00:32:04.799
I promise you absolutely zero financial returns, no airdrops, and your investment may go to zero.
00:32:04.880 --> 00:32:06.319
But you will get good content.
00:32:06.480 --> 00:32:10.240
Check out patreon.com slash the blockchain socialist to support the show.
00:00:00.160 --> 00:00:05.919
But uh at the time of the deployment, we find like the the proof generation is like too small.
00:00:06.160 --> 00:00:15.039
Even we use a very large AWS uh instance, it's still like very slow, which like inspired me to design some special hardware for that.
00:00:15.279 --> 00:00:27.760
For instance, like I want to say like transfer probably like 10 10 bucks online, but I need to prove to the channel or like to some like uh authority that my bank balance is like more than$10.
00:00:28.480 --> 00:00:35.679
So like I can prove that part, but without releasing the exact uh balance number in my bank account.
00:00:36.000 --> 00:00:38.079
What are the implications of this technology?
00:00:38.159 --> 00:00:49.600
Is the ability to selectively disclose specific bits of information about yourself over a digital space, which is you know historically really hard to do because you you produce so much data when you're uh over the internet.
00:00:50.000 --> 00:00:52.479
Seems like in vacancy, you don't need to trust anyone.
00:00:52.560 --> 00:00:57.039
Like you just uh trust the math, and the math like you can do it by yourself.
00:00:57.359 --> 00:01:02.719
And we are not creating like a honey pod like for uh for the hackers to hack.
00:01:03.280 --> 00:01:06.239
Hi everyone, you're listening to the Blockchain Socials podcast.
00:01:06.400 --> 00:01:14.799
I'm Josh, and I am here today actually in potentially my new studio that I'm going to be using uh going on into the future.
00:01:14.879 --> 00:01:18.640
So more news on that that I will share uh soon.
00:01:19.120 --> 00:01:24.560
But before I go into that, um I wanted to introduce my guest for today.
00:01:24.799 --> 00:01:34.959
His name is Leo Fan, and he is the uh founder of Cysic, which is a super interesting project um building called an on-chain compute network.
00:01:35.120 --> 00:01:42.560
But before I butcher trying to explain it, um maybe Leo, if you want to give a quick introduction to who you are and what CySic is.
00:01:43.040 --> 00:01:43.599
Yeah, sure.
00:01:43.840 --> 00:01:45.840
Um my name is Leo Fan.
00:01:46.000 --> 00:01:49.760
Like I uh I got my PhD from Cornell in cryptography.
00:01:49.840 --> 00:01:59.920
After Cornell, like I did some uh research engineering job at NIST and Agrand, then started as an assistant professor at Rutgers.
00:02:00.400 --> 00:02:16.400
Um yeah, so when I was at Argoland, like uh we we delivered the first uh decade bridge became Agrand AC and Agorland Semana, but uh at the time of the deployment, we found like the the proof generation is like too snow.
00:02:16.560 --> 00:02:22.319
Um even we use a very large AWS uh instance, it's still like very slow.
00:02:22.560 --> 00:02:29.360
So which like inspired me to do something to like uh design some specialized hardware for that.
00:02:29.520 --> 00:02:33.599
Um yeah, so the the word like CISIC is a combination of two words.
00:02:33.680 --> 00:02:38.879
The first is like cipher, uh and the second part, the SIC is like coming from like uh ASIC.
00:02:39.280 --> 00:02:44.240
So it's a combination of like building some specialized uh hardware for cryptography.
00:02:44.400 --> 00:02:46.639
So that's like how we like started Cysic.
00:02:46.960 --> 00:02:59.759
Um and like while doing Cystic, like we we do a lot of things, like we do like the GPU and FPG acceleration, and we also did some like work on ASIC design for um for the K.
00:03:00.240 --> 00:03:09.919
So maybe just to help listeners, just in case they don't know what some of these terms mean, ASIC is an application-specific integrated circuit, if I'm not mistaken.
00:03:10.000 --> 00:03:11.599
It's kind of like specialized hardware.
00:03:11.759 --> 00:03:14.240
Often you hear about it in the context of mining Bitcoin.
00:03:14.319 --> 00:03:26.319
You uh you you get ASICs uh hardware in order to uh mine Bitcoin really, really well because it's a specialized hardware that's really good at specifically the thing for um mining Bitcoin, for example.
00:03:26.560 --> 00:03:31.919
Um and then uh if I'm not mistaken, cipher is also a cryptography term.
00:03:32.080 --> 00:03:32.319
Yes.
00:03:32.639 --> 00:03:36.800
Um that I believe is like, I if I'm not mistaken, is it it's the key?
00:03:36.960 --> 00:03:41.520
Is the cipher the key in like a cryptographic secret kind of way to explain it?
00:03:41.840 --> 00:03:48.240
So like in the encryption part, like the encryption world turned in this like painted into ciphertext.
00:03:48.400 --> 00:03:51.280
So ciphertext is uh encrypt kids uh plaintext.
00:03:51.840 --> 00:03:54.240
So ciphertext is kind of like the secret itself, I guess.
00:03:54.639 --> 00:03:54.800
Yeah.
00:03:55.039 --> 00:04:21.040
Um so jumping off a couple of levels from you know uh assuming you guys understand a little bit about cryptography, which is just the math around uh sharing secrets and uh saving data and memory for computers, um zero knowledge cryptography or ZK cryptography is a different type of cryptography that's a little bit more advanced and is kind of on the on the bleeding edge of things.
00:04:21.439 --> 00:04:30.000
Um maybe you can explain a bit what is zero knowledge cryptography and why is it you know, why is it expensive to generate and and use right now?
00:04:30.399 --> 00:04:41.199
So we can like treat the ciphertext we just talk about as encryption or like some representation of the plaintext, but in the in the ciphertext way.
00:04:41.279 --> 00:04:49.600
Uh so like for ZK, you can think about like we so they k to computation is what ciphertext is to uh plaintext.
00:04:49.920 --> 00:04:57.360
So like we basically compress a lot of like a computation into a like very tiny uh B string.
00:04:57.519 --> 00:05:14.160
And this B string will represent this like computation, but this like B string is not like the harsh stream, they still have some like structure inside which will enable people to trust you if they verify that the the B string you handle to them uh satisfies some requirement.
00:05:14.560 --> 00:05:20.639
So that's like how you can see uh the the K the knowledge part.
00:05:20.800 --> 00:05:26.399
So it's like uh a representation of the of the computation you just uh carry out.
00:05:26.720 --> 00:05:27.040
Yeah.
00:05:27.199 --> 00:05:31.279
And and why it is like useful, it has like two nice properties.
00:05:31.519 --> 00:05:37.519
The first one is uh their knowledge less, and the second second one is the succinctness.
00:05:37.759 --> 00:05:47.199
So their knowledge less means like it will preserve the like the privacy uh and without releasing the privacy, the private part like to you.
00:05:47.360 --> 00:06:03.759
Like for instance, like I want to say like transfer probably like 10 10 bucks online, um, but I need to prove to the say to uh to the channel, like to some like uh authority that my bank balance is like more than$10.
00:06:04.399 --> 00:06:15.040
So like I can prove that part, but without releasing like how uh like what's the exact uh balance number like in my eye in my bank account.
00:06:15.199 --> 00:06:27.120
So like I handle you this like uh this like zK proof, and then you can verify and then trust that my bank balance is like is larger than like 10 bucks, and then you can do the transfer.
00:06:27.279 --> 00:06:30.319
So that's so uh the knowledge part.
00:06:30.480 --> 00:06:32.560
The second part is succinct.
00:06:32.800 --> 00:06:50.480
So the succinctness means like the proof is like very short, it's probably like several hundred bytes or several hundred kilobytes, so it's a very small uh proof, which um which like you can use a proof to encode and like a very large net computation.
00:06:51.040 --> 00:06:53.920
So that's the first part of the succinctness.
00:06:54.160 --> 00:07:01.839
The second part is although the proof generation is like takes a lot of time, but the proof verification is like super, super fast.
00:07:02.079 --> 00:07:13.920
You can verify the key proof on your cell phone like in less than 15 milliseconds, and with a very very little uh computing resource, then you can do the verification.
00:07:14.000 --> 00:07:24.160
So next as I say we have this like MySIC app, which you can download it and then to verify the proof we generated like in part of our network on your cell phone.
00:07:24.319 --> 00:07:24.720
Yeah.
00:07:24.959 --> 00:07:30.160
So that's like two nice properties that their knowledge has and succinctness of like the key.
00:07:30.720 --> 00:07:30.959
Right.
00:07:31.040 --> 00:07:31.199
Yeah.
00:07:31.360 --> 00:07:44.240
I think one of the ways that I've also heard people try to explain Zero Knowledge Proofs is the example of like in the United States, for example, where if you want to go inside a bar, you have to show your ID, uh, your driver's license.
00:07:44.399 --> 00:08:02.879
And you know, you give your driver's license to the bouncer, and he doesn't he not only sees that maybe you are or are not old enough to enter your above 21, they know exactly your birthday, they know your full name, they see your address, they see a lot more information than maybe they necessarily need to know.
00:08:03.120 --> 00:08:13.680
So this kind of it's it allows you to say to the bouncer, I'm over 21 without giving any other information, without getting any any leakage of unnecessary information.
00:08:13.839 --> 00:08:22.240
So it's kind of like a way of doing selective disclosure, I guess is this this is the the zero knowledge part at least aspect of zero knowledge proofs.
00:08:22.319 --> 00:08:48.000
And I think that's why it's interesting from uh you know, what are the implications of this technology is the ability to selectively disclose specific bits of information about yourself over a digital space, which is you know historically really hard to do because you you produce so much data when you're uh over the internet and everybody, you know, if you if you own the platforms or if you own a lot of the the piping between everything, then you know uh maybe a bit more or you've given out more information than you intended to give out.
00:08:48.320 --> 00:08:48.399
Right.
00:08:48.559 --> 00:08:48.799
Yeah.
00:08:49.519 --> 00:08:54.240
Um and then succinctness, it's also good for compressing data.
00:08:54.720 --> 00:08:54.960
I guess.
00:08:55.440 --> 00:08:55.919
Yes, yes, yeah.
00:08:56.159 --> 00:08:57.120
It's kind of how I think about it.
00:08:57.360 --> 00:08:57.600
Yeah.
00:08:58.320 --> 00:09:10.480
Um are there any like um maybe specific real-world examples of zero-knowledge proofs and cryptography kind of um in use that like you know the average person may be able to relate to for you?
00:09:10.799 --> 00:09:11.120
Right, yeah.
00:09:11.200 --> 00:09:28.320
So like if so, like we all know like the uh I uh like both uh Bitcoin and Ethereum, they're like public ledger where like all people can check like the the balance or like how how many coins they have like in one uh wallet address.
00:09:28.480 --> 00:09:34.559
But there are some like some privacy-oriented uh blockchains such as like Zcash or Alio.
00:09:34.879 --> 00:09:59.679
They like if you do the they can basically shield the number of like tokens like you have in your wallet so that people cannot check the exact number on all the transfer, probably I would say like 80% of the transfer also like happen privately and which preserves some privacy of like uh the traders or like some other guy like who use the uh the blockchain.
00:09:59.919 --> 00:10:02.720
So that's like on the the knowledge nice part.
00:10:02.879 --> 00:10:05.759
And the second part is on the uh succinctness.
00:10:06.080 --> 00:10:09.519
We can use the succinctness like to scale up another blockchain.
00:10:09.600 --> 00:10:24.480
So like for take Ethereum, Ethereum for uh for instance, like if like how Ethereum like uh validates the transactions that they re-execute all these like transactions in Ethereum.
00:10:24.720 --> 00:10:31.120
So that's like which is like why they put a guess limit for for the like for a block.
00:10:31.600 --> 00:10:38.080
But in instead of like re-execution, you can generate a zK proof.
00:10:38.320 --> 00:10:51.360
Where like since like a zeker proof is like one proof, you will generate and everyone can verify it, and then if it pass the verification, then uh then they will like trust like the the statement in this proof is true.
00:10:51.600 --> 00:11:01.519
So instead of like re-execution, you can just verify some zK proof, which proves uh that all the transactions uh they are like valid.
00:11:01.679 --> 00:11:06.639
And the verification of like the zek proof takes a very, very small uh guess.
00:11:06.799 --> 00:11:14.960
So which which means it will increase the guess limits, which you can like have like more transactions in in one block like for Easily.
00:11:15.120 --> 00:11:21.600
So that's like two uh nice applications of uh they can in blockchain.
00:11:22.799 --> 00:11:23.120
Right.
00:11:23.360 --> 00:11:24.559
Privacy and scaling.
00:11:24.720 --> 00:11:40.960
I think kind of I think what's what's important to understand for a lot of for some users, at least when I've tried to explain cryptography with them is that like a lot of cryptography can be used for both privacy and scaling, and that they're kind of like um the same property.
00:11:41.360 --> 00:11:54.559
Yeah, like the same property of a th of of of this math that you're that you're using in these systems can be used for like things that you don't think are related, but uh they end up being related in in these types of systems, which I think is really, really interesting.
00:11:54.639 --> 00:12:01.360
It's kind of like a when I realized that I was like, oh, this is why it's like you know, quite expansive what what you can think about doing with this.
00:12:01.600 --> 00:12:01.919
Yeah, yeah.
00:12:02.320 --> 00:12:09.840
So you mentioned that like proof ZK proof generation is um fairly expensive or it's clo costly, it's kind of slow.
00:12:10.000 --> 00:12:17.279
Um it's still you know a type of technology that people are figuring out how to do um better, faster, et cetera.
00:12:17.440 --> 00:12:22.799
Um but I'm curious why, you know, where where does that like cost come from?
00:12:23.039 --> 00:12:31.600
Uh why is ZK proof generation so much slower than like say you know other forms of of cryptography that we're already using today?
00:12:31.919 --> 00:12:43.759
Yeah, so like I I guess uh other crypto tools like we are you using today is like uh encryption, digital signing, and also probably like some hash functions.
00:12:43.919 --> 00:12:50.799
So like compare against the these tools, the uh the data handled by uh by the cable is like we much longer.
00:12:50.960 --> 00:12:59.600
So like a typical ZK statement will handle this like degree of like very large uh uh degree polynomials.
00:12:59.840 --> 00:13:04.000
And the the degree is unit, like two to the 24 or something.
00:13:04.159 --> 00:13:06.399
So it's a very large uh polynomial.
00:13:06.559 --> 00:13:13.279
And for and the coefficient for this polynomial is also uh 256-bit none.
00:13:13.519 --> 00:13:17.840
So it's a large polynomial with a lot of like uh data in there.
00:13:18.000 --> 00:13:23.919
And handle this kind of polynomial will require enormous uh computing resources to do that.
00:13:24.080 --> 00:13:39.840
And also the computing resources, like when they do the uh process, they also need to communicate with each other, which means like you also need to have very high bandwidth to support uh the transmission of the data between this like processing unit.
00:13:40.000 --> 00:13:53.039
So that's the like two most important parts in the competition, like very large uh computing resource and then and a very high uh bandwidth to support the processing.
00:13:53.279 --> 00:13:58.320
So which also explains like why it's quite expensive to generate a data proof.
00:13:58.480 --> 00:14:03.840
You need to have a machine that's like powerful enough to do both things like very efficiently.
00:14:04.159 --> 00:14:04.399
Right.
00:14:04.559 --> 00:14:09.279
So it's just uh basically it's more mash, yeah, is uh is the simple answer.
00:14:09.600 --> 00:14:09.759
Yeah.
00:14:10.000 --> 00:14:11.759
Bigger numbers, yeah, bigger numbers.
00:14:12.720 --> 00:14:18.879
Yeah, bigger numbers uh takes longer time to both like computes and also to to transfer the thing.
00:14:19.360 --> 00:14:19.679
Right.
00:14:19.919 --> 00:14:33.840
So I guess uh so this is kind of like my understanding, at least, is that then um you know, like your average computer hardware is not necessarily built at the moment to be generating zero knowledge um proofs.
00:14:34.159 --> 00:14:34.559
Yes.
00:14:34.720 --> 00:14:47.679
Um and so I guess this is kind of why, if I understand correctly from what you guys are doing, uh that you guys also provide hardware, like a a network of hardware for people to use to generate these proofs.
00:14:48.000 --> 00:14:51.759
Um yeah, do you want do you want to talk a little bit more about the hardware side of things?
00:14:52.000 --> 00:14:52.240
Yeah, sure.
00:14:52.399 --> 00:14:54.240
CySink is a compute-file network.
00:14:54.320 --> 00:15:00.000
So by compute file, we mean like we turn this like computer resources into some early machines for you.
00:15:00.240 --> 00:15:08.559
You can plug in your like servers, your laptop, or even your cell phone into our network to do something something like in the computation.
00:15:09.120 --> 00:15:19.200
And of course, like we categorize like uh each we categorize this hardware into the different groups, which is like responsible to do different things.
00:15:19.360 --> 00:15:26.879
Like for instance, if you plug in your like server, public is powerful enough to generate some DK proof in a short period of time.
00:15:27.120 --> 00:15:32.799
And we also use this like uh servers to do some like AI inference for us.
00:15:33.039 --> 00:15:35.840
So that's like the server side.
00:15:35.919 --> 00:15:57.279
If you plug in your like laptop or even your cell phone, which is like not powerful enough to do the the two things I just mentioned, like then we assign this like um they key verification, AI verification job to you, which like you're not very powerful uh computing how uh hardware still can handle that, and also some other things like uh yield.
00:15:57.600 --> 00:16:06.480
And all the things were wrong in the in the back end, so you won't be able to notice any difference like in using your hardware.
00:16:06.639 --> 00:16:06.879
Yeah.
00:16:07.360 --> 00:16:07.600
Right.
00:16:07.679 --> 00:16:07.919
Yeah.
00:16:08.000 --> 00:16:40.720
So there's uh there's generation, the proof generation, which is taking in all the inputs that make up your your zero knowledge proof, and you know, depending on how you how you're uh um like setting it up, you create you generate a proof which is like kind of a reference to you know all these inputs and gives you a kind of answer, but then that answer has to be verified by some by you know the person that receives the answer to double check to know that this proof is saying what you it's claiming to say.
00:16:41.600 --> 00:16:47.679
And if I understand correctly, the generation is much more uh compute intensive than verification.
00:16:48.000 --> 00:16:49.120
Yes, yes, that's right.
00:16:49.840 --> 00:16:53.600
So there's these these these two types of compute, uh verification.
00:16:53.679 --> 00:16:59.279
You know, if you got uh probably any any probably laptop or or phone can do verification.
00:16:59.440 --> 00:17:04.720
Uh the generation maybe you can do, but it will take a lot longer than like you know, a beefier machine.
00:17:05.039 --> 00:17:21.119
Um but you guys are basically, if I understand correctly now then, you guys are creating a kind of network where anybody can can bring whatever hardware they got and they could potentially make a little bit of money off of it because other people are paying for the verification and the generation of ZK proofs uh on the network.
00:17:23.200 --> 00:17:24.079
Nice, that's super cool.
00:17:24.160 --> 00:17:33.759
It reminds me of I don't know if you remember one of these early, I think I think that was uh Golem maybe was like one of the first kind of attempts to do something like this, if I'm not mistaken.
00:17:34.079 --> 00:17:34.799
I think there are other ones.
00:17:34.960 --> 00:17:41.599
It reminds me of like these proof of useful work blockchains that you used to have back in like 20 uh 2017.
00:17:42.559 --> 00:17:43.519
Yeah, maybe yes.
00:17:43.920 --> 00:17:52.960
Yeah, so so that's like yeah, so at that time, like people were saying like uh all this like Bitcoin mining is like a waste of uh computer resources.
00:17:53.039 --> 00:18:01.839
It's like you produce uh some like specialized hardware which will also cost a lot of like electricity in computing the hash.
00:18:02.000 --> 00:18:14.400
So people think like he probably instead of doing that, like we can use use uh like the resources, whether it's like in electricity or computing power to do some like real jobs for human beings.
00:18:14.640 --> 00:18:20.400
Yeah, so that's uh time like in 2014 till 2017, I think.
00:18:20.559 --> 00:18:33.680
Yeah, so since this since like my like before uh coming to Cornell for PhD, and like I did some like Bitcoin mining uh back then, yeah, for for uh for several years.
00:18:33.839 --> 00:18:43.119
So like and the idea of like using some specialized hardware to accelerate the case, also as you mentioned, like this idea was also inspired by Bitcoin mining.
00:18:44.079 --> 00:18:44.960
Okay, nice.
00:18:45.359 --> 00:18:52.319
Um is uh actually um which uh network does Sysek uh connect to?
00:18:52.480 --> 00:18:53.440
Which blockchain networks?
00:18:53.680 --> 00:19:00.000
Uh yeah, so right now we are like proving um so it also depends on the market.
00:19:00.559 --> 00:19:16.240
So so problem like last year, like SySek is a biggest prover for for like several uh large like nail tools, including uh scores, equinq, yeah, things like the these nail tools, they their job is to scale up Ethereum.
00:19:16.400 --> 00:19:26.319
So they generate a lot of like transactions and they need awesome, which means they need a lot of like computing resources like to generate the key proof for them.
00:19:26.640 --> 00:19:42.480
And like so, for instance, like we successfully like reduce the proof generation time from like 1.5 hours to about like 20 20 minutes on a two-car uh 249k uh GPU card like machine.
00:19:42.640 --> 00:19:52.720
Yeah, so so that's why like we so this like layer tools, they they want to work with us to reduce the burden of like generating vaccine proofs.
00:19:52.880 --> 00:19:58.400
But like uh so that's like something uh that's a situation for us uh for next year.
00:19:58.480 --> 00:20:03.119
But as you can see, like all these layer tools, they are not doing very well right now.
00:20:03.359 --> 00:20:03.839
Yeah.
00:20:04.160 --> 00:20:13.839
So so like we shift like uh some of our computing resources to support uh Ethereum and also to support some perp uh perp decks.
00:20:13.920 --> 00:20:21.119
Um yeah, this perp decks, they also need to generate a ZK proof for each uh for each transaction on their decks.
00:20:21.599 --> 00:20:28.079
So like so that's also like a very large number to not large number of like ZK proof to generate.
00:20:29.359 --> 00:20:35.680
Okay, are these perp do they verify those ZK proofs on chain or are they just generating a proof?
00:20:36.000 --> 00:20:46.400
Um so like like our process is like we we like thin, we we can talk about the lifespan of like a ZK proof in inside network.
00:20:46.559 --> 00:20:57.440
So so like for for each like task we receive from say like this like perp decks, we pick like several like random uh proofers uh to generate the proof.
00:20:58.079 --> 00:21:06.720
And then like we pick a large number of like verifiers, which like on your cell phone or on your laptop to verify the proof.
00:21:06.960 --> 00:21:23.039
After a proof is like verified, say by uh eight out of like 10 verifiers, we call it this proof is like settled, then we then we put it on uh sizing network and uh and also additionally we send the verifier proof to to the perp decks.
00:21:23.200 --> 00:21:44.720
So the proof is already like settled in our chain, and some of the problem, like I I guess uh the the perp decks or like they don't have the energy to verify like each proof, they just do some like sampling on the proof uh already settled by us, like to do to check the uh validity of the proof.
00:21:45.119 --> 00:21:45.759
Yeah.
00:21:46.960 --> 00:21:47.599
I see.
00:21:48.240 --> 00:21:48.720
Interesting.
00:21:48.799 --> 00:21:59.759
So you have like uh the network also has a level of kind of redundancy, I guess, where you you want to you want to be able to have trust in the in the proof, and so you have.
00:22:00.799 --> 00:22:06.720
A few at least people run the same proof or verify the same proof uh to make sure that it's valid.
00:22:07.039 --> 00:22:07.200
Yes.
00:22:08.480 --> 00:22:09.200
Yes, I'm sorry.
00:22:09.680 --> 00:22:10.079
Yeah.
00:22:13.359 --> 00:22:17.039
So maybe we could talk a little bit more about compute fi.
00:22:17.519 --> 00:22:18.480
Uh I think it's interesting.
00:22:18.640 --> 00:22:44.640
I think there's there is like this uh uh general I mean I've just I I've I've never really in engaged with any of this, but I've seen some stuff of like um you know people trying to financialize a bit of their their like the computes on their on their computers to sell that as like future compute used in AI inference or something like that.
00:22:44.880 --> 00:22:53.359
Um I I'm wondering if you can explain a bit like how how do you see ComputeFi, what is it, and like um what does this future hold for us?
00:22:53.680 --> 00:23:03.279
Yeah, so uh as you can see, probably like there are like several authors similar platforms such as like iodanets or some some some others.
00:23:03.519 --> 00:23:16.960
So but but the like the iota nets or like some other, they are just creating a platform where like people plug in their uh servers there and and uh wait to be like rented or like used by others.
00:23:17.039 --> 00:23:28.319
So it's basically like AWS or like vast uhai, this kind of platform, but they are in in the blockchain or crypto worlds.
00:23:28.400 --> 00:23:35.279
So like SciSec differs uh from these uh competitors in a very significant way.
00:23:35.440 --> 00:23:48.720
So like in addition to creating uh creating this kind of like platform, we also bring some like real jobs uh to be computed uh by the by the servers in size network.
00:23:48.960 --> 00:24:06.319
Say, like for instance, the the key pool generation is like one thing, and also we have this like uh a lot of like AI products on SISE network, which also consume the under uh consume the tokens uh it's like generated by the underlying machines.
00:24:06.559 --> 00:24:18.240
So like in addition to just uh uh creating a platform, we are we are bringing some like real jobs for the for the servers on SISE network to process.
00:24:18.400 --> 00:24:20.640
So that's a very big, big difference.
00:24:20.799 --> 00:24:21.440
Yeah.
00:24:23.359 --> 00:24:57.599
Do you think like in the future, you know, one one potential well, I guess the thing that kind of you're describing, just to step back a bit, is almost like a you know, this this uh it's almost like a meme to me, but like this meme of the decentralized cloud, so to say, of like computes or computers being able to connect from anywhere to take part in this network and to provide that infrastructure in a way that is resilient uh and redundanced, at least at par with you know other types of cloud providers.
00:24:57.839 --> 00:25:07.359
Um and that people would be able to kind of like you know uh have almost like have have like make money off of the data center you have at home.
00:25:07.680 --> 00:25:08.319
Yes, yeah, yeah.
00:25:08.720 --> 00:25:17.920
You know, rather than necessarily the mass build-out of giant data centers that's maybe going on right now.
00:25:18.160 --> 00:25:36.160
Maybe one alternative is we could use the excess computes that we already have available to us with everyone's consumer hardware rather than needing to overbuild you know all these amounts of data centers, this could be one kind of alternative uh to that.
00:25:36.559 --> 00:25:36.799
Yeah.
00:25:37.039 --> 00:25:37.920
Yes, that's right.
00:25:38.400 --> 00:25:38.720
Yeah.
00:25:39.359 --> 00:25:54.160
So yeah, are there any other uh I guess the um yeah, are there any other non blockchain related ZK um examples that you're you're particularly interested in?
00:25:54.559 --> 00:25:56.000
Um Right, yeah.
00:25:56.160 --> 00:26:02.559
So like as you you can uh as you mentioned before, like the ID thing is like one uh very good example.
00:26:02.720 --> 00:26:07.039
Like in addition to showing up in like your ID, you can just uh generate a digit push.
00:26:07.359 --> 00:26:14.000
Uh saying like you are like uh older than 21, then you can enter the uh enter the uh the bar.
00:26:14.240 --> 00:26:27.680
And another example is like there are some um events like um say like probably like last year or the year before before last year, there is a there is a death count in in Thailand.
00:26:28.079 --> 00:26:40.240
And when like if you are like a local like in in Thailand, then you can apply for some uh apply for a special discount on the ticket for for the event.
00:26:40.480 --> 00:26:50.400
And how how you like proof you are like a local like is like previously like you just upload your your passport and then you are showing that you are local.
00:26:50.640 --> 00:26:55.119
But like that will reveal a lot of information you don't want to reveal.
00:26:55.279 --> 00:27:21.440
Uh so like very nice way of like using ZK to wrap around it, like you download a software and then you generate the ZK proof uh of your passport showing that you are uh from like Highland, and then you upload only the ZK proof to uh to the TK website, and then then after they verify it, then you can go the like special discount.
00:27:21.599 --> 00:27:30.960
Yeah, so that's yeah, it's basically digital ID, and you can use a digital ID like to do a lot of like things, yeah.
00:27:32.079 --> 00:27:32.400
Right.
00:27:32.559 --> 00:27:48.160
I and this is like I mean, this is super interesting to think about just in the current context with the amount of legislation that's being passed right now, um, requiring people to prove that they're you know over 18 or whatever to use certain platforms over the internet.
00:27:48.400 --> 00:27:51.759
I mean, it is something that um is being pushed.
00:27:52.000 --> 00:28:04.240
Different people have different different views on like whether or not it's like the right approach to kind of actually solve the problem that they that they think they're trying to solve around uh protecting kids.
00:28:04.480 --> 00:28:22.480
Um and so yeah, I mean ZK it it what what is interesting to me is that like at least as far as I could tell, a lot of the the technical uh specs of a lot of these identity solutions from uh I think with particularly like the UK and Australia, I know, had some, and I think the US is working on something.
00:28:22.720 --> 00:28:28.400
But um these these identity solutions don't include ZK, as far as I know.
00:28:28.559 --> 00:28:54.480
Um which is interesting, because you would you would think that, you know, um if if you actually wanted to solve the actual problem that they say that they're trying to solve while also not causing any externalities, aka creating a giant honeypot of people's data in someone's databases, like then zero zero knowledge proof should be like your first that should be like one of the first things you should be considering, you know, right?
00:28:54.720 --> 00:28:54.880
Right?
00:28:55.440 --> 00:29:07.759
Yeah, so I yeah, so like we are pushing like the vacay public to the traditional world or like the the mainstream to make people aware of that and then like uh to use to use it.
00:29:08.000 --> 00:29:15.279
Since like in vaccine, like you don't need to trust anyone, like you just uh trust the math, and the math like you can do it by yourself.
00:29:15.519 --> 00:29:20.960
And we are not creating like a honey part like for uh for the hackers to hack.
00:29:21.200 --> 00:29:31.359
They are just the vicate proofs, and they keep proofs are like due to the nice properties of vaccine, you basically cannot get anything out of like of the kid proofs there.
00:29:31.440 --> 00:29:39.200
Yeah, it's it's like more reliable, more transparent, I would say, like compared with the current solution.
00:29:39.359 --> 00:29:39.839
Yeah.
00:29:40.240 --> 00:29:41.200
Yeah, right.
00:29:41.839 --> 00:29:44.240
Um so yeah, I don't know.
00:29:44.319 --> 00:29:56.240
Call I I think it's like we're at we're at a really funny stage in in politics almost where like you know, it's like call call your senator and teach them about zero knowledge proofs.
00:29:56.640 --> 00:30:02.480
Because what they're proposing right now in this legislation like is not is not privacy preserving at all.
00:30:02.559 --> 00:30:10.799
It like really highly highly depends on uh you know someone to build a database that is like just going to get hacked.
00:30:11.039 --> 00:30:13.759
It's not a matter of if, it's a matter of when, you know.
00:30:14.000 --> 00:30:14.160
Yeah.
00:30:14.319 --> 00:30:18.640
Yeah, it's just a matter of time to to hack this kind of database.
00:30:18.799 --> 00:30:19.440
Yeah.
00:30:20.720 --> 00:30:21.359
Yeah.
00:30:21.839 --> 00:30:28.720
Go go to your capital and teach your uh government officials a little bit of math, a little bit of uh zero knowledge cryptography.
00:30:28.960 --> 00:30:30.880
Um yeah, thanks, Leo.
00:30:30.960 --> 00:30:35.759
Is there any are there any other uh last things that you wanted to mention before we close it off?
00:30:36.079 --> 00:30:49.119
Something like we uh we also want to mention is like we open source like the DKVM uh uh code repository, like you can try it out, and we have a lot of like products on on SciSec.
00:30:49.279 --> 00:31:05.039
Uh you can just check out like SciSec uh XYZ for like uh for products and then you can follow us on Twitter, like it's uh CYSIC-uh XYZ to for the latest updates like from our team.
00:31:05.200 --> 00:31:09.680
Yeah, so we have a lot of like interesting products ongoing.
00:31:10.079 --> 00:31:10.480
Cool.
00:31:10.640 --> 00:31:18.319
And how how can can people right now today be able to join the network to be able to make a bit of money off of their hardware?
00:31:18.559 --> 00:31:21.359
Yes, yes, it's like total uh permissionless.
00:31:21.599 --> 00:31:36.400
Like you can use your like uh even if you don't have a very powerful server, you can always use your laptop or even your cell phone to join, you can download like my Sec app on your whether it's like Apple or like Android phone, you can always try it out.
00:31:36.799 --> 00:31:37.759
Cool, great.
00:31:38.000 --> 00:31:44.640
Well, thanks, Leo, so much for coming on and appreciate you explaining Zero Knowledge Proofs to us and sharing about Seic.
00:31:44.880 --> 00:31:45.440
Thanks a lot.
00:31:45.599 --> 00:31:47.200
Thanks enough for having me here.
00:31:47.279 --> 00:31:49.440
Yeah, I had a very nice conversation.
00:31:49.519 --> 00:31:49.759
Yeah.
00:31:50.079 --> 00:31:53.279
If you like what I'm doing here, consider supporting the show on Patreon.
00:31:53.440 --> 00:31:58.640
Your contributions help me keep doing this work and dive deeper into the politics of decentralized technologies.
00:31:58.799 --> 00:32:04.799
I promise you absolutely zero financial returns, no airdrops, and your investment may go to zero.
00:32:04.880 --> 00:32:06.319
But you will get good content.
00:32:06.480 --> 00:32:10.240
Check out patreon.com slash the blockchain socialist to support the show.
