Anna Rose (00:00:05):
Welcome to Zero Knowledge. I'm your host, Anna Rose. In this podcast, we will be exploring the latest in Zero Knowledge research and the decentralized web, as well as new paradigms that promise to change the way we interact and transact online.

Anna Rose (00:00:27):
This week, Guillermo and I chat with David Tse. We talk about his career in research, spanning from his early work on networking, mobile peer to peer and all the way up to his work on blockchain. Along the way we touch on topics like the price of anarchy, his Prism work, and other works that formalized the ways that Bitcoin functioned. We also talk about his work with the Ethereum foundation, which led to his new project Babylon. Now, before we kick off, I do wanna let you know about two initiatives happening in the community at the moment. The first is the ZK Whiteboard sessions. This is part of ZK Hack and powered by Polygon. It's a new series of educational videos that will help you get onboarded into the foundational concept needed to better understand Zero Knowledge tech. The second thing I wanna mention is the ZK jobs board.

Anna Rose (00:01:10):
If you are looking to find a new job, or if you're a team hiring, be sure to check it out over on the ZK jobs board, there are fresh batch of new open roles. These roles are at ZK or cryptography focused projects like Anoma, Aleo or Web3 foundation. This is a great place to learn about relevant projects and the types of roles that they're looking for. So if you're looking for your next job opportunity, be sure to check it out. I've added the link to all of these in the show notes. Now, Tanya will share a little bit about this week's sponsor.

Tanya (00:01:39):
Today's episode is sponsored by Aleo is a new layer 1 one blockchain that achieves the programmability of Ethereum, the privacy of Zcash and the scalability of a roll up. If you're interested in building private applications, then check out Aleo's programming language called Leo. Leo enables non cryptographers to harness the power of ZKPs to deploy decentralized exchanges, hidden information games, regulated stable coins, and more. Visit leo-lang.org to start building that's leo-lang.org. You can also join Aleo's incentivized Testnet 3 by downloading and running a snarkOS node. No signup is necessary to participate for questions, join their discord at aleo.org/discord. So thanks again Aleo. Now here is Anna and Guillermo's interview with professor David Tse.

Anna Rose (00:02:30):
So today we're here with guests, David Tse, professor of electrical engineering at Stanford, as well as the co-founder of Babylon. Welcome to the show, David.

David Tse (00:02:40):
Thank you very much, Anna. It's great to be here.

Anna Rose (00:02:41):
And for this episode we have Guillermo joining as co-host and actually just a note we're here in person. So we're doing this in person. I'm very excited that we get to do more of these now. Guillermo this was a little bit your idea. Yeah. Putting us together. So maybe you can tell us a bit about what you were thinking and why you thought David would be awesome on the Zero Knowledge podcast.

Guillermo (00:03:02):
Yeah, so, so essentially, uh, I guess David and I met kind of early on and, uh, in particular I've, I've always wanted to grab a drink with David and explore exactly how he went from, you know, through the meandering path of, uh, starting from wireless networks and, uh, information theory all the way to blockchain. In particular, there's a lot of your papers that have received pretty high praise. And, and one that always comes to mind, because Tarun always raves about it is, uh, Everything is a Race and Nakamoto Always Wins, which is, um, I, I believe it's he says that there's not, not one paper that comes close to it in this space and there might never be a paper that comes close to in this space, uh, given his current, you know, trajectory on that. But these are the kinds of things that I would love to chat about along with, uh, a number of your more recent papers, you know, some of them regarding Ethereum 2.0, uh, when will the merge happen. I mean, sorry, no, I'm not supposed to ask that actually. I think that was we, we one request, the one request that we had in this episode, don't ask, I can't ask that, but otherwise

Anna Rose (00:03:55):
But now you just did.

Guillermo (00:03:56):
That's right

Anna Rose (00:03:56):
I think it took you about two minutes.

Guillermo (00:03:58):
Hey, I, I think it's, that's a, that's a world record. I mean then there you go. All right. So anyways, um, with that I'm, I'm, I'm quite curious to, you know,

Anna Rose (00:04:06):
To kick it off?

Guillermo (00:04:06):
That's right.

Anna Rose (00:04:07):
Okay. So I think we often start this with a little bit of backstory understanding the journey to get to work on consensus blockchain generally, because you didn't start there.

David Tse (00:04:17):
No, no. Um, I start way back. My PhD thesis was on networking actually.

Guillermo (00:04:24):
Oh, like traditional networking or was it like wireless?

David Tse (00:04:27):
Networking. So in those days there's an interesting story there is that, uh, that was sort of before internet became the, the actual internet we see now at that time it was not very clear on what would be the networking infrastructure that would rule the world

Guillermo (00:04:45):
At, at risk of, uh, being terrible. What, what, what year was that.

Anna Rose (00:04:48):
What era

David Tse (00:04:49):
That was uh, so now you are trying to entice me to reveal my age, but it's okay. It's okay. So this was mid-nineties. So that was when I was working on my PhD thesis. In those days, there was a technology called ATM. Now Guillermo, Anna, your age? You've probably never heard of this technology. Other than thinking, is the ATM machine or something?

Anna Rose (00:05:12):
Maybe me though. Well, maybe I'm a little older than Guillermo.

Guillermo (00:05:16):
Actually.

David Tse (00:05:17):
So ATM was a totally different way of building the networking infrastructure. It was built by, designed by telecommunication companies in those days there was a telephone network, right. remember companies like AT&T it's like the telephone company. Yes and then they thought, Hey, maybe when data, when we need, instead of voice, we have data, we should have like centralized design of the whole networking infrastructure, just like the telephone network, which is for data. In those days, several key phone companies got together and tried to push a standard called ATM, Asynchronous Transfer Mode.

Guillermo (00:05:58):
Oh, okay.

David Tse (00:05:58):
Okay. And so my research was basically how to design these networks for the particular characteristics of data as opposed to voice. Mm. So that was my PhD thesis.

Anna Rose (00:06:11):
Going back to what he said though, what era are we talking?

David Tse (00:06:14):
This was mid-nineties. Okay. Yes. This was mid nineties. It was all the rage. Yeah. Everyone was working on ATM at that time.

Anna Rose (00:06:21):
So I was sentient. I was alive.

Guillermo (00:06:24):
You were there. Yeah. You actually,

Anna Rose (00:06:25):
I was conscious quite, quite conscious.

Guillermo (00:06:27):
I was not.

Anna Rose (00:06:27):
You were not.

Guillermo (00:06:29):
Uh, so,

David Tse (00:06:30):
But the funny thing was when I was finishing my PhD thesis. Okay. Then this whole thing went bust.

Guillermo (00:06:37):
Okay.

David Tse (00:06:38):
Why? Because it was completely a top down push. Right. And that's not how the internet came about. The internet was come from the bottom up.

Guillermo (00:06:46):
That's right.

David Tse (00:06:47):
And so the bottom up effort one and the top down effort lost. And so now I'm stuck, because I spent, you know, a few years working this PhD thesis, I prove a bunch of theorems about, you know, large deviations. There was a mathematics, I was using called large deviations to analyze sort of the performance of these networks in the face of data. And at that time I got a job at Berkeley. So that was my first job.

Guillermo (00:07:12):
I say you were a professor at, you were assistant professor at Berkeley.

David Tse (00:07:15):
Yes. I got a job at Berkeley, but there was one year of transitional time. I was doing a postdoc at, AT&T Bell Labs.

Guillermo (00:07:21):
Oh, okay. So you did work at Bell Labs.

David Tse (00:07:22):
Awesome. Yeah. So I spent that one year thinking, okay, now that my PhD has collapsed, what should I work on? Because you know, you in university, there's a thing called tenure that you have to get that's. Right. And doesn't seem like this area of ATM is gonna support a tenured, uh, portfolio. And uh, so that's when I learned about this area wireless.

Guillermo (00:07:45):
Right.

David Tse (00:07:46):
So if you remember in the early nineties, wireless was like this huge phones and less than a million of them are around in the whole world.

Anna Rose (00:07:55):
The only reference I actually, I like, I actually never saw those in person, but I think on Fresh Prince of Bell Air, at one point he whips out this like gigantic

David Tse (00:08:05):
It's like a water, water tank, a water bottle. Right.

Anna Rose (00:08:07):
It's a brick.

David Tse (00:08:08):
Yeah, it's like a brick

Anna Rose (00:08:09):
And he just holds it to the side of, yeah.

Guillermo (00:08:10):
You were really cool if you had

David Tse (00:08:11):
Yes. In fact, in those days, the bigger, the better because bang is like,

Anna Rose (00:08:17):
They did have like car phones back then though, they had car phones. Were they the same technology or they like,

David Tse (00:08:24):
I believe it's a similar technology. Yes.

Anna Rose (00:08:26):
Okay.

David Tse (00:08:26):
Uh, but those days the bigger, the better, because when you sit and how can you and I put the phone on a table, it's like... a symbol

Anna Rose (00:08:32):
Everyone notice that's...

Guillermo (00:08:35):
Brick, massive a cell phone, I guess. Right.

David Tse (00:08:37):
So in, in those days, a phone is like attached to the wall. Right. That's what people understand the phone. Yep. So although this is very big break, it's still a step up because you can actually move it around. It doesn't have be attached to the wall. So I would say those are mobile phone. I mean, it is kind of...

Guillermo (00:08:58):
"Not-so-mobile-phone"

David Tse (00:08:58):
Stretching it...

David Tse (00:08:59):
But in any way, in any case. And so those days people were thinking now of trying to develop this wireless infrastructure. But then the problem was that the spectrum for wireless is much more narrow than for wireline. So for wireline, you can communicate over many, many megahertz bandwidth, but in wireless you only have this, you know, one megahertz of bandwidth, this

Guillermo (00:09:24):
Is what the FCC

David Tse (00:09:24):
Allows you. That's what the FCC allows you. Otherwise you'll be interfering with other applications like military, etc. And so the research problem at that time that everyone was interested in, how do I squeeze more bits into this very narrow spectrum?

David Tse (00:09:40):
Okay.

Anna Rose (00:09:40):
Interesting.

David Tse (00:09:40):
Because at that point, people were now thinking, remember that in, in wireline, people were thinking about data. And so the wireless people also start thinking, Hey, if the wireline has data, then the wireless being the last hop has to carry data as well. Yeah. And so data has much more demand on the bandwidth. So people are kind of worried. Although at that point, iPhone hasn't arrived, yet people are already quite worried. Hey, how are we gonna squeeze enough bits through this very narrow spectrum? That was the, uh, after my PhD thesis. And so I was getting lucky because during my PhD thesis, in addition to my useless PhD thesis, I've also adopt a side interest. And this side interest is in a subject called information theory.

Guillermo (00:10:28):
Yep. Okay. So information theory is a subject started by a guy named Clark Shannon in 1948. He wrote this groundbreaking paper. Clark Shannon is a very interesting genius. He's very unorthodox. Although he was a professor at MIT, he almost never had any students but he had one sort of his disciple. Ah, and his disciple was name is Bob Gallagher.

Guillermo (00:10:51):
Oh yeah.

David Tse (00:10:51):
And he happened to be my advisor.

Anna Rose (00:10:53):
Okay. Okay.

David Tse (00:10:55):
But at that point he was not that interested in what information theory anymore he wants to do networking. And so that's why my PhD thesis was in networking.

Anna Rose (00:11:04):
Interesting.

David Tse (00:11:04):
But then I learned information theory from him and I thought, whoa, this is a very beautiful subject. But he kept on telling me to do my PhD thesis in networking because that was all the rage networking.

Guillermo (00:11:16):
But wait, you did a bunch of large deviation bounds in your networking,

David Tse (00:11:18):
But that's not information theory. That's just Stochastic process, random process, applied to networking

Guillermo (00:11:25):
Problems. Okay. But I think, are there a lot of tools imported from information theories...

David Tse (00:11:27):
There is some connection clearly from a mathematical point of view.

Guillermo (00:11:31):
Yes, that's right.

David Tse (00:11:31):
But from my research problem, point of view it's very different.

Guillermo (00:11:34):
I have a book right there called concentration inequalities and model selection, which is exactly applied information theory for, I think it's called art. So anyways. Yeah. Okay. So sorry back - I actually didn't know you were Bob Gallagher student that's interesting

David Tse (00:11:47):
Yeah. I had two advisors, one is Bob Gallagher, one is John Tsisiklis two of two, very smart people. One is information theorist. The other one is the Stochastic process person. So that's how I learned the two skills set needed. But then since I learned this useless information theory, but then when wireless showed up, I figure out, hey, you know what information theory is the theory, which would tell you how many bits you can actually push into this very narrow band. And what's the optimal way of communicating. And so that started my totally new research direction. And, uh, that was my first career

Guillermo (00:12:23):
<laugh>

David Tse (00:12:24):
Yes. The beginning of my first career.

Guillermo (00:12:25):
That's right.

Anna Rose (00:12:25):
Where were you doing all this?

David Tse (00:12:27):
So I started doing this at, AT&T Bell Labs. That was when my postdoc and then I became a assistant professor at Berkeley and that's where I built a research group to do this research. So that was my research phase. Now this was 1995 to 2000

Guillermo (00:12:44):
And who was your first student again? I think we'll come back to this very soon.

David Tse (00:12:48):
Yeah. My first student, his name is Pramod Viswanath. Okay. He's now a professor at Princeton university

Guillermo (00:12:53):
Also in information theory or communication or what?

David Tse (00:12:56):
Yes, he, he was my first student and it was in information theory. Yes. Correct. It was wireless.

Guillermo (00:13:01):
Okay. Yes. Right. I mean, for, for context, I guess David has a, probably the most famous wireless textbook. Uh, it was a

David Tse (00:13:09):
Yes. One of, one of the more widely used, I should say one of the is better.

Guillermo (00:13:14):
One of the, just for political purposes. But, but as far as I know, it's, it's the only one that I know. And I mean maybe of course I'm biased also from Stanford, which we'll get to as well. But the one that is used kind of everywhere, I don't know. Single other one.

David Tse (00:13:27):
That was, it was a collaboration actually with Pramod. Oh, okay. That book.

Guillermo (00:13:32):
Okay, perfect. So anyways, sorry. Okay. So, so you've built your lab.

David Tse (00:13:36):
Yeah, we, yes. We, I built my lab. I have a bunch of very good students, super students and I wrote, we wrote a bunch of papers, but then I thought to myself, okay, I have these papers, very nice papers, uh, winning some awards, but who's gonna use it. Who's gonna care about it. Is it gonna make it into some cell phones or not? So then I start thinking, I say, okay, if I want to have an impact, I should go where the rubber hits the road. In other words, places where people are building this technology, not in academia. And so at that time there was this sort of emerging company called Qualcomm.

Guillermo (00:14:11):
Ah,

David Tse (00:14:11):
At that time it was relatively small company.

Guillermo (00:14:14):
That's right.

David Tse (00:14:15):
And uh, my advisor told me, Hey, uh, there are some cool people there. So I decided to spend six months leave from Berkeley to try to see whether my research are useful in solving real world problems, or I need to develop some other things to solve the problems. Okay. And I was really lucky because Qualcomm at that time was building the technology for wireless data. Yeah. And that was called third generation wireless. That was the first generation in which the system was optimized for data, the wireless system. First and second generation was all for voice. Okay. And it turned out that a problem I solved in my research is that basically a perfect fit to a key component of this technology. And so the, the invention I had during those six months resulted in a pattern that was basically used in all cell phones, from 3g, 4g, 5g, even the phones that you are all using is using this technology.

Guillermo (00:15:15):
What, what is this patent actually just curiosity for, for general context.

David Tse (00:15:19):
Yeah. So that, I think that's a good example of why information theory is powerful. Yes. Because in those days, wireless is like a very bad medium because people think that, Hey, the channel sometimes become very strong. Someone's being very weak. It's called fading and it's very hard to communicate over. So it's a bad thing. And people are trying to make it flat as flat as possible. But information theory actually tells you the reverse. It tells you that you should optimistically use the peak of the channel. To communicate, to get a lot bits through. And, and when this guy, you, uh, Guillermo has a weak channel, Anna may have a good channel and I can switch to Anna and communicate when the channel is very good. So this results in sort of an opportunistic scheduling policy, which is what information theory tells me is optimal. And so I introduce this idea to the engineers at Qualcomm they were very skeptical. They said this doesn't work doesn't make sense to us at all. I never thought of the problem this way, but after a few months of lot of work and convincing, they finally implemented it.

Guillermo (00:16:26):
All right. And what was the result? How many, uh, did you guys?

David Tse (00:16:30):
Two X.

Guillermo (00:16:30):
Oh, okay. There you go. All right.

David Tse (00:16:32):
So the throughput increased by two X.

Guillermo (00:16:34):
Holy crap.

Anna Rose (00:16:35):
This might have been before his time there, but like, did you ever cross path with like Anatoly because he, Anatoly from Solana comes out of Qualcomm I think.

Guillermo (00:16:43):
Does he come out of Qualcomm? I think I don't a little later, but...

Anna Rose (00:16:47):
Assuming,

David Tse (00:16:47):
Yeah. I don't think I met Anatoly. At that time. Yes.

Guillermo (00:16:50):
Yeah. Yeah. Which is why he's, he's super into the weird, crazy low level. Uh, tricks.

Anna Rose (00:16:55):
At least that's what I, I had him on like two years ago or so. And I, I feel like I remember him saying that at this point we are at 2005. Where what time?

David Tse (00:17:04):
Yeah, that was, uh, early 2000 now. Early 2000. So that's when we decided to aggregate all this knowledge that we have and write this book that you mentioned earlier, Guillermo mentioned earlier fundamentals of wireless communication, which kind of gives us a unified view of the whole subject. And that's usually the, my research now is that instead of just solving individual problems and call it a day and move on to some other areas, I want to sort of develop a, sort of a more coherent and unified understanding. So my advisor Gallagher taught me one thing, which I think I wanna share with the audience, because this is not just wireless. It's a very interesting way of looking at research is that he said, there's a concept called Knowledge Tree, where each leaf is a piece of knowledge. A lot of people when they do research, they think of growing the knowledge tree, adding your own leaf or your own twig or your own branch. However, he said that a really successful research is instead of growing you prune the knowledge tree.

Guillermo (00:18:05):
Oh, that's fine. Why

David Tse (00:18:06):
Is that? Okay, because if you prune the knowledge sheet, that means you are unifying your understanding of seemingly disparate knowledge, different leaves into a coherent whole. And so, in some sense, a lot of my research in different fields. It's about how to prune the knowledge tree.

Guillermo (00:18:24):
It's this feels, uh, which I think will get to in the very near future as well. But this feels a very Stephen, uh, void like approach. Uh,

David Tse (00:18:32):
Yes. Yes. Because, oh

Guillermo (00:18:33):
I see why you two became friends.

David Tse (00:18:35):
Yeah, exactly. Because you know, pruning the knowledge tree is very tied to the educational enterprise. Because if you wanna teach something, you don't wanna teach the student 10,000 different facts.

Guillermo (00:18:47):
Right.

David Tse (00:18:47):
You wanna teach them one general principle for which that individual facts can be derived and they can go and derive new facts.

Guillermo (00:18:54):
That's right.

Anna Rose (00:18:55):
And would you say like in an emerging space that just hasn't been defined yet? Like people might be like using some first principle, but like they couldn't necessarily describe it yet. And that's what you're trying to dig up. Explain, communicate with those books.

David Tse (00:19:10):
Yeah, exactly. So, I mean, Shannon's paper is a very good example because communication has been around for 100 years before it's paper. Communication started with the telegram telephone in the mid 1800's. Alexander Bell et cetera. So people have been building communication system for hundred years, but everybody is building their own system. So the telephone people building their system, the telegram people building their system, the television people are building the system, radio. People are building their system. They don't talk to each other. And what Shannon said is like, wait a minute, wait a minute. All these people are just doing different facet of the same general problem, which is communicating information from point A to point B. Yeah. And through this abstraction, you are now having a unified method so if you think about all the communication, physical medium that we have here nowadays, optical, wireless, wireline, cable modem, et cetera, they're all based on design, based on the same principle.

Anna Rose (00:20:12):
Were you interested at that time in sort of the decentralized communications happening? Because like around 2000, there was a lot of, you know, P to P stuff happening to move information. Yes. I mean, it's moving, granted, it's moving over wires and rails that already exist, but was that interesting to you at the time?

David Tse (00:20:29):
Yeah. In fact, um, you were asking me what we were doing after my work at Qualcomm. Uh, we spent the time writing the book, but at that time we were also think of finding new research problems. And one of the research problem that people are thinking about is now suppose you don't have a central bay station, and I suppose you have some emergency scenario where you have a bunch of nodes, peer to peer, what is the best way of communicating. And we were in, in particularly interested in a so-called scaling problem, which means as you increase the number of nodes more and more people using, how, how fast can you increase the total rate of communication? So that in some sense, a decentralized problem, because now the question is how do you kind of spread the traffic across many different connections? In an efficient way. Okay. So that was an open problem. And, uh, we actually solved it.

Guillermo (00:21:28):
Oh, okay.

David Tse (00:21:29):
And we find a scheme that achieves very surprisingly linear scaling. That is, as you put more and more nodes, you actually increase linearly.

Guillermo (00:21:40):
Wait, what is this paper? Or

David Tse (00:21:42):
So this paper is actually with Ayfer you know, Ayfer another professor at Stanford. And it says that, uh, hierarchical, cooperation achieves linear scaling.

Guillermo (00:21:54):
Okay. Was Ayfer a PhD student then?

David Tse (00:21:55):
Yes. That was her PhD thesis.

Guillermo (00:21:58):
But what is the title of this paper? Or was this an entire PhD thesis?

David Tse (00:22:01):
Yeah, it's called hierarchical cooperation. Achieves linear scaling where we have a sort of hierarchical way of these nodes cooperating. So typically people think of wireless, uh, multi hop, right? So you got some traffic you fall to the other guy and that method only achieves so-called square root of n scale, square of n scale. So you increase number of nodes, a total throughput actually decreases per node.

Anna Rose (00:22:27):
It's under it's it's under the linear line.

David Tse (00:22:29):
Yeah. Under the linear line, because why, because sort of each node is have to do a lot of work in forwarding different traffic.

Guillermo (00:22:37):
Right.

David Tse (00:22:38):
But we figure out that if you do it in a hierarchical manner, cooperation manner, you can achieve linear scaling.

Anna Rose (00:22:45):
Cool.

David Tse (00:22:45):
Yes. So we got this beautiful results and we're really excited about it, but then we realized that actually nobody's actually building this system.

Guillermo (00:22:54):
<laugh>

Anna Rose (00:22:54):
The P2P stuff.

David Tse (00:22:56):
The, the P2P stuff

Anna Rose (00:22:57):
Anymore. Yes. It was the winter. Yes. Yes. The P2P winter.

David Tse (00:22:59):
Yes. And I think in your, one of your early conversation with uh, Sreeram he alluded to this as well. And, uh, so we were a little bit disappointed because we feel like we've got the optimal scheme. But you know, nobody, nobody is interested. Nobody cares. Right. Nobody cares. So at that point I realized that, Hey, maybe it's time to kind of move on to some other area where there are more promising, uh, pastures

Guillermo (00:23:26):
Also, I guess, uh, for context, you know, while this result is like very, very nice, it does not hold in a blockchain context. I'm sure David can elaborate as to why, but essentially you're assuming that everyone is happily cooperating with each other.

Anna Rose (00:23:37):
Yes. This is the hierarchical one.

Guillermo (00:23:38):
This is the hierarchical case, right?

Guillermo (00:23:38):
So there's a classic question of when you have a bunch of peers connected to, you know, a given for, for a given chain for example, you actually don't achieve, you cannot achieve this O event scale. I don't know if there's an impossibility result on that, but you essentially cannot achieve this O event scaling kind of under, cause you always have people who are doing bad things.

Anna Rose (00:23:57):
I see. Yeah. So assuming a perfect world or assuming one organization running everything,

Guillermo (00:24:02):
No, assuming, they wouldn't necessarily even an organization, but a cooperative world where everyone is, there's no person trying to kind of mess with organization

Anna Rose (00:24:08):
It's not a hostile environment.

Guillermo (00:24:10):
That's right.

David Tse (00:24:10):
Information theories are very benign. You know, they think the world is very good, but they're everyone should work together.

Guillermo (00:24:15):
That's right.

David Tse (00:24:15):
In fact, this similar timeframe, uh, Tim Roughgarden, I don't know if you interviewed him in your shows.

Anna Rose (00:24:21):
Not yet, but I know him.

David Tse (00:24:22):
Very interesting guy. He starts thinking about, uh, a problem where each node is not willing fully cooperate, but cooperate only based on the incentive basis. And he was trying to figure out sort of where are the good Nash equilibrium.

Guillermo (00:24:37):
That's right.

David Tse (00:24:39):
And, uh, so that was another line of work.

Guillermo (00:24:40):
And there there's a lovely term for this. It's called the price of anarchy. If anyone does want to look it up, which it turns out the price of anarchy is the best possible case. If you had a central planner. So if you had one person directing the flow of every, you know, traffic in this case, uh, data, for example, uh, one organization that is the best possible you could do, of course, this person could just benevolently say we are gonna maximize the amount of data that gets through here. Okay. Period. If I said it as law it's gonna happen, the price of anarchy says something a little bit different. It says, if you instead allow people to like act according to their own incentives to pass data or not, you actually achieve, uh, suboptimal result and the price of anarchy, quantifies, how suboptimal it is.

Anna Rose (00:25:25):
Right. It's the difference between those two things.

Guillermo (00:25:27):
That's right. And you can imagine there's a gap because anything that people are incentivized to do by themselves, certainly a central planning organization can force upon them, can say, this is, you know, what you want to do, but on the other hand, it's not true the other way around. Right. You cannot take a very nice plan and then expect it to be kind of what people call incentive compatible. You cannot expect people to follow it just by their own will even, you know, according to their own incentives. Cool. So this is, uh, now I guess, mid to late 2000s. Right? Is that about the timeline?

David Tse (00:25:57):
Yeah. It's late 2000s. It's about late 2000s. Yes.

Guillermo (00:26:00):
And then, uh, I believe around there, uh, you were poached actually, right? Or is that a little bit later?

David Tse (00:26:07):
So be between then and blockchain, there is another story. Okay. But I don't know whether we want to talk too much about that story because we want to get to blockchains, I guess. Right.

Anna Rose (00:26:15):
We do wanna get to blockchains, but let's, let's hear a little something about it.

Guillermo (00:26:19):
A quick summary would be lovely.

David Tse (00:26:20):
Yeah. So information theory is basically about sort of what's the best way of doing things. So at that time I was looking for something else and I just happened to stumble in a random workshop at a place called Newton Institute in Cambridge, where they talk about this problem called sequencing genome sequencing.

Guillermo (00:26:41):
Oh, that's

David Tse (00:26:42):
Right. So basically genome sequence is a real interesting problem. It's a jigsaw puzzle. You take a very long sequence, 3 billion symbols. That's our genome. You want to read the whole sequence from beginning to end, but nobody can do that. No, no instrument can do that. So what you do is you chop it up into very small fragments called reads and then you want to sample it enough search that there's lot overlap between these reads and then you try, try to fill, put the jigsaw puzzle together to get the original sequence. So that's a sequencing problem. And of course that's a very important technology. That's how the human genome or sequence in the first place. But, and as the information theory is asking the question, okay, so you get data, you want to recover the ground truth. Well, what is the minimum amount of data I need to sample. Right. How many reads do I need to sample? Because that's an important question, because if you can get by with few reads, then you can sequence more organisms. For example, using the same budget. And so we start thinking about this question and then we provide solution problem, and then we develop some assembly methods to try to achieve this limit. So that was in now in the early 2010s. Yes, yes. So that was our focus of our group at that time.

Guillermo (00:27:59):
And that kind of culminated an interesting paper, just, you know, before we get to much into the blockchain, which, uh, I, I actually just pulled it up. It's Hidden Hamiltonian Cycle Recovery via Linear Programming kind of explains all of the results you had prior.

David Tse (00:28:12):
Yeah. That was part of the, yeah. That was part of the research, right? Um, yes. So yeah, it's

Guillermo (00:28:19):
A really cool paper, by the way. It's, it's actually like, it's an, it's an insane result, which is very shocking, but, uh, essentially, yeah, that's right. It says something like we take a, this problem is known to be very, very hard for computers to solve. Uh, but it turns out if you have enough samples. So if you, if you take enough reads, it actually becomes extremely easy and doing the, the, you know, the silly thing of kind of putting everything together, uh, this is called, you know, whatever a linear relaxation, uh, actually achieves exactly the right result as if by magic, you know?

David Tse (00:28:49):
Yeah. So a lot of people, when they look at the assembly problem, the standard approach is on a computational basis. You wanna find efficient algorithms to do it, but it turns like if you look at it from an information theory point of view, which is the minimum of data, it turns out it gives to inspiration directly of finding efficient algorithms to get those information limits. And that's what we were working on. So that was sort of my introduction to sort of data science, genomics. And I worked there for a few years. Now one thing I observe about that area though, as opposed to wireless, is that the people driving the research agenda in genomics are the biologists. People who are like me, developing computational tools are like plumbers,

Anna Rose (00:29:36):
They just call you in to fix

David Tse (00:29:38):
They call us in and say, Hey man, we have this data set, or I have this application, can you help me to actually, you know, analyze this data? And I don't know about you, but plumber of course, plumbers make a lot of money, so that's good. But you know, the status symbol is a little bit low.

David Tse (00:29:58):
A little bit low. So I think the problem there is because there is a, the engineering is kind of subservice to the science. And I'm primarily an engineer, not a scientist. Right. I would not be able to identify the next groundbreaking biological problem to ask, I have no idea. And so at that point, I, I realized that maybe this is not an area, which I wanna spend the rest of my life in. And so again, I'm on the search

David Tse (00:30:28):
So it was fun.

Anna Rose (00:30:29):
I'm seeing a bit of a pattern. Yeah.

Guillermo (00:30:30):
It's interesting. Yeah.

David Tse (00:30:32):
Maybe it's a restlessness pattern perhaps.

Guillermo (00:30:33):
Yes. But

Anna Rose (00:30:35):
It's probably, I mean, I think I have something similar, I've switched through many industries and focuses. So at the end of that, though, you do have this, you tend to have a broader view or sometimes you can pull things from these other spaces and be very useful

David Tse (00:30:49):
Yeah. I

Guillermo (00:30:49):
And post things he does

Anna Rose (00:30:52):
That's for sure.

David Tse (00:30:52):
Okay. I believe so. You know, sometimes like my PhD thesis, right. I spent four years doing it and it turned out to be not very useful. It doesn't mean though, It's like a whole lost everything lost because I learned a lot while doing it. Right. Yeah. Trying to find the right problem. Learn some mathematical skills. So it always useful so I tell my students always that it's okay your PhD thesis does not change the world. It's okay. Yeah. But the more important thing is that you should learn while doing it for sure. Because what you learn will be useful for your next stage of your career, whatever you choose to do.

Anna Rose (00:31:26):
Definitely. So I think we have maybe arrived.

Guillermo (00:31:29):
We have actually

Anna Rose (00:31:30):
At blockchain.

David Tse (00:31:31):
Yeah. Yeah. So now we're in 2018.

Guillermo (00:31:35):
Uh, and in fact, actually that is, I believe the year that, uh, we met or did we meet in 2017?

David Tse (00:31:40):
Around there? Yes. We met probably in 2017. Yes.

Guillermo (00:31:44):
So for, for context, uh, I, I think I was your head TA for a, a very specific class with Steven [Boyd] as well, called introduction to linear algebra. Um, Hey,

Anna Rose (00:31:54):
That's the class I took. Not at Stanford at McGill but...

Guillermo (00:31:57):
You should have it. Would've been, maybe you would've been a mathematician then instead

Anna Rose (00:32:00):
If I had done it there. Yeah,

Guillermo (00:32:01):
Exactly. Um, but so, okay. So that that's, that's actually where we originally met, they were co-teaching this class.

David Tse (00:32:07):
Right. So that was when I moved to Stanford.

Guillermo (00:32:10):
Oh, you had just moved to, yeah.

David Tse (00:32:11):
I moved Stanford to Stanford in 2014 timeframe. Okay. 1, 4, 5 timeframe. And linear algebra was one of the course I started teaching that's right. The first two of three years. Yes, that's right. Oh

Guillermo (00:32:21):
No, you're right. It was 20. Sorry. It was earlier you're right. It was 2016. Was when the course was initially taught if I recall, correct

David Tse (00:32:28):
Yeah. I think you were still an undergraduate. If I remember correctly,

Guillermo (00:32:31):
I, I was, I TA'd that and then convexed right afterwards if I recall. Yes. Um, although you did not, it was Sanjay, I believe, but anyways. Okay. So digressing.

David Tse (00:32:39):
So I love teaching freshmen, so that was a freshman course. I love teaching freshmen because, you know, freshmen, they come in with very little bias and it's great to be able to sort of engage with them at that point. Yeah. I think it's great.

Guillermo (00:32:52):
And actually I'm still getting comments about that class. That, that class was actually what turned people into doing more math. Actually. They were like, math is not

David Tse (00:32:59):
Formal thinking. Yeah. Formal thinking because it trains people to think formally.

Guillermo (00:33:02):
But even more generally as like, uh, you know, math, isn't just, you know, let's prove X thing for Y reason. It's very much, uh, you know, it's an art in a way. So anyways, that was interesting because of that, but, okay. So, and, and that, this is, this is around the time when you went down, uh, down the rabbit hole, I believe, I think that's the technical term for it. Uh, if I recall correctly.

Anna Rose (00:33:22):
And what was that first project then? What kind of triggered you working on something blockchain related?

David Tse (00:33:28):
Yeah. So as I mentioned, I was looking for something new in 2018, uh, beginning of 2018, certainly event happened, which was Bitcoin went to the peak at that time, the peak was almost 20 K, almost 20 K nowadays, when we saw, see Bitcoin 20 K is like the end of the world but someone with a little bit longer perspective will tell you that 20 K is actually very high already. Yeah. Because it was the last peak that's right. And so at that time, there was some interest in doing research in blockchains and I was recruited actually in a podcast with Sreeram. He already mentioned he was recruited by Pramod his advisor in working on blockchains and Pramod being my former student also recruited me. And in fact, he recruited a bunch of people, not only me and Sreeram, but a bunch of very good people.

David Tse (00:34:18):
And we start forming a group doing research in blockchain. And I took a year of, uh, absence from Stanford to focus my effort into learning about the subject and trying to understand what are the interesting problems because, you know, blockchain are typically two types of people from academia work on. One cryptographer. And two distributor system, those two are the main branch of science that enters into blockchain, but we are information theorists. That's our background, very few information theorists work in this area. And so we thought, Hey, first of all, there's a lot to learn. But you know, sometimes when you move into a new area, you bring your own perspective. Something interesting may happen. For sure. So we are, so we kind of did a bet on that and we spent a year working on it. And uh, so that led to our first paper, uh, Prism that I think Sreeram also mentioned in his podcast.

Anna Rose (00:35:13):
I should mention that podcast actually. So we have another episode we did with Sreeram Kannan, uh, and I think it came out three, four months ago. We'll add the link to that in the show notes. If anyone wants to check it out, actually at this point, from this point on, there's going to be a lot of overlap in terms of the works, because it seems like you co-authored together quite often.

David Tse (00:35:31):
Yeah. Sreeram and I work very closely together. We, Sreeram mentioned in his podcast, we talk on a daily basis. Because you know, if you wanna learn something new, it's really better. If you have a bunch of people talking to each other than you, yourself trying to read like gazillion number of papers. And honestly.

Guillermo (00:35:50):
That are pretty bad written

David Tse (00:35:51):
Blockchain papers are either like bogus. In terms of what they claim or the rigorous ones, which are very good, but it's extremely hard to penetrate.

Anna Rose (00:36:00):
Oh, yeah. A lot of them will be using their own words for the same things. I don't know if you've noticed that. Oh, it's exciting. It's like, it's a new vocabulary every time. That's right. And then you have to, you're like, oh, that's the, this in this other one.

David Tse (00:36:11):
That's right. Yeah. So, you know, the core of blockchain is consensus. Right. But apparently there's no consensus on the vocabulary.

Guillermo (00:36:21):
Well, there, there might be eventual consensus in the vocabulary, but that might be as T goes to infinity. So, um, you know, if you wanna wait around that long of I'll let you do that. But unfortunately I think, uh, each person has to come up with their Rosetta stone for, for this, uh, this translation layer. So to speak, you know, now that we've gotten to prism, do you wanna describe what actually, I don't know if sure did, there is in the episode.

Anna Rose (00:36:41):
I, I think he did, but I think it would be great to actually hear it again. Yeah. And maybe you, maybe you just find it a little bit differently too.

David Tse (00:36:48):
Yeah. Yeah. Maybe I can take a little bit different perspective on that problem. So yeah. As Sreeram mentioned, the two big problems at that time was how do you increase the throughput and how do you reduce the latency of Bitcoin?

David Tse (00:37:03):
And as information theorists we start asking. Okay. Alright. If you wanna increase the throughput or reduce the latency, are there any fundamental limits? Right. So what are the let's let's have a ruler to figure out what's the fundamental limit first and then we tried to figure out, okay, how far is Bitcoin from these fundamental limits? and let's say if we can bridge the gap. So first observation we had was Bitcoin's really far away from the fundamental limits, because

David Tse (00:37:34):
You know, Bitcoin is a distributor system. Yeah. Right. So distributor system is limited by two things. One is the propagation delay of communication. Right. You've all the nodes in one place. There's no consensus issue because everyone can get an instantaneous update as to what the other guy's doing because of the delay. There is a split view issue. So that's one thing. The delay. Yep. Two is the throughput is limited by either the processing speed of the computer or the communication link, but both are very fast. We know we are fast, very fast computers and we are very fast networking. Right. That was my early career. Yeah. Okay. So our sensitive to that, right? In fact, we measure Bitcoin throughput and we figure out that Bitcoin throughput can fit into a telephone modem. From the nineties.

David Tse (00:38:26):
24.

Anna Rose (00:38:29):
So that's...

David Tse (00:38:29):
28K, Cable?

Anna Rose (00:38:30):
Are you saying, That's not causing the difference?

David Tse (00:38:32):
So that's not causing the limit. That's not besides, okay. See our communication, right. Your cable modem. We're talking about hundreds of megabits per second. Hundreds of mega per second. That's right. So many orders of magnitude larger. So then we start figuring out, so what's going on? Why, why is there such a big gap? And so that's kind of motivated Prism.

Anna Rose (00:38:52):
It sounds like, I don't know if this is at all related, but it sounds a bit to the, like a throwback to your anarchy is cost or something. Oh

David Tse (00:38:58):
Yeah.

Guillermo (00:38:59):
The Price of Anarchy

Anna Rose (00:38:59):
...because you're comparing it to a central, a centralized version

David Tse (00:39:05):
Actually. Yes. In some sense. Yes. Um, yes, but the price of anarchy result also states that the loss of performance from decentralization is not based.

Guillermo (00:39:15):
Slide based. It's in fact constant.

David Tse (00:39:16):
It's constant factor. So that's what the process in a lot of cases. Yeah. So then the challenge is whether or not one can improve Bitcoin to get closer to those limits. And, uh, what we figure out is that Bitcoin looks like a really simple protocol, right? Very simple protocol. Everybody does exactly the same thing like everyone else, which is to mine and keep on trying to solve this puzzle to generate block. And then once you generate block, you work on the next block. So it looks like a really simple protocol, but it's actually packed with complexity. So this fellow Nakamoto, or this group of fellow Nakamoto, really smart, because if you think about it, if I mine and propose a block, right, I'm actually doing multiple things for this block. One is I'm trying to put new transactions into the blockchain, but two is, I'm also voting for all the previous transactions in the previous block.

David Tse (00:40:15):
Because if I add a block, that means I am increasing my trust that this is the correct history and therefore I'm voting for all the transactions previously in history. So actually I'm doing at least two things. And if you think about it, the first thing of putting transaction in is about throughput. The second thing about voting was about confirmation and therefore about latency. Oh. So to be able to solve these two problems first, I need to disentangle these two problems. Yeah, because obviously it's hard to solve two problems at the same time. Okay. So basically prism is about trying to break the blockchain structure into three types of blocks. One purely for proposing two, purely for voting and three purely for carrying transactions. And by breaking into three types of blocks, we can scale up the behavior in a separate way. And that's basically Prism.

Anna Rose (00:41:14):
Were you able to identify in that breaking also, which one caused most of the kind of Delta between the centralized and the regular, like was one of those more responsible than others for any sort of like lack of performance.

David Tse (00:41:28):
Okay. So the problem of Bitcoin is that Bitcoin's kind of interesting because consensus is actually a 40 year old problem. Okay. Right. But Bitcoin is a consensus protocol. It has no connection absolute whatsoever to all the 40 year old protocol. Mm. But yet it's the first one that's widely deployed in some sense in, in an internet global scale. And so the, the, the, the real reason why Nakamoto could do that is because it slows down everything.

Anna Rose (00:42:02):
Oh, you see, he just sort of didn't care about performance.

David Tse (00:42:04):
Right? So you think about his typical consensus protocol. Where you have a hundred nodes. Yeah. Then everybody is born in the same time. So there's a lot of complexity in trying to deal with, Hey, you know, many people are talking and some people may be talking bad things, wrong things, and try to disentangle that as very difficult right Nakamoto says forget it. I am not interested in speed. I'm only interested in a very robust system that can be used on global scale. And so when you did, was it slowed down everybody. So that only one person in some sense can speak at a time. And so when you try to speed it up, that's when the problem starts, because then you are confronting in some sense, the 40 year old consensus problem. But now you have to deal with many other issues, like proof of work or permission list that people can join.

Anna Rose (00:42:51):
Yeah.

David Tse (00:42:51):
So what we basically did was we think of, instead of having a lot of people voting, we sort of allow sort of each person to have sort of a voting chain. So each chain is still a Nakamoto chain. But now all these chains are regulated by Nakamoto problem by Nakamoto mechanism, but yet they can simultaneously vote to speed up the confirmation. So I would say the speeding up of the confirmation was the most difficult part of the problem. Interesting. Because that was

Anna Rose (00:43:26):
When, but it was the voting dimension.

David Tse (00:43:28):
Yeah. The voting dimension was the most difficult.

Anna Rose (00:43:29):
Yeah. But did you do it on each of them? So you were trying to optimize each of them. That was the hard one. How did you do it on the other ones?

David Tse (00:43:35):
Yeah. So we first solved the voting problem, the fast confirmation. And then we realized that what Nakamoto did was the following. He said, if you propose a block, you have to put all the transactions into the block. Yep. And verify it, and then you can propose it. But that means that the, the, the throughput of the system, if you want to increase the throughput, your block size will becomes bigger. If the block is bigger, then you have a lot of delay. And if you have a lot of delay, you have a lot of consensus problem. So called forking problem. And so there's a limit on how big the block size can be. In the case of Bitcoin is usually one megabyte. Okay. One megabyte. And so what we said is, Hey, why don't we just take the transaction outta the story and have a separate type of blocks in this transaction? And just in some sense, have hashes or representation of this transaction in a very small block and you can pass it around very fast, and then you can vote on that and then make references to the transaction. And so this idea actually sort of became nowadays rather popular and is using many different protocols is essentially separate out the execution from the consensus.

Anna Rose (00:44:50):
This sounds oddly familiar. No. Who else has done that recently? Or has talked about doing that recently? Yeah. So this might be doing it soon.

Guillermo (00:44:59):
Soon to you.

Anna Rose (00:45:02):
Cool.

Guillermo (00:45:02):
Right. Essentially. Yeah. So, so, uh, at a high level, I love the, the idea behind Prism actually is, is, um, is quite an interesting one. It's like evaluating, like, I believe actually is a subtitle of the paper, but it's like essentially evaluating the physical limits of blockchain.

David Tse (00:45:16):
Yeah. It's called deconstructing the blockchain to achieve physical limits.

Guillermo (00:45:20):
Limits. Right. And, and here, the physical limits are, of course, those given to you by kind of the traditional, you know, gross metrics such as network delay, uh, or, or bandwidth. Correct.

David Tse (00:45:29):
Right.

Guillermo (00:45:30):
Then starts leading into a number of other papers though. Uh, this one's 2018 or 19.

David Tse (00:45:37):
Yeah. We're finished in 2019 end of 2019. So since this was the first problem that we worked on, we thought, Hey, Bitcoin is the thing to improved.

David Tse (00:45:48):
Now, after we spent all this effort on this paper,

Anna Rose (00:45:51):
Why do I think there's another pattern? Okay.

Guillermo (00:45:53):
Ah, yes,

Anna Rose (00:45:53):
Exactly. It's like, we're gonna hear that now

David Tse (00:45:55):
Now, that's often the case when you work on the first research problem, is that you look at the problem very fundamentally, but in some sense, the interest in the field could have shifted while you are working this problem. That's right. Okay. And what happened for us is that proof work has shifted to proof of stake.

Anna Rose (00:46:14):
For sure.

David Tse (00:46:15):
So you may ask, Hey, your paper's so great. Why haven't I heard of it? This guy? What, what is he talking about? Right. I I've tell you that paper's great in my mind.

David Tse (00:46:27):
But I still agree. I understand why you may not have heard of it. And really the reason is because the center of mass of the field has shifted to proof of stake at that point.

Anna Rose (00:46:36):
Yes. Yeah.

David Tse (00:46:37):
Okay. And in proof of stake the problems are quite different. And so one of the, when we finished that paper and realized that the center might shifted, so a natural thing was okay, can we sort of adapt a protocol to proof of stake because our protocol has still a lot of interesting advantage and that led to a more basic problem, which is okay, can you adapt the longest chain protocol, which is Nakamoto protocol to proof of stake. Right. That's the more basic question yeah. Than we ask. Okay. And so then we start looking at sort of the protocols that people are playing around with. Okay. And we realized that proof of stake has one big problem compared to proof of work. There's a problem called nothing at stake.

Anna Rose (00:47:24):
Nothing at stake

David Tse (00:47:26):
Nothing at stake,

Anna Rose (00:47:27):
In proof of stake?

Guillermo (00:47:28):
I know.

David Tse (00:47:29):
So suppose you have you, you think about Nakamotos protocol, right? The protocol is you should always mine on the longest change. You should always grow on the longest change. It's a very simple protocol. One sentence summarizes it. Yep. But then you ask, well, is there an incentive for the adversary to go and mine on other blocks maybe to create confusion? And the answer is really no, because if he tries to mine another block, then he has to consume computational power. He has only limited budget of his hash power, and then he can't mine on the tip anymore. So there's a very strong incentive for him to just forget about every story action just followed the longest chain. But for proof of stake it's different because stake is not a computational resource. No. And you can use that stick to try to bet on many different blocks and see, Hey, maybe I can win on other blocks and if I can win on other blocks because the lottery is different. So that's nothing at stake problem. And as a result, you can like grow a forest and try to pick out the longest chain among them. And you can win against the honest.

Anna Rose (00:48:35):
Interesting.

David Tse (00:48:36):
And that's actually a protocol that Bram Cohen at Chia , Was building. Okay. Because there, they were using the same concept of longest chain, but on proof of space. So proof of space and proof of stake is very similar because they both have just nothing at stake problem. And so we realized that, okay, the analysis of that protocol is extremely complicated now because you have so many people growing. Yes. So many chains growing. And so we, that led us to sort of try to find sort of a unified framework of analyzing all possible sort of longest chain protocol, whether it's proof of work, proof of space or proof of stake.

Guillermo (00:49:17):
And here we get to, uh, I believe Tarun's paper

David Tse (00:49:21):
Yeah. So that led to this paper, everything is a race and Nakamoto's always win.

Guillermo (00:49:25):
And what is actually, which is kind of funny, because you nerd sniped a very famous who, as far as I know, has no idea about anything about blockchain, which is Amir Dembo right before we, well, yeah. Before we get too deep, deep in the, in the results, how did you even achieve, uh, nerd sniping being a statistician actually?

David Tse (00:49:43):
So at that point

Anna Rose (00:49:45):
Guillermo wants to know.

David Tse (00:49:47):
Yeah. At that point we spent about a year, almost a year working this problem of trying to figure out, in fact our original point was not to have a unified framework. Our original point was just to solve this Chia's protocol problem.

Guillermo (00:49:59):
Sure. Yeah.

David Tse (00:50:00):
Because they didn't have a good solution for this and they were asking us, so, okay. How do you sort of deal with this problem? Because it looks like the security is very bad because you can do this "Nothing at Stake" attack. And um, so we spent months working this problem with my students. We're stuck. And then I met my old friend Alfred Zatoni.

Guillermo (00:50:18):
Okay.

David Tse (00:50:18):
Who is Amir Dembo's buddy. Okay. So they work in probability. Okay. In fact he was the one who worked with me, helped me in my PhD thesis on large deviations because he's the world expert in large deviations. So I knew him from 20, 30 years ago. Wow. Okay. But then he kept having to be visiting Stanford and he asked me, Hey, remember those problems we're working 25 years ago. I'm still interested in those problems. Can we do some more collaboration? I said, no, I've moved on several times.

Guillermo (00:50:49):
As we found out repeatedly.

David Tse (00:50:51):
But then I said, I'm working in this area called blockchains. He looked at me and said, what's this blockchain. Never heard of it.

David Tse (00:51:00):
And then I thought, okay, how do I explain to this guy who knows nothing about blockchain

Anna Rose (00:51:05):
In a way that is gonna make him wanna do something

David Tse (00:51:07):
In a way that

Anna Rose (00:51:09):
Not like there's this thing called crypto? Yeah. Yeah.

Guillermo (00:51:12):
One point, no,

David Tse (00:51:13):
I realize that this nothing at stake problem has a very strong probabilistic nature because you're kind of growing your tree in the independent way at many, not. So it's like a random tree and there's a lot of known improbability about the evolution of such a tree. And so I figure out a probalist can really contribute on this problem. Cool. And so I strip away all the details about the cryptography and just turn it into an entirely a stochastic process problem.

Guillermo (00:51:44):
Right. Okay.

David Tse (00:51:46):
And then I can explain to him, and then we start collaborating and Amir Dembo is his collaborator. So he also brought Amir in and we start talking. And when we came up with a solution, because not only we solved this particular problem of the Chia's protocol problem, but we also solved a unified problem. Because why? Because we're stripped away all the details already. So now we can apply the framework to many different types of problem. So why do we name the paper? Everything is a Race and Nakamoto Always Wins is Nakamoto's paper. There is one page of math in that paper out of eight pages. That page math is basically analyzing the security of this protocol through a race between the adversary and the honest miners.

Guillermo (00:52:35):
And there is a joke that the only distribution that computer scientists know, or is the Poisson distribution. So, uh, there's, there's a reason why

David Tse (00:52:41):
Maybe it's enough already.

Guillermo (00:52:42):
It, it, well, it turned out to be, uh, by,

David Tse (00:52:44):
So it is two, it's a race is a stochastic race because of the randomness in the mining, but he never showed that that attack is the worst attack. And so the question is because in consensus, people are always interested understanding what's the worst attack, because otherwise if you design a protocol that just deals with one attack, what happens if someone else came up with another attack. That you have not thought of yet. And so, uh, that was an open problem. So what we showed in our paper is that in fact it is the worst attack. No other attack is worse than this attack. And not only for Bitcoin, not only for proof of work, but also for proof of stake and proof of space. So that was our main result. And that's why we call the paper. Everything is a race. So all the attacks can be transformed into a race and Nakamoto always wins because they figure out intuitively that that is indeed the worst attack without being able to do the math to support it. But we did the math basically.

Guillermo (00:53:47):
Yeah. Yeah.

David Tse (00:53:48):
So that's

Anna Rose (00:53:49):
To show it. Yes.

David Tse (00:53:51):
To, because you have to show, I mean, it's not trivial because the attack base is infinite dimension.

Anna Rose (00:53:58):
After this. I know that. And stop me if there's something in between here, but I know at some point you also started to kind of work in collaboration with Ethereum. Yes. And there, it seemed like, were you going back to the Prism work or were you also, was it also coming out of this work? Just because I think of the breakdown of the consensus, like these sort of breaking apart of things, I guess.

David Tse (00:54:17):
Yeah. So remember in university research is often curiosity driven, so it's not like you're trying to build a startup, right. It's like, okay, I'll solve this problem. Next time I solve this problem. Research has the liberty of trying to sort of identify interesting problem whether or not it's related to the previous problem you worked on. So that was now bring us into 2020, that was the beginning of the pandemic, I believe. And I was teaching a blockchain course at Stanford online and uh, we have to find some projects to work on for the students. And at that time, Ethereum published a paper on their consensus protocol for Ethereum 2.0, now we just call proof of stake Ethereum, is the Beacon chain protocol.

David Tse (00:55:03):
And that was the one that they will be, uh, uh, implementing towards the merge. Okay. And then I asked my student who was doing a project and said, Hey, why don't we try to understand what this protocol is doing? Maybe we can find something interesting to work on and then we spend a lot of time reading this protocol. It's a very complicated protocol. And one thing we observe is, is unlike any of the traditional protocol, it's not like Nakamoto's longest chain protocol, nor is it like a standard bizentine for torrent. So 40 years history protocol it's kind of a conglomeration of bunch of ideas. And so my question for a student is, okay, two question one is what objectives is this protocol trying to achieve, so why are they building something from scratch? Why don't they just take an off the shelf protocol and optimize that and build it?

Guillermo (00:55:54):
And by the way the student is Ertem, right? Sorry, is this Ertem wait, which student is this?

David Tse (00:55:58):
Yeah, Nusret uh, Nusret sorry. Nustret is, uh, my student. So that was his project. Okay. His project. And also with Joachim, another student of mine. Yes. As well. And um, so that's number one, question. Number two is, okay. Given those objectives. Can they achieve, can this protocol achieve those objectives? In their paper they never actually spell out what objective they're trying to achieve.

Anna Rose (00:56:22):
In the Ethereum paper that's yeah. The Ethereum paper. I mean the objective is proof of stake something.

David Tse (00:56:28):
Yeah. The objective is the proof of stake. They have some idea that they want to be so called accountable. That means they wanna punish people. But yet they want to share some characteristics of their Ethereum 1.0, which is this notion of dynamic availability keeps on growing.

Anna Rose (00:56:45):
Did It also have a little bit to do with just the fact that there had to be a migration and like how to do that migration? And if you do this splitting, then it becomes a little bit part of it can stay the same part of it gets reintroduced. I mean, as I understand now it's like switching out consensus basically.

David Tse (00:56:59):
Yeah. I think that was original thinking was to kind of put a gadget on top of Ethreum 1.0. Yeah. But then it sort of evolves to kind of like something very different. So there's no such putting gadget anymore, but still the research thinking

Anna Rose (00:57:15):
Was coming from it

David Tse (00:57:15):
It's still similar. And so that's kind of interesting thing about a research about engineering development is that often you're trying to solve a problem and the problem has changed, but your thinking is still the same. And so now you ask the inventor, you say, why are you doing this? Is that, oh yeah, that's a good question. It was because we're trying to solve some problem. Oh. But that problem doesn't exist anymore.

Anna Rose (00:57:36):
Right,

Guillermo (00:57:37):
Right, right, right. Okay.

David Tse (00:57:39):
And uh, so that led to our paper it's is called Ebb and Flow protocol. And uh, in that paper we formulated exactly what Ethereum is trying achieve. In fact, we got in touch with them and said, Hey, you know, maybe these are the objectives, this like, yeah, yeah. Actually these are objectives. So we came with these objectives and then we realized that the Ethereum protocol is actually not secure. The one they propose in that paper is not secure.

Anna Rose (00:58:08):
Not at that time.

David Tse (00:58:08):
Yeah.

Anna Rose (00:58:09):
Okay.

David Tse (00:58:09):
Yes. And we find attacks on that protocol. So that's the second thing and the third thing is we constructed a protocol that actually is secure and also achieved the objectives that we dictated.

Anna Rose (00:58:22):
Who's the we in this and at this point,

David Tse (00:58:25):
Yes. So this is with Nusret and Joachim. So your students, Nusret Tas and Joachim Neu. So those are the two students at Stanford in my research group.

Anna Rose (00:58:36):
Got it. I guess you looked at, it shared a lot of these results. Did it change it? Has it changed due to that?

David Tse (00:58:42):
Yes. So standard in security, when you find attacks you have to talk to, you should talk to the protocol people. You don't go and say, oh, okay. So,

Anna Rose (00:58:52):
Alright, go for it, build it. Go for it. I'll be first.

David Tse (00:58:56):
Yeah. So, so we engaged them and then we started a collaboration. So one thing we found was together with them. We found even more attacks.

Anna Rose (00:59:05):
Wow.

David Tse (00:59:06):
Okay. So the first thing was, in fact, we have a paper collaborating with Ethereum foundation called The Three Attacks. So two of them we found one of them, they found. Okay.

Anna Rose (00:59:16):
So who were you working with primarily there was it with like Justin and

David Tse (00:59:19):
Yeah.

Anna Rose (00:59:20):
Justin, Dankrad,

David Tse (00:59:21):
Dankrad And there was a group of younger people. Okay. Uh, and we have biweekly meetings. To talk about ideas and we realized that there were more attacks. And so we proposed a few things to try to improve the protocol. And it's still ongoing work. Really? It's still ongoing work, I think. Yes.

Anna Rose (00:59:44):
But what were so like changes, I guess you proposed some changes, shared disclosures. Did it change the design or was more like

David Tse (00:59:50):
Design changed, design changed. Yes. That's fine. Not significantly. Um, so one thing about research and technological development is that the timing is often important. Yeah. And we know that the merge is happening soon and there are 10 client teams, 10 client teams working on this, implementing this new protocol. So it was a little bit hard to make very big changes in the short time scale.

Anna Rose (01:00:15):
For sure.

David Tse (01:00:16):
So, our contribution was to identify some problems that they could look at and have some short term patch. But at the same time, come up with longer term design that they could switch swap in maybe at a later point of time, maybe not the merge, but maybe beyond the merge. Got it. So that's what we are working on right now.

Anna Rose (01:00:34):
Nice. So the project that you are also co-founder of Babylon, I need to, I think we need to understand what, like, is there a link between the work you did with Ethereum and that, and we should learn a little bit about it actually.

David Tse (01:00:46):
Yeah. So, uh, there's no link from a, uh, from a deployment point of view. But there is a link from a academic intellectual or from a research point of view. So let me explain that a little bit more to do that. I need to explain a little bit more about our solution to the objectives that Ethereum 2.0 was trying to achieve. Okay. So the objective there was trying to achieve two objectives. One is they want dynamic availability. That means they want the chain to keep on growing. No matter what the participation rate is. I think ... already discussed quite a bit of that. So I won't elaborate too much.

David Tse (01:01:24):
So in any proof of work protocol, you have this property and you certain 1.0 is this property and they want to keep this for Ethereum 2.0. But at the same time, they want to achieve a property called accountability, which means that if you commit a crime as a validator, you can be punished Ethereum 1.0 or Bitcoin doesn't have this property. They want this for Ethereum 2.0. And in fact that was one of the things that proof of stake can potentially give. And they want to exploit that. It turns out there's a negative result, which we proved is that you cannot get both at the same time. Oh, you cannot be accountable as well as dynamic available is impossible. Right? There's no protocol which can do that. And so the best you can do is kind of create two ledger where the longer ledger is dynamic available. Keep on growing. And then a prefix ledger, a shorter ledger is accountable. That means what that means that if you or have important transactions and you wanna protect it better, then you should wait a while before the transaction get into the shorter ledger. Right? If you have easier transactions, like buy a coffee. Yeah. Then you say, Hey, I want fast transaction. I don't need accountability. I can move on.

Anna Rose (01:02:37):
Is this like, could you use the word asynchronous here? Like a synchronous consensus? Or am I going, going in the wrong dimension?

David Tse (01:02:43):
A synchronous usually is a description of the network assumption.

Anna Rose (01:02:46):
Okay.

David Tse (01:02:47):
So here is a little bit analogous to maybe I can use the Bitcoin as analogy. Just analogy. So in Bitcoin, when you confirm a transaction, you can ask yourself how many block deep do you wait until you confirm a transaction? Many people wait, two block deep, one block deep. They confirm some people say six block deep. Some other people can say 20 block deep. So well, in fact, there's no single number that can be that needs to be determined system wide.

Anna Rose (01:03:16):
Okay.

David Tse (01:03:18):
So if you think about it, what does that mean? That means that Guillermo who wants to buy a coffee is now trusting the whole block as the ledger, the all blocks. Me who is buying a car is only trusting a prefix of this ledger up to the six block deep. Okay. So in fact that was one of the really cool innovation of Nakamoto. So in some sense, we are applying this principle to Ethereum problem is that the shorter ledger is accountable,

Anna Rose (01:03:48):
But have they implemented that design you proposed?

David Tse (01:03:51):
Yes. Yes. Yes. There's two ledgers.

Anna Rose (01:03:52):
Well, I know that there's a consensus in the execution. Yeah,

David Tse (01:03:56):
Yeah...No, no, no.

Anna Rose (01:03:57):
Two ledgers in the consensus.

David Tse (01:03:59):
Yes two ledgers in the concensus.

Anna Rose (01:04:00):
That I didn't know. Okay.

David Tse (01:04:01):
Yeah. So you remember there's this thing called the Casper FFG. Yeah.

Anna Rose (01:04:07):
Yeah.

David Tse (01:04:08):
It's called a finality gadget and that finality gadget determines the accountable ledger and then there is a so called, uh, ghost, which is growing the dynamic available ledger. So if you tease out the protocol, there are actually two ledgers there

Anna Rose (01:04:26):
But still going towards Babylon. What did you take from that?

David Tse (01:04:30):
Okay. All right. So now I need to say, okay. So if you think about it, accountable is really coming from the history of BFT because BFT is accountability and finalization is very similar. And so the best protocol to give you accountable ledger is actually a BFT protocol, not a longest chain protocol, but dynamic available comes from Nakamoto. So the best protocol, the simplest protocol is the longest chain protocol. And so once we realized that I said, Hey, why don't we just take a longest chain protocol and a BFT protocol off the shelf, both of them. And, but compose them in a nice way so that they can communicate each other in a simple way. To give you this two ledger solution. So our construction is instead of like the Ethereum approach, which is to build a brand new protocol, we try to be composable. So we try to put an off the shelf BFT protocol together with the longest chain protocol. And camera solution.

Anna Rose (01:05:28):
Oh, interesting.

David Tse (01:05:28):
Okay. And so that led us to sort of, uh, sort of a further level in our understanding of consensus is that often when you build new consensus protocols, you don't have to start from scratch. Yeah. You can try to take off the shelf technology. And compose them and try to get the best of both worlds.

Anna Rose (01:05:45):
Mm. So was this the first time, I mean, in the work you had been doing previously, you weren't really working on consensus specifically until I guess Prism, or would you say you were?

David Tse (01:05:55):
No, I've never worked on consensus.

Anna Rose (01:05:56):
Okay. Do you still say you don't because I mean, this is

David Tse (01:05:58):
No, I'm working consensus now.

Anna Rose (01:06:00):
Now it's yeah, no,

David Tse (01:06:03):
The area that I work in in blockchain is consensus.

Anna Rose (01:06:06):
Perfect. Okay.

David Tse (01:06:07):
Right. That's the, that's the area I'm working in. Um, but I've never worked on consensus before, before, but of course, while doing this, I learn about sort of all the history. Yeah. Because why should not ignore history? Right. I think research is about, okay, first you try to look at problem with, with a blank slate but once you have some understanding, you should look back at the literature and see what's been around and not be ignorant.

Anna Rose (01:06:29):
Totally.

Guillermo (01:06:29):
Right.

David Tse (01:06:29):
So going back to Babylon, this idea of composing off the shelf protocol can take it one step further. And the one step further is instead of just composing protocols, you can compose existing blockchains. Okay. So that's a little bit different. Uh, a protocol is something you, can you deploy yourself? Someone already gives you the protocol. Someone gives you sort the algorithm and you can deploy it. Right. That's composing protocol. But composing blockchain is really taking existing blockchain like Bitcoin. A running blockchain and composing it with another protocol. That you could develop together to form a more powerful blockchain. And so we start thinking is that, Hey, you know, there are two developments that we observed. One is that in an ecosystem like Cosmos, okay. There are many they so call applications specific blockchain, which the ideas that they want to give the developer the flexibility to build autonomous blockchain.

Anna Rose (01:07:37):
That can be designed slightly different.

David Tse (01:07:41):
Yeah - Optimized for the needs. For example, the MEV application, et cetera. Right. However, they lack the powerful, shared security platform of the theory, for example. But then we realize that actually there is some very secured resource out there, which is Bitcoin. And so the Babylon project started when you start thinking, how can we use Bitcoin to not to replace the autonomousity of say Cosmos zone, but to compliment the security of the, the Cosmos zone. And so this is a composition problem. We take a Cosmos zone and we want it to communicate with Bitcoin in a succinct way. In order to get extra security that it cannot get by itself.

Anna Rose (01:08:28):
But, can you, when you communicate with the Bitcoin network, are you always just like using the memo field? Like what can you do? How can you lock something in? I actually, this is sorry, this is just like my ignorance on Bitcoin, but I don't actually know how to use it as a checkpoint.

David Tse (01:08:42):
So the key here is really to use Bitcoin as a time stamping device. So, uh, Nakamoto actually mentioned in the paper that what I'm building the Bitcoin is actually a time stamping server. Okay. And so the, the real security benefit is really for the proof of stick chain to timestamp the events that are happening on the chain. So that they can be, everyone can look at those timestamp. So what is the timestamp? Right. A timestamp is really the fact that I'm sending, I'm saying, okay, an event has happened to proof of stake chain, that trigger a transaction that I sent to the mempool of a Bitcoin miner. The Bitcoin miner includes it into the next block it mines that transaction gets enough deep into blockchain secured and now I have a timestamp the timestamp is the which block. It appears in.

Anna Rose (01:09:40):
Immutable.

David Tse (01:09:41):
Now this timestamp is not like a very accurate clock because Bitcoin blocks. Yes. You know, are not, you have one block every time, minutes. So it's kind of like a pretty inaccurate clock at a short time scale. Okay. But at a long time scale, it's a very accurate clock. Because for example, if you get a block 20 block deep in, no one has seen a block has been replaced after it got 20 block in. No one has seen that in history of Bitcoin. That's right. Okay. So it's very reliable at a long time scale, but not so accurate in the short time scale,

Anna Rose (01:10:14):
I think. But, and maybe you just said it, but I didn't, I missed it. But like, but I, I still don't understand, like how do you connect it? How, like, where's the connection made if you're talking about an external blockchain, how does it get a message like into it? How does that...

David Tse (01:10:26):
Okay, so, so, uh, now we get a little bit technical. One of the powerful idea in cryptography is hashing. Okay. Now I have a block I've created this block in a proof of stake chain. Now to timestamp this block, one thing I can do is I can ship the whole block to the Bitcoin finder. And put the whole block into Bitcoin. That's impossible because that goes back to our original statement. That Bitcoin is a very low throughput. Yeah. Yeah. If everybody does that, then Bitcoin will completely swamped.

Anna Rose (01:10:53):
So are you hashing it? So you hash it and then you have a hash.

David Tse (01:10:57):
Yes. And you have hundred, 256 bits.

Anna Rose (01:11:00):
Okay.

David Tse (01:11:01):
And you post the hash as..

Anna Rose (01:11:04):
In the memo field, the memo

David Tse (01:11:05):
In the memfield, to get up with some signatures saying that your validators have approved of this.

Anna Rose (01:11:10):
I think you had said memfield. And I thought of, of mempool, but memfield. Got it. So it's a memo, it's the, yeah.

David Tse (01:11:16):
It puts into the memory of the Bitcoin miners. And now they put this hash as though it was a regular transaction.

Anna Rose (01:11:26):
Yeah.

David Tse (01:11:27):
That's and actually there is a special field. Yeah. Yeah. The memo field it's called op return. Actually, it's a very strange, oh, it's op return. Very nerdy term, very nerdy term. It's called op return. It's for putting arbitrary data into Bitcoin.

Anna Rose (01:11:40):
I see.

David Tse (01:11:41):
And this transaction will get into a block. It will be fixed forever. And so this is a timestamp of that hash and because of the cryptographic co security of hashing, nobody can create another block which have the same hash. So therefore the link between the hash and the block is secured

Anna Rose (01:11:59):
Babylon though, was work project, I guess, conceptually, but is it also now a company?

David Tse (01:12:04):
Yeah. So we had this, these ideas last year and then we, uh, raised a seat round of funding and now we are building the product.

Anna Rose (01:12:14):
Okay, cool. Now, and what is the product? Is it a tool? Is it the bridge between the two things like because it, or is it a zone of its own? That's and is it gonna be Cosmos? Actually? Yeah.

David Tse (01:12:25):
So Bitcoin is here. Cosmos is here, right? So the audience cannot read it.

Anna Rose (01:12:33):
They're far away.

David Tse (01:12:33):
They're far, they're far away.

Guillermo (01:12:34):
They're far away in some space...

David Tse (01:12:36):
They're, far away. So what we are building is in some sense, a Cosmos chain which we call Babylon and this Cosmos chain is like a connection between Bitcoin and all the other Cosmos chains

Anna Rose (01:12:50):
IBC, basically, I guess you're gonna be IBC enabled. So, so you're like, but just to Bitcoin using this particular technique for time stamping the hashes.

David Tse (01:13:00):
So, so the thing is that time stamping has to be done in a smart way because Bitcoin is very limited space. So our technology sort of optimized this time stamping. And then we use IBC to communicate with other Cosmos zones.

Anna Rose (01:13:16):
Nice. And they can go through that

David Tse (01:13:17):
To sub bridge. The security one we can think about is that, you know, bridging is about often bridging token from one chain to another. What we're doing here is we're bridging security.

Anna Rose (01:13:28):
Would you almost call it like, okay, it sounds a little bit like a roll up on Bitcoin. That's IBC enabled. Would that be at all fair? I know rollups have a different thing and I like I've done. I just did an episode about actually the comparison of like rollups to bridges and stuff, but maybe roll ups, the wrong word, but it's like deeply connected to the Bitcoin world.

David Tse (01:13:49):
Yeah. So it does have some similarity roll up. So while we're thinking about Babylon is aggregating many checkpoints from many proof of stake chain and represent that by one checkpoint efficiently into Bitcoin. So in that sense you can think of a roll up can do this job, but we decide to use the Cosmos chain as our infrastructure to build this instead of a roll up, because actually Cosmos SDK is a much more mature technology than building a roll up number one, number two is that IBC is already an established technology between different Cosmos chains. So we can immediately leverage that. So I think that's why we chose as an implementation platform the Cosmos chain, but you do, I do agree from a functionality point of view. It has some similarity as, as a roll up.

Anna Rose (01:14:36):
So although I guess the different, like, there is no transfer of like token though. There's it's not, yes. There's you're not locking on one. Okay. So

David Tse (01:14:44):
The more technical term is that we are not using Bitcoin as a sediment layer.

Guillermo (01:14:48):
Right. That's right. That's right. So

David Tse (01:14:50):
Correct. There's no transfer of token.

Guillermo (01:14:52):
Okay. So here's actually a quick question. So, uh, when let's say I were to, you know, have a Cosmos ecosystem and you know, a chain, right. And I go, and I wanna, I wanna have this checkpointing actually, uh, even just, this is a very type of question, but uh, how do you, so what do I pay you specifically A and B how do you actually, um, how do you even bid for the transactions for the Bitcoin miner? So that's a, that's a non-trivial problem as well. Right. To be included to you, have to yeah.

Anna Rose (01:15:20):
You have to pay for that?

Guillermo (01:15:22):
So yes, in Bitcoin.

David Tse (01:15:23):
Yeah. So in, so first of all, Babylon the chain has to pay Bitcoin. To get the check point in. Yes. But that's the business between Babylon and Bitcoin?

Guillermo (01:15:34):
No, no, of course, of course. But, but somehow it has to get translated to people who are, you know, paying you for the service.

David Tse (01:15:41):
So that's right. That's our cost basis.

Guillermo (01:15:43):
Right.

David Tse (01:15:43):
From a Babylon point of view. Right,

Guillermo (01:15:45):
Right. And you can't just buy fees,

David Tse (01:15:47):
But I can't just keep on paying and not getting anything. So when others proof of stake chains, Cosmos zones put checkpoints into Babylon, through IBC, they have to pay.

Guillermo (01:16:00):
That's right.

Anna Rose (01:16:01):
Actually, is there anything else that you wanna share about Babylon maybe where it's at or like if you have any sort of roadmap?

David Tse (01:16:07):
Yeah. So maybe a few words. So right now we have a, uh, engineering and research team, the researchers mainly come from Stanford, uh not surprisingly, the engineering team is a really good layer one engineers around the world. So we have, uh, five engineers working for us and three or four researchers. That's our team right now. Uh, we are working towards, um, a demo, at Cosmoverse. So I don't know if you're going or not,

Anna Rose (01:16:38):
I would love to, but it's so there's so much travel this fall.

David Tse (01:16:42):
Yeah it's a, it's in Medellin .

Anna Rose (01:16:43):
It's like three weeks before DevCon.

Guillermo (01:16:46):
That's right. Oh, wait. That's right. Because that one is in Medellin and then, uh, EthCC is in Bogota.

David Tse (01:16:51):
Bogota yeah I think so.

Anna Rose (01:16:52):
Or DevCon.

David Tse (01:16:54):
DevCon.

Anna Rose (01:16:55):
But it's not at the same time.

Guillermo (01:16:56):
There's like two weeks in the middle.

Anna Rose (01:16:59):
I know some friends who are gonna like travel in the middle.

Guillermo (01:17:01):
Yeah that sounds pretty lit.

David Tse (01:17:02):
Yeah. So we're hoping to demonstrate our ideas there and get some feedback from the Cosmos community. So that's our next step in the roadmap. And then we'll use that feedback and build a testnet and hopefully we'll find some interested Cosmos zones to do a joint testnet integrated testnet with us. So that's the roadmap in the next six months. Yeah.

Guillermo (01:17:25):
Well, actually I have, I have a question. Why, why wouldn't people, uh, use the, the protocol to bootstrap their, is that, is the idea mostly for bootstrapping or is it for actual, like kind of, you know, continual protection so to speak?

David Tse (01:17:37):
Yeah. So there are two, two or three use cases that we are working on. One is, um, reducing the unbonding time,

Guillermo (01:17:45):
Right? Oh, that's a good point.

David Tse (01:17:47):
So Cosmos zone has a 21 day unbonding time, uh, using a process called social consensus. Which basically means that the off chain you have to agree on what's the canonical chain. We can use Bitcoin to replace social consensus using our technology, and we can reduce the unbound time to one day. So that's the number one use case. The second use case is, as you mentioned, bootstrapping proof of stake has a very interesting characteristic. It has very fast confirmation by very slow unbonding.

Guillermo (01:18:15):
Right.

David Tse (01:18:16):
Very slow unbonding. And the unbonding is 21 days in the Cosmos zone. And, uh, and that's because of something called long range attack.

David Tse (01:18:26):
And basically using Babylon with Bitcoin security, you can solve the long range attack. And that sort of connects to what I said earlier is that Bitcoin is, uh, good on a long time scale, not good in a short time scale so it's in fact, a very good compliment to a proof of stake chain because proof of stake chain is very good in a fast time scale because it does fast finality. But a long time scale is this long range attacks. And so they're very good compliment to each other. So that was the first use case. Bootstrapping is another use case. A third use case is for example, you can imagine if you wanna protect some important transactions then maybe just finality on the proof of section is not enough. You wanna wait until the timestamp becomes sufficiently deep onto Bitcoin, and then you confirm, so then you can have some differentiated service. So those are the use cases that we've been exploring.

Guillermo (01:19:19):
Yeah. Now, now, now I'm wondering if you're gonna enable the first, you know, 35 block, deep reorg of Bitcoin, because someone wants to do something else on a proof of stake chain, you know...

David Tse (01:19:28):
That would be right. Expensive. Yes. That could be rather expensive.

Guillermo (01:19:30):
Yeah. To say the least.

Anna Rose (01:19:33):
So you're doing both you're, you've co-founded this company, but you're also still a professor at Stanford, I guess. So what kind of topics do you focus on today?

David Tse (01:19:41):
Yeah, so I'll be on leave from Stanford. To work on this.

Anna Rose (01:19:45):
Oh, you will? Yeah. Oh, wow.

David Tse (01:19:46):
So in some sense, a lot of my research right now is kind of focused around this idea of sort of, how do you leverage off sort of different blockchain characteristics to get security? For example, here's a research problem we're looking at not necessarily tied to Babylon, but sort of motivated from Babylon is, okay. I talked about checkpoint the Bitcoin, right. But maybe you can checkpoint to multiple chains. Why only one chain, maybe you can checkpoint the Bitcoin and Ethereum and, and Cardano, right? Yeah. And you can say that, Hey, maybe as long as the majority of them are secure, then your protocol should be good. So that's kind of a research problem that we're working on right now, uh, sort of motivated from sort of this idea of borrowing security from other chains.

Anna Rose (01:20:33):
Do you have your own consensus as well? Are you using Tendermint yourself? Yes. Under the hood?

David Tse (01:20:38):
Yes, Tendermint. So again, right. The research is about sort of combining sort of off the shelf protocol. So therefore we're not inventing any new component. Bitcoin is Bitcoin, Tendermint is Tendermint, so we're just trying to integrate them in sort of a meaningful manner to get the full security benefits of both.

Anna Rose (01:20:55):
Very cool. Well, congratulations on the move, the switch from academia. This is the first time you found..

David Tse (01:21:03):
This was the first time. Yeah.

Guillermo (01:21:05):
You, you're kind of already vaguely familiar with the, the, you know, how the whole spiel goes. You went to work for Qualcomm for a bit. And so that's right. Oh, I think this is probably gonna be slightly longer.

David Tse (01:21:13):
Yeah. So in the early days in wireless, often times is done by bigger companies, like,

Anna Rose (01:21:19):
Right,

David Tse (01:21:19):
Right. So to impact industry, you kind of have to work with them.

Anna Rose (01:21:22):
It sounds like we'll be seeing you around the Cosmos events too. That's cool. Yeah.

Guillermo (01:21:26):
Yeah. That's gonna be fun.

Anna Rose (01:21:27):
That's a world, that I've been spending time in, the validator, zkValidator.

David Tse (01:21:31):
Great.

Anna Rose (01:21:33):
We're there.

David Tse (01:21:33):
That's your, that's your another hat,

Anna Rose (01:21:35):
Another hat. There's a couple hats.

Guillermo (01:21:37):
One of the many, actually one of the many, a couple I would,

Anna Rose (01:21:39):
Guillermo is wearing a zkHack shirt right now, which is another hat. There you go. There's a couple couple,

David Tse (01:21:44):
Which the audience cannot see,

Anna Rose (01:21:45):
Which they cannot see. So I told them.

Guillermo (01:21:46):
Thankfully, unfortunately,

Anna Rose (01:21:48):
Um, um, but very cool. Thank you so much for sharing with us, your kind of journey through all of these different fields, bringing you to blockchain, like helping you understand Bitcoin, bringing you maybe into like non Bitcoin ecosystems, like Cosmos and building your own company. This is amazing. Yeah. Cool.

David Tse (01:22:03):
It's so fun talking to you guys. Yeah, really, but thank you. Yeah. I hope it's not. I hope I wasn't too long winded in this. Describing this...

Anna Rose (01:22:11):
The way I think of it is the audience has now been introduced and I think we'll, we'd love to continue our conversations when we have topics that would be of interest to you

Guillermo (01:22:20):
Possible interest.

David Tse (01:22:21):
Thank you.

Anna Rose (01:22:21):
Thanks again. So I wanna say thank you to the ZK podcast team, Tanya, Rachel, and Henrik, and to our listeners. Thanks for listening.