Fireside Chat With Grigore Rosu and Kartik Talwar

We're going to start off the session with a chat with Regory from Pi Squared. So, please give him a big round of applause. [applause]

Welcome.

Um, there's a lot of things I want to actually jump into and I have a a pretty big list of questions. And uh the thing I want to start off with is uh actually having you tell everybody uh what you're doing at Pi Squared and uh what kind of made you start this project.

Um so I'm the CEO and founder of Pi Squared. I'm also professor of computer science at the University of Illinois Urbana Champagne and I started this project because I believe that there is a more efficient way to implement the infrastructure that we need for web 3 and crypto both for humans and for AI agents.

So what is the product?

We [clears throat] have two products. Um one is um actually the name of our company pi squared comes from proof of proof which is a universal verifiable computing uh approach where you can get zk proofs for computations done in any programming languages going through mathematical proofs. So we call them correct by construction. That's product number one. And product number two is the decentralized network on which all these proved claims can be settled which we call fast set.

So I want to focus a lot on facet. Um I obviously searched things up and I got to uh read about how it's set up and what it is and my kind of first question to you is why make another chain and um you've also kind of talked about blockchain. to kind of help us understand what the features are of fast set and what made you do this as a separate problem set on top. Right? So the thing I should say is that fast set is a decentralized network a large decentralized network but not a blockchain and that is an important aspect and why we did something a decentralized network which is not a blockchain.

The main reason is scalability. We wanted a super fast decentralized network that can be used to settle anything that is verifiably true. Any claim that is true needs to be settled. And currently we settle them on blockchains. And we actually first try to do exactly that with our first product, the universal verifiable computing approach.

We had claims that had proofs. All the heavy lifting has been done offchain to generate the proof and then we just had to settle the claim and we implemented it as a blockchain as a layer two on an optimism stack and although there was no computation to be done just settling the claims we could not go beyond a few hundred such claims per second to be settled and that's too low uh it's too low even for humans. So, so what was the bottleneck and and kind of how do you get rid of that for facet?

Exactly. So, what was the bottleneck? Because many many many of us think that the bottleneck in verifiable computing is the generation of ZK proofs which is a bottleneck in itself. But we realize that actually there was another bottleneck. The actual elephant in the room was the blockchain architecture itself which we all love but we also have to once in a while analyze and understand deeper.

So what happens on blockchains to remind everybody is that there is a network of validators and they communicate a lot in order to agree on a total order on all the transactions on the blockchain and they agree on the order and then everybody executes the exact same transactions and they get exact same state. Agreeing on that order is very expensive. It requires a lot of communication. There is a lot more communication and and computing power in this interaction.

And this is specifically talking about the gossip between validators to understand if they have the same order or you mean just the ordering within a validator set

just in ordering the transactions and agreeing on the order. So that is so if I can say metaphorically since we are in Argentina imagine you have a a football field not soccer football field right and you have all the players on one side including the reserves like 50 players on one side and their job is to run as fast as they can to the other side of the of the football field. They will probably make it in 15 seconds or so plus minus one second. Some of them could be a bit slower. But now imagine this.

Imagine that you put a door in the middle and everybody has to run through that door. No matter how good they are, no matter how well trained they are, there will be congestion at the door. Okay? No matter how well you sequence them, there will be congestion at the door. So that's metaphorically what happens.

Uh on the blockchain, we need we have this total order discipline that comes at a cost. Uh no matter how much we try to improve it, it will come at a cost. And we wanted to explore the possibility of going beyond that. Is there a way to settle things that are already true that you know they are true at the speed of the internet as opposed to having a lot of communication that slows down. So I do want to be um not too kind of I forget the audience but there's this reason why transactions are and the order matters especially when you need money and management.

So how do you solve that without having you know and kind of what are the folks that

right? So this is the question. Um so in early 2000s it was firmly believed and Nakamoto was a product of all that firm belief back then that in order to avoid double spending we need to totally order all the transactions because if we totally order all the transactions in a chain which was also a term that Nakamoto introduced. If we order everything in a chain, then there is no way to double spend, right? Because if you submit two transactions um with the same nons, one of them will be sequenced before the other and the second one will be rejected.

So total order will guarantee no double spending. Um however something that Nakamoto didn't know uh is recent research uh published in uh 2019 u it was a scientific paper um from academia initially where it was shown for the first time that there is another way to avoid double spending where the idea is that all the different uh participants clients have their own total order. Remember the soccer the football players, right? Each of them is sequential step after step after step, but all of them can operate in parallel. And all the validators have to do is to keep track of each account and their sequence number.

And if you try to double spend, that means that you submit two transactions with the same sequence number, then one of them may arrive before the other one on some validators. another can arrive in the other order on some other validators and that means that you will not be able to get quorum on both cororum means more than 2/3 2/3 plus one the same security that you have in proof ofstake protocols

so that seems like a problem

exactly so you cannot get quorum on both if you can get quorum on one actually there's a lucky case because you can use that quorum to settle the transaction one of them the one you got corum for but if you end up with like half validators giving you signatures for one, the other half for the other. Now actually what happens and this is the catch you locked yourself. Your account is locked. You cannot continue because you cannot continue unless you get quum on a transaction. In other words, [snorts] blockchains and bitcoin solves a double problem, the double spending problem by picking for you one of the two transactions and the future goes on.

while this protocol don't cheat. If you try to cheat, you will suffer the consequences including that you may end up locking your account.

So, is that lock up permanent or is that a condition that you now have this kind of forum resolution and then something is resolved which means you're going to have to remove the other resolution set like how do you settle that at the end of the day? First of all, that is a rare case that shouldn't happen. That only happens if you make a massive mistake or you try to cheat. For this rare cases, we need consensus. Now, the validators have to talk to each other.

You have to pay a higher fee to unlock yourself. So, but that is the very rare.

That's a good way to then think about effectively partitioning the problem. This is super cool. So my my kind of obvious question is um in a way you were kind of describing parallelism um we've seen approaches like magic development and kind of parallelm can help us contrast this with kind of EDM paralization

yeah a big big push for parallelism because that's the only way to make things faster we cannot make them sequentially faster we have to parallelize um and let me let me first tell you how the protocol works because it's so simple I can describe it in less than a minute. Okay. So, so suppose that you have something which is true a claim that you know is true and you have to convince all the validators and to get quum from them that you are right. This is what you do. You broadcast the claim to all of them.

They don't even talk to each other. They don't need to talk to each other. No communication between validators. Now they in parallel they will all validate the signature of the claim and then they will give you a thumbs up back. They will send you a signature back that uh yes it's valid and now you aggregate the signatures but you need a corum.

Once you have cororum that is your proof that you have global consensus that your claim is true. You see so there are no ordering between claims and then you do the same with the other claims and others do exactly the same in parallel with you with all the claims. The only limit is the internet. How much bandwidth the validators have and how much computation they have to verify signatures basically and you can verify like a million signatures per second even on normal machines these days. This is completely different from blockchains and it's massively parallel by design.

Now getting to blockchains that uh that enhance parallelism like mega, monot, sui. The essential difference between the two approaches is that blockchains use parallelism as an optimization of an otherwise sequential chain sequence of of processes and some of them like mega which I think is one of the best blockchain technologies these days. Um, Mega is also sequences the transactions in such a way to exploit parallelism. So you

but it's still total ordering at the end of the day.

Yeah. Several of them, right? So there is an art how you sequence the transactions so that they independent so you can run them in parallel very efficiently but you still have a total order in the end. While Facet decided to jump all the way to the other extreme. The universe is parallel.

If you need sequentiality implement it. So for us a sequential there there are some some some smart contracts that are inherently sequential like unis swap because the price of the token is given by each transaction in sequence or a or there are several protocols which are inherently sequential. Those need to implement their own sequencer. They have to decide they have to do clustering on their own group. They have right they have to provide their own sequence but they all live in parallel in a completely parallel asynchronous uh universe and there are techniques to implement total order there are techniques to communicate in an asynchronous universe all these are well established distributed systems and concurrency techniques

that's awesome um it's a followup question does this mean fast

um we remember when we started with universal verifiable computing networks with all languages. We really wanted to separate that concern.

So this is just proof aggregation that verifies every

you can plug and play your programming language. Basically as far as you know how to generate proofs and verify proofs for a certain language. Now you can apply plug that language. So you can in principle write smart contracts even in Java and Python. Um but of course EVM is u pre-built.

We have EVM, but you can also add new languages and we hope once we launch a network that you will add your favorite languages to the network.

Amazing. Um the I mean there's so many questions and I feel like the inner nerd in me wants to go too deep but I want to make sure the rest of the audience. Um so I can do a mix and answer this. Um the thing I'm curious about is your background is internal verification and this is kind of what you've been doing for many years before P square. So walk me through how safety is part of fast just general in general and um especially when you say that you can now any programming language in a way that opens up a lot more opportunities for and risk and general bugs.

Um so how do you think about that learnings from the past life incorporated into us? All right. So, um, yes, I'm a professor of computer science. I teach programming languages. I've been doing that for more than 20 years by now.

But I also teach concurrency and distributed systems. Um, so everything basically that we need for for for the blockchain uh world. Um, so I think the the key insight here is separation of concerns. So we wanted to separate the programming language from the consensus protocol. Um and we've been trying we've been doing that for a few years by now.

Um another company I founded runtime verification does security audits based on formal verification using formal semantics. We also use formal semantics in pi squared but for a different purpose to verify executions of programs. In runtime verification, we use formal semantics to formally verify programs that programs do what they are supposed to do no matter how they executed. The symbolic base verification in Piquare to use the semantics to generate a mathematical proof in real time that the execution of the program is compliant with the mathematical definition of the programming language, the formal semantics. And we invented a framework called the K framework.

Before UIC, I worked at NASA. And while at NASA, I started this framework, the K framework, which essentially what a semantic framework does in a nutshell. It reduces computation to mathematical proof in real time, not cryptographic proof, mathematical proof. And now you have a mathematical proof that fu evaluates to 42, let's say. And that mathematical proof is the ultimate correctness argument that the execution of the program executed as intended as a completely separate task.

Now we can take that mathematical proof and we have a zk circuit for mathematics that checks the mathematical proof and then gives you a zk proof that the mathematical proof is correct. So proof of proof by square. Um now we got back to the to the original problem. I have a claim and a succinct cryptographic proof that the claim is true. I need to settle it.

This is where fast set comes into the picture. So you see all the computational correctness and formal verification happening before the

separated and you can do that now even security audits now can become claims on chain because I can have a claim which says that this is a correct implementation of unis swap x * y= k protocol. Actually we've done that for unis swap right so you have that mathematical proof which now becomes an audit which is onchain so you can now if you are bridge you say I'm not going to take a token unless you have a proof that token

so it's it's it's really great when you separate concerns because lots not only that you can optimize beyond obvious but you can also imagine new applications and new usage

that's amazing so how does all that lead to practical metric talks. How do you measure how well asset works? Is this TPS? Is that the metric? We're looking at gas in the market or through size.

What do you think about those metrics and where are they right now?

Right? So [clears throat] because it is embarrassingly parallel, it is expected to be fast. Right? So being fast, we we're not showing off. It's the protocol that is fast because it's embarrassingly parallel.

But even in the first implementation, the first cut after one month and a few weeks actually of of effort to implement it, we got more than 50,000 transactions per second.

And these are just transactions that have just proved the

right say payments. Payments payments are some of the simplest claims because they your signature is the proof basically, right? So just payments like what bitcoin does, right? So imagine a bitcoin but on facet, right? So we could do 50,000 suctions per second in the first implementation.

Then in a couple more weeks we exceeded 100,000. Now the version we have now, we just launched the DevNet 2.0 does I know more than 150,000 I think, right? I think we exceeded 150,000 and we have we have an internal version that we reach 250,000 and we'll make that also public soon. So but but again there is no TPS

because that's just hardware bound at the end of the day.

Exactly. Payments in particular are very important because um um we see a lot of interest in AI agentic companies to to settle all the payments of agents uh on chain uh using web3 technology including the X42 advance and and variations of it and all this requires a super fast scalable. It has to work at the speed of the internet. And if you're doing thousands of transactions per second, which most websites do,

you have to be able to support this.

And and and and this is not known. I take it as a personal mission now to tell everybody that you do not need a blockchain for payments for peer-to-peer decentralized secure payments. You do not need to pay the price of blockchain. That doesn't mean that blockchains are not great because they're great. And Ethereum in particular, I love Ethereum.

I'm big fan of Ethereum. And actually as a matter of fact we integrate with blockchains. Um for example every single block on Ethereum becomes a claim and is settled also on facet. So for every single everything that happens on Ethereum is visible on facet. So bridging for example is for free in some sense and we do the same for optimism the same for mega that means that you have an additional layer of verification and at the same time you have access you are interconnected with all the blockchains.

Yeah, I mean my last question to you is going to be just in a way we kind of went through all the things that you want to kind of kind of parallelism but how does this compare to where Ethereum now is competitive complimentary you can answer that is there any other positions you might want to think about

yes so so right so first of all we want to see everything that happens on Ethereum and also the other way around because we verify with KVM M with the formal semantics of EVM, every single transaction on Ethereum, we get assurance. It's like having an multi-client, another client for Ethereum that checks that everything is is compliant. Um and I personally believe that what we'll see in the next maybe couple of years is um you know a convergence. We come from the massively parallel angle. Blockchains come from the totally sequential model and they add parallelism and um and um I think many of the Ethereum contracts can be run still using EVM as appchains um in um in their own account and then submit the claims the same way the entire blockchain um some now submits claims to to the facet network whether facet or not that's not what's important but what's important is I think blockchains have to go parallel to

there's no other option

because you already know what the limit is right now you won't be able to use it that's going to be the way you want to get

Yeah. Yeah. But you see same like in a car you don't really care what's under the hood in your car whether it's an electric engine or a gasoline engine. You just go from point A to point B. The same with blockchains.

Users will just use blockchains and they don't really care if there's a total order or not in the end. They care whether things happen fast and cheap. Um and I think um slowly blockchains will migrate um into such a beast eventually.

Well, thank you so much for this amazing conversation. I got to tell you things about building asset and please give him a big round of applause.

Thank you. Thank you. [applause]