S17 E52: Michael Sutton on Kaspa, Covenants & Research
Search this conversation
0:00# S17 E52: Michael Sutton on Kaspa Covenants, DAG Knight & Chess On-Chain
0:00**Podcast:** Bitcoin Takeover, Season 17 Episode 52 **Guest:** Michael Sutton, Kaspa core developer, DAG Knight co-author **Host:** Vlad Costea **Sponsors:** Cake Wallet, SideShift, Braiins Hashpower, Orange Rock, LayerTwo Labs
0:00---
0:00## Transcript
0:00**Vlad Costea:** Hello there and welcome to the Bitcoin Takeover podcast. This is season 17, episode 52. We are live with Michael Sutton. This episode is sponsored by Cake Wallet, SideShift, Braiins Hashpower, Orange Rock, and Layer 2 Labs. This is one of the most requested interviews of the year. Starting last year I've become interested in experimental new generation proof-of-work systems. I've had people from Kaspa, people from Quai, and it's quite interesting to see (quite interesting, bad pun) how these projects evolve over time.
0:00Michael Sutton was mentioned during my interview with Yonatan Sompolinsky this summer. We were driving together through Romania and he asked me, "Have you ever spoken with Michael Sutton?" And I was like, "Who's that?" Now I felt quite stupid for asking that question, but this interview has been very requested and people really want to hear what you have to say. You are working on research and development at Kaspa and you're also a co-author of the DAG Knight paper.
0:00Good to have you, Michael. Let's talk about avant-garde proof of work.
0:00**Michael Sutton:** Hi Vlad. Happy to be here. I hope we will be able to give people what they want.
0:00**Vlad:** Oh, we never give them what they want. [laughter] Don't delude yourself. We never do that.
0:00**Michael:** Exactly. They will never try and we always fail. Ah, they're never happy. Okay.
0:00**Vlad:** No, we do our best. That's the most we can do. And I do want to ask you because I know that this is a Bitcoin podcast and there are two branches of questions with which we can start. First of all, do you have a Bitcoin story and something that helped you discover what this space is about? And secondly, how is Kaspa related to Bitcoin?
0:00**Michael:** My Bitcoin story is that I was introduced to Yonatan's research. And I obviously traced back a bit as a researcher and came to the major reference called the Bitcoin white paper. We're talking about 2018, I think. And I had the aha moment of reading the Bitcoin paper. And then I went back to Yonatan's papers and enjoyed very much the beauty of them.
0:00**Vlad:** And how is Kaspa related to Bitcoin?
0:00**Michael:** Well, at the protocol level, Kaspa generalizes the proof of work system of Bitcoin into a more general structure which isn't a blockchain but is a blockDAG. So there are obvious relations. We inherited multiple things from Bitcoin except the proof of work: also the UTXO model. And we'll probably touch on more stuff as we go.
0:00**Vlad:** So you say that you discovered the Bitcoin white paper more than eight years ago. What were you doing at the time? What's your background?
0:00**Michael:** I was finishing my master's degree at the Hebrew University in HPC actually. So not in distributed systems, not in consensual systems. I was actually researching high performance systems, working with general purpose GPU programming at the time, parallel algorithms, a lot of graph algorithms.
0:00And I happened to meet Yonatan at some advanced communication course and he came up to me at the cafeteria and just asked if I would join his startup which was implementing his PhD. I knew nothing. I don't remember exactly the conversation, but it was basically like that. We spoke for the first time and he mentioned it and asked me if I would be interested. I couldn't join at that time exactly, and also I think the startup took a few months to actually be evolved. But I joined about one year later when I became more available.
0:00So that's it. And then I started looking and studying and understanding consensus protocols, proof of work. My first project in DAG Labs, the startup that originally founded Kaspa in a sense, was actually related to future L2s. A circle closed not so long ago.
0:00**Vlad:** Oh, interesting. So you say that L2s are a closed circle. What kind of L2s were you researching?
0:00**Michael:** At the time, ZK systems or validity proof systems weren't practical enough. So we looked at an Arbitrum-like design actually, like a fraud proof system where there's a bisection protocol. You execute off-chain and validators report the state updates to L1, and if there's a disagreement they kind of run a dispute, a bisection protocol where L1 essentially eventually needs to validate a very narrow set of instructions of the L2 VM. That was the sketch we looked at at the time, and we tried to design it over the UTXO model.
0:00**Vlad:** Was this going to be an L2 or was this research for Kaspa?
0:00**Michael:** It was research for Kaspa, for how to build an L2 for Kaspa, already before launch, already two or three years before launch. It wasn't my main research. I only finished some kind of sketch document or a very basic yellow paper. And then I moved on to other questions, one of which led to DAG Knight.
0:00**Vlad:** Oh, that's interesting. DAG Knight, which is this protocol that is rumored, I'm not sure if it's demonstrated yet, but it's rumored to survive like a nuclear winter and be the only one that can outlive humanity. You were thinking about that before you actually launched Kaspa?
0:00**Michael:** Yeah. So why didn't you launch Kaspa with it?
0:00**Michael:** We actually published the first version of the paper in 2022. So that's a year after Kaspa's launch. And then other things were more important for Kaspa's future, the immediate to short, short to midterm. Like getting more utility, getting some kind of expressiveness, programmability in all kinds of forms.
0:00We also, my first main project in Kaspa as an engineer, pre-launch I was mainly a Kaspa researcher. I actually became, I took more engineering roles only after launch. So I helped stabilize the network in the first half a year. It didn't run smoothly from the start. There were hiccups. Because nothing can prepare you for a global peer-to-peer network acting at those rates. You can do as many testnets as you want with a bunch of nodes, but nothing really prepares you for a real-world peer-to-peer network at those block rates. At the time, one block per second.
0:00So my first main project over Kaspa was actually suggesting a rewrite of Kaspa to Rust, which is known today as Rusty Kaspa and is the actual main client of Kaspa. And it led to the 10 BPS we can see today. We had to improve performance significantly in order to sustain 10 BPS without stretching node limits.
0:00So yeah, I touched a lot of things. Pre-launch I was more busy with research, then became more engineering: 10 BPS, then programmability and covenants in all kinds of forms with the recent hard fork named Tokata, which brought a lot of these capabilities to L1. And now we're back to DAG Knight. There's an active team of people finalizing all aspects of the protocol these days, and we're doing a lot of progress, accelerated by AI I think, which is very significant.
0:00**Vlad:** For the people listening to this who are hardcore into Bitcoin and don't know anything about Kaspa: BPS is not beats per second, which is a metric used in music. It's blocks per second, which is something that Kaspa does. In Bitcoin, you have a new block every 10 minutes approximately. On Kaspa, you have 10 blocks every second. And the way that the system works is that it doesn't orphan blocks. Every block discovered is one that's accepted by the network. It's their way of unlocking high scalability.
0:00**Michael:** Yeah. So the blocks, obviously many of them are mined in parallel and they form a blockDAG, which typically has a narrow structure because blocks do propagate. So this blockDAG doesn't get very wide, has a typical narrow structure. And in that sense, it's a generalization of blockchain. It's a blockDAG, but it still has a fixed width. If you look at a chain you'd say it has zero width or one, and we optimize the substructure of the DAG which has a narrow width, called a K-cluster. A zero-cluster would be a chain, and a seven-cluster would be a structure which tolerates up to seven additional parallel blocks beside myself as a block, for instance.
0:00The consensus protocol of Kaspa optimizes this structure. The current consensus protocol of Kaspa is called GHOSTDAG, and DAG Knight is an evolution, I'd say, of GHOSTDAG that removes the latency assumption from the protocol. But it is also based on the idea of optimizing K-clusters, just allows the K to vary and be more dynamic.
0:00**Vlad:** This is very high-level math at this point. I don't think there are too many experts in blockDAG and DAG Knight at this point. Maybe you're one of the few people who actually understand how this works. But I gotta ask you something because I remember when I first interviewed Yonatan last year, he told me that he had a conversation with a Bitcoin core developer whom he did not reveal, he didn't say exactly which one, but he said that what that Bitcoin core developer said he would find impressive about the Kaspa system is if it unlocked 100 blocks per second. That would be like an aspiration. Was he talking about Elai?
0:00**Michael:** Yeah, I remember that saying. Moving from 1 BPS to 10 BPS was part of that kind of thinking. We can talk about the different realms. Going from 1 to 10 BPS still has obvious advantages of UX, like the human eye still appreciates sub-second frequency. Going from 10 to 100 is more about decentralizing within the latency round of the network.
0:00If you heard the term "real-time decentralization." Think about it like this: in a 100 BPS system, you get 100 votes, samples from the mining, from the global mining hardware mining Kaspa every second. So you have a very good statistical sample of the network. And if the miners vote over something, you can actually wait only a second or two and still get very strong statistical guarantee that you actually see the majority opinion.
0:00The realm from 10 to 100 BPS is also something we're seeing. We shifted, I mean, each increase, each significant increase in BPS isn't just a switch. You can't allow the node to be stretched. The normal processing, even current software of Kaspa can process 100 blocks per second. But the node, think about a new node, it has to catch up for the network, it has to sync from a few days ago at least. So you can't make the 100 BPS the limit. You can't put it on the limit. The average normal behavior needs to occupy much less of the resources of the hardware.
0:00So in general, you have to do significant performance work in order to keep increasing BPS. There's also more questions that are more theoretical. But yeah, that's why we can go, I say with current performance levels we can go to 20 BPS already now. But it's still an effort and it requires a lot of testing. So currently we're focused on DAG Knight now. At least some of the current team is focused on that. Not everybody.
0:00**Vlad:** My question was about to be: what do you think it takes to get to 100 BPS? Will it be better hardware or faster internet, or does it just have to be software engineering?
0:00**Michael:** It's not about faster internet. The protocol can tolerate... I wouldn't say it's about faster internet. I'd say it's about reducing the role of a block header. A header is, you think of a header as a sample of proof of work from some miner who votes on something and chose some transactions to include in that block. And a header has all kinds of processing requirements when you process a header. And you would want to allow that sampling of proof of work but require less processing per header. Maybe, I think it requires a significant change. Already in GHOSTDAG we can get to 25 maybe BPS or something like that. But going to 100 requires another, more R&D.
0:00**Vlad:** All fair. I'm just pushing something that I remember Yonatan mentioned as a challenge. And I think the next section is going to be like a loyalty test for you, because you've been with the project for eight years almost. You said it took you about one year since the first proposal to actually join the team. And you also name-dropped Arbitrum as one of the projects that you were looking into. I'm pretty sure that if you joined Arbitrum at some point you would have made more money. And if you looked into these ZK rollup systems that were deployed on Ethereum and Bitcoin, you could also make a lot of money. My question to you is: why did you stick with the Kaspa project, in spite of the fact that you could have been paid much better switching teams and going to do something else with another project?
0:00**Michael:** I'm not sure I'd get paid better. But it never crossed my mind because in Kaspa I get to, first of all, I got to work with Yonatan. We have kind of a special relationship as researchers together, him as an adviser, me as a younger researcher. We're a good complement to each other and we had a lot of good research sessions and research journeys. And also in Kaspa I have a special role of helping to lead and innovating at the cutting edge, and being a part of a bigger revolution in my point of view. I don't know how far Kaspa can get, but I believe it earned the right to succeed at a much higher scale than it does today still. I think things that we did in Kaspa and are doing in Kaspa are very significant. Either Kaspa will succeed or they will influence some future wave. They are part of the common knowledge of crypto now. And it's a more natural place for me. I'm not seeking to get rich. It's not the point.
0:00**Vlad:** That's a good answer. The other question to test your loyalty would be: what do you think it would take for you to start working on Bitcoin? Let's say that Bitcoin activates OP_CAT or CTV or something that lets you deploy some of the systems on Bitcoin. What would it take for you to jump ships and switch teams?
0:00**Michael:** That's a good question. I never thought about that. It never felt like a real question to consider.
0:00**Vlad:** I'm the devil on your shoulder right now.
0:00**Michael:** Yeah, I understand. You can also ask the same about Ethereum. There's a lot about Zcash. There's a lot of researchers and people I appreciate. And there's actually more of these people in Ethereum and Zcash than in Bitcoin. I kind of don't even know. I'm third generation. I call myself a third generation. I think Yonatan is second generation and we need him to step up and lead more with his vision. I call myself third generation. I don't carry the weight of seeing the first generation. I don't necessarily come with the same cypherpunk... punkiness.
0:00**Vlad:** But you're not answering your question.
0:00**Michael:** I think Bitcoin needs to be... I would like, I can imagine a bitcoin where I would want to contribute. Maybe a long shot. There can be such a world if that bitcoin was more, less rigid, like much more flexible. My biggest problem with bitcoin-related research is solving problems that shouldn't be solved.
0:00**Vlad:** Can you show that again to the camera?
0:00**Michael:** Solving problems that shouldn't be solved. Like, okay, Bitcoin has these fixed rules and a fixed script engine and no one can touch it. So let's find the most clever way. And it's probably amazingly clever. But I never really read the details because I don't like the idea of rules of nature which aren't real and are kind of forced, and you have to somehow be so clever to work around them.
0:00Someone asked me on X the other day. He took a screenshot of Greg Maxwell on Bitcointalk saying he has an amusingly bad idea of covenants, of essentially presenting the Schnorr trick, which we can speak about. And he asked, how did you make covenants a good idea? And I answered basically what I just illustrated with my hand: just doing on L1 the straightforward things that make it easier to do things above it.
0:00So I think we did covenants the right way. And by the right way I mostly mean something known as KIP-20, which streamlined and made simple the whole idea of lineage and authenticity of external state, so to speak.
0:00**Vlad:** Should I get into that? Yeah, beautiful. Let me plug a couple of sponsors and then we can unpack a lot of the stuff that you said. Starting with the covenants, and also the names that you think, you said you respect some people in Zcash and Ethereum and I'm curious exactly who. And we also need to get into more specifics about your work.
0:00But first of all, download Cake Wallet. It's a wallet that supports Bitcoin and Zcash and Monero and Ethereum and Solana and whatever. The best part about it is that you actually get to use the coins. You can swap them for other coins if you want, or you can spend them directly using Cake Pay, which is a service that lets you find gift cards and all sorts of goodies that you can purchase directly with your coins. It's very easy to use, free to download. Check it out, cakewallet.com. I think it has like two million users by now, or at least two million downloads, because we don't know exactly what that means in terms of actual users.
0:00There is also Orange Rock, which is this place where you can trade perpetuals mostly, but you can trade without limits, without KYC, without any sort of roadblocks. Because with modern exchanges you have to sign up with your ID and it turns out to be a terrible idea usually, because these exchanges get hacked more frequently in the last few months. And it's much better to not give any kind of data, and you can trade on Orange Rock without any kind of restrictions. You can also use leverage, but I don't recommend it. It's like gambling. Orange Rock is brought to you by the people behind bitcoin.com, so it's not like they're some obscure developers who emerged and decided, hey Vlad, why don't you promote our stuff? It's a real team, pretty solid. Check it out, orangerock.com.
0:00And the other one that I will plug here is Braiins Hashpower, the place where you can rent hash rate if you want to mine. Let's say that you want to mine Bitcoin or you want to mine the drivechains hard fork which is called ecash right now. You go to ecash.com, you spend BTC and you get the hash rate that you want according to the schedule that you expect. It's actually in lots of situations much cheaper than buying your own hardware and running it at home, because especially in Europe the prices for electricity have gone over the roof and it makes no sense at this point. So you can use it for this, or you can try to 51% attack BSV. I'm not saying you should. I'm saying you can. So don't sue me, Craig. But yeah, Braiins Hashpower, check it out.
0:00And now Michael, let's unpack a bit. Let's talk about first of all covenants. What is a covenant and why is that useful? Because you said you integrated that into Kaspa.
0:00**Michael:** Yeah, I'd say we're kind of discovering or seeing practically how many, that it's surprisingly expressive. We can do probably any application you imagine on Kaspa L1, one way or another. If we started with the question of how far can we reach with applications above the UTXO model, the answer is far.
0:00**Vlad:** So what is a covenant? That's the first question. I always enjoy when people define what they are because when I interview developers, they talk about use cases for which they think they're interesting, but the definition is not standardized, at least not yet.
0:00**Michael:** Okay, for me the definition is that authority over the UTXO isn't unlimited. Meaning if you own a signature, that doesn't make you a full authority over the value in that UTXO. There are still more rules you have to follow. And in particular those rules typically require the continuation, the output of the transaction, to obey something. It can be probably anything, but essentially the core is: you enter a covenant in the sense that you don't have full freedom over that Bitcoin or Kaspa value anymore. You actually entered another set of rules beyond your public key authorization, which still probably plays a role in the covenant, not always, but yeah, that's my definition.
0:00**Vlad:** How is that? Well, it's pretty good because I've heard people define it, for example, in terms of creating rules that you cannot break, or in terms of creating vaults or building all sorts of applications. In the context of Bitcoin, I think covenants are like this cure for the problems that were brought upon us by the Lightning Network. They say with covenants we can actually make Lightning work. I'm not 100% sold on the idea because it leads to a lot of situations which don't have unilateral exit, and that's another thing. But mostly in Bitcoin they're used for L2s. In the context of Kaspa, why would you say that covenants are useful?
0:00**Michael:** Yeah. We kind of simplified. We added one signature besides obvious introspection. So we didn't implement covenants as a workaround. We implemented it straightforwardly. You have introspection opcodes that allow you to inspect each input, each output. You can just directly inspect almost all the transaction around you.
0:00I want to explain. We also implemented something we called the covenant ID, which is an L1-tracked entity that kind of travels through UTXOs and proves they come from a common source, so to speak.
0:00A covenant can be used for a vault, for instance. I kind of differentiate two main families of covenants: covenants that are about the Kaspa value they protect, and covenants that represent some other scarce resource. If it's about the Kaspa they protect or the Bitcoin they protect, there's no notion of cloning. Someone else can clone my script over his own Bitcoin, so be it, don't mind. I'm locking this Bitcoin in a vault, or this Kaspa, and I'm saying I'm placing conditions. My heir might be able to access it if it wasn't touched for one year, and even then he might have restrictions of how much he can spend per month. Anyone, the rules are about the main Bitcoin value or the main Kaspa value, which already has preservation of value. No one can create that from thin air.
0:00So I actually wrapped this Kaspa with these rules and I want them to carry the rules forward. That's one type of covenant.
0:00The other type of covenant, which would also be used for L2s typically, is a type where you need some kind of lineage. If I encode in my UTXO some additional state, some bytes saying something, just putting some bytes that say something doesn't give them significance, because anyone can copy those bytes. In order to create applications that have additional state other than the Kaspa they hold, like for instance a native token, some ability to represent over Kaspa another asset, that requires more. If it's only about bytes written in a UTXO, that doesn't give you enough information. You have to know that's originated from some genesis, so to speak, that authenticated and makes sure that anything that enters this covenant is actually part of the lineage of how the state evolved.
0:00Is that clear, more or less? If I create a covenant, let's take the L2 example. Let's think about a UTXO holding a hash of the state of the L2.
0:00**Vlad:** Yeah, go on. How is that typically updated?
0:00**Michael:** So you're probably referring to all kinds of demos and ZK-related stuff where you provide a ZK proof and you transition the state. The L2 state is represented on L1 with a hash, and the transition function is moving from one hash to another hash. You're saying the L2 is updated and I'm proving that your transition is valid with a ZK proof, right?
0:00Now, what makes that UTXO represent the L2 state? Why can't someone else create a UTXO that looks the same way with his own state hash and claim that he's the UTXO representing that L2?
0:00You would answer me that you'd be tracking one specific lineage of UTXOs and anything else isn't related, right? If someone deployed an L2 ZK system over such a covenant, it would have a genesis transaction. "Here I created the UTXO for the first time." And now this UTXO has rules which are recursive. It enforces that the output, there's going to be an output that has the same kind of script exactly. So I'm encoding all the future, the commitment to the future rules in this single UTXO. And then when you would track the system as it evolves, you would say, okay, I'm tracking this specific chain of UTXOs and this is the chain of UTXOs that represents the known L2, has some name.
0:00And now if I come as a new user and I'm asking you to prove to me what's the state of your L2, I know you have a nice ZK design. I know you have a ZK circuit. You're proving on L1. Now show me on L1 the UTXO. If you just show me the UTXO, I'll tell you, who says this is a UTXO that represents your L2? Maybe someone, maybe you just faked it.
0:00One answer would be: okay, I'll show you all the chain of UTXOs from the genesis, and I'll link that genesis somehow to the known L2 for some public information. I'm going to tell you this is the main system. My point is you have to prove lineage for essentially that L2 represents a state that isn't Bitcoin. That state can hold, I don't know, a smart contract system with various assets and whatever, it can hold other stuff which aren't Bitcoin or aren't Kaspa. And that state is external to the L1. It's just a bunch of bytes. What do they represent? You have to prove some kind of lineage. There has to be social consensus about a specific lineage of UTXO evolution.
0:00The way this can be done, a bit hacky, is by some kind of recursive proof. Have you ever heard about that?
0:00**Vlad:** I mean, they're part of the Bitcoin conversation. We keep hearing about CTV and then CTV not being good enough because we need a fully recursive covenant soft fork, and then we've had this conversation over the last three years, I think, and we never got anywhere. But I think it would be useful if you defined all the terms: what is a recursive covenant and what is a non-recursive one.
0:00**Michael:** So we established the need. If the covenant controls some external state, then we established the need for some kind of lineage or recursive proof. The answer of saying "okay, I'll just show you the full chain up to this genesis" isn't good enough. So what you do is you allow the script itself to require, as part of its execution, to show you that when you execute a UTXO, you have access to the transaction that created it, the transaction ID that created that UTXO, right? It's part of the key, like the key for the UTXO: the transaction ID that created it and the output index. So you have that information. And now within the script you ask to get the pre-image of that transaction, and you make sure that transaction had an input script that forced the same rule.
0:00That's kind of an inductive step. Think about it as an induction proof. I can be spent only if I prove to you that a previous transaction was spent with the same rules, and that transaction was able to be spent because it proved the same, until we get to the base case. So I either show you my previous input is actually some hardcoded genesis, or I prove by induction that my predecessor came from the same line of lineage.
0:00This proving this recursion is what allows you to prove. Essentially anyone can create a fake UTXO like that. It's easy, I can write bytes, right? But it cannot be spent. If you fake this and you create such a UTXO from a different lineage, you put in the same script, and that script is looking for a proof that the previous transaction was also spent the same way. And because you just forged it, you actually can fake that UTXO but you cannot spend it. So it kind of becomes useless. Essentially by making it unspendable, we're showing that the UTXO is the authentic one if it can be spent. That's basically how this problem is proposed to be solved in the Bitcoin servers. I don't know the names well enough. I don't know what CTV exactly means.
0:00**Vlad:** Check Template Verify. It's BIP 119 by Jeremy Rubin. He has been pushing this for like five years now, I think.
0:00**Michael:** Still actively?
0:00**Vlad:** Still actively, you mean Jeremy pushing it? Yeah. He sort of moved on and he's trying to do other things now. He's working on something called Char, which is burning UTXOs. That's why it comes from Char, Charmander, Charmeleon, and Charizard. They're all Pokemon that burn.
0:00**Michael:** Okay. I don't know anything about that. But yeah, it's a design. It's clever, very smart. In the transaction, if you look at the script, it's a weird oscillation. You're kind of looking at the transaction hash of a previous transaction hash, and then the signature needs to provide the pre-image of that transaction. You hash it to verify it's the exact transaction. Then you look, you can't even look at the input because the input is not part of the previous transaction. So you have to get the previous transaction. You have to look at the grandparent and show the same. Only then you can actually see the script public key that was the previously spent input, not this one, and you can verify it's the same, and then you kind of see the recursion. This whole operation is not so cheap. It requires bringing the full pre-image of the previous transaction essentially to the signature.
0:00And to me, as you'll see if we make enough progress today about this, you'll see that I think it would be a big limit. The current design we have allows composability between covenants from different lineages. And with that design I imagine a nightmare. If you had to do recursive proofs and still run multiple such proofs in one transaction, it would be very hard. I'll present to you an idea that simplifies the whole thing. It required obviously a change to L1, but it's a very simple change to L1, very local to that transaction.
0:00That's essentially KIP-20. It's a simple idea. Let's attach a covenant ID, a 32-byte hash, to each UTXO, optionally. Most UTXOs don't have it. And if I have this covenant ID, an output can claim to have this covenant ID only if it has an authorizing input, only if it has a specific input with that covenant ID already. So there's a genesis way to create a new covenant ID, but other than that, each continuation needs to show the input. It's a very local operation.
0:00Transaction verification step: you look at this transaction. If any outputs claim to have a covenant ID, you make sure there's a corresponding authorizing input with the same covenant ID. You can't enter the covenant without being authorized by a previous member of the covenant. Does that make sense?
0:00**Vlad:** Yeah, it does. So that removes the need for recursive lineage. Essentially you show a link. I no longer need to look at whether this input came from, because it already has this stamp of covenant ID. So I know it came from the lineage. All that is left is for the person writing the covenant to make sure that the input really wants to authorize these outputs and give them the covenant ID. The input, in the input script, the obligation of the script side is to make sure that continuation makes sense.
0:00**Michael:** If the input doesn't verify who took that stamp and is saying "okay, these are good," then there's a missing link. Consensus is doing one part of the story by just making sure there's a link, and then the script writer needs to verify the forward link. The input has to verify that the outputs carrying the same covenant ID are carrying it justifiably.
0:00So this removes the whole need for recursive proofs, essentially. And that's how Argent, which is... we now have, in this amazing AI era, I'd say two compilers over the script engine, a few months old, and we already have two programming languages which are layers over Kaspa's script engine. The lower one is called Silver Script and it's a single-contract compiler, I say. It allows you to write code which would be the single contract backing a specific UTXO and telling you how to spend it. And Argent is a compiler above Silver Script, which is a multi-contract, multi-app compiler that allows you to express complicated relations between contracts and other apps, which are other covenants, and interact between them and even perform atomic operations cross-covenants. You can design a system where there are two operations happening with two covenants and they both inspect each other and kind of decide together they make progress.
0:00**Vlad:** That sounds really good. I just want to remind people that this year has been horrible. We have seen people getting kidnapped and losing their coins in France. We have seen hardware wallet users getting hacked and losing all of their savings. And we have seen exchanges getting hacked. In the beginning it was smaller ones, but I think like two or three weeks ago we had Binance that got hacked.
0:00So covenants are actually a very smart way, especially if they enable vaults, to prevent users from losing everything. You can have conditions about what can be spent and in how many transactions or in how many blocks. You can actually have this very basic rule-setting setup where you can't lose everything. Of course, it means you can't spend everything yourself if you ever want to sell everything. But if that's what you're planning to save long-term, you want to put that into a vault and you want to set the best conditions for yourself to actually be secure over the long term, to prevent any wrench attack or whatever.
0:00And in Bitcoin, people usually say, well, just use a multisig and put two keys somewhere else and only keep one in your home. I'm not sure if it's the best setup. It's definitely one that works, and one that companies like BitGo built their entire business system around. But if we had a way to program the coins so that you only spend what you want, that's actually useful. And that's actually what I think people in the future will be much more aware about: this kind of attack and also the quantum attack. They will choose their savings and the assets in which they save according to these criteria, and also these assets that give them good ways to have vaults so they can't lose everything all at once, and also quantum resistance. I think these assets will be repriced and they'll be much more valued in the future.
0:00**Michael:** Yeah, I agree.
0:00**Vlad:** I appreciate this academic-level discussion about what a covenant is, but at the end of the day they have to be useful.
0:00**Michael:** Yeah. So one major thing: a group of developers and researchers in Kaspa recently released what we call a KCC. It's like an RFC instance, like a request for comments. So it's a call for convention. And KRC-20 is obviously a standout for native assets using covenants on Kaspa. The standardization of that was a major milestone and a lot of people were waiting. Now you can build an order book application on Kaspa as a DEX. You essentially have covenants interacting. You have covenants representing the assets, you have covenants representing the pool, and they can actually build this kind of real-time system. The expressiveness is pretty amazing, I think.
0:00I definitely agree about the tools for protecting the asset itself, like having smart rules, is very significant as well.
0:00**Vlad:** I think especially for exchanges, so they can't lose everything. They keep some funds in hot wallets obviously, because some of them get deposited and withdrawn in a very short amount of time. They have to adjust to this. And the funds that they keep in hot wallets is what is vulnerable. If they were able to have covenants for this, it would be super useful.
0:00**Michael:** I think it's like you would say, okay, on smart contracts you can do the same supposedly. So the answer would be that I think covenants are more customizable, in the sense of you can control your exact rules. A smart contract system, typically the smart contract controlling the system, it's hard to have each user control its assets in its own way. There are many ways to get exploited. The Binance example is like you said, their hot wallet is so big that hitting it, they probably need that kind of liquidity available. They can't easily choose to not have that kind of liquidity available. It would mean some compromise on their...
0:00**Vlad:** Yeah, sorry. I was looking at the comments. It's kind of amazing that we received some actually good questions from people. So thank you very much for your input on that. I got to plug two more sponsors and then we can move on with the rest of the interview.
0:00The first one is ecash.com, which launched the hard fork of Bitcoin. You can claim your coins for free today. If you hold any BTC in a UTXO, you can split your coins so that you also have some ecash. Don't do that with your largest UTXO first. Test it with like $10 or something so you understand how it works. And then if you like the system, you can test it with the rest of your coins too. Once again, it's free money that you can get from this quote-unquote airdrop because there's been a hard fork.
0:00And if you're wondering where you can sell it or where you can buy more of it if you're bullish on it, the answer is SideShift, which is this place where you can swap between coins. You can swap the ecash that you just claimed for Bitcoin, or you can swap some Bitcoin for more ecash if you're into that. I think they also accept some stablecoins. But SideShift has a lot of pairs and a lot of use cases. Usually I tell people to use it for swapping stablecoins for Bitcoin or Ethereum or Litecoin or anything that cannot be taken away from you, because stability is one issue, but being able to hold the asset over the long term without being concerned that Tether or Circle is going to freeze it, that's something else. You never know when you receive a payment where it comes from. It might come from an exchange hack, might come from some organization that later gets labeled as whatever, and then you'll end up having your funds frozen for reasons which are not your fault. So swap your stablecoins for that on SideShift.
0:00And now we're done with the ads. I actually want to ask you something from the chat which I thought was very smart. You mentioned that you're working with two AI agents. One of them is called Silver something and the other one is Argent.
0:00**Michael:** They're not AI agents. They're programming compilers.
0:00**Vlad:** Compilers, okay. But the question is about AI agents. Someone wants to know to which extent is core development being done with AI.
0:00**Michael:** It's an individual question. Each developer has his own, but obviously it's a big, I use AI all the time. When I code in the node software, it's different. I do it on a very granular level. I inspect each commit. So it depends on what kind of system. But for research, I'm using AI now significantly. It's amazing. I don't know if anyone, you can hear about all these mathematical results, right? But if you're actively trying to prove something or doing some active research and you try it nowadays with the good models, it's astonishing. It's a new world.
0:00**Vlad:** So which models are good for your work?
0:00**Michael:** I use OpenAI. And for deep research I use Astra. I spend my tokens. I've noticed that it gets really good results in some areas of research, also some areas of applied research, like when you have an abstract algorithm you want to design more specifically, or the step before implementation, which is something we do a lot with DAG Knight now. Spec'ing an actual implementation with Astra is amazing. The ability to analyze complexity, to not make mistakes, literally no mistakes. It has its own engine of running audits with sub-agents even if you don't instruct it to do it. And it kind of understands known problems.
0:00Talking about applied research, it's more about these problems that were analyzed from many angles already and many implementations in the world already kind of touched on them, and there's a lot of analysis about them. So when you see some kind of problem like that, it feels like the model understands it in a very unique way. It's very fluent in the options. You can ask for that trade-off and it can easily find a way to improve that.
0:00I don't know how to explain it. When doing harder research, I still feel like the responsible adult, and my role in steering is very significant. It's a very eager student, very fast, very clever, much faster than me because it can do things much faster. It can still make these, I can still tell it, okay, but why are you even looking at that subclass? There's a lot of, I'm important in the conversation still. But with these applied problems, they just solve things brilliantly. And you can also ask things. I don't know, you're researching something and you have something you say, okay, that would be interesting to check, but checking it would take two days and I just can't do it. I don't know if the result is going to be worth the two days. So you never get it. Now it just performs a check in five minutes and you get a result and you can do things, you go through paths you would never go through because of time.
0:00**Vlad:** I hope that passed through somehow. It's interesting because people usually have to choose, especially for the high-level stuff, they have a choice of Anthropic versus OpenAI. But I've also heard people running local models, open source stuff, just because for some simpler operations they're good enough, and also sometimes it's good to have these open source models verify the non-open-source ones.
0:00**Michael:** Yeah.
0:00**Vlad:** I think especially this year has been horrible, or beneficial depending on which side you're on, but for security it's been a nightmare. We had zero-day exploits. I had to redo my website from scratch because it was using PHP. So I had to ditch that and switch to basic HTML, static HTML. And I saw something today about JavaScript being vulnerable so that even if you, you can get access to someone's admin key just by injecting something into the code. I didn't have time to look into how it works, but it's very bad. Linux has had a lot of issues that they keep patching. A lot of the open-source world had to adjust. Previously if they were not using AI at all for coding, now they have to at least verify every line of code with AI to be sure that there's nothing wrong there. But some teams are choosing to improve everything using AI so that it's more resilient to this type of attack.
0:00**Michael:** Yeah. Kind of the hardening phase of the worldwide software.
0:00**Vlad:** We used to say that open source is safe because the code is out there, and just because nobody hacked it in like 10 years it means that it's probably very good.
0:00**Michael:** The problem was that there was no incentive for anyone to look at it, or at least not a large enough reward for breaking it. So hackers just forgot about it. And then the cost of looking at software became so small with AI that anyone could just point a link to these LLMs and say, just find whatever issue, and they would find it in a day and then exploit it.
0:00**Vlad:** Yeah, it's been a nightmare of a year. I mean, in terms of being on a core team of a consensus protocol, a consensus network, you have to be on the alert. You have to be defensive. You have to scan before the attacker. You have to use your deeper knowledge combined with AI to defend better. You have some advantage because you know the system better, but you have to utilize AI.
0:00**Michael:** Yeah, been a big weight, can say that.
0:00**Vlad:** When you know the system better, usually not usually but sometimes you have a subjective perspective on it, and you need a system that is completely non-emotional and not attached in any way to every line of code that you wrote there.
0:00**Michael:** Yeah, that doesn't care essentially about what's happening in your code, that tells you this is wrong, you should redo this, this is vulnerable or whatever.
0:00**Vlad:** I think Zcash had one of these moments when they discovered a bug in their shielded pool that could generate inflation. They patched it internally and they were able to find it before any attacker. Was there any such incident with Kaspa? Did you guys find anything that you needed to patch?
0:00**Michael:** Not a consensus bug, only node crashes and stuff like that.
0:00**Vlad:** Would you say that overall AI has helped Kaspa grow over the last year?
0:00**Michael:** Yeah, of course. In many aspects, both the development was accelerated, like Tokata, the recent hard fork, was definitely accelerated due to AI. And the compilers I mentioned before were written using a lot of AI. And also on the defense side, the node is definitely more resilient now. No doubt.
0:00**Vlad:** Do you think it will get to a point where development is kind of useless and it's just the research side that is relevant, or do you believe that you will always have humans typing the code on their keyboard for Kaspa?
0:00**Michael:** Already now most developers don't write every line of code. They might use AI too. So the question is whether we would need to also know the code, like would we reach a stage where humans don't need to actually verify each line of code but we can actually trust the AI to faithfully implement what we wanted and securely? That's the question. Not about specifically typing a line of code, which already happens less today. There's no programmer today that can afford not using AI to speed up themselves at least a little bit, because it's not cost-effective.
0:00Whether or not, like assembly today, no one knows assembly. We rely on the compiler to generate correct assembly, right? Whether or not I think we're probably going to reach that stage one way or another, probably through formal verification. We would need to trust, the AI would need to formally verify code in order for us to not need to look at each and every line.
0:00I'm fascinated. Some people are scared. I understand the fear. I understand being cautious. But I'm fascinated. If I can take a research question and prove or disprove a hypothesis I had by just asking the right questions in a day, that's very interesting. The sea of knowledge is more approachable, still infinitely large, but you can do a bit more than playing with rocks on the sand, so to speak.
0:00**Vlad:** I think to a lot of creative people, there are two camps. There are the ones who are like doomers right now, and they say, well, I can't do this right now because the AI is more efficient, so I'm out of work. But then there are people like myself who are like, oh my god, I've waited for something like this my whole life. I can do this and I can do that. I can build apps. I can basically turn any stupid idea into a product in a day and then understand why it makes no sense.
0:00**Michael:** But that's the other issue. You also can create software for yourself. So you don't need to purchase software or wait for someone else to create something for you. If you want to make a new wallet or a new website, I think AI is already good enough for these use cases. If you want to build something very sensitive like a banking infrastructure or a consensus model that you don't understand, I think it's safe for you to not understand the code if you're doing something with low impact, just for yourself.
0:00**Vlad:** Yeah. But if you want to sell the software to someone else or to push others into using it, you actually need people to look at what's happening there. I can see situations in the future where governments push these AI companies to essentially attack some teams in their code with updates or whatever. It doesn't have to be specifically targeted at one repository, but it can be at the same time. You don't know. They're not open source. You don't know what they're doing. You're just going to tap into Claude 6 or GPT-7 or whatever. Just like Kimi K3 was very good at finding attack vectors in any system, and now Kimi K3.1 they sort of removed that feature and made it more efficient for other use cases, which made a lot of people not care about it anymore because that was the main point of Kimi. But that one is open source.
0:00**Michael:** That's interesting. With OpenAI and Anthropic, you can basically assume at any point that maybe they don't like something, and if you make the model...
0:00**Vlad:** Yeah, that part is scary. The only thing you can rely on is competition. If both of these companies shut down, stop providing their services tomorrow, many people wouldn't know even how to go back. How do you go back to coding manually? Obviously there are the open-source models, but let's assume there's a significant gap still, and you're very used to these closed company models. The only thing you can say is, okay, they're not going to shut it down because there's competition. There's enough strong models in the world. But yeah, I agree. I don't have enough time to explore too many models. I have this thing working well, amazingly well, so I'm just sticking with one thing I already used.
0:00But I agree with your point. It's cool, at least if you're from the US, to complain about the open-source models destroying the Anthropic IPO or whatever, or diminishing their chances to succeed in their goals. But they're actually useful from a developer's point of view. You can do more honest verification because you can see the code of the model, and you can use another model to verify the code of the other open-source model. Even if you know nothing about coding, you can do more verification with these open-source ones. Of course they're not as good. I think right now what I use locally with Qwen 3.8 is at the level of GPT-4, even slower on my machine.
0:00**Michael:** Oh, wow.
0:00**Vlad:** So it's ancient at this point. But for some applications it's really not bad. If you ask it to do some basic, not coding stuff, just thinking or answering questions or searching for stuff online, it's just good enough. I think by this time next year maybe we'll have GPT-4, not four, five or six-level open-source models that you can run on your home computer. It's beautiful. I hope we get there.
0:00To people listening, and maybe there is a significant portion of the Kaspa community that's tuning into this interview: what would you recommend them to build? Because right now the barrier of entry has been lowered to the point where anyone, someone who has a weekend, can contribute to creating something. What are some applications or wallets or whatever that you think are good ideas and can actually lead into successful products?
0:00**Michael:** That's a great question. Since the hard work that enabled all this, I saw my main role as being an enabler. I realized that Silver Script was a bit too low level, and if you try to create something that is a bit more advanced, you have to do a lot of low-level understanding and coding. So it actually started by me implementing chess as a covenant.
0:00**Vlad:** Okay. Can I ask you: what would you think would be the main obstacle if you try to implement chess in a covenant? What would be the main problem? Think like we both lock Kaspa over a game and we play on-chain, and the winner takes... chain verifies each move.
0:00**Michael:** So the obvious, the rules of chess are not so small. The rules themselves are not so small. You also have to scan the board.
0:00**Vlad:** Yeah, you need an oracle or something, right, to tell you exactly what's happening.
0:00**Michael:** So if you try to implement it in script, you have to unroll many loops, maybe even to scan the board and to verify that the king is not under threat. I managed to implement chess in full through, I don't know if we should go into it, but through a design that splits the rules into multiple contracts.
0:00**Vlad:** No, it sounds beautiful. Go on.
0:00**Michael:** So I call it, there are a few main ingredients. The first one is a multiplex design. It's kind of, okay, on Bitcoin maybe it would be a bit easier because you have Taproot. So you have the idea of a contract under a Merkle branch, right? You can only reveal some of the code when you want to execute some. There's the idea of masked Merkle abstract syntax tree, whatever. But we don't actually have that feature in Kaspa.
0:00So what I did is, okay, first of all, the script was too large. So I had to think about, okay, I'm doing a move, I'm moving, I can move a pawn, I can move a knight, speaking of knights, I can move a castle, whatever. Each one of them has a different validation function: is this move legal? The check is different. So what I did is I did a contract which is a multiplex. You actually just choose your move, you choose what next contract to delegate the game to.
0:00So it's my turn. My turn is coded as part of the state. And then I say I want to move a knight. So the first transaction actually moves responsibility from the multiplexer to the knight contract. And then, okay, if I would be on Bitcoin, I'd have to wait 10 minutes for that and then do the next move. Which is not, I'm lying here, but whatever. I'm just saying that in a high-frequency system, I don't mind having an operation be done in two transactions. So the second move is actually going to execute the knight contract, and it's going to verify that the slot I'm moving has a knight and I'm moving it according to the rules. And then the contract is going to switch turns and go back to the multiplexer.
0:00So there's also, first of all, I was able to split. Now I call this a multi-role type system because essentially it's a covenant where you don't necessarily pass control to the same script, but you pass control to some script out of a predetermined set of scripts. So it's a covenant with multiple possible scripts, and they each form a closed system. They only link to each other. So it still can be deployed once and it already encodes the full possibility of transitioning, but the control can move between all kinds of scripts. That's what I mean when I say multiplexer. The multiplexer moves control to the script representing the knight. That's one major design.
0:00And then Argent generalized that into, Argent generalized this pattern of multiplexing and managing many contracts under the same app or the same covenant at the language level. So you simply say to Argent, I want to become some other entity within this closed system.
0:00I'm just going to close up this chess example with the chess. The other part was I never check if the king is in check actually, because that's a scan and that would require a big loop which is unrolled in script. So it becomes a very big bytecode segment. So what I do is, if your king is, let me remember for a second, kind of switched, flipped the game. Essentially the game on-chain allows you to move illegally as long as your opponent can prove you did an illegal thing, and then you get ruled out.
0:00So if you don't, if you get a check and you don't respond, so I can by design eat your king at the next move, right? If it was a checkmate, anything you do, I can still eat your king, right? I can still knock them.
0:00**Vlad:** It's not really fair, right? Because in a game of chess, if the opponent doesn't move and he's in a check situation, it doesn't mean that you won, right?
0:00**Michael:** Right. So if he has an escape, he'll just move to that escape and then I can do nothing. But if it's actually a checkmate, he either will surrender or he'll ignore the check. But then I can prove in the next move that I can knock down his king through a specific move. So that way I avoid the scan of all possibilities of the king on-chain. I just allow the opponent to prove he actually did the checkmate. And the same pattern exists when you have to prove other stuff, like a bad castling, or a check, a check-style stalemate. It works on the same kind of system. Instead of verifying on-chain the whole possibilities, I kind of flipped the idea and let the opponent prove otherwise. But I managed to, it's kind of, think about it as instead of an execution engine, more of a verification engine. If you know the NP analogy, instead of searching for something, let's validate it after the fact.
0:00So I was able to design, I think, the full chess game using scripts, and it's not even huge scripts. I never deployed it, but it's a demo repo that's public. And that led me to develop a compiler essentially. Like I said, I was seeing myself as an enabler first. I don't claim to have the creativity to know exactly what people will develop over this. I'm trying to do my best role as someone who understands the underlying system and capabilities and knows how to code and knows how to design a programming language. And I try to, and that's significant. Apps are now popping up, built with this, and it's amazing to see each and every example.
0:00**Vlad:** It's quite interesting because we started from the question of what can the community build, and you presented the example of chess on-chain.
0:00**Michael:** Yeah. At the time I was thinking about chess because I was trying to illustrate the uniqueness. I was saying you can play blitz chess on Kaspa, and that's it. You wouldn't be able to do that on most other chains.
0:00**Vlad:** What's the difference between chess and blitz chess?
0:00**Michael:** Blitz chess would be very fast, as you have to respond in 5 seconds, let's say 10 seconds. So if the chain doesn't have the high frequency, it cannot actually measure 10 seconds if there's a block every minute. So below a minute, the chain has no idea how to measure time. A high-frequency network can actually tick very fast, so it has a fast clock. And that clock can be used to, that's why you can play a very fast chess game.
0:00Also interesting, a colleague of mine hasn't published it yet, so I'm not going to, he implemented just for illustrating the idea, a kind of ping-pong on-chain. You actually move whatever the paddles.
0:00**Vlad:** Like the classic game.
0:00**Michael:** Yeah, sides of the screen you have paddles and then you have a ball.
0:00**Vlad:** Yeah.
0:00**Michael:** So you move the paddles via transactions on-chain. It's not too fast, you know, but it's very responsive relatively. It's illustrating the point of the L1 being responsive and high frequency. In general, I see the high frequency as, I see all these covenants or contracts as like these small programs that can interact with each other pretty fast. And I don't know yet how exactly it will evolve, but to me it feels like a new kind of playground, that ability to code small units of code that interact with each other.
0:00Already starting to see it. Someone is deploying an order book DEX. Someone is designing a prediction market over Kaspa covenants. There's a KUSD, a bridge to USDT. There's a design for a Kaspa Dollar which would be a decentralized stablecoin. A lot going on. People building bridges from Ethereum and between Ethereum and Kaspa, because we now have ZK opcodes so you can verify Ethereum finality and you can prove Kaspa's proof of work on Ethereum. So you can have some kind of bridge. Bridges are a big question to me.
0:00That's related to your question of what people can build. I'm not so, I've been trying to think about what bridge would bring a lot of value to Kaspa in terms of connecting it to the wider ecosystem and providing some kind of dual value to both sides that would increase Kaspa awareness and bring more liquidity in. I don't have a definite answer, but it's the kind of stuff I'm thinking about. And anyone asking me, I'm willing to go on a Zoom, I'm willing to bring my input. I'm not sure I can bring the ideas. I'm not sure I can bring ideas like stag hunt, for instance.
0:00**Vlad:** You just gave me a crazy idea. It would be very fun, like these RPGs where you have to make decisions, and let's say Pokemon.
0:00**Michael:** What's an RPG?
0:00**Vlad:** You don't know what's an RPG? There's a kind of a, it's kind of a weapon. That's what I know.
0:00**Michael:** [laughter] We can clip this.
0:00**Vlad:** It's not a rocket launcher. It's a role-playing game. Have you ever played Final Fantasy or Dragon Quest?
0:00**Michael:** I'm not a gamer.
0:00**Vlad:** Okay, so you don't know. But you're a warrior because you know the RPG, the rocket.
0:00**Michael:** Yeah. QED. [laughter]
0:00**Vlad:** So you never played Pokemon? I guess simple Pokemon.
0:00**Michael:** Yes. Probably the trading card game or the actual game?
0:00**Vlad:** No, I played maybe simple computer games, not like interactive social, like I played as a kid, I don't know, classic computer games from 30 years ago, but I don't know modern games.
0:00**Vlad:** Okay. I was thinking of something simple too, like an RPG once again where you get to choose where you move on the screen: up, down, left, right, and then you have to go on your journey to go to the next town and get into the battle or whatever. And I can imagine a situation where you have this RPG and people who want to control the character, they have to make on-chain transactions, and it's almost like a voting system. So the next move is determined by, I'm not sure if it's going to be the highest amount of votes or the largest amounts being sent for that purpose.
0:00**Michael:** So you control your character, or does it, what do you mean, like you have a specific character you control?
0:00**Vlad:** It's a social or financial coordination game where everyone participating has to choose the next move, and to choose the next move they have to send an on-chain transaction.
0:00**Michael:** Mhm.
0:00**Vlad:** And depending on either the number of votes or the weight of votes, which is the highest amount of money being sent, the next move is made, and people have to react to that. And it can be very small amounts, like sub-cents or whatever, just to encourage people to participate. But only something which has very high speed and enables high frequency.
0:00**Michael:** It's a space where there's multiple entities moving in parallel, independently. No, it's just one main character?
0:00**Vlad:** Just one main character, and they have to choose where that main character goes. And also decisions when they get into a battle, you have a menu and you can attack with your sword, you can cast a...
0:00**Michael:** No, what do you win? Like when do the moves evaluate to some reward, for instance?
0:00**Vlad:** Well, the goal is to finish the game, right? You got to get to the last level.
0:00**Michael:** And then what happens in terms of what's the incentive of people to play good?
0:00**Vlad:** That's a good question. I don't know. Just to see the beautiful ending, save the princess or something.
0:00**Michael:** Okay.
0:00**Vlad:** You can also get the bad ending or just get defeated. And I also had these kinds of experimental thinkings of how to, I was mostly focused about, okay, let's find a game which illustrates enough complexity in terms of the contracts and illustrates all kinds of capabilities like the real-timeness and stuff like that. I was actually thinking it was also part of the motivation behind that programming language called Argent. I was imagining a space, kind of a two-dimensional space, and entities moving independently within it. So evaluation of the movement is always local versus the actual physical reality around it. And I think we can do interesting stuff. But I haven't had time to put more time into it, so to speak.
0:00**Vlad:** For some reason you became blurry. Yeah, I can fix that. Okay, it's much better now.
0:00The point is that for this kind of game, you only have six inputs or maybe even five, which is up, down, left, right, and then A or B button. So the next move is dictated by a vote of one of these. But it's much more complex than that because in battle, for example, you have to choose the kind of attack. So it can be like a spell or it can be a health potion or something.
0:00**Michael:** So it also depends how compressed can you make the state, like the reality. How complex is it? The state, I mean the covenant or the contract has to carry the state of the reality, right? There's something happening, that character is interacting with something.
0:00**Vlad:** So that physical reality, how compressible is it?
0:00**Michael:** Now, if the scene is simple and there's only a few objects and I can just encode their locations into the contract, and that model is like the reality, right? But if the scene is more complex, has many moving parts, many another entity which is moving and you're fighting, so you have to know where it is, you have to encode that into a state of the contract essentially. And I'm asking how complex it is essentially.
0:00**Vlad:** Well, I think it's much more complex than chess because you have level design and all of these elements. But I don't think it has to run on the blockchain. You don't have to run this on a blockchain. It can also be something that you run in a virtual machine. And what matters is only how people vote for the next move. It doesn't have much to do with putting this on-chain, which I guess is a much better challenge for yourself and something that you think about. But for something that you put on-chain, you can create something simpler like tic-tac-toe. I think you can do surprisingly complex stuff, but it depends on what value you get from putting the whole logic on-chain. You have to define if it's about illustrating some capability, that's one thing. If it's about making the game make sense and just committing on-chain to the moves, I'm not so sure. But there has to be, you put on-chain when there's value to the chain verifying the change.
0:00**Michael:** But you can get creative.
0:00**Vlad:** I was looking at the chat to see if people are commenting on this. JB83X8R says: "Michael always talks about his chess example, never talks about its real world implications. Has Michael noticed that outside the Kaspa ethos there's not many people jumping in line to use Kaspa?" I think I can rephrase this question: academically Kaspa is interesting, but why doesn't Kaspa have more users? And in spite of the fact that you have a lot of potential and capacity for transactions, why are these transactions not actually being used by people?
0:00**Michael:** That's the whole point of the movement of the programmability enabling. It's only very new. And even though the hard fork was a few months ago, the tools to use it properly took more time. A lot of things are popping up. A lot of people can be working on it in stealth mode, developing a product, a prototype. It takes time for an ecosystem.
0:00So we're actually seeing the first signs of activity now. There is an effort, there is a fundraise for an initiative called the Kaspa Dollar, which is about creating a decentralized stablecoin over Kaspa. But in order to not wait for that day, someone launched a centralized bridge, three out of five signers, people known in the Kaspa community, that is bridging Ethereum and BNB into Kaspa. And you can actually now trade Kaspa to KUSDT on Kaspa L1 and then bridge it out. That's already deployed on mainnet since a few days ago.
0:00So things take time. People were always complaining, like that commenter, always complaining why we're not doing things for utility. So we've done a lot for utility. It just doesn't happen overnight. And there's so much we can influence. From my perspective now as being in the core team, I can do my best to now enable as many use cases as possible. And then you have someone like Yonatan that is actually trying to build something that is novel in terms of full crypto perspective as well, talking about stag hunt and related initiatives.
0:00So there's so much I can personally do. I can slowly enable things. I can do as much as I can to consult, to help others. And I might find the time to develop an application which I'm not going to tell anyone about until I actually deploy it. So there's all kinds of ways to do stuff. They all take time. That's what I can say. The whole effort of the last year and a half is about bringing utility.
0:00**Vlad:** I think it's also worth mentioning that even Bitcoin right now, I'm looking at the mempool, there's not much demand for the block space. And we can complain that Bitcoin doesn't scale, but for what it's being used right now, arguably it scales pretty good because the amount of users trying to use it are not running into any congestions or the issue that transactions are too expensive. With just 19 cents, you can get a high-priority transaction mined into the next block.
0:00**Michael:** Yeah, it's just that we kind of gave up. And by we, I mean in Bitcoin and maybe other communities.
0:00**Vlad:** You go to a Bitcoin conference, you see that most people pay with their credit cards. And it's an industry-wide problem because even, let's say AI, which I like to argue that proof-of-work mining enabled the existence of AI because we created a much more resilient power grid and much more capacity that's being used right now for data centers. And you have some large mining companies, including Marathon for example, which used to mine Kaspa a couple of years ago, and they're pivoting to AI. We sort of enabled this. But on the other hand, if you want to pay with your coins for AI credits, that's a challenge. The major companies don't accept it. I think NanoGPT, and I got to look that up to back up my claim, I think they accept Kaspa or they accepted in the past. I'm not sure. Let me look it up.
0:00But that can be one way in which you can use your coins for actually paying for AI use. But it's kind of disappointing because we don't have more use cases. We have a technology that's getting better and is becoming more usable, but it doesn't mean that humans are actually going to be using it for its intended purposes.
0:00**Michael:** I think Yonatan was pursuing this question of what systems are uniquely enabled by crypto. And one of his answers is stag hunt, like wide coordination. He was saying the internet is an open space but we haven't built institutions for the internet. And I think that's one of the questions he's trying to answer.
0:00**Vlad:** Yeah, I've looked at NanoGPT and it seems like they do accept Kaspa. So you can actually use AI. Remember using NanoGPT for anyone who wants. Let's do a test. I put a Kaspa address in the comments. You can see it on I think YouTube and Twitter. If you send any transaction to this address, it's going to turn into AI credits. And I'm curious to see if anyone is here, like send me a dollar or whatever. I'll send it back to you in Bitcoin if you want. But the point is just to see if the community is willing to do this on one hand, make a quick payment, and on the other hand to see if NanoGPT is going to validate this instantly and top off my account.
0:00There's also the other way around of speaking about AI agents using crypto.
0:00**Michael:** Oh yeah, that's the other big conversation right now. We're talking about AI agents choosing a cryptocurrency to settle whatever between themselves because they cannot have bank accounts.
0:00**Vlad:** Yeah. And the question is always when this conversation, which one is the AI going to choose, right?
0:00**Michael:** Probably not one specific. Can be multiple of them. I want to believe that Kaspa has interesting attributes: speed, maybe also the covenants. I think the covenants are an interesting question. You can actually very cheaply deploy a custom contract between you and another agent. You don't have to deploy a complex, like a deployment on Solana for instance. If you try to deploy just a custom contract, you wouldn't do that for small operations. So speed and compression of contracts are interesting properties of Kaspa that I hope can speak to AI agents or whatever runs them.
0:00**Vlad:** What happened to that address? I can't tell right now because I've changed it. It was asking for $100 and that's a problem. So I've changed it to one. Let's see if it works. Start new payment. Did anyone send you that dollar?
0:00**Michael:** I got to check. What's the best block explorer? kaspa.stream?
0:00**Vlad:** I apologize if it's not going to go through, but if it was asking for $100 and you didn't send, that's fine. That was a default setting.
0:00Wait, how does this work? Because I pasted. So just put the address?
0:00**Michael:** Yeah, the site takes a few minutes to verify. You're not a bot, I think. So interesting.
0:00**Vlad:** Yeah. Let me put it on the screen actually. An address with $1. There are two types of addresses. It says wallet link or address only. I will choose address only. And I'm going to take a screenshot of this and also put it on the screen. Sorry for taking longer. It's just an experiment right now.
0:00**Michael:** I'm not fully following. What should I look at?
0:00**Vlad:** You want to send? That's funny. I can send to you if that's needed. I can send you the address in plain text if you want. I'll put this on the screen. I think it should show up now. Yeah, it did. I'm going to make it smaller. Okay, it's right here on the screen. I'm curious how this is going to work if I get the credits directly into NanoGPT. And then I'm going to ask NanoGPT how I can port Pokemon for building maybe on the Kaspa blockchain or whatever. That could be a fun experiment. It's going to close the cycle.
0:00But once again, I think there's going to be some copyright issue that is going to block NanoGPT from even answering. It will say no, you're infringing on some copyright or whatever. So that means I have to rewrite the game or remove anything copyrighted from the game.
0:00**Michael:** Well, someone said that they sent 150 Kaspa. I don't even know how much that is right now, but thank you.
0:00**Vlad:** 150 Kaspa would be five dollars. Oh, really? Nice. Thank you very much. I'm going to use it for burning tokens. I become an addict when it comes to primarily GrokBot because it doesn't use Grok. I sort of noticed while using it that it's much better than Grok and I was thinking what's happening here. I got the $200 subscription and then I burned the tokens in a day. And then I later found out that it's actually using Claude Opus 5.5 for reasoning and Midjourney for images, and that's why it's burning more. Not using Grok, maybe for some low-level tasks. What they're saying is that it just uses the best model for the job.
0:00**Michael:** I see. Maybe Grok itself is very impressive. I realized that by writing some informational posts sometimes with the right links, you can get, and then you can do a Grok conversation to test how other people would understand that information, to see if it passed by clearly. And it's very accurate. It knows how to access the resources, the links. It has a very deep understanding of things if it has the right context, which wasn't like that a few months ago.
0:00**Vlad:** What I like it most for is just researching tweets because the other models don't have access to the tweets. So I use Grok for, for example, I have this idea for a project where I'm going to take everyone's bad price predictions and put them on a chart. And usually people should post a lot when they say, well, Bitcoin is going to go to a million dollars by next month or whatever. So I want to put that on a chart so you can see exactly who said that and at which price and what actually happened afterwards. It could be hilarious to look at.
0:00**Michael:** Shaming.
0:00**Vlad:** Well, it's not shaming per se. Oh, someone, it does say transaction failed in NanoGPT. Maybe too many people sent this $1 at the same time.
0:00**Michael:** Ah, it's an address that they provided to you. I see.
0:00**Vlad:** Yes. Maybe I did something wrong. Can you send me the address so I can look at the explorer and see? I will send it to you right now. Let me paste it. I basically don't have Telegram on my computer, so I only have it on my phone, which means I send notes from one device to the other.
0:00**Michael:** Updating. Okay, it's this one.
0:00**Vlad:** So if there is any issue, it's because of NanoGPT processing the payments. But it's kind of cool. It's one of the websites where you can use Kaspa. For those of you who are listening to this, I made a statement that the AI companies don't accept crypto even though crypto enabled AI to be more powerful. And even crypto enthusiasts were running GPUs from home to train models. There is now, you got like four people sent you, you have four UTXOs under this address.
0:00**Michael:** Yeah, all from three minutes ago with overall 161 Kaspa.
0:00**Vlad:** Okay. So let me check balance. It doesn't seem to have, oh, there are some pending payments. Okay. It says they're not yet credited. There's only one of 91 cents and it says transaction failed. I don't know why. Let me try another one for five or $10. Not that I'm asking people to send me more, but maybe it's a NanoGPT problem with small amounts.
0:00**Michael:** So what happens when the transaction fails? Is it sender? What do you mean, who reported fail?
0:00**Vlad:** NanoGPT said that it failed.
0:00**Michael:** I don't know what they mean. If they saw the transaction, it screws the deck. So they saw it. So it didn't fail. You didn't tell them you sent them, no other way for them to learn about the transaction. So I'm not so sure what that means.
0:00**Vlad:** Seems like it's stuck in a loop. I'm going to send you the screenshot. There's also a transaction ID, but I think it's internal for NanoGPT, not for Kaspa. It looks like this. I can put it up on the screen so people don't assume that I'm lying or whatever. It's a shame. I had expectations from this integration. I'm going to ask Milan, who is the guy who's in charge of NanoGPT, what's going on with this. But they support Kaspa, which is nice.
0:00**Michael:** Yeah, maybe they have some problems.
0:00**Vlad:** So this is what it says. You can check the addresses, by the way. I put them in the chat. I'm talking to the people who sent: thank you very much for sending. Seems like there was an issue. I just put the error on the screen. This is what NanoGPT says, but I didn't get credited. So let me give it one last try. I'm going to put an amount like $10. Let's say that it's reasonable. Not too much, but also not small enough for NanoGPT to complain. We go to add funds. It became slower.
0:00**Michael:** Should I send something equivalent to $10? Would you want to see?
0:00**Vlad:** Yeah, let's do it. Start new payment. It says, so I'm going to use Kachat. Ever heard about Kachat?
0:00**Michael:** No. What's Kachat?
0:00**Vlad:** It's a chat system over Kaspa. I think so. Let's margin. Okay. So this one will ask for 372 Kaspa, which apparently is $10 in their system. I'm going to put the link in the... So, I sent 512.
0:00**Michael:** But did it succeed?
0:00**Vlad:** Yeah, it always succeeds. [laughter] Yeah, because there are no orphaned blocks, right? Let me send you the new QR code. Actually, you don't need the QR code. You need the plain text address, right?
0:00**Michael:** A new address? Oh, I already sent to the figure.
0:00**Vlad:** Oh, this is such a mess. I feel like I owe you already.
0:00**Michael:** No, I created a new payment link for $10 on NanoGPT. And this is the address. I'm going to send it to you by DM.
0:00**Vlad:** No, but I'll let you know if the other one went through. This is funny. Anyway, that previous address now has 683. I'm going to have to contact their support now. It does say that I have estimated credit of $10.15 from Kaspa that's not yet credited.
0:00**Michael:** Yeah. So that makes sense. You have almost $20 in there.
0:00**Vlad:** Basically this works better. I'm also going to change the image on the screen for the latest QR. This is for $10 if you can make it happen. Any Kaspa whale, let's try this.
0:00**Michael:** I sent you a link to the explorer, to a more responsive explorer so you can see. explorer.kaspa.org. It's easy enough to remember.
0:00**Vlad:** That's another question, because when Yonatan was on the show last year and also this year, I've asked him what's a good Kaspa wallet to use, and he sort of refused to give an answer, which I guess is diplomatic from his perspective.
0:00**Michael:** You can go into kaspa.org. There's a better place to look at this: kaspa.org. Huddle. Assuming that something happened, that maybe something happened.
0:00**Vlad:** Are you still there?
0:00**Michael:** Yeah, I'm just, I'm not sure what happened. kaspa.org works for me.
0:00**Vlad:** Ah, okay. So there's a huddle page and you can see many wallets. kaspa.org, or yeah, there is Kaspium, huddle, bidd, and dag viz. What's dag viz? DAG Vision?
0:00**Michael:** That's the visualizer. Oh, visualizer.
0:00**Vlad:** So to get a wallet, you can have one on iOS, Android, Linux, Mac OS, Windows, and hardware wallets. Which hardware wallets support Kaspa right now?
0:00**Michael:** Tangem and Ledger.
0:00**Vlad:** Oh, yeah. OneKey, SafePal, Tangem, and Ledger.
0:00**Michael:** Oh, and it's cool that you made it look almost like bitcoin.org.
0:00**Vlad:** I think it was inspired. Yeah. The ratings for every wallet, like how much control you have, validation, transparency, and fees.
0:00**Michael:** Yeah. Not me, but yeah, someone did it.
0:00**Vlad:** Looks good, but in dark mode. Oh, you can turn on light mode, but I don't like, dark mode is fine. You prefer dark?
0:00**Michael:** Yeah, dark is fine, but I can't code in dark. I don't know why.
0:00**Vlad:** Most people code in dark. I can't.
0:00**Michael:** Well, you called yourself a third-generation adopter or something, or contributor. I think those who started with MS-DOS or Borland Pascal, Basic, they just had this blank screen which was either blue or black.
0:00**Vlad:** What was your first computer and programming language, by the way?
0:00**Michael:** Where I did my undergrad studies, they actually started with, I think it was C++.
0:00**Vlad:** Interesting choice. It was C++.
0:00**Michael:** Do you know which version? Dev C++ or...
0:00**Vlad:** We're speaking about 2012, I think, 2011 maybe. It was just, I don't know, I used Visual Studio 2010, that was like the, I don't know if you...
0:00**Michael:** I know Visual Studio. Yeah, now we have VS Code, but then it was Visual Studio. Visual Studio 2010, I think, was the first I was using.
0:00**Vlad:** Yeah. So across the addresses that I shared on the screen, I have 10 Kaspa in the one that is right now on the screen, and 683 in the other one, which is like $18.51. I'm going to have to contact support for NanoGPT and let them know about these addresses. Maybe they're just too fast for them or something.
0:00**Michael:** Yeah, maybe the integration needs maintenance. I don't know. You never know. You saw some indication of something happening, right?
0:00**Vlad:** Said something is pending. Maybe they have a confirmation policy that takes, I don't know, a few minutes or an hour. You never know.
0:00**Michael:** Yeah, it does countdown to an hour, but I don't know what that means.
0:00**Vlad:** Let me check the first address too. So I have three of them that I shared tonight.
0:00**Michael:** Where should I look?
0:00**Vlad:** Because the first one was asking for $100. So I presume that nobody sent to that. I hope so. No, someone's... 380 Kaspa in this one. So in total I have like $30 in Kaspa that I received from you guys. Thank you. I hope I'll get this credited on NanoGPT for using them to burn some AI credits. This was like a useful testing because NanoGPT is one of the places where you can spend your Kaspa, and it seems like they can't keep up or something.
0:00**Michael:** Someone mentioned it to me. I don't remember who. Someone mentioned the usage of NanoGPT. I remember hearing about it. Yeah, how good is it as a model?
0:00**Vlad:** It's not a model. It's like a marketplace for models. You top up your account and then you can use anything and only pay for what you're using. It's not like a subscription.
0:00**Michael:** So you can use GPT or you can use Claude or whatever.
0:00**Vlad:** So the kind of anonymity is like the usage of crypto and also the no-KYC and stuff like that. No KYC, but also you only pay for what you use. That's kind of the sales pitch, because with subscriptions, sometimes maybe you pay $200 but you're on vacation and you don't use it too much. With this one, you only pay what you use.
0:00**Michael:** Yeah.
0:00**Vlad:** Is there any other heavy topic we want to touch on?
0:00**Michael:** Of course. [laughter]
0:00**Vlad:** Oh, you think I'm going to let you go so soon?
0:00**Michael:** We can go on some more, but not too long.
0:00**Vlad:** Okay, I respect that. We're not breaking records today. No, I think Yonatan spoke for like seven hours and then Shai Wyborski spoke for about eight hours, and that was the longest episode for a while.
0:00And actually I want to reference this because the longest episode was with someone called Moonsettler who's working on Bitcoin covenants, and I had a Twitter exchange with him today where I firmly disagree with him. He said that the value of a network is in network effects, and if a new network or a fork doesn't have the network effects, then it gives no benefit to the users. And I disagree there because I compared, I said that Bitcoin forks are just like Linux distros. You spin up a new one and it has new features and people can choose to use it. And he said, well, there's no benefit in using it if you don't have the same network effect. But you objectively have features like contracts, like speed of confirming transactions, like privacy. And by privacy I don't just mean coinjoins or whatever. I mean hiding amounts or pruning the previous blocks so that you don't have a permanent storage of everything in plain sight. You have some features that people want that can be used directly on day one, and people will want to use it only because they find it useful. The network effect can't be the argument for freezing a technology for so long.
0:00**Michael:** I also agree with this. But the argument right now is that there is so much value, quote-unquote, on the Bitcoin network that you can't just give up on it or tell the investors to fuck off or whatever, because you have institutions, you have ETFs, you have Michael Saylor, and somehow they're important as voices and you have to please them. And when they say that something should not happen, especially Saylor who said that covenants add complexity and they're not welcome to the Bitcoin code, that somehow becomes influential.
0:00**Vlad:** What does it mean, add complexity, by the way? Does he mean that on a technical level, like the script engine is complicated?
0:00**Michael:** I don't think he understands what he's saying. I think he is ill-advised, if you ask me. But I think what complexity means is just extra lines of code, and somehow adding any extra line of code is perceived as an attack surface.
0:00**Vlad:** Yeah, there can be formal ways to verify. We're not talking about complicated changes. I don't see that as, it's not a valid argument.
0:00**Michael:** I know, but he's not, I mean he is an engineer but in other stuff. I think if I understand correctly, the objection to OP_CAT is not about the fact that OP_CAT can take two entries on the stack and make a new entry which is longer, which is a concatenation. I don't think that's the complexity problem. More about the fact that the Schnorr trick exists, so there is a way to use OP_CAT to implement covenants. I think that's the problem. The abstract entity called covenant that doesn't control value 100% but kind of establishes internal establishments within the portal. I think at the end of the day that's the fear from something like OP_CAT.
0:00**Vlad:** I've never heard anyone express fear about that, only hope that this can and will happen. When it comes to OP_CAT, people were mostly complaining that ordinals people will be able to set up much better marketplaces and automated market makers or whatever on-chain. And this was one. But the biggest one was that you can have what they call unknown unknowns, meaning that at some point in the future, someone smart will figure out a way to use this in a way that breaks Bitcoin, but we don't know yet.
0:00**Michael:** It's possible to prove what systems can do nowadays. That's very funny. Who said there's no, okay, did they prove that the current Bitcoin is safe, or are they relying on the fact that it hasn't? Let's ask them for the same kind of, why is, prove to me that Bitcoin is safe. And whatever proof you use, I'll use that kind of proof system to show you why additions can be made. Makes no sense.
0:00**Vlad:** No, it doesn't. But we have been here for years. And every time someone says, well, but it isn't about the abstract idea of covenant, like the fact that you can create a covenant, that's, isn't that what scares people? You don't think so? It's only about the technical complexity of the script engine having a new operation.
0:00**Michael:** I'm very convinced at this point that you can put the cure for cancer on the Bitcoin blockchain and nobody will use it because everyone is huddling. You can bring the best possible use case and the best technology out there, and we have seen some of these ZK rollups that launched and then they have only a few users. It's very disappointing because you expect Bitcoiners to be happy that their chain becomes more useful.
0:00**Vlad:** And this will not come without a change in culture. I think this experiment that we did right now with asking people to send some coins to this address and they responded immediately is proof that the Kaspa community uses Kaspa for payments and they're very used to sending transactions. If I did this with Bitcoin, they would be much more skeptical, or I don't know, they would be like, well, I don't have any coins on my wallet because I keep them in cold storage.
0:00**Michael:** I sent it to you from the Kachat because that's where I have small money. And the chat system, you can actually give me your Kaspa address and we can chat encryptedly over the chain. We send transactions to each other. We establish a handshake and there's a protocol someone else wrote above Kaspa that encrypts the messages over the blockDAG. You can actually just chat. The fun part is that you can interact with someone in chat and you can send him value, a coin over the chat, as part of their identity. Their identity is both an owner of coins as well as represents them as an entity in the chat system. They actually have, you can write, you can have groups, shared chats. I don't know all the features, but someone, a very dear person called Kaspa Silver, I forgot his real name, but he writes a lot in the Kaspa community. He has videos, he has a video stream program, and he's developing now, he's maintaining this Kachat, which is very cool. His main goal is to bring the usability as much as, to make it very practical to normal people.
0:00**Vlad:** Can you also send more transactions from the chat?
0:00**Michael:** I can send. What do you mean? You said that there's a handshake between the two addresses. I think it's a UX thing. Once you establish the chat interface, can you say type in the chat like "send 10 Kaspa to this person" in words?
0:00**Vlad:** It doesn't have to be words. Can be something.
0:00**Michael:** Yeah, in chat you just open the, you're in the chat and you have chat and you can just the plus button. I can show it to you. I might need, okay, so there's a plus button and you can simply say send Kaspa.
0:00**Vlad:** That's cool. It's exactly what Facebook was trying to build with Libra, if you remember that.
0:00**Michael:** I remember the name. Yeah, it was their idea. If you can send messages in your chat as quickly as this, just typing and clicking a button, why can't you send money like this?
0:00**Vlad:** Yeah, it's also about being censorship-resistant. You only need a node that you can obviously control and maybe a lightweight indexer. I'm not sure. And you can just chat with someone without no central authority being able to censor, to stop you or ban you or whatever. You don't know you need it until you do. It's part of the, but this is actually a system built over Kaspa even pre-Tokata, before covenants. Covenants are not even integrated deeply into the system yet, because transactions were able to carry payloads also before. So now as far as I know, he's working on more features which are deeply connected to covenants. It can make it much more interesting.
0:00**Vlad:** Yeah, someone just sent me another one Kaspa. Beautiful. Unfortunately, if you're trying to establish a connection with the Kaspa chat, it will only work with NanoGPT because it's their address. I was hoping that I'm going to get a top-up directly into this, but it doesn't.
0:00**Michael:** You can download the Kachat on Android or iOS and then you'll have...
0:00**Vlad:** I'll do it right now just for the sake of it. I remember I was using another wallet like one or two years ago. I think three years ago is the first time when I tried Kaspa, when you could mine it from your computer. And I was not very successful in mining.
0:00**Michael:** CPU mining.
0:00**Vlad:** Yeah, CPU mining worked for a few months only. But the client kept it, I think.
0:00**Michael:** Yeah, you could run it, but on mainnet if you run it now, obviously it's not going to produce blocks. If you ran it at the GPU era, you would still get blocks every few days, maybe.
0:00**Vlad:** How is it now? There is someone in the chat who was angry and was saying, why can't I buy Kaspa ASICs?
0:00**Michael:** I don't know. Maybe the main ASIC producers were both, what's the name of the biggest company? I forgot their name. Bitmain.
0:00**Vlad:** Yeah, Bitmain had custom miners. I don't know if they maybe they don't have an active product now or they don't have a new line yet. And then there was a specific company, Ice River, that was founded for building miners for Kaspa.
0:00**Michael:** Interesting. I don't know. I haven't tried to buy an ASIC recently, so I don't know.
0:00**Vlad:** Are you currently mining Kaspa?
0:00**Michael:** No, because I feel like it would take some time for maintenance and I can use my time better for Kaspa specifically. Actually it is related to something I wanted to mention before: RTD, real-time decentralization. There's now ability, the capability in Kaspa to have miners write something in their payload, and we can aggregate that information and make it some kind of an oracle system of miners attesting what's happening in the world. The Kaspa Dollar initiative is planning to use miners as oracles for the price. And I was thinking to myself that once Kaspa is used for RTD and there's a real system using it, I will want to mine myself because I can see the added value of decentralization.
0:00I'd say it's always important, but there's a trade-off in time that I can invest, and I'm not an expert in mining. It's not my domain. But at that point I'm going to try and get an ASIC and maintain it because I want to contribute to the decentralization of RTD.
0:00**Vlad:** Let me ask you the most Bitcoin question ever. What kind of hardware do you need to run a Kaspa node? And do you run one?
0:00**Michael:** You need, 16 gigabytes RAM would be enough. You need, in order to be safe, an SSD of half a terabyte or a terabyte. And usually it uses much less, but in order to be fully safe you need, I don't know, eight cores at least, I'd say. We use parallelism a lot for processing. If it's going to be a laptop, you'll note it, so it depends on activity. You might notice it on your laptop. But yeah, that's the kind of hardware. We were always aiming, we only shifted to 10 BPS when we knew that reasonable hardware can run the node.
0:00**Vlad:** So you're saying that the latest generation of Intel and AMD processors that have at least eight cores and 16 GB of RAM, which I guess are kind of expensive right now but still findable, and an SSD, you can run a Kaspa node with that. Like a $1,000 laptop can run a Kaspa node.
0:00**Michael:** Yeah, definitely. Have you tried on Macs?
0:00**Vlad:** Personally I don't have a Mac laptop, but people have run, yeah obviously. I'm not sure about the M series of processors, how many cores they have, but I think any M series processor has more than eight cores and 16 GB of RAM. I think with a new MacBook you can run a Kaspa node.
0:00**Michael:** Yeah, most reasonable laptops have that spec. Oh, that's very cool.
0:00**Vlad:** Yeah, I mean I came from high performance and I did my share in making the Kaspa node efficient and performant. The Rusty Kaspa software was designed with parallelism in mind from the start. When you submit, when the network has multiple headers that are in parallel to each other, they're also by definition, you can validate them by definition in parallel because they don't know of each other. So their state cannot depend on each other, right? So we use that to parallelize, to do things in parallel because the blockDAG itself is not sequential. That's one part of the system. And we do it all over: transaction level, multiple inputs, whatever, classic parallelism. But the system is fully designed so that things happen in parallel all the time. There's like a pipeline where headers are processed and bodies and UTXO validation. It's all, everything is always busy.
0:00**Vlad:** There's someone in the chat who gave possibly the best explanation as to why NanoGPT might have rejected the deposit. Maybe they're searching for the exact amount that was requested for that invoice, and anything lower or higher than that will just get rejected. So I'm going to contact their support.
0:00**Michael:** But then they need to ask for a specific amount of Kaspa, because you can't know what $10 exactly are at that moment. So I think it did, but I didn't mention that.
0:00**Vlad:** Okay, I was in a hurry. But in that situation, I'm just going to ask them to return to the sending address.
0:00**Michael:** Someone is complaining that I'm repeatedly saying it takes time. Well, oh yeah, I read that. Yeah, things take time. And we're not waiting. We, as developers of Kaspa, we did as much as possible from my perspective in order to enable things. Now it's, we can't build the whole ecosystem. Some core developers now are transitioning to building apps, like the Kaspa Dollar initiative is led by a core developer, and others will take initiative. Some other core developers are developing the prediction market. People take their time developing things. They want it to be secure. They want it to be audited. That's just the facts. I mean, the fact, saying it takes time, but we did a lot of actions to make things happen. So I'm not going to apologize, but that's...
0:00**Vlad:** I don't judge the team very harshly because I've seen what happens in other projects and how slow they move. So in comparison to what we see in Bitcoin, you guys are very fast. And I've also seen over the last couple of years what happened with Zcash. That one was considered dead for a while and then it came back because it had the right technology for the right time. It was what they call product-market fit.
0:00**Michael:** Yeah. And I think it's amazing to witness. It's very interesting and fascinating to watch the Zcash developers operate as professionals. I don't know. It's very inspiring to see sometimes how they communicate, how they perform, how they move fast, how they release code, how they formally verify whatever they can, how they try to be transparent and open about things, handle tough situations nicely. I don't know. It's just been nice to watch.
0:00**Vlad:** Yeah, it's inspiring. But this was only the last year. I think previously they had, they have a long history. It's a 10-year-old chain. And they launched with that famous ceremony from which I actually own one of the laptops used, one of the six. And afterwards they made some promises about the tech which were right about the privacy but wrong from the perspective that there was an inflation bug that they needed to fix. And then they had to get rid of the trusted setup and they needed to scale the signatures because they were too large. It was like a constant improving process.
0:00**Michael:** We're talking about before 2020?
0:00**Vlad:** Before 2020, yeah.
0:00**Michael:** So it took them almost a decade to get to this point, and it paid off, to their credit. It really paid off. It's exactly what it was supposed to be, and it's at the right time when there is actual demand for this.
0:00**Vlad:** I didn't ask you about privacy in particular, but you did say that you worked on ZK layer twos. Do you think that Kaspa can have shielded pools or something like that?
0:00**Michael:** Yeah, it's possible. I've been observing Zcash, like I said. I also witnessed recently Solana deploying some kind of privacy system called rings. And Kaspa, since the recent hard fork, supports ZK verification, two types of ZK verification. One of them is actually unique. We actually can verify a full STARK proof directly on-chain, meaning we can verify ZK proofs which don't require a trusted setup and are post-quantum. They're obviously big. They occupy, a transaction verifying a STARK proof will fill up a block in Kaspa. So you can have 10 of these in a second if you want to saturate the blockDAG.
0:00Anyway, ZK opcodes naturally allow a lot. Their main aim is typically in our environment more about scaling, right? More about, okay, covenants are expressive, but what if I have a very complicated program which can't fit in Kaspa script and I want to ZK prove it? That's not a use case for privacy, that's a use case for scaling. We have the idea, an idea called based app, but ZK codes can also be used for privacy.
0:00**Vlad:** Yes, I would love to see that. And I love when, for example, Vitalik talks about adding privacy to Ethereum. I think it's very courageous of him, and supporting the Tornado Cash developers. And I also think that the window of opportunity for normalizing privacy is narrowing down. I think that based on the results of the elections that we saw in the US, the openness for privacy tech will be much lower going forward. And maybe we're going to have four years of just Operation Chokepoint or trying to attack anything that's not happening on banks, which is really unfortunate. We had, I think over the last two years we had a pretty good time building stuff. And it's interesting how the Zcash movement seems like they always have a plan for anything.
0:00**Michael:** So what I would be curious, not that I wish that kind of, you probably mean more like strong regulation pressure against privacy?
0:00**Vlad:** Regulation, yes. But also they can criminalize the use of anything private.
0:00**Michael:** Yeah. Because they can see when you use it, right? They can't see the private transaction, but they can see that you withdraw your coins from an exchange and then shield it.
0:00**Vlad:** Yeah. They can track the entries and also the exits. And Monero doesn't solve much either because they keep getting delisted and their parallel markets are not very liquid. So if you actually want to use it for amounts... And Zcash doesn't get delisted because privacy is optional on most exchanges. I think all of them except for Gemini: when you withdraw Zcash from the exchange, it's transparent.
0:00**Michael:** There are also some swapping services that let you withdraw shielded Zcash directly. So that's kind of cool.
0:00**Vlad:** But if we don't actually build markets for private stuff, we might be losing this battle. And while we are going to have the technology, we're going to end up just like Railgun for example on Ethereum, which has filters and blocklists and whatever.
0:00**Michael:** I guess you know very well how politics can change, and something that is perfectly normal today will become illegal tomorrow and somehow it gets criminalized. And they will always say, well, it's the worst of people that they're blocking. But in reality it's political dissidents. It's people speaking against the government, people trying to escape their countries or whatever.
0:00**Vlad:** Yeah. But anyway, I sort of took you on a forked path. I want to get you back on track. Let's end this conversation with DAG Knight because it's your special topic. You are a co-author of the paper, and I want to keep you focused on something that you know very well.
0:00**Michael:** Okay. That's actually occupying my mind now very much. What are you curious about?
0:00**Vlad:** Let's answer a question from the chat. Someone really wanted to know when there's going to be a testnet or any production code that people can see.
0:00**Michael:** So we're moving on three fronts in parallel now, all of them feeding each other. Finalizing the full proof of the protocol, which changes small details here and there, or subcomponents. Then there's a yellow paper effort, actually spec'ing algorithms in the language closer to code. And also the implementation is going forward. I mentioned on X a few weeks ago that I think we will get to a testnet by the end of the year. I still think so. We're going to do a big release of materials and stuff in a few weeks when it's all ready, and the progress will be very clear.
0:00**Vlad:** For people who are unaware and they just joined the stream or they're listening to anything Kaspa-related for the first time: what is DAG Knight and how is it different from blockDAG?
0:00**Michael:** BlockDAG is just a data structure, or as Yonatan called it, it's the problem, it's not the solution. The solution is a consensus protocol that utilizes a blockDAG to reach consensus over a high-speed system, so to speak.
0:00DAG Knight essentially doesn't... the consensus protocol, the transcript or the node code that runs it, sees a blockDAG and runs an ordering. You need to order the blocks in order to evaluate. The node obviously has a UTXO state, right? That state is a product of a sequencing. If you sequence all blocks in some order, you can compute the exact resulting state. If two blocks are trying to spend the same UTXO, then the order dictates which one of them wins. The first block in the order is the one that still sees the UTXO as available and executes that UTXO. The second block that's trying to use that UTXO will be in the blockDAG, but the transaction will not count and will just mean nothing to the state.
0:00There's a transcript running this order protocol, and in DAG Knight it doesn't assume any explicit network delay bound, unlike, I'll explain in a minute, unlike Bitcoin and unlike current Kaspa.
0:00What does that mean? This has two implications in two different realms. First of all, a protocol that doesn't assume a delay bound and is still somehow secure can confirm transactions in time which is the real delay in the network, in a sense. If you explicitly or kind of explicitly encode a delay bound in the protocol, then that's kind of the, then confirmations typically are always proportional to that assumption, because if a protocol knows the delay bound, it's going to use that information in order to make a decision or to converge. So once you encode the delay bound, or if the protocol needs to know the delay bound, that delay bound affects the confirmation times. And if the protocol does not need to know the delay bound, then confirmation times can actually be faster.
0:00The other edge of the spectrum is what you mentioned: the resilience to a world where if there's a catastrophe in the internet and internet communication slows down significantly, then a protocol that assumed a specific delay bound, let's say of a minute, while now internet communication takes 10 minutes, then the security of the protocol actually is compromised. And it doesn't matter how long you wait, you actually have no guarantee that the honest majority can use its majority anymore. And if the protocol doesn't know the delay bound, you essentially get resilience also to worst-case delay. So there are two edges, two sides to this, and they're both important.
0:00That's the high level of the story. A lot more nuances. I just don't know how to do them short enough.
0:00**Vlad:** There are some questions in the chat about the power of DAG Knight, like if it can actually keep the network running if they presented a hypothetical situation where Iran cuts down the internet cables underneath the ocean.
0:00**Michael:** It depends if that means that the internet, the worldwide internet, is actually partitioned, like there's an internet in Europe and an internet in America. Then, okay, so for how long? I think you need to model what happens. Are we assuming wartimes mean that communication, there are still valid links but they're slow, so eventually a message arrives but it can take, I don't know, tens of minutes? Or are we talking about a disconnection of days?
0:00A disconnection of days long requires, DAG Knight is an ingredient, but you need more components to allow the system to converge. You also get to the question: what happens, how do you make progress on each partition at the time of the split? How can I, if we operate in Europe and I can see your transaction, but how do I know if on the other side of the network you're performing the same, you're doing a double spend, and only when we try to converge history in a few days we'll realize one of the sides wasn't valid? So there's a question of how to make progress in such a network. I think that's very extreme. The first goal I'd say for DAG Knight would be to target a situation where the internet is very noisy and tolerate that.
0:00**Vlad:** That sounds good. I've downloaded Kachat. Is that how you pronounce it?
0:00**Michael:** Kachat, I think so. Yeah. People say Kachat, Kachat.
0:00**Vlad:** You can send me some coins if you want. If you want to chat with me, we can give this a try. I'm actually interested if any community members are interested in clipping this long interview, creating short clips of various statements from this interview that could be interesting. And maybe I can reward you with like $10 in Kaspa for that, for your efforts.
0:00**Michael:** Let me try and scan panic as well. Can you show it?
0:00**Vlad:** Well, let me put this on the screen too. I'm actually curious to see how the UX is on it. You showed me a very brief glimpse of it from your phone. So it seems like with five Kaspa I can send like 2,500 messages, and with 50 I can get my own Kachat name for a while. I don't know what "for a while" means. I guess it means that it expires.
0:00**Michael:** Yeah, it expires. My scanning through whatever I saw on screen didn't work. The resolution wasn't high enough. So I'll do it later.
0:00**Vlad:** It's fine. I also put this on the screen. I can send you the address separately. What's happening here is that it's asking me to connect either to the default node, to my own node, or to auto-search for nodes. I'll choose auto-search for nodes just because I'm extra paranoid, like Electrum, you know, it chooses a node randomly. It says I'm all set. This is what the UX looks like. I hope people can see. And I have no message requests for now.
0:00**Michael:** Does it have GIFs or emojis or anything fun?
0:00**Vlad:** I think it has emojis. Yeah, it has emojis. Could be interesting. So to everyone who sent coins to NanoGPT, I'm going to make sure that you get refunds. I got to message them because it's really not fair.
0:00I do have a message request from caspacouple.kchat. They say, "Hi there Vlad. Caspa couple here. Nice to see you are using Kachat." I called it Kachat. And I can accept it. But if I want to send a reply, I need to fund my wallet.
0:00**Michael:** Does it let you purchase Kaspa directly from this?
0:00**Vlad:** I think he was working on this feature. I don't know if it's already... I'm sure that some people thought about doing something like this in Bitcoin, but it's not very reliable.
0:00**Michael:** Yeah, and it's also not responsive enough.
0:00**Vlad:** Well, they could do it on Lightning, but you get a lot of failed payments. Oh, and that's nerdy. I like it. You can see to which node you're connected. Pool status, active nodes. I like it. It's not designed like some sort of corporate messaging app.
0:00Oh, someone sent me 100 Kaspa. Thank you for that. It's very cool. Caspa couple. Oh, nice. So now we have funding. I do. I'm going to... So what should my name be now that I can do this? I can get a domain. Oh, I can get Kachat and .kas or .k. I like .k.
0:00**Michael:** It's a covenant. It's a covenant name service. It's very cool. I think also Kachat name is a covenant name service. When you buy a K name, I don't know if you can do it through this app or there's a website called K something. Let me see. k.name. That's where you can register a new name to that address.
0:00**Vlad:** Interesting. But how do I, oh, I have to go to another website to register. I don't think he integrated the K from this app yet.
0:00**Michael:** But what about... he recognizes the... so the Kachat is a native one. So I believe it is supported. But addresses, chatting address.
0:00**Vlad:** I sent you a DM with my address. Someone told me getkaspa. It's dog. I like this. I'm going to spend some time playing with this. Thank you very much for sending me the 100 Kaspa at caspacouple.kchat. I keep calling it Kachat. I'm going to give this a try.
0:00There's a question from the chat which I think is very good for ending the interview. Man from Bogota says: ask him how he sees Kaspa in two years' time, and if he is concerned about Kaspa's emission schedule.
0:00**Michael:** Of course anyone is concerned. That's the main motivation for bringing utility, so the fee market can provide more income to miners. There are also other relevant ideas like RTD providing more income opportunities to miners as well. It's basically the main motivation for everything I've been doing in the last year and a half: bringing utility, trying to find the product-market fit, trying to increase the capacity, the utility of the network, so that it can self-sustain without printing more money, so to speak.
0:00I believe that the capacity of the network, the TPS it can reach, the volume is high. So if we manage to be that infrastructure for this kind of utility and get more liquidity, that's going to... so there is a possibility that this should be enough to compensate for the emission schedule. There are more answers, but it's a broad discussion. I don't think I should go into everything.
0:00**Vlad:** This person wants to know: what are you going to do when Kaspa is a complete project?
0:00**Michael:** I don't know what the definition of complete is. I myself would like to do more things over Kaspa or focus on research that can contribute to Kaspa on the upper level, upper layers. So I'll find my way to contribute in such a scenario. I'll stay around. I'm actually hoping for the day where I can be not so obligated with the ongoing R&D and actually can be more creative and think about my own initiatives above Kaspa.
0:00**Vlad:** There's a question. Please give a short answer to this one because we can go on for another hour. 33 Ethereum wants to know if now that you have covenants enabled, if you can do smart contracts just like Solana or Ethereum in Kaspa L1. That's a different question. He's also asking if off-chain computation is necessary for that or if there are any trade-offs.
0:00**Michael:** If you want fully, like a complicated program that's not easily, that cannot easily be written in script, then that's why we have ZK, and we have a based app design where you can actually write a complicated program and it's still committed to the sequencing, the exact sequencing of the L1. The term "based" means that the app is obligated to follow the exact sequencing of transactions on L1, and it executes payloads from transactions in that sequencing, and it must prove to the L1 it executed all these transactions in order, fully. That's part of the ZK proof. So there's a whole design of how to build more complex programs on Kaspa. They still have their limitations. Related topic called vProgs, which is not fully implemented in Kaspa L1 yet. But yeah, that's a broad topic as well.
0:00**Vlad:** I was actually answering to a message in the chat. Cool. People can use this to message me. I'm on Kachat now, so you can actually send me messages there. Once again, I will establish a $10 bounty for short clips that you post on social media. For every good one, I'm going to give you $10 in Kaspa, so you can use that. I'm going to make your currency also more useful because I acknowledge that it has value and I'm willing to exchange my money for it so that I can pay you. This is how adoption happens.
0:00**Michael:** Yeah.
0:00**Vlad:** And what I want to ask you, Michael, and maybe this can be the closing question. In the beginning of the interview, you were talking about making Kaspa successful, and I want to ask you: what is the definition of success? At which point do you look at Kaspa and you say this project is successful?
0:00**Michael:** I don't know if seeing use cases, like utility use cases for programmability, is the only answer. Kaspa can be successful also as a peer-to-peer payment system if it had enough volume and adoption, if it became more and more accessible. I think Kaspa as a concept has more value than only a peer-to-peer system. It's a fast proof of work. So it has unique attributes of sampling decentralization very fast. So it can bring other unique services related to that kind of decentralization. It has what Yonatan calls fast censorship resistance. So combined with all these, yes, Kaspa more, in terms of more, you obviously need the market cap of Kaspa to be larger to say it's utilized correctly in terms of money. And the other alternative is widespread usage of all kinds of apps over Kaspa and various layers.
0:00I think you can build complex systems now above Kaspa and you still have a decentralized and secure and fast network. And these properties, not always so obvious why they're needed, but their existence should prove itself eventually to bring unique solutions. So I'd say more usage, more utility, maybe a unique product over Kaspa that utilizes Kaspa's properties correctly. It's a very high-quality project and it should win eventually. And I believe crypto is a long-term game. It's very volatile, but I think value is seen over time. Very good technologies cannot really be ignored. In general, I act under the almost, say, very definite belief that Kaspa will succeed. I kind of know it's going to happen.
0:00**Vlad:** I was not aware you can do this, and I gotta say it now that it happened. Someone sent me a domain in a message. Now on Kachat I am "takeover." I'm no longer this, I mean I am this address basically, but now I have a fancy name: takeover.
0:00**Michael:** Takeover.
0:00**Vlad:** So thank you whoever sent me this. It's cool. And also want to comment, it's midnight for both of us right now. I want to comment that your metrics for success are not something shallow like saying, well, I want my local shop to accept Kaspa or something like this. They have more, of course you're a technologist, you're an engineer and you're a researcher, not just someone who works on applications. And you have expectations which are mostly technical about the network, not necessarily how the network is going to be used. Because I can presume the first use cases, and the ones that usually emerge when you have a fast payments network, they're going to have something to do with gambling and maybe some video gaming stuff, which I think could be novel or interesting.
0:00But beyond that, is the goal to replace Visa? Would you say that replacing Visa for payments would be a metric of success? Or is that not the goal anymore? Because we are in a world where instead of replacing Visa, we're integrating our coins with Visa to spend them on their network.
0:00**Michael:** Replacing Visa is still a goal, I think. Maybe an ambitious one. So you have to define shorter-term goals in order to make it realistic. I also imagine, again, that's going to be an engineer's answer, I also imagine that the internet needs these protocol layers which are trustless and provide state management, provide decentralized trust. Like TCP is a protocol for communicating between parties, and many routers on the network participate in every TCP message. So imagine the sequencing capability at high frequency used as a backbone of the internet eventually.
0:00I don't know what role we can play in that, but I believe we can play a big role. So I also have that kind of vision of an infrastructure or the ultimate sequencer or all kinds of stuff like that. But first Kaspa needs to find more usages in the product area in order to gain enough volume, in order to gain enough liquidity. It's kind of a flywheel. So you can't win the big wins first. You have to win more specific wins. You have to get adoption one way or another, and then things lead to things, and you can then achieve perhaps the more grand vision. It's my perspective.
0:00**Vlad:** You think Kaspa is for internet payments, or do you also see people doing in-person payments to each other, like walking into a shop, paying with Kaspa and leaving?
0:00**Michael:** Yeah, definitely. Because it's fast enough so it can be used for that. I would, okay, so if you had worldwide usage for small payments, I'd say it like this: if there's capacity, it's obviously a good use case. But worldwide consensus about the order of things as a valuable service doesn't necessarily justify every dollar I send to you in a coffee shop. You get from the system much more than what you need. Even if the system has capacity, of course you can do that, and it's fast enough so it's perfect. But essentially, if you still look at this from an efficiency perspective, you still see a system that provides a much more valuable service that actually provides you the ability to send something to the other end of the world securely at the same speed. So it provides worldwide coordination and consensus, and that's like a very, so in that sense I don't think chats or local coffee payments utilize the engine to what it really provides.
0:00**Vlad:** So that's, I don't know, I hope I answered. There's a mean comment in the chat which I think is still relevant. Someone says, well, you want to...
0:00**Michael:** I'm used to mean comments.
0:00**Vlad:** What's that?
0:00**Michael:** I'm used to mean comments. That's the internet for you. [laughter]
0:00**Vlad:** So he says, "You want to replace Visa, but you can't even replace Litecoin." And I'm not sure if he means the market cap of Litecoin or the network effect of Litecoin. But basically Litecoin benefited from people walking door to door asking businesses, hey, will you accept this? And they also had payment terminals that were easy to integrate with because Litecoin is very similar to Bitcoin. And a lot of the legwork was done by advocates in 2013, 2014. The software was already there. Payments processors were already accepting it. So it was very easy to integrate. In the case of Kaspa, I think it's a whole different challenge because you need to get them to first of all get exposed to the brand and to the existence of this currency, and then you need to give them the custom software which is not compatible with what we have with Bitcoin. Or maybe you can get payments processors to do this for you.
0:00**Michael:** Yeah. I'm not sure if BitPay is still big nowadays. I haven't heard anyone use BitPay in a long time.
0:00**Vlad:** I'm not claiming to know the answers for everything or claiming to be able to do everything about Kaspa that requires that kind of adoption. I try to do what I do best. And that should be the power of a decentralized movement. And Kaspa has a unique community in the sense that there are growing circles of people that are very motivated and try to work on it and build on it and market it. And yeah, I think I'm doing my part.
0:00**Michael:** I'm doing my part.
0:00**Vlad:** I just got the meanest comment ever and I want to thank you, Kaspa Blutely. He says, "Vlad, Michael is answering your question about what he's thinking of when Kaspa is successful. It's not really polite being busy with Kachat and not paying attention to his answer in my opinion." Okay, I'm sorry. I apologize for that.
0:00**Michael:** Both of you might... No, you can multitask.
0:00**Vlad:** Yeah, I'm trying my best here. I have a chat of people commenting, I have Kachat now of people sending me messages, and also have to ask good questions and get good answers and come up with the next question. But I'm very grateful, Michael, for your time. This interview has been like two months in the making, or even longer because Yonatan mentioned you this summer.
0:00**Michael:** That was August. Romania.
0:00**Vlad:** Yeah, August. He was kind of on his way to us.
0:00**Michael:** Oh, you were in Romania?
0:00**Vlad:** No, we were in the Netherlands and he came to us eventually. We had a secret, I call it core retreat.
0:00**Michael:** Core retreat.
0:00**Vlad:** Well, I hope you guys had fun and it was a productive time.
0:00**Michael:** Not so much about fun. It was the kickoff for DAG Knight essentially, for the main effort for DAG Knight. DAG Knight was developed by Coder of Stuff before, but that was like the... And he obviously is, there's a guy named Coder of Stuff who's been working on DAG Knight for a while, and he couldn't come to the Netherlands because he's a pseudonym and he wouldn't break his identity.
0:00**Vlad:** I think I've actually inspired someone. Coincidentally, there's someone Yonatan in the chat and he said, "Would be cool to make a private server on RuneScape running fully on Kaspa." RuneScape is this online RPG. Once again, you thought RPG is like the rocket launcher, but it's actually a genre of video games.
0:00**Michael:** And if someone pulls that off and I inspired people to run video games using Kaspa, then I guess this podcast was more successful than expected. [laughter]
0:00**Vlad:** Yeah, I'll take that. [laughter]
0:00**Michael:** Oh my god, there's like a never-ending flow of haters. I can't please everyone. I'm sorry. Thank you very much, Michael. Is there anything that you want to say in closing, like where people can follow your work or follow you on Twitter or your papers or whatever?
0:00**Michael:** Yeah, I share, I try to stay active on X. My username is @MichaelSuttonIL, one word. You can follow my GitHub activity. I'm happy to speak to anyone. And to the Kaspa community, I want to say I love you all and I can see the criticism sometimes. I see people get, I mean, some of the people can be obsessed, but I think it's also a sign of care, right? If you don't care, you don't write anything. So even seeing someone complaining all the time can be, there is obviously a minority, most people are really inspiring and supportive. I'm just, there is constant criticism even though I have no idea what I'm supposed to do beyond what I'm doing. So I also appreciate those voices and they are kind of a constant reminder of keeping pushing. So let's all keep pushing and we can build a lot of interesting stuff about Kaspa. So go ahead and do it. Don't wait for others.
0:00**Vlad:** Last advertisement that we have on the show. There's this guy Triage the Machinist who says, "Vlad, whenever you want to buy a Supra, I can sell you one and I accept Kaspa." Auto... service. That's the name of his business. He's been accepting Kaspa since 2023.
0:00**Michael:** What's a Supra?
0:00**Vlad:** I've been saying for the last four years almost that Supra is like the pleb's Lambo, like the Toyota Supra. It's a sports car. It's more reliable than a Lambo. It's fun, but it costs like 10% of the price of a Lambo. So you get like 90% of the experience for 10% of the price. And I've been saying that one day when I make it, I'm going to buy myself a Supra. And then I didn't buy myself a Supra.
0:00**Michael:** So what is he saying? You can buy it with Kaspa?
0:00**Vlad:** Yes, he accepts Kaspa at his auto service. That's what he calls it.
0:00**Michael:** There are a lot of companies and services accepting Kaspa. You'll be surprised when you see the list, wherever it might be. I don't know. But yeah, you can go on vacations with Kaspa, you can book hotels, there's one specific website which I forgot the name of. Anyway, yeah, I guess we're falling asleep.
0:00**Vlad:** Thank you very much, sir. I appreciate your time. I wish you all the best. I hope when we do this next time, you're going to give me more status updates on DAG Knight. Maybe that one will come out in full production. You said by the end of year?
0:00**Michael:** End of year to a stable and secure testnet. That's my hope. That's my plan and hope.
0:00**Vlad:** Okay. And then 100 blocks per second. That's what you want to see.
0:00**Michael:** No, I don't honestly. [laughter]
0:00**Vlad:** Yeah, I think, I mean it doesn't matter what I think because I'm not like a big Kaspa community member, but I think it just makes it more difficult for everyone.
0:00**Michael:** So creating capacity for something that doesn't have demand yet prematurely is like, yeah, it's not about capacity though. It's only about the frequency, and if it provides more value to RTD, like faster whatever. There's arguments that are not about capacity. I agree that capacity is only relevant when we see saturation of current capacity. It's not like, yeah. Anyway, premature optimizations are almost the root of all evil.
0:00**Vlad:** Yes. Like do this when there's actual demand and people use Kaspa so much that you need more blocks.
0:00**Michael:** Yeah. Buy cars apparently.
0:00**Vlad:** Yeah, cars people can usually wait 10 minutes before they get the car they pay for it. But I'm not sure if it was Auto's Out Service or it was supposed to be Auto Scout Service and he has a typo. So that can be a possibility. But anyway, enough advertising. Thank you very much.
0:00**Michael:** Okay, good night. I'll see you on the next one.
0:00**Vlad:** Good night too. Thank you so much for having me.