Launching the AI Compute Network | Nockchain

for the long term, but as you all know, at present, the blocks are not full of transactions, and so while the settlement is much faster, we're not taking full advantage of all that throw put yet. But the network upgrades and the block time upgrades, all of this is keying us up for the next upgrades that we're doing to the chain. And as you all know, in May, we modified the issue and schedule to be smoother over time, less aggressive up front to better improve the supply dynamics of issuance to miners, and we had that immediate having that occurred. So we're coming up on another having in a few months. I guess it won't be quite half this time. It'll be 25% or 33% reduction along the new kind of yearly to case schedule. We've been eating through blocks really, really fast as the mining competition's been heating up. That's just kind of a recap of what we had been doing, technically, but a lot of us are kind of, I'm sure, interested in all the organizational developments that have been going on that we announced. As we mentioned about a month ago, I guess at this point, we've really decentralized the protocol a lot. We've basically created these two new organizations that are carrying on not changed mission in different ways. We've made the NOC community co, which is a Delaware non-stock corporation, which we're aiming for nonprofit status of some kind, to basically aim to support the NOC community and work for the best interests of NOC and of privacy technology and of all of that beautiful stuff. The general idea there is that NOC community is NOC community co is in charge of all of the bridge. It's in charge of looking at and maintaining the NOC chain repo as you've seen. It's recently launched a bug bounty program where we can try to basically find security issues and get them fixed with the protocol. It's been very successful with that. We can talk more about some of those EK proof of work exploits that happened over the past few weeks and the really fast and efficient response to all of that that occurred. In general, the NOC community co has begun its operations around taking care of the ecosystem and doing community programs that are in the best interest of the ecosystem and of making sure that everything starts learning nice and smooth. As you all know, I recently made this company national compute with Shelby. We've started working on this idea of building an applied AI company that can tap into NOC chain's new AI compute network and access those consumer GPUs and start directing useful economic demand toward them. We can talk about some of these components, but that's the general idea. Zorpe has transitioned out. We've made these two new companies, NOC Community Co, which is primarily oriented around looking to become a nonprofit that's really focused on the benefit of everyone in NOC community and stewarding the protocol and national compute, which is focused on the AI efforts of building useful and interesting applied AI products and directing demand toward NOC's AI compute network. As you all know, we're launching NOC chain's AI compute network finally after it was first announced in May this Friday. I'm really excited about that. I've been working on it pretty much full-time all summer. It's been a long, damn summer of doing audits and more audits and more audits. Not a testing, but I'm very excited about it. Let me tell you all about it. NOC chain's AI compute network allows miners that are running inference for customers or for themselves or whoever they want to also be mining and securing NOC chain at the same time. If they've got customers paying them, they can have customers paying them to be running the latest frontier open-weight models while also being subsidized to do that by getting NOC. Not only can they do that, but every time that they want to, they can make a ZK proof. I know you guys have learned all about ZK proofs from us talking to you about them for the last few years. But they can make that same inference work that they're doing verifiable so that you can know that they ran some particular input past some particular specified model. All of this is wrapped into NOC chain's new useful proof of work algorithm oriented around AI. That's what we're shipping on Friday. Miners can start doing that very soon on Friday. It is compatible with the Perl useful proof of work puzzle. Miners that already have a good implementation of a Perl miner can direct that same thing at NOC chain. We just have our own proof format. You can imagine that over the years I've become pretty opinionated about how to do zero knowledge proofs correctly and securely. And so I wanted to build that whole thing from scratch myself so I could basically have control and find rain decision making over the way that we would build all of that. So we did a full re implementation of that whole ZK proof stack using the Plankey 3 library and integrated it all natively into NOC chain. So I'm really excited about that because that's going to give us control over the direction of the AI compute network over time and the way that we can make it more efficient and follow our own needs and not be beholden to anybody else and their implementation as we move forward. But we're starting with int 8 inference support in the NOC chain AI proof of work puzzle. And so what that's going to let us do is you can take any int 8 quantized model whether it's a dense model or a mixture of experts model. And you can compute that as a supported work type for no chain. So it's likely that we're going to launch some quantizations that are kind of recommended of various models on hooding face so that you can pick up some stuff that you know will be compatible with no chain out of the gate. But in the meantime you can pick up any of the stuff that's already in the pearl hooding face or you can do your own int 8 quant or find an existing int 8 quant and use that. So I'm pretty excited about this. So as I mentioned, you can use one of the existing GPU minors that already exists for Pearl. We're going to be launching a reference GPU minor in the repo prior to Friday. I'm working on that right now. So I'm pretty pleased frankly with with all this. I think it's going to incentivize a lot more compute to join the network and start securing the network. So anyway, one of the things that's really unique about the way we're launching compute networks on on not chain is that not chain is able to support multiple proof of work puzzles running in tandem, mying the same chain. So I don't know. I haven't looked at every blockchain that's ever existed, but I'm pretty sure we're the only people that have ever figured out a way to have multiple proof of work puzzles mining on the same chain simultaneously. So one way to think of this is that these compute networks are kind of like bit tenser subnets in that they incentivize a specific type of compute to join the network and they subsidize it. But unlike bit tenser subnets where you have these validator committees on not chain, every one of these proof of work puzzles is fully proof of work and they're just merge mining the same chain. So regardless of whether it's the old ZK proof of work puzzle or the AI proof of work puzzle that we're launching on Friday, they're both mining, not giving all the miners the knock token, and it's all just mining on securing one chain. So this is a really interesting mechanism. You can read through how the code works, but basically what we have is we have the assert algorithm that we use for difficulty targeting and difficulty adjustment, modulating the difficulty of mining each one of those two puzzles independently with different block-time targets to adjust to a ratio of blocks of each kind of puzzle. So we're starting out with aiming for 70% of the blocks being ZK and 30% of the blocks being AI. And that may change over time, but I think that it'll be a lot of fun to see how miners respond to these subsidies. So that's the starting point, which is we're launching this AI compute network on Friday, and it's going to bring a whole bunch of new miners that are able to mine inference as a verifiable inference capacity to knock chain. And then in a related fashion, we've been discussing the National Compute Company, which we're building as a part of all of this AI work. So National Compute is a company that shall be and I found it. It's an applied AI company, and we're building out and driving demand use cases for inference, and we're intending to use some of the consumer GPUs on the knock chain network that are mining the AI proof of work as a part of some of this applied AI capacity. Because the way we figure it is, first off, obviously the applied AI space around agent services, et cetera, is growing rapidly. But second, we think that it's a competitive differentiator over time to be able to have access to cheap, subsidized consumer GPUs in a cloud that are already doing useful work in being subsidized by something like knock chain. So I'm really excited to be able to continue the work on building out and making knock chain great that that work that I had started at Zorb of working on knock and and launching the whole chain. I'm really happy we were able to figure out how to how to keep working on the project and decentralizing the project and a lot of people really stepped up to take responsibility. And frankly, it's amazing what we've been able to build over the past year and a quarter. 15 months ago, knock was just an idea and then we launched it and it's grown to be this huge thing where there's all these different people that have worked hard to make it real to secure the network, to work on the network, to make the code better, to tell their friends about it, to all this. So I mean, I appreciate you guys working on the vision with me, working on it with these other guys. I know the knock community code board members intend to do their own Twitter space next week. And so I think that they're going to try to, you know, I know you know a lot of them, but I think they're going to be introducing themselves and talking about how they see the knock community code and what they see its role as being. And in general, I mean, now instead of having one company's orb that's working on trying to build a ZKVM and bring that to market and be real focused on just stewarding, notching and also trying to be a for profit thing. Now we've got basically national compute, which is this unrelated AI company that's building out the do to market for this AI compute network. And then now we've got knock community code which is in charge of the stewardship of the knock community. So I think the incentive alignments frankly a lot better now that we kind of have these like specific purposes for specific organizations. And in general, it's been amazing to see, you know, the different people from, you know, knock box from Southwest pool, from Lambda collective, and just from the community, you know, whether it's, whether it's former's orb employees or whether it's just other people that have been working on and working in and around the community stepping up and finding ways to be useful as basically, you know, work needs to be done. So let's talk a little bit about the two exploits from the past few weeks on the ZKVM work puzzle. So basically we uncovered that, you know, all these people were doing crazy stuff with, with AI to figure out ways to mine knock more efficiently. And they, they were not able to in any way break the ZK or anything like that. All the cryptography itself was fully secure and has been the whole time. But they were able to figure out a way to kind of like skip parts of the puzzle. So one of them was they were patting the polynomial with zeros, which is pretty dumb, but, but it worked, right? And so they were able to, you know, basically fake a really high proof rate and get it up to 100 million proofs per second. So I'm really proud of what we were able to do there. Chris reported that exploit to us and within 24 hours I had a fixed rolled out on the network and I had worked with and coordinated with everybody to make sure that they were all ready to upgrade at the right time so that we didn't have a chain splinter or weird fork or anything.