Lilypad Network: Serverless Distributed Compute for AI

GM, GM, and we are live. Welcome, everybody, to what's going to be another fantastic DevNTell. If you didn't know what DevNTell is, DevNTell is a 30-minute podcast held by Developer DAO to showcase founders and companies and weekend hackers' projects for those who have a passion for building awesome applications and whatnot, just like we're going to have or going to see today on our show. If you are interested in AI and Web3 and seeing how potential use cases can come up with the two, this is the show for you this week. Today, I am ecstatic to welcome a couple of Developer DAO OGs. We want to welcome CEO Alison Haire, also known as Developer Ally, as well as DevRel Engineer Phil Billingsby, also known as P Billingsby online, of Lilypad Network. If you didn't know, Lilypad is a groundbreaking serverless and permissionless distributed compute platform, originally developed at Protocol Labs, which democratizes access to high-performance computing. If you stick around for today's show, you'll learn all about the platform and how to use AI on Lilypad, a decentralized AI approach that matches the ethos of Web3 but is just as easy as an API call. All right, let's get into it.

[Music playing]

GM, GM, welcome to the show, Phil and Ally.

GM.

GM, hey!

How's it going?

I love the balloons. Yeah, I love the balloons today.

I love them.

I'm super thrilled to have both of you on. Viewers of the show, if you didn't know, as I mentioned before, Phil and Ally are some of the original DD members, back when DD was spawned in 2021 or 2022, one of those. I am thrilled to have both of them on to showcase some groundbreaking work both of them are involved in. But before we get into that content, Ally and Phil, did you want to give a brief intro about yourself for viewers who might not be familiar?

Yeah, I mean, what a wild ride. I came into the crypto space a couple of years ago, and for those that don't know, I was working for Filecoin and Protocol Labs a couple of years ago. In the earlier days, I think it was earlier rather than later, I remember Nader Dabit starting up this experiment called Developer DAO. It was just like a few folks minting an NFT, no idea what we were going to do with it. And look where we've come now. It's just like such a massive testament to the community in Web3, and to the developers in Web3, that this is where we're at. We've met so many great people along the way that are doing some significant work here as well. I think you guys are some of them. We were reminiscing in the background as well. I think we met at a graph hack in San Francisco one night. We were all together around a fireplace, I think having tequila shots. That was a good night, but yeah, I'm Ally for those that don't know. I'm founder and CEO of Lilypad, as Narb mentioned. Great intro by the way, Narb. And we're looking to democratize access to high-powered compute. So what that means is we're creating a network of GPUs and CPUs that people can get paid for having on the network. We are trying to make it really easy for anyone anywhere to access kind of inference AI or other compute jobs. We've created it as quite flexible as well, so you can create your own jobs on the network and even earn some money off them as well. It's kind of a three-sided marketplace in that sense. It's multi-chain. You can run it off smart contracts, you can run it from a CLI. In the future as well, we'll have APIs and SDKs that make it a lot easier. But Phil will go through all those, so I won't rattle away too much. And yeah, had to hire the other Australian in the room, didn't I? Naturally, naturally. Go ahead, Phil.

So I'm Phil. Most people know me as P Billingsby with the red PFP on Twitter and all that. I'm a developer and a DevRel and a mix of other things depending on what's needed. Recently joined Lilypad a couple of months ago, and it has been an insanely exciting ride. The team, the tech, the vision, the mission, it's all just really exciting to me. I'm not used to being in such a vibrant community of folks just looking to kind of change the way we use AI, not only in a Web2 sense but a Web3 sense. But bit of a background, I came from a non-traditional tech background. I used to work blue-collar jobs for like 17 years. Went into fintech, thought that was complete and utter boredom, so I found Developer DAO just by the stroke of luck. I didn't even have a MetaMask wallet at that point, but I got one just specifically to mint the Developer DAO token. And yeah, it's kind of just been my life ever since is just exploring Web3. And now suddenly this whole AI emergence, so I've just kind of leaned into that. And then found out Ally was doing Lilypad a couple of months ago, like last year sometime, and I was very intrigued by it, but I just wasn't sure if that was my path forward. And then a few months ago decided to explore it a little bit closer and now I'm here working amongst these incredible humans. So yeah, that's the gist of it really.

That's wonderful. Yeah, I love stories like this, just getting into the grassroots of things and seeing how people kind of evolve over their careers and as humans as well. But yeah, this is wonderful to see, and I'm really excited to learn more about the Lilypad here.

I don't know which one of you wanted to drive the slides. All right, I will bring up the slides and we can get going.

Right, thanks Narb. So we couldn't go past chucking these ones up on the slides. Oh yeah, these are classic. The original pixel art ones. Yours is cool. You've got like a whole necklace. Is that a second version of the pixelated items? I think there was a V1 of pixel devs and then they did it again with the V2 and mine's a V2. Must be, must be. It looks like it anyway. Anyway, super cool, super fun.

So the Lilypad vision, I think I've kind of gone through it a little bit. We want to try and democratize access to these decentralized compute platforms while giving people the power to create their own jobs on the network. One of the ways that we differentiate actually is with this job-based system rather than a time-based system. So what that means is if you're trying to run a compute job or an AI job, then you don't need to go in, hire a GPU, set up the system, set up all the packages you need, set up the infrastructure, and then go ahead and run your job. What you can do is kind of use our system like an API almost based system. Like you just hit an endpoint and you get the information that you wanted back. And that endpoint hits a distributed system and runs on one of and is matched to one of our GPUs that matches your kind of order, I suppose you could say.

So we're currently live and in testnet already. So we have a marketplace already going. I've kind of explained it a little bit. It's compatible with all EVM chains. Currently, we've deployed to Arbitrum and we've done that because it's got fast transaction speeds and really low costs at the moment. But in future will be deployed to multiple different chains so you'll be able to use us from your favorite chain, whatever that happens to be, or whatever is good for the job that you're or the dApp or app that you're developing at the time. You know, as developers that's how we make decisions at the moment. We want to make sure that this is accessible for you wherever you are as well. We're also building out, you can currently access this from the CLI as well. We're also building out easier ways to use this from a SaaS layer almost, which will obfuscate a lot of the Web3 elements and enable you to just kind of put this in an app like or a front end like you would other probably NPM packages, for example.

So some of the technical details, I've kind of gone through that a little bit. We use our own token. At the moment we're only in testnet so this token doesn't exactly have value. And we'll have some exciting announcements at the end of this that I won't give away yet. But we're aiming to get to mainnet towards the end of the year here. We also use IPFS kind of as a hot storage layer. So what that means is you can do things like inference on our, do things like fine-tuning inference on our system. So you could have a whole heap of data in IPFS which you then train a model on, which you could then use that model for something else. So you could fine-tune a model rather than train an entire like ChatGPT model, but fine-tune a model to specific parameters that you have. For example, maybe the Developer DAO docs, you wanted to fine-tune an LLM model to that knows information about specifically about Developer DAO. Then you could fine-tune to that and then use that model in your applications. So that's where we're aiming to get to.

This is just a little bit of how the ecosystem works. So at the bottom layer we have this decentralized verifiable platform. I think it's important probably to note that building a distributed compute platform which is verifiable is kind of non-trivial, which is probably why you haven't seen it spring up before this, before this kind of massive demand for AI came about in the last year or so. I was actually just talking to Phil and Narb before this and mentioning when we first started here and we were looking at hackathons. We had one I think in June last year that we could participate, one AI across crypto thing, and now you have a look around and you definitely have like a massive choice of AI/crypto events that you can go and be a part of as well. And you know, I think that's right. I think AI is like fundamentally changing how we are going to live our daily lives in the next 10 years or how we're going to do our jobs or how you know we're going to do a lot of different things. So you know I think if you are sick of the hype I think you know there's a reason for it. And there's some really interesting things happening in the space, including in decentralized science, I would say. There's a lot of you know rare disease applications happening or other like kind of scientific research that is being enabled because you can kind of share between computers with systems like this. Our head of research is Stanley Bishop, he is a very, very big advocate of these kind of projects and is working on a couple like that as well. So it's really cool to see what AI is helping enable as well. Just on a complete side point there.

Anyway, this is our stack at the moment. We kind of have the underlying library, the smart contracts on top of it. Now these smart contracts are what give the kind of on-chain guarantees to our network. So you can imagine if you've got a bunch of GPUs sitting in in space here somewhere and you're sending a compute job to it, you want to know that that whatever GPU that job hits is going to actually do the job and not just you know try to cheat and send back something that means they don't have to use their GPU power but they still get paid. So in order to enable kind of this verification we use kind of these on-chain guarantees and some game-theoretic methods to achieve that. And to achieve that trust which is like you know one of the fundamental things about blockchain is having trust in your networks and and codifying those rules. So that's that's what we're doing here as well in this compute system. Obviously you can't actually perform the compute on-chain because you want to be able to perform these really big compute jobs and and that's just not viable on a blockchain. And nor should it be, blockchain doesn't have to be used for everything. But these features of blockchain definitely fundamental to our process.

So on Lilypad we have what's called a module. Now this is just the compute job. This is like an SDXL text-to-image job or a LLM command asking about the weather in Spain or whatever it is because that's where you're next conferencing. I don't know. So these these jobs are called modules in the Lilypad ecosystem. They're basically Docker containers or arbitrary WASM. So if you can containerize a compute job you can run it on Lilypad basically. It doesn't have to just be AI or ML jobs either. We're looking to use it for a bunch of different things within our own systems, including kind of CI/CD at the moment, which would be a kind of interesting use case.

So I think, Phil, am I handing over to you here or am I still going? I can't remember. So compute modules. We have quite a few already on our system, and over the next few months we'll be looking to incentivize having more folks build out modules. So it's kind of a community collaborative development part of our ecosystem, the modules, because anyone can build and run a module, and that can be accessible to anyone else as well. So there will be kind of a module marketplace that will pop up in the next couple of months as well for you to look through and use. But we do have some available already as well.

I think Phil, I'm going to hand over to you here because it looks like kind of this is your spot. But yeah.

Thank you, Ally. No worries. Do you want to drive the slides now and swap over or should we just go from here? I mean if you just want to like push down every now and then once I kind of get through it. Tell me to click, no worries, I can do it. Excellent.

So yeah, this is pretty much how to use Lilypad. There's a couple of different available tools that we can use to run Lilypad jobs. One of them being the CLI, that was pretty much the first tool that was released by Lilypad. Also recently released a beta JavaScript wrapper so you can basically use a POST request on any HTTP client, Postman, you know, Axios, fetch, anything like that. And you just pass in the module name, the private key that has your ETH and Lilypad tokens on it, and then whatever inputs you want, whether that be the prompt, the tunables, the options. So really simple to use. Let's go down one.

So yeah, the first thing you want to do is fund your wallet with the tokens. So you can get that at the faucet. I kind of missed the slide to add the network here, that was supposed to go first, but we can go to the next one.

It's meant to be on this slide, Phil.

Yeah no, I'm here. Maybe you stop sharing yours and I'll take over because it might be easier here. Yeah all good. Okay so add the network, then get some tokens at the faucet. Yeah let me know when you're ready to take over.

I'm going to jump over now. How's that? There we go. Okay, so first step you want to do is add the network to your MetaMask or whatever EVM wallet you're using. All of the information for this is also on our docs at docs.lilypad.tech as well, so don't worry if you miss this screen.

Next thing you do is load your wallet up with testnet funds. Same thing you go to faucet.lilypad.tech, put the the wallet that you choose to use or the private key you want to use to run jobs in that input there. I think yeah I think it's 100 tokens every 10 minutes or an hour, something along those lines, but those tokens will last you so long so don't worry you won't run out.

Then you'll want to if you're using this locally, if you just want to run it through the CLI, we have the installation to install the binaries on the docs page. With the smart contracts and the actual JavaScript wrapper you don't need to install anything, it's just readily available for you. So yeah, CLI instructions there.

This is an example of a Stable Diffusion image generation job running from a CLI. As you can see it spits out an IPFS link also, so it also uploads your results to IPFS. That takes a little bit to generate the results so I use the local version to kind of illustrate how that works. But yeah, here's an image of an astronaut floating in space with a white background. So pretty cool stuff that you can just do this on your computer just in one line of code in your terminal.

Here's an example of a prompt that you would run from a CLI. Basic no tunables or options for this one. This is the return, so pretty cool, pretty cool, unicorn in cyberspace.

This is with some tunables. So you have the steps which would be the number of iterations to refine the image and then the seed is kind of a kind of a stopping point so you can basically refer back to refer back to this number like, 'Oh I love this image and I want to be able to reproduce this the next time I run this command.' Add this seed to kind of keep track of that.

And now we have Stable Diffusion video. So you can see we have another seed here for the image, prompt, steps, and then video steps, and then we get this incredibly refined astronaut in space and it moves. So really cool stuff there.

We do have an Ollama pipeline that we've recently added. So that's just an open-source open-source module developed by Meta. This is cool for creating chat agents and things like that. You can run this from the CLI and kind of ask it anything really and and get a pretty unique response and knowledgeable response.

This one is Ally mentioned earlier we have some researchers looking into like medical side of AI and this is one that I'm really excited about. It is way beyond my scope of knowledge, that's for sure. But it's a protein folding tool that they're using to kind of come up with therapies and remedies for medicinal ailments and things like that. So definitely keep an eye out on what we're deploying to Lilypad with new modules because this should be coming soon.

Now we have the CLI wrapper which is the HTTP endpoint that we mentioned earlier. So basically all you need to do is very similarly to the CLI is add your private key, the module name, and then whatever inputs you have whether that's prompts or messages, tunables and things like that.

This is an example of running that from a client-side on the Cowsay module. So you put a message in, it's going to generate some ASCII art, I don't even know what the word is for that. Takes a few moments to generate and then it's going to give you a response down here. So that's pretty cool as well.

This is the same job being run from Postman client. Same result there. This is also in beta, so we're definitely working on modifying this a fair bit to add a bit more functionality and and kind of reduce the latency in this as well, so bear with us on the API.

And then here we have a Lilypad Hardhat template which is an end-to-end smart contract environment for running Lilypad jobs. It essentially deploys the token you'll use to pay for it, deploys the job creator, the mediator, and all the kind of inner workings of Lilypad behind the scenes. This is a very good kind of repo to get a rough idea of how everything works under the hood so I definitely suggest checking this out. But here again I'm running the Cowsay module, it spits out a job ID that goes to an IPFS CID and displays the results.

And then here we get into the other side of Lilypad which is providing the compute for us. And Ally, do you want to kind of take over on this side of thing or I don't mind continuing on? Where are we at? Yeah sure. Just like the Run a Node section.

Yeah, open to any questions on running those different modules on your system as well while we're here. I know I had a few questions there that I was just answering in the chat on how we add verification to our network. Anyway, so exciting news at the moment as I'm looking at this screen. We're just about to launch our, so we've been in testnet for a while now in different iterations of testnet for a while now where you could kind of play around and experiment with a few different models running on our network. Mostly we were running the GPUs in the background there. But exciting news we're just about to launch our incentivized testnet so keep an eye out for that next week. This will run in stages all the way through to our mainnet launch when there'll be a token that has value for use on the network and our tokenomics is fully completed. But this incentivized testnet is going to run in stages aiming to kind of bootstrap and test different parts of the network. So one of the things that we really want to reward first is going to be folks that are bringing their GPUs to the network.

So if you want to run a node on the Lilypad network we have docs on how to do that and there'll be some more information and videos coming out early next week to help you to be able to onboard your GPU easily to the network. And you'll be able to earn points then. Oops, I don't know about your job module thing. I think we were meant to move this slide anyway. So and you'll be able to earn points then. Now the points are going to be worth more, we're calling them Lilybits, they're going to be worth more in the early days. If you come onboard in the early days obviously you can get more points as an OG and then they'll trail off. Now there'll be different for the Developer DAO members as well that want to contribute, there's going to be different ways to earn some of these Lilybits rather than just contributing a GPU too.

So in future there'll be incentives for making modules on our network, there'll be incentives for building out tutorials or education content or contributing to the open-source code. So there'll be lots of different ways that you'll be able to earn these Lilybit rewards that will then when we move towards mainnet be redeemable for a certain amount of token as well. So keep an eye on that. I think Phil have I got a different version of your slides here or something? You might do. Yeah, the slides on screen are Phil's. Okay, oh they're Phil's right okay. You want to switch to yours?

So I've got Build Your Own Module here. Phil, did you want to go through that as well or should we just have a look at what's next for Lilypad? Yeah I think we should. Yeah, you want to have a run through? We've clearly practiced this, it's the back and forth with the presentation that's confused me a little bit. Yeah I mean we have a guide in our docs at the moment for building your own job module. It's very straightforward when you kind of look at it, but there can be a lot of complexity. So if you do go through this guide and have any troubles just jump in our Discord. We have so many people that have kind of battled through the same kind of issues and we should be able to get you the help you need. But everything to do with building your own job module is in our docs. Definitely check it out and yeah, just keep us posted on how you find it, what kind of issues you run into and if there's anything we can do to kind of help.

And the best way for people, sorry Phil just to ask, the best way for people to give you that feedback is through Discord or is there any other means?

Yeah, currently we kind of have a single protocol on help in our Discord. But we have plans to kind of create more resources to help people out, but at the moment just pop on in and say what's up.

Yeah, I think we've got some slides at the end here if you wouldn't mind swapping over to mine Narb that might be useful for folks if they want to reach out to us as well. Yeah, we swapped. Yeah, so this is just our kind of brief timeline on where we're going. Currently we're about to be at incentivized testnet which I was just mentioning before, then we're going to go multi-chain as I mentioned earlier and we'll also have that module marketplace where you can kind of pick out jobs that will be useful for you or contribute jobs that you can potentially earn money off as well. You'll be able to add a small fee to a job you make so if you make something really cool that people want to use all the time then you know you can kind of make some amount of money from from that usage. So three-sided marketplace there and this is kind of the great thing about crypto as well that it distributes the value more fairly across different parts of the ecosystem rather than being concentrated as well. So that's definitely what we're aiming to do.

We're also aiming to you know build out a lot of plugins and integrations for those kind of things that you're going to need if you want to build in decentralized AI. So things like you're going to need storage, so we'll have multiple integrations to different storage solutions or you will you know obviously need to access the open-source models or host those models. Or you might want to have autonomous agents running across this chain or on your jobs so we're going to enable many of those use cases along the way as well. Really trying to like you know bootstrap a decentralized ecosystem here, a decentralized AI ecosystem I should say. Anyway, aiming for mainnet launch towards the end of this year.

Here is a cool diagram of our incentivized testnet and some of the stages we're looking at which I was kind of speaking about before. Now these will be our main focus but most of the reward points will run the entire way through as soon as we release them. So keep an eye out. We'll be looking at putting out a definition and understanding of more information basically is what I'm looking for on each of these stages as they happen. So we're just about to release a bunch of information on the onboarding of GPU stage, then we'll be releasing all the information on how you can endpoints in the module onboarding stages. And most of these stages will also have these side tasks that I was mentioning, so things like helping contribute to the open-source code or which I hope you're doing anyway as Developer DAO members. But you'll also be able to earn points for that contribution as well. So keep an eye out for all that sort of thing as well.

Yep, we've just deployed to Arbitrum as I mentioned.

And here is an easy QR code, this is what I was looking for, to get in touch with all our socials. But yeah mostly we all hang out in the Discord so you definitely can reach us all in the Discord there. Follow our Twitter as well. Our marketing team including Sam who's another Developer DAO member has been putting out some great stuff recently so they're really working very hard getting our numbers up there or getting out information there really and there's so much information to get out. This is quite a complicated project so major props to our marketing team for keeping on top of all that as well. Yeah otherwise open to questions here Narb. I think this is the best way to get in touch with us.

Yeah 100 percent. And yeah all this information will be shared in the YouTube description after the fact, but yeah we can keep this screen here just in case people want to scan ahead of time. You briefly touched on it towards the end of how complicated a project this is, and it certainly is quite involved and it's amazing to see all of that kind of packed into one platform. Tippi had a great question. I think you answered in the chat but for folks who aren't reading the chat, can you talk about why it was so hard to put this thing together and some of the obstacles you overcame to deliver such groundbreaking hoodie?

Yeah awesome. I think hard maybe is the wrong word, like it's challenging in so far as there's a lot of different ways that you would want to use compute, much like there's a lot of, I'm coming from Filecoin so there's a lot of different ways that you want to use storage. Maybe you want it to be temporary, maybe you you know want to overwrite the information in there like a database quite consistently, maybe you want archival storage. So there's a lot of different ways to use storage and there's a lot of different ways to use compute as well, probably even more I think if you start thinking about it. So that's one of the challenges, how do you you know cater to all of the different types of use cases of compute? The next challenge is how do you make it easy for people to use? And one of the major challenges is how do you build verification into a permissionless system that doesn't run on chain? So these compute jobs don't run on chain, so how do you ensure that there's trust that these jobs are running? Now as I was kind of mentioning in the chats I think there's a few different ways that you could go about achieving this. So one of them is just using reputation, right? So you could have nodes that have good and bad reputation and then people are only going to really rent out the nodes that have a good reputation, so you're incentivized to make sure that you have a good reputation or you fall off being able to earn money. So that is a good way to do it but it really lends itself to kind of renting out of the whole GPU which you've got to set up all the infrastructure on and a lot of systems do that to good effect in the ecosystem here as well so Akash is probably one of those systems.

Another way to potentially do it which we were well, another way to potentially do it in the future more than now is probably with ZK. So people are probably thinking why aren't using ZK to verify calculations? These are big calculations. And ZK is already computationally heavy. You really can't use ZK to verify AI calculations yet. Now maybe in the future that is definitely going to be a possibility but currently there's just I mean unless you want to wait for like kind of a few days for your like rainbow unicorn to come back then it's probably not the best option which I don't think you do. But in future definitely when ZK catches up or when other cryptography methods like FHE catch up, we'll definitely be able to plug and play these kind of options into our system. Now initially we also built off a verification method that was optimistic reproducibility. So this enabled us to have a really performant chain still, you know still have those quick response times while still enabling verification through game theory basically. So what this meant was we had mediators on the system that would re-execute X percent of jobs randomly on the network to check that that node had run them correctly. Now if that node hadn't run them correctly their collateral that they had to put up in order to bid for that job would be slashed. So you know they're incentivized to run the job and earn the rewards and not get slashed. So this is kind of a game theory mechanism like a ticket inspector on a train, you know if the price of the ticket is low enough, the chances of getting caught are high enough and the price of getting caught is high enough you're going to do the right thing and just buy a ticket. So that's kind of the method that was involved there. That meant that we had to have determinism in the system though, which meant that every job had to be deterministic and it's quite difficult to make all AI jobs deterministic. Some you can which is an interesting you know little foray if you're interested in AI at all, there's an interesting a lot of topical discussions around how much of AI is deterministic as well that you can go down the rabbit hole on. But you know for the long term I think people are wanting going to want to do more than have input out equals input in equals input out. So we're moving to a different system currently which involves a blended amount of work. So it'll include reputation, it will include model fingerprinting and I've temporarily forgotten the other blend that we're using because we've got a lot on at the moment. But we're using we're going to use different algorithms and there will probably also it's also likely that we'll use different algorithms depending on what you want to do on the system. So the job itself will kind of say I want to use this verification method and and that will get matched to the GPU that will use the same verification method. So that's one of the difficulties. In fact that's one of the big difficulties here is doing the research into how you create trust in this kind of decentralized system and we're only trying to do it for inference, right? So we're only trying to do it for one-to-one GPU networks at the moment. If you look at someone like Gensyn, they're trying to create this decentralized model training network. So they want different GPUs to run different batch jobs and all come back together to create like ChatGPT for example and still have verification. Those guys are crazy. That's a hard, hard job. Exactly. So that's definitely one of the things that we do a lot of research around here and then we want to add privacy to the network as well so researchers out there if you love this sort of stuff we'd love to hear from you as well.

That's quite the hurdle and a wonderful answer. Great question there, Tippi. I have a couple questions of my own. Crowd if you have any more, please feel free to comment and we'll bring them up. Phil when you were showing the image generation, were those models that were doing the work built in-house by Lilypad or are they offloading that to some other model like OpenAI or what not?

Yeah so Stable Diffusion, very commonly used across AI, so yeah someone else built it, we just kind of run it through Lilypad. Gotcha. I think if you yeah if you scroll down there'll be an image. Where'd they go? I don't know so weird. I think our our PowerPoint's tripping out a little bit but it's okay.

I'm just going to add to that answer. So we don't use OpenAI, we don't use these API endpoints. These are centralized models. And that's great, like they're definitely I use it every day, I'm not saying I don't. I'm not that you know ethically inclined. However, you know these are central models that we don't really know where they're going and these companies have control over where they go in the future as well. We're using open-source models and if you have a look at the AI landscape at the moment there is a little bit of a fight between or you know a discussion, a heavy discussion maybe, happening between those who think that AI should be open and those who think that AI should be closed and protected and private. Now I don't think I'd find too many people in the blockchain or Web3 space that are going to agree with closing down development for your own safety. So yeah and we're also kind of seeing a lot of these companies that have control over these AI models you know pushing for regulatory change which is just insane to me. Like pushing for regulations that mean that other people can't have open AI systems or actually open AI systems I should say, can't have like you know AI models can't use AI models. That is just crazy to me. These AI models should be open, people should know what's going into them, people should be able to kind of you know verify their security before they run them. You know I think the community should decide the future of AI basically and so we're using open-source models in all of our efforts here which are like you know really good. There are a lot of people pushing for this Meta for example putting out Llama, this LLM model, really powerful LLM model. You know it's great. It's actually surprising Meta's almost fixed their image recently by putting out like these open-source AI tools which is very interesting yeah. Very interesting yeah. Curious to see if the trend continues there and maybe they inspire other companies to do the same but they have actually. There's a consortium now of companies that are dedicated to open-source AI releases actually. So there's an AI consortium now which also includes IBM and forgot who else but yeah which is great. It's great to see some of the bigs jumping on and going 'well actually you know we're not sure' and you look they probably selfishly motivated because clearly OpenAI is so far ahead at this point. But still I think it's a good thing that we have options.

Yeah, definitely. One last question and seeing how we're running over time here. So Eric asks, he wants to run his own module and how to do that, and that kind of segues into my last question is what's the best place for people to to start with Lilypad if they want to start playing around with the tech?

Yeah absolutely. So if you do want to build your own module, basically all you need is a job that you've Dockerized or containerized and you can then go ahead and run this just with an overlay. So all you have to do if you do have a Dockerized let's say SDXL job already, which is a text-to-image job, and you've already Dockerized that ready to go then all you would need to do is add a Lilypad spec to it which is just basically a GitHub repo with some of those matchmaking terms that we need to match the job to a GPU. And underlying this we actually use Bacalhau as well so some of this if you've been in the Bacalhau ecosystem it's a peer-to-peer compute protocol developed by David Aronchick who was actually the lead of Kubernetes back in the day so he started Kubernetes and then he worked at Protocol Labs which is where we started on this with him as well. He's one of our advisors and friend now. But he he built Bacalhau while at Protocol Labs for the last couple of years and this is kind of the incentivized version of that. So if you do want to build your own job module and you'll notice in the Lilypad spec there's some Bacalhau terms there if you're familiar with that, but there's a whole video on how to build your own job module online on our YouTube and also in our docs. You can look through how to build your own job module as well. It's a little bit non-trivial at the moment, as we head towards the next phase of Incentivized testnet which will be module marketplace, the DX on that should improve as well, so we'll make it much easier for you to be able to contribute to the network. But thanks for the question. Yes great question and great answer again. Unfortunately this brings us to time, but yeah this was really wonderful to kind of get a bird's-eye view of everything that's going on at Lilypad and really appreciate both you and Phil taking the time today to share it with with me and the people watching here today.

Dude this has been so much fun. Thank you for having us on. My pleasure Narb. My pleasure. Hopefully run into you again this year somewhere. Yes yes yeah definitely you're more than welcome to come back on to give us another bird's-eye view. I know I think we need shots around a fireplace for sure. Yeah yeah yeah or that. DevNTell IRL. I know, but just before everybody leaves today, I'd like to share the QR code, sorry for covering your face there Ally. The QR code for people to scan to claim their NFTs for being an attendee here on DevNTell today. So if you stuck around until the end, definitely you'll want to scan this QR code, fill out the form, and you'll be airdropped a NFT on the Base network within a couple of days for being an attendee live with us today. So you'll have let's say an hour and a half from now to be able to fill that out, and you should see the NFT land in your Base wallet within a couple of days. And with that I just want to wish everybody a happy Thursday, happy Friday, wherever you may be around the world. Hope you enjoy your weekend, that's going to be coming up. And that we will see you all back here next week for another great episode of DevNTell. All right everyone. Thank you. Thanks everyone. Cheers. Bye.