Competing For Quality Compute | The Marketplace Of Inference

of the Hermes three llama three point one eight b dot q eight underscore zero point no. Okay. Well, no. We're not gonna do that. We're definitely not gonna be getting into technicalities today, even though the topic is very much on it. So we're gonna be taking a high level overview of the compute bucket, one of the four c's in the Morpheus project. What is it? How can one participate in it? What is its purpose?

Where are we at now? What might be coming in the future? And just its changing role in the Morpheus environment as a whole. So let's jump into

the compute. The compute is too damn high, and there's a lot of competition going on for compute at the moment. If you're concerned, I guess, for for AI in general, it's like, oh, deep sea is going to come out and we're not gonna need GPUs, we're not gonna do any of this. Like, it's it's, it was too much.

That's incorrect. There's definitely the way technology works is you build out a lot of it, it gets cheaper, so then people can use it more. So therefore you need more. And that cycle just repeats over and over and over again. So there's a lot of competition for for compute going on at the moment. Now, as I've talked about previously, it's one of the four buckets of the Morpheus ecosystem.

24% of the tokens go to it over time and distribute distributed to people providing these compute resources. And pretty much you can just think of this as people with GPUs and fancy computers. You'll get rewarded not only for the amount of compute that you provide to the ecosystem and why do you need it? Well, if you are running agents, decentralized agents, they need access to inference, I. E. The, the kind of token, the message request being sent back and forth to these to these LMS, and

that this needs to be provided by someone. And it's it's got to be through a service. It can be a centralized place, I guess. But in terms of the the whole decentralized infrastructure of of Morpheus, you you'd want to have a kind of lot of marketplace, I guess, of decentralized compute. So it's determined not by only the amounts that you provide, but kind of the quality of it.

It. And this is determined by people who actually want to use it as well. So you need to both the, you know, the supply of commute, compute, and then also the demand on the other side, and they've they've got to interact. And I definitely see the the demand side for AI agents coming in the future and why you need particular ones which are coming from a more decentralized place because I just don't think the the the ones that we'd actually want to use and are really useful are gonna come from the

the centralized companies because they're not gonna let you have access to wallets. They're not gonna be able to, let you do a whole lot of things, which I which I think the, ones from the decentralized world would. But in any case, let's say like I'm a I'm a person here. I've got my little gaming laptop that I use for podcasting,

and I've actually only got half a GPU here because the I don't know. It was like the mothering board, the soldering board or something screwed up on one of them. So I'm running on half a half a GPU. And, let's say I wanted to provide this as as compute. Well, it's not gonna be that attractive because it's slow. You know, it's offline all the time. I close my computer, and I use it for other resources. So it's gonna degrade in quality.

And I'm just not providing a whole lot of it as well. So if I tried to connect to the the network and provide compute, it's not going to get well, so I'm not going to get rewarded for it. So it's going to be very much dependent on the market demand for AI models, which I think is booming. Some might need to continue as drip feed of small amounts. Some might not need these high adrenaline doses of compute coming in. And

BowTide Bluefin in the discord, nice name, puts it like this. Rewards are purely based on usage of your compute rather than your availability within the node. If it depends if, consumers choose to create sessions with you. This is based generally on your rankings. And so there will be a ranking system with this being, based on bid slash price, compute TPS and transactions per second, compute t t f t, time to first token, number completed successful set sessions, etcetera, etcetera.

This is me talking now. The the the ranking system is gonna be very important in terms of how compute is is measured and who is providing good quality and therefore getting rewarded by the Morpheus ecosystem. Anyone can be a provider, but not everyone will have demand and thus rewards flow to them. With that being said, if consumers come in and look for a cheaper session that has smaller compute, then there could be definitely demand for the basic build.

Cool. Alright. I think that's a pretty good explanation of of just why this is needed, what is wanted. Let's get into the nuts and bolts. And no memes here, unfortunately.

This is the the serious portion. So from the GitHub, at the simplest level, the compute node employs a proxy router that functions as an information transfer mechanism and helps create a marketplace for consumer slash prior provider transactions. Both parties host this proxy router, which is used to or router, I'm not sure how people say that, which is used to create sessions or engagements to send and receive prompts.

Okay, so what you can see on your little screen here is you might have a consumer, they'll have more staking to access compute, they'll create a session with a model or provider subnet, send prompts to the provider, this is me actually using it or using an agent

I'll receive this will go to the provider who will receive the prompt, exit execute within the LLM, and then I will get this back. And they the provider themselves will have registered a a subnet and creating a bid to offer compute resources. So this is where the the marketplace comes in.

There's also another diagram just below this, which I think kind of helps out in seeing what the the individual is probably like one of the cleanest I've seen of the the Morpheus ecosystem as a whole, which is showing, you know, how the distribution of the the token goes out to capital code as community, and it should show compute there as well.

How the AI model is actually probably provided by the provider and not by the user, they just want to access this thing goes through a session router. And then there's all of these other little accesses showing how these, the the router, the router is providing this go between between the the provider and the actual user. So in my analogy, I guess the nuts would be the the proxy router router, jeez, probably should have. I'm gonna just say router from now on router,

known as the Lumen model. And the bolts would be how the prompts, a. K. Inferences per second, go through here. I'm not technical. I'll give that right at the start here. So I'm not going to even try to explain how the Morpheus Lumen model works. The Lumen

network was created by Ryan Quadran and was basically to decentralize hash power on the BTC network. That was the original purpose of it and I think still actually works like that based on all of good things of not having a single point of failure, one software solution, minimal on change transactions and getting towards the Morpheus side of thing, blockchain agnostic, node security, etcetera, etcetera.

There's a another Morpheus podcast called Brainpower, definitely worth checking out, especially, episode seven if you want to hear a interview with him. And he mentions how BTC, decentralized compute is easier because they all use this thing called Stratum and, the Stratum protocol. And there's a little actual data being transferred if you want to distribute your hash power or decentralize your hash power.

With the Morpheus, it's a bit different because you're you're having to send inferences. And so there is a little bit more data being transferred back and forth here by but letting anyone provide GPU compute and keeping its purposes slash use anonymous, which is also really important in in this context, is much harder. So decentralized compute is possible. It's just very, very hard to actually get going. And, this is probably one of the sections of the more Morpheus ecosystem that I think is

taking the longest to kind of tick up to actually get some adoption to actually get people using it. It needs a lot more work in the back end rather than something like the capital contracts, which were being used at the start or people creating front ends and the builder community or even just coders creating, you know, all of the websites, all of the additional things related to Morpheus. So it's certainly one of the harder portions of the the whole network.

And speaking of immigrants, do you know yellow trumps white in the color scheme? Or maybe yellow bolts are wetter than better than white bolts? You can see I'm struggling for for jokes here because this is the serious section. But this is the one of the differences, I guess, between the white paper. The probably one of the main ones actually, which

had a lot more details about how compute would actually work. And it was talking about sending tokens, the more token back and forth as kind of like the bidding process or paying for things. Instead, what we use now is the Yellowstone compute model, which was created by Eric Voorhees, and probably some other people at Venice as well. So instead of paying for your compute with more, you're staking your more and pointing it where you'd actually direct it. And this is,

the kind of staking rewards are what goes to paying things. So once again, this is actually too technical for me. But the point is to maintain privacy of prompts and minimize on chain transactions while still allowing anyone anywhere in the world to access compute as long as they are kind of participating in the network and and paying for it with the Morpheus token, which is one of the whole points of of having this this token. So I think we're a bit discordant right now.

Where are we at? Well, shit. I actually have no idea. So I've trawled through a lot of the the compute subnet on the on the Morpheus discord. And there's been some updates here and there. Well, but lots of talk in there. Over the last year or so, I can tell you what has happened. So there was a compute testnet that started in June 2024 on Arbitrum Sepolia testnet, I guess. And Ryan Condren gave some updates at the August, before the compute maintenance went live on the 11/18/2024.

So it is actually live now. It's had a couple of audits, and people have managed to get some sort of hardware running on there and providing compute for there. The only dashboard at the moment is the testnet one, which is kind of defunct is just showing not no one using it. Because of course, why would you use the testnet if the mainnet is live? So I actually have no idea how many people are providing compute, how much is being provided,

all those sorts of things, like a little bit of a black box to me asking around, I wasn't able to find out either. So I'm not sure. Not sure is is where we're at now. So discordant discordant is a good way to put it. So it's morphin time. If you're beginning to morph, stand back.

How do you actually start to provide compute if if you want to participate in this? Well, the first caveat is that it's not really meant for hobbyists. Like I said, me with my little shitty computer, or even funnier, if I tried to use my like old MacBook from like 2010, I think we're not going to have a good time, no one's going to want to use it, it, it would be essentially useless,

worse than useless. So I'm sure it would just tie up resources and I would break things. So it's really for dedicated machine with the uptime, if you're looking to provide and to get a good reputation on on the marketplace. The TLDR of this, from the GitHub would be something like this or a Gitbook. Run a Morpheus Lumen node on a computer and make sure your ports are configured for inbound traffic.

Download and configure the model you want to provide, register as a provider and on chain provider registry, claim rewards periodically from the session router contract. So there's a couple of things just to note here. Because this is still early, it's not as pinned permissionless as one might hope. So to make it easy, and once again to stop farming, as we saw on one of the last episodes,

there's two things in place, the plug and play models, as well as the subnets. So the models themselves are are easy, and you provide compute for people using a certain type of open source model. So this could be the Mistral 70 instruct v 0.3 dot IQ 1 underscore m or that Hermes one I listed at the start, or there's a whole bunch of others, which people can provide. So this is the actual model that people would want to use. And I think in the in the future, it might even connect to

the agents themselves. And it's like a user wants to use an agent. And the compute provider could actually provide that as well. Or it could be the kind of three way thing that we see, which is, you know, user wants to use an agent, so they connect to the agent. And then the agent is the one accessing the the compute. So how that'll actually work, you know, we'll see. Those that are listed off were seemingly popular in the test net, so are likely being used right now as well. Now

the next step, I guess, is once once again to stop farming, like we saw in the capital bucket. So this is once again getting into the technical stuff I'm not so great at. You might need 10 k more stake to be a subnet provider. What's the difference between a subnet and a provider? From what I gather, only subnets get paid directly from the protocol. Providers can get paid directly from staked more of people who want to compute.

But then providers can be part of a subnet. I'm not too sure of how that works. Hardware and software are it's funny. Like, you'd think software would be more frustrating because you can get bugs. It's kind of like you don't know how exactly it works. You have to go scrolling in, try to find, like, where's this fucking bug? We've got to find this thing. I certainly get pissed off when when stuff doesn't work on my computer, and it's just like, why aren't you working?

But what's even worse is hardware, man. I hate dealing with shit that just breaks. So I'm the probably the worst techno technologist out. I certainly wouldn't ever claim myself to be one. Because problems related to technologies piss me off. So this is kind of like also getting towards that area of Bitcoin where, you know, I'm diving in, I'm learning more about it. And then I just get to this point where it's like, okay, SHA $2.56.

Okay. There's, you know, crypto elliptic curves and shit like this. And I just tap out. I go like, the smarter people who can do better jobs at this of explaining it. I'm sure it's very important. Do I need to know this to,

you know, be confident in this networks working? No, I'm just gonna have to put my trust in other people, in this case, that they that the things are actually working out how they're working. So morphing out, That's probably the closest I'm going to come to explaining compute ever again.

What if there's more to come? What if there's more to come? So what does the future hold? I guess for this bucket? There's been some suggestion, hints, maybe even theories, as conspiracy Keanu, is looking very confused here. You could have more compute borrowing and lending. So in terms of instead of just providing things, you could actually have like a market on top of a market related to compute borrowing, borrowing and lending.

So if you wanted to develop something without actually exposure to the the more token price, as one in the chat here is saying more to the moon, maybe. Who knows? We don't I don't encourage that type of talk on here. The

you could do you could do this as well. Eventually, maybe a centralized decentralized compute could come here. So, you'd, I guess, eventually want to see the the likes of Akash, BitTensor, all those sorts of decentralized compute places or the the infrastructure decentralization, I guess, is is what they're doing. I'm I'm not a % sure. I haven't dived into those projects particularly. That would also play into this as well. So that's I guess my my little summary here.

Ain't nobody got time for that. Morpheus AI compute. What? Karen, you haven't done a good job of explaining this to me and convincing me why this is important and why this needs to happen.

Oh, lord Jesus. It's a fire, my bronchitis. So you might be getting a bit tired after this and thinking like, dang, isn't grok easier just to use? Why why is this even important? And you know what, if this was solely based on text, LLMs? Sure, there's the there's the problems, I guess you're seeing of of those being censored and the lack of privacy. But even then there's open source models out there like Venice, you know, why is this even important?

And if it was just related to LLMs, I'd probably be inclined to say, Yeah, why does Morpheus need to have its own compute, decentralized compute and things like this? And who knows, even in the future, this, this could change just because like, I have no idea what's going on there. So, you know, is it even being used? Is it even that important? I'm leaving that into hands of other people.

What are the things that I can say, though, is that, and especially because people don't care about privacy, that's I think that's been well established, in general, and it's more about the UI and UX. So the open source models are likely to be sorry, the closed source likely to be faster and nicer and easier to use. The big draw in my mind of why somebody should have time for that

is the AI agents themselves. And I think it's just you probably need this the central component to to work in with all of these things, in terms of the capital providers, in terms of the incentives for people to create the agents, and then for the coders to actually work on these things and, you know, UI UX front end and things like that. It's one of those ones where it's kind of like, you know, Bitcoin nodes and mining in, in essence. So nodes in particular.

How many people have nodes is like 10s of 1000s, if that is securing a what is it now $2,000,000,000,000 network or something like that. Like most people, I actually was tempted to create a node for myself, a lightning node in particular, which ends up having to be a full node. And I and this was very important to me because of the podcasting two point o, be able to receive

payments directly via lightning within podcasting apps. If you haven't seen any of that, it's definitely worth trying out an app like fountain and seeing how you can connect with your favorite creator, content creators, directly via that way. On the audio side of things with with YouTube and video, it's a bit different. And, you know, I I almost bought a raspberry pie. I almost,

started doing this. And then I realized, like, dude, I don't want to deal with liquidity channels. I don't want to have to deal with this thing overheating. I don't wanna have to deal with, like, the even the miniature amount of noise that comes from a Raspberry Pi and the fan and making sure, you know, it's working all the time. What do I do when I travel overseas for months at a time?

And so I ended up just going like this, this is fucking stupid. I'm not going to do this. I like it makes no sense for me to do this.

And I think most people come to this conclusion, yet the nodes are a critical part in the way that that Bitcoin works. And so even though it's like technical, it's hard to use and it's not going to be important for the vast majority of people, I get the feeling this is one of those things where it's like this actually does need to work and you need to have at least some people enough to make it decentralized, to form part of the backbone of the whole network.

And that's kind of how I view the compute side of things. So I actually like you'll notice I do this a lot comparing an aspect of Morpheus to Bitcoin. And it's one of those ones where it's like just because it's similar doesn't mean it's going to take off, go to the moon as one is said in the chat here. But I also do think that the more similarities a network has, it can actually, provides it's an indication of more sustainability

of something that will actually be useful and used. And this is why I still continue even though there's, you know, things in this project, which I don't understand, which I still kind of don't like, like, I don't like that I can't see the compute stats and how it's going easily.

That's and that's fine. That's one of those things that it's just worth monitoring and seeing in the future. And if it continues like that, then I'll start to get a big concern and be like, okay, what's actually going on here? Especially because 24% of the rewards go out to this bucket. But there are times where it's like, I have to tap out. I can't personally add more into this. And me trying to learn more about this would either take a ridiculous amount of time

of me spending some time that I don't actually enjoy. Like I don't I trolling through the GitHub and trying to understand Lumen model was not fun for me. This this week's episode was certainly I've made more memes this week because it kept me sane up to trying to actually understand this stuff. But I think that is is kind of my overview of of the compute bucket. And I've I've hope I hope that's helped provide some value to you, because it certainly

wasn't super valuable for me to, to try to dove into all of these things. So we're going to end this episode here today a little bit shorter than normal.

This is a value for value podcast For any of the value I provided today, I hope that you got some. I really hope you do. I just ask that you can provide that back. This channel itself, the mere Morpheus, it is run on the value for value. And in particular, I am very against advertising. I'm very against sponsorships. I'm very against any sort of form of monetization in particular that will affect my judgments of how I view this podcast of the network, the the information that I provide you.

So that does mean and I want this available anytime, anywhere for anyone. Anyhow, that does mean though, that I ask for some value back in in return for what I've given you here. You can do this in many different ways, time, talent and treasure. So sharing this with a friend's word-of-mouth, spreading the word of this podcast helps out just more eyes on it and more ears on it.

Gives me more feedback, it actually does make the show better, funnily enough, to self fulfilling prophecy and cycle, giving us a like providing feedback joining in on the live 11AM Australian Eastern Standard Time on a Monday like Juan did here. Talent. I've got graphics.

If you are a meme curator, if you want to help play a part in the show, produce the show with me, I'd love for you to to join in. Maybe you'd wanna touch music. Maybe you would have suggestions on how I can improve things, on how you could contribute on topics. Hell, even when I'm in the discord and asking questions, replying to those and giving me answers to that is very, very useful and valuable, much like Anton has and, and David of David Johnston

smart agents on on Discord have they've been very helpful, so thank you very much for to them. And then finally, you can send some more my way. I do have a link down in below of that referral link, which I'll keep referring to. There's also PayPal and I mentioned on fountain, you can send lightning

via that way or stream lightning payments of Bitcoin to me, and also send a message at the same time. So any of those ways are very much appreciated. If you do that, let me know. So I can shout you out in the episode as well. I will let you know the podcast does look, I'm I'm going to be traveling in May. So there's going to be a bit of an end date in anyhow, at least for the season one.

But if you know this, if I don't get any value back from this podcast, I'm going to assume that people don't find it valuable. And so I might not kick it off again afterwards. So there's one of those ones where it's it's a give and take here people like if you enjoyed it, if you want

to see more, I need to know that you're you're consuming it. There's nothing worse than being a podcaster and talking into the black box and it just going nowhere. And unless people let me know that they found it valuable via any of the methods that I just said, I'm just going to assume it's not valuable and so would end the podcast that way. So, what's coming up in future episodes? So I've got two more episodes on the coding and on the community builders.

I'll do a tokenomics one, and then I think I'll probably have a couple more. So something on the atomic governments, something on how there's no team, probably open source and then separating AI from crypto. And that'll probably just about wrap me up for this season one of of me and Morpheus. So tune in for all of those. Thank you very much for everyone who's helped to contribute and participate. And until next week, ciao for now. Karen out. Hi.