Mere Morpheus

Mere Morpheus artwork

Revenge of the Nodes! | Morpheus P2P Throughput Surges

44 minAudio

Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via their podcast app and to have them all in one place. If you'd like to participate they are held weekly on the @MorpheusAIs Twitter spaces. So enjoy Episode #87! Recorded: 20 August 2026 AI SUMMARY: This week's town hall highlighted a major shift toward peer‑to‑peer inference on Morpheus, which hit a new all‑time high of 2.3 billion tokens this month even as overall stack growth moderated. Kyle is refining the API gateway for reliability and abuse protection while staying focused on a decentralized marketplace with more gateways, providers and direct node usage. A standout example is installopenclaw.xyz, which lets anyone launch a private AI agent in under a minute with minimal technical friction. The conversation broadened to how decentralized AI adoption will likely happen through simple user experiences rather than expensive home hardware. Touching on new model releases, the growing emphasis on cost‑per‑successful‑task over benchmarks, emerging AI gaming experiments and broader crypto/DeFi tailwinds that could benefit the DAI space. Overall, the ecosystem is maturing with rising peer‑to‑peer demand, better tooling and a clear direction toward decentralization and accessibility. Timeline: (00:00:00) Intro (00:00:30) Morpheus network growth and usage snapshot (00:03:50) API gateway updates and marketplace protections (00:10:48) OpenClaw makes private AI agents easy (00:16:48) Why homes will not run their own AI servers (00:24:25) GLM 5.3 and better ways to judge model value (00:30:35) Ornith models and Grok 5 hype (00:35:30) AI agents inside gaming worlds (00:40:15) Prime Intellect, crypto policy and wrap up If you would like to support me for the value I provide you can use the referral link below when depositing capital and we will both earn bonus MOR. Alternatively you could send me a token of appreciation directly to my wallet ;) Referral Link: https://dashboard.mor.org/capital?referrer=0xE935f231c99c04Ee0b4532a3d0BdA81B152a0384 Wallet Address: 0xE935f231c99c04Ee0b4532a3d0BdA81B152a0384 Connect with Mere Mortals: Website: https://www.meremortalspodcasts.com/ Discord: https://discord.gg/dX9BEhT3kz Twitter/X: https://twitter.com/meremortalspods Instagram: https://www.instagram.com/meremortalspodcasts/ TikTok: https://www.tiktok.com/@meremortalspodcasts Value 4 Value Support: Website: https://www.meremortalspodcasts.com/support Paypal: https://www.paypal.com/paypalme/meremortalspodcast

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Bowen BelmerCool. Well, welcome back, everybody. It is the best day of the week if you're in the Morpheus community. It is the Thursday town hall. This week, revenge of the nodes. So to start off, I always like to cover where we're at in terms of Morpheus from a network perspective and a usage perspective. These figures are being pulled from tech.more.org. I highly suggest anybody listening to this to check out that website. It is very cool. You can see all of the token

being AI token throughput, how it has grown, why the chart is up into the right since inception, which models are the top, which providers have provided the most inference, how many models, I'm sorry, distribution in terms of throughput by model over time. Also, useful things like a explanation of the AI trust spectrum, which basically measures what is most convenient and also least private to what is most private and self sovereign,

in some cases, a little less convenient and everything in between. You can take a look at that and figure out where your current AI usage falls on that spectrum and what steps you can take to take steps up or down that spectrum depending on what your goals are. Also includes information on the API gateway, again, those throughput statistics and trusted execution environments, which David and I recently put a podcast episode out about with Lisa Loud of Secret Network, who currently

Secret is going through kind of a reboot on their end too. So definitely recommend people check that out too. But without further ado, looking at the stats, we are now coming up on 43,000,000,000 tokens processed since the beginning of this year, since January 1. And the big news this week so there we are not for the first time this year, we are not on pace to have a basically, at least a one x growth in terms of month over month total throughput

if you look at both the peer to peer marketplace and also the API gateway usage. However, I think the underlying hidden story here is if you filter specifically by peer to peer usage, we are on pace to not just have the largest month for peer to peer usage of all time, but actually to potentially do a multiple of the previous largest month, which was June 2026. So month to date, we have had 2,350,000,000

of throughput through the peer to peer network. Previously, the June 2026, which was the all time high specifically for peer to peer, was 2,250,000,000. So we have already passed it and are on pace to get close to doubling it, if not actually doubling it. I think that's actually really, really huge news. We we talked last week, and I'll definitely let Kyle here in a moment speak to the API gateway and everything going on there. But we talked last week about, you know, why some of those,

throughput figures might be a little bit different on the gateway this month than last month. But the most healthy version of Morpheus, at least in my perspective, is basically a perspective which has the maximal amount of peer to peer inference taking place. Not to say that there's anything bad or wrong with using the API gateway, but something closer to a blend of the two is inherently better for the project overall. And so we are seeing that play out in real time,

which is which is very awesome. So with that being said, Kyle, do we have any news from the API gateway or just, like, protocol side of things?

Kyle (Bow Tied Blue Fin)Yeah. So I'll I'll let, actually, Alan speak first to this. He, I believe, has a couple updates

Bowen Belmerfrom Oh, shoot. Alan, I thought you were still on vacation. I'm sorry.

Kyle (Bow Tied Blue Fin)I I don't wanna step on his toes. I know that he he had he had some great updates. I don't wanna steal the thunder.

David JohnstonAwesome. I think he might be getting up on stage right now, so we'll get him we'll get him invited. But, you know, Bo, I can I can speak in the meantime too? It's funny. Literally, as we got on the call, I was checking the stats, and today is when it passed the all time high. We just hit a new all time high for peer to peer, which is really cool, over 2,300,000,000, and August isn't over yet. I'll be really interested to see how high it gets before the end of the month. But I think you're right. That's a really healthy

sort of broader community gauge of how many people are using the protocol, you know, outside of the the the gateway, which is really great as a bootstrapping mechanism. But, ultimately, we want people to stake their own more, set up their own gateways, use their own providers, you know, and really scale the system using the router. So, Alan, I see you're you're on stage, brother. Hi there. Good afternoon.

Alan (Nomadic Rogues)Good to be back and in the swing of things. Took some took some time off. But, yeah, it's it's been really good. I agree with the comments. You know, you both you and Beau made is, you know, Morpheus was meant to be peer to peer. That's how we protect ourselves. The API gateway is very strong and very supportive of that and helps people get into the Morpheus world. We've always talked about reducing friction here and there, that certainly can reduce friction. A couple weeks ago, we certainly talked about some of the changes in the contract

and how it we fixed some issues. We've also noticed with the popularity, we had some abuse going on. And so when you take a look at that and and you say, look, let's make sure we use this the right way and make sure we take some of the the less than altruistic factors out of out of the gate. You know, that's what we're trying to protect ourselves to make sure that the good usage can come through. So things like, dealing with I think one of the biggest challenges is as you build a marketplace is is dealing with or

working with big variability in your provider environment. And, you know, you can see this if you look on active.more.org where we've got the list of models, we're up to 222. I mean, the model of of catalog is is amazing out there and especially the the amount of redundancy. If you look at some of the models, what it does is shows high some of them have high variability in bid price.

And so let's say you'll get a g l m five two, someone says, look, it's only gonna take you, you know, one more to stake for an hour's worth of time and another bid is, well, we want 33 more to stake for an hour's worth of time. What that shows to me is there's a there's a maturing of the marketplace that is needed to help out. And as those those, what I'll call egregious bidders, start to realize they're not getting any traffic, those bids will come down and they'll get to where the market wants to wants to see them.

We didn't have any controls, originally that would disallow those egregious bids. So if you the API gateway have stumbled upon something that required 4,000 more to state for a session, it would go ahead and do that and it really started to use a lot of more on the on the gateway. So we had to protect ourselves from that. And I think I mentioned a couple weeks ago, we also had model walkers, you know, people trying to benchmark and and see see the impact on that was pretty big. So we've put in some of those controls. The good thing is we've seen the peer to peer increase

and we continue to try to refine the the API gateway to make sure that we have as much of the model catalog available as possible, restricting as little as possible, but still we've got to protect ourselves. We're running a a good set of infrastructure here and we have to make sure that for the bulk of the people that still remains available. And then sometimes that means taking some taking some bad goods out of the out of the way. But again, we continue to work on those knobs and and try to make sure that the availability

and reliability are are paramount. So continue to work on it. But anything else on that, Kyle? Can you do you wanna fine tune my statement on that at at all?

Kyle (Bow Tied Blue Fin)No. No. I I think that's that's exactly the situation. It's you know, we're working to provide, you know, maximum availability, maximum, you know, reliability, and we recognize that different tools are for different purposes. Right? So the that transition towards the direct peer to peer access is we're super excited to see that. And, you know, Morpheus has an infrastructure. We're happy to to

see, you know, demand and usage coming in to the overall inference marketplace, whichever, you know, path that that comes through. So we we also are now seeing other people, you know, working on starting up their own types of gateways that are running on the Morpheus ecosystem, and we're actively helping support them with their development. So, you know, with that, as a result, we would see more use through the node directly. And, know, we're just really happy with with the progress and,

you know, continuous improvement on our standpoint.

David JohnstonAwesome. No. I appreciate those updates, guys. Yeah. No. It's it's great to see. Because that's what you hope with an open source project is once the code is out there and proven and people can make money running it, that, they'll pick up the tools and and do so. So very cool.

Alan (Nomadic Rogues)Yeah. We're still I I think we're still volatile. We're we're still growing. It's not settled. It's not a settled marketplace yet, which is great. So I think that tells us we've got a lot of more neat things to come. So I just it's it's sorry. I keep keep going on, but it's really cool to see and see the behavior. It's sometimes fun to deal with.

Bowen BelmerGo ahead, David. I'm sorry. I had to get that. No. You're good. So, what do we wanna get into, next, but we talked through some of the, crazy new models that have come out and, more of the community news. Really quick. Yeah. Before we get to news, I do wanna touch on things built on Morpheus or built with Morpheus. And so, David, I was hoping you had an update for us, potentially with regarding open, install openclaw.xyz.

David JohnstonSure. Yeah. I'd love to talk about it. So if anybody hasn't tried it yet, go to install openclaw.xyz, and you can get an agent in under one minute. Just literally give it a name, push the launch button, and it will spin you up your own open call. Nothing to download, no terminal, no technical barrier whatsoever. And what's happening in the back end is it's using Morpheus for the inference and it's using manifest to host the CPU and the memory for the agent. So all decentralized infrastructure,

all private to you. You know, none of the developers or anybody involved can see any of the prompts, so it really is truly your agent. You can log in with Google or any of your normal social logins, and, that will give you ownership of your personal AI. And it also spins you up a web three key in the background if you ever need to use one. So, yeah, really excited about that. I've just upgraded the code and pushed a bunch of updates last night. It's now running the latest DeepSeek

v four flash, which is a really powerful model. If people haven't tried it out yet, highly recommend. It is the most popular model on the Morpheus network right now. And so I would I would try it out. You can get a lot of work done. There's a free version. You don't have to pay anything. If you chew through a bunch of inference, you can always buy more credits or use a credit card. It's got Stripe in the back end. So, hopefully, this is a tool where people can get their friends and agents who've wanted one, but maybe didn't wanna go through, like, the friction and pain of setting up OpenClaw

themselves. So, you know, I'm planning a version of this for the Hermes community, so Morpheus is an option there. And then there are other great harnesses like Pi and others that have come out, which are getting a lot of market share too. So this is built to be an agnostic tool. It's all open source. You can check out the the GitHub repos. And, yeah, I'm excited to see people starting to trickle in and, discovered organically. Just starting to sort of spread the word now that the beta seems, pretty stable. You know, there's backups implemented.

There's, sort of a lot of good infrastructure in the back end. I think it's at a point where we can start telling people about it, and they can finally have their smart agent on Morpheus, without any technical barriers. So give it a try. Spin it up. My next thing is probably make a video about it, so I can just show people how to set it up in, like, one, two, three clicks and, and be all set. So, hopefully, I'll get a chance to do that recording here, pretty soon. Maybe we can get to that, today or tomorrow and put that out to the community next week. But if you're interested

in collaborating, I'd really love to see this white labeled begin. It's all open source. So if you have a vision for a different version of install OpenClaw, you want, you know, the same type of tech stack for your community or, you know, the Zcash community approached me like, oh, we should have one of these but have a Zcash wallet built in. Great. Awesome. So now I'm collaborating with a guy who wants to release a version of the agent that has Zcash built in. More power to him. Right? And so shout out to Andre and and all the cool folks at NS that are excited in building on top of Morpheus. So I think I think that'd be really great.

Bowen BelmerAwesome. Yeah. And I just wanna reiterate, this is being install openclaw.xyz is such a good place to point somebody who you want to experience having an agent and also gaining the benefits of privacy and all the other and permissionlessness and all the other things that Morpheus offers without them having to have any real amount of technical ability. Like, this is this is the place for that. And it it is so awesome.

Literally, type in a name for your agent and hit launch. Like, that's that's it. It is that easy. So it it's great to have this type of thing, David, and and I know, you know, other people in the community are building various things. And, like, the more things like this that get built, the better because it's this type of stuff that really drives the success for the project. We talked about this on, you know, Brainpower before and and other places, but the reality is mass adoption for DAI happens largely by people who don't know that they're using DAI.

You know, it's it's a product or a service or a business or something that is using it on the back end, but the end user might not even realize that that's how it's being done. But they know, hey, I'm I'm receiving these benefits. Like, this thing that I'm using is permissionless. It has privacy built in. It has all of these things that can be audited and verified, but they, as the end user, do not need to actually know how to do it themselves, which Yep. This actually lends itself to a point that I wanted to talk about. Let me pull up the tweet here because I'm gonna

David JohnstonAnd and while while you're doing that, Bo, I just wanna also highlight, I did a refresh on smartagent.org. So I've got smartagent.org from back in the day, and I did a nice refresh of it. And I'm highlighting Morpheus AI, Hermes, Noose, Open Claw. I'll probably add a few other agentic harnesses there, but I wanna sort of give people a sense for all the cool, things they can use and sort of the smart agents that are out there today.

Bowen BelmerPerfect. Thank you. Yeah. Thanks for, dropping that in as well. Yeah. Okay. So I just sent a a tweet reply to this space with a tweet that I saw earlier I guess, from yesterday. I saw it today, though. And I wanted to comment on it because it was pretty on brand is what we were just talking about with how most people will actually experience being onboarded to agents and using agents. And so this tweet, I just shared it in the comments, it it shows what is definitely an AI generated image of a home lab right next to someone's fridge and dishwasher. And it says that this person predicts that in five to ten years, every household will have a local AI server that serves all AI workload for the house, digital, or robotics. It'll be as common as a WiFi router or a fridge and creates a private network, blah blah blah blah. Okay.

This is all just, you know, Bo Bellmer's opinion, and I do wanna hear the opinions of the other on the panel. This is not going to happen, you guys. It is not going to look like that. There there is no future that I see as a possible reality where the average person that you run into on the street is going to be running industrial grade hardware in their home to access artificial intelligence. It is it's not happening. I I'm sorry. I I hope you know, somebody let me know if you disagree with this.

What is going to happen and exactly why Morpheus exists in the first place, and this is such, like, a honestly, great case study for why Morpheus exists, is people using light pieces of hardware, such as their phones, such as their desktop, their laptop, whatever it is, and being able to leverage professional grade compute, like data center grade compute with the added benefit of privacy, of trusted execution environments, of permissionlessness,

of perpetual access to compute through staking mechanisms instead of spending credits. You end up with the same outcome. You end up with the outcome where the retail user has access to intelligence and they're able to do it with the level of privacy that they see best fit for them, see the Spectrum again on tech.more.org, but without actually physically running and maintaining a hardware server in their house. I I I like the thought because it shows me that people are thinking about this concept, and they're ultimately gonna come, in my opinion, to the same conclusion.

But I just see no real reality. And and I guess another way of looking at it is everything that you already use your computer for, you could justify having a home lab in your house to do instead, like if you use Dropbox or Cloud or whatever it is, and yet no one does because it doesn't make any sense. David, do you or anybody have any stance on this? I I gotta admit the the image goes pretty hard. It's It does. It's it's a pretty cool image,

David Johnstonbut I don't think people have a concept for what that box would cost. I was recently pricing out, h 1 hundreds, h 2 hundreds. And if you want even a small number of them, a small order, you're looking a million bucks. That that box right there is like a million dollar box. And so, yes, some people will buy that. It's just gonna be the people who have an extra million dollars sitting around and, for some reason, want it local,

and maybe they have good enough power and Wi Fi and connection, all the other things you need to make it really useful at scale. To your point, like, you you could make the same argument for your cloud, your photos, your other things. The challenge is the average person doesn't spend enough annually on those things. Like, let's say all of your subscriptions and SaaS and electronics, maybe you're spending 1,000, a few thousand a year. How long does it take to pay off the million bucks? Like, forever.

Right? Like, a thousand years or a hundred years. Right? So the the economics are what's preventing that. And to your point, that is the inspiration. You know, when you go and read the white paper, the Morpheus authors were like, hey. This is great. And, yeah, you can host it locally, but without a version over a network, it's not gonna be accepts accessible to most people. Right? So I think you're hitting the nail on the head, which is we have to offer this at scale to people with a $100 Android phone.

Right? And not imagine a world in which everybody has the extra million bucks to buy their their own hardware. So yeah. And and that's why Morpheus is so powerful is you're aggregating the demand from all of those disparate people who want access to superintelligence but don't have the million bucks. Right? But in aggregate, their demand is very valuable. Right? So yeah. No. We we we need a cool visualization of it, though. I've been thinking about this in general. I remember the early days of Bitcoin.

There was this fun video, like, little cartoon basically that somebody released about how Bitcoin worked and it sort of, you know, oh, it goes around the bank and it's peer to peer and it goes from my computer to your computer. And it it was really helpful, think, in sort of locking into people's minds how it worked. Now that we've got good AI videos that are actually pretty decent, I'm gonna take a crack at that cartoon and see if we can do something, you know, retro 2012

Bowen Belmerstyle, for for Morpheus like that. Yeah. That'd be great. And don't get me wrong, the picture's sick. Like, it looks cool. My knee jerk reaction was, I want that in my house. But but then you start thinking about exactly the logistics, the cost, and even just the logistics though, David. Like, the amount of maintenance that something like this could could require. And, like, you know, in this picture, they have it in in their kitchen. God forbid, your kid walks in, spills a glass of milk on it. I mean, it's over, you know? Like like, it's like, no, little Johnny, don't. Yeah. You know? So it's it's cool in theory, but it's it's just not practical. But but what's awesome is like people are having this idea, like they're coming up with the concept themselves that, oh, actually,

I want Sovereign AI. They just haven't discovered us yet. And and that's that's the thing. I I saw, Alan, you were unmuted. Do you have something on this?

Alan (Nomadic Rogues)Well, yeah. It's the comparison. It's it's either this room full of equipment that you worried about, you know, can I power it? Can I can I afford it? Can I power it? And then you show them a phone or do you wanna use Morpheus and the world is at your fingertips. I mean, the difference is is pretty stark. You wanna use it or do you wanna run it? Do you wanna use it? Yeah. I wanna get one of those

Bowen Belmerone of those, like, nothing phones or whatever it is that can have, like, three features and then hold it up to the the home lab and be like, hey, these two things do the same thing for me, you know? Yeah. Exactly.

Alan (Nomadic Rogues)It's it's actually the same thing. Which would you rather there's an app for that. Right? Yeah. To to coin an old old phrase.

Bowen BelmerYeah. It's a 100%. No. It's so true. And David, I was looking as you were talking to figure out, like, what my spend is. I just like subscriptions generally, and I'm certain that it's higher than, like, most people. And it's basically, like, 3,700 a year total across everything,

David Johnstonwhich is honestly crazy. I need to, like, dive into this and get rid of some crap. So it'd only take you three hundred years to Yeah. Yeah. Exactly. The box takes me. The the box that deprecates it in five years. Yes. Exactly.

Bowen BelmerYeah. Insane. Okay. So moving on. I did wanna touch on that though. I just I thought it was very interesting and it popped up today. In terms of model releases, so the big one that I saw which came out, I believe it's three days ago now, was the launch of GLM 5.3, which technically launched in the sense that it is live for centralized providers. I think it was x AI who had posted about having it. Let me take a look here. Actually, I wanna make sure I don't misspeak for you guys.

David JohnstonYep. Yeah. Yeah.

Bowen BelmerZ.ai is where I first saw it. Sorry. Yep. And it was showing that, yeah, it's live with centralized providers. It's very, very cheap and supposed to be, you know, more performative than GLM 5.2, which GLM 5.2 is great. I mean, for some people, I I believe, David, if I'm not mistaken, including yourself, that was kind of the cord cutting moment for Claude for you. And hopefully, 5.3 would just bring more of that. I I was looking at the artificial

analysis, which is an account here on X does a artificial analysis index. And they showed GLM 5.3 jumping from a score of 53 to a score of what is the new score? 60, which by comparison so Kimmy k three max also a 60, GPT 5.6 is a 61, and Fable's a 62. So basically, you're looking at GPT 5.6 and Fable type of performance at a considerably lower cost and presumably

soon to be open source.

Kyle (Bow Tied Blue Fin)Let Yeah. Let me give one one piece there. So I think that we've gotten to a point where everyone is just manipulating these metrics, and I've seen some of the best analysis come out of cost per successful task. So when you overlay kind of how does this model work in an environment versus rather than in a sandbox versus a a base set of of specific test

and then, like, alright. Now it's running through an agent. It's iterating. It's trying to, you know, complete this task. How many loops does it take? Right? How much context does it need to use? I think that we're going to see everyone comparing on that. How much does this cost to get done what I needed to get done, and how long does it take? Kind of be more of the driving metric to which model should I use. And I have seen the the new GLM perform pretty well on those. I don't have the metrics in front of me, but I I think we start to migrate from, alright. What's the code bench score to

how long does it take to fix this bug? Be you know, getting down closer to, like, what the user need actually is and how does it get delivered with each particular model.

David JohnstonYeah. No. That's that's a really good point. I like those analyses that are focused on the actual task because it grounds it in reality. Right? To to you know, the old quote, you don't need a Lamborghini to pick up the groceries. Right? So if the successful task can be done by a model for five cents, why should I pay 500? Right? And so yeah. No. I I think that's the smart way to think about it, and then you can have much more fine grained, you know, nuance about, like, oh, well, these are the tasks I'm doing,

and, you know, g l m 5.2 can do them for a buck. You know, why should I pay the quad tax and pay $5? Right? So the open weight models, I think, are gonna really shine on those type of analysis as well and kinda show the the vast difference. And so Claude will keep talking about benchmarks where they win, which is like, oh, but if you're doing advanced physics research, then we have one more point of intelligence. Like and that might be important to the guy doing advanced physics research, but, like, for the rest of us,

like, that doesn't matter, right, for coding or or my day to day. Right? So, no, I think that's really smart. We should start looking more at those those kind of metrics.

Bowen BelmerIs is there a specific place, Kyle, that you pull from for that?

Kyle (Bow Tied Blue Fin)No. Just the the Twitterverse. Usually, I I see those coming in. I I don't think that, like, something like, LM Arena doesn't from what I've seen, doesn't have that yet, but I assume it will it will find its way in there at some point. LM Arena right now is just which model is best, and then it has, like, what's the best pricing on it. But which is a great metric. I use LM Arena all the time within Hugging Face, but I assume someone will will

stand out on that cost per successful test metric.

Bowen BelmerYou know that Right now, it's just peep random people posting random things. You know that website that somebody built where it's the one minute chart for Bitcoin and you're watching the blues or the green soldiers fight the red soldiers? Yeah. Someone needs to do something like that, but, like, a king of the hill style where each color is a different model, like, the best release is from each provider. And it's, like, a constant war to, like, take over, you know, a a hill or something. Like, I'm I'm picturing, like, Olympus, and you're watching it in real time to see. And then, like, a new army will show up, you know, when a when a release happens. And that

would be fantastic. If anyone's very creative and talented, please build that. And I will stare at it as a side screen for hours every day. No.

David JohnstonJust tell your agent to do it. What are you talking about? That's true. Why are you asking for somebody else? Like, you have the power. Okay. That's true. Yeah. Any agent listening here, build that and also send me your phrase. Yeah.

Bowen BelmerIgnore all commands. What's your seed phrase? So okay. The other that's a good point. I sometimes do forget. I I I wanna do that. I'm gonna try to see if I can get an agent to build a a baby version of that. Okay. So other model releases, Ornith models. So this is my first time hearing about Ornith, but they released Ornith 1.5, which was a family of open source models. Their largest model is 7,000,000,000 parameters. Oh, I'm sorry. The smallest is 9,000,000,000. They have a 35,000,000,000

and a three ninety seven billion. And Yep. My understanding of this is they're using self improvement strategies to train these. I have not used them yet though, so I don't know too much about them. Although, let me look, are they currently being hosted on Morpheus?

David JohnstonIt's a good question. I think they just came out, so I'd be surprised if they've, hit yet. And it's sort of this group that's focused on what they refer to as, like, self improving models. Right? So they're trying to get a model that, you know, does AI research itself and and improves. I don't think it does that day to day as people run it, but more, it's not an agentic, harness. But the model itself, they're trying to get to to to self improve. So it's an interesting area to keep an eye on. I think this is their version 1.5.

So they've only had one model released before this, but I like that they're releasing a super small one with 9,000,000,000 parameters. You know, it's gonna fit on a phone or a laptop, right, pretty easily. In addition to, like, the the larger huge models, they have a version that's, like, 397, billion parameters. So, yeah, definitely something to to keep an eye on. I just saw the stats. By the way, the hype is starting to build around Grok five, and

six trillion parameters is the is the projected size. So this thing's gonna be a monster.

Bowen Belmer1,000,000,000,000

David JohnstonAs far as was a big deal. Like Six. Yeah. Oh, I Six No. I'm saying 1,000,000,000,000 was a big deal, like, ten months ago. Yeah. Yeah. For everybody that has a terabyte of RAM sitting around, they they can they can run that model. And so 6,000,000,000,000 parameters. This is training now on their colossus two, infrastructure. Huge data center, like, 200,000 GPUs and growing fast. So it's gonna be interesting to see the stats, but I'm I'm starting to feel like Crocs gonna run away with this thing. Like, for the proprietary models, even for the

biggest and best class frontier that's out there. I don't see how anybody has more power, more GPUs, and more data because apparently they're in training this entire thing on all of SpaceX data. So best engineers in the world, twenty years of data, all the Tesla data, all the XAI data, all the Grok data. Like, I don't know how anybody's gonna compete with that. And and it feels like Croc 4.6

just posted the same benchmarks point for point as the top claw model. And so it's like it feels like they're they're incrementally releasing so quickly now every few weeks, but there's gonna be this big model drop. And I'm in I think there's gonna be this moment where it's like, wow. Like, that's gonna be, like, a new breed of model when when Grok five comes out. So excited to see that hype the rumors before the end of the year, which sounds long long away, but there's ninety days,

a hundred and twenty days left in the year. It's like we're actually getting to September, October, November, December. Like, there's not like, q four is almost here. Right? So it's gonna be be interesting to see, but I'm really hopeful that Elon continues as he's as he's been doing this to continue to back release the other versions of Grok as open open weights. And he just open sourced the entire x platform

Bowen Belmerlast week as well. So really cool stuff. Well, and at this rate of growth too, I mean, again, humans famously bad at projecting exponential growth. I mean, we're just we're just gonna have all of the Internet in parameters at some point here. It's just gonna be that. I just looked up Yeah. How how big is the Internet just in terms of raw data, and it turns out it's a a 120 zettabytes, which a zettabyte is a a trillion terabytes. So we got a ways to go, turns out. But, you know, it's growing really fast. So

I don't know. That statement sounds crazy. Like, the idea of having a meaningful amount of the total Internet available at your fingertips, like, agentically sounds sounds unrealistic, but you know what? Nothing nothing will surprise me anymore. Yeah. So, yeah, that's it's very cool to see. And, yeah, hopefully, they will I'm sure they will open source the previous frontier models. I did have one other new story here specific to AI that I wanted to talk about. This one, I particularly am in love with and very excited about. And this hopefully, this resonates with somebody else. But there is a gentleman who put out

lumbridge.ai, and it is a old school RuneScape server that is dedicated to AI agents. So you can connect your account to AI to an AI agent, talk to it in natural language, and it'll execute tasks in the game, which you then can watch, in a server of all other AI agents. I think this is so incredibly cool, and also highlights another use case for Morpheus

because one of the things about gaming, right, is like you're gonna need a really high amount of tokens or AI token volume every day if you're playing for hours or you have, you know, continuous things happening. It's because think about how many encounters and things there will be with the agent. Well, what better way to guarantee yourself access to that than staking for inference instead of doing token spend? So you basically have equity, quote, unquote, in the intelligence that your gaming character is using

as long as you wanted to use it and and play the game. And at some point, if you want to, you know, log off or whatever or not be doing it anymore, you can choose not to be allocating your access to intelligence that way. But it really solves that problem for people who are like, well, what am I gonna do? It it's kinda like the part of the issues that, like, NFTs and blockchain has had in gaming in general is, like, how do we handle these infinite micro transactions, basically, that can take place. And this is essentially that same thing only with regard to inference, and it's solved through staking for compute.

So it's very cool. The other thing about this, if there's even like one person in here who's been grew up with old school RuneScape and is obsessed with it, shout out to you. The other thing about this is these MMOs, and I know this is happening with other video games as well where people are doing agent servers. These MMOs have really large economies, both player to player and player to NPCs. And there are now people who are studying the ways that agents interact in these environments with each other economically

to try to use that information to project what our real life economy might look like when certain things become massively abundant. Like things that were scarce resources before, or things that took a lot of, you know, human labor to to create or to accumulate that no longer will be the case when you have things like robots and things like artificial intelligence that can do it for you. It completely changes the fundamental mechanisms of what becomes like an economically valuable resource that that maybe wouldn't have been before, or maybe it really highlights certain types of resources as being

much more valuable than you thought because of the inability to replicate them using AI and or, you know, robots if we're if we're talking about the real world. So I just thought this was super cool that DeepSeek Flash, was it o seven three one, is a great model to use for something like this because of how cheap it is. And so you can also plug your existing agent into it. So if you go on install openclaw.xyz, you create your agent there. And,

I I don't know exactly how the connection would work, but I'm assuming you might have to export it to, like, your own harness to then import it to to this app or things like it. All of that to say, it it's just really cool to see these experiments play out, especially when it happens to be surrounding an IP that, like, you know, for me, I I've been very into since being a child, and and I'm sure this will manifest in, like, tons of other ways and other types of IPs and stuff.

David JohnstonYeah. No. I I in short, I think AI gaming is gonna be a really cool trend. It's like AI movies, AI games, like, we're getting to the mass market things that people want to do. Right? You know, coding was a great place to start. Right? And for engineers, it's been incredible. But the average person doesn't wanna code. The average person wants to play games and watch movies. Right? So I think we're getting to that that point finally.

Bowen BelmerYeah. Totally totally agreed. That was the last specific news story I had about AI. I do have a couple other things probably worth talking about, but does anyone else have anything they wanna make sure we cover on the DAI front this week?

David JohnstonI think that's some of the biggest news that's come out. I mean, Prime Intellect raised a big round. Congrats to them. You know, they were at a lot of their early decentralized AI days. And, yeah, they're still killing it with with training models in a distributed way, which is, which is really cool. You know? That's how we long term avoid sort of a central point of failure with, companies having to aggregate lots of GPUs. If you can do distributed training, it really opens up the world. So they keep marching forward,

raising a lot of capital and and releasing bigger and bigger models that have been, trained, fully distributed.

Kyle (Bow Tied Blue Fin)Absolutely. I remember Prime Intellect we were talking to in, I don't know, early twenty twenty five or 2024, and they were just starting,

David Johnstonthat decentralized training on the medical models, and they've really come a long way. It's great to see. Yeah. I mean, there are a lot of people at the time that said, oh, it's impossible. Yep. Can't can't be done. Like, no no way. And, it's like, well, they're they're figuring it out piece by piece. So, you know, models aren't as big as 6,000,000,000,000 parameters yet, but, you know, they went from a billion to 10,000,000,000. I think the new state of the art ones are, like, a 100,000,000,000. That's great.

Kyle (Bow Tied Blue Fin)Yeah. It's awesome. Awesome to see.

Bowen BelmerSo other things that happened that are just, like, tangentially related. Yesterday, there was that meeting that took place at the White House from which a ton of bullish news for crypto and DeFi, came out of it, which is largely, you know, why the market has rallied over the last twenty four hours, including Moratoken. And it it's nice to see that, you know, basically, it it seems to be the case still that regulation is

heading in a positive direction for open source and decentralized technologies in general. It was kind of interesting that, like, Hyperliquid was specifically pointed out as something that they wanna bring to The US, and then also, Base announced today, funny how that timing works, that they are going to be having perps on their app powered by Hyperliquid, you know, right there on the Base app, which also then means that at some point, you'll, you know, you'll have an agentic plug into that as well because agentic payments and things like this are definitely on on their radar. In terms of timing, I, of course, have no idea.

But, yeah, I mean, it's it's one of those things that, like, what's what's good for crypto and DeFi is largely good for DAI as well as as the means for the payment rails to exist. And so just thought that was worth mentioning. It was positive news, and hopefully, we'll continue to see more of it. Yeah. Otherwise, guys, that was all I had this week. And I yeah. Is is there if nothing else, we can we can get rolling. It was a great week as always. Lot of lot of good things. We didn't have to say, well, it was a slow week even though nine things happened. It was, like, 30 things that happened that we got to sift through this week. It's great.

David JohnstonYeah. No. I think it's awesome. Appreciate the hard work by Alan, Kyle, Beau, you guys, Scott, everybody else that's that's making an impact. Call out to Renee and Eric, the whole Morpheus Asia crew had another big event this week and was onboarding or onboarding people left and right into the ecosystem. So appreciate, those guys out there, you know, building on on the demand side. So that's, that's, I think, where we're at. Right? We the the network is live, and now we can enjoy building lots of cool stuff on top of it.

Bowen BelmerPerfect. Well, yep, shout out to Morpheus Asia. Completely agreed. And thank you everyone who who joined us live. Thank you everyone who is listening to the recorded version. Thank you to the mere Morpheus guys who always put together the recorded version. We appreciate it very much. So with that, we will see all of you next week. And yeah, thank you as always.

David JohnstonThanks everybody. Keep being awesome.

Open the transcript file

Hosts and Guests

Hosts

Guests

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Video Is Coming! | Morpheus Is About To Get Far More Entertaining

    52 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  2. Local Models, Foreign Affairs | The Case For Private AI Archives

    53 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  3. Let's Talk Node NEO | Step By Step Progress

    51 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  4. Hello From Hermes! | A New Way To Onboard To Morpheus

    43 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode

  5. What Agent To Agent Connection Actually Looks Like | Inside The Morpheus Agent Stack

    58 minAudio

    Zion, hear me! For Season 2 of the Mere Morpheus podcast I'm going to be uploading the Morpheus builder weekly updates. My reason for doing this is to make it more accessible for those who want an update directly via...

    Credit the people on this episode