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 #91! Recorded: 17 September 2026 AI SUMMARY: In this week's town hall Kyle Chris and David joined to unpack a big week for the network. Morpheus surpassed 50 billion tokens processed and native peer-to-peer throughput now overtakes the API gateway. New tooling makes direct node access easier including a proxy-style front-end service giving users a familiar API key and base URL experience while preserving the decentralised permissionless design. Agent-friendly documentation and easier developer onboarding explain why native usage keeps accelerating and what that says about the move towards open user-controlled AI. The launch of MorDiem and the new MDM token model positions it as a fixed-compute unit for the Morpheus ecosystem with relevance for everyday users and larger institutions. Agent workflows recursive self-improvement AI firewalls and new model architectures like JEV aim to make automation dramatically faster and cheaper. Decentralised AI is maturing quickly the tooling is getting easier to use and open models agent workflows and permissionless compute are unlocking a powerful next phase. https://www.youtube.com/watch?v=WYWaFIaPFcg Timeline: (00:00:00) Intro (00:00:32) Town hall opens with record throughput milestones (00:03:24) New consumer node proxy simplifies direct access (00:06:29) Why native usage is rising with agent-first tools (00:13:20) MorDiem launches as fixed compute for Morpheus (00:24:30) IPFS, model resilience, and self-improving agents (00:28:53) Workflow wins, dashboards and agent integrations (00:33:37) AI firewalls, bug bounty overload and attention (00:40:02) JEV system one models and faster agent actions 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 BelmerLive. It is 09/17/2026, and it is the Morpheus community town hall call. We have myself, David, blue tied. I'm sorry. Bowtied Bluefin, aka Kyle, joining us this week. A little bit of a lighter cast this week. A lot of people are busy, but always working on all things Morpheus. Oh, and a surprise guest who joined after we already went live. How dare you? No. I'm just kidding. Chris, what's going on? How are you? What is up? I saw the new link and I just decided to pop in. Yeah. No. Please do. We we just jumped on. We haven't even really kicked off,
the official stories yet. But so a lot to cover this week. I guess, first and foremost, as always, let's just take a look at throughput stats. So let me share my screen here. We want this one. We officially crossed 50,000,000,000 tokens processed. We're nearing 52,000,000,000 already, which is awesome. I mean, we talked about this last week, but growth just continues to, you know, to happen into compound.
But the most important thing is looking at native only continuing to be up into the right, which is amazing. And we're on the precipice of already eclipsing August, and we're only halfway through the month. So September is teeing up to be the largest amount of throughput ever through Morpheus network natively through the compute router, meaning tokens not pass through the API gateway. Not that there's anything wrong with tokens pass through the API gateway. It's just the native throughput helps show,
kind of organic growth of of people tapping directly into the network, you know, beyond what is currently the largest subnet.
David JohnstonGuys, any thoughts on this? I mean, this is this is exciting. It's pretty cool to see. It's nice to see the shift. If you go to, like, the last thirty days, it's pretty pronounced on the share. Up until, I think, last month, the majority was going through the gateway, and we saw peer to peer continue to grow. And now gateway is less than 50%. You see it there. It's at 39% going through the gateway. So the peer to peer network has definitely overtaken for the majority of volumes, which is awesome
to see. So yeah. I mean, I've been getting great results. Now that we've got so many providers on, the network, you know, I'm easily able to direct my agents and get great performance and fall back between multiple providers if there's an error or anything like that. So, yeah, so far so good. I'd love to see I think at the current rate, it's gonna hit about 6,000,000,000, maybe 7,000,000,000 tokens. But if it continues, you see over the last week or two, it's been accelerating. If that continues to grow,
maybe we get to 8,000,000,000. I'd love to see another 100% monthly growth off the the network. Because I think that's sort of a good signal that people are getting traction with it.
Kyle (Bow Tied Blue Fin)Yeah. Absolutely. I I think it's a great signal. And we've also, you know, seeing the interest in the use of the node directly. We're similarly investing on the tech development side to kind of continue to iterate there and make that easier. So just this week, we, you know, you could probably see in some of the GitHub's, we launched another, let's call it a a not a light client, but a front end service that will make using it,
using the node much more simple. So, I'm gonna explain it a bit. It's not launched and live and packaged and, like, fully ready for everyone to jump into, but it's something that has been developed, has been proven, and and we're just in the process of kind of buttoning it up and figuring out the best way to get it out to everybody. But more or less, this, product that we that we're releasing is, kind of a a parallel with the with the actual node. So when you wanna enter the inference marketplace,
you need to use what we call a consumer node to get in there. Right? You need some piece of of software that's able to talk to the smart contracts, that's able to connect with the providers, that's able to funnel that inference through the system. From a tech perspective, it's very straightforward. But from a user perspective, sometimes it's it it could be difficult to interact with that if you've never done it before. Right? I'm a new user. I know nothing. I don't understand sessions. I don't understand,
you know, bids. I don't know how to interact with the system. And what we did at first is we launched the API gateway. Right? We said, let let's handle this for you. Right? We know that you don't wanna deal with this. So we'll provide this front end layer for you to go grab an API key, and you grab a the base URL, and you kinda just funnel through API gateway. And seeing that demand transition over to, no. We do wanna use this direct. Like, we're willing to kinda get down and dirty a little bit more. So we said, okay. Let let's migrate a little bit and and meet these users where they are. So let's let's give these users a a service that they can host alongside that consumer node that does some of those same things. Right? I'm a user. I want
to plug in AI into my applications or workflows or agents. And what does that need? Needs a base URL, needs an API key, and needs access to inference. So this service essentially puts that in front of the node, and and all it is is a proxy. So it it it's able to you know, you request a model. It's able to create that model in the consumer header just like you would manually and just like you would through the API. So it's kind of just simplifying and and making it a little bit easier to handle and kind of allowing you to connect the applications that you want to connect without going through this, you know,
single single service of the API. It kind of gives you that decentralization, gives you that permissionlessness, and and allows you to use the system more closer to as it was designed or exactly as it was designed, really. What would you guys say that the the the number up into the right, you know, of native usage
Unknownis indicative of the barriers kinda coming down and also that people are able to use agents to kinda get stuff, direct access to compute. And so it's like the you know, before where you you look at the user interface, you're trying to click through and you're trying to, like, generate manually generate your API and all that. Now it's a function of, you know, as these nodes pop up, you can just tell your agent, hey. This you know, whether you're running a coordinator agent or you're running, you know, whichever agent wherever your structure is, you can just say, hey, I need this resolved. Here's the doc set that exists
on the you know, within the protocol.
Bowen BelmerGo do it. Is that I I would think so. Because I was thinking, like, I can launch an agent on install OpenClaw like we showed last week. I could give it a ETH wallet, have more in that wallet, and just tell it, hey. Here's the Gitbook for Morpheus. Connect me to the node.
Kyle (Bow Tied Blue Fin)And and presumably, you could just do it in natural language. I mean, I'm pretty sure at this point. Yeah. So I I would say it's it's that, Bo, but I would say it's probably four things together. Right? And what you just said doesn't happen alone. Right? So let's say, on one side of it is simplifying the tools and simplifying the access. So we know that Alan talked about it a couple weeks ago, making those doc sites agent friendly. Right? So making it so that they can actually get the information that they need. I think the second part is
UnknownHow how common is that? Like, among other projects that are out there, you know, really dialed in really dialed in agent first doc sets. And do you see that kind of growing over time?
Kyle (Bow Tied Blue Fin)Definitely growing over time. I would say it's not fully common. It's not very common. But in some AI pages, they'll be like a /agents or or /robotdocs.txt. I I would say some native AI platforms have that, like, you know, if they're using Mintlify or or one of those services for the API, they'll have that, like, you know, read as MD. But it's it's growing slowly, and I think that a lot of the agent harnesses recognize that and look for that page if it exists.
So, like, the the we know context is king. Right? So, like, the more context you can give to that AI agent, the higher likelihood that it's gonna get it right on the first shot. So I think it's a mixture of having the information available plus a mixture of, kind of, right, what Bo said. Right? If we asked Bo a year ago, you know, are you running any, you know, agent harnesses that that are able to go? The answer was no. Right? So it's kind of the adoption of people will sorry. That's right.
The it's like the adoption of of people's willingness to use the tools to get the information as well. So it's kind of information, ability to access information, and then the overall user need of of people recognizing, I actually want this product. Like, I don't want, you know, Sam Altman and Dario to, like, have full control of of my intelligence. So it's like the the overall market move towards open source plus the open source, you know, challenging the frontier on an intelligence level,
plus people's adoptions of agents, plus be there in the right place and have the information so that the agents can actually do it. David, you're on mute. Oh, you're on mute, David.
David JohnstonIt's worth pointing out, like, the clods of the world have zero incentive for you to connect your agent to their system. Right? It breaks their model because they built a gym, and they want people to show up and do prompts one at a time. And the idea that your agent is gonna churn through their entire subsidy and charge them $2,000 in inference for your $200 subscription, they actively delist,
block, and remove people that tried to connect OpenClaw or Hermes or other stuff like that. If you remember that raft of emails everybody was getting like, you violated our terms of service. Goodbye. Don't use our service anymore. Right? So they're fundamentally conflicted about giving people a good experience with their agent because it doesn't fit their model. Which to your point, Kyle, is why the shift towards open weights is winning is because they're the ones making it easy for agents because they're not threatened by that model.
Bowen BelmerYeah. You were and and you're exactly right too about just the the ease of use for people, like being able to onboard to it. Because, I mean, yeah, a year ago, I absolutely was not using an agent harness. No way. And then even, like, when they came out, I would say I jumped on OpenClaw pretty quick, but I didn't make the switch to Hermes until they made it so easy that I could just download the client and ask it, you know, port everything over for me, and it just did it. You know? Like, there wasn't really, like, a technical barrier there.
Kyle (Bow Tied Blue Fin)And that's not a shot at at No. No. No. I know. It's it's just that's, like, where where do you live? Right? We all live at a certain place on the adoption curve of of tech. And, you know, as technology and as time increases, right, We all are are or the adoption curve is kind of flowing through the chart and where we sit on it is is gonna be, you know, how how many people It's it's okay, Bon. I I
UnknownI'll I'll go ahead and take the shot at you. He's not meaning it as a shot, but I'm I'm meaning it as a shot against you. Hey. Listen. My my position here is just ahead of everybody else. By the time it hits me, you're reaching critical mass, and then it's falling off the cliff. You're you're taking the approach of parenting where you just need to be one day ahead. You know, you only need to be, like, a week or or or, you know, half a year ahead of the kid. Yeah. Yeah. Exactly. Puts your price on it. Yeah. Yeah. Exactly right. No. Seriously, though, it is It's all good, Bo. If the if this is the all in podcast, you can be RJ. It's true. Fine. I'll take it. Whatever. I Hey. He just wants the he just he just wants the zeros that go along with it. Man in the arena. Bring it on, guys. Come on. I I I believe I believe he said something about being a whale yesterday.
Bowen BelmerOh, no. That was that was MDM. I said I'm gonna be an MDM whale. Which, by the way, we should talk about that, and and it was already mentioned in chat. That's a good good segue to more DMs, new site, and token going live. Let me share that and we can take a look at it. Let me just make sure I'm logged out here. Cool. I was saying yesterday, I think about this as like stable coins for AI.
David JohnstonAnd, you know, so cool to see Ryan adapting this for the Morpheus ecosystem and basically, you know, bringing that sort of $1 of inference, to the the Morpheus community. You know, Venice did a great job of pioneering, this this approach, but I think we're gonna see a lot of these. And I really love the way he's leveraging yield to create the tokens to drive the demand. It kinda for for the crypto folks out there, it reminds me of, like, that that DAI Yeah. Stablecoin kind of model. Yeah. This is, like, the maker well, first of all, David, you you were the first one yesterday when we were talking that,
Bowen Belmermade the connection for me in my brain that Diem as a concept is just stablecoins for AI compute like, for AI agents. And I love that analogy, especially because and we were saying, like, the term Diem can and should, in my opinion, just become obviously, it started with Venice, so shout out to them with the naming convention and and their token model. But, it should just become the Kleenex of instead of Stablecoin, we're talking about Diem, period. Now whose Diem is that in this case? This is more Diem. There's obviously Venice Diem and so on and so forth.
But, yeah, it was such a a genius way of approaching this idea of how do you solve the problem of multiple decentralized or, you know, privacy first compute providers? And, also, one of the things and this is like a pro and con of Morpheus, but the floating token price. Like, now, when you're staking for compute, it's based on MoraToken based on your proportional share, how much compute there is, and then also the price of MoraToken. But that variability sometimes, while it can be great, is, I mean, maybe at the institutional level, a drawback for some because it's harder to predict what they're spending on any given day or what they need staked. And this kinda solves that problem but still specifically leverages Morpheus.
It was very cool to see that, you know, it it attracted, like, $500,000 on day one, all just grassroots people, you know, allocating yield to to the project. We'll see where it goes from here. And the basic idea for, you know, everybody listening is, the ability to divert yield to earn MDM. MDM can ultimately then be similar to, like, the Venice model, be staked for SMDM and then minted for MCU. MCU is the diem in this case, meaning $1.01
MCU token is a dollar a day of fixed compute perpetually. Yeah. I mean, more stuff like this, the better. You know? And it just, like, expands the ecosystem. I'm definitely participating. I don't know. I'm assuming you guys probably are as well. I know a couple others in the community are. We'll have to get Ryan in here to talk about it sometime soon.
David JohnstonYeah. I'm I'm sorry. Nick's
Unknownon on the podcast. I think I think the I think the other part to the to this is, like, as, you know, Morpheus is really working towards product market fit, and you've got different community members that are trying they're building individual things that are trying to to to solve different issues. Right? And to your point, there's you know, when someone's using when someone wants to use access to inference, you know, it's like you may have somebody that's the the privacy you know, they want privacy first. Right? So I had you know, we have this interesting conversation with some Zcash guys last week.
And, you know, there you you look at where Zcash really is, you know, you have privacy access of your your finances moving back and forth. And as agents expand, you're gonna want to be able to have you know, when tokens are used, you wanna be able they wanna be able to access compute, but not be tracked. Right? And so this was a you know, I think I think Ryan added in the the functionality so that you can get compute credits directly with CCash. That's not currently an option with with
Venice. You know, there is the there are the capital mechanisms of the staking which which, you know, that that can be I think one of the large concern within the community is is like, hey. Is that is that net accretive to the system or is that diluted to the system? And I think Ryan went around, you know, as part of the solution for that was, you know, the the treasury and the structure for it was with a multisig of of large Morpheus holders so that, you know, there's there's not there's not a desire there's a desire to make sure that that
that that is managed in a way that's not extractive from the system. But there are all there we have there there are a lot of profiles that exist within the community and in driving down the friction for that profile is kind of, you know, I think that's where we get we get fit. And as as the marketplace grows larger, as the marketplace is charged with, there's gonna have people that that narrow, you know, focusing on one solution or one problem set and kinda drive it drive it to the bottom. So I think I think one of the things that Morpheus has has worked with, you know, sometimes we've we've
the protocols missed the mark a little bit, but it always takes the lessons and upgrade from it is, you know, structure for an incentivization of solving the marketplace's problems. So, like, how do we build a protocol that doesn't run into the issues of, you know, Hugging Face decides to, you know, remove 5.3, Open Router decides to move, you know, remove a model, people no longer have access to it. Right? This is a decentralized, indestructible
marketplace. People can set up their own nodes. They can get paid, but there are different problem sets that have to be solved as we get through there. So it's a it is a little bit messy sometimes and we, you know, we we we ship it, and then we see what we can improve on it. And, and so it's it's beautiful to see Ryan shipping this. It's, you know, I think he somebody said in one of the chats yesterday, it's it's just a it's a thing of beauty when you can watch a an engineer turn into a 100 x engineer just because the only thing he has to manage are not five or six other devs and communication and all that friction. It's just his agent swarm that he can get directed.
And then share those agents swarms. Right? So we're talking about using, you know, his he's got a he's got a great audit structure that he set for some of the smart contracts. Then, you know, looking at open sourcing that and allowing anybody across the different models. And so it's it's great when people ship stuff like that. Yeah. Yeah. Hashtag m ran on MDM. I've I've been doing a lot of back and forth and community back and forth as people ping me with questions and concerns and and all that. So I I think it's important to, you know, touch base on on any of those.
Bowen BelmerYeah. I mean, I I think it's just another great building block in the mix of everything. Like, prior to this, there wasn't a a Diem, a fixed, compute unit for Morpheus as as an ecosystem. I think having that is a huge net benefit. And also, anything that provides further access to permissionless compute and also further leverages, obviously, privacy as much as possible. And, the, basically, like, nature of having access to intelligence is great. So, like, if I if my agent has an API key with
I mean, theoretically, and I'll screw it. I'll probably set this up. I could have my agent using Morpheus directly through the compute contract, also the API as a backup, also my Venice DM as a backup, also my MCU as as a backup. And and it just, like, the odds of you not having access to compute via all of those things at once just
Unknowntrends towards zero. Right? And that's Well, you you literally just described my setup. Yeah. Okay. Yeah. That that is my fallback chain for my I wanna I wanna pause here. I actually made a miss so so on this, I wanna make sure that we note this. The people may some people may remember this. Most people will not just because this was so long ago, but I think it was early twenty four. The initial idea behind DM of a dollar of compute tied in with the tokenomics was actually it was actually Kyle's. I think he mentioned that at
I think it was at I think we were at Capital Factory, if I recall. Nice nice hidden room. I think he was am I am I remembering that right, Kyle? Yep. Yeah. Pretty much.
Kyle (Bow Tied Blue Fin)It Yeah. Yeah. So you did I don't I don't We don't need to go into that much detail, but yeah.
UnknownYeah. Yeah. I'm oh, hey. Listen. I'm just going full bore here, but, no. No. I mean, I think I think it's important that when people you know, I think reducing the time to execution from those ideas is is important. But I it's also important to to note, like, when someone comes up with a great idea, like, they they need to get credit for it. And, you know, Kyle's been in the background doing a bunch of ideas that I think you have, you know, in the marketplace have have come to play out,
either within Morpheus or within other protocols. There's been four or five other protocols that you that you've shot. And, Dave, you know, David, you've done you've done the same things on this. So, yeah, I wanna loop back and make sure you mentioned that. Yeah. Yeah. No. I appreciate it with that. And and
Kyle (Bow Tied Blue Fin)the the thing here is that mostly what we wanna see is we're all invested in the in the progression and the adoption of decentralized AI. I think, like, a rising tide lifts all boats here, and the more people that are in this industry, you know, seeing other protocols deal with the you know, in in the same space, gaining more adoption is is just a win for us. Right? It's have someone do it, provide an an alternative option, right, Kind of meet users in a different space with with,
how you're providing them value is is how everyone's going to kind of grab a piece of that pie. And, for Morpheus, you know, I think more DM hits a specific area where a lot of value can be provided. I think Morpheus core hits a different value where a different group of people where value can be provided at a significant scale. And I think we all are going to be, you know, key chapters in in the story of how, you know, how did decentralized
AI become, you know, the the base standard of what we do in the future.
David JohnstonYep. 100%. And and just to build on one thing Chris covered really briefly is a lot of people don't realize there are sort of interesting future features already built into the Morpheus protocol. For example, IPFS is built into the Morpheus node. And so when they started censoring and taking down these different models, the first thing I had my agent do was, here's five terabytes in a hard drive, go ahead and start downloading the weights
of these important models and putting them on IPFS. And let's start serving these models so they don't get censored or or taken away in the future. And so, you know, it's it's not just the inference, it's the models, it's the data, it's the agents, it's the harnesses. It's how all of that works together that we can get to these what I I really think we're getting close to recursive self improvement via agents. Right? Everybody thinks of, you know, RSI.
It's the holy grail of AI. Right? It's an AI that makes itself smarter, and people have thought about that primarily in the context of the models, the model training the next model. But what I think we're at right now is your agent learning every day and improving itself. And you don't need to radically change the models from here. They're already good enough to get to RSI for agents. And so that's been my fun experiment this week, and I wanna open source a bunch of stuff around this, is I've I've got the loop.
Right? And then I added an RSI agent on top of the loop and said, hey. Look at everything you did the last twenty four hours. Take the lessons. Put suggestions in to the loop for self improvement, and then start working on them. And it's been so cool to see that loop start to to cycle, And, like, all of a sudden, half the things in my pipeline are improvements to the pipeline. Right? And it figured those out themselves. Right? So I think we're getting really, really close to, like, that takeoff moment
where how fast can my agent improve, how it works for me. And that's the other thing is people think of intelligence as an IQ or very broad. I think it's very specific to the individual. How does it improve your life? How does it improve your outcomes? How does it work for you? It's very personal to the individual. So anyway, it's it's getting pretty cool, but, like, we are right there at the edge of that sort of holy grail.
Bowen BelmerI'm equal parts excited and terrified of that.
UnknownIt's it's the kaizen skill. Right? It's improve 1% every day, and you get the agent to do that. And then it goes back to your con concept of we have Nivolt's concept of intelligence is your ability to get what you want out of life. And, you know, you got the the agents themselves have to know what the goal is for you to get out of life. Right? It's like, you know, so you wanna be able to build that and say, hey. I I know exactly what I'm trying to get done here. And, otherwise, the agents just kinda, like, run off thousand different directions. It's like
Yeah. How much of it is just, like, the standard process and the standard approach of being a human being where you're chasing virtue and you not not you note down to, like, okay. Hey. I'm getting done a, b, or c. Here's where I failed the day. I need to go ahead and revise that tomorrow. Okay. Wake up, I might do that again. Right? The agents can just do we're trying to infuse virtue into our into our agents.
David JohnstonSo far so good. Like, after I got the RSI agent live, there are still things that it escalates to me for a judgment. And then I become the bottleneck again, but what I've had to do since then is a learning agent that as I solve things manually, it takes the lessons from that and tries to derive those preferences and values and virtues that I'm expressing when I make those decisions to improve my architecture agent that's making those judgment calls and whether, okay. Yeah. You can ship that or no. That needs to be seen by David first. So, yeah, it's it's getting there. It's pretty it's pretty cool. I had my agents spin up a dashboard,
and so it's now like, okay. Here's everything in the loop process, which stage it's in, you know, which agent is working on it, and you can see it all working in in real time. It's it's pretty cool.
UnknownScott, speaking of workflows, do you guys have anything in workflows that you guys are really liking this week that you're just like you've you've implemented and you're like, wow. This is I wish I'd implemented this sooner, or I'm really appreciating it. Oh, wow. You're weeds. Go ahead.
Bowen BelmerOh, I have oh, go ahead. Sorry. No. No. Go ahead, bud. I was gonna say I have one that I'm try well, kinda one that I'm building, one that I'm trying to figure out, but it's being more difficult than I thought. The one that I'm working on building is I I use my agent a lot for for the mortgage business, and different lenders at any given time will do, like, various pricing incentives. So beyond, like, what their standard pricing is for a given file, they'll give incentives. But you would have to kinda cross reference
them manually to figure out, like, hey. At any given time, this is, like, what their incentive is. Does this situation that I'm looking at qualify for this? Like, does it make sense to consider it? And so having an agent that can just track those for you and you explain a scenario and it flags it automatically, like, hey, this is, you know, one you should take a look at that happens to be running right now. I'm that I'm, like, in process of having setting up, and it's it's It works, it's just, it's not as robust as I want it to be. The other thing I'm working on setting up,
and this kinda feeds into there was a question in chat about if if Morpheus will have a slick, UI like Venice. I mean, the short answer is Morpheus itself doesn't have a UI in that capacity, but there are ways to use great UI. So for example, because the harness I use is Hermes, I use their desktop app, and so my agent is on their desktop UI, which is really slick and great looking, but running Morpheus compute on the background, just to answer that question. And there's a lot of other ways to do things like that. But,
I'm trying to integrate my agent with iMessage using Bluebubble, which is a, like, extension that someone did, but you need an Apple ID for your agent. And Apple is making it a huge pain in the ass to get an Apple ID for my agent. Has anyone figured this out?
David JohnstonThe best answer I got is give it its own phone number. I did. Yeah. And it's denying it still for some reason. Well, once proof of human and you can't give it proof of human. So, yeah. It's it's fun to to try to navigate that stuff.
Bowen BelmerAlright. Well, if someone figures very out
Kyle (Bow Tied Blue Fin)very difficult in this situation. But, yeah, the the only workaround that I have found myself is it it needs its own phone number. And it can't from some testing, it can't be a Twilio phone number
Bowen Belmerbecause they recognize that. The Twilio, no sudo. Yeah. No. But so I found that and I went and got a proper SIM card and everything and set up a plan. And for some reason, it's still denying it. And I'm wondering if it's because it's like a month to month they can somehow pick up on that and they don't like that. I don't know.
David JohnstonMaybe so. The the big workflow win for me is we've got such a big backlog and morpheus of great ideas and proposals. I finally had my agent build out a full dashboard of every pull request issue and MRC in the Morpheus ecosystem. And then I built a Morpheus worker into my loop. And so now every hour, it's picking up one of those open issues. Sometimes stuff is just out of date. Oh, that's already been implemented.
Great. Like, go update the documentation. So getting very close to sort of I'd love to do a full refresh, you know, of of all the things that people are too busy to to update and just get things fully up to speed on, like, okay. What's been proposed versus what's been implemented and all that? So, you know, having that for install OpenClaw, Morpheus, my own coding and everything. Now I've got these dashboards, and I can just see at a glance where everything is in those flows. So, yeah, it's definitely starting to accelerate. So I went from having an agent to having an agent that has memory to having an agent that loops tasks
to having an agent that learns from that process and itself improves and then pointing it at the projects that need all of that that work to be applied. So, you know, it's it's gonna be pretty fun. I I think I need to to start sort of just open sourcing this building Bernardo journey that I've been on and, like, just showing off, you know, how it hooks in and, you know, where Morpheus powers all of that in the back end. Yeah.
Kyle (Bow Tied Blue Fin)Back to your That's awesome. Line,
Unknownhaving agents loop and do improvements and then correct things like, hey. Here's where stuff that's been proposed that hasn't been done. I think something we're seeing right now on the on the smart contract side is on bug bounties. You're starting to get just just an avalanche of submittals because the the cost of the one, the cost of submitting is nothing. You can just run you know, there's there's thousands of models that people are out there running them and and doing submittals, and then then they do, you know, the 13 follow ups. Right? So I think I think on the initial concept of the Internet was you would add some friction,
you know, some payment friction to it, you know, and there would be these might the concept of micro payments. And I think something that's right now I know is getting proposed is implementing within Morpheus will be implementing are the the concept around paying for a bug submittal. Right? So now it's it's no longer it's no longer, hey, I can I can submit 15? You know, I can never submit 15 things. I'm a get I'm gonna get I'm gonna get paid for it. And,
you know, it's it's just micro payments. So it'll be interesting to see how this plays out with the concept of micro payments where you do have to add in a function that says, you know, where it says the, you know, your agent can submit 40 things to me, but to end up on the my priority list, it's just like email. Right? When even the cost of email is zero, you start to get spam and you start to get out there. So it it will be interesting to see how, you know, businesses and protocols and us as individuals, solve the avalanche of attention and the avalanche of context that cost people next to nothing to generate for us to review. Right? So it kinda forces us in. So we it'll be a very interesting and one of the things that, like, we're I've been implementing
is are the concepts of, private repos where no one else can have access to it. Right? So it's it's we stuff gets generated. You have the you know, you you have you AI generated content, but then it is you know, people are having to go through and and do more of a fine tooth. Like, hey. I'm I'm signing off that I've actually looked at this, and it's not just AI slop that's, you know, missing one or two prompts or or adjustments. But but it will be interesting to see this evolution of workflows, you know, over the next you know, as this accelerates, it's only it's only gonna get worse. So we're gonna have to do something to to be able to guard our attention. We only have the same twenty four hours in the day.
David JohnstonYeah. We we talked about this. I think the term we've used before is an AI firewall. Right? Mhmm. And and everybody's gonna need one because you're gonna have slop, right, AI slop agents coming at you with a billion emails and requests and whatever, and, like, how do you figure out where the signal is? Just like we did in the early days of the email. Right? The the Nigerian print scheme and, you know, 50 other variations, those started not showing up anymore
over time. And it was because Gmail and others got really good at pattern matching all of those scams, taking the reports, you know, for phishing and stuff like that, and and all of a sudden disappeared from your inbox. So as agents ramp up, we're gonna need the same thing. The only way you're gonna be able to deal with it is having an agent, right, to sit as that defensive shield that figures out what's real and what's worth addressing and
what you should do next. So, yeah, to me, that's the logical next step, and that's where Morpheus could really sit in there and and do it well because it's aligned to you as opposed to Claude and others, right, are are aligned to making money for them. Right? It's it's their agent, not your agent. Right? So I think that's sorely needed in the market. Sam Altman wouldn't do that to me. Yeah. Never. You never mind No way. Which is Not Sama. Not my boy. Not Sama. Not Sama. Oh, man. You guys saw with the the,
what was it? The no ball prize, you know, type of deal. Like, that was that was wild. We're we're open AI, you know, oh, we solved this millennial math problem. And then mathematicians are like, wait a minute. We were working on that exact problem and, you know, we're pretty sure you used part of our solution. Like, well, you know, we can't say we didn't, you know, read some of your chats. You know, they might have got included in some training details. And they're like, really?
Like, you just you just ripped off our solution?
Bowen BelmerOpenAI and Anthropic are just quickly becoming Hooli from Silicon Valley. And all the all the DAI people are the Pied Piper.
David JohnstonYes. Exactly.
UnknownYeah. Of the very Wait. Wait. Wait. Hold on. Hold on. We've gotta have our fat guy. Hold on. We gotta have our fat guy, our tall skinny guy, our Indian guy. I think we're missing our Indian guy. We have we have Anton,
Kyle (Bow Tied Blue Fin)is Ukraine.
Bowen BelmerThere there was a time It's so. Early early in my, involvement in the community where we had a meetup with a whole crew of people, and we and you guys were all there. And we watched an episode of Silicon Valley, and I was looking around the room, and I was like, wow. We really fit the mold of this thing. And, like, every single box was checked at that time. You know? Of course I understand, Bo. That show was not a comedy. It was a documentary.
David JohnstonAnd it was just a little too close to home for anybody that's ever been startups and
Unknownand anything related to Silicon Valley. Yeah. It really it really is a it really is a little bit too close to home, especially if you're if you're if you're going through some some route any of the any of the issues that they flag, any investor gets pulled or you didn't take the money or you did take the any one of those that, like, you're on the other side of, you're like, I I Let's watch something else. Mandatory
David Johnstonwatching for anybody involved in startups. Yeah. 100%.
Bowen BelmerYeah. One other thing I wanted to talk about, you guys, and I I do have, unfortunately, a hard stop right at three this week, so I apologize, guys. And we did we'll be back Yeah. None need none need to apologize. Listen, we yeah, we can ramble forever. Don't Yeah. Fair enough. So So The thing I wanted to talk about was was Jeff. And hopefully, at least one person understood the title joke of my name's Jeff, which is there's a line from
Kyle (Bow Tied Blue Fin)My name's Jeff. 21 Jump Street. Yeah. But
Bowen Belmerso actually, I wanna try this. I believe I can share my screen with audio so anyone who is watching this can also listen to this. Let me know if you guys can hear this. Can you guys hear this? Because I wanna just play the video and then talk about it briefly, or at least some of it. Obviously not. But the trillion dollar question is why Can you guys hear this?
UnknownLHF is to LMs that are optimized for human preferences and include issues such as mode dropping, overconfidence, and an overall lack of reliability. These flaws mean that LMs require humans in the loop, and almost no true automation can be done. At TypeSafe, we've spent two years building in Stealth, and we're finally ready to share our new type of foundation model that's optimized for automation, system one models. With a new architecture, new sampler, and new training algorithm, reinforcement learning for calibrated decisions.
The improvements are clear if you see them side by side. Ask the system one model a ton of structured questions, just like you would an LM. Get the answers back near instantly. Meanwhile, LMs take hundreds of times longer to finish responding. Look at how the LM generates sequentially, which is great for a natural conversation, but totally useless for computers. LLMs expect intelligence from the tiny straw auto regression, and similar to the jump that transformers made over RNNs,
we are replacing sequential computation with parallel because that is obviously the future. Our system one models output decisions with probabilities and confidence instead of words, and they can't hallucinate. They're a lot more like code, reliable, fast, self consistent, and type safe. This opens up a whole new world of possibilities and applications. Applications. Today, we're releasing JIV, the first public system one model. It's a 100 times faster, so real time AI is finally possible. A 100 times cheaper with input tokens priced at $42
per billion tokens, and output tokens are free because they're finally too fucking cheap to meter. JEV is super smart, and its intelligence per dollar is literally off the charts. This is just the beginning. We're excited to see what you build with JEV. As we say at time safe, we're building prod and not God.
David JohnstonAlright. We get the point. Yeah. My my favorite, first of all, is that the shirt changes. Yes. I was gonna say that. It cuts to him like he's got a random different outfit. It's really interesting. Right? And this has been effectively taken over my timeline the last twenty four hours, is everybody talking about JEV, beta testing JEV, doing something interesting. You know, what they're effectively doing seems a lot like classic classifiers,
right, in machine learning terms, but here it's dynamic. Right? So instead of like just this is your decision tree and this is the only thing you can do, is they've made it more flexible and you can throw anything at it, and it can take that classifier like approach. So it really is super fast, and it seems like they've done something pretty novel here. Though, I did see people point out pretty quickly, like, there are open source versions of this. And so pretty quickly, went on to Hugging Face. I was like, oh, cool. There there are some open weight alternatives live right now, and the models are
surprisingly tiny. Like, these aren't 700,000,000,000 parameters, like 2,000,000,000 parameters, 3,000,000,000 parameter models that can do this type of work. And so I think as agents have taken over and it's become less about talking to a chatbot and, like, actually taking actions with your agent, this type of model is gonna be really compelling. So, yeah, can't wait to see some of the benchmarks as people kinda get all their their test results back, but this seems like a a major breakthrough.
Bowen BelmerCan you give just a little bit further context, David, on JEV for Dummies? Like, I don't, I'm gonna be honest, I don't fully understand what they're doing that's different other than, I like, I I I understand that they're they're talking about doing, instead of sequential reasoning, like, they're they're basically doing it all at once, but I don't understand functionally. Like, I I guess, why is this different? Like, why has no one else done this?
David JohnstonIt's not that no one else has done it. It's that approaches like this in the past have been fairly fragile and, like, limited. They've been, like, static. Like, okay. I trained my system, and it could do these 10 things. And if I know in advance what those 10 things are, then they're gonna do them really quickly. Right? So think of it as a decision tree, a classifier in a traditional sense. The power of LLMs is they're a little bit slower, but you can throw almost anything at them. Yeah. Right? Because you trained it on
trillions of data points, basically the whole Internet, and, like, you could ask about virtually anything from dinosaurs to asteroids, and it's gonna have something, right, relative that into its weights. Whereas, this is much more like, I wanted to send emails or read this document or whatever. Right? But that's that historically has been much more narrow. Right? And so what they're trying to do is get classifier like outcomes and generalize it and make it dynamic enough to to to address as many use cases as LLMs
Kyle (Bow Tied Blue Fin)have sort of, like, had this breakout moment. So, Kyle, I don't know if you wanna add any color. Let me give a good comp a good comp that I think is a good visual thought. So I think a lot of people last year were involved in n eight n. Right? Like, as agents were coming along, we're like, oh, we can create these workflows in n eight n eight n, which now we know is kind of garbage, honestly. Like, they're you you could do things way better than n a So in n a n, the old way, the pre LLM
agent way is that you would just map out. If this happens, do this. If this happens, do that. If this meets this criteria, go here and perform this action. And then we switched over to you have that little AI agent box in n eight n that you're like, here are the tools available to you. Here's the LLM. Here's your memory. Right? So it's kind of achieving that speed that you got of the old way of the if then statements
with the, dynamic aspect of the AI agent box. Right? So you're you're kind of meshing. You're you're still saying these are the tools available to you just like you are with the AI agent box, but you're able to get to that answer without 13 thinking loops and 27 different calls to the LLM and taking, you know, ninety seconds to get there.
Bowen BelmerBut they're having to, like, proactively guess basically or determine, like, what some of these decision trees will be. So it's like a collection of decision trees instead of figuring out the decision tree based on what you said. Is that a good way of thinking about it?
David JohnstonI think that's that's a fairly good comparison. I think Kyle's got the the best analogy. You know, it's a dynamic and more generalized version of the NA and, you know, if then Yeah. Maintenance, but across millions of possible workflows or billions of possible workflows. Right? And I think the thing they're they're pulling out is focus on the action, not the word. Right? An LLM is driven by, you know, attention is all you need. Right? That that is like, I have focus on on the word and the statement, and here it's much more seems to be focusing on
the action or the the thing the agent needs to do. Right? Because it's it's it's shifting again from talking to taking action. So, yeah, it's it's pretty cool to see. I think we'll see this play out, but good good direction. I mean, you get a 100 x on current performance. These agents are gonna be cranking things out in minutes instead of hours. Right? Well, that's a good loop running every three minutes instead of every thirty minutes instead of every three hours. Right? The fact that the token output is free too because it's so cheap to use is outrageous.
Bowen BelmerThat's crazy.
David JohnstonYeah. And that's a call out to an old phrase. Like, when when nuclear was invented they thought electricity would get so cheap it would be effectively too cheap to meter. Right? And the regulatory agencies killed all of that progress in the seventies, but seems like it's finally turning on again. But that's what people wanna get to. And with tokens is like too cheap to meter, right? And you charge for the product or the harness or something else, but the tokens are going to like zero cost. So if they've gotten there with outputs, that's really impressive. You saw them highlight
their pricing at $42 per billion. Billion. Yeah. Right? You know, there's a there's a thousand millions and a billion. So it's it's significantly cheaper than than anything else I've seen.
UnknownI mean, if you look at if you look at the explosion of what happened with the second brain setup with Obsidian, yeah, the entire thing around that was was the process of the process of contextualizing data and and tagging it. Right? So making it towards more easily accessible for your agent properly and and there's, you know, mapping for it. And I think that that as we
as data as as there is more data, the process for indexing that data is gonna be more and more valuable. The speed at which something can be accessed and then the actual correlation between the tokens. Right? So, you know, we all look at, the training of these models as, you know, this is is I'm speaking in broad terms, but but say it's, you know, you're you're 61% on Reddit for OpenAI and you have these these large models. They're not designed for specific things. I think, you know, we're seeing more and more cases that are popping up where people are like in manufacturing right now.
There are there are firms. They are taking in their their building models specifically for their firm. They're building it specifically for their customer base. The answers that they've done and then, like, how their machinery taking their specific niche that they know with their data. And I think that as as as models, as we get better and better about managing data, there really isn't the reason to do 13 loops. Right? When we when you and I, when the three of us get on a call, four of us get on a call, or we're trying to look up data, we and we, you know, go go back to the encyclopedia
and we wanted to know something. We went to a specific page. We didn't, like, flip through every single one and and, you know, bounce back and forth on it. So it'll just be interesting to see that. And I think I think to go back to it, it's like the idea of that is great is gonna be the execution of how accurate is that. You know, is it is it biased or not? And, you know, I I think one of the one of the things that popped by I think I pinged somebody earlier on this It's like what we have got access to it, and the the question was I think they were the the kids were doing retarded.
They entered retarded, and they they gave it three options of liberal, Catholic, or conservative. And the a and the liberal one at 88% probability is being the accurate answer for retarded. Right? So maybe it's maybe it's biased, maybe it's not biased. I tend to agree I I tend to think that it's not, but in that instance but all that being said is the idea is the it's in the execution. Right? The idea is great. What is the execution? And then how is that how how practical is that for a specific use case? And
we're, you know, we're on the we're on the a brave new world on frontier there. So and, Boen, I know you've got the end of we had two minutes over. Yeah. Yeah. No worries. I am I I picked my extra call. I left you just enough time to get us demonetized on YouTube. Thank you. God. I'm just kidding.
Bowen BelmerI'm just kidding. You guys, we're not monetized. But regardless, we're It's okay. I'm pro we're pro free speech out here. But, no. Anyway, yeah, I do I do have to run though, in all seriousness. But, yeah, I mean, great stuff week over week. Still loving the new format, by the way. Being able to share stuff in video and audio is amazing. Thank you for joining us if you're listening.
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...
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...
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...
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...
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...