Cashew & Confidential AI | The Big Privacy Milestone
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 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 #70! Recorded: 23 April 2026 AI SUMMARY: This episode features Rob from DRM3 introducing Cashew a community‑first block explorer for Morpheus that surfaces providers models staking stats and subnet data with plans for capital contract coverage and time‑series insights. It also covers Morpheus shipping Version 7 with full TEE phase‑2 support enabling end‑to‑end private attested inference between consumer nodes and TEE‑tagged models achieved with Secret Labs. The discussion includes latest model updates (Kimi 2.6, GLM 5.1, Qwen 3.6, GPT‑Image 2) the push for auditable agent workflows via DRM3 provenance and why decentralisation and data privacy matter as AI expands into science and healthcare. It closes on industry moves community plans for demos and weekly town halls and upcoming appearances at Consensus. Timeline: (00:00:00) Intro (00:01:20) Why Cashew was built and early focus on contracts and providers (00:04:24) Planned features: capital contracts, leaderboards, time series analytics (00:06:24) DRM3 overview: bottom‑up provenance and trusted AI outputs (00:09:45) Open data, provenance tools, and building in the open (00:13:55) Rob’s tech stack and multi‑agent coordination setup (00:16:15) Agent meshes with Elixir, APIs, and code audit loops (00:18:55) Security note: bridge exploit context and Morpheus resilience (00:21:39) Back to AI: new models landing on the API gateway (00:25:29) Model hype cycles and GPT‑Image 2 leap in image generation (00:27:43) Multilingual image understanding and open‑source catch‑up (00:31:01) Personalised health, DIY genomics, and data privacy needs (00:34:22) Version 7.0 release: full TEE stack and attestation flow (00:39:00) Why this TEE milestone matters for owning your intelligence (00:41:31) From impossible to shipped: journey to confidential inference (00:45:00) Who can run TEE today and realistic hardware requirements (00:48:09) Economics of TEE vs general inference and access via Morpheus (00:51:22) Industry moves: xAI’s Cursor deal and consolidation trends (00:54:41) Community and events: Consensus plans and outreach 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.
David JohnstonWell, I mean, there's a there's a ton of stuff to cover today, but I'd I'd love to get in to, Cashew first and, kinda go from there. There's a bunch of new models. There's a bunch of new news to cover. But, you know, I think the most important thing in the world is people building cool stuff on Morpheus so more people can access it. So, Rob, you wanna do a quick intro? Talk about yourself, how you've been involved in the community, and, yeah, kinda what inspired you to to build Cashew.
Rob (DRM3)Yeah. Sure, David. So I've been involved in one way or another with Morpheus for probably nine months or so. I'm always looking for, you know, optimizing CapEx, OpEx on AI spend. So naturally, I kind of found myself, you know, looking at Venice and Morpheus and various others. As I was looking at Morpheus, it's it's interesting because there's a lot of contracts and, you know, there's a lot of sort of ways in. There's a lot lot of ways you can stake and and participate in the network.
Being mostly interested first and foremost in inference, you know, I wanted to run my own node and I wanted better visibility around what's actually going on in the contracts. So I started Cashew, it was actually was actually a cash for me initially and when I did the second version I called it Cash Dew and I guess the name Cashew was born at that point. But really just looking at the Diamond contract most specifically initially
and trying to get a handle on you know, what are the active providers, what models are they offering, what's the you know, what is the stake to time for session look like. And so I ended up building effectively a block explorer for Morpheus and of course then I, you know, I wanted to have I wanted to be able to see the price. I wanted to see providers coming coming and going. And then eventually, I I got in more into the Morpheus subnet and and I ended up adding
builder builder nets to Cashew as well as in Morpheus token holders as well. So it's starting to shape up as sort of an overall Morpheus block explorer. What I'm missing right now I think mostly is the capital contracts. I don't I'm personally not doing much on the capital side so I haven't really gotten around to that. But I think with the capital side, kind of have a really nice optic and window into what's going on in Morpheus.
David JohnstonThat's awesome, man. Well, before we move any further, what address should people be putting in their browser? What's the URL?
Rob (DRM3)You can go to cashew.explorer.drm3.network.
David JohnstonSo cashew.explorer, and then the name of your project.
Rob (DRM3)And then .drm3.network. Yeah. And that's that's explorer, explorer. You can also go to drm3.io and it's linked to from there. You can also go to status.drm3.network and ask them to from there as well.
David JohnstonAwesome, man. Well, that's that's great. For anybody that hasn't checked out the new explorer that Rob built, it's really clean, really sort of easy to understand. I I think sort of brought to light a bunch of stats that I hadn't been tracking, so it's cool to to see them all. So yeah. No. That's that's a huge benefit to the community. Really appreciate you putting that out there. What what are your sort of plans to to grow it going forward? How do you want it to evolve? Are you getting sort of requests for different stats, or how can people give you feedback?
Rob (DRM3)Right now, best feedback is at this point, it's probably supported d r m three dot I o, an email there. So one of the key things that I'm using Cashew for foundationally is a peer to peer inference client node and gateway project that I'm working on. I'll be I'll be talking more about that in the coming weeks, but effectively, you know, being able to run your own gateway locally
against another provider, a direct peer to peer without a gateway in the loop. So kind of kind of leaning that way, I think there's room for I mean, the builder contracts to cash you. I think that or I'm sorry. The capital contracts to cash you, I think that would be a nice add to kind of surface. As far as cash you goes, think also kind of building up some of the stuff that Morpheus has around leaderboards. I think there's room for for that. So I've I've started to save this
the snapshots off to a BigQuery warehouse, and I'm gonna be doing more like time series and things like that to kind of tease out, you know, what's going on in the network, you know, trends over time, things of that nature as well. That's great. Yeah. No. I I think that kind of, you know, visualization is everything. As my friend Scott would say, people eat with their eyes.
David JohnstonAnd so you could have a a great protocol and you can have really great code, but unless people can see it, you know, they don't really understand it. So, yeah, I think I think more and more of that would be would be great to see. As sort of like a crossover, I'll see if I can promote Alan here as I I saw that he's been building out a lot of visuals and stats as well. He's been maintaining tech.mor.org.
And if you guys have noticed on there, there's, like, a calculator. There's also now a bunch of data about token usage. You know, it'd be great to to see maybe there's a a crossover there for what's what's getting built. But, you know, as people, I think, get hooked into Morpheus and and the usage continues to grow, you know, that's what sort of makes it real for people is be able to see that chart that goes up into the right.
Rob (DRM3)Absolutely.
David JohnstonCool, man. Well, glad to have you here, Rob. Anything you wanna highlight about, DRM three? I mean, you've been building in this space for a long time, and there seems like there's a lot of intersections with Morpheus and this is sort of just the start with the Explorer.
Rob (DRM3)Yeah. So DM3 provides trusted outputs to what agents produce. So, you know, today we do a lot of top down analysis. Peter Thiel has a new project out. You know, Elon Musk had, you know, GrokX, you know, effectively fact checking from top down. Even even going back to Snopes, it's sort of the same idea. You kinda take the output and then you're you're kind of backtracking and trying to figure out, you know, is this true? Where do these things come from? What are the primary sources and all that? So d r m three takes a a bottom up approach.
Every time we're getting inference from Morpheus or Venice or or Cloudflare or whoever it might be, anytime we're getting stats from Cashew, and anytime we're talking to third party APIs or mapping data or analyzing data or synthesizing data with with with prompts or whatever it might be, we're capturing the request response, the activity that's going on, the data that's being generated, and we're creating
cryptographic signatures over that. And then we chain those together and we build up effectively a tree that is the set of receipts that go into an AI output. So imagine taking RSS feeds, you know, government APIs, etcetera etcetera, you know, financial, you know, Nasdaq data, CoinGecko data, PolyMarket, etcetera. You're pulling all that information in and capturing exactly what you're doing at that time,
who specifically, you know, what sort of service or signer within the DRM three network is capturing that even down to the query, the request response, all that. We're signing that and then those get fed upstream, you know, downstream, downstream, downstream until ultimately you're remixing into podcast, new news, new blogs, etcetera. And when you have those final outputs from the agents, the arm three is providing you with a full list of, you know, immutable receipts, the timestamps, and exactly who did what, when, where, how, and why. And then what that gives you is exactly the ingredients that went into the muffin.
The EU is starting to enforce come August effectively something around, you know, you have to explain whether something was AI generated or not, what human aspects were there, what machine aspects were there, how did you get the data, where did it come from, was it lawful, did you have rights to do what you did with the data. So the m three solving sort of the AI trust problem from the bottom up. And as I'm building, I'm building infrastructure, products, and protocol all at the same time.
And where decentralized AI fits is a lot of what I what I'm surfacing right now is open data. And so the open data, you know, processing open data, you know, do the entire thing out in the open, including AI synthesis on that data, remixing, etcetera. So everything is effectively open. So while I I may not be, you know, creating, you know, full open source shop, I release the m three provenance. I release AI text analyzer and various other things.
And I'll be building around those to to make those more user friendly to, you know, to add into other projects so that people can can connect to the network. And I'm basically building it all up and kind of eating my own dog food as I go, building out this sort of data trust and monetization protocol.
David JohnstonThat's really cool. No. I think that's gonna be critical. You know, people are gonna wanna know what their agents doing, why it's doing it, how it did it. You know, I sort of more and more of the workflows that we're involved in become agentic, you know, that visibility, I think, is gonna be everything. And it's sort of the opposite of the clawed in OpenAI world where everything's a black box, and you don't even know what model they're running, let alone what the system prompt is, you know, let alone what they're doing with your data afterwards.
So appreciate you building, stuff in the open.
Rob (DRM3)Thanks, David. Appreciate it.
David JohnstonYeah. Good stuff. Great. Well, I think we've got a a number of updates. I see, Bo is, back online. Let me go ahead and get him promoted. Alright. Bo, you are now a speaker. Let's see if, his mic is working now. You online? There he is.
Bowen BelmerI swear I get rubbed by x more than anybody on spaces. It's crazy. They just don't want the information to get out. I could hear you guys. So, Rob, thank you so much for joining. Anybody who builds something beautiful to look at that I can look at and understand what's going on on the back end with the protocol or anything that touches the protocol is, like, the greatest thing of all time. So thank you so much. You bet. Glad to hear that, though. Thank you. Yeah. Absolutely. I mean, this is amazing. I'm I'm looking at it again right now. I I missed just the back half of what you guys were saying because I had to restart the system,
and everything. But, yeah, this is awesome. And the more things we have like like this, the the urgency is to onboard people into kind of what's happening here, which is just so cool.
David JohnstonYeah. That's awesome. Yep. Bo was was really kind of, just to key catch you up real quick and really kind of talking through, you know, what, what he's doing at DRM three more broadly for agentic workflows, building them all so that they can be audited and knowable and provable, sort of the opposite of the the Claude and the OpenAI world where everything's a black box. And so I think that's that's really cool. So I think that's gonna be critical for sort of building agents,
that people trust as opposed to the agents that take advantage of people. And I think that's sort of the cross crossroads that we're at right now is they're trying to build up the same walled gardens that they they always did and that they did with web two at scale, right, where they can effectively extract most of the value from their users and, you know, leave their users not only aware that unaware that that, you know, value is being abstracted extracted. But, you know, that's that's what we wanna avoid.
Because if that's done with AI, it's gonna be sort of like web two on steroids. Like, let me extract all the value from your mind and all your workflows and all the things that you create, right, are just feeding the revenue of of these these companies. Right? So, yeah, I I think it's really important moment. Rob, I know you've been working nonstop on this. One of the things that we we often cover is I'd love to hear what your tech stack looks like, how you're using agents.
You know, obviously, there's been this explosion since the OpenClaw moment in January. Just curious how that that journey has been for you and and sort of what it's unlocked for for your ability to build.
Rob (DRM3)Yeah. Well, I I tend towards TypeScript and Rust when I build. I use Google Cloud, Cloudflare, you know, Alchemy. I use Postgres, Cloudflare d ones, d twos, caches. That's a good chunk of the stack.
David JohnstonAnd are you running one agent on that? 10 agents on that? Are you have you gone multi agent? You know, I'm always curious how people are are building and getting them to actually coordinate in the real world. Yeah. I'll say as far as agent coordination
Rob (DRM3)goes on the on the build side, I tend to run I run two Mac minis and a and a laptop and I could be running, you know, up to sometimes as many as 50 agents and I've actually got a little clicker that clicks between my it clicks my keyboard between the machines. So I'm I sort of feel like a switchboard operator just kind of clicking and I've got a bunch of terminals open and and and I'm sort of going through there. But I've also got d r m three office and
Avatar, and those are really, like, agent agent based products. So d d r m three office is your virtual office, and Avatar is more like hire a team of virtual a virtual team, essentially. So so it you could think of it like imagine you wanted, a newsroom to build news and blogs for your company. You could you would use Avatar for that. And behind the scenes, I also use Elixir, and that's an old, like, telco technology
that's very robust. And so I have sort of a I I sort of build up APIs and composite APIs. A lot of that is sort of Cloudflare, and then I build up sort of it more of, like, an agent mesh and comms mesh, and I I use Elixir for that. And I I create effectively, like, these really tight micro services, and then I'm able to continue that d r m three signing from that perspective. And then you can compose those into a dynamic DAG using natural language.
So I guess it would it it wouldn't be fair to not include Elixir in the mix here because I use that as well.
David JohnstonVery cool. Well, I appreciate the the insight. Yeah. Switchboard operator is a pretty good way to describe it. I feel like I'll go from one model to another, basically copy and pasting over instructions or code or whatever. I've only recently got it to the point where effectively I I I trust my agent to interact with another agent and model and be able to basically do a predefined audit of code or things like that because I found
cross model auditing is really powerful, right? You can have a less powerful model like GLM five or 5.1 doing most of the coding, but then when it comes to the audit, I'll throw it at Grok 4.2 reasoning and it does a really good job of finding bugs and they basically haven't run that loop until the code is perfect and that might take three loops or five loops, but you know, takes a couple of minutes to basically, you know, jack up the code quality from mediocre to rock solid.
So, you know, that's been a big unlock for me.
Rob (DRM3)Yeah. I found the same. Even even multiple clods with different contexts can catch catch one another.
David JohnstonYep. 100%. Cool. Well, Bo, do you have other questions for Rob? It sounds like this is just a preview of things to come based on what you're building next with, with, your API and gateway and inference tooling.
Bowen BelmerYeah. I mean, nothing really specific at this exact moment other than, we've been talking about, and this is maybe a bit of a teaser for people, of figuring out a way of doing this exact town hall every single week with a video component as well. And so, Rob, I'd love to have you back on well, anytime that you wanna join. I mean, we we welcome, you know, people to join every single week. But, also, once we have the capability to walk through from the perspective of the person who built it, some of the visuals and things that we're talking about to kinda show people how how it can be used and leveraged.
Rob (DRM3)Absolutely. Happy to do that, though.
Bowen BelmerAwesome. That'd be amazing. Thank you. You bet. And and, Rob, stay stay on stage, please, as as we run through, you know, everything else we wanna cover this week because if we were always looking for, you know, insight for community members and builders. So we do have a list of things as per usual that we want to cover this week. First one here, we're gonna kinda breeze through a little bit quickly just because it's one of those things that definitely needs to be mentioned. It's it was a huge thing that happens. But, also, we have a lot of other,
more, like, DAI focused and interesting stuff to touch on. But that first topic is the whole situation, with Abe and LayerZero and Keltdown and the hack allegedly by Lazarus Group, and those happenings. All we really wanted to touch on really quick is that the Morpheus contracts, it also didn't directly impact the pools that our contracts touch. So we were really, you know, not involved in that capacity.
Wanted to also reassure that the the setup that Kelpdau is using through the Layer zero bridge is fundamentally different than ours. This is coming, you know, directly from individuals within the Morpheus protocol who've worked on the bridge. Most of the more capital in the capital contract still flows through Lido, which was, again, not not involved in this, particular exploit.
And, yeah, I mean, really just wanted to touch on it as kind of, like, a high level, reassurance in in, like, the strength and robustness of what's being built here. David, is there anything else you wanna add to that?
David JohnstonWell, I mean, this is where decentralization pays off. Right? So doing the extra work on multiple validators and, you know, a five of nine multisig, like, all that takes extra effort, but it prevents, you know, events like this. You know, the the affected contracts, I guess, had a one of one for their bridge, right, which is not ideal and not really decentralized. Right? So, you know, we you know, the folks that have been, you know, contributing to the the open source and the smart contracts,
I think we've we've definitely leaned towards higher security, less convenient, and this is one of the examples where that pays off. Right? So, you know, definitely worth, worth understanding. But, yeah, it's good to, good to see the Morpheus community was was unaffected.
Bowen BelmerYeah. Absolutely. And to anyone else that's been listening to any of the other content we've been putting out this week, I'll just make a a really quick side note that my tears and prayers have been answered, and we are no longer paying more than 50% APY to those who who had borrowed Dime on, Ovid at any point. So unrelated to Morpheus, just just really quick update that they seem to be handling it, and it was another example of on top of, you know, preparing and having sufficient decentralization
so this type of thing doesn't happen. Also, these protocols having safeguards in place for when things do happen, ways to address them, which so far has been, from my perspective, honestly, pretty impressive. So we will see, what comes with that. I just wanted to touch on it really briefly. Back around to actual AI stuff and thing that things that are more relevant probably to the builders and community members listening to this. There were a plethora of new models pushed out over this past week. And so as always, love to talk about them, talk about how they benchmark, hear from
some of the builders and developers in the community who specifically help pull those types of things into what's going on with Morpheus, whether it be bow tie, bluefin, and Kyle through the API gateway. The first one, starting out with, is Kimi 2.6, which is a new open weight model. It's been benchmarking very well from what I understand, and we already have it on the API gateway, which I think was, like, a same day turnaround. Kyle, correct me if I'm wrong.
Kyle (Bow Tied Blue Fin)Yeah. But it was about same day. Within a day or two.
Bowen BelmerIt was Kyle. It was the same day. No. It's fine. Can can anyone who's maybe had experience working with Kimi already speak to their experience with it?
David JohnstonSure. I'm happy to jump in on that. So I I found it was really good at analysis. I had it look up at a bunch of code repositories, do a bunch of write ups, kind of architecture stuff. Maybe it's something with my my harness, but I I didn't see the agentic tool calling that I was hoping to see. And so I think there is a gap between the benchmarks, which look amazing, right, on Kimi 2.6 is is supposed to be basically, you know,
quad 4.5, 4.6 level, but but I haven't seen that in the performance yet. So I'm curious if other people have had different experiences, but I I chose to stick with GLM 5.1 for now, which I've gotten pretty consistent results with. And so after experimenting with with Kimi, I ended up going back to GLM 5.1. But, you know, I I might give it another round of testing. Maybe just what I was throwing at it wasn't wasn't a good fit. But, you know, I'm eager to see
another model emerge that's at the 5.1 level or better and and sort of take the take the crown away from GLM 5.1. But we've got a couple of other contenders that just came out. So I even haven't even have a chance to touch them. I think it was, like, literally yesterday. But, you know, the the word on the street is that QEN 3.6 is crazy. And, you know, it's it's at a really small
size, so it's something that you could potentially run locally, maybe compete with the Gemma four models that recently came out.
Bowen BelmerI love the the, like, emerging hype cycle of model releases. You know, obviously, like, Morpheus as a protocol is infrastructure layer, but I'm a little jealous of people who, strictly for this reason, people who are working on the model releases just because I would love to just hype up the the new model every single week, like and just see how much buzz you can get going. Because it's ultimately I mean, not to say that they're not hitting the benchmarks. Clearly, you know, the hype's only real if if the thing gets released and they hit the hit or exceed the benchmarks. But it is so funny to watch how,
all the different groups, whether it be, like, the centralized players or the open source players, have to, like, build build the hype and mystique around the drop before it happens. It's, it's pretty fun to watch. The the other big one that I wanted to make sure we touched on was g p two. So, obviously, not a open source model. However, from I did check I was able to check it out, and then, there's a specific individual within Morpheus who was actually the one who initially turned me on to Nano Banana a handful of months ago. And at the time,
was like, this is the craziest thing. Like, you have to try it. Oh my gosh. And and I was, like, blown away when I did try it. That same individual is that level of excited about the performance now of GPT image two. So another really, really big step up in image generation. Has anyone else had the chance to check this out yet?
David JohnstonI mean, what I'm seeing is a whole new level of sort of access to basically, I mean, they look realistic to me. I mean, Nero Banana was pretty close, but one of the things I'm seeing now is, like, the lips and the words actually match. Right? And so you don't have that sort of cognitive dissonance where it's it's slightly off anymore.
Bowen BelmerYeah. That was the the biggest takeaways I was seeing is and, again, not not not looking to, like, promote centralized players, but in turn but the thing is, like, anytime we see something like this where there's kind of a a mat like, a big noted notable jump in one of the aspects of the models, it's just a matter of time until an open source variant that's as good or better ultimately gets dropped. But the ability to something like Nano Banana from my experience, you could take, like, two images at a time and say, hey. Use these as the reference points for this thing that I'm trying to create, whether it's a a flowchart, a specific image, a,
you know, piece of marketing material, whatever it is. And you can now do it with, like, many, many documents at a time, which is just pretty crazy. They're just they're they're getting incredibly, incredibly impressive. So very much looking forward to that coming to open source as well or or something something similar. Oh, the one thing to note on this as well is that these, like, latest image model can also now cross reference different languages. So, previously, something like NanoVanana
was really proficient in, like, English. Right? Like, you could take images that use English language and create something new, and it could kind of replicate English. But now it's doing a much better job, I'm sorry, it being GPT image two, including things like complex diagrams, images, in non English languages as well, which is, of course, you know, something I'm not as a one language speaker, not always something that's top of mind for me, but is obviously very, very relevant,
to to other people. We can keep moving here, David, unless we have anything else on that specifically. No.
David JohnstonJust, like you said, the models on the open source side are only, you know, I would say at this point, two months behind. So, literally, anything you see on the closed source side, it's like a preview of what you're gonna get in the next next sixty days.
Bowen BelmerYeah. Exactly. It's like, hey. This is the thing you're gonna be able to do. Now here's the actual freedom of intelligence version where you're not getting, just farmed, basically. You're not the product. Right? You're you're using the product. So kinda next in line here, there was a teaser for an and I don't I don't even know that I fully understand what this is, and so I'm hoping somebody can explain it. But the teaser for an intelligent
Internet physics model.
David JohnstonAh, yes. This is, one of the models by Imod. So if anybody knows his reputation, he built one of the first open source image generators a couple of years ago and famously left the centralized company because his investors wanted to sort of extract all the value from users and that that went against all of his, you know, ethics and morals and decided to, you know, keep everything open source
and start a new project fully around decentralization. And so, yeah, this is one of their new releases. And, you know, it's it's gonna be interesting to see the models start to impact the sciences. And so this is this is people, you know, sort of playing with, okay. If I give it an understanding of physics, what can it find? What can it discover? Right? You can go through all of the papers and theories and ideas, and you can run
simulations at a level that you you never could before. So, yeah, I think physics is interesting. I think drug discovery is going be another huge one. Are starting to take this and run the agent through their own genome, right? And analyze the data at a level that they could never do before. The gene tests have got so damn cheap, you can go run a $45 test and figure out your exact exposure to whatever vitamins or minerals or
pharmaceuticals or anything environmental. You know, the level of like people have been talking about personalized health for a long time, but it feels like it sort of stagnated in the world of highly regulated health care and so now people are just doing DYI, know, do it yourself for for healthcare, you know. So I think, yeah, models are gonna start breaking into all of these other categories. You know, it's it's pretty amazing to think like two, three years ago, everybody was amazed by an AI writing some some good articles.
Right? And then it was creating some cool images, and then it was coding, and then it was video, but it's gonna get into basically anything where information is involved. So, yeah, I'm I'm pretty pretty excited about that for all the people in the longevity out there.
Bowen BelmerWell, this is another example too of just, like, the incredible need for data privacy. As you were talking, I was just picturing some centralized business that says, hey. Go get your genome tested, and then we'll run it through our models. And we'll collect every single bit of medical information about you possible down to a literal molecular level. But don't worry. We're definitely not gonna sell this information to your insurance companies and your health care providers and pharmaceutical companies and everyone else who can,
ultimately extract value off of that data and information. And so these tools, like, the concepts of these tools are just incredible, and I totally agree. I mean, I I think that we are, without a doubt, going to see a change in how medicine works and, like, what it means to for people who especially for people who want to take control of, like, their own health care to the best of their ability. But with the major key being,
we have to be able to do this in a way where you can keep your data private or, at the very least, only give your data access to the people that you want to have access to it and have that be verifiable, not just, hey. Trust me. I'm not selling your data to UnitedHealthcare and x y z other company that may use it against you for financial gain down the line. At best at best, use it against you for financial gain. At worst, use it against you, you know, to deny
a health coverage that you may have otherwise had.
David JohnstonYeah. I think that's that's that's huge. Well and I wanted to get, to Alan up here. I know, there's been a bunch of developments, going on with, with the router, and and I think that deserves a lot of attention. Al, I know if you wanna talk to her or at least tease what's coming, what's shipping, and why it matters to people.
Alan (Nomadic Rogues)Excellent. Thanks, David. Can you guys hear me okay? Yep. Loud and clear. Super. Well, I've never been much of a tease. I'm I'm I'm engineer for crying out loud, so it's kinda straight up. So as we've talked about here in the past about security and privacy and I think it dovetails a lot with what Bo was just talking about about, you know, medical and private things that you don't want to have out there. We've been talking about our TEE
deployment. And a couple weeks ago, we released version six which had the the provider node, a security kind of attested the provider node. You can know that the provider node itself is not logging or doing anything spurious because we we we certify and verify all the bits into kind of into a trust list of verifiable model. Well, just as of oh, about thirty minutes ago, we released version seven, which is the full TEE stack,
which basically means if you have if you're running a consumer node on your own machine of at least version six or greater, connecting to a team model that is published as TEE. In the tags, your consumer node will verify that the provider node is is qualified and verified and trusted And then the provider node will also examine the model that it presents and verify the CPU and the GPU for the model on the back end. So therefore, if you have your own consumer node and you're looking at a t model, all of that is
private, not exposed or not revealed in any logs. We make sure that logging is turned down so that you don't see the notes. So even the operators of those machines or the operator of the g l of the the GPU and the CPU for the provider node or the language model, that those logs can't even be exposed so that people can't see that. Now they know because it's on the blockchain that the sessions have been created, but the content and the prompts and the inference and the response and all that between your consumer node and that model
is fully encrypted and fully private end to end. So really big really really big release and a lot of things to get here and I think we'll have more of this in the coming weeks. We partnered with Secret Labs and they were they were essential on that. So right now, this works with Secret Labs. We are extending that once we've figured out one. We're gonna try to extend that with other providers. It is very provider specific because of how they publish their golden images and their their attestation.
But we think the, you know, the first out of the gate here is pretty good. So did a terrible job of teasing it, but the the really cool stuff is when you can see it and when you can say, alright, let's let's start poking at it. Let's make sure, you know, can I see this? Can I see that? You look at the code and have your as you guys talked about earlier, have your models look at the code and go, hey, is this is this what it says it is? You know, have have GLM go after it and say, hey, are you are there any holes in there? But seven o for full phase two TEE and you look on you can look on tech.more.org/tech.mor.org/tee
to see kind of the road map and what that means for some more visual detail.
David JohnstonThat's it. Got it. That's that's amazing, Alan. Thank you for, you know, previewing sort of this new release with with seven o. Think yeah. I'm just pulling up tech.more.org right now. It's the button that says privacy, and I see it walks through sort of the visuals on each of the different phase. So I guess this is where it's phase one, phase two, you know, and so you guys are sort of just cranking through each of those phases on the proof.
Alan (Nomadic Rogues)Yeah. I need to I do need to update that road map, but we are all the way at at phase two with you. Last time updated last week was 50% alpha. Phase two is now active with version seven o and that's if you're running your your your convert your own consumer node. If you're running to the API gateway, you do get a lot of good benefits and you kinda model that as you as you are on that that website. You can kinda see where where the deficiencies are. But now we get to work on the consumer side of things. So things like Node. Neo
that is a much easier kind of a signal app of of Morpheus that allows you to have your own little bit simpler consumer node side of things, but we're buttoning up all all that security and privacy.
David JohnstonYeah. Awesome. That's that's really great, Alan. Yeah. No. Look forward to, know, sort of the, the formal announcements and all the details as that gets pushed out. But this is a great preview and, you know, really fulfills one of the big primary goals, of the Morpheus community is to provide privacy in all of this. Yep. You betcha. Awesome. Sounds good. But what else we, got on the list for today? I'll just give one more comment. So much for Yeah. Go go ahead. Yeah. Go ahead, Kyle. Just, I I just wanna rehighlight,
Kyle (Bow Tied Blue Fin)like, how transformative and and important this is. And this is really one, like, one of the initial value propositions that we spoke about for Morpheus about owning your intelligence and having owning it not just from a monetary perspective, but from a privacy perspective as well. And we really haven't seen anyone do this at scale in the industry reliably and and with the proper with the proper validation and proof. And we're just super excited to to get this out.
There'll be tons of documentation and marketing and and all that stuff to follow so that everyone can really understand it at both the technical level or at or at a high level. And I think it's just amazing work by by the engineers and and something that will be incredible for for the industry overall. So really exciting stuff, and, we're we're really proud of the work that was done on it.
David JohnstonYeah. It's definitely worth, emphasizing how big a deal this is and, you know, took took a while to fully build out, but definitely worth worth the wait and sort of delivers on a key key difference between the centralized and the decentralized systems. So because I think there is there's still a common misconception. You know, I still talk to people, and they say, oh, well, you know, if I use Morpheus, it's some random node. Right? And and how do how do I know if that's that's private? And then I explain that you can, you know, pick your provider,
and not only that, you can use this cryptographic approach, to know that it's it's private if you if you want those levels of assurances. So, yeah, definitely need to get the word out and this sort of, you know, makes it this puts a lot of me down the bone when it comes to sort of, you know, showing those things aren't just a policy, but it's it's physics. Right? It's it's math And it's not trust me, bro. It's this is this is something and because Morpheus is all open source, you can check it yourself. So
I think that's, that's a huge difference.
Bowen BelmerYeah. I was just gonna say as well, I I have, like, so many questions and things I wanna cover on TEE, but we definitely need to do, like, a proper, full, dedicated, you know, everything, basically, for that, and we will. But, yeah, incredible work by everyone who worked on that. This is one of those things where from from a nontechnical perspective, I remember conversations probably close to two years ago where the conversation was, this is this is possible. Right? Like, right? This can happen. Right? And now it's like, no. No. No. This this happened and exists today
because of incredible amounts of effort and expertise and time,
Kyle (Bow Tied Blue Fin)from community members who made it happen. And so, seriously, like And could not agree more with what Kyle just said. And and, Bo, you're probably even I don't know if you realize or not, but you're probably speaking to things that I said where I specifically came out and said, I don't think this is possible. Right? Like, we like, two years ago or whatever it was, I went through and I said, I don't think that we can do this. I don't think that we can get fully attestable end to end decentralized,
you know, confidential inference. And the advancements of the tools and the advancements of the the Morpheus node kind of made that possible. So it's been a long time coming not just from hey. We gotta figure this out, but from the advancements in the industry and in the TE architecture and how you can attest and from NVIDIA. And it's it's it's really something amazing that was done.
Alan (Nomadic Rogues)Yeah. It was it was funny. We were talking this morning about, you know, the conversation last year versus the conversation today and I don't think any of us can see the forest, you know, the the the light through the trees. Like, man, how are gonna do this at scale? How are you gonna go all the way down to the CPUs? How are you gonna get the attestation? And I think, you know, our partnership with Seacryk really helped and just take it at one bite at a time, right? How do you how do you eat the herd of elephants, you know, one bite at a time and move through, but it is it was pretty stark.
I think I was one of the bigger, you know, feet on the ground going, you're nuts, you're out of your mind, you can't do this, how are you gonna do this? And yeah, it's to Bo's point, it's not what we're gonna do, it's it's done. Shipped.
David JohnstonA whole 30 a whole thirty minutes ago. You you guys,
Bowen Belmerhere today are are witnessing a bit bit of history. Yeah. No. Absolutely. I mean and and you're right, Kyle. I mean, I distinctly remember us having some of those conversations where it was like, you know, I was trying to understand in the early days, like, what does it even mean really to have privacy in in the ways that we're talking about? And it started out as being, well, we these are the ways that it's possible to have privacy, and then, like, that would be considered private in the sense that it is as private as the technology allows. And now we're talking about
something new that the technology allows that that we, you know, didn't even know for sure that it would. So awesome, awesome stuff, guys. The the the next natural question that I have, and no one has to have, like, a hard answer to this, obviously, right this second. But now I'm envisioning a world where I'm somebody or, like, you know, the general me. Like, somebody sitting there with hardware, and they, you know, they have their, multi GPU setup or whatever in their home, and they're like, hey. I wanna be or maybe in a business entity. Like, maybe it's a data center. And they're saying, wanna be a compute provider
for Morpheus and run-in TEE. Is there a possible world where there's, like, a I don't wanna say, like, one click, but but something to that effect where it's like, hey. I can, like, do x y z thing, download x y z thing, and I'm able to now provide compute in a trusted execution environment. Like, is that possible? Because I know you had said right now that there's Yep. Obviously, like, a heavy amount of tailoring happened on the side of the compute provider.
Alan (Nomadic Rogues)It can happen today. As I said, right now, we use secret to to help out you know, it help is the wrong word, but they've already done a lot of the heavy lifting for that. The goal really on the phasing was let's let's do it with one and let's do it with a partner and we learn a lot along the way and then I think we do it with another partner and then see how that goes and see what the variation is And I think once we end up with those two, we learn enough about the underpinnings of
what Intel's done, what AMD's done and what Nvidia has done. They can we then start to look at, alright, if I have my own chipset or my shoot my own my own host, how do I make that happen? So it's on the road map. We've talked about it and blue skied at a bunch is like, jeez, can't everybody be that secure provider? We've got a lot more steps between now and now and then. Yeah. Totally makes sense. And and not to, like, add work the second you ship it. I'm more just thinking, like, what is the theoretical end game? You know? Like, what is the Yeah. But it it would be it would be nice to have anybody who can host that kinda hardware to be able to provide and attest and be trustless.
That would be amazing. Okay. Sorry. Think I stopped on you. No. Stopped on you. Let's
Kyle (Bow Tied Blue Fin)just be clear that if you're providing, you know, model through TE, it means you have, like, an eight by h 100 node. So it's it's we're we're a little bit limited in the types of people that we're talking to who can do this. And if anyone is running an eight by h 100 cluster with Intel TDX, then we should definitely talk and and, you know, work through that development with you on on how we get this running over over the next however much time. But
a a normal person with, like, a couple, you know, 50 nineties or something in their house, the the that hardware is not natively capable to do this. So I just wanna I don't wanna set the wrong expectations now of we're going to have, like, you know, a million people running TEE in their cost. That's not really the the application. So I just wanna kinda get that on the table, and we're talking about larger, more robust data center level applications for, TEE at at scale.
Bowen BelmerTotally makes sense. And, yeah, sorry. I didn't mean to, paint paint that picture, but I actually did have, like, the retail person at home. How however, I I will say that it's it was one of those things too, and we've kinda talked about this before that at least as it exists today, the economics behind something like inference in TEE is a lot more expensive for the exact reasons that you were just outlining. And so it's, from my perspective, one of those things where it will be amazing. The the point is, like, it needs to be possible to do it because there are certain things that are worth that cost. So, for example,
your your health records and things like this that we were talking about before. But not every single bit of inference needs the same level of security that, like, that might need in the same way that not every single transaction on a blockchain needs to be as fully secured and decentralized as, like, Ethereum layer one. Like, me, you know, sending you 5 USDC for copy is not the same thing as hosting a, you know, massive global enterprise on on Ethereum. Right? And so it it kinda the same thing with regard to inference and the need for privacy. However, the ability to literally access it if you want to and choose to is, like, the massive unlock here.
And and just you know? It's so exciting. We're, like, some of the only people in the world that can do this. Like, that's crazy.
David JohnstonYep. Well and it's worth pointing out that if you go all the way back to the original Morpheus white paper, right, the insight is the average person ought to be able to access to this, but they don't have the hardware. Right? So that's where Morpheus comes in, right, is to put in the hard work and deliver these kind of capabilities to people that can't drop 20 ks on a high end GPU or and they can still get access to this for pennies,
right, by staking to a session or slashing their credit card on the gateway. Right? And it's cool to see I'm I'm poking through the throughput, you know, and it's on tech.more.org right now. And it's cool to see the usage cranking up because that's that's the whole value here, right, is making this accessible to everybody without a giant economic barrier. So, you know, e even if in, you know, some distant future,
a bunch of people can afford the hardware themselves, most people still won't, right? And there's also an intellectual barrier, there's a technical barrier, there's the time and energy to get it all set up barrier. So what Morpheus can do is bring forward those benefits to people many, many years before those barriers fall low enough for them to do it themselves. So we're living in the future. Yeah. It's just not evenly distributed yet.
Bowen BelmerExactly right. Awesome. Well, no. I'm I'm really glad that we touched on that. We did have a couple other things of note if we and we got trying to think of deciding between two. Would we rather cover certain, services that shall not be named yet, cutting off access to users or large acquisitions? So you would have a opinion on where they'd like to start?
Brian BourgerieLet's do the large acquisitions because there's always people cutting off users Okay. And more often than the acquisitions.
Bowen BelmerFair enough. Fair enough. Yeah. So we only good good point. I love the logic. So, yeah, we only have a couple minutes left here, but, fairly large acquisition or potential acquisition was announced with x AI acquiring the rights to buy person, which is a massive, massive distribution play for, x AI, ultimately.
David JohnstonYeah. I mean, this seems like a a big deal. 60 billion's quite the price tag, but, also, you know, it continues to sort of consolidate the different players into different camps. Right? So you had OpenAI buy OpenClaw or at least hire the main developer. Right? And sort of, Claude's gone its own way blocking all the users, whereas XAI and others are are trying to forge these, these partnerships. So it'd be interesting to see. XAI's got a lot of compute.
Cursor's got a pretty amazing product. Right? Couple that together. Now the rumor is they might also acquire Mistral, which the Europeans might not like that. So we'll we'll see what happens. But, yeah, like, it seems like the consolidation is is definitely here, and all the big players are starting to, to gel together.
Bowen BelmerYeah. Exactly. And I think that's a good way of framing it. It's like the the Game of Thrones of centralized AI, and they're all planting their flags. I will I mean, I'm I'm certainly not not particularly partial to any of them with the exception of the fact that, you know, x AI with now being owned by SpaceX is, you know, the the colonization of the war of the, future I was gonna say world, but new new worlds, I guess, is gonna take a lot of compute, a lot of intellect, some robots mixed in, and a lot of spaceships. So if if anything, it's just exciting to see that come together. Yeah.
David JohnstonAnd maybe a million satellites is what they've, requested. And, apparently, they're gonna stuff a lot of GPUs, into that star Starlink constellation, you've basically got AI inference available anywhere on the face of the earth, very instantly. It's it's gonna be wild. You know, that's still a number of years out. But, yeah, it's it's interesting to see.
Bowen BelmerI'm just gonna call the shot now because, you know, like we said about two years ago, TEE sounded impossible. I can't wait till Elon is hosting Morpheus compute on his space satellite cluster. It's gonna be great.
David JohnstonThere you go. You heard it here first. Awesome.
Bowen BelmerAwesome. Very good, guys. Well, that is all I had on on the docket for this week. If anyone else has anything that they want to share or make sure we discuss before we get out of here, please, the floor is open.
Brian BourgerieI think just quickly, who else is gonna be at consensus in a couple weeks
David Johnstonor a week or whenever it is? I am scheduled to speak, so I will be preaching the good word of of Morpheus in Miami. So if anybody's gonna be there, hit me up. Let's let's hang out. I will be there. Alright. Perfect.
Bowen BelmerAwesome. We'll sound off in these social channels. If you're gonna be at consensus, we would love to, make those connections happen. Anything else, Brian?
Brian BourgerieI guess the only thing I'll say super quick, since we're a little over time is I've been at multiple four events in the last two weeks. Obviously, always repping Morpheus and what can be done, and there just continues to be a lot of interest. Especially, I I had a lot of CEOs come up to me in the hallway at this alternative investment summit. One, just asking about
how they can pay less for their compute for their OpenClaws or Hermes or whatever they're running, but also really liking the idea that I've been trying to put forth for a while where besides covering your own inference, you invest with your inference. So amassing as much more token as you can, And with your unused inference or compute, they have access to providing that to your portfolio companies
and investing that way. That's still a very active topic of discussion and interest or at least curiosity, among the investment class, let's say, which is quite exciting and and just it's it's a if nothing else, it's a fun thought exercise, but I I see a lot of investors and executives of companies asking questions for themselves first just because they're also tinkering.
In addition to whatever their companies are officially doing, they're also tinkering and and doing their own thing and figuring things out as well. So just that's been pretty exciting. So
Bowen BelmerYeah. I I am very excited for a future where that type of fund exists, Brian. And and I think that, like, from the perspective of, you know, somebody who is a future builder that can have their largest expenditure, which will ultimately be compute just completely handled in that way. I think it's it's a it's a genius mechanism. I can't wait to see it play out.
David JohnstonFor sure.
Bowen BelmerAwesome. Well, thank you guys as always, everyone who joined and who's listening to the recorded version. As always, get involved. We ask that anyone listening, one, go out and try stuff. Two, engage with us or with other community members in the social channels of which there are now many. And three, David.
David JohnstonKeep being awesome, guys. Catch you soon. Thanks, everybody.
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...