Geopolitics Meets AI | Brainpower Is Back Featuring Paul Allen (But Not That One)
54 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 #71! Recorded: 30 April 2026 AI SUMMARY: This episode covers DeepSeek V4's strong benchmarks versus top models and its real‑world workflow value, contrasting large open‑weight models with smaller, efficient alternatives. It discusses coding with GLM 5.1, multimodal debugging, and using models like Grok 4.2 for cost‑effective code audits. Claude's outages highlight the need for resilient, provider‑agnostic agent stacks. Updates include Figure AI's robotics ramp, geopolitical shifts in AI access favouring decentralisation and community news like Morpheus migrating to Base and upcoming Consensus talks. Timeline: (00:00:00) Intro (00:02:11) Hands-on coding: comparing DeepSeek v4 with GLM 5.1 (00:04:10) Personal benchmarks: when to switch models and agents (00:08:26) Hybrid strategy: cheap coder, premium auditor (Grok for reviews) (00:13:21) Community SOPs: sharing 10-step build flows in Morpheus (00:16:51) Grok 4.3 release, Heavy consensus mode, and expectations (00:19:42) What broke: Claude outages, bans, and KYC roadblocks (00:22:54) Humanoids ramp: Figure AI throughput and vision+LLM synergy (00:25:38) Manus blocked: geopolitics, talent controls, and jurisdiction (00:31:57) Global adoption vectors: restrictions, app stores, and SE Asia (00:35:29) Sovereign agents: trimming third-party dependencies and risk (00:37:00) Avoiding the cloud tax: build to Base and reduce lock-in (00:41:06) Consensus in Miami: privacy for agents and TEE on Morpheus (00:44:05) Podcast drop: Brainpower S2E2 with Paul Allen (SOAR, Ancestry) (00:47:37) Tech plus policy: making decentralised AI durable and honest (00:49:57) Media: DevCon Buenos Aires interview foreshadows OpenCall (00:51:37) Final notes: aliens.gov, humour and next weeks tests 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, Bo, what do you, what do you wanna cover first, man? There's been, a lot going on. Do you wanna get into, to new models to kick it off? Yeah. Sure. Let's do it.
Bowen BelmerSo the big one that I saw was the release, and this actually happened right, I believe it was Friday of last week, so kinda just after our town hall, which is why we would have missed it last week. But it was the launch of DeepSeek v four. So China doing big things again and doing so allegedly for very, very cheap.
David JohnstonHave you had a chance to, use it yet? I actually haven't.
Bowen BelmerI was looking yeah. The the short answer is no. But I was seeing that it was seeming to benchmark, like, pretty solid. I mean, I'm looking at well, actually, here's another thing while we're on the topic. I it's so funny how as all like, all these things are moving so fast, they're developing, and it's like breakneck speed to keep up with all of the different models that have hit the market. I also feel like it is breakneck speed and extremely difficult to keep up with all of the different
benchmark standards, because there's just like seemingly an endless number of them. But I'm looking at a image right now that was benchmarking DeepSeek v four versus Opus 4.6, GPT 4.5, and Gemini 3.1, across SimpleQA, HLE, Apex shortlist, Codeforces, SWE Verified, Terminal Bench two point o, and Toolathon. And DeepSeek benchmarked
higher on almost every one of these measures than the other three with the exception of CodeForce oh, no. Actually, sorry. With the exception of sim SimpleQA, still had Gemini outpacing it. But but otherwise, in terms of knowledge and reasoning and then agentic capabilities, it seems to be performing really, really well.
David JohnstonYeah. I had a chance to use it, and I got it to do some good coding. It seems, in my estimation, sort of on par with a GLM 5.1. But I I couldn't get it to do some other tasks. So it was interesting sort of where it where it fell over, and I ended up going back to GLM 5.1. The other thing that's interesting is it's such a large model. Right? I mean, depending on which version you use, it's like a trillion, a trillion and a half parameters,
which means somebody needs a rack of 1,500, you know, gigabytes of of RAM somewhere to run this sucker, which is, you know, huge. There's a few other models out there this big, but it's sort of the opposite of where, you know, Gemma is headed in the Google models with their, you know, compressed small stuff. And the GLM stuff is sort of somewhere in the middle. Right? You have to have a decent laptop or a good a good setup to run it, but, like, you don't need a monster server.
It's gonna be interesting to see how all that plays out, but it's good to see more open source models or open weight models getting to that same parity. I'm I'm sort of looking forward to the next breakthrough where it adds more of the multi modal stuff. Right? Because a lot of people are using, and I was hearing this from other developers is, you know, they use a lot of images with their coding. They just wanna take a screenshot of, you know, a bug or an error or console output, right, and drop it into their AI and, you know, go fix it, go figure it out, right, instead of having to figure out how to copy and paste and, you know, get all of that content
to format over to the model. And that's something I hadn't been doing, but makes a lot of sense as a as a use case. So, you know, I think that's gonna be sort of the next, maybe one of the next breakthroughs.
Bowen BelmerIn in terms of these launches, David, you know, I was just touching on this idea of, like, having to track not just the model releases, but also the various benchmark standards. Do you have, like, your own internal David benchmark standard that you kind of test models to? And then also, I I'd love to hear people's thoughts too on for, like, the I don't wanna say layperson because the reality is anyone even listening in my opinion, at least, anyone even listening to this is so far ahead of the curve in terms of paying attention to the developments than, like, what the actual layperson would be. But just in terms of somebody who's, like, using agents and and whether it's in workflows or building stuff, like,
the process for determining when it's time to upgrade and when it's time to change what you're using, I'd love to hear, like, your thoughts or Kyle or Alan if you guys have thoughts on what that looks like.
David JohnstonJust for me, does it work for my use cases? It's really all that matters. Right? So can it go off, follow instructions, you know, follow the standard operating procedure, go build what I wanted to build, and not basically get stuck halfway and me realize it's been, you know, off for an hour waiting for me to poke it. Right? So can it get through my coding standard operating procedure all the way to stage eight where it's supposed to come back to me and ask for permission to deploy?
Right? So that's that's my benchmark is like, does it do my primary use case especially for coding? Because almost all of them can do pretty good writing at this point, summary of notes. A lot of the other use cases I have are are a lot simpler. And so the the real benchmark right now is like, it can it build what I'm I'm trying to build without, you know, me having to babysit it every ten minutes to sort of get it to to keep going.
Kyle (Bow Tied Blue Fin)Yeah. I agree, David. I think the I kinda split that into two categories as well. Right? Like, how much do you have to babysit it is one, and then how many curses do I send in my prompts to the AI per hour is probably a good measure of of how good the model is. But joking aside, I I think that these models have gotten so good that
you're you're I kind of look at it from a debugging side. Right? Like, I'm gonna write code using, let's let's say, Opus and then using GLM five one, and, I'm going to see how many times I have to go back, how many times my tests are failing. Make sure that if a test is failing, is the the model going to produce thinking to create another tool call, to create another action, to go test it again with fixes. Right? Or is it going to need for me to interject? And that's kind of using the same agent harness. Right? It's not you're taking the harness out of the equation, and you're looking at a at a direct comparison.
So I think that they have come a very a very long way, and and, GLM five one, from my experience, is a very, very strong coding model. Yep. As David mentioned, my problem is they they they gotta figure out a couple things here and there. Like, GLM five one isn't a vision model, you can't use screenshots. Right? Kimi or minimax, whichever it is yeah. No. Minimax doesn't support structured output in some ways. Kimi isn't great at certain types of tool calling. So I think that that everyone is really focusing and optimizing
towards coding, towards agents, towards here, towards there, and there's just a little bit more learning that has to be done on their side and training that they have to incorporate to really get the full user experience for the application, like, covered. And it it's you know, look back six months ago, we were nowhere near here. So I think we're we're right on that crux of at frontier levels, and now we just need frontier levels of intelligence and frontier kind of feature parity, we call it that way. But, super exciting.
David JohnstonYeah. And, you know, we've talked about this before. One of the best hacks, that I've seen is I'm sort of supplementing GLM 5.1 with a quick check with a more powerful model. So I like Grok, and Grok 4.2 reasoning is the highest rated debugger in the world right now, and it has a privacy mode, and I can call it with an API via my agent, and they're open sourcing models that are behind the frontier.
Right? So if I'm gonna pick any model, I'm gonna probably go with Grok. And I found, like, for literally pennies a day, I can have my agent call and get an audit done of the code it's just created. And I force it to go through this loop, And it'll go two, three, four, even five times until Grock considers the code perfect, then it's ready to ship. Right? And ever since I've started doing that, you know, the tests all passed now. It it you know, it's a lot more stable. It has a lot less edge cases. If you get a different model to audit it, you know, you can basically do most of the coding with a cheaper model and then end up, you know, auditing with something powerful. But, you know, you're just passing the tokens related to the audit. So the cost of the powerful model is really, really low. Yep.
Kyle (Bow Tied Blue Fin)No. I I agree with that as well. I I think that over the last month, I would say, and I was talking to Bo about this a little bit earlier, I've transitioned for less audits. Right? And it kinda sounds weird when you hear about it, but if you spend more time on the upfront and the requirements and the testing, then you've kind of built the audit into the execution. So it's I don't remember who I was looking at on on Twitter, x. It was it was probably, like, Shaw or someone who
put out saying that every single, you know, aspect of the code should have a test written for it, even if it's a very simple one, when you're going through in that plan. And I was explaining to Bo that my planning was about, you know, eight to ten hours for for something that I was doing. And my execution was, you know, forty five minutes to an hour, for example. And that's because once that plan is so well developed and once those tests are so well developed,
then it's really just going through, and it's able to iterate on itself and over and over and over, until it can pass the tests. And and, then your audits are are rather simple, rather than fundamental, like,
Bowen Belmerinfrastructure audits. Yeah. And just to give more context to that, it was actually pretty hilarious. I was trying to do something that should be, to to most users, probably very, very simple. I just could not figure out how to get my agent to do what I wanted it to do. And Kyle was like, hey, here's literally one line. Ask it to do this. And it was like, oh, turns out you need to know how to provide the correct context so that it's like the let help me help you, right, of of this Yeah. What the agent's ultimately saying.
Alan (Nomadic Rogues)But Yeah. No, seriously though. Appreciated the help earlier, Kyle. Well, I hate I hate to I hate to say it, but it's funny, you're right. You know, comp sci one zero one, you know, thirty years ago was it's all about the requirements and as as a new student, they're like, you're gonna create these books and books of requirements that then, you know, developers are gonna go in and and work to. You know, we kinda got away with that with agile, like, we can just sort of define what the next step looks like. It'll be cool. We're back to requirements. Right? So what you just described was if you've got good input,
you know, you get garbage in garbage out, but if you put good input and what good looks like, oddly enough, the LLM will do what you need to do. And I think your Kyle's comment on the iteration, it knows what the test success, what success looks like, and it gets better and better. So I just think that's funny as it as you look at the LOM evolution as we had, like, you know, the the horror of you're gonna have to spend 90% of your time writing requirements. We drifted away with agile and now we're kinda back to, hey, to get the LOM to do what it wants,
we're back to writing requirements and what good looks like. It's amazing. Have you ever seen that
Bowen Belmermeme? It's the guy handing the chimpanzee the a k 47? No. Yes. Oh, I'm that's literally me with this tool. This this conversation's making me realize though that a really good, like, tool for the community would be some type of standard operating procedure, like check and balance or document or or interface or something that helps people work through those SOPs if they don't already know how to establish them.
Kyle (Bow Tied Blue Fin)Yeah. That's a good point. We should, maybe that could be one of our kind of ongoing blogs or or something along those lines because that's something that really can't be static, Bo. And and the reason is that people are figuring this out on the fly. Right? Like, as the models get better than and as the models change, the way that you go about harnessing them changes as well. Like, we yeah. I mean, we all know the basic process of set requirements, set tests, and then developing them to like Alan said, that's, you know, kind of been the process that we've been following for for many, many years. But I think
it would be good for for us to share our learnings about how to go about these things on a more ongoing basis. I think that would be helpful for people.
David JohnstonYeah. That's a great suggestion. I'll put a markdown file into the Morpheus skill repo and just take my existing standard operating procedure. Right? So that way anybody can fork it, copy it, use it, or improve it. Right? And so I'm up to about, like, 10 steps now. And just like Alan was saying, the whole front part of that is do the research. Right? The you know, then you need the specifications,
then you can really architect the code, then you get to, you know, what tests do I need and actually coding it up, and then the audit and the dependency checks and the regression checks and everything else. Right? So I think even something simple like that would be a good starting place for a lot of people and something that can evolve over time.
Bowen BelmerYeah. Absolutely. I I also I had one other note on the this is off topic, but I just wanted to touch on it before we move on, with a completely new topic. So you're talking about DeepSeek and their launch. Did you see that did you guys see that this was their first model that was trained? It was trained using, I always mispronounce this, Huawei chips, not NVIDIA chips, which is kind of interesting. So, basically, they were able to use non NVIDIA chips and accomplish a frontier model, which is, to my understanding, the first time China's done this.
David JohnstonAnd that might explain some of the difference in performance because I'm sure they're benchmarking on, you know, native hardware. Right? And so if you release the open weights, I mean, it's gonna perform differently on different hardware with different setups. And so there's definitely a gap right now between the benchmarks that are, being set and, like, what people see in reality. Right? It might not be a huge gap, but it's, like,
one to two month kind of performance kinda gap as far as, like, how fast the models are evolving is there's they definitely take a hit when they go on to Western hardware and and sort of less optimized setups. So, yeah, it's it's interesting. Though, I think it's it's something that can be overcome over time.
Bowen BelmerYeah. No. Totally agreed. I just thought it was interesting and, just showing, like, the direction that development are there's like a a divergent path now, you know, officially where it was definitely being talked about previously, but now it's something that's, you know, happening in real time. Other big model launches this past week. So looks like Grok 4.3 moved out of its experimental preview into a full release and has also been performing pretty well. David, I know you are a huge Grok guy. Have you had a chance to dive into that one at all?
David JohnstonI just come out, and I've been on a plane. So I'm gonna I'm gonna try it out, I can report in on the next space about how it performs. But, I mean, I'll be interested to see what they really changed. I mean, honestly, their performance with Grok 4.2 and and I've got a subscription to Heavy. Right? So it's their most powerful version that has 15 agents work on the prompt simultaneously and then come to sort of a consensus, which sort of wipes out any hallucinations
or edge cases because there's independent agents all doing the analysis in parallel, and then they aggregate it together. I found that to be extremely accurate, and don't know what it's missing. So, I wonder if 4.3 just accomplishes that, but with if you use the fast or one of the lighter versions. But I'll have to play around and and find out.
Bowen BelmerOkay. Awesome. Well, I'm definitely, looking forward to hearing on that end. Yeah. Other releases over the past week, GPT 5.5, and then I know GPT Image two as well, which we did touch on that a little bit last week. Was mentioning how a community member who was like the first one to turn me on to Nano Banana when it first came out was and that like completely blew me away, you know, about probably coming up on a year ago now.
But this was like the same type of knock your socks off moment for that person with the release of GPT Image two. And I must say I had I had some time to play around with it and was very, very impressed as well. And I saw I've seen other people post some stuff that they vibe coded with it, like some game design things that just look incredible. I mean, it's another one of those things where the first thing I think of when I see this type of thing is like, wow, art well, first thing I think of is like, wow, artists are gonna totally hate this, but then it really should just be
a tool just like anything else. Right? It would be like saying you should hate photography if you're a painter, which obviously isn't true. But yeah, that one I did have a chance to mess around with a little bit and thought it was very, very impressive. That was kind of it in terms of model releases this week. I know, you know, we touched on this when we first hopped on, but it is kinda true that a quote unquote slow week in 2026 is like a three or four new model release week. I guess there was nothing like huge that was groundbreaking or like anything that broke, which is usually the bigger news on any given week. Well,
David Johnstonwe do need to get into what broke, and that story continues to be clawed. Right? They had another big outage the other day. They continue to sort of ban a lot of users that are, accessing it via agents. They're pushing forward with their KYC requirements, so a lot of people can't access it if they're not from the right jurisdiction or don't have the credentials for it. It's it's really amazing. I'm starting to see a lot of memes where, like, all the other AI companies are just watching Claude, you know, shoot itself in the foot again and again and again.
And, you know, like, people have started describing, you know, their uptime graphic, you know, because it's all multicolored. It's like the Claude rainbow. Right? It's it's not green. It should be all green, but now it's red and yellow and orange and all these other colors. So I don't know if you guys ran into that, but, like, that was my feeds the last few days of just people complaining about, you know, you don't really have to worry about finding break times anymore. Claude will just force you to have a break. Yeah. Built in break times that that you're paying for, by the way.
Exactly.
Bowen BelmerYeah. It didn't impact me directly, but I I have been seeing it on the timeline. And it's funny. Like you said, it's it's, like, gotten to a point now where it's almost normalized and expected, which is not good. We don't you know, Morpheus didn't go down just for those you can check on chain just for those keeping score at home.
Alan (Nomadic Rogues)Well, at some point Well they'll list it as they'll list it as a feature. That's true. Just wait. Well
David Johnstonand and that's pretty critical for anybody that has something, you know, where, like, it can't go down. Right? You know, imagine you're in a health care industry and, you know, you're relying on this tool and, oh, I'm sorry. It doesn't work today. Right? Like, that's that's not acceptable. Like, most professional software deployments are guaranteeing nine you know, 99.99% uptime. Right? Four nines. And Claude is doing one nine right now. It's like 98% uptime. Like, that's that's not something you can build
a real, you know, production level business on top of. So it's gonna be interesting to see, you know, as people have access to tools like Morpheus, right, you can just, you know, never go down. Right? If if a particular provider has an issue, you can instantly switch to another. Right? And that is just so much more robust for anybody that needs that type of reliability.
Bowen BelmerThat's so funny that you use healthcare as the example, David, because I've I've had robots on the brain recently, and I will talk about that in a second. But as as you were talking about the downtime, I was literally imagining, like, you're in surgery, and then the the robot running on the large language model who's performing your procedure goes down, like, halfway through it and just pauses waiting for more uptime. Yeah. That would be bad. Can't can't have that. So, yeah, the good
thing. If if only people were building a solution for this type of thing, as you said. But I have had speaking of, you know, robots, which will ultimately use stuff like what we're talking about as their brain, I I have had them on the mind recently. There has been some news in that world. I don't know, did you guys see the latest out of Figure AI? Are you guys familiar with Figure?
David JohnstonYeah. I mean, other than Optimus, it's the the most developed US based robotics firm. And they like 20 x'd their manufacturing in the last couple of months. Yeah. So I was gonna say, so they're now claiming to be pumping out one per hour. It's a good start. It's a good start. I think it's 24 per day. You know, need to those those are rookie numbers. You need to get those up. But, yeah, like, I could see them getting to you know, I think, you know, the target for Gigafactory
in Austin is something like one per second or one per minute or whatever. So, yeah, I think that's gonna finally start delivering. Same for for Neo. Right? And in the AI context, like, all of these need, you know, an AI to provide any real intelligence. Right? Okay. It can walk around, and it can do some basic tasks. But if we really had robust vision models and you pair that with with a good LLM, now it's gonna be really general purpose. And I think that's what sort of the breakout moments can be driven by. It's not like, oh, Neo can do my dishwasher.
It's like, I want the lawn done, and I want, you know, the room cleaned, and I want 50 other things. And I don't want somebody tele operating it in my house, which is creepy.
Bowen BelmerYeah. And also that plus downtime. Like, you want the security and also the fact that it will be up when you need it to be up. It's not gonna, you know, start I don't even know. I'm trying think of an example. It's not gonna start mowing your lawn wouldn't be dangerous. What's a task that would be dangerous for it to mess up halfway through? I'm sure I'm sure there's plenty.
David JohnstonAnything with hot items like cooking. Yeah. True. True. Like Oh, sorry. I poured all the vegetable oil on you. That, you know, that must really hurt. You know, let's try again. Like, no. Like, yeah. No. There's a lot of stuff like that. Yeah. It's dumping the hot grease,
Bowen Belmeryou know, and it accidentally just drops it everywhere, something like this. Yeah. Exactly. Well, that no, I know. This is something that I'm like very much looking forward to myself as a as a future where these things are just out there, and then you have them you know, like I'm imagining owning your own hardware in that way, but then having its brain be your agent or like a spin off of your agent through Morpheus, and just like how incredibly powerful that is.
David JohnstonYep. A 100%. Well, the other big thing we should talk about is what happened with, Manus. The, the authorities have decided Facebook cannot have Manus, and they would like their Manus back.
Bowen BelmerYes. So, basically, my understanding here is China blocked Meta's ability to make this acquisition. And this happened on their side, not on our side?
David JohnstonYeah. I mean, I guess they've got some committee that's trying to keep domestic talent and IP from going overseas, which is a bit weird because I'm pretty sure Manus is a Singaporean company. But I guess the two founders are ethnically Chinese and made the mistake of visiting the mainland, and now they're not allowed to leave the mainland anymore. So the the CCP has decided no manus. It was it was funny. I was I was meeting, you know, some users at an event in Dallas, and one of them was a big manus user. And it's like, sorry, dude.
Using permission systems is apparently, you know, at the purview of the state. So no man is for you. So I'll I'll be curious to see if it's actually enforceable or whether The US basically just ignores it. Like, I have no idea what's gonna happen. Right? The founders, do they have a kill switch or is the company really truly in Singapore? Like, we're gonna find out. But it's a great example of what we've been talking about for years, which is, you know, if AI is, you know, only available at the whim of the state, then you're gonna get all this fractured,
you know, sort of regionalism where, you know, each country tries to regulate who can access intelligence or who can, you know,
Bowen Belmerprovide it or create it. Oh, yeah. I mean, I think that was over this past two years, like, so far in The US, one of the biggest dodges that has occurred was, you know, early on there were I forgot what the bill was called or their proposal, but it was basically this idea of, like, a federal licensing system for who can and can't create and operate artificial intelligence, which is basically I mean, essentially every industry function probably literally every industry functions that way in China,
which to your point, David, is kinda funny. I mean, I of course don't know the ins and outs of of the founders of Manus and if they were I know the claim is basically that they relocated the business on paper to Singapore, but there you know, there's, like, question as to whether or not they actually relocated. But it it just shows another example of how China can basically with the a huge portion of the East just basically be like, you're Chinese because we said so. So
give us that. Yeah.
David JohnstonYeah. Well, it's definitely gonna affect more and more people. Right? Is, you know, I I don't want my access to intelligence to be, you know, so fragile that, you know, any court or jurisdiction could just turn it off. Right? You know? And and this is sort of what we've seen with the Internet, right, increasingly is, you know, oh, well, people in this country aren't allowed to access that app. Right? And, you know, with AI, I think we have an opportunity to if it's open source, if it's open weight, if it's, you know, permissionless,
then those type of orders become, you know, unenforceable. Right? And people are gonna have a a much more free access to intelligence than, you know, that's the way it used to be with the web, but it's been become extremely and increasingly regulated.
Bowen BelmerYeah. I mean, I this always comes back to I I want people to realize this just based off of, like, this type of story and realize why what we're working on matters. But I always have this huge fear that it takes like something really bad happening to move the needle for enough people to care about it. And it doesn't have to be bad in terms of like, you know, it's not like tons and tons of people have to die or anything like that, but it just almost has to be like something that becomes such a large inconvenience
for people that it pisses them off enough that they're like, oh, actually I understand now why, you know, I I don't wanna use this system that can be taken away from me. Because I'm trying to think, I mean, we kinda saw that a little bit with the whole TikTok debacle about a year ago, where like the Trump administration was, you know, basically threatening. And I think they did for a brief time shut down TikTok until there was a new agreement made, and now there's like a TikTok US entity.
But that was, like, disruptive enough for a lot of Gen Z that they were legitimately angry about it, at least online. But I don't think we've seen that moment yet with AI tools specifically, where it's like, okay, we've crossed this kind of mass adoption curve for people, and then it gets, you know, stripped away or taken back. And I don't know exactly what that looks like, but I think it's probably necessary to win over the hearts and minds of, like, most people.
David JohnstonSo, you know, hopefully that doesn't happen in The US, but we know that's happening overseas. Right? Jurisdiction by jurisdiction is banning certain applications. Right? You know, in The UK, especially in the EU, they've got a ton of regulations, and they've come out and said, you know, oh, you know, these are unrestricted apps, and therefore, it must be banned from the app store in our country. So, you know, just like it was with crypto, you know, it's not primarily US people that needed it. Right? It was the guy who wanted to send money to his family and couldn't get around some arbitrary restriction of his local jurisdiction or the the local currency was, you know, trash.
Right? You know? So it may well be something, you know, built and and flourished in The US, but it becomes a lifeline to people that don't have an alternative. And I know there's been a bunch of restrictions in Southeast Asia and others for different reasons. And so I think it's it's likely to come, you know, from those places where a lot of the demand we see a lot of the demand.
Bowen BelmerYeah. I agree. I guess just not being like boots on the ground there, I'm not seeing it as much. It certainly doesn't mean that it's I mean, it's definitely happening, like, without a doubt. Well, and, I mean, we do have, obviously, a very huge blossoming Morpheus Asia community as well, which, by the way, I love every time they post pictures of events and stuff. It's awesome to see all of the participation.
David JohnstonYeah. No. They were just having a big developer event and posting about that. So it's definitely a global thing, and I think that's where a lot of the adoption's likely to come from. But I assume we got Brian on stage. Brian, don't know if you wanna give anybody an update. What's what's going on with you, man?
Brian BourgerieHey, guys. Yeah. Sorry. I got in a little bit late, but school pickup sometimes takes long. So on my end, like always, I've just been talking to a lot of kinda, like, executives and investors in in general and what they're doing. It's been interesting. I've been seeing a lot of, like, CEOs that are both trying to build their their AI strategies
for their companies, but also personally building out in parallel. And it's just been interesting to see that because at and what I noticed is the ones that are trying to build personally in parallel, whether it's agents or wanting to know what harness to use or or things like that. They're those ones are coming to the conclusion quicker that not being dependent on big tech centralized AI is ideal, at least in the circles that I'm running. I can't speak for, obviously, everyone, but
which is a natural conclusion to come to, especially when you start working with personal data or health data or other things like that. It's just been interesting to see it kinda, like, click for them to where they're like, hey. What was that Morpheus skill, or what was this thing, or how do I do that, or how do I get more token? Or or or sometimes it's starting even earlier. They're like, one of the guys, actually, who we met in Dallas, he called me the other day, and he's like, okay. I need you to walk me through cryptocurrencies,
then I wanna understand the AI stuff better. He's like, because everything I've learned about Morpheus sounds great, I wanna understand it. So, like, I was walking him through just even just the basics of things, and it was resonating because what he was telling me, he's like, I have made lots of money. I can make money. He's like, but I wanna align with something, one, that I care about, and two, that's helping preserve the things I care about. And and that's one of the things that, I guess, for him, has been sticking out for Morpheus, which obviously I resonate, and I think most of us also resonate with.
So it's just been interesting, those conversations I've been having on that front just to see, not just in the Twitter sphere or wherever. And I'm sure you're you've been seeing too at the events you go to. People are starting to get it in in their own right and in their own places. So, yeah, that's that's one insight. Another thing as well and and this is like a a bigger, broader subject, so I don't know if we maybe don't wanna dive into all of it. But
I've been and I was talking to to Rob with CRM three. I think he's here too. But about this is there's so many skills and everything. There's a lot of open source stuff that you can get distracted with these days or wanna pull in and implement or try out, which is great. But many of them just have a lot of third party dependencies that are unnecessary baked in. So I've set up a entire process
as I'm assessing or having my AIs assess what open source model structures, memory, whatever that I wanna maybe implement into my stack, so to speak. I have a specific scan to see, are there any, like, third party dependencies, not just on the inference side, but on anything quite frankly, that don't need to be there. Because many times we'll get we'll default to using just whatever is out there, but it has more dependencies, has more surface area for security risks, and things like that.
So I've just been optimizing my structures and just trying to make sure that my agents are as as sovereign as they can be even beyond just the inference side of things.
David JohnstonVery cool. I appreciate that update, Brian. Yeah. And I also see that Rob's here. Maybe we can get him promoted as a speaker as well. But yeah. Well, it's I think people are seeing now the relevance and the importance. You know, and and it feels like it's worth like a post about, you know, avoiding the the cloud trap that most companies fell into for paying the giant cloud tax in in the web two context. It's like so many people, you know, thought they could use AWS and that it would be cheap, then they sort of rack up the price more and more over time.
And most companies end up spending 50% of their budgets on cloud services. And they don't effectively have an alternative at that point because they're they're locked in. Right? And so it's like, no matter what you build, if, you know, the the cloud companies, you know, can effectively take your entire profit margin away from you, you know, that's not really sustainable. So, you know, I think if we can get people to avoid that trap, right, that's gonna be the the key to them building stuff that they can actually make money off of in the AI context. So, hey, Rob. I see you're on on stage, brother.
Rob (DRM3)Hey, David. Hey, guys.
David JohnstonYeah. Well, welcome.
Rob (DRM3)This week's been a big heads down week for me. Lots of building. I noticed there were a couple of comments in the discord this week around cashew around numbers. I think one of the pages around the builders, I wasn't filtering out the other stake pools in cashew. So I've got that updated and I'm I'm putting a tab up on the cashew for,
you know, anything that anyone finds, you know, David, you had asked about, you know, feature requests or numbers that don't look good. So people will have a place to submit those types of issues here, hopefully, by end of day today. And and I kinda had a I was I was looking into the capital contracts and the code contracts, and I wasn't sure if that's been moved over from Arbitrum yet or if that's still on Arbitrum or or or what the deal is with those two pools.
David JohnstonThat's a great question. I think there's pools on Arbitrum and base and and elsewhere, but, you know, it's sort of all run by the community. I'd have to I'd have to check.
Rob (DRM3)Yeah. I think it might be a good idea to to start maybe drilling down on those a bit. So there's a there's a pools tab on Cashew. It just has the builders and the and the compute right now, but it'd be good to get kind of a full full eye view there. So I'll probably keep keep moving towards that a bit. And then I also heard Morpheus may be out of consensus coming up here in April. I don't know if I'm speaking before we wanted to mention that or anything like that, but
not sure if anybody has anything on that.
David JohnstonThat's a good question. Well, I was just looking up the details and just to answer the question on the pools, the distribution is set to go to Arbitrum right now, but we've been migrating everything towards base. And so the expectation is all the and when when you say pools, we're talking about emissions pools of MOR will eventually be on base. So that adds a lot easier path to integrating the smart contracts going forward. It's been a real point of friction to have them across
multiple chains. So, yeah, that's that's the expectation is is it's all gonna it all end up on base. So happy to send you some details on that, and I think that will really, you know, help, you know, sort of clarify everything for people. So yeah. But that's that's a good point.
Rob (DRM3)Yeah. I was kinda wondering would it make sense to try to, you know, get on top of the Arbitrum stuff or wait for this to to move over to base? Probably, ideally, all on base, but I can I can surface up the arbitrary ones if that makes sense, if that's useful to anyone? Okay.
David JohnstonI I would build towards base given that's where it's headed and it's in the relatively short term. So, you know, the contract for compute's already moved and, you know, so I ex I expect everything will end up there. So no no need to create some duplicate work.
Brian BourgerieAnd I think, David, you're you're you'll be at consensus. Right?
David JohnstonIndeed. It's coming up here next week in May. For anybody in Miami, come and check it out. I'm actually gonna have a whole dedicated panel to how to do privacy for agents. And so that's directly applicable to all the great work that Kyle and Alan and the developers have done on TEE and privacy on Morpheus. So really excited to kind of talk about that to a much broader audience. And so, yeah, I'll be speaking a couple of times at consensus. It should be a good event. I like bear bear market years because it tends to be more focused, more interesting, more substantive,
right, than sort of gets very noisy in the in the bull market. So if you're gonna be there, yeah, come come say hi a 100%.
Bowen BelmerYeah. There's probably not a Bored Ape Yacht Club rave after party this year, I would imagine.
David JohnstonOh, I I hope not.
Bowen BelmerDavid's David's actually hosting one.
David JohnstonYeah. Yeah. I've I've never been to one of those, but, you know, I'm I hear they exist. Hey. I'm So Don't worry. Eventually, there will be some type of
Bowen Belmerthe time is a flat circle. So there will eventually be some type of agentic version of this, and it'll be just as ridiculous as it was that time. So that'll be your top signal, you guys. When we're talking, you know, whether it's next year, whenever it is, and we're talking about consensus, and we're like, hey, are you going to the non fungible agent monkey bot rave party, and we're all like, yeah. Well, we won't all be like that. I might be like that. You should sell everything when that happens. So just reminder right now.
No. That's awesome. No. I totally agree though, David, that in general, all of the events are such high high signal during the bear markets relative to any other time.
David JohnstonI mean, that's what it was like at East Denver. It was sort of like, oh, you wanted to talk to the head of Coinbase's, you know, agentic engineering? Here, go spend an hour with him and it's you and one other guy in the workshop. Right? Like, you'd the access you get during the Dow market is like there's just so much less noise. So, yeah, it's a good time to build relationships, build partnerships, really let, let things scale.
Bowen Belmer100%. The only other news I had this week, which I and I'm happy to dive into any other topics that people want to, of course, but I have to mention our release of Brainpower season two episode two with the great Paul Allen. If you have not had a chance to listen to it yet, please go check it out. Maybe share it with somebody else who you know that will enjoy it as well. Paul Allen was an excellent guest. It was it was actually my first time having the opportunity to speak with Paul. So for those who don't know, Paul is the cofounder of SOAR AI Studio,
who I know is exploring decentralized AI and Morpheus specifically with some of the stuff that they're doing, and was also a co founder of Ancestry, or you might know it as ancestry.com. And so usually, when we go to do a recording, right, we'll, you know, we'll schedule the recording, we'll have him hop on into the digital studio, say, hey, how's it going, you know, for a second before recording. With Paul, I had to, like, emergency start recording immediately,
because, like, the second he started talking, everything out of the guy's mouth was interesting and insightful. And so, yeah, it was it was a ton of fun, Really, really great conversationalists and some really fun takeaways on just kind of the climate of everything. We talked a lot about geopolitics and economics. And so if you're into those types of things on top of decentralized AI, I think this will be kind of a a episode for you. Well,
David Johnstonand anybody scratching their head, this is not the deceased billionaire Paul Allen from Microsoft fame. Oh, yes. This is this is this is the freedom technologist, creator of ancestry.com, and Soar, and all of these cool projects, Paul Allen.
Bowen BelmerSo, you know, we'll we'll call him Freedom Paul instead of Microsoft Paul. That would have been such a funny episode name featuring Paul Allen and then in in brackets, but not that one.
David JohnstonBut not that one. Yeah. No. Paul Paul was an amazing interview, and we really went in-depth, you know, the way his team is, you know, accessing inference, you know, and their support for for Morpheus has has been great to see, you know, featured at an event they recently hosted credit to to Brian and others that pulled that together in Dallas, you know, and put that in front of a bigger audience. You know? There's just there's a people who are building that type of application
are hungry for, you know, a solution that isn't spyware. Right? That doesn't, you know, defeat the whole purpose of of what they're doing. Right? And so, yeah, that's that was it was a great episode, and and looks like we got some other big names coming up pretty soon. Yeah. Absolutely.
Bowen BelmerOne thing I really did like about Paul's perspective as well was both both from the, you the work they're doing with SOAR and and other adjacent things, and then also just the work being done by people he's connected with, it really touched on, like, the regulatory and policy side of things as well, and the importance of making sure, of course, that technology is done right, but then also so that the correct inroads are made with policymakers and design
of regulations and and all these other things as we look to the future, which are in some ways equally well, I was gonna say equally as important to the technology. They're not in the sense that the whole point of the technology being decentralized is that it, you know, isn't strictly reliant on, approval of anybody, let's say. But it is, of course, important. Like, we you know, I I I used to say this with regard to just like anything that lives on a blockchain.
I don't support it because, because of the ability to, like, bypass the system. I support it because it's the best way to have an actual honest system. Right? And, like, decentralized AI has that same future ahead of it if the technology is done correctly and if the policy and regulation is adopted in a way that makes sense. And so this kinda touches on both of those things at the same time, which I thought was really insightful.
David JohnstonOh, I'm I'm perfectly fine going around evil and corrupt systems. But, yes, it also puts pressure on them to be more honest over time. I I think you can have both. That's that's okay. Yeah. Well, I mean, like, the system itself is inherently neutral.
Bowen BelmerMeaning meaning the the decentralized systems themselves. Like, as tools are inherently neutral. But they create the ability to hold accountable the corrupt evil systems that you're referring to, which I which I agree with. But I'm Yes.
David JohnstonThe more the more things we can move to math based, you know, deterministic consensus that is neutral and honest and, you know, really creates property rights that are, you know, rule of rule of law in the sense that code is law, yeah, the better versus, you know, okay. We're gonna use, you know, force and violence to instantiate this this system or this order. So the more of that, the better. The more math, the better. So Yeah. Absolutely.
Bowen BelmerThat was from my side, that was actually the only other thing I had this week that I wanted to make sure we mentioned it. Take any opportunity we can to promote, obviously, on the I'm mostly kidding. But, yeah, we do have some really great upcoming stuff to that end as well.
David JohnstonYeah. Next interview is gonna be epic. One other great podcast, which, you know, is prominently featuring Morpheus and me wearing my favorite Morpheus logo was the one from Buenos Aires. So the DevCon interview we did with JenLayer just just went live. And so they did a really good job of cutting and clipping and, like, putting together a lot of the highlights. So they're just, like, churning out more and more posts about that that interview. But it was really, really cool discussion.
And it it's a funny sort of, you know, moments or capsule in time because that's November before the OpenCLaw moment. And we're talking about exactly what OpenCLaw would become a couple months later. Right? I want a bunch of agents. I want them to do recurring tasks. I want them to work across all of my tools. You know, and and we're we're talking about all the pieces that existed right in that moment, but hadn't yet sort of crystallized
in the open call moment. So it's it's highly relevant now and, you know, at the time was sort of very future looking, but then only a few months later, it became reality.
Bowen BelmerYeah. That's why we have to and we do our best to pump out the content, like, just in real time as it's happening because it it moves so fast. But I do love the like, this example and things like this where, yeah, you're literally predicting the future, which is just so awesome. I I I just think all we're gonna continue to see that, though. I think that all of the people again, this is just now, you know, Bo Balmer's opinion. But I I think decentralized AI as a system is just, like, so far ahead of the curve on the vision of what's coming.
And it's it's so cool to see.
David JohnstonYeah. Well, if you build the future, it's easy to predict. That's the that's the only hack. So just make the future you want. That's, I think, why we're all here.
Bowen BelmerI I had only one other topic actually jotted down on my sheet, and I don't even know if it's worth getting into. But if let me see. Does anyone have anything else that they wanna cover, I guess, this week before I dive into goofy? Oh, goofy is maybe not the right word. Potentially goofy stuff? Going once? No? Okay. Aliens.gov. Keep an eye out. We're two we're two and aliens again. So that's yeah. I just wanted to mention that. I'm sure everyone's kinda seen that by now, but if you haven't seen it by now,
it is very funny to me that, you know, people constantly joke of, like, the, like, look over here. There's something big happening, so we need to, like, pull the attention cord. And for a long time, the joke was that, like, aliens was the thing, and now it's we're just fully we're just fully on aliens.gov, which is just hilarious. Like, we just really have gotten to, like, the end of the spectrum with that. So
David JohnstonOh, yes. It it feels like the joke should have died during during COVID. But but, yeah, it's you know, we'll we'll see what they yeah. It feels very much like a distraction. It's like, hey. Look over here. Like, you know, don't don't worry about, you know, these other problems. Yeah. It's like, okay. Well, I I well, let me be clear. To people listening, I want it to be real. Like, I I very much want that to be real and and all of the declassification
Bowen Belmerof information and and that topic in particular. I think there's plenty of stuff that, you know, we don't necessarily know about, and I would love to. But it's just the timing and the way that it like, the announcements are coming and all of those things that are just like, I can't help but laugh because it's just so it's just ridiculous.
David JohnstonYep. Good times. Alright. Well, no. This has been awesome. Thanks for cohosting Bo, Kyle, Brian, Rob, Alan, everybody that, jumped on stage. You know? A light week in AI is only three, four, five major announcements and, and new things going on. So, sounds like we'll have plenty to, talk about next week. I'll have to, go start poking at Croc four point, 4.3.
Bowen BelmerYeah. Absolutely. Thanks, guys. And for everyone who joined us, as always, we appreciate it.
David JohnstonAlright. Keep being awesome, guys. See you.
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...