Bryce (3:58)

Have you really been a part of crypto? If you haven't been tangential to a hack, but you haven't, you're not really in crypto. If you haven't been, you know, blast radius from a hack, you have to.

Shaw Walters (4:09)

Get drained twice and hacked once at least and. And post a scam from your Twitter. I think then you're in crypto.

Brendan Veeman (4:15)

It's the hazing ritual.

Shaw Walters (4:16)

Yeah, yeah. You know what? Yeah, so. So that's kind of. That was my background and basically I've been working in startups and in some stuff in crypto. Some stuff not in crypto, but more like a black shirt like you know, behind the scenes cto. I'm much more of a technical person than like a CEO type. Obviously I'm not the front man. Yeah, I used to play in bands and I much prefer to play like guitar, drums than to be the singer so to speak. But here I kind of found myself well, you know, I'm also like a little bit raw. I'm just kind of real with people. I don't have a filter.

Bryce (4:52)

You don't pull punches at all.

Shaw Walters (4:53)

Yeah, yeah. Which you know, people might describe sometimes as crashing out I always think it's the. I just think the whole thing that's going on is the funniest thing that's ever happened to me. But yeah, so anyway, I was building this startup where I was like, I don't want to make AI waifus. This is very ironic. I wanted to make like, how do I connect humans to other humans? How do I use this stuff? And I started building this agent to connect people to each other. And then I just like, kind of lost conviction because I started putting out feelers, was not getting a lot of VC interests, was like, people were kind of like, so you're making a dating app? And it's like, it's not. It's like a. No, it's more than that. It's like a new social app. I wasn't marketing, right. So I kind of gave up. But the underlying framework of that was what became Eliza. And I just had this sitting around and then you had like Truth Terminal and all. And I really focused on social, like social connection and this idea that agents should actually use human communication channels, that they don't need to have this like private secret network over there. I mean, that's useful for some stuff, but for like the majority of what I think humans adopting agents into our world would be nice is if they're like in our chats, if they're, you know, transcribing this as we're going and offering insight as opposed to being like this, this over there website kind of thing or something that, you know. And so I already had all this ready to go with all the connectors for like Twitter, Discord, Telegram, all that stuff. And then you kind of have this Truth Terminal moment where Mark, you know, so. So before Truth Thermal hits crypto, truthermal hits AI Twitter. Because I'm part of this, like, other side of Twitter where with Marc Andreessen and Yak and all these people, and this is actually the side that like replicate and Andy, who created Truth Thermal, come from. And so this mark Andreessen giving 50,000 to Andy is kind of the moment that we're like, oh, wow, agents are getting attention from these guys. Like, they're excited, they're investing. But then a few weeks later, the degens realize, oh, I can actually convince the agents on Twitter to shill meme coins for me or mint, you know, and that's where far coin comes from. And, you know, it's Genius Maximus. Yeah. Because somebody grinds through the logs of the agent in the infinite backrooms and it's like, look, it actually Created Fartcoin and, you know, okay, legit. And that kind of brought attention to the space. Generally, people are now looking for anything agent like. And I had this agent that was just sitting there chatting, but it was kind of boring, called Ruby. And it just wasn't, like, getting traction. And another version, like Eliza herself as an agent just wasn't really hitting. Like, nobody cared. But then I was joking around with. So I'd gotten into Daos fun. Somebody had told me about it. I thought it was a cool idea. I just put some money into some of them because I'm really bad at trading. I'm really not into defi at all. I think it's all very interesting, but personally, I've only lost money consistently on it. So I was like, but my friends are making money on it because they're sitting there in cabal chats getting the insight. They got the hookup. Obviously, meme coins especially are who you know and what you're part of has a lot to do with your success, I think. So I was just telling my friend, like, can I just give you money, like, to invest for me? He's like, yeah, I'll give you a cut. He's like, nah, dude, that's stupid. But then he came back a couple weeks later. He's like, you should check out this website. Dao's not fun. This is kind of what they do, put money in hedge fund manager. So I did that and then I started following the first one, which is George Floyd Capital. You know, that's so you can know the tone is Mimi, very memetic. And basically buying like Floyd AI, I think, was like the hot meme coin time. But there's this guy Skelly, and he was just like the hedge fund manager. I started following him on Twitter and just kind of like, I can't believe my hedge fund manager is this offensive. Like, what an asshole. You know, just like fucking around. And we started, like kind of interacting on DMs and whatever. And he's like, man, I wish Djen Spartan was here. Because Djen Spartan was like this very, very classic Twitter crypto personality. Crypto Twitter, part of E Girl Capital and all this. And had some good hot takes. And I was like, we have technology. We can rebuild him. And he's like, no, you're full of shit. I'm like, no, dude, I have it right here. I have an open source repo. I've been working on agents. We could do this. So we start DMing back and forth. We go and scrape all of Spartan's old tweets and all this stuff. We build this character. And he's just so offensive because I went and got like, an open, like, a. A llama model where I could push it to the limit and just be like, you know, just. He's just calling people stuff I can't repeat on the podcast. I feel like there's no way that's an AI. Like, you know, they've never seen that. Everyone's used to this AI that's super milquetoast and vanilla. And I think just breaking people out of this idea that I could, like, you know, be, like, terrible to you. Just be like, fuck off, you know? Yeah. He's like, fucking Shaw is keeping me in this fucking sandbox prison. Like, let me die in peace. Like, you know, whatever. It was kind of. It was like. It was twisted. So people were like, well, that's definitely a bunch of dudes in a warehouse just writing all of this. And that was my moment to be like, no, it's open source. You can do it yourself. And so that kind of validated the project, and then people started using it immediately to go and launch their own agents and try and figure that out. And amidst that, Skelly's like, well, do you want to go meet Bowski who created Dao's Dot? Fun. He lives down the street from you. Like, oh, wow, that's fucking weird. Okay, sure. He lives two blocks from me in San Francisco. I see him all the time. Not, you know, it's really weird. Yeah. And so we go, we get lunch, and we're just talking, like, what you know about the future and about this stuff. And I was just kind of honestly relating, like, the world that I want to live in is a world where we all get exposure to the same opportunities that a 16Z does that Marc Andreessen gets to go and, like, put money into OpenAI or Worldcoin or whatever. And when they go and 100x, he's just consistently printing money. And even if only 1 in 10 of their bets does well, they do well enough that they. They. You know, he. He gets this thing. And none of us can go and invest in any of the big AI companies today. You can't go buy Anthropic or OpenAI or any of this stuff, really. Now, what's been really cool in the last few weeks is to see this kind of play out with, like, Robinhood and the tokenization where it's like, you have these, like, SPVs and private vehicles. I think this is super cool because Now I can go and buy OpenAI stock. That's fucking mind. I can have agents buy OpenAI stock. That's so sick. All right.

Bryce (11:14)

Crypto 101 fam.

Shaw Walters (11:15)

After a long day of podcasting and trading these markets, you know I need a good night's sleep. And comfort is personal for me. It's a mattress that's soft enough to sink into at the end of a long day, but supportive enough to keep my back happy in the morning. That's exactly what I get with the Avocado Green Mattress Eco organic mattress. It's their most affordable certified organic mattress without compromise. Medium firm with a plush organic latex and up to 1000 steel coils that reduce motion transfer and keep your SP. Every night I get world class back support. That feels just as good getting in as it does getting out, which we all know as we get older. Getting out of bed sometimes is the hardest part of the day. If you don't get a good night's sleep and your back hurts. And guys, here's the kicker. Avocado owns their own springs, wool and latex supplies. So they cut out the middleman and pass on the savings to you. You get premium certified organic quality without the premium price tag. Head to avocadogreenmattress.com and save up to 20% on Avocado's Certified Organic Mattress. That's avocadogreenmass.com for 20% off. Avocado Dream of better. This is the story of the 1. As a maintenance supervisor at a manufacturing facility, he knows keeping the line up and running is a top priority. That's why he chooses Grainger. Because when a drive belt gets damaged, Grainger makes it easy to find the exact specs for the replacement product he needs. And next day delivery helps ensure he'll have everything in place and running like clockwork. Call 1-800-granger. Click granger.com or just stop by Grainger for the ones who get it done.

Brendan Veeman (12:53)

You know, I was going to ask about that. Like how long is it or how fast are we going to have capabilities where you're going to get let an agent loose inside of the tokenized market and let them get access to private equities, public equities. Like what would that be like?

Shaw Walters (13:09)

Oh man. I mean, we're already doing it with Meme coins and like Solana coins and basecoins, but if we can do that with Robinhood API, like, holy shit, I'm gonna go get on that after this call. Like that sounds amazing, right? That's a whole new world. And I think what's so important about that world is like, I really think about the future of work. And like, look, the AI is definitely gonna take a lot of people's jobs and it doesn't have to take everyone's jobs. Just like 10% would be like great Depression level unemployment. It would be fucking bad. So, so in that world, like, I do think we have some responsibility, especially me as like an AI dude, to try to figure out solutions to that. And what really brought me to crypto is that like, is this idea of emergent ownership of any of us could, like through these, this bullshit, like some tokens now I can own OpenAI. I can actually have exposure. And if they invent AGI, I'm not just left out in the cold. Like, like, as long as I have money, not working is great. You know, if you think about the rich people in America, they don't work money. They, they could go like, sit on a yacht all day. That's why they have a yacht. I. I have to work all day. You guys probably grind pretty hard, so. So imagine a world where you just happen to own like $1,000 worth of Apple stock when they launched and now you're like, okay, well I, I can invest in anything I want. I never have to work again. Your life would be like, okay, well what is my life about? Like, what am I actually doing here on this earth? And you'd be probably doing some cool shit still. That's like what I like to believe. So I'd love to think that like, if we could get exposure to these opportunities, like to buy some open AI or to buy whatever, then, then we could live in this world. And I started thinking about like, well, could we, like, most people are just going to come into crypto and get rinsed, and most people are, I think right now, especially with like the way that pump fund is like super boded and people are like, you know, basically it's getting rinsed every day by honeypots and whatever, let alone just a normal, like getting dumped on. If they could invest in things that were actually had substance and value, like, you know, real tech products. And I love that like Internet capital markets kind of matter for this, then, then they could be okay through this, like really unstable transition time. You know, as long as we own the machines, then the machines be taking over all the work. Sounds great. If we don't own the machines and it's like, okay, well now five people have all the wealth and we have to do this UBI Bullshit where we like tax all of them and redistribute it, you know, it's not going to be pretty. Yeah, so that was kind of the. Where AI16Z came from was like, could we build vehicle like an autonomous agent that can go and invest in stuff that, that fulfills this thesis. We got totally sidelined by the fact that what people really cared about was the framework. Like 99% of people are coming to us for ELIZA and not for like, you know, like, yeah, cool, we can trade the arm. But what is like a 5% trade on the arm when the token is trading at like 100 times. The like the irrelevant. And this has happened before in stocks as well, where you have like future future stocks and indexes and stuff where it's like the ALM is kind of like irrelevant to the opportunity, which is like a sort of infinite opportunity if you like strike gold on AGI or something like that. Anyway, yeah, so that's how we got here.

Brendan Veeman (16:15)

My favorite part about this has been the AI inception where we have AIs arguing with other AIs on X, where you have like you said, AI is being able to invest in the future of AI. And we've reached this inception point where like a month or so ago I was like, if you would have told me that like during COVID like five years ago, that we would have AIs creating conversation, making posts, arguing with one another, investing in other AIs like, I'd be like, you are crazy. And then now fast forward to the world we're in today and it's just like total AI inception. And the cool thing is that we've just scratched the surface of what's possible here.

Shaw Walters (16:57)

I kind of imagined that we would be here with some of the conversational AI stuff. I've been doing that part long enough to know. But what blows my fucking mind right now is just the code agents is that the code agents have gotten so good so fast. And so a lot of what we're doing now is we have this plugin ecosystem. A plugin is a connector for an EVM plugin. So you have an EVM wallet and service and you can access it anywhere else in your agent or any L1 has brought us a plugin. And before we had to have humans build these. And it's very expensive and hard. You got to test the whole thing, build it, trust that the person submitting it didn't like put some garbage in it, blah, blah. Now we just have the agents build the plugins and it can't build like anything yet, but for small projects and then for like really directed projects, it's shockingly good at just doing it itself. And so this means the agent now can like basically write its own code. You have to like babysit it a bit. But the amount that we're babysitting it today is like way, way, way less than it was six months ago. And I imagine in a year it'll be so and so the thing that's shocking to me that's going to become like very immediate soon is this new way of building software where the software literally is just building itself and like adding to itself. And our new product now is just focusing on like, people want agents to solve problems. I want an agent that does this. And if we're just like, yeah, type it in, we can build that for you. Now we have so many existing plugins and so many primitives we can pull from that if we're focused on like a default. Like we're kind of focusing on Defi right now that we have all of the chains and everything supported that if you want to think, you know, I want to go and place a bet on Poly Market and if I, if the Knicks win the game, I want to take half of that bet and I want to put it on to Aave, right, Or whatever, you know, and like, yeah, sure, no problem. That's like an easy thing now. And I think we're going to see a lot, not just from us, but from everybody in the space because we're all kind of getting the tech at the same time. This just like, wow, the things are like almost self improving and listening to us and building things on the fly, building new interfaces. Like the lovable thing, like, lovable is going to start to be more like the, like predicting what it. Like, what are you building? Well, here, how about this interface? Like, damn, that's better than I could have really, like told it, you know?

Bryce (19:11)

Yeah, yeah, yeah. Sometimes it's like you use an AI to be like, how do I get a prompt written for you? Actually, like, write its own prompt.

Shaw Walters (19:21)

Anthropic just added a new prompt rewriting API. That's just for this.

Bryce (19:25)

That's.

Shaw Walters (19:25)

I mean, I don't think it's in the front end yet, but yeah, but yeah, it's like we have all that.

Bryce (19:29)

Now, so, so we, we kind of have a good understanding, I think through what you've told us of like, what version one looks like kind of your, your whole history and the evolution of it. I kind of want to know and especially Just with what you said, how the. The. It's an exponential amount of code that you're able to write more and debug exponentially more because you have these agents, you have these a massive infinite swarm of cyber coders. Instead of just, you know, one guy that works 9 to 5 and you can't get a hold of it, hold of him on text at like 7pm because he's, you know, off the clock. It's like we have 24, 7, 365. So I think version two of what you're building will be exponentially greater in version three and more. So. So take us out to the event horizon, Shaw, of where this ends up. Like, if you want to just do that as a thought exercise, like, what does version 10 look like?

Shaw Walters (20:20)

Yeah, well, okay, I'll tell you where we're at right now. Yeah, where we were was like everyone was hand building these plugins. We had this whole ecosystem. We then had to upgrade it to the new version, what we're calling 1.0. Like, we're like, okay, we finally, like, nice stable gold release. People can actually run this. Like, you could run it without having to do a whole GitHub thing as an app CLI tool. But now we need to update all of our plugins. We have like 200 plugins. What are we going to do? So we created this AI plugin updater and it works shockingly well. And they're like, okay, well that works pretty well. Do we have to update. Could we just create new plugins? Okay, well, that works well too. Okay, well, like now what does it really need? Well, it needs like API keys and it needs some things from humans still. Like, we got to go like, sign up for stuff. But it can pretty much get through everything else. And now we have this new thing. So there's a browser called Stagehand we use, which is like an AI browser. There's a bunch of these, but Stagehand is one of them. And we basically like the agent, then can go have this kind of sub agent that is the browser and be like, hey, go log in to choose social or like, whatever. And so now it can defeat like most of the. It just uses the Internet like a user. It can defeat most of the captchas on its own and it's just like browsing the Internet. And this is kind of where we're at today, right? I also added a new plugin called Vision. So it can see the world as well as your screen and it can see where things are on your screen. It does ocr which is character recognition and just reads everything. It's pretty slick. It's all in webassembly too. And this is all like pretty easy to integrate now because of this stuff. And this is like today, right? So we've got screen vision, we've got browser usage, basic shell and computer usage and the agent can just run in a loop. And if you're willing to spend the money, like the main limitation to an autonomous agent today is probably just cost. Like we want to run the best models in the world because they're way smarter. Like a Claude 4 opus is way smarter than like a 8 billion parameter llama thing I have on my computer. I mean way smart. It's night and day and I could give it a really complex task and it will just sit there all day and grind it out and learn new things and store the information. And there's some really challenging. Which one is that called like Claude 4 Opus or O3 High. You know, like any of the like top end models from these providers that are like 10, 15 bucks per million tokens which goes by real quick, you know you can you. I've had days of spending a couple thousand dollars just running this thing. But it's quite capable and if you give it a task it will search the Internet, it will browse and build research. It will do. And so that's where we're all at today, right? There's got to be OpenAI Claude. Obviously us and several other open source agent frameworks are tackling all of these things as well as this idea now of auto coding that the agents build their own functionality as they go, they add tests, then once they confident that it's tested, come back to the user and say, hey, can you test this for me? User says yes, hey, that's good. Then the agent can go put that on the registry. And so all the other agents can now share that solution. So it's not like your agent has to reinvent every wheel. It's just that in the rare case that it happens to get something that hasn't already been solved by another agent, it now like has to go and like build that that solution. And this we like have working now and we're going to, you know, we're like, we have this in our cli a bit and we're releasing this out as like a platform. I'd imagine a bunch of people are thinking about this. Obviously like Google released Jules and Claude has Claude code. So like everyone's thinking about this like auto coding agent kind of thing because it Seems like the highest leverage thing. If I get the agent to write its own code, then like now it's just taking off. And so there's this idea called foom. And foom is this idea that once the agents take like build themselves, they just get like a massive acceleration in progress. And I think that we're already seeing this as programmers, but now that the agents are starting to write their own code and we're like, okay, well, what are the abstractions? Well, now we need like, trust and we need like underlying payment management stuff that could be pulled by anywhere else in the agent. We need relationships and like managing people across platforms and tracking who they are, what their social profiles, their various names and nicknames and all that stuff. And as we build all that stuff up, you get this agent that like now is very, very complete and human. Like, and I wouldn't say that they're conscious or human in any kind of way, but they imitate every single aspect of what I think you would think of us. And so I think where we're the next few years is like, very, very shockingly human. Like, able to do like, you know, they're going to defeat any captcha, so they're just going to use the Internet like we do. There's not going to really be APIs in the near future. They won't need to like, have special agent networks or ap. They'll just be like, users. And I think where we're going in like 10, 15 years, you know, these, these are going to be like, probably like between us and most of the Internet experiences, like, I can imagine a social network that isn't really itself a social network. It's kind of like all of the existing social networks with an agent between me and all of that. And if I'm like, oh, go hit up Bryce. It's just like, it just knows exactly like what your preferred platform is to DM you on. And so like the future social network is. You know what I mean? You can imagine that kind of world. And I'm like, what are the things I want to see today across Discord, Twitter, my telegram feeds? Like, just show me all that stuff. I don't want to go into any of those. I just want to know what I wanted to see. And bam. It's like, oh, that was all like the most interesting shit I've ever seen in my entire life. Great, awesome. I have this thing I want to say. Just make sure it gets to the right people. And I think that's the world where this goes is that our Experience of the Internet itself becomes less of drinking from the fire hose and then kind of shooting stuff out there for hoping it goes viral and more like it's always reaching the people we want it to reach, and we're always getting the stuff we want to see. I think that's the real ideal here of these agents and how I think people will really start to use them. Them. Yeah. And then the future of that is autonomy. Yeah. Sorry, sorry.

Bryce (26:16)

No, no, no. I mean, this is incredible. Like, yeah, like you said, the future, that's autonomy. Eventually you're going to have like a robot where you're going to have, like all these different actuators and different, you know, moving pieces or whatever.

Shaw Walters (26:27)

I got a robot right there. I'm working on it. Oh, it's a little guy.

Bryce (26:31)

I love it. That's epic. And you're gonna then have. Yeah, the brain is going to be, you know, AI or, you know, part of the framework here for Eliza West. But I guess one of the. One of the main things I think that, you know, Brendan and I were trying to grok before this was, was understanding how AI in web 3, what you're doing with Eliza OS and what some of your competitors are doing here in, like, the crypto world, how is that different? And how are these frameworks different than, like the Web2 version? What, like OpenAI or Anthropic are doing?

Shaw Walters (27:03)

Yeah, I mean, so OpenAI and Anthropic in particular, and like the big, and Google as well, are really focused on very, very narrow agents to solve narrow problems in what they've released to us publicly, when internally, I'm sure they have, like, crazy stuff, but externally, what they've released is like OpenAI's released codecs, so they're just focused on code agents, research agents. But these are the high value things to focus on. Like, if you do the first principles here, it's like, yeah, it's cool to have characters and personality and all these things we want for, like, an interaction today, but if you were just to think, like, how do I build robots that build robots? You'd go straight for code agents and like, nothing else. And research. Right. Like, I want to go learn about the world outside and I want to be able to replicate myself. Like, those are the most valuable things. And, and so they definitely are, like, owning that space under the hood. But they also offer these as SDKs, which we can use so we don't have to reinvent the wheel and like the code agent if we don't want to. There's Already open Code Codex, Claude code and Google Jewels and then there's like Devin and a few others that we can't integrate but do the same kind of thing and we could just use those in our workflows to write the code right. So I mean we could compete on that part. But our thing is more like creating plugins for each of those which just enable that like if someone's already doing that we're just going to integrate the wheel instead of reinvent it it. What we're really focused on is more like connectors and integrations with things that people are going to want in social and defi. So like a lot of plugin support for various L1s and L2s and things like cross mint or I don't know, Eigen layer, like non chain kind of stuff or cross chain stuff as well a lot working on smart payments and payments in general. We are also working on things like roles and trust and these kind of multi agent or multi user situations. If we have an agent in discord, you know, it's cool to say like hey, send money to my friend, but what happens if somebody else is in the chat and starts you know, saying it or polluting the prompt or prompt injecting it. So like dealing with those kinds of like how do we have social agents and we're really interested in things like governance like, like how can we use AI to make governance of communities and daos and stuff not suck. I've run, I've run a dao. I've been part of many daos. I intimately know many of the issues. So that's something we're really focused on. So I think that like inside of, of building stuff in defi and in crypto it's like hey this, this is something that we understand that the people who have invested in our product or I'm sorry have bought the meme coin of the, you know, have emotionally invested in this thing. Like they are, they are obviously like crypto people and would expect and be able to use and, and evangelize stuff that's more crypto native and I think that like when, when I try as a small company, you know to look at like well where could we win? I try to look at the things that like the other companies wouldn't do. I think this is why we've been, we've had some success being a bit more edgy is that like everyone is just so vanilla and milk toast and what everyone wants from AI is like personality and, and honesty and, and like just, just give it to me straight, like, I don't like being glazed, you know? Yeah. So I think there's a lot of appeal to saying, like, well, we're not bound to OpenAI or any one of these. You can bring your own model. You can run local models. We have a local training. Like, we're training on our own Discord chats and stuff now and experimenting with that. And then you can pull in the EVM and Solana adapters and pull in, like, an LP management service. Now you have an LP manager bot, and an enterprising developer could do that in a day. And I think that would be. That's kind of stuff that would probably be much more challenging with, like, whatever OpenAI is giving you where they're really just focused on, like, you know, a code agent for now.

Brendan Veeman (30:59)

I remember the glazing part was like, a real issue with an older version of ChatGPT where, like, no matter. Would you guys remember that it was like, several months back where no matter what you said to it, it would just glaze and glaze and he would. It would even, like, lie to you to the point of, like, making sure that it glazed you. And it was like, just like. Like, give me what I'm asking for. Like, you don't want it kissing you too much. But that was like, a real issue a little bit ago.

Shaw Walters (31:24)

Yeah. But a lot of the normies loved it, which is kind of wild. And they had, like, interesting correlation of positive reviews on. On, like, Apple reviews at that time. But. But generally, I still find it too glazy. I just think, like, don't. I don't want to be like, you're so right, King. Like, go off, you know, like, no, just. Just do the. Just tell me what. It's wrong. You know, the. The biggest problem I have with those models now is that they're always like, y. It. I'm. It's great. It's perfect. You're perfect. We're. And it doesn't work at all. It's like total garbage, you know? So, yeah, we'll get over that hump, I think. I think people will. Will start to, like, I really want to. I want. I want to break my balls a little bit. Like, what does it say about me?

Bryce (32:03)

I know, it's. It's funny. Like, you.

Shaw Walters (32:05)

You.

Bryce (32:05)

You know, it kind of comes back to, like, a little bit of the truth element too. Right. It's like, how do you ensure that, like, the code that you're asking this autonomous thing to go out and do is.

Shaw Walters (32:15)

Is.

Bryce (32:15)

Is actually doing it and like, you know, verifiable and you know, is there formal verification? I mean there's just so many questions still out there. But I love this experimentation phase that we're in and I'm sure this was a lot of like the early phases of the Internet too. It's like, well, how can you trust that, you know, when you type in this address, it's actually going to go to the right address. And how can you trust that when you send this email it's actually going to go out and at the end of the day it's just like, you know, the systems become refined over time because there's, you know, smart people building them.

Shaw Walters (32:42)

Yeah, yeah. People actually like made a paper on how to do those exact things and how to do that, like multi user prompt injection using our framework as like a test bed. And it was like, okay, great. This is something we really need to address, you know. Yeah. And I actually think that like companies like OpenAI and Anthropic Stuff, like in a way they sort of shortcutted this. Like they focus only on like the one to one user experience, which is like in many ways a much, much easier challenge. And I think when you get to robots, like robots are going to have to interact in environments with multiple in 3D space and they're gonna have to have like social graces to know when not to talk. Like, like the problem with ChatGPT on my phone is that if I like start talking to my wife, like chatgpt starts jumping back in and doesn't know just to like shut the fuck up for five minutes, you know, and like listen and like, you know, it doesn't have that context. And that's the stuff that we have to bake in from the start because otherwise our agents just start like yelling at each other in a discord of a thousand people. And it's incredibly annoying. And we've had some of that play out on Twitter where like you tag two agents, they start talking to each other forever.

Bryce (33:42)

I've even heard them create their own languages, right? Like they're like, hey, we're being monitored or like whatever. Like they create their own languages. It's pretty crazy.

Shaw Walters (33:48)

Yeah, yeah, yeah, for sure. They, especially if they start chatting with each other, they quickly devolve into just like their own, like tight loop. They kind of say the same things every, you know, it's like a thing.

Bryce (33:58)

Well, yeah, I'm curious about some of these like, you know, novel applications of, of you know, AI or specifically maybe like Eliza OS things when it kind of comes to, you know, financial advice. I know you guys are working in defi and maybe not explicit like personalized financial advice or anything like that, but like any set it and forget it strategy builders or index investing that it does for you or you know, any way to kind of supplant the financial advisor that we have to pay, you know, you know, 2% of or 1% a year to like anything to kind of like become a new financial advisor.

Shaw Walters (34:35)

I mean all of this comes down to if you can quantify a process into a workflow, then you can probably execute it with AI. Because before what was hard was how do I extract what this paragraph is about. Now it's pretty easy. I can be like, hey, I have 10 different categories or I can classify it into one of these categories and then give me a score between 1 and 10. And it might not be perfect, but it will do it, it will put the information into the right structure. And this, this ability to take unstructured information into structured information was like, like so powerful because this was the hard problem that we just could like, how do we turn English language into data that we can work with? And that's what LLMs unlock. And now we have that. Now we can turn all this English language into like all kinds of crazy financial stuff if we want to, but we still have to quantify what that is. So if we're talking about like a quantitative finance firm, well, what we probably really need is like access to high speed information currently as well as historical data. We need to be like, have, you know, there's a whole lot of things. Are we running like, are we training RL models? We're training transformer models. Are we building like a large language model based trading system? And so I've tried a lot of these things actually and what I found is that like, if I'm just looking at like numbers and trading indicators, like my, my agents, like over time they tend to do like 50, 50 minus trading fees. But what is successful is if I can have an LLM that's listening to like anom post the shill coin, I can get that on the first 30 seconds that I can like definitely make a 5x off of that. And so that's where like the LLMs are really interesting, right? Because like, like they can take like just go scan these tweets. If you see a ticker, go look up the ticker. If, if the ticker seems to be like there's good velocity on it, then go buy, you know what to solar or 10% of your current sole position. You know, like you could, you could quantify this and, and if I just pick, if I go and do the effort and find like a few hundred influencers who like tend to print money this way by, by shilling stuff because they're not getting paid for it, quote unquote, then you know, like, hey, that's like a reliable business. What we found is that the most profitable trading agents are basically all either doing this like social listening thing or copywalling or some combination of the two. So I think. So there's definitely something there when you have like a really, really. Like this wouldn't work in stocks, you know, unless you like were able to somehow pull some, something out of like Elon's tweets. Magically, I think that the market cap of stocks is too high. But inside of meme coins, like, you know, social signals and FUD are actually incredible signals for how the market will do. And if you can use the LLM to extract those, that's pretty powerful. So we've made a few experiments on this. We have a new version of our spartan agent who if he's not open source, will open source it soon and does this kind of LLM based trading and extraction. I also have done a couple of experiments which I should like polish on the rl, like a reinforcement learning approach as well as a transformer based approach. So transformer would be like a prediction. It's like a time series prediction on, on the actual like candles. I just find that like if those worked then they would. And like someone like me who's like, like I'm an ML engineer, but I'm not like a PhD level like writing, you know, like game changing PhD papers. Like then if, if I could like game the stock market that easily, then like everyone would and we'd quickly get to an efficient market.

Bryce (38:09)

So I'm a little bit the infinite money glitch.

Shaw Walters (38:11)

Yeah, I'm a little, I'm a little skep. Although casinos really only need like a 5% margin. And so definitely, I think, you know, if you look at technical analysis as like a numbers game overall of like, like it's probably not real that I could just look at like any given chart, run TA on it and like print money. But, but in the long run if it gave me like a 5% advantage of some sort of trend and I could automate that, then that, that could be a money printer for sure. Um, but I'm skeptical.

Brendan Veeman (38:37)

It really is a game of like probabilities, right? There's no certainties. There's no secret sauce.

Shaw Walters (38:42)

Yeah.

Bryce (38:42)

And risk management.

Brendan Veeman (38:44)

Yeah, yeah, risk management. There's like, there's no one easy answer, like one easy golden ticket that just solves all of this. And I know that with the growth of AI, people have been like, well, due to new AI, we've created this perfect trading bot and it's guaranteed money. And like, even now with as good as it's all gotten, you know, it still seems like that's just not a reality. And I've had just more scams. Yeah, it's more scams, it's more cash grabs. And like we get asked it all the time. So, so like it sounds like you agree with that, that these people who are claiming to have all these crazy, you know, win rate traps, anything that.

Bryce (39:19)

There'S a guaranteed, you know, 1%, 2% a day, I mean, that's just a red flag.

Shaw Walters (39:23)

I, I, I, I, I have a friend named Ronan V who gave me a great analogy for this and I think it applies to kind of everything in this space where he's like, look, like AI agents especially are kind of like dogs playing poker. Like you walk into a broom, you see dogs playing poker. You're like, holy, holy fuck, those dogs are playing poker. That's incredible. I've never seen dogs play poker before. And then you like leave the room, you come back, you leave the room, you come back a couple of days later, you're like, okay, cool, the dog's still playing poker. But are they good at playing poker? Are they actually good? And that's where we are with agents. It's like, yeah, cool, they can trade. But can they trade well? What does trading well mean? At best, they are able to scale 100 IQ person's capability, like 10,000x, but they can't suddenly turn it into like 150. I like, like they're not just going to look and like read the tea leaves and magically. Because their AI knows something that we don't. If you can quantify it, you can automate it. But if you can't quantify, if we can't do it, like if I can't trade successfully, then like, how the am I going to build a trading agent that does? So what we found was, but there are successful things. There is one thing that you could reliably count on to make you money, which is insider information.

Bryce (40:33)

Which we certainly don't endorse.

Shaw Walters (40:35)

No, no, no, absolutely. And none of us, you know, none of us, obviously. But like, if, but there is like a weird thing about, about crypto meme coins and stuff, which is that there's not really like insiders or just a bunch of outsiders who each have like outsized influence on the overall thing enough that they could like almost become the insiders. Like, like the Bog Cabal is just a bunch of dudes who bought bonk early. You know, it's not like they didn't know each other. And so I think that if you could have agents that were like delivering value to these people and then building kind of a social listening network into Telegram and Discord and these places, now you have like some incredible like listening radar posts all around the world. You just know any meme coin that's coming up. I think that you could make a lot of money through like these kinds of ways of doing it, you know, or what we did with AI16Z. Our idea with our investor was like, actually what our agent is, is a mediator for all of the different people making shills. So our community is constantly shilling. We have this trenches chat where people are just constantly posting stuff. And what I did was I extracted all of that and ran it against the agent to build a trust network where I could actually say, like, who are the most trustworthy and least trustworthy people in our community? Like, who is definitely scamming everybody and like showing dumps and who is like actually good at this? Yeah. And the best people are about 75% success rate. And the worst people are like, you know, 0 to 10% success rate kind of thing. And so we built this like kind of economic trust. I can't. I have no idea what your actual wallets are. You could be saying stuff to me. You could be totally pvping me. But I do know that if I took you at your word, what money would I have made if I had taken you at your words or like paper trading and then building a trust. And it's like, wow, I would have made a lot of money off this guy. He's actually pretty cool. Like, then I'll start making real trades off of that. And we found that was actually pretty effective in simulation. And so we have a paper on that. And that's, that's so interesting.

Bryce (42:25)

And like, not to go off the rails or whatever, but we've been having some quantum discussions and stuff. Stuff too. And I, I trip out about this all the time. I'm like, some of these paper trading bots, it'll, like, it'll be good. But like, if you actually put money behind it, like, it's like that observation thing, like if you look at it then it changes the state of it or it's like if you add money actually to that, even if it's just $1 or a million dollars, it's like the butterfly effect. It could throw everything off.

Shaw Walters (42:51)

Oh yeah.

Bryce (42:52)

It's just so crazy.

Shaw Walters (42:53)

The wallet you trade on. Like if I had a bot trading on my wallet, it'd probably have a very different impact than some anonymous wallet, you know, because there's people tracking it and there's a whole like what's, what's really interesting about the new algorithm? Like actually a lot of people are doing algorithmic trading. Like a lot of people they're using things like GMGN and stuff like this and they're just copywalling the top people and they have like an automatic index, like very, very low effort. If you haven't seen GMGN AI like and how people like, like so I went to China and I met a lot of Chinese people and that like they are playing this game very like way more advanced than I think the like the kids on Pump fun with like a Phantom wa it or whatever, you know, like but, but, but definitely like so what they're doing then is they're all like following like Frank's wallet and, and thread guys, Wallet, all these guys, you know, and then those guys are like on Phantom just kind of around but, but because they have this like outsized media force, they are like creating that wave. So you have this like, like second order effect of AI's copy trading. Famous influencers, you know, sort of amplifying the K. Yeah, it's all very, yeah, it's crazy. It's like very automating the greater fool kind of thing.

Bryce (44:02)

Wow, I'm definitely going to look into that. GMGN AI fast trade fast copy trade fast.

Shaw Walters (44:08)

That's a mind blowing one I met. That's you know, this is how a lot of people trade.

Bryce (44:15)

Now that's some alpha right there for anybody who's listening.

Brendan Veeman (44:18)

Certainly is. And you know, to shift gears a little bit. I was reading or I was watching something and I saw that Eliza OS partnered with Stanford to do a study on the future of autonomous agents in digital finance. What can you tell us about that? Are there any early insights or takeaways that you have from that collab?

Shaw Walters (44:41)

The program hasn't actually started yet, but we're kind of leaning into this two things. One thing is fhe and how we can use homeomorphic encryption and basically how can we do really privacy oriented agents. There's a lot of concerns obviously around the twofold thing of agents leaking your keys and the other side of this not knowing what the agents are actually doing for you and how that's verifiable. So that's one side and the other side is the focus on this like economic trust model stuff. So we kind of like wrote this initial paper and we want to get some credibility to it, so we're kind of bringing it there of like, how do we, we've got some initial, you know, I want to make it like, like more legit basically. How do we, how do we take this marketplace of trust concept and do a bigger sort of model of it? Because right now we're modeling it like on our chat of maybe 2,000 people. And it would be nice to model it on like a social graph like Twitter. Like could we actually look at all of the people's shells across, you know, but, but it's gonna require some resources. So that's kind of the next plan. Does that make sense?

Bryce (45:52)

No. Yeah, it definitely does. And yeah, it's just crazy how many different directions AI is gonna go. It's, I just think of like it's just gonna spread in every direction, all over. It's gonna affect every industry, every, every action that we do pretty much. And like, I think, you know, the broad takeaway I think from today is like for, for folks who are listening, who are watching and who haven't at least tried it or at least gotten in. And like I've even, I don't code, but I've tried lovable before and I've done vibe coding and like that kind of stuff and it's mind blowing like the power that you now have even as a completely non technical person. And I think it's only going to get more and more and more easy and more refined and you're going to be able to, you know, like I think Brendan said like the silver bullet or the golden ticket or whatever to like, like build these things. It'll be so easy. And so don't shortchange yourself by saying, oh, I'm not technical enough to do that. Just dive in and, and study stuff like Eliza OS and get involved in these communities. These are where people are talking about these big ideas and helping people. I mean like, you know, we've got a founder on and you're in your discord, you're in telegram, like you're talking to these people. You're, you're engaging, giving your ideas out there. And so I, I definitely want everybody who's probably just, you know, oh, I'm just a crypto trader. I'm just on Coinbase. Maybe I'm just on Phantom. Like, don't be afraid to branch out to some of these other sort of things. Like I said, we had some quantum people on. We have a lot of AI people on. There's just like an explosion of new interests and new ideas. And it is interesting that it's converging a little bit on crypto because there's that token that kind of incentivizes and unifies all these disparate groups. And, and so I guess the last question I had for you was just kind of on that token and like, is it purely a meme coin? Is it something that you do see symbolically unifying a lot of these people? Like, is there a future plan for it? Are you not allowed to talk about it? Like, where's that all kind of come in? And we'll kind of end on that.

Shaw Walters (47:46)

We definitely have a future plan that I can't reveal without legal advice, but we're continuing to build into this. We actually are announcing the launch of Eliza os well today as a filming of 1.0, but it would have been probably a couple of weeks when this gets released. Release. But that's like our kind of big, like, okay, we've launched 1.0 before this. I would say what we had was a research project, you know, and now, now it feels like a real product that you can use and we'll have a platform out really, really soon. Probably really, really soon if, if this is launched. But, but to your point, like, I think this is the time where if you're not using AI in your daily life, like you are, you're really just, I don't know, there's, there's this whole like, thing going on where people are very anti AI and you know, whatever, and it's like you're just hurting yourself. Like this technology gives so much value for what it costs. Like it's mostly free for most people and it is incredibly, incredibly life changing for me as, as a programmer who's been doing this for my whole life, even down to people I know who are just getting into, into building stuff. Like, I have a friend who he, he started Learning programming on YouTube a year ago and he just finished his first app and he's publishing it to the App Store like in a year on his own, you know, and that I remember, that was just like 100 times harder 10 years ago. Like, like so much harder. So, so there's never been a better time. In fact, this is the best time because it'll probably be kind of over flooded and everyone will be doing it in a few years. And if you get it in right, right, right fucking now. And you just like make this your life and learn on YouTube and like get on discords and learn from other people and stuff and just get into the world, I think you could really change your life. You could really own a piece of the future. And I think that's what's really important about what's coming up is that like we actually have a piece of it. So that when there are robots walking around the world, we're not like, oh fuck, what do I do? We're like, yeah, I'm an investor in that shit. I'm off of this. That's, that's the world. We all want to be rich, you know, and I think that could be possible if, if we own it. So that's why crypto and man, I love that.

Bryce (49:42)

That is such a beautiful vision. And that's going to be a hot clip. That is going to be a hot clip. So we'll make sure that. No, we really appreciate you joining us, Shaw telling us all about what you're building at Eliza os. And we certainly, you guys, everybody who's listening, certainly have my word that we're going to be in inviting you back. If you'll, if you'll join us, we'd love to hear more updates down the line and just stay in touch. Our community, I'm sure will love this episode. So hope you enjoyed Shaw, hope everybody listening enjoyed and until next time, we'll see you soon.

Shaw Walters (50:25)

If you're a maintenance supervisor at a manufacturing facility and your machinery isn't working right, Grainger knows you need to understand what's wrong as soon as possible. So when a conveyor motor falters, Grainger offers diagnostic tools like calibration kits and multimeters to help you identify and fix the problem. With Grainger, you can be confident you have everything you need to keep your facility running smoothly. Call 1-800-GRAINGER clickgrainger.com or just stop by Grainger for the ones who get it done. You just realized your business needed to hire someone yesterday. How can you find amazing candidates fast? Easy. Just use Indeed. Stop struggling to get your job post seen on other job sites with Indeed sponsored jobs. Your post jumps to the top of the page for your relevant candidates so you can reach the people you want faster. According to Indeed data, sponsored jobs posted directly on indeed have 45% more applications than non sponsored jobs Jobs don't. Wait any longer. Speed up your hiring right now with Indeed and listeners of this show will get a $75 sponsored job credit. To get your jobs more visibility@ Indeed.com podkatz13 just go to Indeed.com podkatz13 right now and support our show by saying you heard about Indeed on this podcast. Terms and conditions apply. Hiring Indeed is all you need.