Hacked - The GTA 6 Leak With a $250,000 Crypto Question

Episode Date: October 2, 2026

It's a good old chatty chat. We dig into the GTA 6 leak, with a strange crypto angle to it. Harnesses like Open-claw become consumer products with Muse and dots. And other stuff? Lots Hacked is presented by NordLayer. NordLayer is a network security platform for modern teams. NordLayer gives companies centralized control over who can access their systems, keeps every connection fast and encrypted, and requires no additional hardware or complex infrastructure. nordlayer.com/hackedpodcast   Learn more about your ad choices. Visit podcastchoices.com/adchoices

Welcome to Hacked. Welcome to Hacked. I want to start this episode by talking about a very big, very public, very profitable video game leak. Normally leaks aren't associated with profit, Jordan. Tell me more. What a time to be alive. On August 18th, a group or person or like agenic swarm is really hard to know nowadays. Something calling themselves cyber leak starts publishing gameplay footage from Grand Thefts

Auto 6 that the developer of that game Rockstar had not released yet. Clips of the game. Scott, the big game. The big game that everybody is waiting for. The world is waiting for. People are signing up for Netflix to see trailers in advance of. People just were hungry for clips of that of that crime Florida life, that five star light. I guess it's six stars in this one. You can go up to six stars. People, all these clips start going live of people fighting in the game, driving in the game, basking in the Florida energy. There's maps in the leak. There's videos of the characters, our cool crime protagonists. This was all before that big Netflix reveal that you mentioned, Scott. So people are like, they're hungry for this game. This leak of all of this video is a

really, really big deal. 17 of these leaks fall over the next 10 days. And alongside the footage that they were leaking, they were kind of framing the leaks as a type of protest. The group CyberLeak published a set of like gaming demands on the website. And it's pretty classic stuff. End digital only pre-orders. End downloadable content held back from the game at launch. Guarantee that the single player games remain playable without a server connection. Like hurrah gamer stuff.

And each leaked clip links back to this website. which cited kind of the big reason for all this is Rockstar's $100 ultimate edition and the decision not to release it as a physical disc as why they were the target of this. Big leak, big public statement. They're selling a box with a download code in it, boo. High level, that's it. For like a week and really nerdy corners of the internet, that was a thing. So this is like a cyber protest.

Exactly. done by what I would call, you know, people of our generation and older that care about physical media, owning games, offline play. Old hacker shakes fist at clouds type stuff. Yeah. But at the same time, and this is where it gets interesting and I wanted to talk to you about it, the group also launched a cryptocurrency. Why not? Why not? It's called CyberLeaks.

Just call it the same thing. Who even cares? It goes up at the same time, at the same day as that first leak. In a post about this token, because like you hear that and you go, what's the story there? They write, quote, cyber leak is raising funds for a secret project. The specifics cannot be revealed yet. Let it be absolutely clear. This is not a cash grab.

No. Scott, no laughing. It's not a cash grab. Sorry. Yeah. Sorry. I was too busy thinking about whether the. special project was some form of prediction market or online gambling system because I feel like the

whole I bet on crypto make money on crypto is now gaming prediction markets like I feel like that audience is moving and you know just the sheer fact that they went crypto is kind of an old school thing to do again furthering my theory that these are like 40 year old people sure you're like they could have been on cal she making fat stacks yeah yeah they could have had a way I got you. I will GTA 6 see a leak? Sure. By the entire yes side of the argument,

pump up the no side, and then retire. A little while later, a couple days. Rock stars, you brought this up already. Big GTA6 reveal is going up on Netflix. A 27 minute long,

like, to call it a trailer. It's like it's just like a little movie. Did you watch it? I did not watch it. You did not watch it? I was somebody that's excited for the game to come out and was a big fan of GTA5.

I could not have been. bothered to watch a 30 minute pre-launch trailer. Oh, I got less shit going on in my life. That thing ruled. Do it actually? Yeah, it's really good. So a few hours before Rockstar puts its 30-minute trailer for the game up on Netflix, um, at what you might call the apex of the value of cyber leaks leaks,

up until this point, they're the only source of these, of this video footage. 30 minutes of it is about to go live, the value of those. leaks is about to go down. That wallet on the Salana blockchain where they coined CyberLeak makes four transactions. Rugpole. It's a rate of the speed at which you said that.

Rugpole. Say it. Was it a rub pole? It's a rugpole. It's like a car driving pass. This is according to on-chain data reported by CoinDesk. They converted 146 of these coins.

They did it. Blah, blah, blah, blah. They realized by kind of realizing these fees about a quarter million dollars in profit. Not bad. Not too shabby. And so here we find ourselves this rag tag thing guy group. Agentic swarm called CyberLeak,

exposing themselves to great potential legal retribution, which we'll talk about at the same time as having this very earnest message from gamers to a gamer giant. At the same time, really muddying the waters of their, messaging with this alleged quarter million dollar payday. And I just wanted to start the episode here. In this episode brought to you by our title sponsor Nordlare, the network security platform

for modern teams. Tell you about them more later. I just wanted to start here, Scott, because I think this is such an interesting story. It is, I'll say it feels very 2024. Sure. Yeah. But it is interesting. I took a peek into their subreddit because I saw that they had a little subreddit.

read a bit of their manifesto-ishy stuff. But that's about where it stopped for me. So I was intrigued to talk to you about this, hear what your thoughts are, hear what your takes are. I think everybody in the world knows that there is a demand for leaks for popular games. And if you were making the most popular game in the world ever, probably,

I can see the demand being really high. It seems like an obvious target. And they've been the target of multiple hacks before, if we recall. For sure. For sure. I started with kind of like, do we know anything about the hack itself? Like, do we, do we, have we learned anything about the intrusion method? Because there's something to be said for them saying, well, we targeted this game in particular because of this choice.

But then there's just a question of like, well, did you target it because of that reason? Or did you just have access and you reverse engineered the reason? That was a question that was. pretty prominent in my mind in reading through all this. As of right now, we don't have a confirmed, like, attack vector for all of this. There's not how it happened. Yeah, there's not like a smoke and gun fishing email where we figured it out or this credential leak where we know for sure.

Best information is that it's about a year old content. Lucia, one of the protagonists, wasn't in it. There was a year old song. People were trying to get into the forensics of it. But the thing I found interesting on that front was it seems as though it wasn't Was it just a tranche of pre-recorded video and image files? It seems as though CyberLeak had access to a build of GTA6. And that's pretty important because it could have just been a bunch of reference files, screen recordings.

The strongest evidence of this and maybe the coolest part of it for me was August 20th, CyberLeak, like an old-timey, like gangster, shot their own name into a wall in the game. they shot the word leak So they were able to play They were playing it So like a game that went Like they call me cyber leak And they wrote leak Into a wall in the game

So that was the best evidence That this wasn't just stolen videos This was like an actual build Of GTA6 out in the wild Can I make a hypothesis Please A far-fetched classic Scott hypothesis Sure

Major video game development Which I've worked in you've worked in indie development. Yep. Have entire teams of people called QA people that just play the game. They get dev builds of the game pretty much every day, if not multiple times a day. Scott suspects the bug testers. They, well, here's the thing is they're typically not the highest paid people on the team.

And they're typically subcontractors. So they really have no connection to the core team that's been developing the IP. They're not really bought in. They're just mercenaries for hire to come in. and play video games all day, every day for, you know, the lowest compensation tier in the game team. Sure. Often. And they're super valuable.

This is not, I'm not in the middle of the game design process. Extremely important to the build process. You are not begrudging Keyway and bug testing. It is like actually a deeply important part of game design. But I know what you're saying. It's the subcontractor part is really what's most important. Totally. You're a subcontractor and you're not the highest paid subcontractor, which is a sweet spot.

Like having worked on some of those teams, some of those people, they're not even, they're from countries that you haven't heard of. Like, these are, it's not like they're local people to get brought into a sub and they end up on a QA team. They're people from Georgia and people from, you know, Eastern Europe, like Asia, like they're from everywhere. You can only imagine that even if the subcontractors getting paid a North American rate for their time, the person who's actually doing it is making very little. So my hypothesis goes that this chances are this was somebody who said, oh, I've actually got a live playable version of GTA6 on my computer at home. And somebody said, if you leak that, you could make like tons of money. $250,000 US dollars is depending on where you are in the world. Like that's a large amount of money for by any standard.

Totally. Totally. It is generational life changing money. So it's. Yeah, of course. You get sent to build and you're like, so many eyeballs will be drawn to this organically. And we'll talk about how the cryptocurrency and the leak worked in a second here. But like it's a very short line to an extraordinary sum of cash.

The incentive is really like natural and like I understand it. And a quarter million dollars in, you know, the United States or Canada is is a specific line in our life. A quarter million US in Bangladesh is. Yeah. You know, royalty money. A hundred percent. So the economic incentive is there. Just that's my, that's my hypothesis.

That's where I would start looking if I was the FBI going through the key. Well, if you were take two's lawyers, because that's where this is actually all starting is. Okay. Basically using like DMCA grounds, like copyright grounds. Yeah. Their lawyers have started filing subpoenas, New York court DMCA subpoenas. We haven't gotten to a suit yet. This is just fact finding for them.

They're just going like, hey, Microsoft, a bunch of this stuff was hosted on OneDrive files. We would, here's a subpoena. We would like as much. information about who that account belonged to. Hey, Discord. Um, we would like account information pertaining to the cyber leak account and a few other ones, which we'll talk about because there were streamers that then covered this stuff who had nothing to do with it, but who have now gotten embroiled with this set of subpoenas. It's all very murky. Hey, X, who does the cyber leak account belong to? Uh, and I think a YouTube subpoena as well. As I mentioned, a streamer,

dark viper, a, you got swept up in that. Um, they've denied volume, the involvement. They were like, I was just streaming and showed this footage that they were streaming. I had nothing to do with this. Now they want, I think his audience, like it's the legal creep is occurring. And there's a little bit of a like slushing in slushing out of how many people get caught in the drag net of figuring out who leaked this good, good crime video game. Having worked on games and having seen a leak from the inside. The first time that I saw a leak on a highly anticipated pre-release title.

I immediately thought that it was pre-planned by the team. Like it was a marketing concept. Like guerrilla marketing. Yeah. Yeah. Because the audience is like, you know, it's there and it's excited for it. But they're not really engaging with it yet because it's not out yet. But every time a slow drip of data comes out, a small leak, bit of this, it creates

this fervor of communication and the Reddit's explode and everybody's talking and getting more excited and asking about pre-release days. And so the first time I saw it happen, there was, I think, probably a similarish witch hunt that went on. I had entirely thought that it was just a pre-planned marketing promo stunt because to me, it was very effective at that. So yeah. So the same thing that goes here. Every time I see a game leak, I'm like, is this a leak? Or is this just the PR team earning their annual bonuses?

Yeah, right. I think that skepticism is often warranted. I think in this case, I don't smell that. Like, I think, and it's, you know what it is? It's honestly, it's the gulf in production value between the Netflix trailer and the like poorly streamed low quality. It's like this didn't have the veneer of a PR's broadcast. It was like there would have been a lot more pixels. You know what I mean? Like there just would have been a higher quality, a lot more curated. It was like kind of a bad first impression. compared to what went live, then I'm like, I actually believe that this was just a big, really, a really big whoopsie. And then the lawsuits would be like, wow,

you're really playing with fire to have your lawyers sue your, sue you. Sue yourself. Sue yourself, a great personal expense. Yeah, I think in a lot of cases, it probably is in this case.

I'm like, no, I'm pretty sure. I think there's, I think your earlier theory, smells more accurate to me. Well, the last time there was a big leak on this game, it was Lapsis, wasn't it?

The Hacking Group Lapsis? Yeah, shiny hunters, yeah. Shiny hunters adjacent, but Lapsis. There were two. There was the Lapsis hack in 2020, two with Cortage, and then there was another thing in, I think this past year with shiny hunters

who are technically related. I'm cutting you off, but yes, there's two incidents there. Your, uh, your memory is impeccable. Yeah, but I think Lapsis actually stole in-game content. Like they had the ability to show pre-release, you know,

what was being worked on kind of systems, where I think shiny hunters took like internal information. Right. That's kind of their MO anyway. Okay. But the, anyway, yeah. So it had been a long time since somebody had seen the state of the game. Yeah, right.

So I think that that's just it for me is there's the constant desire for people to see what they can't what they're not supposed to. Totally. I think that's a very powerful marketing lever that if I was VP marketing of a big AAA studio,

I'd probably consider pulling it. Just with thumb drives with all this leaked content, just like placing it on the concrete and the parking lot out front, looking a journalist in the eye, being like, where did my thumb drive go? I wonder if, like,

if you could get the viewing metrics from Netflix, for their 27-minute, $25 million pre-release launch trailer. Yeah. I wonder what the viewership and audience of that is versus what a major leaks audience is because I bet it's comparable and it cost you nothing. What was interesting about the Netflix thing was that it wasn't even,

it's like you can't guarantee perpetual exclusivity. It was like a Netflix exclusive window for like a little chunk of something. Exactly. 24 hours. It was like a little chunk of time. Yeah. Fascinating that Netflix.

paid money for that. I bet a lot of money changed hands. Considerably more than on this crypto side of this whole thing. Yes, yes. Please take us down. Yeah. It's interesting. The only thing I wanted to talk about the crypto, we could dig into the like specific ratio of like, well, they they burned 20% of the supply and this is how we're like we could get really into the crypto weeds. I don't think it's that interesting. The thing I found fascinating was that they launched this thing right before the leak goes viral. every time they shared that content, every piece of content contained a link back to the coin. So all of the news coverage, all of the Reddit posts, all the Twitter posts promoting this stuff were inadvertently driving traffic back to this.

The sheer amount of traffic influx caused the price to spike onto sort of like hype alone. Sure. Algorithmic traders saw massive trend influx of this coin being mentioned and scooped them up. And then operators cashed out on the trading fees that they had accumulated about 250,000 U.S. real dollars, at which point they traded that out and then the price collapsed. So it's like you could, there is maybe a, it's not a traditional rugpole where it's like, I'm holding a lot of the coins you drive the price up and I sell them. But it's sort of like cousins.

I'm not going to call it a rug pull. I'm just going to notice that it shares the shape of a rug pull. Sure. So they knew the marketing lever that they had. and they said, how can we turn this into money? Creating a cryptocurrency was that way to turn it into money. Even if it was more on the legitimate side transaction charges, not the, hey, everybody buy in, I'm going to rug pull you, Hock to a girl style.

You kind of just look at it and you're like, we can have a debate about the mechanism of what occurred here, but you did walk away with a significant sum of money, which they primited by saying it is for a project. But it, it, and this is sort of where this all ends for me, is it just, muddies the water of your messaging a little bit because there there is a message in there that I'm like we can talk about realism and where the market and where the gaming industry is going. But I'm like, I'm sympathetic to your basic intent here. But you have muddied the waters of that message with this rug pull shaped operation you've pulled off here. Well, they got to fund the new product. They need the $250,000 to fund the product. You got to fund your project, which is an even

vaguer word that's like, I have a project to have a new house. Rockstar's August 26th statement in response to this. They basically said, like, sorry to the fans that this is how you're seeing this for the first time, which is like kind of a like, interesting way of framing it. They didn't address any of the criticism, but they did say a thing that felt real to me as a person who makes games. I imagine it would be to you, quote, it would be an understatement to say that having videos of Grand Theft Auto 6 gameplay leak in this way has been heartbreaking for our team.

And this is obviously not how we intended for you to see the game. after all this time. Like, yep, that's how I would feel. I get it. Yeah, especially because it's old footage. I think that's it. Like a recent build copy. This could also just been like a crypto locker thing where it's like maybe a QIA

computer got crypto locker. Literally. They stole the data, exfiltrated it. And then found out that they had a copy of GTA6 and we're like, whoa. Can you imagine? Can you imagine you're kind of just like fishing? And you hack this QA company and you're like, maybe they have a, cool, oh my God, it's GTA6.

Like, you can imagine the feeling of it rushing to greet you when you, like, open the EXC and you're like, oh, no. Oh, no. I'm in, I'm in more trouble than I was a minute ago. Yeah. I better make a lot of money because of how trouble, how much trouble I need. Yeah, like, I better find a way to rug pull off. This kind of engenders empathy a little bit, sort of.

That's kind of almost everything I have for this. The thing I wanted to end on was from a group called stock. stop killing games because at the heart of their manifesto is a sort of games preservation argument. And I think there's a really good argument there that unless we're talking about a live service game, if I give you 50 or 60 or 80 or $100 for it just through corporate attrition, my product shouldn't go away. If it's a single player game, the idea that it has to like ping a server to check if I'm allowed to play it is like, well, then I didn't really buy it. So can I really pirate it? You get into this whole murky thing here.

I like the idea that if I spend 50 bucks on a video game, I own it. And we should cling to that as long as we possibly can. Recent physical media games aren't even games. They're literally like downloads certificates. Yeah, no, it's all codes. Even the cartridge for Switch 2 is just a link to a download code. I'm with you. But we rage against the dying of the light scot.

Stop killing games. who is like, I would say like an important part of the digital preservation movement. They're not about physical. They just, they just want these things to be standalone pieces of software that can run so that they can archive them to preserve them for the future. And Cyberlux Manifesto kind of echoes their language. They distanced themselves from this thing with a hundred foot pole. Stop Killen Games was founded a few years ago.

It was by a YouTuber after Ubisoft shut down this game that they love. Their whole idea is games should be left in a quote, reasonably functional state after discontinuation via like an offline mode, private service, or whatever, the game just shouldn't die. Their line about this was this kind of act is destructive and helps no one worse. It risks tarnishing the reputation of our communities and our chances to make real change. And I found that really fascinating, that the real games preservation folks are like, thanks but no thanks.

This looks way worse than it serves us. And unless there was like an under the table quarter million dollar donation, you didn't really do it for this project, did you? remains to be seen what they do with that money. Maybe they pop up a VIN's whatever that website, like a ROM emulationy thing. Maybe they prop up something real. You think about archivism and archives.

You know, pop culture stuff, advertising campaigns from the 30s, 50s, things that helped shape the social norms all go into archives. You know, they're collected, they're indexed. Video games since 19, I don't know, like,

I would say since the NES when they got huge, like even the, even the Atari, like in-home video games, but even before then you had arcades. They're massive social things. Like the amount of people in the world that play video games at this point and have played for a good majority of their life is, you know, substantial. And to not archive them and to not consider them part of like the cultural data of a generation and of an era is crazy to me.

So I'm a big fan of this idea. I love the idea of the video games archive, stop killing games, lots of these people. big fans, big supporters. I think what they're doing is really good. And the truth is, nobody likes the digital only games. Like, I don't know anybody that does. Sure, it's convenient, but it's like nobody likes it. Yeah, it's an option thing.

And I get both sides of it, to your point, games preservation and archivism is like, it's important. This is like a medium and a genre that we have directed so much energy and love as a culture towards manufacturing. It's like, we should preserve it. It is a huge, like, output of human effort and labor and energy and creativity. It's like, yeah, don't burn the library of Alexandria for this. Like, it's, it matters that those things persist.

It really does. Totally. On the flip of it, I'm like, I get it if you're selling games to be like, most of them are selling. Why are we paying for all this production? I get it. I, there is a legitimate thing there, but how we handle a digital only future, it's not binary. It's not like you either just let go of the steering wheel and hope the industry

functions well on its own. And it's like, no, there's better practices and there's worse practices in a digital only world. We should be pushing towards better ones. Stop Killing Games seems to thread that needle really beautifully, that kind of way of thinking about it. So I should expect my physical media version of Rebounder by 33 games sooner than later? I'm not saying. I'm just saying. Not saying. I'm just saying. Before we totally jump off this is a tiny little touchback to GTA that I wanted to make, which is shiny hunters. You mentioned it. Yes. Yep.

Recently, there's a big arrest in the Netherlands. They recently arrested a 24-year-old Amsterdam man, had a role in the shiny hunters, allegedly. His name, I could try and mess up and we'll choose not to, but it ends with Vanderstap, which I get. And given that I speak a bit of Dutch, I should know how to say the first name of this person. Pepin, Pepin, Pepin van der Spap. Okay. Prior to prior conviction, got arrested. I think he was arrested on the 15th, but they didn't break the name.

news until the 28th, 29th. Kind of a big deal. You know, Thorne in the side, obviously, Shiny Hunter's famous group that steals data. They go into massive corporations and exfiltrate large amounts of private personal trade data. And then they hold it hostage. And if you don't pay, they give it away. And they have proven to do such thing. So just wanted to throw that in as I do have a link back to GTA.

They had hacked GTA for it. We alluded to this earlier in the show, but they had a, it wasn't content about the games, but it was like internal financing docs, internal ops. Like there was a big sort of like Excel spreadsheet. I think it was a lot about how the GTA economy worked with like GTA money. The numbers. How much people were spending on it and using it and stuff like that. Interesting.

Probably a fascinating leak to read, honestly. Before we get to the next story, this episode, it's brought to you by our title sponsor, Nordlayer. The reality of running a modern team, your people are working from different devices different locations, different networks, and most businesses have no real visibility into what that looks like from a security standpoint. Nord layer is a network security platform that fixes that. Nord layer gives you centralized control over who can access your company systems, lets you grant to revoke access in seconds, and keeps every connection fast and encrypted and does all of that without any additional hardware or complex infrastructure. You can verify users by identity and

device block malicious sites and risky domains and stay compliant without slowing anyone down. It's built for the way teams actually work now. Check it out at Nordlayer.com slash hacked podcast. That's Nordlayer.com slash hacked podcast. And thank you, as always, to Nordlayer for their support. And we're back. Scott, do you want to talk about OpenClaas and its many ancestors? It's family.

It's the family tree that has emerged over Open Claws. I would love to talk about Rockbot, Muse, Open AI, Dots, Instinct. I'm trying to remember. There's a few more. They said it was going to be the year of the personal assistant. And boy, were they right. It harnesses abound, you might say. It harnesses abound.

So we're talking about two of these consumer products were released, announced and released the week that we are presenting this, Muse by meta and dots by open AI. to go back a little bit, to lay the groundwork of the story here. We started talking about OpenClaw at the start of this because these are all in some way related to that. So I think it's worth going back a little bit. It was this kind of like hackery project started as a weekend hack called WhatsApp Relay, 100,000 GitHub stars in two months. It started running. It was this like harness, this little AI agent that runs on your computer that you talk to you through messaging apps.

You text it like you would a person. And instead of just replying, you can actually use the files on the computer. That was the big change. It was a harness for one of these models. It was originally Anthropics Claude, which is why it was called. It was Claudebot and then it's Claude mod, mold bot, open claw. LLM is the brain. This is this harness that gives it access to your files.

So you can give it kind of instructions like you would a person using a computer. Go check those emails and do this with it. Go grab this file, edit it and send it over here. It led to the fact that you couldn't buy a Mac Mini for like a year. Yes. Because that was the standard operating platform for these things. Everybody would go up and buy a Mac Mini. They would set up OpenClawn on it, which eventually I think became Hermes bought.

There was a family that descended out of OpenClaw of like better versions of it that were less jumbly. Sure. The de jumbleification is kind of what this story is about in terms of the consumer product side. And like you said, people bought these things because they realized pretty quickly that if they're willing to pay for it, you could just sort of have this thing running on a Mac Mini 24-7 getting up to some real trouble. But that's the high level of what a some real hijinks, Scott. That's sort of what a harness is super high level. I think is a good way of thinking of super high. An agentic harness, yeah.

Yeah. Sure. Which is. Yeah. Yeah. If there's another type that most people need to know about, please do let me know. Well, harnesses are comes in all shapes and sizes. Harness is really just a name for a framework you put around an LLM that controls its execution flow.

So that could be plan development, execution, verification, quality assessment, and it could be domain specific. Like you could have an AI harness just for coding, which is like what cloud code is. Sure. Or you could have an AI harness. That's a good distinction. Yeah. just for educational delivery, which is kind of what Alpha School does. You can have, so there's various different things.

A harness is more of the general term for, we put an LLM brain in a box and we want to make sure that it does what we want it to do. So we build a bunch of systems that control the LLM. It's kind of like the nervous system for an LLM. And in the case of these products, we'll be getting to in a minute. Just the idea of like a consumer personal assistant has been. been since this technology came out, this guy Peter Steinberger, I think is his name, comes up with it.

Sure. Thank you. He comes up with it. And everyone goes, oh, this could be the thing that becomes Jarvis. You know what I mean? Like the personal computer assistant that just does what you want because it has hooks into all of your digital life. Uh, sky Peter Steinberger comes up with it. It's burning money for him because it's like super popular and it's like $10,000 a month.

So he, I can't remember if he got bought by or just hired by, but basically he and up at OpenAI. He kind of takes it over that direction, please. He got not Aqua hired, but hired. Right. Okay. The OpenClaw, I believe, is still held by the OpenClawe foundation. Peter's still the head of that foundation.

But then he went to within weeks, he had a, what I can only assume was a very lucrative offer to go join the Open AI team pre-IPO. Sure. to lead, I think, their hygienic solutions team. Interesting. Some other things have made this super famous, like just to go into the research archives that I keep in my brain, is Jensen Huang, NVIDIA founder. One of the, if it's got to be one of the top three richest people in the world, if not second richest under. I think so.

Yeah. Those leather jackets don't come cheap. Famously, and if you didn't know this, and I probably mentioned this on the show, but his cousin is, the CEO of AMD. So their family reunits have to be a blast where you have two of the richest semiconductor, you know, moguls in the world.

There's a sky ship somewhere. Totally. Yeah. The, the, the, the, the,

the, the, the, the, to loop back, Jensen was on all in. And he was talking about open claw right after it launched. It's like one of the largest progress. progressions of

AI kind of that had happened to date. The ability, like, the one of the things that made OpenClosso like, so functional was that it had access to a full computer and it had the ability to understand how to use

that computer. So that's what really gave it the ability to start doing and automating things on behalf of the, I don't know what, your master, I don't know what the proper term for user, yeah. I think user's still good, still software, you know? Yeah. Or masters, you could go that direction, Lord. Like, I don't know, but I think, I think user is good. Old timey. Yeah, users. So the,

the user that was controlling it had had, uh, the ability to do far more than any of the existing AI chat solutions. Because that's kind of where things were still at in that realm. There were some development harnesses like Cloud Code was out. Codex might have been out. Uh, but there wasn't really anything for the person to help facilitate daily life. And OpenClaw was one of those things. And it spawned off an entire ecosystem of multi-agentic, you know, coordination and collaboration. And there's all kinds of worlds that came out of this. For sure. And part of the structure of this thing, OpenClaw, is important for where we're going with the consumerization of this technology. Every OpenCly agent had four files. I didn't know about this, but now the whole world knows about

this because this is now becoming a recurring structure. You had agent, which was like the operating rules for this thing, like just sort of super high level, how it's supposed to work. You got user, which was like the users or masters in your case, preferences and habits. Yeah, yeah, data on the user. Yeah. Memory.md, which is like its long term memory. Soul.md is its hyperbolicly named like personality tone.

It's like a real comprehensive system prompt. There you go. So you have agents, user memories. soul that is the sort of like a big part of the structure of this thing the guy who makes it gets swooped up by open AI meanwhile now meta jumps in um they didn't win the higher they were apparently in fierce competition over peter that's just allegedly but they were trying to get them they lose the higher but they jump in on this and september 8 they launched this product muse so a user asked muse to compare itself against open claw like an open claw repo and they were like immediately like

well, you got a soul.md and memory.md, a user.md and an identity.md in the directory. And meta was basically like, oh, yeah, we were super heavily inspired by OpenClaw. The quote that I loved and the reason I wanted to talk about it was, well, they got those things exactly right. Let us all not forget that the head of AI safety at Facebook deleted her email with OpenClaw in the week after it launched. Getting things right, Scott. Go for it. I bold through part of that. Yeah. No, the, I want to jump back in time a bit if you want to go.

Because actually Grockbot was the first of these kind of new, modern, multi-agent. I haven't used Mews that much. I installed it and tried it. I haven't. I just don't trust Facebook that much. So. Oh, yeah. We need to talk about that.

But I want you to compare it to Grockb. Yeah. Let's definitely come back to the meta of it all, because I think that's, a huge part of this story. Fall is falling. Winter is coming. And that only means one thing. Holiday shopping.

If you have an interest in starting an online retailer store, sell something. Holiday season is when you want to be in market for it. Shopify helps you launch a store and starts selling in a few simple steps without needing to be a big high-falutin e-commerce expert. And sidekick that Shopify's AI assistant. and can help with business tasks along the way from store setup to planning promotions. They got pre-built themes, an AI site builder that makes it easier to create a professional

storefront. They got a mobile app that lets you manage key parts of your store from anywhere, including what have we got? Inventory, customer support, and payments. Shopify checkout with ShopPay converts up to 36% higher than the competition. Launch before Black Friday, a time of year when more people are already shopping and ready to buy. And launching before Black Friday means you can.

take advantage of all of that juicy extra online traffic. If you're ready to launch your business before Black Friday, go to Shopify.com slash hack to start your free trial. That's Shopify.com slash hacked. Go to Shopify.com slash hacked today. To hear your first. Your phone and laptop don't stop connecting just because you stop using them. They can still share your location and sync data in the background.

Those connections can help big tech and. and governments track you and give hackers a way in. That's why we started using Silent. They make Faraday bags that physically block wireless signals to and from your device you put inside, you close the bag and the connection. What does it do? It stops. No app, no settings, no subscription.

And it's just a bag that I carry. A simple way to take some control back over your privacy, check them out at silent.com slash hacked. That's s L-N-T-no-vowels.com slash hack. Use hack to check out to get yourself 15% off and free shipping on qualifying orders. So the, so Grockbot kind of spun up their own version of this and it was out in market, I think, months ago. And it is, I would say, exceptional. Having used it quite a bit, you kind of have the ability to spawn up multiple agents.

This is a funny thing. And it's something that I talked to. There's a guy Theo. He makes T3 code. It's like another coding harness. And I fired him a Twitter message one day. And I was like, hey, can you add the ability to have your agents message each other? And he was just like, no.

And I was like, okay. And then Grockbaugh came out. And one of Grockbot's most powerful things is you can have group chats with agents. You can tell them to talk to each other. So you can like delegate something to like a production manager who then controls their entire. team of resources to like, you know, whether it's right code, they've got a QA person that's reviewing things. So you can kind of build an entire organization in Grockbot, which then they can kind of self-manage a bit, which was kind of where the second generation of like,

when people got a hold of OpenClaught, they started saying, how do I get multiple of these things talking to each other and working together? So Grockbot kind of took that concept and built it into a commercial platform. Every single one of these agents, including Muse and including OpenAI's dots, which I'm sure we'll get to. One of the things that makes them so powerful is they essentially have a container virtualized computer. So every time you spin up an agent,

it gives them their own computer to use where they can download stuff, they can open web browsers. They can, you know, they are fully agentic with full computer backing. They have the ability to download stuff and install things. So yeah, anyway,

I don't know where I was going with that, but that's a bit of the history. No, it's, it's, I think it is relevant. to this because it there's there's a little bit of a spread. Grockbot runs primarily on your computer. I think it has like it can talk to a one of X's servers, but it's primarily it's more like codec like it's controlling files on your system, right? The inside of a virtual container.

Okay. And that virtual container actually lives in the cloud. Oh, okay. Interesting. So it's more akin to Muser dots in that regard. It can control files on your computer. So like say you have it installed on your computer. if your computer's running, like say you have a Mac Mini from your OpenClaw days and you have GrockBots installed on it, it can actually go into the local file system and do stuff for you.

But then it also has its own virtualized VC that it works in. That was what I was curious about because I was, Claudebot seemed like it was like, you need to buy a Mac Mini because you're, you're the server now. Like it needs to run everything locally. And Grockbot, I had a sense it was a bit more cloud based. And then now we've ended up with Muse and Dots where it's entirely cloud-based. It's like there's just a little virtual machine running on their servers.

You don't need the, I mean, you need hardware to interact with it, but you don't need the Mac Mini functioning as the server farm. Like that is sort of where this productization has taken us to is like you don't want to have to have a home hub for this stuff. So we'll just run that in the cloud encrypted. Well, OpenAI made news the other day for buying something like 40,000 of the new Mac minis. I'm wondering if open AI dots just isn't a Mac mini farm. It seems as though it was just like he kind of one-shotted OpenClaw. Like he didn't. And I shouldn't say one-shot because it implies that he vibe-coded it, which maybe

did. He agentically engineered it. It took him like a little like couple months maybe. He nailed the structure is kind of where I'm going with that. It was like the basic architecture of it is like it will evolve, but it was like he kind of figured out the rough skeletal structure of one of these things. sure, eons of advancement and stuff, but like the basic, the basic bones of it. And though he ended up at OpenAI, Meta then announced this new consumer technology

product called Muse. It is a application. It is a little Tamagotchi style. I think they called the charm. Same as Grockbot. Does Grockbot have physical hardware? No. Grockbot is essentially the exact same thing as Dots and Mews.

But they. the, oh no, Mews has the Tamagotchi. Dots and BrockBots have an app. Little agentic icons look almost exactly the same. Oh, really? Just the little shapes, the little, they remind me of breakfast cereal. Animated eyes and stuff. I called them breakfast cereal.

I'm like, it's the little stars. It's the marshmallows in the Lucky Charms. So they roll out this application. It does very, very well. Meta has no problem building out a social graph. Like between Instagram and threads, they can just direct so much. much traffic that this thing naturally Facebook climbed up the charts did very, very well.

And it's number one for two or three weeks at this point, I think. Basically, they're trying to say like Grockbot is really, really good. It's like you can wire it into your Slack and your discord and your email. And it's, it's very like a work productivity aimed thing, I would say is at least how they're marketing it. I'm sure it has personal application. But it's really marketed as like a work thing.

You can tell they're really trying to market muse. between the app, the vibe of the videos, the little muse charm Tamagotchi keychain thing. They're really marketing this as like a consumer personal assistant. The use case that everyone keeps talking to is like, check my email and find all those, I don't know, Home Depot bills or this type of file and do this with them. Like something that's sort of a brute force task. Sure. Unsubscribe me from all the mailing lists that fill my inbox up every morning.

100%. And then the other feature is just this idea that it's a little bit running in the background. It is spending a little bit of time, say, when you're sleeping, like checking stuff that happened through the day to see if it missed anything so that you can wake up to some kind of message from it, kind of like a personal assistant getting headed this type of thing. The thing that is important, however, for all of this, is that it is being made by the company, Meta, formerly known as Facebook, still run by Mark Zuckerberg, a company with, and I'll put this generously a reputational trust issue. And this technology requires, say you want this thing to organize your email,

you have to let it log into your email. And read all your emails and associate everything about you to you in the Meta Marketing Network. Yes. Look at every receipt of everything you've ever bought. Yep. reply to your messages, just give me your messages. There's a reason I don't use Grockbot.

But it like, it, and it is the same thing as like, uh, there's a, there's an expression. It's about physical hardware. So not about software. It's software is different, but it's, uh, friends don't let friends use, uh, gen one products. Totally. Open claw. And I'm like, pardon me. Fuck me.

This is a gen one product if I ever wanted to not use one. Like, how muse? All of them. Like I'm, I'm, I'm waiting. I'm not saying there's no application for this. Like, there is a point. where I'm like, you know what? It's kind of like if I search for a photo on my computer and the file name is like

7-1-2-3-4-895, but it's a photograph of a tree. If I search tree on my computer, I would like it to surface that image. That is a relevant application of this type of technology that's useful to my life that can be done securely in a way that makes sense for me and that if there were two products, one of which had it, I picked the one that had it. This isn't there yet. It will become there. it will get to a point where I'm like I wouldn't buy some computer that couldn't organize my bills and my email for me

but I'm not going Gen 1 on it just personally I had this conversation the other day with my brother-in-law about AI and like this I was just bringing him up to date on the current state of it sure he's still living in a co-pilot you know chat dpT like chat window it's pre-hardness Yeah, and he was asking me about it and we were talking through all the different platforms and developers and what people are doing and the controversies around them and I was like really the only one that I really kind of trust is Google.

They're generally considered corporate enterprise. They have data security agreements in our enterprise accounts. The people that I trust to uphold the law. On the meanwhile, you've got Dario from Anthropic running around saying Anthropics is going to be the only company in the world in like 10 years

because they're just going to take over every industry after they steal the intellectual property from them. I'm not, that's not a verbatim quote. I'm just like expanding and, and commenting on it. I know, I know the sentiment you're talking to.

He is, let us say, taking a buck wild approach to a pre-IPO PR tour. I'll just call it that. It's like strange tactics, sir. Let's see how it goes. Yeah.

And the same kind of energy from open AI, but a bit different. Mm-hmm. You know, a little bit less altruism. and a little bit more consumerism. And then you've got Elon who's kind of Elon.

He's kind of like I'm a free speech or die kind of person, which kind of so am I. And you know, you love them or hate him. And that's on you to choose. Really, there's nobody that walks in the space that's kind of like clean. And Google,

I would say is the cleanest. I know, I know what you mean. Google gives IBM. Google gives like, our lawyers are here. Like the lawyers said we, can do. I'm like, okay, that that, and they also have built an empire on selling seemingly private data for

it's like they're an advertising empire. Like I'm not going to give them a big badge for privacy. But like, I know what you're saying. Um, open AI is like, well, Sam Altman's something of a specter. I don't know. Like I'm just, I'm just like, I'm almost, we're all kind of just watching to see what kind of ruler of the world he's going to turn out to be. But I think generally speaking, essentially, skepticism is warranted in all of these cases, not just about him. Grock, X, I don't know if I would call, you know, a musk, free speech warrior. He's a musk speech warrior, but I understand his reputation. I'm not giving it my private information, but best of luck.

And then meta is like, hard no for me. Like all of the rest of them, there's this sense of like, let's see how you do it. That one, I'm like, I just fully don't trust you. Like, it's sort of in its own little. box for me over here. I'd love to interview Mark. I really would. It would be very awkward. But I would sure like to do it. I have questions. I don't have a great deal of trust. And it is based on history of performance. Do you know what I'm saying? Like the thing for me that really jumps out is I think the entire AI space is moving at this clip of ask for forgiveness, not permission.

100%. And they're doing it very intentionally, I think. And like you're seeing that with things like the, we never covered it on the show, but we can have a brief chat about it now is the mathematical problem, the Navier Stokes solution that Open AI came up with. And I'm air quoting for the listeners that aren't watching on video. Sure. Theoretically, they got the theory and hypothesis that they submitted a billion dollars of agent time at from training data that came from conversations that like math researchers were having with codex. to try and work through the thoughts of their problem. Hasn't been confirmed or denied. I'm saying this allegedly.

But that is, that's kind of the thing. They're all looking for what is a great PR story that I can release. Hey, if we cracked one of the Millennium Prize mathematics problems, that would be awesome. Hey, guess what? Our new training data set has this like really promising theory that comes from all of these very promising PhDs in math that are working on solving this as their life's work. and we're just going to submit, you know, $100 million of compute time over 30 days and just knock it out.

And then, hey, look at our press launch. Like, look what we just did. Look what AI can do. That's so interesting. And that's kind of a recurring thing. Allegedly. No, allegedly. But that's what's interesting is we also said that on the security side of things.

These companies have these models rigged up into this agentic structure to broadly gloss over. that if you then combine that with functionally unlimited capital of just like how many servers running for how long, those kind of questions that most mortal people would just be like, well, there's just a limit. If I run it for more than a week, it will bankrupt me kind of thing, that they can kind of just be like, point all of it, like the fire hose of the entire tech industry perched atop the entire economy, pointed to the security problem. and this like Dragon Ball Z energy beam blasts at the energy problem.

And they're like, oh, cracked it. Pointed at this math problem and the Dragon Ball Z energy beam just at this math problem. And maybe they sounds like in some cases they've cracked it. I think that's the shape of PR for these companies for the foreseeable future is like in a way that is not reproducible by the general public. Can we just point this thing at something? thing and see what it can do. And then we get to put out a press release that's good news instead of bad news. I'm kind of here for it.

Like I hear what you're saying, but I'm also kind of, I'm also kind of here for it. If we're going to do this and we're going to bring in and like the power consumption and all the rest of it, if we're going to go down this road, we should do it for the betterment of society. And if they're going to choose that going after Millennium Prize math things and, you know, touring situations, like any kind of major development in science. is where they want to point that fire hose, I'm kind of there for it.

I don't like the like IP, alleged IP theft side of it, but I, but I don't hate the fact that, you know, the mathematicians that were working on it probably never would have could have, like probably never would have solved it

because they didn't have the fire hose of intellectual capacity that was provided for. So, so I like, I like the idea that we're using it for good. No, it's better than, then turning that fire hose on like,

consuming Jordan's emails and figuring out what how to best market something to him or serving the ads at a scale and specificity never achieved before in human history I hear what you're saying two things one in a in a first for this show when it comes to the question of IP theft do we have to say allegedly and the second thing is the second thing is like I was I wasn't even criticize I was just describing. Like, that's not even like a criticism that you shouldn't be doing that. I'm like, I guess if you were to be doing anything, because it's not happening at scale too, like I'm so much less worried about the some of the ethical dimensions of AI because it's like,

well, that, that's not happening at scale. You're just, there's just steam shooting out of one room. I have a different kind of set of questions about that. I'm intrigued by it. I think if you. you can cure a disease or solve a math problem is a good PR day for you. Have had it, guys. Like that seems like a good, an instance of a good use of this technology.

I would agree. So that's where I'm living. Like Google's alpha fold like protein models. Yeah. They've got LLMs custom trained on proteins. They had a theory about kind of why we age. And it's this jumbling of amino acids and proteins that kind of gum up together and our body gets worse at clearing out, kind of like the dead waste inside of our cells.

Sure. So they, they kind of took this idea, ran it through Alpha Fold and said, hey, build us something that'll break this stuff down. And so they ran it against it. They came, the Alpha Fold came back and said, hey, we think these compounds or amino acids will help kind of clear this garbage out of your cellular, out of your cells. They fabricate this stuff. and then they start using it on trials.

And it's like wicked good. They take donor skin from like 75 year old people and apply it in a cream form. And the skin rejuvenates back to a 35 year old skin. They start looking at inflammation and, you know, other byproducts of aging that happen in our physical bodies, arthritis, things like this. And it's cleaning it up. And I'm like, this is what I want for my AI future. Yeah. You know, I want my, my 80s to feel like my 30s.

And, you know, the people that I love to be healthy because we've got RMNA vaccines for cancers coming out. I don't know if you saw that. Moderna had a clinical trial successful. This is the, this is where I want the fire hose pointed. No, I care less about, I can send my own emails, thanks. I think, dude, I honestly think that like the speed at which this technology became such a cornerstone of the tech industry, which is.

itself such a corner start of the entire economy is like not to its benefit. If the state of this technology right now was a bunch of brilliant people in a skunk work situation being like, we're here toiling to try and solve this very specific problem. And you just had that in a bunch of places. And there wasn't the gigantic like magnetic pull of propping up the economy with like, can I get you to buy a $39 a month subscription, please? Like if it wasn't all oriented towards. that and it was much more about the skunk works vibe of just solving it solving intractable problems

and we heard about it a hundred of the best researchers from every major challenge we'd be hearing about AI four percent of the time that we're hearing about it because it doesn't need the market that all the PR and spin and bullshit there's it would just be a totally different ecosystem and probably for the better but instead I have to ask the question can amuse make me live forever Scott and that's where we find ourselves instead. No, a muse is going to anger everybody you email by sending horribly written gen.

AI content. Why did they give it a fat booty, Scott? This is the thing I don't understand. Why did they give Muse a booty? I didn't realize that was a thing. Look it up. They gave Muse a booty. It's so strange.

Anyway. I think the only other thing that I really want to talk about is there was kind of a new model launched in the last couple weeks. So instead of like a GPT general purpose transformer, I've been kind of in my head calling these things like GPC general purpose classifiers. So really what they do is they, you send them in a bunch of options and it weights them based on what it thinks is the appropriate output. These models aren't really for human consumption. They're for agentic consumption. Like they speak almost entirely in structured, structured data.

So if you have a software system that's doing something, it needs to evaluate, like say, I send you an email and it needs to evaluate if it's spam. It says, hey, general purpose classifier, tell me if this is spam or not. It reads the email and assigns it essentially awaiting a probability percentage of whether it's spam or not. And then it returns that structured data back to the software system that says, okay, this is 96% spam, throw it in the spam box. So type safe release this thing, early access September 15th. It's almost free to use. Like the API costs are like negligible. It kind of broke the software engineering AI agentic world for a few days.

Because typically if you ask an LLM a question like that, it burns all these resources, thinking tokens. And then it returns you like a text prompt. A natural language prompts. Yeah. Distill into like a binary. to be like, did it say yes or did it say no? Skips that. And this skips all of that.

So in OpenAI Dots fashion, OpenAI released their Decisions API a couple days ago at their new release at their developer days. So they've essentially saw the motivation behind what's called Jev, Jev by TypeSafe, founded by a former OpenAI employee, a reinforcement learner. He went and founded TypeSafe, they make Jev. Now OpenAI has the Decisions API, which is pretty much the same thing. So you can kind of send it binary information or like a, not even binary. You can send it like a list of questions with potential outputs and it will score the outputs

based on what the likelihood that that is. So this is truthfully from like somebody that likes to, to write code and deal with agentic harnesses, this is a huge step for us, because this will allow us to make a lot of quick decisions, especially when you think about a harness. So if you've got like a verification step, so say you and I were to build a harness

to do like some deep research. First thing we do is we ask it the question or give it the topic. It takes that topic and says, what is my research plan? It develops a plan for how it wants to execute that research. Then it goes through it and figures out what it can do in serial or what it can parallelize, like it has a bunch of steps.

But then say at the end of a return point, it says, okay, this is the output that the models prepared for me. For this section of the plan, does this meet the needs? Am I verifying it? Was it, are any of these hallucinations? Is everything properly referenced, et cetera, et cetera?

You could just fire that into something like a Decisions API or a Jav and be like, are these conditions met and it'll score it? that's typically something that would have been much harder and consumed way more IP, like, API tokens to ask an out Lent to do. So these models are going to be big game changers when it comes to actually developing the harnesses, truthfully. Fascinating. So I just, I just wanted to note it because, you know, we noted we talked about GPT when it came out.

Yeah, it's like, it's more little like organs inside of this organism of like, okay. nervous systems developing. Totally. It was like, I don't know, the nerve cells in your toe versus your brain. And up until this point, it's like every nerve system, every nerve ending is a whole brain. Like we're going to point all that energy at the little one spot on your foot or whatever. And it's like, no, what if we had a much streamlined lighter energy use faster thing for these sort of like little quick categorization problems? That's kind of what I'm hearing.

Yeah, exactly. Like the transformer architecture was always an enhancement. inherently amazing classifier. So like I've actually written some code that has used a transformer architecture as a classifier. And these people essentially took that one step farther and turned it into essentially its own model. So it has all of the knowledge of a real model, but with essentially a really light API front

end that you can just ask it classification questions. And it will score the different outputs based on what the likelihood is, which is kind of it generates tokens of output, right? Like, we send it a question and all it's doing is looking at the next highest probability token to send out. That's all it's doing. So really, that's all this is doing. But instead of returning the token, it's returning the score.

Something that's very fascinating to me as a developer that builds harnesses. So I wanted to mention it because it is a big, it is a big deal and it is a big, big enough deal that within weeks, Open AI had their own announcement of one. Oh, it's another one of those, sure. Someone rolls one out and everyone else is like, we have one of those too. I think the speed at which these companies can all sort of just gobble up the ideas from their competitors is like in a world where copyright infringement is no more. Everyone moves kind of at the same speed. Totally.

When code is free, like if you're open AI or Claude, you've got the infrastructure and the compute horsepower that you can generate pretty much anything you want. Yeah. And you've got teams of some of the best. you know, harness engineers in the world. Yeah, sure. And if it's already been done by someone else, you can just give the finished, give the Git repo from whatever project you want.

Give the, just give a design document describing what it achieves. And it's like, sure, we'll just take one of those. Send the web link. It'll write its own design document. I mean, that was the speed at which dot, like I'm sure dots has been under development for a long time.

The, I want to say 3D custom animated little dot. guys, which is interesting, uh, someone took a while to work on those. Like, I believe this was in process for quite a while, um, but you can't imagine the situation where Muse came out and they were like, you have 24 hours to make a comparable product. It doesn't have to launch, but it has to look finished enough that we can do a big one more thing style, uh, marketing announcement. The question I want to end on, Scott is dots and muse, which one is cuter? I don't want to talk about functionality. I don't want to talk about privacy.

Dots. Dots. Yeah. I do want to talk about something. Like, I want to push back on something you just said, even though we probably should end this episode and move on. But the,

uh, if you're still here, you can just stop listening at any point. We appreciate you. You've made it to the end. But we're, we're in the trough now. Please.

Dots does feel very bolted on to me. Like it is inside of the codex app. Right. It's, it doesn't have, like the interface is literally just codex, but with a new logo on the side. Sure.

Like when you, when you deal with Muse, or you deal with Grockbot, it feels like they've built, they've built an application optimizing for the delivery of the specific service, where Dots just feels like, hey, we're launching this,

shove it into the next, like, release of the app. So to me, to me, funny enough, even though Open AI is the one that hired, the OpenClaw founder,

they, to me, feel like they have released what I would consider to be the most bolted on feeling of the products. Interesting. Yeah. Yeah. I don't think I would disagree with that.

That's pushing back on something I said. I think that that's probably true. Because like Muse feels very like we've been building this. It's the, the hardware is for as much as I don't want that hardware, the hardware is the evidence that like you've been, you've been working on something. Like you had to make this little Apple watch without a strap at some point.

Like, yeah. Yes. To me, to me, you could tell that meta's been structuring into this. Yes. Grockbot felt like somebody took OpenClawn and said,

what is the best version of this? And let's build it custom now. And they did that and it's really good. Dots to me feels like OpenClawn, number one in the app store and it just got bumped to number two by Muse. Where's our solution for it? Get it out the door.

We have an IPO to get to. Yeah, sure. Kind of my read on it. they are cuter though the dots the dots yeah did you look up the grockbot dots no grokbot dots

oh they're flat oh so this is interesting I have more of an opinion and insight into this type of thing in terms of like the visual design of these things grok bots and dots are very very similar

grock bots are just flat and dots are sort of a faux 3D rendery Pixar thing. The other thing I find fascinating about the visual design of these things is that muse, so dots are a series of non-choices. It's like, what do their eyes look like? And it's like, yes. It's like some of them have little recessed black dot eyes.

Some of them have big Pixar eyes with the white circle around them. Some of them have drawn line eyes under a pair of like glasses, just like a swoopy line. So there was no choice made there. Whereas muse eyeballs are just always these little. dead eyes. Just like a little two pinprick black things. It always has the same sort of expression on its face. Grockbot's kind of in the middle.

It's these little flat shapes, but they're different shapes each time. Color code them. You can choose the shape. You can choose the eyes and you can choose the color. Yes. I feel like dots is going to be very much the same. Where a muse is like a creepy teddy bear that.

It's a creepy little teddy bear. I'm sure has been, I'm sure there's been so many reasons. research groups meet about the design of this thing that it would make your head spin. 100% empires. Empires. Millions of people have been paid to sit in in research sessions and respond to how they feel about every version of this before it got released.

And they're very customizable. In one of the mockups, it was like a dog instead of the little like teddy bear thing. So it's like, oh, they're definitely AI generated. They seem to have been like curated by hand to have certain aesthetics. but they're fundamental. The other thing I find fascinating from like a UI, UX perspective is the pluralization juxtaposition where dots are plural.

You can have dots. And they're meant to be these little like Navi from Zelda-esque things that flutter around and do things. You have many of them. I think meta and muse seems to have made, I think, a pretty important decision to make it like, no, it's your muse. You can change what it looks like.

You can change how it talks. But there's one and it's, yours. And that's like a non, that's an arbitrary choice that has no impact on how it works, really. But it's this like emotional relationship thing. The back end would be like, okay, spin up another one and don't tell me and say it's the same thing. But it's this like emotional design choice. I disagree. I'm going to push back again. I think that the big difference like between so Grock bots, Grock bot, which has multiple bots. So you literally spin up a persona, a soul.m.D., for everybody, for every team member support person, however you're using it.

Sure. I see what you're saying. So you're spinning up agents for specific roles. Those agents can then collaborate on a larger project. There's kind of a more organizational structure to it. That's crock over here. That's croc over here. Dots I have not used because it literally hasn't launched in Canada. Thanks. I believe it is the same structure. You can spin up multiple agents where, so you're kind of getting domain specialists and you can engage with them directly or you can put a manager bot or like a personal assistant bot or an executive assistant or whatever you want. A chief of staff is a classic term in the bot in the agendic bot world.

Yeah. Who coordinates all the bots underneath you. So when you ask them for something in market research, they'll talk to the market research agent. If they ask them for this, they'll talk to somebody. Anyway. Yeah, great.

Muse is meant to, I think, be like a personal assistant. That's what I'm saying, yes. So you only talk to the orchestration bot. If they want to spin up sub agents to do tasks in the background, they will do that. But I don't think they'll ever, I don't think they have a hierarchy of independent bots that use different computers. They just have the ability to kick off subagents to go, hey, go look up the 10 cheapest flights to Cabo during this week. And they'll kick off an agent to go do that. where if you were using something like Dots or GrockBots,

you might have like a shopping agent who that's their entire role is just going out to find you the best prices. And they'll come back and ask you a bunch of questions about what you're looking for. And you interact with them much more like a staff member. Yes. So I think there is a structural and architectural difference between the two. I hear what you're saying.

I just disagree. No, but it's like, how do I put this? It's like those are the same thing, but meta is making the design choice to never let me talk to the subordinates. So yes, that is an architectural difference, but it strikes me as more of a design choice, where it's like we, we're just going to never let you talk or spin up any subordinates. We're going to keep it high level with clipy.

So you are right. There is a structural difference, but to me it expresses itself is just like meta wants this to just be, meta doesn't want you spinning up a sales agent. The sentence spin up a sales agent is like arsenic to these products. I would imagine from meta's perspective. They're like, talk to tell the robot what to do. But I want to spin up a subagent that's going to coordinate with the, then this isn't for you, brother.

Tell the robot what you want it to do. And it will spin up the teammates and the subagents and all that stuff. Is the sense I got from watching it is that they're trying to like, like de-hackerify OpenClaw. They're like, do you want to create subagents that can do it? then don't use this. Yeah. This is for.

This is the family friendly one. You just talk to, you talk to Jessica. Yes. Jessica does what you needed. And it doesn't matter if it's figure out my meta marketing strategy or sell my old tennis racket on Facebook marketplace. Jessica will take care of it for you. Actually, good use of it. Yeah.

Well, the funny enough, I've read some, well, actually the, I read some X chat. about Muse this weekend and somebody had given a control of their Facebook marketplace, said, hey, you know, deal with these people that keep messaging me about all these things. It accepted low balls. It was like,

it was just, it was just going to town. Yeah. So I'm not sure if it's the best. Yeah, I don't sure if it's the best for that stuff. Yeah, sure.

Oh, that's interesting. And I mean, that's the other, we haven't even talked about the like efficacy. We've talked about that, not efficacy, but like the privacy.

We've talked about all those. dimensions. Then there's just the question of how good is it going to work. And we're recording this too soon after it came out for anyone to know, which is like a really interesting moment. So in a couple of weeks when we do another chatty chat episode, we will probably report back on this. There is a to speak about efficacy. I could talk about these things all day because like I literally am fascinated by this. The once you go to a muse model where say you have a one conduit into a team, your token spend and use goes way up. Because if you want to speak to the sales agent and you go directly to them and say,

hey, how were sales last week? What's our plan for next week? Or whatever, however you're using it. You go straight to them. They ask you the direct questions. You kind of get into that cycle with them. And it only consumes the tokens and they have their own cash for the cash data. So it's much cheap.

The second you go to the orchestrator and then you ask them and if maybe they go to like the Jordan's sales team manager and then they ask the person who's in charge of that exact sales plan, you're paying for tokens and caching and inputs and outputs and redigestion and reformulation all through the tree. So if you're looking to maximize the amount of compute required, creating a. corridor through somebody like a muse is way more costly than then allowing people to go to direct agents that understand the context of the question you're asking it in rather than going through the like a conduit. That's fascinating. I think that how these things are architected probably will become like a sales differentiator, like not to turn the tech talk into like the marketing business side of things.

But it's like eventually these things are going to live and die by like are people willing to spend money on them and to buy the hardware that comes along with them. And I'm putting the number of metamuse Tamagotchi's I'm expecting to see out in the world at zero. Open AI, maybe a better shot at it. See what Johnny Ive does. Let's find out if that $6 billion was worth it. Apple and Google with the pixel line have some funnel for getting you to carry hardware around. Like just the basics of not making a product called pervert glasses.

Like they just have a totally different relationship with the consumer hardware market than meta does, and certainly that Open AI does. I would say that I see, the amount of meta ray bands that I see these days is shocking. They're very common. They're a rounding error to smartphones. I guess this may be where I'm going with that. You know what I mean? But I think that matters.

I think that matters. That is the cluster. I'm not going to show my phone right now, but that is the cluster of cameras and microphones and GPS trackers that we already carry with us in the absence of a killer product. that will remain the hub in which all of this stuff happens. So I'm still thinking of it as like it is the phones race to lose as the heart of the AI hardware thing that everyone uses. Apple's trying to be like, you already have AirPods.

Could those have a camera in them? And it's like TBD. If people will go for that, they've already done it with the Apple Watch. But I'm like that constellation of accessories orbiting the phone. I'm like, that's, that's probably going to be it for the foreseeable future. and two companies have a stranglehold on the software operating system of those devices. They just do. To your point, like, yeah, I'm like the phones, like phones blew up.

The iPhone blew up because it was extensible, right? People had the ability to custom build solutions and tools to install on it. You know, the marketplace, the app store is really what, like, I would say one of the top two reasons why, you know, the iPhone changed the world. It changed our relationship with access to information in a self-discovery sense. You know, notifications are not so self-discovery anymore, which is why I have them mostly turned off because they drive me crazy. But I think our relationship with access to knowledge and information and the extensibility and the ability to extend the product through the development of applications,

installation of tools and utilities you wanted is what made the phones cool. Yeah. And not just cool, life-changing history defining. The person who needs to make the AI, physical AI tool, needs to think about those situations, think about how our relationship with knowledge is changed. Think about self-discovery as still a key aspect of it instead of forced discovery. So that's why I think the Google XR project was the one that I was most interested in, because I feel like they had the best perspective on it where like I don't.

And once again, going back to my trust for the meta corporation, not so high. I don't know if I want to slap an always on set of glasses on my face. My face sees a lot of stuff. And I think that this is like a, this is a problem for the whole wearables market is the, the floor of privacy expectations is so much higher on a thing that is always on my face. Well, we also to talk about efficacy again. the second you put a live video stream that you're piping into an LLM, asking it to do visual recognition and then knowledge generation and then delivering stuff to you in a real-time thing.

The token used for that. Like you'd hit a $200 monthly subscription in minutes. The compute implicate 100%. And those models will get more efficient. You'll find a way to run it locally on blah, blah, blah, blah, blah, but 100%. I'm in agreement that the core functionality promise of these things, which is the robot already knew is like the implications of that are dizzying. Like you you are you are it is a again to firehouse metaphor of data that every human being is

producing unless you are finding a way to run that locally on a device that I guess is just plugged into the wall because you would still have energy use. You have like punted the problem of this. We we can do it. It's just not efficient enough to do it in the hardware that you're talking about in many of the prescribed use cases. Yeah, and I think like the general purpose classifier type models, the Jev's and the Open AI decisions, I think that they're going to figure out ways to optimize and make those very lightweight,

like be able to really crush the amount of dimensions in them to make them very usable in a specific medium. So all of a sudden you'll be walking around with your glasses and it'll be feeding a frame being like, is there a caradness? And it'll be like 68% yes. It's like, okay, that's not high enough. It doesn't meet our threshold. And et cetera, et cetera. So I think that we're going to see some of those highly tuned micro models get really good. And especially once you have them built so that they're so fast and so cheap to run,

if they could run something like that locally, even if it was paired to your phone and it was running on your phone, that's when I think we open up a lot of the gates that allow more of that knowledge discovery to come in to play without being so token intensive. Yeah. But that would also be my key argument for not having it forced knowledge discovery and making sure that like it only responds when I ask it for something. I don't want constant notifications every time I look at something. If I look at something and say what kind of car is that, I want it to tell me. I don't want it to just be constantly telling me everything going on around me.

Like I've got enough things in my mind already, let alone have the overwhelming amount of information flashing in my eyes 24. hours a day. Everything you just said is the reason I don't want the architects of the suck up your attention and monetize it with an algorithm. Like the people who I most associate with that kind of toxic behavior. In like in a court setting, the dopamine cycle, the like let's get away from the linear feed. Let's get to this algorithm that pulls you in. It takes things that we're to show you the thing that's going to bring you in. And if we show you something that kicks you out, we're never going to show it to you again.

The smartest people of a generation working on that problem of my attention. And now it's like, you just described like, and I don't want hardware that works that way for this new computing style. And I'm like, guess who I don't trust? And it's not just because of privacy leaks and stuff like that. It's not that service level thing. It's about how they approach design problems and the preferencing of the company versus the user.

I'm like, I harbor a skepticism about that. Anyway, Mews. Muse has a booty. Anything else? Thanks again to our sponsor Nordlayer. Check them out, Nordlayer.com slash hack podcast. Thank you, as always to Nordlayer.

This was a fun one. If you're still here, hey, we're really happy to have you. I don't think we've done one of these since we've switched to video. And not, don't drop us a comment, not for like the engagement harvesting,

like classic YouTubers or Spotify podcasters. Just drop us a comment to know if you made it here. Yeah, drop us a comment if you made it to the end. Thank you again for joining us for another one. This is fun. We haven't done a chatty chat in a while, so it was a good time,

and we'll catch you in the next one.