Stop Using Web UIs: Why LLM APIs Are 10x More Accurate

[00:00] Dave Pengelley: I think I gonna put it out there. I want a longer bumper if we've got the 30 sink cat out and then we've got this like bumper and that, that's like we are here. have you guys played around with any ai, video platforms of any sort? Anyone? [00:58] Marno Brits: Playing withs field. Oh, you have Matt? [01:02] Matt Slager: Yeah, I'm not a lot though. [01:03] Matt Slager: I can need to do it more for content and stuff like that, but yeah, through, through like Fal and then through Gemini directly and all that kind of thing. [01:10] Dave Pengelley: Fal? [01:11] Matt Slager: Fal, yeah, it's just like a reseller. So it's kinda like open router for media generation. So FAL ai, okay. Fantastic platform. [01:21] Dave Pengelley: Yeah. Nice, nice. [01:22] Dave Pengelley: Because, one of some of the big stock photo groups are doing their own kind of thing with that too, where they're bringing multi levels under one house. What's the main one? I can't remember. Used to use them anyway, it doesn't matter. What are, what are you using? Mine, [01:34] Marno Brits: I wanna start using, Hicks Field. [01:36] Marno Brits: I've gotta create one for tomorrow. I've got a demo for a client, [01:39] Dave Pengelley: Hicks. [01:40] Marno Brits: I'll be using Hicks Field to create a, a copy of a grandmother that can then just talk and answer fun questions. So I'm quite keen to see how that's gonna turn out. [01:50] Dave Pengelley: Oh, see, Grandfield. Higgs. [01:53] Marno Brits: Yeah. Hicks. Hicks Field. Yeah. Is it [01:54] Dave Pengelley: Hicks Field? [01:55] Marno Brits: No, no, no. Hicks Field. Hicks Field. [01:57] Dave Pengelley: Like the scientists. [01:58] Marno Brits: they seem to be the, the leaders in it, from my perspective at least. They're doing the best of the best. [02:04] Dave Pengelley: All right. Are they doing like, like avatar type stuff or they're just doing anything? [02:08] Marno Brits: Just doing anything. [02:10] Dave Pengelley: I see [02:10] Marno Brits: again, I know of it. I've not had to play with it. [02:13] Marno Brits: I've just seen content and tutorials on it. But that's the, the source you go, then they choose what you wanna try and achieve, but then that again connects to the other models. [02:21] Dave Pengelley: Yeah. They're like foul then and Nana, banana Soul clinging all them. What about you, Richard? Do you, you play with AR video. I know you like making a few comics and cartoons here and there. [02:31] Richard Webbe: Yes, I do. I enjoy that. one of my, I think there was one that I did with me going like this with all the AI symbols around me, like someone who's overwhelmed by ai and just to be honest, 'cause I'm not that smart. I didn't know I created a video. I just did it by itself and I can't remember what platform I was on and I thought it was a picture and it came up with this. [02:53] Richard Webbe: But it was a really cool video. I liked it. People said they liked it 'cause it, it kind of metaphorically really related to what I was thinking, what I was trying to communicate at the time, about being overwhelmed by it. I, I, I actually, you know. my kids, I dunno about you guys. My kids hate AI because they've been taught that this is killing creativity. [03:13] Richard Webbe: And I come from a creative family, so a lot of my family members go, this is all wrong. And I have a theory about that. And that is AI is like the sum net average. It's like music in the seventies, it was all the same, right? And it wasn't until people like Pink Floyd and a few others came out and did something completely different that we just were happy consuming the same thing. [03:34] Richard Webbe: And that's how the mega bands got started because Yes. Yeah, yeah. Sorry, kisses. So whether it was branding or, or music, right? And so people say to me, oh, you know, I'm a creative and my life's gonna be destroyed and my career's gone because, and I said, no, it's not. Because all AI does is create what I call the sum net average of what people are thinking and asking. [03:58] Richard Webbe: And so you won't stand out at all. It'll be a good way to communicate. It'll be an automation, then a speed up way of communicating an issue or saying something. But if you wanna stand out competitively or be different, you're still gonna need your creative input into that view of the world. And, yeah, so I think, I love AI video. [04:17] Richard Webbe: I love AI pictures, but I think people are gonna use it a certain way for a while. And God, I clever. And then pretty soon that's gonna get boring quick. Like, I noticed a lot of prompts for certain video AI and, and all being populated. So everyone's gonna be doing the same thing. So what's that saying on the Incredibles? [04:37] Richard Webbe: Like the buddy says to the Mr. Incredible, I've got tools to make everyone a superhero. And if everyone's a superhero, no one's a superhero. [04:45] Marno Brits: Mm-hmm. So [04:46] Richard Webbe: it's not that helpful if it's just doing the same as everyone else. So getting your creative input, then putting it into production quickness, I think that's where the cleverness comes from. [04:57] Matt Slager: I love it. How good's, the incredible reference. [05:00] Dave Pengelley: It's incredible. [05:02] Matt Slager: Yeah. The, there's a word. There's a word for what you're describing there, Richard. It's, it's called Taste. Oh, thank you. [05:11] Dave Pengelley: I thought, I thought you were gonna say equality. oh, you mean, you mean the musical thing? Yeah. I mean the Thein and, and even the electric guitar, like, did people think this is gonna destroy music, this electrification of the music? [05:23] Dave Pengelley: But then we've got amazing things and the's a bit of a deep, [05:26] Richard Webbe: Matt, you had a little bit of a tag to go on the back of taste there. [05:31] Dave Pengelley: Good [05:31] Matt Slager: taste is literally what you're describing. It's the, it's the spice that each individual brings and, you know, when everyone can go to one tool and just generate an amazing website or an amazing piece of software instantly without any trouble. [05:45] Matt Slager: The differentiator, the thing that makes it not just all the same, you know, everyone being all a superhero is your unique taste and what you bring to the table. So [05:54] Richard Webbe: yeah, [05:54] Matt Slager: that's, that's something that I always really stress about when people are trying to like, make clones of themself or write, get AI to write content for them. [06:02] Matt Slager: It needs to use like your unique taste. Otherwise there's no point. [06:07] Richard Webbe: Yeah, that's true. So we can automate everything as we've been doing for 3000 years. Mm-hmm. Back to the Pharaohs. And every time someone automates something, someone says, we're all over. It's all done. I hear it on the radio every morning, but we're not. [06:21] Richard Webbe: Yeah. And then like you said, the taste bit is that little trick, that nuance. As my dad once said to me, he said, who made the most money outta the hairpin, Richard? I said, I don't know the guy that invented Hairpins. He said, no, the one that put the little cink kinks on the end. And it's a true story apparently. [06:41] Richard Webbe: So I'll go with that. Yes. Yes. And I love the incredible, what a great show. Mm-hmm. Great bird. An iconic genius. [06:50] Dave Pengelley: Yeah. Yep, yep. And everyone's trying to mimic and make their own things. Now everyone's trying to make their own cartoons and thing. There's video gen stuff we talked about. It is becoming easier, but it comes back down then to the, the creative behind it. [07:02] Dave Pengelley: So if everyone's got the same tools, and here's that foul one that you were talking about, I think Nana Bananas and Clings and everything. And then, oh look, let's go to this other one, which has Nana Banana and Cling. which is the one you're talking about, mano Higgs field, and then traditional players. [07:19] Dave Pengelley: I remember the name Envato, who do not sponsored. None of these sponsored Feel free to sponsor us. That would be awesome. Envato who are traditionally stock photography and, and video and sound things. And, for creators, like I remember when I first started trying to do creative stuff three years ago. [07:35] Dave Pengelley: AI generation didn't exist. And so I signed up and paid way too much money for Envado for the amount that I used it. Mm-hmm. And but now they've got AI gen building in there as well because they know they need to keep their audience, like having all the new toys as well as the old toys. [07:52] Richard Webbe: I grew up with this creative disruption 'cause my dad was a commercial artist, an art director, and he started his own advertising agency, many, many years ago. [08:02] Richard Webbe: And he used to tell me, because he loved technology and we always spoke about it and he brought his kids in, right? And so me and my two brothers were all technology nerds like you three. I'm, I feel honored to be a, consider myself in a group with U three and, and so my dad embraced technology where everyone else didn't. [08:20] Richard Webbe: So you guys are too young. You're all in. Nappies. But in the old days, the, the old age Harold son, the old Sydney Morning Harold, down the side were these hand drawn pictures of either ladies' clothing or houses that were the biggest things that got, and you couldn't do any type setting of pictures. In those days, it was very hard. [08:39] Richard Webbe: So a lot of everything was hand drawn or hand put in. And Dad said, oh, these computers coming along, Richard, they're gonna change all of that. And I might be out of a job. But he worked out very quickly that the art direction of how to use those tools to create the message was what was important. Not the automation of the computer. [08:56] Richard Webbe: So he kept going and got even better 'cause he embraced the technology. And that's where we're at. Video pictures. Yeah. And letters. People writing all these emails and letters to people. Give me an email from the LLM that will convince Mano to, you know, give me something. Yeah. Everyone's writing the same email. [09:16] Richard Webbe: You can help anyone. [09:17] Dave Pengelley: I was writing a connection request to someone on LinkedIn this morning, and I was, well, I wanted to reach out to them, around something that I'd seen on their profile. And you got 200 characters when you connect and say, add a note and you only get three of them a month if you're not paying for LinkedIn. [09:30] Dave Pengelley: So I was like, Ooh, I wanna get this right. and so I drafted something up and then I, I went to the LLM and said, Hey, this is what I wanna talk to them about. and this is my contextual, like anti-gravity environment there knows heaps about me. Like it's not just generic chat GBT, but said, I want to talk to this person, gimme a about 200 characters. [09:48] Dave Pengelley: And it was rubbish and it was just so generic and didn't say what I wanted to say. And I was like, what about this? And, and it needs to take into account that, you know, it needs to say hello because it's a, it's a connect note and things. And I was just, eventually, I just wrote my own one. I went, what do you think of this? [10:00] Dave Pengelley: And he goes, brilliant. You've hit all the notes. I was like, thank you. Thank you for the confirmation. I'll take confirmation and use my own words on that connection [10:08] Marno Brits: request. I feel like it's definitely there just to amplify like a great analogy that I've seen. I think it was Gary v. Talked about it. A fa everyone used to be farmers, like we, you hunt, you make food, come back in, and then you eat the food and then you go back to work the next day. [10:21] Marno Brits: And then we have like industrial revolution of like, now we have tractors that rather than one person having to do it, it's the work of 20 people in one machine. And what I'm trying to communicate there is like, AI is not there to necessarily replace anything or take over anything. It's gonna amplify the existing skills that you have. [10:39] Marno Brits: Exactly. If you don't have that skillset. Same with Match said a couple weeks ago. Like if you communicate to a team, you can communicate to ai, but you need to have the skillset of, like, you've had, you've been in the role like David, where you can say, well, I've worked with a sales team, or I've worked with creative teams. [10:54] Marno Brits: This is what I needed to accomplish. Otherwise the outputs is gonna be slop. Like you need to give it that direction. And that's the scary part of like the education system. Everyone's going for university using Chacha PT to cheat on the exams and stuff. Then they come into the real world. They're like, crap. [11:09] Marno Brits: I never had to think for myself. Now I need to create content or deliver on this role that's paying me all this money. But they've got no idea where to begin. And they have to then learn while being in the shit, for lack of better word, they percent. [11:22] Richard Webbe: Yeah. Domain expertise is the an experience. [11:25] Marno Brits: That's [11:26] Richard Webbe: it. [11:26] Richard Webbe: It's actually going to become more important. And people like us, when we create AI solutions for our customers, we're gonna be leaning on that domain expertise of those individuals who think their gonna lose their job or something, but not their input, their experience, their connection with people. I used to sell SAP and e, A-P-E-R-P solutions, big ones. [11:50] Richard Webbe: And, with BHP and a few other big company. And I remember I walked into one particular company and the lady that did the, The logistics and the transportation and that, you know, had a giant whiteboard. That's how she worked, coordinated everything. She was the best, excuse me, how rude is that? She was the best, out of all of them. [12:10] Richard Webbe: And, and I said to her, wow, this is great. Look at all this stuff you've got on the wall there. I want to talk to you about how we capture your expertise put into the air. And she just got up and swore at me and stormed out because, you know, I was the evil, yeah, coming to take her job now. I'm not sure how her career ended up with that company, but her job didn't go. [12:35] Richard Webbe: Her job just got more powerful and got more efficient and got better and, and doing so. I'm not sure if she stayed or left or, you know, the whole emotional, and that's what we're talking about. Everything. There's no typing calls anymore. Right. But we didn't kill people to have the brains on how to communicate and write a good letter. [12:52] Richard Webbe: There's no, you know, Excel didn't kill all the accountants, although LLM may kill all the lawyers, but I'm not sure about that. I'm just, [13:00] Dave Pengelley: I, I miss, I miss people actually writing their own content, though, like you LinkedIn. Yep. There, there's two main issues I have with platform like LinkedIn. It's a, it's all the generic AI written slop. [13:09] Dave Pengelley: And I'm not immune from actually using AI to help me write stuff, but I'm not just doing generic ai as, as we discuss, as we know here as AI operators, we're operating a level above the average punter. [13:19] Richard Webbe: Yes. [13:20] Dave Pengelley: So there's that, but there's also just the way the LinkedIn algorithm and the way they feed and the UI mechanics work to just show you that first sentence and they make you click through. [13:27] Dave Pengelley: And so people write everything as click bait and engagement bait. And they don't just give you something and they wanna insist, comment this word to get this thing. And I'm like, I'm just so sick of all the engagement farming. [13:38] Marno Brits: Mm-hmm. [13:38] Dave Pengelley: I, I just, can we just go back to sharing ideas, information and if I follow someone. [13:44] Dave Pengelley: I wanna see what they're putting. Don't, don't go. I'll show you what they pointed out three weeks ago if it managed to somehow get popular. Mm-hmm. Like the feeds used to be like, show me what's new, show me the latest stuff. And now it's all this algorithmic stuff, which is just confusing and messy. And I get all these promoted things that I didn't ask for. [14:02] Dave Pengelley: I get that we're not paying for it. So with a product, but it's, it's very frustrating to use these, these things because of the abuse of ai. [14:10] Richard Webbe: Yeah. It's a sum net average. It's not, that's it. Not the, the, the peak. And, I used to, say to my friends when we used to go to the nightclub to meet partners and stuff like that, you know, you know, don't, don't talk about yourself and don't do what everyone else is doing. [14:26] Richard Webbe: Just turn up and ask the person about them. [14:29] Dave Pengelley: The, the problem is the algorithms don't reward that. The algorithms go well, everyone else is doing that thing and that looks pretty popular. So you better do that. And if you're not doing that well, we dunno what to do with your content. We're just gonna, we're just gonna deran that because it's clearly different and weird. [14:43] Richard Webbe: Well, we've got ways to fix that, haven't we Dave? We've got ways to do that and, and, and, and getting that technical expertise. Matt, I'm sure you've got a view on this, getting that technical expertise and then capturing it in an automated environment in your own sovereign AI to represent to the world that that, that that's not hard to do, is it? [15:02] Matt Slager: Not unnecessarily the, the idea of what Dave's talking about with the algorithm serving you certain stuff and, having your own sovereign ai. I'm not sure where that links, but yeah, there's, there's a lot there. I, I kind of follow everything on X now for that reason, Dave, like I've sort of stopped following things on LinkedIn. [15:22] Matt Slager: LinkedIn is like the show off feed. You know, it's here is one small marketing email after another. [15:28] Dave Pengelley: Yeah. [15:29] Matt Slager: XX [15:29] Dave Pengelley: isn't exactly immune from engagement farming though. [15:33] Matt Slager: No, it's not. It's different. But like [15:35] Dave Pengelley: what, what, what do you notice about this picture? It's like, what do you mean? Like, what do you mean what do I notice? [15:41] Dave Pengelley: This is a random picture. They like, they deliberately put vague, meaningless comments out there to make you go, oh, I think it's because I saw that little curve on their shoulder. I think it's this. And so they just engagement farm, I mean the algorithms are broken. Human interaction, humans, were not built for discussing things with thousands of people every day. [16:00] Dave Pengelley: At the same time, were built for one-to-one interaction. And I know it seems weird as the AI guy to be complaining about all these AI and digital interactions. Mm. But I think it's important that, like you were saying, Richard, the power of all these tools to generate things is based on the people behind them. [16:14] Dave Pengelley: Yeah. and if we're just abdicating everything out and just overloading ourselves with these things, then we're not having the time to focus and think about. Why we're doing what we're doing. We are just doing the next thing and we become, [16:26] Marno Brits: yeah, [16:27] Dave Pengelley: algorithmic computers just task and task out content and content out. [16:30] Richard Webbe: Intellectual spam. [16:32] Dave Pengelley: Yeah. [16:33] Marno Brits: I feel like I would've to disagree there boys for specifically x.com, I don't use it heavily, not nearly as much as Matt, like I'm on social media much, but I find out of most of the platforms, x.com is the only one that consistently, or from my feed, most people are there just to provide value. [16:51] Marno Brits: Like a great example is levels. He doesn't do any engagement farming. It's just a bunch of tech people that are like, I learned something, how about it? Same with YouTube, like PewDiePie, like it's not about, Hey, I wanna get a bunch of money that they've reached this pinnacle where like they're doing it for the love of the game. [17:07] Marno Brits: Yeah. and it takes probably a while to get your algorithm to that level where you stop just getting slop and like engagement where they constantly want to keep you on board. But building that, I guess that dopamine reset where you just get to see value and read value where it takes time. once the old group picks up that that's what you are after you get served a lot more of that. [17:26] Marno Brits: From my [17:27] Richard Webbe: experience, man, no, man. No, I agree. And, and the altruistic view when you look at it for humans is actually much higher on their pecking order than money or prestige. And that's how Jean Rotary wrote, star Trek. It was about the future where there's no money. [17:44] Marno Brits: Mm-hmm. [17:44] Richard Webbe: And your currency is the value you add to the people around you. [17:48] Richard Webbe: Now, funny enough, right on the money, it's interesting. There was a, and right on the money, that's what Elon Musk was saying a couple of months ago. He said, I might, we might see in this generation the end of money. Mm-hmm. And then you become altruistic, like, you know, whether you're a Malcolm Turnbull or Trump or Clinton or Hawke or any, they were already multimillionaires. [18:14] Marno Brits: Mm-hmm. [18:14] Richard Webbe: When they became leaders, I don't believe any of them did it for money. They all did it for deep down altruistic reasons. They measure, measure themselves differently. Yeah. But no one goes into public office and leadership to make money. Well, maybe some do, but they, they get found out. So I agree with you, mana. [18:31] Richard Webbe: I get this whole altruistic, how can I help the people around me or, yeah. And, and it, and that does come back to your ego, right? [18:40] Marno Brits: Mm-hmm. [18:40] Richard Webbe: But if, of course, if you're sitting back there as a cheap sales person or a cheap architect or A-A-L-L-M, give me an answer to this. Everyone's already got the answer. [18:49] Marno Brits: Mm-hmm. You make a very good point there. Yeah. I love that. I never knew, I've not watched Star Trek, so I didn't know that was the, the purpose [18:57] Richard Webbe: of it. That's the core belief. That's Fascinat that he believed. That's why Star Trek, I think, resonated with a lot of people. And, and, and, and I think Star Wars took its feed from that, Lucille Ball was the executive producer, believe it or not. [19:10] Richard Webbe: the Desi studio and his whole idea, and he was very unforgiving. This is, we're gonna paint the picture for our kids on what the future looks like, doing the right thing. And that's why every show had a moral imperative. Every show wrapped up a certain way. And I, to be honest, I'm a tricky, but I do subscribe to that view of the future. [19:30] Marno Brits: Mm-hmm. No, I like that view of the future a lot. That, something similar to that but not really on value is the movie time with Justin Timberlake. Oh, I [19:39] Richard Webbe: like that. [19:40] Marno Brits: But that movie, I've watched it probably two or three times. 'cause it was the first time where it was like retraining my brain. It was like, wait, the real true currency, is it money? [19:50] Marno Brits: It's your time. Like how much are you exchanging for this cup of coffee? Like there's a great quote, of like, the true cost of anything is the amount of life you exchange for it. And that perfectly summarized that movie and just sell. Some people take that for granted on how little time or how much time they have. [20:10] Richard Webbe: I agree. I agree. very, very well. Matt, you looked like you had something to throw in there. [20:16] Matt Slager: My brain jumps out with fireworks on every single sentence. So, yeah, like if I start talking, it's gonna be about a topic that's completely random but parallel and, but Matt, [20:29] Richard Webbe: you and I are the same like that. [20:30] Richard Webbe: Go for it. [20:31] Matt Slager: So I was just thinking before when Mana said it takes a while to train your algorithm. That's a hundred percent true. Like, if you've never been to a particular platform, like if you create a new account on a platform that you already are on and you see what it serves you on a brand new account, you'll be horrified. [20:47] Matt Slager: You're like, what is this [20:48] Marno Brits: man? Yeah. [20:49] Matt Slager: So yeah, like it, I think you liken that to, I'm just gonna pretend like I'm someone really social and I have a lot of friends, where, you know, you get to the point, you know how people say, like if you. You just hold onto friends for no reason. And it's, it's okay to let them leave your life. [21:05] Matt Slager: And you, you maintain the ones that you have really good connections with and that sort of thing. It's the same thing with social media. If you don't wanna see something, unfollow it, block it, tell the algorithm that you don't want that and promote the stuff that you do want by, you know, liking it and all that kind of thing. [21:21] Matt Slager: That's, that's what they want. So [21:23] Dave Pengelley: it does help. I, I went and prune a whole bunch of people from previous jobs and things outta my LinkedIn thing. 'cause I was like, I don't need to see their content. And a lot, the thing is like, I think only 1% of people on LinkedIn regularly post, like, LinkedIn doesn't have a huge percentage of population versus content creators on there. [21:40] Dave Pengelley: Mm-hmm. which is wild. So if you're seeing lots of things from the same person, you're sick of that, then you can unfollow them, you can p prune them and, and then that changes it. But also, like you said, what you do, like what you follow, like I did manage to change some of my algorithm by pruning some people liking other things. [21:54] Dave Pengelley: LinkedIn specifically does seem to show you stuff about the stuff you post about. So if you want to see things post about those things and then it will feed those things back into your algorithm. 'cause it goes, oh, you post about, did ROI in AI algorithms, and then all of a sudden you start seeing other people talking about ROI and AI connected together. [22:13] Dave Pengelley: and so like, I'm not saying they're completely useless, but I'm just saying I am sick of the spam and the engagement farming. And I wish that humans would learn how to connect with humans again. because we are gonna be connecting with robots a lot more. Robots are very useful. I talk to mine every day. [22:28] Dave Pengelley: but as long as we don't lose that ability for humans to connect with humans in the process, and remember, we're doing it all for humans, not for robots. [22:36] Matt Slager: What you're saying is that if everyone started to do more realistic raw human content on LinkedIn, it would all of a sudden just shoot to the sky. [22:48] Richard Webbe: That is it. And isn't that an incentive, Matt, at, at, isn't that an incentive for people to be more direct? Mm-hmm. Less salesy, more articulate, and more problem solving. So AI again, may shape how we approach 'cause I know about you, but we're all very sensitive to, our brains are pretty smart, right? Yeah. And they spot patterns very easily. [23:10] Matt Slager: Mm-hmm. [23:11] Richard Webbe: And you know, when a computer program or a conversation is grooming you in a certain way and you react a certain way to that, and we don't want that. Right? We want new creativity, all that sort of stuff. Right. [23:22] Matt Slager: You know, Richard, that sensitivity. At sensitivity, I think I've mentioned previously, I, I forget now we're up to episode six and it all kind of goes into a blur, but, that, that like little radar that you've got that you know, little blip when you see something or you feel something. [23:41] Matt Slager: I've used a phrase in the past to try and describe what that is and there's been many of other little analogies to describe it. But your one there is perfect. That is literally cognitive security in action. [23:55] Richard Webbe: Yeah. Oh, cognitive security. I'm gonna use that. I like that [23:58] Dave Pengelley: because what do you, what do you talk about Richard? [24:00] Dave Pengelley: The, reticular activation system? [24:02] Richard Webbe: Well, there is a, I used to give a talk to executives about how, how do they spot patents and how do they do things? And we have this thing after there are any psychologists listening. I will say, again, I'm not a scientist, I'm not a doctor, I'm not a psychologist, but I've been around a little bit and we spot patterns, right? [24:19] Richard Webbe: And there's a thing called the reticular activating system, which is a logical system in your brain, not a physical one. I think it does sit near the hippocampus or something like that. And what it does is it connects your memory, your experiences with your five senses and your needs. And so what that says is when you're driving down the road and you need to take a pee, you are looking for a toilet, not a place to buy a drink. [24:41] Richard Webbe: And your brain automatically tells you that. Now, the best part of that is as your brain gets smarter and you train it better, like any algorithm, you don't consciously think that. It finds it out for you. And I used to do this on a talk called The Lady Salesperson taught me by a great guy called Barry Freeman many years ago. [24:59] Richard Webbe: And so I get, you know, large groups of executives, and one of the great one is I say to them, when you buy a new car and you're on your way home with a new car, you know, what's the first thing you notice? And it looks like everyone else bought your car. [25:12] Marno Brits: Mm-hmm. [25:13] Richard Webbe: They're on the road everywhere. Oh my God. I just got into my Corolla and I'm my way home and everyone's in a bloody Corolla. [25:18] Richard Webbe: What happened to me? Of course, all it is, is that you've been pre-programming the algorithm of your brain. Every time you look at a Corolla, you imagine yourself in a Corolla, you look 'em up, the prices, you're pre-programming. And so there's, oh, Joanna has joined us. Good morning, Joan. You're running a bit late over there in the West, but thanks for joining us. [25:36] Richard Webbe: And and, and so I love having a national audience and so, I teach this to my kids now and I say to them. Be clear in your life what you want, and your brain will take you there. It's the whole goal setting thing. Yeah, and I think AI does that for us, and I think if we can get the merger between, you know, technical business company and people, I think that'll just speed up, get us getting to the objective that we wanna get to without alienating people. [26:06] Richard Webbe: It's still about people, Dave. [26:08] Dave Pengelley: So, so you're saying this pattern recognition is why when I look at this, I know that that's not the four of us. [26:16] Richard Webbe: God, hang on. The AI version of me is much better looking. Can I change? I have [26:22] Dave Pengelley: to [26:22] Richard Webbe: agree. [26:23] Dave Pengelley: I, he's a bit [26:23] Richard Webbe: cheesy. [26:25] Dave Pengelley: Yeah, yeah. Like I, I just, I was trying to read, redo our thumbnail because Richard, I've kind of hacked you in like ar originally did the three of us together and it kind of made us look like us. [26:33] Dave Pengelley: although it did, it did rebuilt Matt, but I think it still looks pretty Manish. Whereas I was like, okay, my older [26:38] Richard Webbe: brother, [26:39] Dave Pengelley: I was like, I was like, can we, can we do a new version and, and drop Richard in as well? And it just rebuilt all of us as different people. [26:46] Richard Webbe: So that's a technical question I have for you. [26:48] Richard Webbe: Three Einsteins in the brains trust. I get a picture of me and I tell an LLMI wanna produce a photo with a brand and I wanna put some words on it so that it's got some personality. So my customers go these Richard face, this is the message he's saying, you know, the quicker communication thing, right? [27:05] Marno Brits: Mm-hmm. [27:06] Richard Webbe: and then. A lot of the time it'll just put my picture in. It'll be great. Ah, it's Richard. He said that. Good. I get the message. Thanks. And then a lot of the time it does what David just showed. It creates someone who's not me. Why? [27:20] Marno Brits: Who wants to answer it. [27:22] Matt Slager: You can go [27:23] Marno Brits: from my understanding, it's, especially with chat GT specifically. [27:27] Marno Brits: It was great when it first came out, when 4.0 came out, it was exceptional. It was about a week where you could make whatever you want. It was just a lot of fun. And then there would be people that are obviously gonna use it for bad or they're gonna start impersonating other people or create content that shouldn't be true. [27:43] Marno Brits: and then you start finding where it'll be you, but it'll be slightly warped. So it's not exactly you and it's the most horror photos because you can see yourself in it, but it's not really you and it's just a security mechanism. If you wanna bypass that, use something like rock. Elon Musk has done a great job there where it just bypasses all of it. [28:00] Richard Webbe: What's your ROCK? [28:02] Marno Brits: GROC. [28:03] Richard Webbe: Oh yeah. Rock. [28:05] Marno Brits: GROK. Sorry. Yeah, I know. Another one is, stable diffusion. [28:10] Dave Pengelley: I mean, those ones you guys were just talking about at the start. The, the, the, aggregate platforms, because that's more of a direct API thing. They can have less guardrails on a hundred percent trademark people as well, whereas A and CHATT comes and goes, like, I generated their original thumbnail through like six weeks ago, and it was fine. [28:28] Dave Pengelley: It, it kept us as us pretty much, and now, you know, they've, they've tweaked it a little bit. They've changed it. The guardrails have shifted. So now it's like, okay, now we would need to rebuild these people. We can't just use the, we can't just crop and, and recolor the photo version of them. We need to make a new version of them. [28:43] Dave Pengelley: tragedy 5.3 is coming out any second now, and who knows if that's gonna be better or worse, but, I think. It's probably worth remembering all of these different tools that you use if you're using it through whichever website or chat interface, that's not the direct actual underlying model doing all the answering because they always have this, it's, it's a harness, it's an ad like, like it's how you get in. [29:08] Dave Pengelley: And they've actually put all their own rules above that, that actually are part of the interpretation of your prompt. And so if you just go onto the chat cpt.com website and plug in something that's gonna be very different than if you go to NAN and build your own custom chat bot and then tie it to the same model via the API because you're not getting all of chat GPTs, web logic being injected on top of your prompt. [29:34] Marno Brits: Wasn't there one study, something like 90% accuracy or 99% accuracy if used the API, but 20% accuracy if you use chat GT's interface. [29:42] Matt Slager: That was Blake. That was literally Blake. [29:44] Marno Brits: That was [29:45] Dave Pengelley: insane. One of our AI friends, he, he ran a whole model on that and when, if he gave a PDF through chat GT and asked it to, to reference things from it, it was like 15% accurate. [29:55] Dave Pengelley: But if you did the same thing through the backend, directly through the API, it was like 98% accurate. [30:00] Marno Brits: Ridiculous. [30:01] Matt Slager: Huge. And is another massive thing is the context window, which probably came into Blake's testing then, which yeah, like if you are just running CHATT threads and you know, I've seen people maximize their thread size where it literally won't let you send another message and they have to go create a new one like that. [30:18] Matt Slager: You are not using any of that context anymore. When it's, you know, I think it's something like 32,000 tokens is what the maximum is for the chat GPT model. Mm-hmm. Whereas if you go direct through API, it's like 400,000 tokens. So that's, [30:35] Richard Webbe: Joan has just asked us a question from the West about the memory situation in lms. [30:40] Dave Pengelley: Well, that, yeah, that, that's, that's this, so that's that ability for them. The, the memory of an LLM is the context. And so in all these web chats, it every, every, every discussion whenever you hit enter, it basically copies and paste the entire chat between you and the robot so far into the top of your question. [30:57] Dave Pengelley: So read all of it and then read your last question and then make the decision on how to respond to you. So it doesn't have memory. These things are stateless, they don't exist, but every time you press send, it grabs that, grabs whatever context of things is necessary and sends that through to the model. [31:12] Dave Pengelley: The model processes it and responds with an answer. So in a platform like chat, GPT, it's gonna have all the chat GPT guardrails around, you know, ethical use and all the things that that says you can and can't do. apparently you can use it for creating autonomous weapons. we, we can dive into that a bit later if you want. [31:30] Dave Pengelley: but it's got its guardrails and ethical boundaries and then it's got its other rules and then it decides from all of your previous chats what context it should pull in and apply that plus all your current chat thread and then your question. Whereas if you just raw, raw dogg it straight to the model and say, here's my question, it'll answer that question. [31:48] Dave Pengelley: Now if you give it a follow up question, it doesn't know the previous answer. It doesn't know the previous question because it is completely stateless. It just takes every prompt in itself. So if you're building your own agent, you have to give it memory where it'll repost the previous X number of messages, for example, back in with your new comment. [32:05] Dave Pengelley: That way it can reference the previous chat history as well. [32:09] Richard Webbe: David, you did, you did warn me about that. And I, I know, you know, I'm editing a manuscript, a quite large one and it all fell apart as, as you described for me, I told and 'cause I'm flipping between iCloud and Dropbox and all of that because sometimes my phone will recognize it and won't, it just became a bloody mess with my manuscript spread over many different files. [32:30] Richard Webbe: And I'm gonna, I did actually find the one you told me to export so I didn't lose it 'cause it, it was a lot of work in there. So I've got that saved. So I'm gonna get the new fresh manuscript. And the new fresh writings and then drop them back in and say, can you please help me edit, edit this, and hopefully it'll absorb it all. [32:48] Dave Pengelley: We, we need to talk offline. I'm gonna set you up with something better than just using chat GPT for, for editing your manuscript. [32:54] Marno Brits: I do wanna just add for, for stats, Matt was on point. you have the free version of chat, GBT, you get 8,000 tokens. For the context window. If you have the plus version, it's 32,000. [33:08] Marno Brits: And if you are using the direct open AI API, then you have 400,000. So, wow. Massive response. Even the response that you can get, you get a eight K response. So 8,000 token response, that it's allowed to respond in like the response size where if you get the API response, it's 128,000. So yeah, more than 10 x. [33:31] Marno Brits: just [33:31] Richard Webbe: define for me again, I know you said this last week, what's a token? Is that one conversation or? [33:37] Marno Brits: No. So Matt probably knows this more about me, but more than me. But the token is, like, not necessarily per word, but it's how they, like, how big the conversation is or how much you're asking of it to use. [33:50] Marno Brits: And that fluctuates. I don't think there's a real per word type of equation we can do there. The last Correct. [33:57] Dave Pengelley: Okay, thanks Mana. Yeah. You wanna, do you wanna weigh in on tokenization, Matt? [34:02] Matt Slager: No, there's no point. It's just technical nonsense. It's fragments. It's fragments of text, that's all it is. [34:08] Marno Brits: Yeah. [34:08] Matt Slager: So if you have a hundred words of just generic ps, you probably have around about. [34:16] Matt Slager: Like 150 or so tokens. So, well, easiest way to do that is the way around. So a hundred tokens is about 75 words of, pros. Yeah. [34:27] Richard Webbe: Like metric and imperial again, [34:29] Matt Slager: kind of, yeah, just fragment. Like there, there's token, analyzers. You can go to websites and you can put in a bunch of text and it will literally tell you what the tokens are and it will show you in highlighted colors so that you can actually see and have a bit more of an understanding. [34:43] Matt Slager: 'cause what a lot of people don't realize when they're building AI systems. Is the amount of just rubbish context that you're sending to and from, like Dave mentioned before, an API call to an LLM even before you even look at an agent, just an LLM thread, you know, just a series of calls in a row is, like you said, it's just your entire previous chat history shoved all the way in. [35:06] Marno Brits: Mm-hmm. [35:07] Matt Slager: Every single time. And at a certain point that has to get curbed, right. A certain point. Those are gonna just drop off. So to Joe's question about memory, it's what, what do you want the thing to not drop off? [35:21] Dave Pengelley: Mm-hmm. [35:21] Matt Slager: You know, which things are actually shoved into context every single time or which things are looked up specifically. [35:28] Matt Slager: You know, which things do I have to go find to make sure I have this in context before responding? So that's where an actual agentic retrieval system is built. And I think I made my mentioned in previous episodes that. Most common retrieval stuff you see is rag retrieval or gen augmented generation, which usually uses embeddings and similarity search, but it's not always accurate. [35:55] Matt Slager: And then you always gotta like ground that with a keyword search as well. So proper retrieval then with angen search on top of that, is this actually relevant to the request? And all of that happens in the background, so you [36:07] Richard Webbe: can't Okay. Ignorant technical question around what you're saying, Matt, 'cause I'm getting a good understanding here and I really appreciate it. [36:14] Richard Webbe: if I throw Sovereign AI into that mix, what does that mean? [36:19] Matt Slager: It means that you own all of those inference steps, nobody else is watching it. [36:26] Richard Webbe: Right. But does that change the call, the overhead, the capacity, using it in a sovereign AI matter? Excuse me. Yeah. [36:36] Matt Slager: Bless, bless you. [36:37] Richard Webbe: Thank you. So [36:38] Matt Slager: yeah, absolutely. Your, your power usage, your power. [36:42] Matt Slager: That you will have to spend your power bill to run that model is going to be, at the beginning, it's gonna be much less efficient than just doing the API call directly. [36:54] Richard Webbe: Yeah. [36:54] Matt Slager: But the economy of scale is when you are running hundreds, thousands of requests every second, all of a sudden your power bill is negligible compared to what your API costs would be. [37:05] Matt Slager: So yeah, from a running cost perspective, your own inference at scale is the reason why people do it. [37:14] Richard Webbe: So what you're telling me, let's say I'm a big company and I've got Snowflake from a data, and I've got, I've site, you know, I've bought a couple of LLM choices in architecture and I've got some key applications running over here and a bunch of recalcitrant users over here that, that there's a real important scaling architecture. [37:32] Richard Webbe: Strategy that you have to come up with before you start to just plug away on these things? [37:37] Matt Slager: A hundred percent. Yeah. If you look at, if you just go to, the status page for, for Claude, so for Claude ai, the Claude API Claude code, all of their various products, you'll see that there's, they're up time graphs don't look great. [37:52] Matt Slager: And that's not because they're a terrible company that are really bad at inference infrastructure, it's that they're scaling and expanding constantly. So at every little plateau, there's gonna be issues that you potentially haven't catered for. [38:07] Richard Webbe: Wow. [38:08] Matt Slager: Yeah, there you go. There's a direct article and well [38:10] Richard Webbe: done, Dave. [38:13] Matt Slager: Yeah, it's nuts. So yeah, if you're, if you're a company with 10 people, you probably don't have to worry about it. But if you're a global company and you've got millions, then yeah, you're gonna have to consider it. Especially, it's [38:23] Richard Webbe: amazing. I was doing a speed test on my phone on the weekend between my wifi and my 4G, 5G and all that sort of stuff. [38:31] Richard Webbe: And one of the tests I had in there was Claude. I'd never seen it before, but Claude is one of the, the yardsticks, return calls speed. [38:39] Matt Slager: Interesting. [38:40] Richard Webbe: Mm-hmm. I was, fascinat had all the different apps, you know, used to be our outlook, Microsoft, you know, general, and now Claude was, quite high up on that list. [38:49] Richard Webbe: I was, I was quite surprised. [38:50] Matt Slager: I think it's because, if you look at market share and actual sort of, you know, if you just compare open AI to philanthropic. OpenAI gets a lot of their revenue directly through chat chip t subscriptions, whereas Anthropic gets a lot of their revenue directly through Enterprise API volume. [39:08] Matt Slager: And so for them to actually offer more things and more features to the enterprise community, if you wanna use that word. [39:16] Richard Webbe: Yeah. [39:17] Matt Slager: you, you'll see a lot more businesses like employing people to train their team how to use. Claude Cowork. [39:23] Richard Webbe: Yeah. [39:23] Matt Slager: You know, that it is just gonna be a thing. 'cause it's the approachable version of a, a really good agent assistant. [39:30] Richard Webbe: Mm-hmm. Fabulous. [39:33] Dave Pengelley: Yeah. It's interesting that, everyone's all of a sudden dropping chat. GPT, like it's hot over the White House stuff. I mean, this is, I'm sure there's, there's some merit to some people doing it. I think there's also gonna be the bunch of people on the bandwagon. Like any of these, sort of sociopolitical causes where like, I'm gonna signal my virtue by making this choice. [39:53] Dave Pengelley: and so each their own, I'm not judging that. If people want to switch providers, I use lots of providers. I, Claude, Claude Anthropic is one of the ones I haven't actually spent money on. 'cause I'm like, I'm already spending money there getting some value. I'm spending some money over there. I don't need to spend money everywhere. [40:07] Dave Pengelley: At this point in time, I nearly, nearly bought Claude. Not because of any, white House business, but just because I. Needed extra code capacity. And then finally my, my Gemini reset and I'm like, yes, it's working. [40:19] Richard Webbe: And we're back in the, we're back in the TV streaming conundrum. [40:25] Dave Pengelley: Right? Yeah. Which, which, how many subscriptions do you need to which, to which platforms and why? [40:28] Dave Pengelley: Right, [40:28] Richard Webbe: exactly. [40:29] Dave Pengelley: I mean, the difference is the TV one, like it's what channels and what things have they got with the AI LLMs, they're all converging as far as they can all kind of do the same job. Some are gonna do a bit better, some gonna do a bit more efficiently, some are gonna ask you better questions. [40:42] Dave Pengelley: but it's the amount of usage you get. And so it's almost, but like the old dial up days where I'm gonna pay for, I'm gonna pay for 90 minutes of internet a week, and that's all my internet I could use. And then mm-hmm. I remember the name of them. I think my dad found one called Dingo Blue, and it was one of the first that had like a fixed rate, unlimited plan, unlimited minutes. [41:03] Dave Pengelley: It was unheard of to the point that my dad. Went and bought an extra phone line and we had a dedicated phone line for our dial up internet. It's like 56 KBPS. This is like, like mid late nineties. Pre A DSL. He put in a dedicated phone line because we had this unlimited internet plan. So before it was before nearly anyone had this, I had 24 7 internet in my house as a teenager. [41:26] Dave Pengelley: It was just insane. That's [41:28] Marno Brits: wild. Why are you not a millionaire? [41:31] Matt Slager: Like the idea of getting dial up at 56 K is salivating worthy. I used to download things at four kilobytes a second. [41:39] Richard Webbe: I was at a tele being, having a telecommunications background. I was at a, I forget where I was. I was at a, might have been aug or one of those, you know, user group events. [41:47] Richard Webbe: And I'd just given a talk there and I was walking around and, someone said to me, oh, Richard, You've gotta check out some of these kids over here. And there's, these kids are like 15, 16 walking around, like they own the place. And I said, what's he doing? He said, well, they're all aggregating the internet and they've all had ID Inlines put in. [42:08] Richard Webbe: And he said, that guy over there had, they've run out of IDN capacity at the exchange to put another ITN line into his house so that he can sell high speed 24 hour internet access to his ca. So a lot of these early internet service providers would run outta houses. [42:23] Dave Pengelley: Yeah. [42:24] Richard Webbe: With a couple of servers and a bunch of ISD inlines. [42:27] Richard Webbe: Wonder how long before we get that with ai? [42:29] Dave Pengelley: I, I remember when I met with brand new and it was very similar to that. Like, like I, I inet the, the huge mega one. I remember going into like, there were like one floor in office and they had a one rack with some modems on it. Like it was like completely different back in the day. [42:42] Dave Pengelley: But I mean, Matt just, you know, it's not 56 kilobytes, it was K BPS kilobits per second. You weren't pulling down at 56 kilobytes. You were maxing out at like 4, 5, 5 kilobytes. [42:52] Matt Slager: Yeah. [42:52] Dave Pengelley: By the time we do the math. But, yeah, I'm, I'm not a millionaire mano because I, I had a, troubled youth where I downloaded video games instead. [43:01] Dave Pengelley: Sorry. Oh, good man. Good man. Yeah, I'm, I'm sorry for all the, the hardworking developers that wrote them, and I didn't, [43:06] Richard Webbe: Dave, none of us stayed 24 hours in our homes playing computer games when we were growing up. Was none, not [43:13] Dave Pengelley: nothing. No, definitely. No, no. I, I just mean I wasn't paying the developers for their work. [43:16] Dave Pengelley: I don't feel bad about that. [43:18] Marno Brits: I'm curious. I like that idea a lot, Richard, 'cause I've been playing with that as a bloke. I met last week, week before last, and he set up his own local LLM just for him and the family to use. And he training his kids. His kids are homeschooled. Yeah. Teaching his kids. How do you interact with the terminal and how do you communicate with the future, essentially? [43:36] Marno Brits: Perfect. but I like that thought exercise of like how long until we have someone that's in town that runs a local LLM. And rather than you paying a large corporation in America somewhere, you instead pay Joe down the road and. They'll just host it for you. [43:51] Richard Webbe: Well, just letting you know, [[Syllogism|Syllogism]] of AI will be releasing their new LLM in the next No, just kidding. [43:58] Richard Webbe: Could be something to aspire to. [44:00] Dave Pengelley: Yeah. I think, I think as, as businesses and I've been working on, on my own Agentic harness, which, you know, watch the space. I, Am currently iterating version two of it. 'cause version one, like most early pilots went down in a ball of flames. because you engineer it down a particular path and you think it's the right path, and then all of a sudden it's just fighting with itself and doesn't work. [44:24] Dave Pengelley: the summary that I got from my early testing, the Gemini said that it was focused more on, bureaucratic ceremony than actually getting the job done. So it started kept telling you about how much it was doing the job and how it was gonna get approval for this, and it was gonna do that. 'cause I was trying to build this governance hierarchy and, and I was like, great. [44:41] Dave Pengelley: I was trying to build a productivity engine. I built the Australian government instead. so man, I'm, I'm working on, on version two, but I, I'm imagining the future of these kinds of tools and if, if, if someone wanted to run, like, you know, you've got Telstra with what, 50,000 employees across Australia. If they wanna give every one of their employees an LLM, what if they wanna give the multiple LMS. [45:02] Dave Pengelley: That's a lot of tokens that's gonna rack up very, very quickly. Mm-hmm. At what point did someone like Telstra go? I need to host my own LLMs on my own Telstra server hardware? Because the cost of paying open AI or philanthropic directly is, is just brutal. And [45:18] Richard Webbe: it's gonna be on, it's gonna be on water power networking. [45:21] Dave Pengelley: Yeah. [45:22] Richard Webbe: And networking would really solve that. That's getting up to enough speed with, you know, fiber to the home and all that stuff. And solutions. what's the story on power for AI in Australia? Do you guys have any view on that? [45:35] Marno Brits: Matt, a bloke who lived in Kalgoorlie, he was a heavy duty electrician, so he was the one that was working on the power lines. [45:41] Marno Brits: He got his license to my understanding is like, like jump from a, a helicopter or be connected to a helicopter to work on high heights, on electrician stuff. And before he went to New Zealand to just explore your snowboarding career, I was like, what's your thoughts on AI and like, power? And he was like, bro. [45:59] Marno Brits: From his perspective to be crude is like, we're fucked. Like we don't have any infrastructure. We are running at capacity. Yeah. I mean at the time Kali lost power twice that year for about a week at a time. and it is like, we don't know what the hell we're doing. No one's prioritizing it. We're just, we keep shoving it. [46:17] Marno Brits: We keep putting on a bandaid fix. and we're not prepared that that's what his conclusion was basically. [46:22] Richard Webbe: And when you see all those private companies like Amazon and Microsoft and all of that, building their own nuclear power stations [46:29] Marno Brits: mm-hmm. [46:29] Richard Webbe: I would've thought a private company go back to building their own power stations. [46:32] Richard Webbe: But that's what they're doing. [46:34] Matt Slager: Yeah. Well I actually saw a really awesome, interview that was with Elon and a couple of journalist guys that didn't really know what they were talking about. and they got Elon talking about power and about power usage and demand. And you know, the idea of scaling data centers means you need to scale the actual power source as well. [46:55] Matt Slager: You can't just plug more into the power board, you know that that power board can only take so much before it burns up. So. Yeah, the idea of actually running the physical cables from a literal power plant or power source of some sort, is something that people don't really even think about. And then the material it takes to make that cable, and then where all that ends up coming from in the first place. [47:18] Matt Slager: Like you just mentioned, nuclear. do we do nuclear? Do you do solar? You do coal as Joe's just mentioned. Yeah, [47:24] Richard Webbe: yeah, yeah. Identical ana there for the free metal doctor to send a smoke into Perth. But anyway, [47:29] Matt Slager: yeah. So what, what Elon actually said, which I, I think this is incredible. Like, you know, say what you want about him as a, as a person. [47:36] Matt Slager: But, the idea is that we won't have capacity or the ability to continue to build data centers on the earth. It's just. Not gonna be a thing. [47:47] Richard Webbe: That's right. He's talking about space data centers, wasn't he? [47:49] Dave Pengelley: Well, he, he wants to go to Mars. I saw someone interesting talking. [47:52] Matt Slager: No, this is not Mars. This is literally putting data centers into orbit. [47:56] Matt Slager: Into orbit. [47:57] Dave Pengelley: No, but, but, but from, from Elon's point of view, I saw someone breaking down and saying, if you think about Elon from a systems point of view goes, why does he have these disparate businesses? Why is he doing electric cars here and solar panels there and Tesla batteries there and optimist robots here and, and people going, he's very clearly talked about humanity needing to leave Earth and he wants to go to Mars. [48:14] Dave Pengelley: And so he's building all these technologies and systems here on earth because you don't have atmosphere. And so robots can work there and be forward deployed. you've got no real estate issues there. So you can deploy as many solar panels as you like on Mars. You can, you need vehicles, so, so battery powered vehicles 'cause there's no fossil fuels on Mars. [48:33] Dave Pengelley: All these sorts of things that people are predict. Some people are saying that, you know, this is actually 40 hs and he is actually prototyping and building all these infrastructure pieces that he's gonna need. To deploy in space. So space data centers. I think that it all ties into that big systems program diagram of what he's trying to do. [48:49] Dave Pengelley: A [48:50] Matt Slager: hundred percent. Anyone who like rebuts, that is an idiot. Like, the, the it is that it is 40 chess, but it's not really even chess. It's just like a basic Sudoku. He's literally just, you know, lining up the puzzle pieces. Mm-hmm. It's the same as, Peter Steinberger with Open. Like, I feel like we need to stop talking about it every single episode. [49:13] Matt Slager: But, the point is there that if you look at his GitHub, every single one of his projects that he's put together and built. Has become a core component of open cloud and the reason why it exists. So you are right. You need to get all that technology done and the infrastructure in place in order to be able to use them elsewhere. [49:31] Richard Webbe: It's funny, you're you're right Matt. My it is funny people. So a classic example of what we're talking about historically and 'cause I'm old, I can talk about it, is the SM Morgan family, one of the richest families in Australia, actually one of the richest families in the world. And the, I met, a lot of the s Morgan family members over the years and I was very lucky to do some work with some of them and they created a family dynasty, right? [49:55] Richard Webbe: And and it's gone very well and very successful. And I think it was Vince s Morgan and apologies to any of the s Morgan family, if they're listening. He set up his, his. Small goods deli, when he came out from Hungary, I think many years ago in Sydney Road, Brunswick. And all he did was he was making these beautiful Hungarian sausages and people and going, gee, they taste all right. [50:15] Richard Webbe: We'll have some more of those, right? And we'll have some more of those. And then as he's doing it, he goes, gee, it's really expensive for me to get access to my meat. So we looked up the road and why am I paying the abattoir? So then he had set up his own abattoir, and then he set up his own abattoir out at some Somerville, I think. [50:32] Richard Webbe: And then he saw the packaging company next door and said, well, why am I doing the, why aren't I doing my own packaging? Why am I paying them? And Elon's doing the same thing, but as a urist, he is assembling all of those pieces to get his product to market and to get the objective exactly as you said. And he's just building them. [50:48] Richard Webbe: Yeah. And if you can get them to be autonomously profitable or at cost. Then there's the building block problem solved. Let's work under the next challenge and the next challenge. [50:58] Dave Pengelley: Yeah. That, that vertical integration. [50:59] Richard Webbe: And he's Yeah, exactly. And he's investment in quantum technology, of course. Because if we have a data center on Mars, what's the latency gonna be like? [51:08] Richard Webbe: So with superposition and quantum physics, we'd like the space time continuum to be zip. And if anyone read an article on the weekend just to get really excited, they believed, they photographed actual, the movement between, two particles in superposition, [51:24] Marno Brits: what the hell? [51:25] Richard Webbe: Yeah. And they, it looked like it was really cool. [51:28] Richard Webbe: If you look it up, you said it looked like that unity, symbol, you know, the, the yin and yang and the black and the white one, right. Was exactly like that. But they did find out that the change when they changed one, the change in the other one was not perfect. It wasn't instantaneous, but they suspect it was faster than the speed of light. [51:50] Dave Pengelley: Yeah. Crazy. [51:51] Richard Webbe: Wow. That's exciting, isn't it? So, mm-hmm. And who's, who's investing money in these things? Not [51:57] Dave Pengelley: Australia. [51:58] Richard Webbe: I get a headache thinking about it, right? But here's Elon, like you and Matt are saying, sitting over there going, okay, I am building a house. I need the bricks. Okay, I need a wall. But walls haven't been invented yet, so what would a wall look like? [52:11] Richard Webbe: I mean, we just need these people, [52:13] Dave Pengelley: right? Mm-hmm. And in a, in a, in Australia, we we're, we're got a problem. I, I don't know where these stats came from. so, so Philippe, I trust that you've given us good stats here, but, he's ridden 65% of Australian organizations plan to increase the a AI investment. But we are the world, we're the world's 13th largest economy would account for less than 2% of global AI investment. [52:33] Marno Brits: Yep. [52:33] Dave Pengelley: And that's, that's just the AI investment, like you said, that that probably doesn't take into account power infrastructure or other bits and pieces, which are, are huge gaps Australia, Both sides of the political are, this isn't a political comment on one side or the other, have dropped the ball over the last two decades. [52:49] Dave Pengelley: and we have lost sovereign capability in manufacturing the steel for the cables, let alone the power infrastructure, let alone any water infrastructure. When was the last time we built a dam? All of these things are actually critical gaps that successive governments of both persuasions have allowed to happen under their watch, which have now left us in 2026 with significant sovereign capability gaps that would be underpinnings of being able to take advantage and leverage into this new AI generation. [53:16] Dave Pengelley: And so we, we are well behind the eight ball before we even talk about AI in our ability to actually manage and do these things is crazy. [53:23] Richard Webbe: And we, the geopolitical situation, so I'll show up in a sec. Mana, they have today is the 50th birthday I think of the Australian Made in Australia symbol. And that means, you know, and everyone was talking about, oh, it's better to help local. [53:36] Richard Webbe: I, I say one thing about by Australia, we could get isolated, not just for fuel, but as we know, toilet paper, everyone worries about that. And so if we're not doing it ourselves [53:46] Marno Brits: mm-hmm. [53:46] Richard Webbe: Right. We had to in the old days 'cause no one came out here. [53:49] Dave Pengelley: Yeah. [53:50] Richard Webbe: Isolation is innovation. sorry man. You're probably gonna say the same. [53:54] Marno Brits: No, it was, it was all Matt. Go for it. [53:56] Richard Webbe: I've [53:56] Matt Slager: got three really fast points and I think that these are, these are cool. They just really have heavy heating things about that space stuff. Again, the cool thing about data centers in space literally is there's no cooling [54:09] Dave Pengelley: because they're cold. They live in zero degree icy air. [54:12] Dave Pengelley: Yep. [54:12] Matt Slager: Yep. And then the second coolest thing, which is super weird, and you wouldn't really expect this at first, but it's always sunny in space. [54:20] Richard Webbe: Yes, it is. [54:21] Matt Slager: So you can have permanent, permanent solar feed with it, with no battery. You don't need any battery. So there's no battery replacement or anything like that. [54:29] Matt Slager: Mm-hmm. And then the third crazy thing is a supply chain thing with regards to building more power on earth, apparently, literally, in order to get turbines, like if you were going to go and construct and build your own power station today, you would need to find a source to purchase your turbines from, right? [54:49] Matt Slager: Mm-hmm. Well, that person, that company that assembles them, they get their turbine veins forged from a specific individual. And that individual is like flat chat, you know, fully booked out. So you literally cannot get veins to manufacture turbines to install in your power plant. [55:07] Dave Pengelley: Yeah, [55:08] Matt Slager: like, it's like the Rams Roy, like, you know, it's crazy. [55:11] Dave Pengelley: I mean, this was one of the arguments when, It was discussed, nuclear in Australia and you know, the plan to have it by 2035 or 2045. It's like we needed to book it 10 years ago. Like, but I mean, that doesn't mean you go, oh, well it's too late. We never do it. Like we've gotta, you gotta pull the trigger. [55:28] Dave Pengelley: The best, the best time to make a decision was yesterday, the next best time is today. Mm-hmm. For any of these things. But I don't know if our nation has the people in Canberra, like I said, on any side that are gonna make the decisions we need to make for this sort of stuff. So what can we do? We can just use the resources we have, the AI tools we have to differentiate and, and try and outcompete and do what we can. [55:50] Dave Pengelley: yeah, [55:51] Richard Webbe: because those concern is that the Joe's concerned that the nuclear power station circling the planet will malfunction and enter the atmosphere and destroy us all. [56:02] Dave Pengelley: That's the, the data centers. We're not putting nuclear in the, in the atmosphere. Solar, Matt just said it's all solar up up there. [56:08] Richard Webbe: Yeah. [56:08] Dave Pengelley: Solar's great in space. Just not on our farmland. [56:10] Richard Webbe: Yeah, yeah. [56:11] Dave Pengelley: Not to [56:12] Richard Webbe: now. it's interesting, isn't it? So he is gotta solve the problem of speed, data speed, but it's not that big a problem from the data center up and down. And if you're looking at a casing situation, or an aggregating situation, you could easily solve those problems with low orbit and high orbit and solar and stuff like that. [56:31] Richard Webbe: I love the way Elon thinks without restrictions getting in the way of his brain, we should all think that [56:39] Marno Brits: his measurement is like if, if, if it's, if physics says it is possible, or if select doesn't say it's impossible, then it's possible. Like if anyone comes to say no, we can't do it. Like the great example that's public is the Twitter when you bought it and said, Hey, kill these warehouses like these database centers 'cause we don't need them. [56:58] Marno Brits: Then all the technicians said, engineer said, no, we need them. Like they're, if that dies, then Twitter dies. And he said, no, kill them. Like, and then they said no. So then he fired all of them on his way to his Christmas Eve celebration with the family. His brother suggested, let's just turn around. The plane goes to the data centers. [57:16] Marno Brits: He rocked up, got the security guard to give him the knife, and he just cut the cables and everything was fine. And his approach was just like, if physics doesn't say it's impossible, then he firmly believes there is a way. [57:28] Richard Webbe: Yeah. [57:28] Marno Brits: and that's like a more like plain way of putting it like just for the service. [57:31] Marno Brits: But that's his approach. And his mentality is like, it's all possible unless it. Physics says it's impossible. [57:38] Matt Slager: You know what, like the whole point of XAI existing in the first place is another one of these puzzle piece pieces along the track. Yeah. Which, the point of XAI is to develop new physics so that they can do the things that which saying, and they can transport fast enough and, and do all the things that they don't know how to do yet. [57:58] Matt Slager: Mm-hmm. Like, it's incredible. Hey, [58:01] Marno Brits: oh, I love it, man. I'm a massive fan of this and just, yeah. I'm excited for the future. The, the fact that what, what was the speculation is by the end of this year, AI is growing such rapid pace. It's gonna start discovering new technologies by the end of 2026. [58:15] Richard Webbe: So it'll solve the problems for us, knowing the parameters that we won't. [58:18] Richard Webbe: But [58:19] Dave Pengelley: it's giving it the right context and giving them the background information [58:23] Marno Brits: mm-hmm. [58:23] Dave Pengelley: To, to do that. which I mean, that's one of the things Open Claw is famous for. It's like that heartbeat mechanism where it gives itself jobs and it gives itself like goals. And then every time it iterates it thinks about those goals and, and can give itself new goals and the future. [58:37] Dave Pengelley: And it's that self iterating process. And as long as it's getting fresh insights and it's not gonna suffer from entropy and just sort of drill itself down to the lowest con, common denominator and die. [58:46] Marno Brits: Mm-hmm. [58:46] Dave Pengelley: that's, that's where these things are headed. I think it's just another power usage. The cost of doing them. [58:51] Dave Pengelley: Who's gonna run them? What information do you give them? They're as good as the human and the creativity that the human injects into them. To begin with. Sorry, sort [59:00] Matt Slager: of, but like to push back on that, that just creates a mirror. You're just getting back out what you put in there. I would, I would like you all and all of the listeners to look up a term latent space. [59:13] Richard Webbe: Latent space. [59:14] Matt Slager: Yeah. Latent space. And even the concept of entropy. 'cause entropy is where new things are created. [59:23] Richard Webbe: Yep. [59:24] Matt Slager: You know, you can't create something new without it not existing. [59:28] Richard Webbe: Yeah, yeah. Yes. All creativity comes from stress, conflict and problems. [59:34] Dave Pengelley: But the, the mirror thing, Matt, I'm not, I'm not saying that it's just gonna circle on the idea that's the whole point. [59:39] Dave Pengelley: 'cause entropy would be, be that where it just sort of like degrades itself because, you know, the third law of the thermodynamics, which I know well from the Muse song, around a system built on infinite expansion without new energy, it's unsustainable. That's the song. It's that concept of like, you program this thing to bring in external sources. [59:59] Dave Pengelley: 'cause it's connected to the web, it's connected to other things. So it's not just circling a mirror back and forth of what you gave it, it is looking up new things, coming up with new ideas, seeding that back in that, that, that's what I'm talking about. Not purely just Yeah. Circling into oblivion. [01:00:14] Richard Webbe: Hmm. [01:00:15] Dave Pengelley: Which is a trap, but I mean, you know, you're saying it dies and then what the Phoenix arises from it. Maybe [01:00:22] Richard Webbe: what happens, by the way, do we put our warehouses up in space? [01:00:27] Dave Pengelley: All just, just in time logistics. We're just gonna 3D print everything. [01:00:31] Richard Webbe: Okay. [01:00:32] Matt Slager: In dark factories. [01:00:33] Dave Pengelley: In dark factories, [01:00:35] Richard Webbe: zero gravity manufacturing has got a special, you know, special key, clever key to it. [01:00:41] Richard Webbe: It's, it's very good at perfect manufacturing and perfect engineering. So I'm imagining a 3D printer in zero gravity. Producing some complex things that we need and then to get 'em down here, they just drop 'em with a parachute or a drone. [01:00:56] Dave Pengelley: The the whole zero gravity is wild. And what that means for any of this manufacturing. [01:01:02] Dave Pengelley: You think about the micro nanometer chip sets. So they've gotta create an earth vacuum and stuff. When they start creating CPUs and silicon chips and wafers and, you know, the quantum computers, that might be an argument where they're building those in space, maybe. But it's just that, that's like, it's just a complex that makes your brain hurt thinking about it. [01:01:20] Dave Pengelley: It's pretty wild. That's [01:01:21] Matt Slager: very cool. [01:01:22] Dave Pengelley: I mean, have you ever read, Enders Game by Austin Scott Card? [01:01:27] Matt Slager: I've been picturing that a lot recently 'cause there was something to do with Claude. Responding to somebody not knowing that it was responsible for some of the recent conflict. So, and yeah, I was instantly picturing Ender's game. [01:01:39] Dave Pengelley: Yeah. Because, because, not, not, not, not a spoil the end, but just even the, like the battle simulations were in that 3D space and, and just trying to picture, I know they made a movie, but it's not the same. But when you try and work out, like they're battling in these 3D zero gravity zones with these blocks and they're moving around, it's just like, it's such a different environment to think about than anything. [01:01:57] Dave Pengelley: We're used to sort of living on a flat gravitational plane. [01:02:00] Marno Brits: Mm-hmm. Pure mess. [01:02:02] Dave Pengelley: Yeah. [01:02:02] Marno Brits: I'm still a bit stuck on the latent space. I feel like if, is that Matt Soft launching as YouTube? No, [01:02:10] Dave Pengelley: that's JJJ, Jeff Huntley is called his new, blog. Latent Thinking or something. I [01:02:15] Matt Slager: think. Latent Patterns. Latent, yeah. [01:02:17] Matt Slager: That's gonna be his course. Yeah. So anyone that actually wants to know what's important, definitely keep an eye on Jeff Huntley and latent pats for sure. [01:02:27] Dave Pengelley: Yeah, he's the Ralph Guy. For those that are unaware, we're the, we're the, our guys. We've talked a lot about different things. We didn't touch too heavily onto the, the White House stuff, but that's because we're not a super political channel. [01:02:39] Dave Pengelley: I mean, it sounds like we are sometimes when we talk about nuclear power things, but rest assured these problems exist no matter which side you vote for. and there's no perfect answers, but we do need leadership and people to make choices and start moving forward on these things continually just kicking the can down the road. [01:02:57] Richard Webbe: well, that's what Australia's done for a while, and that's dangerous. I agree with Matt when he said that. [01:03:02] Dave Pengelley: Yeah. So we, we, we need to. Hopefully see some change, for Australia to stay competitive. But Australia is a big place. And even if the whole nation has issues with com competition, you as an individual, as a small business, as a big business, as an enterprise leader, you can grab the reins and take the competitive advantage by leveraging into ai. [01:03:21] Dave Pengelley: So if you wanna know more about how to do that, reach out to any of us. Our links are in the description at the bottom. Find us on socials, find us on LinkedIn, reach out, drop a dm. We would love to have a chat around how AI can be leveraged to give you the competitive advantage, for more interesting chats, philosophical discussions around ai, hands-on ideas around the execution, and how all the engineering works. [01:03:45] Dave Pengelley: Make sure you keep following us here at AI operators and we'll see you next week at the same bad time, same bad channel. See you boys. Thank you all. Thank you, [01:03:54] Richard Webbe: Matt Mano, Dave, Joe, other people online. Thank you. [01:03:58] Dave Pengelley: Bye [01:03:59] Richard Webbe: bye.