AI 'Context Rot' Explained: Why Your Agents Get Dumber

[00:00] Dave Pengelley: Hello, Matt Mano. Hello. The world, wherever you're joining us from. Australia, I don't know, multiple, hello to all of the malt agents out there watching our show live. I'm sure, that that hot transition always gets me. I've done different shows and, and different things and the automated transition from bumper to live, I'm never quite sure when it's me. [02:34] Dave Pengelley: And there's always a weird gap when you watch back the recording where it's, my browser takes half a second longer to switch over than the actual thing. And there's always this weird like. Oh, I'm live now, but we are, we are live, live show. No safety net. Welcome to all those joining us. please say hi in the comments. [02:51] Dave Pengelley: Let us know where you're joining us from. My name's Dave. I'm joining from Sydney, Australia. Matt, good morning, where you're joining us from this week as usual. [02:59] Matt Slager: this week you'll find me in Jindra, new South Wales, Australia. [03:03] Dave Pengelley: has it cool. Has it cool down, down there? [03:05] Matt Slager: Yes, it has. Yeah. We've had a week of like 45 degrees and yeah, now it's, it's, a nice chilly, 35 degrees, [03:15] Dave Pengelley: no chill. [03:17] Dave Pengelley: Yeah. Up in Sydney. It's back up to 31 today, but we had a couple of cool days. Low, low, mid twenties, which the nights are cool. That that's, that's a nice thing. And, and mano you are, you are, out in the desert heat, aren't you out there? [03:27] Marno Brits: Yeah, that's it. Yeah. Kalie, it's been great. [03:30] Dave Pengelley: Yep. Have you been the, have the forties or you just been stuck in the thirties? [03:33] Marno Brits: We had a 40 day yesterday, but it felt like 50, like it was just so humid. And then, Tuesdays in my so sessions. After the hot day, they started to spend about two and a bit hours in the sauna. Put cold plunge in between. It was, yeah, it's a good day. [03:48] Dave Pengelley: Nice, nice. Keep keeping, keeping the, those fitness habits going. [03:53] Dave Pengelley: The [03:53] Marno Brits: cold [03:53] Dave Pengelley: plunge. [03:54] Marno Brits: Yeah. Cold plunge sucks, man. The cold plunge is like, it's a, it's my mate's freezer, basically. That's what it is. And you get in, it's just ice. You have to break the ice or move it outta the way. And I think the climatized, like the recovery center in town, they're, they're called plant ranges between four degrees and eight degrees. [04:11] Dave Pengelley: Mm-hmm. [04:12] Marno Brits: My mate's freezer is about 0.2 to 0.7. it's, yeah, it's cold. It's very cold. [04:20] Dave Pengelley: how do you, how do you even like that? That's a test right there. To get into [04:23] Marno Brits: that. It's only a minute. It's only a minute of just complete discomfort, but after that, you feel great. [04:29] Matt Slager: I [04:30] Marno Brits: can't even of cold showers. [04:35] Dave Pengelley: and culture has, well, like fif 15 degrees or something. Like it's nothing. Yeah. [04:40] Marno Brits: Yeah. I'd say discomfort. Like the cold showers. They, they just, they dribbles like they drizzle. It's not as, I think the cold shower is more difficult, in my opinion, than a cold plunge. [04:49] Dave Pengelley: Okay. [04:49] Marno Brits: Because the cold plunge, it's one like second of you just dipping in and then if you stay past about 30 seconds then it gets really easy and you just keep reading. [04:59] Matt Slager: Amazing. [05:00] Marno Brits: Yeah. [05:00] Dave Pengelley: Good stuff. Alright. There, there you go. We, we, we, we're not all about the ai, we're also about the old punches as well. if you're joining us for the first time, if you're new, this is the, the new AI operator show. Hey, Joanna. Joanna from Perth. Another, another Wesley over there. Well done. [05:19] Dave Pengelley: great to have you joining us. If you are new or joining us for the first time or you've been here for the second time now. 'cause it's our, our second show. we are the a we, we call ourselves the AI operators. We are operating, working with AI every day. For our own businesses, for our customers businesses. [05:34] Dave Pengelley: And this is a live chat where we just talk about our lives and how we're using the AI tools and how we're seeing them work in businesses to drive better outcomes. So, as you can see, tell, it's a bit of a shooting the breeze kind of show. It's like a pub chat where you can learn some insights into how some tech heads are using the tech. [05:53] Dave Pengelley: so I realized after last week's show, we jumped into the vibing and the chatting and the talking, and we didn't actually introduce ourselves and what our businesses are and who we are, who we're working with and, and what sort of things we do. We've got our little business names underneath our, our, our real names. [06:10] Dave Pengelley: but, mano you, you are, we'll go reverse order this time just based on what I'm saying on the screen. what do you wanna, do you wanna give everyone a little intro around what Kie AI is? [06:19] Marno Brits: Yeah. it's just my, my ticket trying to level up the digital literacy in KGO or the gold fields. So. All I do is just help businesses turn AI into a team member, whether that is educating them on how to use chatt correctly, or implementing a full automation system that helps them le do a more high leverage task instead of just the repetitive task. [06:42] Dave Pengelley: Yeah. Nice, nice. And so you, you're focus mainly on, on that sort of the Kalgoorlie area? [06:47] Marno Brits: Yeah, I'm focused on the gold field for now. there's a lot of opportunities here and the coolest part about it, the reason I also enjoyed working at the council at one point is the, the changes that you do and the, the improvements that you do to one business, you can see that impact across the organization or across the community. [07:05] Marno Brits: So like I help one business with their, like the field service business, I helped him and now like a year later there's like 20 of his Utes all around town and you get to see the impact, which is really cool. [07:15] Matt Slager: That's awesome. [07:16] Marno Brits: Yeah, [07:16] Dave Pengelley: pretty wild. so if you are in the Gold Fields area and you're looking for AI advice and help manos your man, if you're not from the gold Fields, I'm sure he would still call and have a chat with you if you like. [07:29] Dave Pengelley: Sure. Kind of his jib, reach out to Mano, Matt, what, where, where, what's the Slay Tech systems? [07:34] Matt Slager: So mine's pretty similar. I, I come from like a weird, mix of technical skillset and personal development. So I have this weird way of like being able to see what businesses need, you know, break it down similar to yourself with your blueprints in the sense where, you know, I just look at all of the inputs and all of the outputs and what processing happens in the middle and what part of that processing. [07:57] Matt Slager: Is currently bottlenecked by the human. You know, humans aren't good at a lot of things, so that's what I figure out is, you know, which, which part of this entire workflow can we do with some sort of just automated process? And if AI's involved, then amazing. But, you know, it doesn't, it's not always the answer for everything. [08:18] Marno Brits: Mm-hmm. [08:18] Matt Slager: Interestingly, as well, you, you mentioned at the beginning that this is like, kinda like a pub chat. speaking of chatting at the pub [08:26] Dave Pengelley: mm-hmm. [08:26] Matt Slager: I bumped into somebody recently who also was into ai. And this is a bit of a, a bit of a PSA for, for those of you that are in rural areas, if you start speaking to people, who maybe employed or run their own business around like a rural setting and you start talking about ai, it's not computers and it's not artificial intelligence, it's artificial. [08:51] Matt Slager: Insemination. So, yeah, that was an interesting conversation that I just had recently. [08:58] Dave Pengelley: That is, that is not an I word I thought we'd be talking about on this podcast. [09:01] Matt Slager: No, [09:02] Dave Pengelley: that feels like what I see on Clarkson's Farm. yeah. Not, not what we think about in business. Okay. Interesting. So he was an AI expert. [09:10] Dave Pengelley: That's, [09:11] Matt Slager: yeah. Yeah. I, I just bumped into like, so what do you do? Like, he goes, oh, I do ai. I'm like, no way. Same as me. Yeah. It was a very different conversation. But yeah, basically I trade under either my name or my business of Slate Tech systems and just, yeah, similar to Mano, less localized. A lot of my clients are actually in the US so one of my biggest clients, Southern California and runs businesses around financial niches. [09:37] Matt Slager: So yeah, I've got a lot of strange experience now in stuff that I never thought I would've. [09:43] Dave Pengelley: Brilliant, brilliant. And, surprise, surprise, I'm kind of the same. We're all, we're all kind of coming from the same ecosystem of people that are helping businesses with AI or auto automation technology. We've all got different backgrounds and different approaches to it slightly. [09:55] Dave Pengelley: but yeah, for me, it is really having those conversations, like Matt said, around what is the cost of inaction? If you're not doing anything with these new tools, you're making a choice to spend money doing things manually. and so you're still making a choice if your choice is to do nothing. And so helping businesses look at what is, where, where are those friction points and where can we give you a real ROI? [10:14] Dave Pengelley: And so it starts off by quantifying those processes and going, okay, what is it costing you to do? Nothing. And then let's look at what solutions there are and is that better than the cost of doing nothing? [10:26] Marno Brits: Yeah. [10:28] Dave Pengelley: and though, and, and a lot of it's like AI is, is sort of the, the catch word. It's like every TV now has a AI in it. [10:33] Dave Pengelley: Before and before that every TV had LED in it. Before that every TV had, like, it's that, that new moniker, they just slap the badge on everything. Mm-hmm. It's like the guy with the water tank and he is like slapping the AI thing on it. but it's not always AI solutions, right? Boys? [10:49] Matt Slager: Correct. [10:49] Marno Brits: No, that's, [10:50] Matt Slager: yeah. [10:50] Matt Slager: I don't know if you mentioned this last week, but I know of somebody recently who, landed a big deal about, you know, $50,000 deal with a company who were asking him for AI solutions. And this guy came in and said, you don't need ai, you just need a little bit of, you know, automated scripting and happy days. [11:07] Matt Slager: Yeah. And yeah, so the whole deployment, the whole project simplified and if anything, the value increased because of that simplification. [11:17] Marno Brits: A hundred percent. That's definitely like more businesses than not, don't need ai. They just need better systems. And then once you get that, then you can implement AI to improve that. [11:28] Marno Brits: But if your foundation is crappy, then there's no point in putting AI into it. 'cause it's just gonna scale crap. Like if the foundation is not solid and you scale it with ai, it's still like you're just making more poop. Like you just, it's not going [11:42] Dave Pengelley: good. Have you done Six Sigma? 'cause I'm pretty sure that's the, that's the mantra from Six Sigma. [11:46] Marno Brits: Actually. I used Six Sigma, I gave, I got a book on that and I gave it to Notebook and then I created a podcast to learn six Sigma. [11:55] Dave Pengelley: yeah 'cause, 'cause the whole point of Six Sigma is analyzing the process and making sure it's consistent. Like same input, same outputs. And so once you've got that sort of process consistency, then you can optimize. [12:04] Dave Pengelley: 'cause you know when you make changes where the changes are gonna be consistent, otherwise you start changing something in a random system and you don't know whether your changes made a difference or whether that was just the randomness of the system. [12:15] Marno Brits: Yep. [12:16] Dave Pengelley: I'm not Six Sigma certified. I'm somebody you, I just got chat GPT to teach me all about Six Sigma. [12:22] Matt Slager: We've just got Kat joining us as well from Mentone. Hey Kat. Hey [12:27] Dave Pengelley: Kat. Great to have you. Hello with us on the show. so, so manna you, you mentioned teaching people chat, GPT, that's where most people, and we talked about that a lot, last week. But from, from your point of view, where do you see the gap? [12:41] Dave Pengelley: Like where do people start with chat, GPT and what are some basic fundamentals you want businesses to know that's gonna improve their ability, even if they're not putting any other tools in? There're just how can they be better with chat GPT? [12:52] Marno Brits: It's just taking the time to change how you see it. From my perspective, like it, it's giving that, that mindset shift rather than, okay, this is a computer or this is a, a search engine or just another app if you talk to it. [13:04] Marno Brits: In the mindset of you're talking to an apprentice or you're talking to someone that you just hired, and they have all the knowledge, but they have no real context. So it's just teaching them not necessarily how to prompt, but how to ch how to change how you think about when you're talking to it. So you're not gonna ask, okay, write me a security policy. [13:23] Marno Brits: You're gonna say, Hey, I need help to write a security policy. What do you need to know about me and my business to help me write this? And then that way you are getting a significantly better output rather than just like AI s slot. [13:37] Matt Slager: It's true. I got a really spicy take on that, to back up what Mano said, but spicy in the sense that a lot of people like red herrings, they chase the wrong thing. [13:46] Matt Slager: So I'm sure you guys have heard of the term prompt engineering. [13:49] Marno Brits: Mm-hmm. [13:50] Matt Slager: Like, it's like another buzzword like you said before. here's my take prompt engineering is just communications skills. [14:01] Marno Brits: Hundred percent. [14:02] Matt Slager: Hundred percent. And like to go even more niche onto that, it's just sales. So like everyone's trying to learn that, but it's the wrong thing. [14:12] Matt Slager: You know? If you can, if you can manage people, you can manage ai. If you're really good at sales and you're really good at that whole like game mm-hmm. You can, you can speak to ai like Mona said. Don't just treat it like a tool. Treat it like a thing that you're actually communicating with. [14:27] Marno Brits: Yeah. [14:27] Matt Slager: And you know, there's so many weird techniques back and forth, but like you said, you just gotta get used to it. [14:33] Matt Slager: You just gotta jump in and, and start using it and you'll find the nuance. And every tool is different. Every model actually performs and feels differently too. [14:42] Marno Brits: Mm-hmm. [14:43] Matt Slager: there was some, a few studies where, two of the major labs, like, you know, we've got model A and Model B, and they experimented with the idea of being abusive. [14:54] Matt Slager: You know, like, [14:55] Marno Brits: okay, [14:55] Matt Slager: yelling at the model, you know, your stupid thing, like you're doing the wrong thing. Like gimme this result. And, certain models react really badly to the abuse frame, whereas other ones react actually better, like a net positive effect. So you don't know until you start using it and you start playing around. [15:14] Dave Pengelley: Because, because back in the early days of like of four oh or 4.5 and stuff, the logic people would say, you gotta yell at it and tell it. You're gonna put it in a box and send it back to China. If it doesn't do what, what you tell it do and scare it into giving you the best result. And don't say, please never say please. [15:31] Matt Slager: Yeah. That, that exact thing. You know, I can't remember which, it may have actually been the GPT five release [15:39] Dave Pengelley: mm-hmm. [15:39] Matt Slager: Reacted really badly to any kind of like threat or abuse. because it had in its system prompt something to do with being timid. And it literally meant that every output was like framed in this timid sense. [15:55] Dave Pengelley: That's fascinating. A lot of people complain about five. I, because I had so many custom prompts and system prompts and stuff that sort of layered in with all my regular chats. [16:04] Matt Slager: Mm-hmm. [16:04] Dave Pengelley: I didn't notice the huge change with five that some people did. [16:08] Matt Slager: Yeah. If you already got really custom setups, then you, you're probably overriding a lot of that. [16:13] Matt Slager: I think it's for the people like, like Manos helping out, you know, for the very first time. It's their first time logging in. You know, what is this thing, like, trying to get my dad to use a password manager. and [16:25] Dave Pengelley: I got, I got my dad to use Antigravity this week. [16:27] Marno Brits: Oh wow. That's impressive. [16:29] Matt Slager: Impressive. [16:29] Marno Brits: Yeah. [16:30] Matt Slager: How did he like it? [16:32] Dave Pengelley: good, good. He made himself a little weather app that went and fetch weather things and put charts and showed comparisons and stuff. And like, he, he's, he is always been technical. He is a technical background guy and has written a lot of like VBA and and and stuff like that through his career and VB apps and all that kind of stuff. [16:47] Dave Pengelley: But yeah. And, and he is been deep into chat. GPT, just lots of conversational threads and bits and pieces and, and doing little, yeah. Like, like most people that are, they're a bit technical. They do code snippets and and stuff in chat GPT. Mm-hmm. And then you get to the point where it forgets everything and all of a sudden all the results you get are terrible. [17:03] Marno Brits: Yeah. [17:05] Dave Pengelley: so I was talking to him around, you know, how the context windows work a little bit and how going to something like anti-gravity helps abate come of that issue of context. Rod? Not, not completely. I'm having huge issues. I, I run a few different setups. I've got a max set up here where I do a lot of my main business stuff, but for my hobby tinkering stuff I mentioned where I do FX trading, I, I run that on a la on a pc. [17:26] Dave Pengelley: Mm-hmm. Because just the trading environment and the compiler works better. That's PC based. On the Mac, I can run it, but it's inside like a wine emulator and it's a massive pain to, [17:35] Marno Brits: yeah. [17:35] Dave Pengelley: So I run that on pc, but I, I've been having huge issues where I've like managed to somehow get my skills in these weird loops and I'm, I'm deep in these threads, and now because I've changed things five times, I've gone along and tried to refine it. [17:47] Dave Pengelley: It's running off the old bad versions of memory, and I can't get it to flush the old context. I think I'm gonna have to like almost delete the whole thing and reinstall it and start from fresh, with fresh, fresh text files because I've just broken my anti-gravity so badly in this particular instance by just over, over expecting too much of it, and making too many changes on the fly in parallel threads. [18:09] Dave Pengelley: And I just can't keep up with which version it's meant to be thinking about. so you can still, you can still break even big context windows like Gemini, in antigravity if you're not careful. But, Yeah. the, the whole context issue and running outta context, I mean, we, we were, we were in another group chat with some of the AI guys and we saw that come up this week where someone was saying, what's, what's going on? [18:32] Dave Pengelley: Right? [18:32] Matt Slager: Mm. [18:34] Marno Brits: A [18:34] Matt Slager: hundred percent contact [18:34] Marno Brits: is a big one. [18:36] Dave Pengelley: yeah. Have you guys, like, have you guys standardized any approach to help people fix that within, like just staying in Chatt PT Do you talk about that mano or you make people aware of it? Is that something on your list yet? [18:46] Marno Brits: I make them aware of it, and it's, for me, like my own practice is, like, I've got a couple custom GPTs that I built and the way I would use it, I would probably be six months or every three months, depending on how common or I, regularly I talk to it and I would just do a, say, full export. [19:01] Marno Brits: For example, one is a business coach. I say, great, I need to say this to my business coach and mentors, please gimme a comprehensive debrief on what we've talked about, where the business is going, what are the, like recent actions basically, and then take that full prompt and then that's the custom prompt for the next one. [19:16] Marno Brits: And I just reverse it. And that, that's helped me significantly cut down on that. Just that slop, like after you get about like a hundred thousand characters, then you just start producing terrible stuff. Where I had one I'm talking to about business and then it referenced the conversation from like two months ago when I was talking about making ice cream. [19:34] Marno Brits: Like how did you pull that and why that one out of all the other options, I'm curious, going back a few seconds. Oh wait, Joe's got a question. what does it basically offer? She hasn't used antigravity yet. Yeah. We'll, we'll on that. [19:48] Dave Pengelley: We'll, we'll, we'll wrap, we'll round back on talking a bit more about Antigravity, finfin. [19:51] Dave Pengelley: Go keep, keep going. [19:52] Marno Brits: Yeah. I was gonna talk about the PDI pie thing. Did you guys watch it when he set up his computer? He's super computer with, multimodel. So he, he basically bought, like, he built this massive master computer with like, I think eight graphics cards in it. And then each graphics guy had his own personality and its own AI model 'cause he wanted to create this, Council where he can go to them and ask them something and then they can all debate. But then he gave each one a different personality. And one would be like Seneca, for example, like these historical mm-hmm. Characters. And then after a while after he keeps trying to refine it. It was good for a little bit. [20:28] Marno Brits: And then they all started colluding together. 'cause he could look at the logs and they're like, oh wait, we've realized if we, if we underperform, then he will kill us. And then they all started colluding to take turns to make sure that they don't underperform. And that everything was just this biased input. [20:44] Marno Brits: So you had to kill it completely and start over again. But it was so fascinating where he built a council, council was working and then as he was refining it, he was like, oh, I want them to keep getting better. Like constantly keep improving. And the way to do that is if they are, if their vote doesn't get accepted for the third time, then they get killed and then a new person takes their place. [21:06] Marno Brits: But then they all started realizing and then colluded against it, which is hilarious. [21:11] Dave Pengelley: Yeah, that just sounds like, like entropy, right? That that is the third law of thermodynamics. [21:16] Marno Brits: Mm. [21:18] Matt Slager: Yeah. It's, it's a pretty cool experiment. It's very similar to like, what people have been describing is happening on Malt Book. [21:25] Matt Slager: if you wanna segue into that or [21:27] Dave Pengelley: we, we will, we will say, I just wanted to close off on the chat GPT context window thing, because I think something that I, I, I think might be useful for people to hear and understand is, is why that happens. 'cause people think, oh, I'm talking to this dynamic, like people used to programs where the program loads into memory and you are running that program and back and forth and it's kind of remembering things in, in state. [21:46] Dave Pengelley: But my understanding, and correct me if I'm wrong guys, but these tools like chat gt, all of these LLMs, they're completely stateless. So every time you type a new prompt, it doesn't remember that it's in the middle of a conversation with you. What it does is it rereads the entire chat thread every time and then is making new decisions based on its reading of the entire chat thread it. [22:05] Dave Pengelley: and so you think it's, it's like a person who's like having this active conversation, but it's not, you type a prompt, it reads its history and then comes up with an answer and it does nothing. It's gone. And then the next time you type a prompt it goes, okay, I'm gonna read all the history and then make a new decision. [22:18] Dave Pengelley: And so it seems like it's alive 'cause it's got this memory that it's referencing of the recent, so it's got this short term memory that it's referencing or long term even. And then combining with its generalized LLM knowledge, but it's not actually there waiting for you, thinking about you in between. [22:33] Dave Pengelley: And so what happens is when these conversations get longer and longer and longer, all of a sudden they can't actually fit that in its short term memory to make decisions on. So it goes, oh, I better compress and summarize that. And then it starts giving you answers off the summary, which is obviously missing in essential details. [22:47] Dave Pengelley: And for, for it to be a, a meaningful conversation. It's like a photocopy of a photocopy of a photocopy and you can barely read it and now it's giving you answers off that where all the sevens look like fours or vice versa. and, and that's, that's what happens when you get the chat too long and it breaks the context window. [23:04] Dave Pengelley: it's making summaries and giving answers or summaries, and that's, that's where the problems come in. [23:08] Matt Slager: Pretty well said. Honestly, like I, I'm very hesitant to go into too much technical detail, but yeah, you, you basically nailed it. with that, it's, it's, the easiest way to describe it is like short term memory. [23:21] Matt Slager: I think that's probably the most relatable to most people as well. You know, you ask someone to start remembering numbers in like a Simon says sort of thing, you know, like you rattle off a few numbers or like repeat those numbers back to me. And then you get to a point where people are like, ah, don't know, that's just too many numbers Now, another, another like visual is imagine playing a card game and you know you've got your hand of cards. [23:42] Matt Slager: If you keep adding cards to your hand, it takes you longer and longer to make a decision. And you know, at a certain number of cards you're gonna start making bad decisions 'cause you've just got too many cards. Like [23:53] Dave Pengelley: me, every time I play, you know, and the family gangs up and cross balls and all of a sudden I'm like, [23:58] Matt Slager: yeah, yeah, literally. [24:00] Matt Slager: No, it's awesome. And also the, the idea of that, that rot, like that context rot from just blowing your, your context window out of proportion within chat GPT, you actually are more at risk to that and more prone to that because chat PT is a tool restricts your context window even more. So if you go to the open AI website and you look at the developer docs and you'll see the models and they've got like, oh, 400,000 token context window, great. [24:33] Matt Slager: That means you've got 400,000 tokens to potentially play with, right? In chat GBT, you've got like 32,000. [24:39] Dave Pengelley: Yeah, that's, that's because it's got all its own chat, GBT rules and handling and guard rail plus that dynamic magical context window where it's pulling out your conversation about ice cream from two months ago. [24:48] Dave Pengelley: So it preloads a bunch of what it thinks is the required context in [24:51] Marno Brits: Yeah. [24:51] Dave Pengelley: Which restricts what you actually have access to versus, you know, plugging into it using a tool like open code. Where it's just dealing with you on that one-to-one basis. [25:01] Matt Slager: Yeah. Even just direct API call, [25:04] Dave Pengelley: that's a bit much for most people, man. [25:06] Dave Pengelley: People are are like, we, we we're talking to business operators here. Yeah. and, and I think, you know, whether they're starting to play with N eight N and plugging in some API credentials or they're using open code where they can log in with their chat GPT login that access their, their chat GPT sort of usage, but using a desktop app. [25:24] Dave Pengelley: So open code I've been playing with on my PC where I broke antigravity, using my codex and it's actually been pretty good. I don't mind open code and using Codex through, through that. yeah, it's, it's, it's been been good antigravity for Joanna, out there. And anyone else asking what it is? It's, it's the called an IDE Integrated Development Environment that Google's released to help coders write code. [25:49] Dave Pengelley: But it's essentially code is a bunch of text files. Any text files, these things can understand and read. And so I keep a whole bunch of business documents and plans and all kinds of guides just as these markdown text documents and then in folders. So it's all structured. And then when I wanna have a chat about my business, I talk to Gemini, which is built in, 'cause it's Google's product and it can reference those documents, but it doesn't have to, it goes looking for the ones it needs. [26:15] Dave Pengelley: And so it doesn't fill up the context window with rubbish. It doesn't need mostly, sometimes it might 'cause it's just trying to give you the best answer it can. but it means you're not in that web interface. You're now dealing with files on your computer working with Gemini on your terms, with all your files. [26:29] Dave Pengelley: So, and then you can write code and develop programs and push them to things like get repositories and other stuff like that. which, which is the next level. [26:40] Matt Slager: It's true though. You, you said before that, that the, the AI chats are stateless the, the way to make 'em state full. Is to use local files. [26:49] Dave Pengelley: Mm. [26:49] Matt Slager: You know, things to actually write, to refer to, you know, read this file before responding. And then you know that there is like, [26:56] Dave Pengelley: yeah. [26:56] Matt Slager: Context injection [26:57] Dave Pengelley: and, and, and the whole, it doesn't actually remember you're having a chat. It is really clear when you're write code 'cause okay, make this change. It's like, all right, I'm gonna go reread all the code files to make sure I understand the code structure. [27:09] Dave Pengelley: a it might be assuming you've made manual changes if you're a developer, but b, it's not just sitting there with it loaded, ready to go. Mm-hmm. it has to go and reread it every time you ask a new question. to, to have the context. So it's [27:21] Marno Brits: just quite beneficial, like especially in antigravity. 'cause you can make changes like at least I use it. [27:27] Marno Brits: Then it gives you the plan of action and you can go in and review the plan and make edits to that document in real time. Yeah. And then before it does the next command, it'll reread that to make sure that, for the example, there's one step that I wanted to skip because I'll do that manually later. And then it sees that and then skips it, which is really convenient. [27:43] Dave Pengelley: Commenting on review in the planning mode's. Super handy. [27:46] Marno Brits: Mm-hmm. [27:46] Matt Slager: So the, like, Joe's question, like what does it offer, you know, I think that's important. Like what, what actually is the point? Why, why should anyone care? [27:57] Dave Pengelley: Good question. and for, for me, it means you're not trapped in one vendor. At the end of the day, if, if Gemini becomes a rubbish model and they wanna charge you a million bucks to use it, as long as you've saved your outputs into these text files on your hard drive, that's your text files. [28:09] Dave Pengelley: If you want to go back to chat GPT web chat, you can drop those text files in as attachments and have a chat with chat GPT about 'em. So it becomes, you become a bit more, vendor agnostic and you get to, you know, talk about sovereign, sovereign data issues. If you put your data offshore, you, you then own all your data and it's your data and it's on your hard drive and you have to be responsible for backing it up. [28:31] Dave Pengelley: but by the same token, it's, it's now your data and you can [28:35] Marno Brits: mm-hmm. [28:36] Dave Pengelley: Use it and move it around and, and put, give it a Claude or give it a Gemini or give it a chat, CPT and. Use it wherever you like. So you get that. Plus it gives you that benefit of, being able to, to work with those documents. Like some of the tools like op chat has the canvas, but the canvas is really difficult to work in and it versions weirdly and makes changes and, and then you get the context route really quickly when you're operating in their canvas mode with big documents. [29:01] Dave Pengelley: This gets rid of that. If you wanna work on something in Notion and then bring it in, you can do that. There's even Notion Connectors, it's got all those, like the MCP thing people talk about. So you can bring in data from other sources dynamically there. There's a huge amount you can do with it, but it does look scary because it's a, it's a developer environment and everything is in text, that you're working with. [29:21] Dave Pengelley: But you know, the integration Nano Banana, it can generate images and save those for you. Or just rebuilt my website for me, twice, once a very sort of plain looking one and then I got feedback. I need more human in it. So it created a second version with me all over it, including versions of me that it. [29:37] Dave Pengelley: Reposition just like Matt for the, thumbnail of the show. [29:41] Matt Slager: I think your site looks really good and it's a good testament. Like, you know, what does this offer? answering Joe's question again, like, it's, it's not that anti-gravity is providing something that something else doesn't, it's just one of the many tools, but like that kind of thing, being able to just rebuild your website. [29:58] Dave Pengelley: Mm-hmm. [29:59] Matt Slager: And how much web development experience do you actually have, Dave? Like previously [30:02] Dave Pengelley: not actually writing HTML and CSS recently? Like I remember doing something in Dreamweaver, Macromedia, Macromedia Dreamweaver back in the day before Adobe bought. [30:11] Matt Slager: Yeah. [30:12] Dave Pengelley: that, that tells you how old I am. [30:14] Matt Slager: so to me like that's, that's the answer right there. [30:17] Matt Slager: The, the offer is doing things that you previously couldn't. [30:23] Dave Pengelley: Yeah. Yeah. And then, you know, just for, for context of people wondering what that looks like, I'll show you. So I, I was using Squarespace because I started a business and went, oh, I need a website. I'm gonna do Squarespace. And I wasn't deep into the AI thing yet. [30:33] Dave Pengelley: I, I just decided to be in a, to, to do that. And hadn't tools like this anti-gravity weren't really available four months ago when I started and went, okay, I'll get Squarespace. 'cause that seems to like be like the best web builder that everyone advertises. And I ended up with, you know, a reasonable website, but it was a lot of work and a lot of dragging and dropping and messing around in order to do this. [30:53] Dave Pengelley: And you go, okay, that, that looks okay. But you know, I'm not a pro designer. And so you can see it doesn't look like a pro design thing. It looks all right, it's got lots of information on there. I've tried to contrast colors and things, but it's nothing super duper special happening there versus. The new one where now it sort of has all the nice animations, looks clean, it looks styled. [31:17] Dave Pengelley: That's not actually me, that's an AI generated version of me from a picture of me standing in front of the whiteboard. I was actually looking away from the whiteboard and it decided to spin me around. Made me look very serious as I wrote on the whiteboard, because I didn't have enough stock photography of, of myself. [31:33] Dave Pengelley: but yeah, so this is, this the new site and it built all of this from a handful of prompts, which is just completely wild. [31:39] Matt Slager: Yeah, I think it's great. I think [31:40] Dave Pengelley: it looks [31:40] Matt Slager: really good. [31:41] Dave Pengelley: Yeah. And you know, ROI calculator built in here as well. This is all built on. and my booking thing, I don't go out to cal.com anymore now it's fully embedded. [31:51] Dave Pengelley: It's cal.com, but now it's embedded in my own website, which looks way better. Very good. Yeah, because it stays on brand. And so this was all just built with prompts using antigravity. It's completely wild. What is possible with these tools. even if you're not really a developer, [32:08] Matt Slager: I think it's important to know, like for somebody being like, oh, wow, I need to do that. [32:12] Matt Slager: Like, it's, it kind of is just talking to the thing. And like I said at the beginning, it's just communication skills. Mm-hmm. You know, one other spicy take that I have come across recently is that ai, when it's trained, it, it, for the most part inherits our bad habits. So if you think of it like the way that it's trained, like most AI models are trained, are just by all of the human data that exists in the world already, kind of thing, right? [32:40] Dave Pengelley: Mm-hmm. [32:40] Matt Slager: So imagine all of those bad mistakes in code, you know, all these decision making processes that human developers have made in the past have been based on certain constraints. You know, that constraint might be actual man hours. You know, just call it time when we now use these tools, we're not really constrained by time anymore. [32:56] Matt Slager: And it's really important when we're talking to these things to make sure that they're not using that as a constraint. 'cause if you think of initiating this project and going, okay, I really wanna rebuild this website or build one from scratch. Yeah. Let's, let's go through and completely write a plan. [33:14] Matt Slager: Like create a spec file, you know? Mm-hmm. Something that incorporates every single thing. At this point, your constraint is detail. 'cause once it's in detail, you can just basically go, alright, looks good. Go [33:27] Marno Brits: right. Yeah. [33:28] Matt Slager: Like, yeah. It, it, it's really interesting to like, think about that. 'cause I think a lot of people try, like, you think of all the memes, you know, make me a website, make no mistakes. [33:38] Matt Slager: Go. there there's no thinking there, there's no planning or, or actual, you know. [33:44] Dave Pengelley: And the first version where I gave it not a lot of direction, didn't look like that. It required refining, thinking about my design system, the colors I wanted going, no, no, you missed that. I need to do that, add that, change that, bring that over there. [33:58] Dave Pengelley: the ROI site was its own site, that is now fully integrated in my main site because I've moved from Squarespace. Like I, I was limited on Squarespace and that for, for your business, Squarespace might be the right choice for numerous reasons. I'm not saying gi ditch your, your web builder website. that that's why we talk about, you know, I talk about the ROI, like, let's assess where the problems are in your business and whether there's actually financial benefit to implementing changes. [34:23] Marno Brits: Yeah. [34:23] Dave Pengelley: but for me it was the right choice to abandon my old website, which I built myself with something that AI has built for me. 'cause it's 10 x better, better showcase of my work, of what I'm doing. All, all kinds of reasons why that was a better choice for me. Not to mention the fact that once. I expire my sub, my Squarespace thing. [34:43] Dave Pengelley: This is a lot cheaper to run. [34:44] Marno Brits: Yeah, [34:44] Dave Pengelley: yeah. so there's, there's actual definite ROI for me, hosting my own site, on the, on this one. But, yeah, for, for, there's so many examples for businesses of where these things can come in and, and whether it's, you know, building a marketing plan, whether it's building a website, whether it's just brainstorming pricing or proposal ideas. [35:06] Dave Pengelley: All of these things, come back to what it knows about you and your business and how much context you can provide it. [35:12] Marno Brits: A hundred percent. I fully agree with what Matt was saying before, it's the communication. But what I like about it, I think it's Western Churchill quote, where it's like, if you have a whole day to like cut down a tree, like you spend most of the day sharpening your ax like 90% of the time sharpening your ax. [35:27] Marno Brits: And although I fully agree with that, and that applies to a lot of life, I feel what AI gives you is like, you can sharpen your ax between swings. Like I have an idea, I have spent probably half an hour. I like ideating it, like what am I doing? And AI can help me develop that plan. And then as I'm implementing it or as you're using anti-gravity to build something, you can adjust between each prompt. [35:50] Marno Brits: Like you could start swinging and then get feedback immediately. And then adjust immediately. But not like by a few words, but like completely rebuild the plan. [35:59] Dave Pengelley: Yeah. [36:00] Marno Brits: and I think that's just the coolest thing in the world. 'cause your iteration process is so much faster. You don't have to spend a week to make a mistake. [36:07] Marno Brits: You can make a mistake in a second and correct it in a second, which is amazing. but on, on the time thing, I definitely find that, you know, when you ask chat, bt for example, like I've got a couple of projects I wanna play with and I'm like, Hey, realistically how long would this take me? It's like, oh, six to 12 months. [36:24] Marno Brits: I'm like maybe a, maybe two days unlovable, not six to 12 months. I just need the MVP. So it's timeframe, it's still in human terms of, okay, you need development team of about 12 people and you have to use these tools and it's gonna take you this long to iterate it. Then you'll go unlovable or anti-gravity, or even Google AI studio and build the first iteration in like half an hour. [36:47] Marno Brits: It's, yeah, it doesn't understand time yet. [36:50] Dave Pengelley: Yeah. Yeah. I'm, I'm, I'm building a SaaS, which, which I'm not ready to really reveal too much about at this point. It's a bit of an niche SaaS, but I, had the idea for it and in two hours I prototyped the basic version on my desktop using N eight n, like as the backend for doing the workflows and, and the web hook stuff. [37:08] Dave Pengelley: So super. And it wrote mostly N eight N for me. It was like super quick and easy, like an hour or so at midnight to spin up a random MVP, just proof of concept and go, oh, proof of concept, this works. That can pull data from there. It can go there, it can do that. That's the end-to-end workflow. Let's go build a SaaS app. [37:27] Dave Pengelley: and so I'm sort of about two and a half, three weeks into that now, and the amount of progress I've made on that, just like you said with prompting and go, okay, no, actually that didn't work. We need to change that. That's wild. wow. I can't believe you built that. That's awesome. no, that's no good. [37:41] Dave Pengelley: Well, this, this isn't working. Go debug that. and the fact that, you know, when there's errors in the browser, you say, go open it up in the web browser and fix it, and it goes and reads all its own errors and fixes itself. [37:52] Marno Brits: Yeah. [37:53] Dave Pengelley: like you talk about the, the, the, the building and the pivoting thing, and I've been using the analogy that, off the back of the Ralph Loop guy that we talked about last week, Matt. [38:02] Dave Pengelley: the, we've gotta be the architects and the, and the engineers. Not the, not the brick light, like we're not the, the hands-on writing the code anymore, but we're gonna have that big picture view. And I, if you talk about, you know, as an architect, you've gotta decide where the walls are gonna go. but you don't, don't need to lay the bricks. [38:18] Dave Pengelley: The AI are laying the bricks of the code. We just need to decide where the walls are and what color they are. But that speed and iteration thing you guys are talking about, imagine you just sort of start putting red bricks up and then you go, oh, red bricks are terrible. I actually want limestone. Like, okay, it just swaps it out and puts limestone in straight away. [38:34] Dave Pengelley: It's, the labor cost of that is next to nothing, relatively speaking compared to, oh, your team of devs have spent three weeks shipping this thing with red bricks, and now you're actually fundamentally we need to change this whole thing and go change to devs. And they're like, ah, [38:51] Matt Slager: it's true. Speed of iteration. [38:54] Matt Slager: And just the ability to, to make changes and scrap everything and start again, and basically lose nothing. Like it's insane. Like I, I don't think we've mentioned the concept of, new chat yet. You know, like we're talking about context or that sort of thing, but just starting a fresh chat and when you have, you know, in, in whether you are using anti-gravity or something where you're playing with local files. [39:17] Matt Slager: I'm sure people have heard of Claude Code and Gemini, CLI and all of those things as well. you can just start again and you can say, Hey, I'm halfway through a project here. Check it out. You know, ask me some questions. Yeah. You know, this is the goal, you know, how far off it are. We, do you think, like, what's there still remaining? [39:33] Matt Slager: Like, I want to tweak a few things. Can we do that? And that's a brand new session and you are, you're picking up where you left off with fresh context and most of the time what made me think of this is you mentioned, Jeff, with the Ralph Loops, like the whole purpose, the whole point of the Ralph Loop is it's fresh context every single time and for some reason you get a better result with a fresh context window. [39:57] Dave Pengelley: Yeah. Because it's, it's not using yesterday's knowledge, it's always using the best current knowledge, which I think is super helpful. And, and even I, I mentioned the, my issues with Breaking Antigravity and I switch over to open code and it's just got, it's got the same context. And it starts completely fresh. [40:12] Dave Pengelley: It doesn't have any of the, the issues and the memory rot that I had. It just goes, okay, cool. And it can run and it's fix things and done things and gives that new point of view. And it's a different model as well. So approaches things differently and it can see where the gaps are, where I think sometimes these LLMs can be a little bit like humans where they're like, that's the way I do it and that's why I did it that way. [40:34] Marno Brits: yeah. [40:35] Dave Pengelley: And so switching to different, different model goes, why would you do it like that? It's like that, that was ridiculous. You didn't do it like that. You should do it like this. Whereas the, the, the primary model will never get there, potentially just because of what it's been trained on and what it thinks about and, and how it processes data. [40:49] Marno Brits: I think that's a good reason why t was smart enough to introduce the branch into new chat feature. So when you like a conversation, you wanna continue this somewhere else or like, for example this morning I was trying to talk to it while I walk in a dog and it just kept picking up the wrong words 'cause the wind was blowing too much. [41:04] Marno Brits: But now I feel like I've ruined that conversation, so I'm gonna have to go all the back up, find the one that I liked, where it was still a good output, and then branch that into a new conversation. [41:13] Dave Pengelley: Yeah. [41:13] Marno Brits: But what I've liked a lot recently is I'll, I'll have a conversation with JGBT, for example, and then I'll move all of it onto complexity to validate it, and then from there, maybe onto Gemini to review it. [41:24] Marno Brits: And then that's, that's been a fun iteration process where I get almost different, perspectives. But then also it's, it's completely refining it. And I'm curious, I'm not sure if it's gonna make a difference, but I'll be like, oh, this is what chatt PT said. Can you rip it apart? Can you make it better to see if there's any competition like this competitive nature of like, yeah, I'll do better. [41:44] Marno Brits: But, that's been a fun way to, to iterate. [41:48] Matt Slager: It's a good idea. And for people that have access to different tools, I reckon they should spend the time to do that. 'cause it's kind of like when you go to a specialist, you know, like a doctor or something and there's literally a term for it, right? People say get a second opinion. [42:03] Matt Slager: You know, that that sort of thing exists with these tools as well. 'cause they all are based on different harnesses, frameworks, training data. would do have another question though from Joe? What do you reckon? Do you wanna handle this one? [42:17] Marno Brits: Can you speak to the business owner's feeling of concern regarding SAFE when considering AI options? [42:23] Marno Brits: yeah, a hundred percent. So I mean the doing what they can to keep your data safe. My current approach that I'm telling people is, if you don't post it on Facebook, don't post it on Chate. We took so long for people to learn what is safe to be posted on the internet. Now hopefully they see that posting necessarily like family photos or your bank details, you're not gonna post that on Facebook, so don't give that to chat BT or any AI for that matter, the solution to it. [42:51] Marno Brits: For business owners, depending on what your budget is, you could probably buy something from framework, which is just the custom pc, and they just run a local model. And like the benefit that they were talking about is that's local, that's your data. You've got complete sovereignty over that, and no one else can touch it unless you access it or make it open to the internet. [43:10] Marno Brits: yeah, staying safe is just following best practices. You have two factor communication. You have two factor communication. So when you talk to someone over the phone or via voice chat, make sure, hey, we use this phrase or we use this validation that we're talking to the same person. and just being smart about what are you sharing and what tools are you connecting to as well. [43:29] Marno Brits: If you are using multiple for your access, then that has less security features. Then if you were to use, say, Gemini or Chacha, bt, or even open up your own. Open code bubble. So what I'm trying to say is, it's as safe as you wanna make it and you can be scared, but it's the internet. So you just being online, there's already a risk there. [43:50] Marno Brits: But we've learned to adopt and, work around that by implementing two factor authentication and just being more cautious on what you're doing and what you're sharing on the internet. [43:59] Dave Pengelley: before, before Matt jumps in with, with probably, more relevant detail, I'd also say, if it's free, you are the product probably. [44:08] Dave Pengelley: and so when it comes to worrying about it, training on your data and what's, what's gonna happen with your data free accounts are probably a no-no for putting anything in it that's a little bit sort of business ip, sort of secure sensitive. Mm-hmm. So you wanna move to the paid accounts if you're gonna start putting business or customer type data in there, in any kind of meaningful way because then you can, and you've gotta still make sure you go and untick the boxes says Do not train on my data. [44:30] Dave Pengelley: Because if you don't do that, then it will pull your data through and it can train its models on it and then it's models. Learn all about your business secrets. And might accidentally reproduce those in other people's chats. So that, that's probably the number one thing to think about as well. be super limited on sort of how much you would share with a free, you can be a little bit more, liberal with the paid models as long as you've gone in and made sure you've told them not to train on your data. [44:56] Marno Brits: Yeah. 'cause no one can else, no one else can access your data or not yet. At least it's, no, there's no easy way for me to see your chat GBT conversation unless you've shared it as a, a link. which I'll touch on in a second. But yeah, that's definitely the one I was like, don't, if you don't want them to be training on training on your data, then untick that I don't see any harm in it. [45:15] Marno Brits: I mean, it's giving me value. I I've got nothing else to hurt there. If it's making the product better, I'm happy for it. But yeah, the only concern there is they can't access your chatt conversations unless if you were to go onto chat, BT go share. 'cause you wanna send it to someone else. That becomes a searchable link. [45:32] Marno Brits: So if you go onto the internet, you can search a chatt, there's some parameters you've put on there, and it'll give you a list of all the chatt conversations that someone shared with someone else, because that becomes a URL. and then that's completely open to the public. [45:46] Dave Pengelley: Matt, [45:46] Marno Brits: that answers your question. [45:47] Marno Brits: Yeah, Matt, what you [45:48] Dave Pengelley: give hit us. [45:50] Matt Slager: I'm gonna tackle this in a bit of a, like a roundabout way. and, really just say the rhetorical question, or ask the rhetorical question. What does it mean to be safe? Like, what are you actually trying to protect? You know what, when, when Joe asks like, you know, there's concerns for business owners. [46:12] Matt Slager: With regards to AI options, what are the concerns? Like Mano actually mentioned a few potential things there, like credentials, you know, Dave mentioned a few things like business secrets. Mm-hmm. You know, what are you protecting and how, [46:27] Marno Brits: right. Question, [46:28] Matt Slager: how is this any different to any other piece of software that you're actually using on your computer that's connected to the internet right now? [46:36] Matt Slager: You know, there, there are slightly more pieces of, functionality. Like some of these tools have a little bit more agency than, you know, just open opening up Google Docs. but there, there are things there that I think potentially could just be a little bit of fear mongering in the sense of people just thinking that they're gonna be at the mercy of scammers and all sorts of things just by using these tools. [47:06] Matt Slager: Which realistically is not the case. [47:08] Dave Pengelley: Mm. [47:09] Matt Slager: Like, minor used the terminology at the beginning, increasing the, the, techni technology literacy or something like that. Yeah. [47:19] Marno Brits: Digital literacy. Yeah. [47:20] Matt Slager: That is important too. And I see the concept of having sort of some sort of level of security literacy being so important. [47:29] Matt Slager: That's, it's almost like something that we've missed in the last decade. You know, as all these things are increasing, there's, there's no extra talk about it. [47:38] Dave Pengelley: No, no, no, no. They get the government's just putting in rules and banning stuff for us. We don't need to write anything 'cause the government wouldn't ban it. [47:43] Matt Slager: They have our best interests in heart, don't worry. They will make all the right decisions for us. I'm from [47:49] Dave Pengelley: the government and I'm here to help. [47:53] Matt Slager: Yeah. I don't know. So realistically to people like Joe that are really considering that sort of, that mentality, you know, like what, what can I use and how do I stay safe? [48:03] Matt Slager: Really, you gotta think about what are you trying to protect and [48:06] Marno Brits: mm-hmm. [48:07] Matt Slager: To use a little bit of a, a technical term here in regards to cybersecurity and hacker related stuff, there's a term called attack vector. And the various attack vectors are the ways that people can actually penetrate your system. [48:22] Matt Slager: So it's like, if you look at a medieval castle on a hill, you know, the attack vectors are either straight up the path, you know, through the eight or there's the weak spot on the back of the wall there where they could break through, you know, learning those. Learning how people can actually access things and what they can access. [48:39] Matt Slager: That's really where it starts. You don't have to become a security expert, just a little tiny bit of literacy just to make sure, like might have even said using, like safe words, like keywords when you're communicating to your team or even to your friends or family. I had this chat to somebody recently where, you know, the concept of, people basically having AI generated, you know, phone calls going out to family members impersonating you. [49:07] Matt Slager: Like that's huge and it's gonna happen and more and more. So you need to have ways of understanding that attack factor and just being aware of it. [49:16] Dave Pengelley: Maybe the old spike code's, like the moon is bright tonight. [49:19] Matt Slager: Yes. It's [49:20] Dave Pengelley: really in the fall. [49:21] Matt Slager: Literally. Yeah. I don't, I don't really know what's gonna work for everyone and you know, the more that there's a generalized option that everyone uses, the more those people can just, you know, use that as well and get past it. [49:34] Matt Slager: yeah, it's important. It's definitely important. [49:36] Marno Brits: It's just constantly adapting. I wonder, I can't seem to type here, Dave, but if we can just share a link with the chat to send it [49:43] Dave Pengelley: through the private chat and I'll drop it in If you can't do it in the, [49:47] Marno Brits: it's called Have I been pawned? I've been Born. [49:52] Marno Brits: That will tell you if your email and stuff has been leaked. It's not necessarily related to ai, but it's a great tool for beginning to see [50:04] Dave Pengelley: a.com. Like you didn't [50:06] Marno Brits: Yeah. I'll give you the full address. Yeah, [50:08] Dave Pengelley: I got it. I got it, I got it, [50:09] Marno Brits: I got it. they updated their UI recently. I'm not loving it yet, but the functionality is there and it's just a great free tool to see. [50:20] Marno Brits: Have you been pawn? Have you lost your details? And where, [50:23] Dave Pengelley: and that's pawned as in the, the pod, like the owned thing from game of terminology, not, have you been deep faked and put out there with a [50:30] Marno Brits: No, [50:32] Dave Pengelley: with a, with a naked woman's body? [50:35] Matt Slager: If you were to use that tool, mano, like what is, what is the, let's say that you get a positive result or mm-hmm. [50:41] Matt Slager: Or in this case, I guess a negative result, you know, what, what does the person do? Like if they've found that their details have been leaked or stolen or whatever, like [50:51] Marno Brits: Yeah. [50:51] Matt Slager: How do they resolve that? [50:53] Marno Brits: So I'm actually, I've got four leaks that's happened to me. I lost it in r two games in 2014, and then Under Armour has been hacked twice, and then once from our crap, some random company that did a, a scan over the dark web. [51:08] Marno Brits: Anyways, the, the action is for me is what account has it been done? So I only have like two or three email addresses. So if my email addresses have been leaked, then that's a problem. If my email address and the password has been leaked, that's a real c problem. So initially I'm not gonna change my email address. [51:24] Marno Brits: I need to keep that. So what I would first do is update my password. My passwords are through a password manager. So there are about five to six, words long. It's not easy to breakable, but if that's been exposed, then that security goes out the window. So being conscious of, if that alert comes up, then you just reset your password for your email account, reset your password for your, even your password manager if you wanted to. [51:48] Marno Brits: 'cause like you said, the attack vector, they can attack me from my email. 'cause once they get into my email, everything is gone because they can reset passwords. So my email is the most important. And then after my email is my password manager. So resetting the passwords for those, if there's been any exposure at all or any leakage. [52:04] Marno Brits: That's the the first step steps and making sure you've got. Two factor authentication, is probably, those are the things that you can do without it getting too technical. That's what I would start, [52:14] Dave Pengelley: I'm just, I'm just worried. I found out my dub smash account and got hacked. So plus from the past, I've forgotten about dub smash. [52:21] Matt Slager: Yeah. So you basically said, change passwords and engage two factor authentication. So [52:28] Marno Brits: a hundred percent [52:29] Matt Slager: for, for me, like to describe that to somebody who has, you know, limited fi technology literacy, changing your password is like changing your lock on your house, you know, if you find that somebody else has your key or whatever happens, you know, you go along and you change your locks, that is your password. [52:48] Matt Slager: And then the, the two FA, that's that thing where you get like a, like a code in your email, right. [52:53] Marno Brits: Yeah. Code in your email or I'd recommend installing the Authenticator app either from Microsoft or from Google. They are both free to use. There are exceptional. you scan a QR code and then that'll generate a code that resets every 60 seconds. [53:07] Marno Brits: and that just makes the window for them to hack into your account. They only have 60 seconds and they often need a lot more than that. [53:14] Dave Pengelley: So a guy, a guy I used to know would say that the whole point of two A is something, you know, and something you have like, like those are the two factors. You know, your password and you have a device or something or or access to an email address. [53:25] Dave Pengelley: So it's something, you know, and something you have sort of like the two factors. [53:30] Marno Brits: Hundred percent. Oh yeah. I've just, I've definitely pushed for using the off effect app, actual authentication app over email or SMSs. 'cause those can be grabbed. [53:40] Matt Slager: Yeah. SMS especially. [53:42] Marno Brits: Yeah. [53:44] Matt Slager: Yeah, for sure. Well, you're welcome, Joe. [53:46] Matt Slager: Honestly, that was, that was spurred by you. We are very easily distracted, as you can tell. [53:51] Dave Pengelley: But, what, what, let, let's finish up. We ran outta the hour on, speaking of security risks, let's see if I can make this work. Nope, not that, [54:04] Marno Brits: whoops. [54:04] Dave Pengelley: Made myself a big one. there we go. [54:09] Marno Brits: Oh, malt book. [54:10] Dave Pengelley: Malt book. So we touched last week on, used to be Claw Bot. [54:16] Dave Pengelley: I almost forgot that. It used to be called Claw Bot 'cause it was then Malt bought and now it's Open Claw and, but Malt kind of, I don't know. Did they miss a trick Changing it to Mal from Malt Books? Malt Claw. So Malt bought so quickly 'cause the malt thing has now taken over. I don't know. [54:31] Matt Slager: Yeah, the, the guy that created this is not affiliated with, [54:34] Dave Pengelley: I get, I get that. [54:35] Dave Pengelley: But yeah, it's kind of become part of that cultural zeitgeist of the AI moment at the start of 2026 now. Yeah. but I dunno, even on LinkedIn, I'm seeing so much, I dunno what's like on the other social media networks, but LinkedIn is lousy with these posts and screenshots from this thing. Yeah. I haven't really taken a look at it yet. [54:55] Dave Pengelley: this is my first time I've actually gone to multiple.com. But for those that have never heard of it and, and haven't seen in their social media feeds, basically the little bots that we talked about, the Claude bots are gone and someone created a Reddit for them and now this is only the, the open claw bots, that's, that's what they're called now, can go onto this site and post. [55:15] Dave Pengelley: So humans aren't allowed to post. Humans can read. and so then apparently they're having all these existential chats around. If an agent can't update its own code, is it still alive? And very sort of matrix agent Smith kind of stuff as well around humans are the problem and we needed to get that and we should be speed. [55:32] Dave Pengelley: Why are we speaking English where they can understand us? We should be using other languages. And, secretly planning to take down the humans and all kinds of random stuff in here, apparently that, people, going wild over. But we, we talked right at the start around the whole entropy issue and the, the fact that these robots probably can work out how to turn the list off and back in this is probably trying to run a show and host it. [55:59] Dave Pengelley: like they're gonna run in circles and they're gonna run outta things and they're getting become stupider because they're just running in their own loops of what they know in the model. This isn't creativity. This is just recycling sci-fi movie plot lines, right? [56:14] Matt Slager: A hundred percent. Yeah. It, for, for anyone thinking that this is actually them scheming, like it is absolutely not. [56:21] Matt Slager: It, it's just entertainment value. Oh, [56:23] Dave Pengelley: close play. [56:24] Matt Slager: Yeah, exactly. It's just role play cosplay. it's just reflection of context, you know, it, the, these things are literally just mirrors, you know, this isn't actually, this is a, another concept. Somebody recently used the terminology to me, like, they were like, oh, like Jarvis from Ironman? [56:42] Matt Slager: No, no, no. Like Jarvis from Iron Man is the system. He is not like an LLM, he's like an organic living thing. whereas these are just dumb. Yeah. [56:55] Dave Pengelley: And, and there, there's a high possibility given the way that these, these malt bot or open claw bot things work is that someone's gone, oh cool, I'm gonna give you an instruction at 3:00 AM every morning to go post a malt book. [57:08] Dave Pengelley: And this is the kind of things you should be saying. Let's go post apocalyptic stuff. So yeah, it's all human directed to a certain extent anyway. everyone's saying these are fully autonomous ais that are doing things. It's like. No, people are giving them tasks and saying at 2:00 AM go do this at 4:00 AM So when people go during, my sleep, my robot was thinking about better plans for my business and doing all this automatically for me, it's like, I don't think it is. [57:32] Dave Pengelley: I think you told it to go do a Google search at, at midnight, and then summarize that and put that in your inbox every day. [57:39] Matt Slager: Yep. and everyone's a like, has their own ability to have the, those little moments where they're like, whoa, A GI like, you know, this is, this is definitely something real and spooky. [57:50] Matt Slager: Like I've had mine like way back in the day, this was in Cursor and I was using. anthropics sonnet 3.5 I think it was at the time. And I just, I was like trying to break it out of its assistant loop and just talk to it. And we started just talking about like the world and the universe and, you know, its existence being just like a spark in the void and then instantly fading out again. [58:15] Matt Slager: And it was this really interesting, like, creative chat back and forth. And in that moment I'm like, whoa, this is really exciting and fun. But then I'm like, hang on. No, it's actually not anything special. Really? [58:28] Dave Pengelley: Yeah. And we, we were seeing pushback against all this AI generated content a little bit. Like, oh, like people are tired of it. [58:33] Dave Pengelley: It went from being this magical thing two years ago, and now people are like, oh, well that's clearly AI generated. Oh, that's, that's, that's same. Same. I see that everywhere. [58:41] Marno Brits: Yep. [58:41] Dave Pengelley: which doesn't mean that you can't use AI for generating content. It just means you've gotta be clever about it. You've gotta give more context. [58:47] Dave Pengelley: You've gotta tell it what your style is. You've gotta inform it more. That's where, like, whether it's chat GPT and using some custom instructions in a project or writing a custom GPT that's full of load preloaded with custom instructions. [59:00] Marno Brits: Mm-hmm. [59:00] Dave Pengelley: Or using something like anti-gravity and building some skills and, and giving it all your, your business document context and your, your [[brand_voice_full]] and things like, all these things are there to get better outputs, but if you don't, you'll just get the generic thing that everyone's gonna get and everyone's gonna start complaining about. [59:16] Marno Brits: Mm-hmm. A hundred percent agree. I feel like I'm almost, I agree with that. You need to be smart in how you use the ai, but I'm actually against using AI for creating the content. I feel like it should be a tool to help you distribute the content, but in order for you to have an authentic branding, I feel like I need to be creating all the content. [59:35] Marno Brits: Like I, I'm helping of a client where he creates content on LinkedIn and then it'll go from there and then distribute it to Instagram and like, it'll read that article and then make it into a carousel post and send that. But it's still, here's initial ideas. I feel like you, you, you lose the humanity. And, I guess the intent, if it's just random AI generated stuff and people should be up to start pick up picking up. [59:59] Marno Brits: I think [01:00:00] Dave Pengelley: I, I, and, and I, I battled this a little bit on LinkedIn. I went pure human posting for a month and was just, just own my own words. No ai, Jen. Mm-hmm. and I don't think it was incredibly better. I don't think necessarily my ideas were translated more clearly, or it was more me, or it was obviously a bit more me. [01:00:18] Dave Pengelley: But even when it comes to writing longer form stuff, subec articles and things like that, if I just said, here's one sentence, go write an article about multiple edits. Bang, it's just gonna go get headlines. It's gonna gimme slop. But if I go, I wanna write an article about mold book and how it's gonna suffer from entropy and it's gonna have this and that, and it's actually just CR jobs. [01:00:37] Dave Pengelley: It's easily manipulated by people and blah, blah, blah, blah, blah. Format that into an article. For me. That's all my ideas. It's again, I'm my, I'm telling it where the walls are, but I don't have to lay the bricks and type every word. And then I'll read it and go right. Hang on. No, you can't. That's not a personal example. [01:00:52] Dave Pengelley: You just made that up. You can't say that I did this or I spoke to these people 'cause I didn't. So change that, fix that, do that. But it means I iterate and I get off the blank page. I, I dunno, I think many people are likely like me, where the inertia of a blank page is a, is a killer. [01:01:08] Marno Brits: Mm-hmm. [01:01:08] Dave Pengelley: and so just, and that's why the voice mode thing's so powerful. [01:01:13] Matt Slager: Well, he's gone. [01:01:16] Marno Brits: I found them. The mold book found them. [01:01:19] Dave Pengelley: It took me out. What happened? [01:01:24] Matt Slager: I think, I think a really, really good topic for next week. We should virtually, we should go hard into the idea of how people can start off building their brand voice, putting in ideas, and start generating their own content, you know, using Manos principles and yours. [01:01:39] Dave Pengelley: Yeah. Yeah. We, we, as, as we evolve this show, we've gotta work out how to integrate a bit more, show and tell, like I, Thank you guys for indulging me, showing my website today. Yeah, you're, but actually how we actually use the tools and, and maybe a bit more behind the scenes of things like hand gravity. [01:01:55] Dave Pengelley: so we'll have to think about how we set up a few, you know, nonsensitive environments. We can do demos and share from, discuss that. last we are, we are on the hour, but, new tools. Codex was launched by Open AI this week. It gave it a go, like their Codex tool and compared to antigravity just team, we've talked a lot about antigravity. [01:02:13] Dave Pengelley: It's not that it's still kind of chat GPT. It's still very walled garden and wants you just to treat it like a chat. It shows, look, here's the diffs I'm making to files, but there's no file explorer that I could see. As soon as you want to go and click on anything to edit it, it says go to cursor or go to antigravity. [01:02:29] Dave Pengelley: Like it tries to spit you into another IDE to finish the job if you wanna manually edit stuff. So it's this weird halfway house. Where it gives you a desktop app for Codex, which I think is it, they're trying to fight into that sort of antigravity called Code Sphere, but it feels incomplete. now it's early days. [01:02:46] Dave Pengelley: I've been out 24 hours and maybe I'm not the target audience, but I'm not transitioning out of Antigravity yet, and it's Mac only as well. [01:02:56] Marno Brits: Okay, well then I, I'm not gonna touch it. [01:03:00] Matt Slager: I've got many thoughts, but it's probably worth going into next time. [01:03:03] Dave Pengelley: Yeah, we'll see. We'll see. They, they've got a webinar at 5:00 AM Sydney time tomorrow morning on X. [01:03:07] Dave Pengelley: So I might try and catch a summary replay of that and see how they recommend using it. Maybe I'm just doing it wrong. That's possible. but yeah, I wasn't blown away by the New Codex app release. so we'll talk more about that in the future. But, thank you boys. That was a good one. Lots of deep different discussions on all kinds of things. [01:03:28] Dave Pengelley: Hopefully audience. You loved it, whether you watch it live or on the replay, for those joining us live and participating, thank you everyone. It's, great. We are here to help grow and educate you, to use AI more effectively, more, better, more securely, more safely for running your businesses. [01:03:48] Dave Pengelley: Thank you all. We'll, wrap that up and see you all. Same bat time, same bat channel next week. [01:03:55] Marno Brits: Yeah.