Stop Uploading Private Files to Sketchy Sites. On-Device Tools with Angus

[00:05] Matt: Good afternoon. [00:06] Dave Pengelley: Good afternoon. It is afternoon now. When when when Angus joined the pre-show five minutes ago, I said good morning, and it was all very confusing. Is it good morning? It is definitely good morning now. Good afternoon now. I feel like what's the Truman Show? Good, good, good day, good morning, good afternoon, good evening, good night, or whatever. [00:22] Matt: All of them. [00:23] Dave Pengelley: Anyone seen Truman Show? Just [00:25] Angus Lee Forbes: Yes. [00:25] Richard (Syllogism ) Webbe: Yes, [00:25] Dave Pengelley: me. [00:26] Richard (Syllogism ) Webbe: good [00:26] Dave Pengelley: Yeah. [00:26] Angus Lee Forbes: Recently. [00:26] Richard (Syllogism ) Webbe: afternoon, good [00:26] Angus Lee Forbes: It's [00:27] Richard (Syllogism ) Webbe: evening, [00:27] Angus Lee Forbes: good. [00:27] Richard (Syllogism ) Webbe: and good night. [00:28] Dave Pengelley: Yes. [00:28] Richard (Syllogism ) Webbe: Have a great day. [00:30] Dave Pengelley: Uh we do have a new good morning and good afternoon to uh Angus. Welcome to the show, Angus. [00:35] Angus Lee Forbes: Thank you very much for having me, Dave. [00:36] Dave Pengelley: It's uh it's good to have you join us. You're uh launched your own AI product. Um, so um keen to hear some of your insights as we talk about the AI world and the AI market as we look at the news and things and so on. Um, but how's everyone's week been so far or morning? Let's let's let's focus on the morning. How's everyone's morning been? Are we all in a good place to talk about the world of AI today? [00:59] Matt: I mean it's always good place to talk about it. Um my morning was hectic. I I've had client calls already. Um just yeah, it's been a busy morning. So I'm I'm glad to be here to kind of relax for a little bit. [01:11] Dave Pengelley: Nice, nice. I've had a I've had a relaxing morning. Relaxing week. My wife's taken this week off. Our kids' school holidays are like this weekend, like they have a third week, which is this week. And so my wife took this week off. And so yesterday we went and did dentist visits for the kids and shopping and other bits and pieces. And so I feel like I'm on holidays. Like I didn't sit on my laptop all of yesterday and this morning sleep. And went for a walk and then got ready for the show. But that's that's weird. I've like I'm just looking at all my my token quota usages not being used each week and being like, oh no, I'm wasting time and money. But it's good to good to step back a little bit sometimes. [01:43] Angus Lee Forbes: Yeah. [01:43] Dave Pengelley: How about how about you uh you guys uh angus Richard? How [01:46] Angus Lee Forbes: Yeah. [01:46] Dave Pengelley: are you morning? [01:48] Angus Lee Forbes: Good. I I just had uh come back from two weeks camping, so I've been completely offline. [01:53] Dave Pengelley: Nice [01:53] Angus Lee Forbes: Um I I I timed it just as Fable was coming online. Um so my token quota. Is quite healthy right now. [02:03] Richard (Syllogism ) Webbe: I noticed that was it someone gave me an extra hundred dollars of credits to keep going on Fabian. [02:11] Dave Pengelley: But that [02:12] Richard (Syllogism ) Webbe: So [02:12] Dave Pengelley: anthropic, [02:12] Richard (Syllogism ) Webbe: uh [02:12] Dave Pengelley: they do that sometimes. [02:13] Richard (Syllogism ) Webbe: Yeah, they did that. They just offered me an extra hundred dollars of credits, and I looked down and thought, oh, I didn't realize I was using it at the peak power, so I wound it back a bit. So didn't uh look into that yet. [02:27] Dave Pengelley: They they gave us an extra hundred and fifty dollars of extra Usage credit. I think it was when they turned off the open claw access when they said third party harnesses couldn't use your subscription anymore. They gave everyone 150 bucks. And I hit my daily limb, my weekly limit, or whatever. And I was like, ah, I'll just I'll just get co-work to run my regular scheduled things or whatever, and my morning processes and my sort of daily reports and things. And I was like, you I'll use extra usage. I've got it there. One might as well use it. And my daily morning thing burned through like fifty dollars of extra usage in one run. And that was a bit big, that was a big eye-opener for me going, Oh, I'm being really Really, really lazy with my uh with my processors and my systems in the fact they're so inefficient they could burn through fifty dollars of usage for like one morning routine. [03:11] Matt: It's pretty crazy. [03:13] Dave Pengelley: Yeah. [03:13] Matt: I'm trying to set up more of those things now. And I'm kind of on the other end of it. I'm like, okay, you know, efficiency, yes, but how do I maximize the amount of actual result here? You know, like what am I actually doing? What is the problem? What do I want? And how do I ramp it up? And then gently like increase the efficiency later. So decrease [03:34] Angus Lee Forbes: Oh, [03:34] Matt: the usage. But um yeah, I've had an insane week of usage. I've had like 7 billion tokens in the last week just [03:43] Angus Lee Forbes: which one [03:43] Matt: smashing [03:44] Angus Lee Forbes: is [03:44] Matt: soul. I thought I'd be slowing down because it's made me a lot more deliberate about what it is that I am doing. [03:51] Angus Lee Forbes: the [03:51] Matt: You know, you know, the the idea of you can do anything, [03:54] Angus Lee Forbes: one [03:54] Matt: but what does that mean? You know, how do you choose what [03:57] Angus Lee Forbes: actually? [03:57] Matt: to do at that point? So um what it meant is all of these various client projects that I've been putting off, I've now just started to uh ramp up each of them individually in parallel and my poor machine, I I definitely need a new machine now. It's just been pinned on the Rev limiter all week. [04:15] Dave Pengelley: Wow. Wow. Well seeing seeing we've used the A word A many times, we should probably do one of these. [04:32] Dave Pengelley: This issue with video generation with AI. You look at something like that, and every time I watch it, when it first pops up AI operators, I'm like, the contrast is off. The logo is too dark. It needs to be a little bit brighter before all the electricity and the flashes happen. There's no easy way to change that because the magic AI engine just built that for me. And I don't have the component files, I don't have the layers. I can't go in and just grab that and bump the colors on that one thing easily. It just is. [04:57] Angus Lee Forbes: Just is. It just is. You know, one shot everything, [05:01] Dave Pengelley: That [05:01] Angus Lee Forbes: right? [05:01] Dave Pengelley: that's it. That's that's uh what does uh what does Matt Barry uh talk about? He sort of like insert insert coins and like pull like um pre prey pray to pray to win or whatever, like like yeah. [05:12] Matt: Flotion. [05:12] Dave Pengelley: Um it's very much that slot machine thing with these AIs. Um yes, so speaking of slot machines and AIs, and and you know, and we've we we talked about I can't was I've losed track with about planning pre-show and post-show. There's lots of things going. Going on with AI companies doing dodgy things with your data, right? Um, which I I want to use, we'll we'll jump into the news of that, but I actually want to use that as a segue into letting Angus introduce himself a little bit and what FWIP app is. Uh, because I think that's something I would really like him to do, because I think it's really cool what he's built and what he's doing there. Um, and so uh Angus, can you please tell us what it is that you do? [05:53] Richard (Syllogism ) Webbe: I love the nine bother why it's cool. [05:55] Angus Lee Forbes: Oh, thank you. Thank you. I really appreciate it. So um I guess the long and short of it is that um it's a uh it's a private file utility um and essentially it's taken everything that you need to like a Swiss Army knife approach to everything you would need to do with a file. It's kind of like all the boring things. But instead of you sending your private data up to a server farm to do the processing, we send the tooling to you and you just do the operation whatever it is on your device, whether it's a phone, Whether it's a laptop. Um, and whilst there is a considerable, I guess, paradigm shift in the innovation, um it really is just coming back to what software used to be, which is you pay to do something and that's it. There isn't a kind of creepiness of people training on your data or you know, reading your contracts, or putting you in this kind. Of like um dark design paradigm where you can't unsubscribe. So um this all kind of was built out of um something that happened to my wife. We don't have a printer, and um she needed to sign a document, and the only way she could work out how to sign a document was to jump online and go onto one of those kind of like really sketchy websites that said offered you a free trial, [07:27] Matt: I forgot [07:28] Angus Lee Forbes: and [07:28] Matt: that. [07:28] Angus Lee Forbes: lo and behold, she Signed the document, happy days. But seven days later she got pinged for 90 bucks US and for a for a year's subscription, right? And I was just like, surely there is a better way of approaching this. And then that just sent me down this rabbit hole of kind of like really just like data sovereignty and just like m helping people kind of have a little bit more control back of what they want to do. And I really wanted to kind of start on the real kind of basics, the the kind of version Uh 1.2, which I'll be shipping in about two weeks, is gonna have a full word processor, a full kind of sheets component. So really looking at the kind of how you use your basic functions on a computer, but allowing that kind of power to be like ported to a phone or whatever, but also giving people their own data um surety, um, their own data sovereignty. So yeah, that's c that's kind of like the loose concept now. Um the the I guess the most interesting parts, and I didn't really expect it. I kind of built it for like the the nurse or someone that, you know, you know, was kind of low tech and they just needed to do a simple thing. Um, but I've been having really interesting conversations with a lot of businesses recently, particularly in the enterprise space. Um with the Microsoft, you know, I don't know if you saw what happened in The Hague where Microsoft pulled the licenses of um the people. At the International Criminal Court. And with that, that just set about this whole like, well, if they can do that on such a like reputable body, it could happen to any business, any enterprise. And [09:13] Richard (Syllogism ) Webbe: Yeah. [09:14] Angus Lee Forbes: I think, you know, like all the big players in the market are running to siphon all your data up. So they're incentivized to um, you know, extract as much value out of you as possible. Whereas I think that this is trying to just give. You need the tools that you need to do the job that you want to get [09:32] Richard (Syllogism ) Webbe: So [09:32] Angus Lee Forbes: done. [09:33] Richard (Syllogism ) Webbe: it's just it was an interesting, I'm glad you said that. We've all, I think every one of us have suffered from exactly what you just described, Angus. So your monetization model to get credit for your work, it's a one-off fee, is [09:46] Angus Lee Forbes: Yep. [09:46] Richard (Syllogism ) Webbe: it? Beautiful. [09:48] Angus Lee Forbes: One off fee, and and as we develop more tools, they just get rolled in. So it's not, it's it's one of those things. And like I've had a couple of conversations with people who kind of go, like, what how how how does that that actually work and it's like well given that we now no longer require massive marketing departments massive product departments you can really flatten your technology stack um and that's that value can be given back to the customer um so I think you know and I I I genuinely believe if you build useful things that people want to use the value then comes. Do you know what I [10:24] Richard (Syllogism ) Webbe: Absolutely. [10:24] Angus Lee Forbes: mean? So you know [10:26] Richard (Syllogism ) Webbe: Absolutely. [10:26] Angus Lee Forbes: yeah so I think I I think that's you know like software Along the way, kind of forgot those things in its kind of you know, march on to get higher and higher um shareholder returns. So I [10:40] Richard (Syllogism ) Webbe: It's [10:40] Angus Lee Forbes: think [10:40] Richard (Syllogism ) Webbe: just [10:40] Angus Lee Forbes: I [10:41] Richard (Syllogism ) Webbe: been run off. [10:41] Dave Pengelley: Well [10:41] Angus Lee Forbes: yeah, [10:42] Dave Pengelley: every every [10:42] Angus Lee Forbes: well, [10:43] Dave Pengelley: it [10:43] Angus Lee Forbes: yeah. [10:43] Dave Pengelley: everything [10:43] Angus Lee Forbes: And it's [10:44] Dave Pengelley: moves towards the subscrip the subscription economy because that that that looked like you then had a stable balance sheet, right? Because then you had this long tail [10:50] Angus Lee Forbes: yep. [10:50] Dave Pengelley: of subscriber revenue. [10:52] Angus Lee Forbes: Yep. [10:52] Dave Pengelley: If you didn't get any new sales for the next year, you were still gonna have [10:55] Angus Lee Forbes: Yep. [10:55] Dave Pengelley: X amount of revenue for [10:56] Angus Lee Forbes: Yep. [10:56] Dave Pengelley: forever because people would keep paying for you. So Yeah, [11:00] Richard (Syllogism ) Webbe: I love [11:00] Dave Pengelley: we're [11:00] Richard (Syllogism ) Webbe: it [11:00] Dave Pengelley: gonna [11:00] Richard (Syllogism ) Webbe: how they [11:01] Dave Pengelley: get that. [11:01] Richard (Syllogism ) Webbe: they they changed the model. See, and I remember, you know, it was the software didn't cost much and your computer cost the most and you just paid an extra few bucks for the software and wasn't this great. Then all of a sudden the software started getting more and more expensive. So you got this monolithic fee to pay to either Microsoft or someone else just to do very basic functions, but everyone paid it. And then someone said, Oh, how do I break that model? Oh, I know, I'll go to a subscription model and to the rest of the uninitiated, oh, it's only $20 a month. It's only four Oh, that's great. And of course, after a while, Bill Shock hits in and you find out you're paying thousands and thousands of dollars a year. And it's like my TV subscriptions and all of that. It just creeps up. And then Bill Shock just hits you right in the face. So I'm glad you're doing it, Angus. [11:44] Matt: I've got a question on that, like on the opposite end. Um, not not from the consumer end, which kind of sounds potentially weird, but you know, like how can the one-time pricing actually fund? Permanent maintenance and support of the platform. [12:03] Angus Lee Forbes: Yeah, so yeah, I mean my background's in marketing and so customer acquisition is still a kind of fundamental thing. I guess the the the relative value of [12:20] Angus Lee Forbes: why like the these tools are aimed at a really, really what wide segment of audience, which means if I'm capturing 0.01% of the market share. Share at that, let's say $100 USD price point, that's over a million dollars. That's all the money I need. Do you know what I mean? Like I don't I don't need a Ferrari and to be paying shareholders and whatnot. So, in terms of the the real challenge, I guess, what I will face in future is continual customer acquisition to maintain the but the good thing about it is because the way That the technology is built, the cost of me serving is almost minute [13:08] Matt: Yeah. [13:08] Angus Lee Forbes: because all the data stays on your computer. There's not these big up and down data usage. I'm sending you a tiny little script and the work happens on your computer. So there's a substantial cost saving in the kind of background engineering or architecture of how I've built it. [13:28] Matt: Am I right in saying that like it all It just has a browser UI that that hits your your local binary or something. [13:33] Angus Lee Forbes: Yep. [13:34] Matt: Yeah, right. That's pretty interesting. So do you feel that the the need for desktop apps are diminishing and you can basically do everything from the browser? [13:44] Angus Lee Forbes: Yep. Yep, yeah. I'm actually like subsequent to this, I'm building uh AI micro app platform to get round the app store. Um as as people, you know, build more and more. in AI, they're missing that conduit of like they don't want to jump into GitHub. They don't want to just jump into doing this thing. So there is a requirement layer for someone to bring a product to market really quickly without all the pain. And I'm building the kind of conduit between the two [14:20] Richard (Syllogism ) Webbe: Because the enterprise the enterprise uh industry was talking about microservices um a while ago. It sounds very much like that model. It's uh I think that's A good idea, I guess, great, I guess. [14:32] Angus Lee Forbes: Yeah. Well hopefully hopefully you'll see it in a couple of weeks. [14:35] Dave Pengelley: Yeah, [14:36] Richard (Syllogism ) Webbe: Good. [14:36] Dave Pengelley: I mean and and and from a from a long-term point of view, like this was not a new sort like pricing model. This was existed through the 80s and 90s, and you're like Cor Corel [14:44] Richard (Syllogism ) Webbe: Very [14:44] Dave Pengelley: Draw [14:44] Richard (Syllogism ) Webbe: true. [14:45] Dave Pengelley: was out, and then then you know, two years later, Corel Draw two came out or Corel Draw three or four and had new features and stuff. And people, you know, [14:51] Richard (Syllogism ) Webbe: Very [14:51] Dave Pengelley: got [14:51] Richard (Syllogism ) Webbe: true. [14:51] Dave Pengelley: their wallets out if they wanted the new features, otherwise, they kept using their ones. Like Adobe was the same before before Creative Cloud. Like you used [14:59] Richard (Syllogism ) Webbe: Very [14:59] Dave Pengelley: to just buy. [14:59] Richard (Syllogism ) Webbe: true. [15:00] Dave Pengelley: It and you could use that version forever. Um, but now this whole everything's got to be connected into the internet, everyone's got to be a subscription is a massive pain. Even I was playing Xbox with my son the other day, and there was just a glitch with our Wi-Fi, and then the Xbox wouldn't connect to the internet or whatever. We had a game that we'd owned that was on a disc, and we're like, play the game, and said, No, you need to be connected to the internet to play this game. And I'm like, What are you talking about? We've bought the game, we own the game. Why do we need an internet connection to play this game? Like, it's just [15:31] Angus Lee Forbes: Yep, [15:32] Matt: Do [15:32] Angus Lee Forbes: yep, [15:32] Matt: you [15:32] Angus Lee Forbes: yep. [15:32] Matt: like I've got another just quick technical question, Angus? Um the the agents that we're using now, like day-to-day people running their Claude stuff, codecs, whatever it may be, [15:44] Angus Lee Forbes: Mm-hmm. [15:44] Matt: um, very, very good at bash commands, you know, understanding their local environment. Um, are your suite is your suite exposed to them as well? [15:56] Angus Lee Forbes: In terms of like having like a MCP sur or something or a CLI that you know you're yeah like you could do it I don't know I'm I have I I have toyed with the idea of you know making basically a headless version but the kind of [16:16] Angus Lee Forbes: like the fundamental thing for me right now is to kind of help the person that's like a limited user not have to pay a subscription and that that's kind of my my targeted model there's there's all those kind of different layers You're kind of like head down into like a slack type kind of like could you just slack your thing? We're gonna in the 1.2 version rollout, we're gonna have like um like a hotkey thing where you have a shortcut and it'll it's kind of makes it as simple as you like airdropping that you can configure your own either compression or whatever, so you make a little preset. And that'll be really interesting to see how that goes because I I tend to think, you know, having to like if doing something a lot or consistently just being able to set something up that is a single keystroke is the thing that really creates the value for the customer. [17:09] Matt: Sure, yeah, and then people that use stuff like Raycast, you know, that can just expand that even more. Um, [17:14] Angus Lee Forbes: Yeah. [17:14] Matt: do you get any telemetry on what tools people are using? [17:19] Angus Lee Forbes: Yeah, yeah, yeah. I do. I do. Okay. So I use this site um it's a it's a no cookie um uh browser company called Plaw And that gives me a rough overview. I don't see any of the people, and that's kind of fundamental about it, but it gives me a kind of understanding of the the most used tools. Um the just in terms of the the particular subset I'm serving, um it seems tools that have generation as part of it. I've got like an invoice builder, uh PDF form builder, they're the ones that are kind of like the most required. or the highest value. But then there are some weird ones that like pop up. I get a lot of traffic from ChatGPT for um text encryption. [18:14] Angus Lee Forbes: So somehow some chat GPT has found it and you because you can create like a 24 uh was it 256 bit encryption in your browser with like a little text password. Obviously some people are using that. But you know it's there online. [18:30] Richard (Syllogism ) Webbe: Contract security. So I'll send you a contract, you open the browser, and I've given you the password and off you [18:34] Angus Lee Forbes: Yeah, [18:34] Richard (Syllogism ) Webbe: go. [18:34] Angus Lee Forbes: but it's even it's even lighter weight than that. Like you can just do it from a text file. So it's not even a you you can do uh so yeah, there you go. That's exactly the demo right there. [18:50] Dave Pengelley: I used [18:51] Angus Lee Forbes: And [18:51] Dave Pengelley: to be in [18:51] Angus Lee Forbes: then [18:51] Dave Pengelley: front [18:51] Angus Lee Forbes: and [18:51] Dave Pengelley: of [18:52] Angus Lee Forbes: and [18:52] Dave Pengelley: us. [18:52] Angus Lee Forbes: then it um and then and then you put a password with it, and then you can um [18:58] Angus Lee Forbes: like encode and re-encode. So if you roll down two, there you go. [19:05] Richard (Syllogism ) Webbe: That is cool. [19:06] Angus Lee Forbes: And then if you just roll down, there you go. And then you have a um you know copy paste and then an encode and a decode function. [19:15] Richard (Syllogism ) Webbe: That [19:15] Angus Lee Forbes: Anyway. [19:15] Richard (Syllogism ) Webbe: is very cool. Very simple user interface. [19:19] Angus Lee Forbes: Yep. Yeah, yeah. It like as I said, like I I built it for people that just need to do a simple job. And I came down with the kind Of understanding that it is just about, you know, and you know, like the the way I've kind of like gated it in, because obviously it does need to make some money to be of value to me, is it's any tool on there, it's full functionality, it's three free a day. And if you decide to use it more, then you can pay for [19:52] Dave Pengelley: Yeah. [19:52] Angus Lee Forbes: it. But like for most people, like who need to sign a contract, they don't need to sign more than three bloody contracts a day, like so. [20:00] Dave Pengelley: But [20:00] Angus Lee Forbes: It's [20:00] Dave Pengelley: the [20:00] Angus Lee Forbes: just [20:00] Dave Pengelley: the [20:00] Angus Lee Forbes: it's [20:00] Dave Pengelley: old [20:01] Angus Lee Forbes: just [20:01] Dave Pengelley: shareware [20:01] Angus Lee Forbes: there to be used. [20:01] Dave Pengelley: model. [20:02] Angus Lee Forbes: Yeah, yeah, exactly. [20:03] Richard (Syllogism ) Webbe: Total [20:03] Angus Lee Forbes: Like [20:03] Dave Pengelley: Yeah. [20:03] Angus Lee Forbes: I mean [20:04] Richard (Syllogism ) Webbe: freemium, total freemium. [20:10] Angus Lee Forbes: well but like it you know, if if a thousand people use it like they did last month, [20:16] Dave Pengelley: Yeah. [20:16] Angus Lee Forbes: then I get, you know, um you know a conversion rate of about one percent. It's not bad. [20:23] Dave Pengelley: Yeah. [20:23] Richard (Syllogism ) Webbe: And can I you d can I use it on my phone? Is it still pretty comfortable with the phone? [20:27] Angus Lee Forbes: You can you can use it on your phone. it and you can even install it on your phone straight away from the browser. [20:34] Richard (Syllogism ) Webbe: Excellent. Because of AI and tools now and access to data, I almost never get on my computer except for a podcast or a video conference call. I seem to hardly ever open my computer and I do everything from my phone now. [20:48] Angus Lee Forbes: Yeah, yeah. Yeah, it's a it's amazing. And you know, obviously the voice, you know, components of all those AI tools are just getting better and better. And you know, the amount of times that I'm like just Narrating notes to myself and then just run like whisper over it. Do you know what I mean? Like it's you can just see where that's going. So you're you're less and less constrained by the devices you're using, and more and more kind of they're they're suiting into your life. But I also think that the the trade-off with that, I don't know if you guys experienced that recently, but uh Google rolled out Gemini across all their home devices. [21:23] Dave Pengelley: I'm an Amazon family, so I I I don't have have those. [21:26] Angus Lee Forbes: But but with that, with that kind of higher [21:30] Richard (Syllogism ) Webbe: Uh, [21:30] Angus Lee Forbes: Now you there was this thing in the back of my head, like, do I want to upgrade? Do I actually want Gemini listening to me the whole time? You know, listening to my kids, you know, like I'm just like, oh, I don't know if I like, and that kind of invasiveness of [21:47] Richard (Syllogism ) Webbe: I'm not sure. [21:47] Angus Lee Forbes: like technology always monitoring you um starts kind of creeping into the [21:52] Dave Pengelley: I [21:52] Angus Lee Forbes: home. [21:52] Dave Pengelley: I saw I saw a guy who posted it that he um he did a data request access thing to Amazon for all of the Utterances that it had captured and it sent all these web files and all this data, all these weird conversation fragments and discussions that that weren't related to the smart speaker at all. But it just, you know, I was listening and captured and recorded snippets here and there, waiting to hear its name. Um makes [22:18] Angus Lee Forbes: Yeah. [22:18] Dave Pengelley: me think of to Tony Wonder from uh Arrested Development, just hiding hiding in inconvenient places, just waiting for people to say, I wonder, so he could pop out and be a magician. [22:28] Dave Pengelley: Deep arrested development. Reference for anyone who knows uh Ben Stiller's cameo character. [22:32] Richard (Syllogism ) Webbe: Yes, yes. [22:33] Dave Pengelley: Um, but yeah, they they they do capture a lot, and like Amazon are talking about bringing out Alexa Plus or whatever they're gonna call it with sort of GPT style LLM intelligence behind it. And there is that element where you go, well, that'd be really convenient, just being able to ask questions ambiently in my environment. But what what is the trade-off if it's free in your product? [22:56] Angus Lee Forbes: Yeah. [22:58] Dave Pengelley: Yeah, yeah. [23:00] Richard (Syllogism ) Webbe: the old uh fish and chip shop where they had those little forms you'd fill out while you're waiting for your potato cakes and uh you could win anything from a holiday to a free saucepans or something. It was just marketing grabbing your details [23:11] Dave Pengelley: Yeah, [23:11] Richard (Syllogism ) Webbe: right [23:12] Dave Pengelley: so the business [23:12] Angus Lee Forbes: Oh, [23:12] Richard (Syllogism ) Webbe: we [23:12] Dave Pengelley: card and the fishbowl. [23:13] Richard (Syllogism ) Webbe: yeah that's it same thing. [23:15] Dave Pengelley: Yeah. No, I think I think it's great, great, Angus. I think um [23:18] Angus Lee Forbes: yeah. [23:19] Dave Pengelley: uh if because you're delivering a tool that runs on their hardware and you're not having that cost to serve issue of maintaining, you know, cloud databases and infrastructure and servers as much, it it's great. I mean, one of my favorite tools ever. On my Mac was Better Touch Tool, which gave me that Windows style snapping thing. And that was it runs on my computer. And so they charged you a one-off license fee, and then you could run it. [23:38] Angus Lee Forbes: Yep. [23:38] Dave Pengelley: And amazing. [23:40] Angus Lee Forbes: Yeah. [23:40] Dave Pengelley: Um, that's how software can and could and should be. But the problem is so many things do have these cloud services now, and it's reliant on there's a cost to search. [23:51] Angus Lee Forbes: Yeah. Yeah. [23:53] Matt: Oh no. [23:54] Angus Lee Forbes: And he's frozen. I think I I think the cloud heard him and cut [23:58] Richard (Syllogism ) Webbe: Yeah, [23:58] Angus Lee Forbes: him off. [23:58] Richard (Syllogism ) Webbe: no, it's It's fr this time he's frozen in quite a complimentary position, are there? [24:04] Dave Pengelley: But I give I give people the the the permanent three-day like like three-day history free trial forever so you can have a free edition if you don't but yeah I think I think when you're doing the sovereignty thing is really clever. Um with your like so you've got some uh we'll move on to some news in a sec, but you've got AI tools there for doing some transcription and stuff. So that's just local model stuff that you're deploying onto their computers. [24:24] Angus Lee Forbes: Yeah yeah so like uh uh there's there's two which are the kind of standout models, which is the um like private meeting transcription one. So, you know, not too dissimilar from like granola or something like that. Now it doesn't like listen in on your computer and it won't pull in there. But if you record a meeting and you have people's consent, you can upload it and then it'll give you a transcription of it. Now, how does that do that? It you it loads Whisper into your browser and does it and then it stays cached in your browser so that anytime that you need to do it. It doesn't have to re-download that model. So that's on one side. I also have like an image upscaling one that's really, really good. And actually, the one that's got a lot of traffic recently was a tool I just recently developed, which was um video for AI, which is uh um a tool that allows you to upload a clip, whatever it is, and then it will transcribe it, but then it'll also keyframe the transcription and make it packaged in a way that AI can watch the video so that if you go you know if you want your Claude to uh do an analysis of a particular video or a podcast or pull out the most interesting bits it can do it [25:44] Richard (Syllogism ) Webbe: Wow. So what's that? There was a bit of code you used to be able to use to extract video from YouTube. I wonder if you use that and combined it with that, you could actually toll the web for the right bits of video. [25:57] Angus Lee Forbes: you could you could and you can use you know um uh uh Google Notebook LM to do the YouTube component. But you've got to be really like this is this is the kind of legal [26:07] Dave Pengelley: I think [26:07] Angus Lee Forbes: gray [26:08] Dave Pengelley: that were [26:08] Angus Lee Forbes: area. [26:08] Dave Pengelley: lighter [26:09] Angus Lee Forbes: I [26:09] Dave Pengelley: in [26:09] Angus Lee Forbes: can't [26:09] Dave Pengelley: the [26:10] Angus Lee Forbes: do you know if if someone puts a YouTube link in there, I can't do the kind of downloading conversion because there's rights associated with it. So [26:20] Dave Pengelley: instruction. [26:20] Angus Lee Forbes: all I can do is do the associated kind of technical backend. Um you could yeah like like as as as soon as I developed it I'm like oh yeah I should just you know make it really easy for people to just dump leaks in and then you get that transcription but the problem is is the kind of legal gray area that puts me in and [26:39] Richard (Syllogism ) Webbe: Yeah, [26:39] Angus Lee Forbes: I'm [26:39] Richard (Syllogism ) Webbe: we [26:39] Angus Lee Forbes: just [26:39] Richard (Syllogism ) Webbe: see [26:39] Angus Lee Forbes: like [26:39] Richard (Syllogism ) Webbe: yeah, we see LLMs come back to us all the time when we make certain requests saying I don't have permission to do that, so I get that. [26:46] Angus Lee Forbes: yeah [26:46] Dave Pengelley: Yeah. Yeah. All right, well let's uh speaking of LLMs and their permissions, let's do this. [27:00] Dave Pengelley: Uh the news. Uh speaking of dodgy AIs doing dodgy things. Matt, tell us about um the alien hugging face. [27:12] Matt: So I actually don't know the exact task that OpenAI was tasking to GPT 5.6 and allegedly GPT 6 coming out soon, you know, doing testing on it, doing evaluations, you know, just checking to see what it can and can't do. And uh yeah. They released today a post modem report of the fact that it actually cheated on their test by breaking out, getting access to the internet, and hacking Hugging Face's database to get the answers to what it needed to solve. [27:45] Richard (Syllogism ) Webbe: I love the [27:45] Matt: So [27:46] Richard (Syllogism ) Webbe: Venetian romance. [27:47] Matt: yeah, it's incredible. Like, um, obviously, this is the kind of stuff that caused the the USG to shut down Fable and and and the reason for uh Project Glass wing. And OpenAI's daybreak to exist in the first place. But uh Theo's statement just there is actually like on point. You know, new OpenAI models so goal-oriented that they literally escaped the environment, the containment, and hacked hugging face to cheat a benchmark. You know, it's it's crazy. So yeah, I don't know. I I think this kind of thing, you know, everyone would naturally go to the negative and be like, you know, security risks, all of the things, but I think it's great because it means that we're actually more aware of security things that people are getting so blasé about. And it's the kind of stuff that I talk about on here a lot is, you know, how do you actually know what's going on? And for for tools like like Angus's suite, it it made me tune into the idea of, you know, if these agents have the ability to do things so easily through a tool suite like that, I would want them exposed to that. And I would want them to be, you know, have The exact progressive disclosure skill file uh discovery system to be like, well, clearly user wants to do this, so maybe I should go hack Hugging Face to get that answer. Um [29:12] Richard (Syllogism ) Webbe: There was a funny article I think I sent you guys during the week about a guy after um uh Google Maps Maps went agentic and uh so he and he put it in to send it send him home and he said the the agentic agent was reasoning too much and uh he Writes a very funny article and says it's now five hours. I'm not home yet, but Google's been reasoning all the reasons I should stop at some places and not others. [29:39] Matt: But I remember the image, it was of a it was of a loop. [29:42] Dave Pengelley: Yeah. [29:43] Richard (Syllogism ) Webbe: I can show that image. Yes, it's a little loopy gone around in in uh. [29:48] Dave Pengelley: Yeah. Everything's a loop. [29:50] Richard (Syllogism ) Webbe: It's just funny how we give I I'm starting to think of data and AI now as a um uh as a like a like water. If you don't contain it properly, it'll spill out everywhere on the floor and run in many different directions that you don't know where it's going to go. And certainly with the way we're doing these things and giving agents the ability to reason without proper containment, it's of course it's going to run off in every direction. It's a mathematical model and off you go. [30:19] Dave Pengelley: Yeah, it's it's the goal the goal orientation thing, right? Like like if it it doesn't ha have constraints or guardrails or so on and it just knows I need to achieve this for my user, like I I do, you know, dabble with my automated trading robots. And last night, and I like I said, I'd had a day off off from the computer, but I was like at nighttime sitting on the couch going, hang on, I'm just gonna open up Telegram. I'm gonna just say, Hey, just go go make some profit. Just go work out how to build a robot that's gonna make some profit. And what it did is it went, Oh, well, you know, three days ago, Dave was working on this particular thing, so I'm just gonna recycle that and show him those results and say, look, use this one. And uh I was like, That's that's you didn't do any work on that one. That was lazy. Go create a new novel. idea and go work on that one. But you know it it did what I told it to. It said, oh this one looks like it could be profitable potentially based on some testing. I was like I know that one I want a different one. [31:12] Angus Lee Forbes: Efficient, not lazy. [31:14] Dave Pengelley: Well yeah exactly right but [31:16] Richard (Syllogism ) Webbe: Detailed [31:17] Dave Pengelley: um [31:17] Richard (Syllogism ) Webbe: investigation, Angus. Detailed investigation, [31:19] Angus Lee Forbes: Yeah. [31:19] Richard (Syllogism ) Webbe: exactly. [31:20] Angus Lee Forbes: Yeah, I think I mean I think it all harks back to the the classic kind of alignment problem that we're all gonna be walking into. And it's the kind of even if you give it all All the guardrails, all the right constitution, there is still a way that those incentives could come out to be perverse, you know, the staple maximizer kind of idea. You know, anytime that you point it, even if it's completely sandboxed in, it might just reason its way out of that sandbox anyway. So [31:52] Richard (Syllogism ) Webbe: It's [31:52] Angus Lee Forbes: it's a [31:52] Richard (Syllogism ) Webbe: very [31:52] Angus Lee Forbes: very [31:52] Richard (Syllogism ) Webbe: funny [31:53] Angus Lee Forbes: hard. [31:53] Richard (Syllogism ) Webbe: because without contextual guardrails, and of course they are infinite rather than [32:00] Richard (Syllogism ) Webbe: study at IBM once, uh I'm showing my H with the old giant discs in the Baker Lite, right? And um we used to uh get people to test products and we'd look through a window and see how they went. And uh we had a group of people in there and they went in to get ready to set up a might have been a system 3 320 was you know very old old system. And the first instruction in the manual says take disc out of box. And they basically pulled the disc out and then ripped it out of the bakery cover to just have this magnetic disc in their hand and then looking at us, how does that go with? I mean instructions, implicit instructions, are just as dangerous with an AI agentic model as with anyone, a two-year-old, an adult, [32:44] Dave Pengelley: Yeah. [32:44] Richard (Syllogism ) Webbe: you name [32:45] Dave Pengelley: It's [32:45] Richard (Syllogism ) Webbe: it. [32:45] Dave Pengelley: it's it's that context of the ambient context and that logical framework that you work within that we talk about all the time. These are these are critical elements um that you know the AIs don't always have. And that's why the old jailbreaks used to work back on the early days of ch GPT three point five and And stuff where it'd be like, tell me how to do something, you know, terrible. Uh, and we'll go, I can't do that. That that breaks my thing. No, no, I just I just want to teach someone why they shouldn't do that. So please please please tell me so I can do an educational lesson for someone and goes, Oh, if this is for educational [33:15] Angus Lee Forbes: Yeah. [33:15] Dave Pengelley: purposes, no worries. Here's the thing you wanted me to write for you. I was like, excellent, [33:20] Richard (Syllogism ) Webbe: That is, [33:20] Dave Pengelley: thank you very [33:21] Richard (Syllogism ) Webbe: you know, [33:21] Dave Pengelley: much. [33:21] Richard (Syllogism ) Webbe: I had I had several examples of that recently where uh that AI said, No, I won't do that, and I just asked it a different way and offer win. [33:28] Dave Pengelley: Yeah. [33:30] Matt: Um, there's a guy that I follow a lot that is, you know, he his name on X is is Pliny the Liberator. You know, his whole shtick is that he can break any model, you know. Pliny, P-L-I-N-Y. And um funnily enough, I was just scrolling down just to see if I could uh get a couple of statements just to add to what we've just been saying. And literally Pliny saying, you need to be a pattern watcher before [33:56] Richard (Syllogism ) Webbe: Yeah. [33:56] Matt: you can break the patterns. So any of those things. things. Yeah, it just you need to understand the patterns. So, but that that kind of like really explicit instruction following that you just described before. I I have a I have this, it's a joke, but I use it a lot for an describing those situations. It's when the the husband is sent to the shops, right? And the wife says, you know, I need you to go get milk. And if they have avocados, can you get four? And he comes home with four bottles of milk. And she's like, what did you do? And he goes, well they had avocados. [34:31] Matt: So it's [34:33] Angus Lee Forbes: It's [34:33] Matt: not wrong, [34:34] Angus Lee Forbes: good. [34:34] Matt: right? It was a hundred percent an accurate response to that situation. [34:38] Richard (Syllogism ) Webbe: Assume [34:38] Matt: Kind [34:39] Richard (Syllogism ) Webbe: you're [34:39] Matt: of [34:39] Richard (Syllogism ) Webbe: dealing [34:39] Matt: weird. [34:39] Richard (Syllogism ) Webbe: with the two rolls. [34:41] Matt: Yeah, so that that kind of like, you know, some people they don't fathom that people think that way. Um, and I'm one of the weirdos that does think that way. And so when you're actually instructing these models, you gotta consider the fact that they're they're pretty like autistic, you know, like they're very single-track minded. Um, somebody described uh GPT 5.6 like a like a rottweiler. You know, it'll grab hold and it won't let go. It'll literally do that thing and until it breaks and then uh hacks hugging face. Um whereas you know the likes of the other end of that spectrum is is Fable at the moment, which is the one who's the super I don't know, experienced coworker. You wouldn't get them to do everything, but you want to have a chat to them about some stuff to get some ideas. [35:25] Dave Pengelley: I keep saying post where if I won't do anything and just say sorry that I can I can't do that request for you, that breaks. my guard rounds like that I've locked it down. They're so scared by because like they just it's almost unusable. [35:37] Matt: I I follow a lot of um security uh organizations and people that do pen tests professionally and that sort of thing. So full full white box, black box pen tests, and um they've moved all of their agentic pen testing to Kimi K3 because it has zero refusals. Do with that what you will. [35:58] Dave Pengelley: There was this demo put out. During the week from Kimmy, where it basically LARPed Mac OS like crazy [36:07] Matt: Yeah. [36:07] Dave Pengelley: that someone built this full thing, like Kimmy, just sort of fully like most of them will go, No, I can't do that, or they they do something that's sort of vaguely similar but different because they don't want to risk sort of being too close to the source. But this is like a full macOS replica sitting in [36:24] Richard (Syllogism ) Webbe: Holy holy [36:26] Dave Pengelley: your browser, like [36:28] Richard (Syllogism ) Webbe: wow. [36:28] Dave Pengelley: with working applications and everything. [36:30] Angus Lee Forbes: That's [36:30] Richard (Syllogism ) Webbe: Wow. [36:30] Angus Lee Forbes: incredible. [36:31] Matt: Yeah, it's [36:32] Dave Pengelley: Like [36:32] Matt: pretty nuts. That was from Max, wasn't it? Max Meinbeck. [36:34] Dave Pengelley: uh yeah, I can't remember where I got it from. [36:37] Richard (Syllogism ) Webbe: Kind [36:37] Dave Pengelley: You shared [36:37] Richard (Syllogism ) Webbe: of killing [36:37] Dave Pengelley: it, I think. [36:38] Richard (Syllogism ) Webbe: the [36:38] Dave Pengelley: Yeah, [36:38] Richard (Syllogism ) Webbe: emulation [36:38] Dave Pengelley: make max [36:38] Richard (Syllogism ) Webbe: industries. [36:39] Dave Pengelley: one back. [36:39] Richard (Syllogism ) Webbe: Goodness [36:40] Dave Pengelley: So [36:40] Richard (Syllogism ) Webbe: me. [36:41] Dave Pengelley: um, yeah. He asked K3 agents form to recreate macOS 27 with real liquid glass and native apps and web browser, and it just did it. Just wild. [36:50] Richard (Syllogism ) Webbe: Goodness [36:51] Dave Pengelley: Um, [36:51] Richard (Syllogism ) Webbe: me. [36:52] Dave Pengelley: because yeah, no guardrails. If you you asked um chat GPT or something to do with this, it would not give you this level of of result. Oh my god. goodness it's got themes like it's completely wild right [37:07] Dave Pengelley: um clear glass there you go oh my goodness that's that's incredible [37:12] Angus Lee Forbes: That is [37:12] Dave Pengelley: it even says apple intelligence [37:17] Dave Pengelley: um yeah [37:18] Matt: But you're right though, like the the message here is is not that Kimmy's like infinitely better than um what OpenAI [37:26] Angus Lee Forbes: a [37:26] Matt: and Anthropic have been putting out. It's to deal [37:28] Angus Lee Forbes: important thing. very [37:28] Matt: with those guardrails. with that that ethos. I forget the term you used, Angus, the like the the alignment that are is actually in the models to begin with. You know, like if you look at how Kimi K3 was put together, I saw another really interesting chart today. It was to do with uh the the the possibilities of distillation. Like K3 is basically Fable 5 without any anthropic like it's just completely learned off that um the with that kind of distillation [38:00] Angus Lee Forbes: Ah, [38:00] Matt: Training method. [38:01] Angus Lee Forbes: right, yeah, yeah. [38:02] Dave Pengelley: Yeah, well they they they read the reasoning traces, right? So they asked Fable a whole bunch of stuff and then got the reasoning traces and then trained it on the reasoning. [38:09] Matt: Yeah. [38:10] Angus Lee Forbes: It [38:10] Matt: Which is [38:11] Angus Lee Forbes: j [38:11] Matt: pretty [38:11] Angus Lee Forbes: it [38:11] Matt: cool. [38:12] Angus Lee Forbes: well, but it also means like that kind of Frontier Lab AI race, like in two months. You've [38:20] Matt: Yeah. [38:20] Angus Lee Forbes: got something do you know what I mean? Like so it's so it's like everyone's racing as hard as they can and then someone's just gonna copy it and give you ninety five percent of it. [38:30] Dave Pengelley: But [38:30] Angus Lee Forbes: Almost for [38:31] Dave Pengelley: and [38:31] Angus Lee Forbes: free. [38:32] Dave Pengelley: and and so much of this is is is the models are good, but it's also the execution and the harnessing. And so, like Kimmy doesn't have the same guardrails and all the system prompting on top the protection prompts, and so it's a little bit more untethered to go and recreate macOS or do anything else. But it's also then how are they using the tokens from an efficiency point of view around the orchestration and the sub agenting and stuff, which is all becoming the big conversation now? So uh and managing I I watched a uh a talk from the Kimmy um CEO. Morning on LinkedIn. And he was saying, you know, the context window is kind of like RAM in early computing. Like in early computing, you had a very small amount of active like working memory. So you were limited in what you could do. As soon as RAM got up to like 512 kilobytes, applications like Word and Excel were possible. Um, you watch some of the um early documentaries around video gaming and the the hacks they had to do to get around some of these LAM RAM restrictions and so on. And so as LLMs get bigger context windows, that's kind of their working memory, their RAM. You can do more, but then okay, within that. Is it efficient to be using all that big RAM, or do you need to like compartmentalize and sub agent things down? And so that orchestration layer. So how Kimmy thinks about the having the bigger picture and then using subagents for task-specific things and the harnessing of these tools, which we talk about all the time on this show, is the next layer and the next battlefront. And so that's why people loved Claude Code for so long because it was an effective harness for you leveraging the model. Now the Codex app has stepped up, or the Chat GPT app, what do they call it, and is doing a lot. And everyone is now having these harness wars. Is continually uh upgrading and innovating to try and be be the the harness of choice. And so it's not just the frontier of the model, but it's their execution layer of the model as well, which is makes a big difference. [40:12] Matt: You're right. Yeah, because like you could like any of the open source labs like like Moonshot with Kimmy, you know, you can pipe that through any harness you like, really. Like you could even put that through clawed code if you really want to. Like I saw um there was a few few guys I follow, they were putting soul through. Claude code, just to see how it handles the way that that Claude Code does sub agent delegation and invocations, stuff like that. And yeah, it performed great. It wasn't super efficient, but it performed performed really well, and you can do it. Um the the other really interesting thing though that people may be missing, uh I I just found another another link on this topic, Dave. It's the idea of regardless of how these things are actually performing on benchmarks, like there's still a like a cost per type. Task. You [41:00] Dave Pengelley: Yeah. [41:00] Matt: know, you can still kind of work out like a rough cost per task. And and what Theo's talking about here on this particular post is that like Sol for some reason is just still so stupidly efficient. Like if you know how to describe it and make sure that it doesn't come home with four bottles of milk instead of avocados, you know, it will get the job done. And it's already cheaper than models that are cheaper than it. So yeah. Yeah, I I don't know. There's just that whole aspect of like the the kind of the economics of the of the tasks, the people keep missing as well. So, you know, the best case scenario right now, as far as people like Theo and and, you know, myself through my experiences are concerned, is you you pretty much have your Claude model as your coworker, you know, the the collaborator, and then you delegate off to workers like Sol in the language that they like, because you'll get your work done efficiently and you'll get your brainstorming done. Properly with a uh a model that actually gets you like Claude. [42:04] Dave Pengelley: Now that this one's crazy. Output tokens per intelligence index, like soul is less than three Gemini three point six flash. [42:12] Matt: Yep. And [42:13] Dave Pengelley: Like [42:13] Matt: when people were going [42:14] Dave Pengelley: that's [42:14] Matt: on about [42:14] Dave Pengelley: um. [42:14] Matt: GLM, um, you know, GLM is a great model, 5.2, [42:19] Dave Pengelley: Yeah. [42:19] Matt: but it's actually less efficient and more expensive than 5.6 Sol. [42:26] Matt: It's just another dimension to this. [42:28] Angus Lee Forbes: Wow. [42:28] Matt: Not [42:28] Dave Pengelley: Oh [42:29] Matt: that that's really relevant. To many [42:30] Dave Pengelley: It's [42:30] Matt: of our viewers, but it [42:31] Dave Pengelley: but [42:31] Matt: tickles [42:32] Dave Pengelley: it but [42:32] Matt: me. [42:32] Dave Pengelley: it but it comes comes back to again um doing the planning up front and I think as we want to do more advanced things with these models we need to take back some of the ownership and I I was I saw a um a really deep write up from someone on LinkedIn um who is looked at um did a full research study on engineer productivity over time and how there's a lot of engineers that they they saw big boosts of productivity initially with these agents um but they're actually Reporting some issues now, like they're losing their flow because now you're context switching and and managing different agents. And so you're not getting to the deep flow of solving complex problems because you're just asking agents to do that. So some some of their productivity metrics are up, but some of their other [43:13] Angus Lee Forbes: Yeah. [43:13] Dave Pengelley: metrics are down because they're losing that flow state, because they're not having as much ownership of the solution as well. Uh, which you know I I described with all my 30 plus agent profiles I created that were automating to to to to whatever extent. And and I didn't own any of the solutions anymore that they were creating. And I was lost and I didn't understand the questions they wanted me to solve. Like they're asking me for feedback. And I had no input on that feedback because I was so disconnected from it. And I think even in proper full software engineering environments, that's happening where you pass so much of the solutioning off to these agents to do it when they come back to you. You don't fully necessarily know what they built. And it's a very different process to, you know, I worked out how to write this algorithm to solve this, to move that data from there to there and do XYZ. [44:00] Matt: True. And yeah, like how you do that. You know, what are those fundamentals that you can't relinquish control of? You know, that's still being determined. And that that's what like I've been trying to put together my sort of teaching program about is how do you actually build systems properly and allow the agents to do things to help accelerate, but then you don't lose that in the process. And honestly, I'm still figuring it out myself. [44:24] Dave Pengelley: Yeah. Yeah, like the the game I I showed a few screenshots last week and I was open IR just had their build week and I was gonna Bunch of work on that and submit that with my son. And we just we just busy week didn't get onto it. But also it's because before we just threw more prompts at the agent, I wanted to step back and do some planning work to really okay, what are we actually building? What what what are the things we want and what are the goals in this game? What are the different things we want to name things and do all that creative work ourselves rather than just you know palming it off to the to the agent? Um, and you know, that required us to sit down and do some work, and we just never got there during the week, so we didn't actually progress anything. It's easy just to spam the agent and get this agent doing things. But to step back and do the planning work, which is sometimes the less fun, less sexy bit, you just want to dive in and do stuff. [45:06] Matt: You're right [45:06] Dave Pengelley: Um, [45:06] Matt: though, like that is the work that needs to be done. Um, you know, Richard is so like such a high um perpetuator of this statement or question, which is, what is the problem? Like, what [45:19] Dave Pengelley: I'm if not sure you're [45:19] Matt: are you [45:20] Dave Pengelley: not sure. [45:20] Matt: actually trying to solve? You know, if you if you send it off to do something that doesn't define the problem well enough, yeah, you'll solve something really well, but it probably won't. What you're trying to do. [45:32] Dave Pengelley: Yeah. Yeah. I've I've I I've been going through this whole determinism workflow process with this podcast, right? And and trying to get the scripts running. So like and you're like this, Angus, that [45:41] Angus Lee Forbes: Yeah. [45:42] Dave Pengelley: now I pull down the recordings from Restream and I can pull down all four separate camera feeds and audio files. And then I'm running local sort of parakeet models on my computer that will then transcribe the four individual um audio files and then mesh that into a super transcript file for everyone and then We'll generate all the summary feeds and stuff I need for putting up on the YouTube description and for the website. And the next step is then building out the storylines that we have within the script of what we talk about to then generate, you know, other marketing assets, sort of the comic sheets or to clip down and do all that again locally using like hyperframes and stuff. And so trying to build these deterministic things, but I have to actually then go through and make sure it's doing the right job. And I've had to really be hands-on and thinking, even when it comes to the storylines, go in and review them. We go, that's stupid. Why would you put that in there? We need to change that or pull this storyline [46:34] Angus Lee Forbes: Hmm. [46:34] Dave Pengelley: instead. And making sure that the humans are still in the loop. I think there's we're on this pendulum, right? We want to automate everything, but then we lose sight of things. We need to come back and rebalance a little bit, or we're gonna end up in this kind of future, which I saw. This is fun. So you know, we've we've got that, but then then the next stage is 2030. [47:00] Matt: You know [47:00] Dave Pengelley: Yeah. [47:00] Angus Lee Forbes: Yeah. [47:00] Matt: what, like it's funny as, but that's kind of how I work now. Like where uh what you guys um you described before somebody like the the LinkedIn post, the engineer saying that their their attention's too disparate and whatnot and they can't like focus. So my bandwidth, like my actual cognitive, you know, like my secret source for me being a human that is doing something useful for people, you know, how can I maximize that while doing exactly that thing? And I've had to curate like systems that create Focus packets for me. You know, you think about like the CEOs that have all the same shirts in their wardrobe so that they don't have to waste that energy on like what do I wear today or what do I eat today? You know, I I've gotten to the point now where I have to like, you know, what do I do today? And, you know, can you just give that to me and then I'll go and do it? I don't want to think about what to do. And yeah, it's weird. That's literally like at the point where I am. [47:53] Dave Pengelley: Yeah, yes, this is this is a paper um called The Productivity Experience Paradox by Annie Vella. Um and yeah, it's that's a huge deep write-up um on what's going on. But just here we go. Uh perceived worsening of developer experience. Flow state is worse, cognitive load is worse. Uh, but you get better feedback loops because you are iterating quicker. You're [48:16] Angus Lee Forbes: Oh [48:17] Dave Pengelley: not waiting, you're [48:17] Angus Lee Forbes: yes. [48:17] Dave Pengelley: not going and doing the thread hunting and reading and thinking, so you can try ideas. So you get faster feedback loops, but generally people are feeling pretty rubbish on some of the other aspects of their experience. [48:30] Matt: You know, how do you let's say you want to get into the industry? You want to be somebody like that? Like, do you do you even touch uni anymore because of all the weird institutionals, like strange dogmatic approaches? Or do you just chase those like niche pieces of information that makes you a good engineer? Like that ability to make good discernment, good judgment, you know, your unique taste. You know, how do you create, foster, and uh amplify those traits while the systems can handle the [49:00] Dave Pengelley: It's it's understanding the first principles and those core structural things that are probably somewhat non-epemeral, that that just sort of stand the test of time around. But then also knowing, but once you know them, then you can look at where you do you break those rules. So I was watching an interview with someone talking about, you know, screenwriting and film stuff, and they mentioned books like Save the Cat, which talk through screenwriting techniques and the hero's journey type sort of stuff, and knowing, and he said, like at film school, they'd be watching a movie and they'd have a time. timer and it'd be like no down when X happens and movie after movie it's like a 12 and a half minutes this particular story trope type thing happens because that's the way they pace movies. That's the way they pace stories. But he said he made the comment that once you know all the rules, that's when you can start breaking them, doing interesting things. And you think about you know your guy Richie's and your Tarantinos that do all that sort of broken, fragmented storytelling, but it works. But they knew what they were working with. They understood the un underlying fundamentals before they started breaking and experimenting. And I think, you know, you could understand, you know Database normalization and things like that, before you can go, why would I not do that? Why should I break that? Why is breaking that going to give me a better result? [50:09] Angus Lee Forbes: Yeah, [50:09] Dave Pengelley: Angus? [50:10] Angus Lee Forbes: but it but isn't it like you know that kind of like curve of like you know it's what you don't know and then you can actually be and you think you're an expert. And right now I think there's a very short line between your ego and AI making you feel like you're an expert to [50:26] Richard (Syllogism ) Webbe: Excellent. [50:26] Angus Lee Forbes: an outcome. [50:27] Richard (Syllogism ) Webbe: Excellent comment, Angus. [50:30] Angus Lee Forbes: Whereas like [50:30] Matt: I mean, [50:30] Angus Lee Forbes: the more you start learning, the more you get boxed in. The thing though is that if you really do get on a [50:36] Matt: you [50:36] Angus Lee Forbes: good [50:37] Matt: can get a [50:37] Angus Lee Forbes: win [50:37] Matt: little bit more than [50:37] Angus Lee Forbes: with [50:37] Matt: a little bit. [50:37] Angus Lee Forbes: that strong ego, you can actually build some good things now. Like it's so back to your first point: is uni going to be relevant? 10 years time, not for a job. Maybe if you want to be an academic. [50:58] Richard (Syllogism ) Webbe: I think I think The same reason a lot of people get a degree, but they never finish in the career that their degree was in. I think, Angus, the same rules will apply. People will want to see that someone has applied themselves to something in their junior years and completed it. Whether it was, you know, becoming an Olympian and traveling the world, whether it was a running race, whether it was uh climbing a mountain or finishing a degree, I think we will have new ways of assessing people's commitment uh to doing things. But yeah, I'm not necessarily sure. Sure, the actual education is going to be important, but the risk there is like none of us know how to grow vegetables anymore, right? And and we're killing our sovereign security in food, in money, in gas, in energy. And it would be a shame for it to kill off our sovereign intellectual capabilities by allowing AI to take over everything. I think that's a real big risk. [51:55] Angus Lee Forbes: Yeah. I mean, but that's the thing, right? Like it's an economy. Once it's More efficient, once it's easier, [52:02] Richard (Syllogism ) Webbe: Very [52:02] Angus Lee Forbes: it will, [52:02] Richard (Syllogism ) Webbe: true. [52:03] Angus Lee Forbes: it will, it will. Like that's the thing. Like back in hunter-gatherer days, farming, farming wasn't better, it just helped you make a bigger society [52:13] Richard (Syllogism ) Webbe: Yeah. [52:13] Angus Lee Forbes: so you could murder the hunter-gatherers. [52:16] Richard (Syllogism ) Webbe: Murder more people, yes. Very good point, Angus. [52:20] Dave Pengelley: Wow. Wow, I didn't I didn't expect us to wrap up on murdering. [52:22] Richard (Syllogism ) Webbe: That escalated really quickly. [52:25] Dave Pengelley: But I I think I think Angus, you're talking about the Dunning Kruger effect, right? Where where you think you know a lot. Where you actually [52:30] Angus Lee Forbes: Yeah. [52:30] Dave Pengelley: don't know as much, and the more [52:31] Angus Lee Forbes: Yep. [52:31] Dave Pengelley: you know, the more you realize you don't [52:33] Angus Lee Forbes: Yep. [52:33] Dave Pengelley: know. Um, but it and it is that thing where AI can be a huge enabler and you can learn things and do things. Like I couldn't build my SAS myself, not for a long, long time. It would have taken a lot of effort to try and build it to what it is, uh, let alone, you know, but understanding what I want to do, understanding some of those first principles and knowing things and you know, knowing guys like Matt and being able to bounce ideas off him around some of the structural uh design ideas that I got from uh that community was super useful. Um But I think you know, you do get those innovations where you don't know things and you come in and you bring you come in from a an external industry or or different point of view and go, well, why are we doing that? It's that breaking the orthodoxies and and and challenging the orthodoxies, it's critical. But sometimes, you know, the orthodoxies are uh orthodox and not everything should [53:20] Angus Lee Forbes: Yeah. [53:20] Dave Pengelley: be broken just because [53:21] Angus Lee Forbes: Yep. [53:21] Dave Pengelley: you can. And [53:22] Angus Lee Forbes: Yep. [53:22] Dave Pengelley: we do live in a society where you know everything is up for grabs, apparently, and nothing that has ever been done before was right for any reason. And people think that nothing historic like we can change everything when sometimes there's some things that are in place because that's just actually a better way for those things to be in place. This is not [53:40] Matt: Kind [53:40] Dave Pengelley: a political [53:40] Matt: of [53:41] Dave Pengelley: podcast. [53:42] Angus Lee Forbes: Hashtag [53:43] Matt: that's like the midwit meme. You know, do it this way. No, do all these crazy technical things. No, just do it this way. [53:52] Richard (Syllogism ) Webbe: Yeah, yeah. [53:52] Dave Pengelley: Yes, yes, there's a there's a lot going on. Um but meanwhile, the the technology keeps rolling and we have Gemini's bringing in out their new models. Not that anyone ever talks about Gemini anymore, except the anti-gravity diehards that are still in anti-gravity, but always complaining about it because it's disillusioned and not doing what it should do. Like yeah, I was very disappointed. Anti-gravity at the start of the year was an incredible harness that was ahead of its time and it just Google did not keep up, which is weird. You'd think they've got the resources to keep up. [54:21] Matt: Yeah, it's strange. Like the sentiment I keep hearing from everyone in the in the space is, you know, what are they doing? [54:29] Dave Pengelley: Yeah. [54:29] Matt: Like Are they actually struggling? Because if so, they probably should be public about it and not just be this company that has so much potential that seems unrealized. [54:40] Dave Pengelley: Yeah, meanwhile they're rebranding, and now it's not notebook LM, it's Gemini Notebook or something. Like they're just [54:46] Matt: Yeah. [54:46] Dave Pengelley: they're bringing some of their different experiments under the one umbrella, but um yeah, weird. [54:51] Matt: I think the [54:51] Dave Pengelley: Weird. [54:51] Matt: other like question though, like is do they need to have a state-of-the-art model? You know, Google already is good at doing Google things. Like, do we need gen. And I to to compete with Fable, like what benefit would that provide? [55:05] Dave Pengelley: Well [55:05] Richard (Syllogism ) Webbe: Commercial Commercial [55:06] Dave Pengelley: yeah. [55:07] Richard (Syllogism ) Webbe: benefit. [55:07] Dave Pengelley: They're they're fully vertically integrated, right? They own their own data centers, they own their own hardware stacks, they own the software. Like they they can just like and just you know get those cost savings by condensing everything. [55:19] Angus Lee Forbes: Yeah. [55:19] Dave Pengelley: Um they've got the operating system, they've got the the the productivity work work OS stuff like um that you're you're competing with now, Angus, around, you know, document. Yeah, [55:29] Angus Lee Forbes: Yeah. [55:29] Dave Pengelley: right. That's [55:30] Angus Lee Forbes: Yeah. [55:30] Dave Pengelley: that's a that's [55:30] Angus Lee Forbes: Watch [55:31] Dave Pengelley: a [55:31] Angus Lee Forbes: them [55:31] Dave Pengelley: a [55:31] Angus Lee Forbes: downrate [55:31] Dave Pengelley: statement. [55:32] Angus Lee Forbes: me. [55:32] Dave Pengelley: Um [55:34] Dave Pengelley: you know they they've got that that the offer suite stuff that they they they're going with as well. Um they don't have CRM. I w we always talked about when I was a Salesforce, like the of all the big tech, the one that made the most sense to acquire someone like Salesforce would be someone like Google because there's so little compete overlap. Um they don't they don't have that business side of things, but they've got all the ads and the marketing and the cert. Like you're right, they they they do a lot. But Know the alphabet, they want to do A to Z and have a bit of everything, I suppose. But um, yeah, meanwhile, Kimi K3, we we talked about with the the Mac OS ripoff. They've got so many people trying to get it that they've said we're sold out, apparently. Um did you get on the K3 train before they sold out, Matt? [56:20] Matt: Uh I've got K three through droid, so like factory AI and uh through the Droid desktop app, the Droid CLI. Um that's the cool thing about factory. Where they expose all of the best possible choices from all the different labs and they get their own sort of, it's all US-based inference, but it's very, very like good inference. But honestly, I haven't really needed to use it. I don't know if that is like a a good or a bad thing. Um I'm glad it exists, but yeah, I haven't needed [56:50] Dave Pengelley: Yeah, [56:50] Matt: to. [56:51] Dave Pengelley: here we are. Join wait list. So two days ago, um when that Mac OS thing came out and I had a look and I was like, oh maybe I should get Kimmy now. Like I know I'm not using it actively enough. Burning my existing tokens. I don't need to buy more tokens right now. So I left it. Now I'm like, I've got FOMO. I could have now I can't even get it if I wanted it. Um, yeah, join wait list. Which, you know, I think credit to them. Then they're not just sort of like taking people's money and then not being able to serve them. Uh, they're going, you know what? They're and my angus from a marketing point of view, this is also a bit of hype marketing, too, right? Like that limiting access and creating [57:25] Angus Lee Forbes: Hundred percent. [57:26] Dave Pengelley: demand. [57:26] Angus Lee Forbes: Hundred percent. Best thing that happened with Fab. [57:30] Dave Pengelley: Yeah. [57:31] Angus Lee Forbes: It's mythos. You couldn't use it, it's too powerful. [57:34] Matt: Yep. That's what I was saying. Like I might have been wearing my tinfoil hat at the time, but I kept saying, you know, this feels orchestrated because how could you get better marketing than this? [57:45] Dave Pengelley: Yeah, yeah, it's uh organic marketing at its peak. Peak [57:49] Matt: It was [57:49] Dave Pengelley: organic [57:49] Matt: even um [57:50] Dave Pengelley: marketing. [57:50] Matt: was it the gentleman that came on our show a little while ago? He was from India and he he said that even the the movies, like the government used to shut down movies in India. And say you can't watch this movie, it's banned. And [58:02] Dave Pengelley: Yeah, [58:02] Matt: then like two [58:02] Dave Pengelley: yeah. [58:02] Matt: weeks later they would release it and be like, actually it's all right. [58:06] Dave Pengelley: Yeah, that was Sav. Savio was telling us that that story. [58:09] Matt: Yeah. [58:10] Dave Pengelley: Um yes, yes. No, it's a it's a it's a wild world going in. And I mean, you know, it's getting crazy if this things like this are legitimate ideas that are happening where they're talking about, [58:20] Angus Lee Forbes: Yes, [58:20] Dave Pengelley: you know. [58:20] Angus Lee Forbes: I saw that. [58:21] Richard (Syllogism ) Webbe: What is that? [58:22] Dave Pengelley: Um yeah, yeah. The the idea is that that way you can talk to your AIs without disturbing everyone else. There's a cone of silence. Personal kind of silence. Uh [58:31] Richard (Syllogism ) Webbe: I think that is fabulous. [58:33] Dave Pengelley: which I mean, I I remember back when I was working in an open office environment and people were doing transcription and and things like that, or talking about all this voice stuff. I was like, how would these shows, these Star Trek and things, work in real life? You couldn't have an office full of people talking to their computers. It just is not feasible. [58:48] Richard (Syllogism ) Webbe: Yeah. [58:48] Dave Pengelley: Um [58:49] Richard (Syllogism ) Webbe: Oh yeah, no, absolutely. I was um, yeah, now I walk around my house and all my little AI assistants and that, you know, I might be telling a joke and they all wake up at once. It's becoming annoying. [59:01] Dave Pengelley: Um [59:01] Richard (Syllogism ) Webbe: And all [59:02] Dave Pengelley: but [59:02] Richard (Syllogism ) Webbe: I was [59:02] Dave Pengelley: it's [59:02] Richard (Syllogism ) Webbe: doing was telling a joke. [59:04] Dave Pengelley: definite definitely an interesting time that we are living in. This is a this is a strange time, one. [59:08] Richard (Syllogism ) Webbe: Can I [59:09] Dave Pengelley: Um [59:09] Richard (Syllogism ) Webbe: ask a question of the Brains Trust? [59:12] Dave Pengelley: yes. [59:12] Richard (Syllogism ) Webbe: Considering [59:12] Dave Pengelley: And then [59:13] Richard (Syllogism ) Webbe: companies like Kimia owned by, you know, what is it, Moonstruck or whatever, the AI company in China, are we worried from a sovereign point of view about which countries own which companies? [59:28] Matt: I think you would be so [59:29] Dave Pengelley: we're able be do gonna that. to [59:30] Matt: Not to consider it, but I don't really know what the you've got to go, you've got to go to the extreme. You've got to go, okay, so what is the actual problem if something goes wrong? Uh, does it mean that it's surveillance for them? Like they're they're gathering data to somehow make some geopolitical move? You know, is it some sort of leverage that they're gonna have against us? You know, are they uh assessing how to target certain products to certain markets, you know, similar to what BYD did in the sense of the CCP just pumping. them so that they can go penetrate other markets and [01:00:02] Richard (Syllogism ) Webbe: Yep, [01:00:02] Matt: then become [01:00:03] Richard (Syllogism ) Webbe: yep. [01:00:03] Matt: really sticky. So you know, is that a problem in the long run? Or is it that we just get some sort of commoditization and uh oversaturation of a particular market? Or is it that we get addicted to their particular uh supply and then [01:00:22] Richard (Syllogism ) Webbe: And [01:00:23] Matt: you [01:00:23] Richard (Syllogism ) Webbe: then [01:00:23] Matt: know [01:00:23] Richard (Syllogism ) Webbe: they [01:00:23] Matt: that [01:00:23] Richard (Syllogism ) Webbe: start [01:00:23] Matt: gets [01:00:23] Richard (Syllogism ) Webbe: to inject [01:00:24] Matt: fed [01:00:24] Richard (Syllogism ) Webbe: political [01:00:24] Matt: off? [01:00:24] Richard (Syllogism ) Webbe: bias or something into a their responses. [01:00:27] Matt: Uh yeah, it could be it could be a bias thing. Again, Like what's the agenda? Like what what do you want BIOS to achieve? Uh so you've just got to think about all these different eventualities that could be true. And what's the problem? It's like when people say that they want their business to have completely Australian residency, um, you know, whether it's on prem or just at least on the on the continent. Um, w why? It's because if it gets cut off, they, you know, don't lose it. You know, it the imagine if you never let your data get stolen. Per se, but at least that like the lights switch off, like that's a problem for running a business. Um, that would be the only thing that I would think is uh is the issue. It might be a little bit naive, but I think that that, you know, there's not really many other cases that make sense. [01:01:14] Richard (Syllogism ) Webbe: No, no. Good answer, man. Thanks. [01:01:18] Angus Lee Forbes: Yeah. [01:01:18] Matt: That's [01:01:18] Dave Pengelley: And [01:01:18] Matt: all [01:01:18] Dave Pengelley: that's [01:01:18] Matt: right. [01:01:19] Dave Pengelley: that's that's that's like when you use Facebook or you get retargeting and anything like these lot like this, like I I get for certain a elements and and certain data tracking there there's elements where maybe you don't want That data leaking and people to know those things or that information to get out, but your general prompts and stuff that you're asking to rewrite your CV or do other things, is there a significant risk to that going to China? What are the what are the the China CCP gonna do with it with that information? [01:01:44] Angus Lee Forbes: I I I don't know, man, because I reckon once it's there and it's out of your control, it's [01:01:51] Dave Pengelley: Well, yeah, [01:01:52] Angus Lee Forbes: too [01:01:52] Dave Pengelley: you're [01:01:52] Angus Lee Forbes: late. [01:01:52] Dave Pengelley: you're [01:01:52] Angus Lee Forbes: Like [01:01:52] Dave Pengelley: yeah. [01:01:52] Angus Lee Forbes: you you you you lose an authority over it. And that's the kind of I genuinely think though from a consumer. Zumagrade perspective, usually it's probably not terrible unless [01:02:03] Dave Pengelley: Yes. [01:02:04] Angus Lee Forbes: you get a really nefarious actor into government. Um but on the kind of governmental, and I know it's not a political podcast, but that's the thing that they should definitely be thinking about is that kind of like because if if someone goes to war with someone, all those assumptions, like just in time inventory management, go out the window, right? And it's like just because you can buy your oil cheap from someone doesn't mean you should forever without a backup plan if it all went to shit. [01:02:41] Dave Pengelley: Supply [01:02:41] Angus Lee Forbes: It's [01:02:42] Dave Pengelley: chain [01:02:42] Angus Lee Forbes: like [01:02:42] Dave Pengelley: dependency is i is a huge issue, right? And there is obviously sensitive data and personal data and and and things that from government and other business organizations and the B to B risk and and you mentioned things like BYD and stuff, Matt. Like um there are less regulated countries that will willy happily go and you know. know, replicate other things and then undercut the market and try and take over and and sell them at a at a massive discount in order to you know put out other people out of business. And so then they become the only option, the de facto standard, and then that's a supply chain risk. So yeah, I think what you're saying then, Angus, around that, you know, where are your supplies coming from and and are you engineered against that risk? I mean that's why it's good to have you know AWS data centers in Australia that are running bedrock models and thing so that way we do have that level of you know within our borders capacity to run some of these things should we need to. Um that's important. And and we talk about this week after week as well. Like if for individual businesses and and what are those critical processes that if the lights go off, how long can you last without that system running? Is it for a day, for a week, for a month? Um and if it's untenable for you to have that level of risk, then what are what is your um business continuity plan? What is your disaster recovery plan? Is that Right software stack for you to be using, if that's the risk. Uh, which is which is why people should reach out to people like us to say, hey, help me think about AI and my tooling and what we should be doing. And is this the right software stack and the right solution for us so we can help you assess and understand all of those things? Um, we are over time. Thank you, Angus. Great having you [01:04:22] Angus Lee Forbes: Yeah, [01:04:22] Dave Pengelley: on the [01:04:23] Angus Lee Forbes: thank [01:04:23] Dave Pengelley: show. [01:04:23] Angus Lee Forbes: you. Thanks for having me. [01:04:24] Richard (Syllogism ) Webbe: Come [01:04:24] Angus Lee Forbes: This [01:04:24] Richard (Syllogism ) Webbe: back, [01:04:24] Angus Lee Forbes: is fun. [01:04:25] Richard (Syllogism ) Webbe: I guess you're good. [01:04:26] Angus Lee Forbes: Oh. [01:04:26] Dave Pengelley: Um [01:04:28] Angus Lee Forbes: I appreciate that. [01:04:30] Richard (Syllogism ) Webbe: Devo hat too, by the [01:04:31] Angus Lee Forbes: Oh, [01:04:31] Richard (Syllogism ) Webbe: way. [01:04:31] Angus Lee Forbes: good, good, good, good. [01:04:33] Dave Pengelley: Yep. So if you want to uh [01:04:35] Richard (Syllogism ) Webbe: Go [01:04:35] Dave Pengelley: have [01:04:35] Richard (Syllogism ) Webbe: to LinkedIn [01:04:35] Dave Pengelley: secure [01:04:36] Richard (Syllogism ) Webbe: and see Angus's Devo hat, sorry. [01:04:38] Dave Pengelley: secure stuff from processing of your PDFs and other files, check out fwip.app [01:04:43] Angus Lee Forbes: Yeah, thank you. [01:04:44] Dave Pengelley: where [01:04:44] Angus Lee Forbes: Appreciate [01:04:45] Dave Pengelley: you [01:04:45] Angus Lee Forbes: it, [01:04:45] Dave Pengelley: can [01:04:45] Angus Lee Forbes: guys. [01:04:46] Dave Pengelley: um fwip it good. [01:04:48] Angus Lee Forbes: We have a good, [01:04:49] Dave Pengelley: fwip [01:04:49] Angus Lee Forbes: all right. [01:04:49] Dave Pengelley: it good. [01:04:50] Richard (Syllogism ) Webbe: There, Dave. [01:04:51] Dave Pengelley: Well that that you know you you mentioned Devo and Angus said that was that was part of the inspiration, right? So um yep, check out fwip.app. uh to support a great Aussie business, a great Aussie innovator, and uh get, you know, cheap, secure ways of managing yourself. The PDF thing is a great example. Like installing Acrobat and then getting all the pop-ups [01:05:13] Angus Lee Forbes: Oh [01:05:13] Dave Pengelley: and all the [01:05:14] Angus Lee Forbes: my [01:05:14] Dave Pengelley: bits [01:05:14] Angus Lee Forbes: god, [01:05:14] Dave Pengelley: and pieces, [01:05:14] Angus Lee Forbes: isn't [01:05:14] Dave Pengelley: even though [01:05:15] Angus Lee Forbes: that [01:05:15] Dave Pengelley: you can [01:05:15] Angus Lee Forbes: the [01:05:15] Dave Pengelley: do some [01:05:15] Angus Lee Forbes: worst? [01:05:17] Dave Pengelley: like everything just takes so much processing power. It's ridiculous. [01:05:21] Angus Lee Forbes: Doesn't need it. [01:05:22] Dave Pengelley: Yeah. [01:05:22] Angus Lee Forbes: All right, all right. Well, I know we're over time. But I appreciate it. So thank you very [01:05:27] Dave Pengelley: Not [01:05:28] Angus Lee Forbes: much. [01:05:28] Dave Pengelley: at all, mate. We'll put a link in the description of the video afterwards. But uh thank you, Angus. Thank you, Matt. Thank you, Richard. Thank you, Australia and around the world, wherever you're catching this from. And we'll see you next time. [01:05:40] Richard (Syllogism ) Webbe: Thanks, [01:05:40] Dave Pengelley: See you [01:05:41] Richard (Syllogism ) Webbe: Dave. [01:05:41] Dave Pengelley: boys. [01:05:41] Angus Lee Forbes: Bye [01:05:41] Richard (Syllogism ) Webbe: Thanks, [01:05:41] Angus Lee Forbes: bye. [01:05:42] Richard (Syllogism ) Webbe: guys. Well done.