Read the full transcript
JR Reynolds: Hey folks, it's another episode of Yelling at Robots. Let's get right into it. It's a big juicy pile today, so without further ado, Yelling at Robots. Action! Get ready. It's time for world famous Yelling at Robots. Starring tech legend Joshua Raphael Reynolds and Ian Thelonious Bosberry. The show starts now. Fawz, welcome. Welcome, friend. Welcome, Yui. It always does feel weird that you're doing all the welcoming and the questioning, although I guess that's just our, that's our dynamic so far. reason, dear Fawzworth, that I think I welcome you is actually because I start the recording. I think that's really the only reason. Yeah, that might be it. Also, you just called me Fawzworth. You know how many times I've had to answer. Is your middle name Thelonious? ⁓ that makes me really happy. Just and just a throw away. Just so throw away. But yeah, it's working. This is the first thing that popped in your brain. Yeah. Important news from the podosphere. Fosworth. ⁓ We are the number 25th technology podcast in the country of Finland. What? are 25th. Adams Adams routing his VPN through Finland for privacy reasons. Of course. mean, to be honest, this is probably something we could exploit to juice our numbers. So what we do, Fuzz, is we get all of our friends VPNs if they don't already have them and we have them all pick a country, any country. The more like technologically unadvanced, probably the better because it will be ⁓ easier to juice the market. Now we're thinking like creators. What are we talking about today, Josh? ⁓ So we'll have a bakery update. I met with Thomas from Also, shout out to Thomas, one of our early listeners, who was, I think, a little bit surprised to hear how much I've talked about him. But he was cool. He's cool with it, man. He's cool with it. We'll talk about the great Claude Shakedown. So this is... There are several things happening, but most important... Yes. Deeper! Go was legit adjustment. That was just a legit adjustment. Definitely staying in. It is, but it's a nice adjustment. Oh, great. mean, man, our production value is just going up day by day. Wait till I get that microphone. You're just a treasure for playing along with this. I'm asking Fuzz to get a third microphone. will take some, not the majority of the responsibility on this one. I used to do this professionally. I once worked as a sound engineer. It's my dumbass fault for not remembering the difference between condensers and dynamic mics. ⁓ Although I was told at work that this mic is giving Larry King vibes. ⁓ like the look of it. It's got a great look. ⁓ yeah. No, it's a beautiful mic. Absolutely. I would love it if you could like pull it down from the, you know, just kind of... ⁓ like we're starting a boxing match. ⁓ let's get ready to rumble! Absolutely. Anyways. The Great Claude Shake it Down both... some continued degradation in people's experience using Claude. But most importantly, the fact that I ran out of tokens in the middle of generating a slide deck and then was asked to pay money to finish it and I paid and I'm pissed. Trend slop, the dangers of reasoning, the dangers of thinking that these models are reasoning. And then we're going to vibe code our website. All right, I'm starting the timer. I've asked Fuzz to try an experiment. today in our recording and we'll see. We'd love to hear your feedback. Also, don't forget to hit that like and subscribe button. Share this with your friends. We'd love your feedback on the quality of this episode because we are trying something new, which is to limit the amount of time that we ramble on any particular topic. Now, this is an experiment. We'll see what happens and I'm curious and I don't want to know. Surprise me, Fosworth. What will happen when the timer goes off? Like will he make a Like, there be a sound? Will he wave his hands? I don't know. Let's I'll give you a warning. I will tell you right now, you are cutting well into your bakery time. Did you start the timer? I said I was starting the timer. I'm gonna cut that out and then you're gonna look like an asshole. First on the docket, Barteen Bakery Update. Bakery update. Today I met with Thomas from Bartine Bakery and went into great detail about the application. he show you the app or apps or anything? Did he show you anything? He did. I mean, the thing's done. The thing is pretty much done. It's done. Oh yeah. Thomas is the he's essentially kind of the operations manager for Bartine. Bartine is a bakery in the Hague. Sourdough Bakery as Faz mentioned. They How do I wanna, like how much time is left? How much time do I have left? This is good. One minute, 30 second warning. I'm gonna need to, okay, okay. We're gonna do, we're gonna do what we, so Josh and I are also part of the thing we learned to do well at our previous job at Reactor was facilitate workshops or be a good facilitator. We're gonna do whatever a good facilitator is gonna do. We're gonna add more time. We're going to just add more time because it's Does anyone need more time? OK, we'll add another five minutes. Thomas, the operations manager who's leading kind of the effort of building this app, got to a place they wanted feedback. He wanted to talk about it. And as Faz mentioned, throughout the course of this, they've been asking extremely software development, technical questions, interesting stuff. And I've really been, and it really stands, I'm super curious to see how this evolves. again this is for me a perfect use case. Now when I sat down and we talked I heard about a not so perfect use case which is they are drunk on power. Like they are drunk on software development power because I said yes I mean it was it was something to see because they like he's like okay cool yeah so you know we We started off by kind of talking about and I asked what he wanted to get out of the session and it's like, well, he wanted to hear about like what they should be looking out for. How do they go into production or how do they roll this out? What to, you know, just to get feedback on the process, stuff like that. And then he told me that like as they're doing this, they're already beginning to plan and develop their payment processing application for a new location that they're opening. We spoke about this. We spoke about a payment processing. ⁓ about being drunk on power? No, not being drunk on power, about what are they building? What are the things? Because you, you, did speak about this in previous episode where you were like, ⁓ they're doing, they're going to do this. And I, I, my first, one of my first questions was like, they're not building a POS or something like that. Right. And you were like, no, of course not. ⁓ my fudge. Yes. Yes. Yes. Yes. And so indeed they are. And so poor, poor Thomas, like I literally spent the next 15 minutes just categorically like just categorically being like here are the 18 reasons why this is absolutely bad sh** insane and obviously I didn't say it like that but ⁓ it is in fact absolutely bad sh** insane and they're drunk on power. think when I when it when what I was talking about really hit home with with with Thomas when I started when I started talking about compliance And see, this is my first thing. It was like the I built like a credit card processing. Oh, you did for my startup. Yeah. Yeah. Yeah. Before we like got into text and all that stuff. Sure, sure, So I built like an like a checkout and a cart and everything. And I had to go through like PCI and all that stuff in America. I can only imagine what it's like in Europe. know what it's in Europe. It is like this will be your full time job and more so because You are taking on so much potential liability. I talked him off the ledge. I'm quite sure this isn't going to happen. ⁓ You put the fear god into him. I really did my best, I did my best because it is... fear of the AI gods. It really is. And it's like, and it really did get a sense of like, there's nothing we can't do! ⁓ And more power to him, maybe we just want to dial that back. And so the domain itself, which is why it's a really good use case is pretty narrow and well scoped. Um, now it has a UI, but it's the same thing where it's like, fill out what you want in production and then it back calculates all the ingredients you need and timings and things like that. And it's great. The things that were missing were tests, any testing. kind of, and, and I know, you know, I didn't see anything unlovable for how do you have test automation or validation that this thing still works the way you think it should when you make change. Right? Yes. Yeah. So they're using they're using lovable to do this. Yes. Correct. And and that that's not like a default. As far as I can tell, there's nothing. There's nothing. What the **** you doing out there, man? I don't know. I could be wrong, but we Google because I think it's just I think it is just OK. So lovable is not like cursor. It's like more. It is a full it's a platform that is designed for creating really web applications. So, but it, it's like, it's, like, ⁓ it's visually driven, not it's visually driven. is a platform itself. So it also hosts and deploys and does all the things. So you don't have to do or manage any of that. my overarching takeaway is the tools are there to some extent. So they, they were able to create what looks like a pretty good. tool that meets their need a pretty good piece of software that meets their needs. The environment they're working in lovable. don't know other ones are seem appear to do be doing an absolutely shit job of providing the level of context and support for how to actually make software and use it safely and a lot of my time was spent just trying to transmit some of that experience and not just technical but also kind of like You know, this was he and his partner building its vision. They haven't done any user tests. No, they'd they had, they had showed it to one of the bakers, but they hadn't done user testing. They hadn't done user research. They'd never heard of a user story. They had never heard of kind of some of the things of like, well, what does it take to actually make a piece of software successful? So they're on the path, but, but he was also super into it, super appreciative. They're on the path. It's happening. They're on the path. I think that the ecosystem is pretty shitty to help people. to effectively help people along the path. But yeah, it's exciting. We'll see what happens. I think they'll go into production sometime in the next few weeks and I'll bring an update then. How insightful. Up next, Trendslash. So we wanted to talk about Trenslop. Josh shared yet another research thing with me, which I did not read any papers this time around. But Trenslop was we did a sh** job of preparing this time. What we did do? Yeah. ⁓ No, no, we did. It's fine. Sure. Sure. Sure. Sure. I was reading a different paper about the mathematical proof on why AI layoffs are going to crash the world economy. the economy. Right. Right. Yeah. It's proven mathematically. That's right. spoiler alert, we can't stop the train, guys. We're f***ed. Perfect. Why? But see, that's liberating. Stop trying. We're free. We're free. I mean, it's like there's a freedom in powerlessness. know? Anyways, but that's for another episode. Maybe the last episode before we all die. It's the last? The last episode is just... Alright folks, we're signing off. The proofs are in. As you can see, the flames have erupted outside. We all saw this coming. If you'd read the paper, We've been warning you about this for... why two months it's taken to get here. But in the meantime, trend slop, trend slop is another great use of the word slop, slop squatting, trend slop, I like a lot of the words. It's so versatile, slop. Slop is such a versatile word. Slop's squatting, trend slop, pop slop, slop and lock. But trend slop is the idea. So researchers asking LLMs for strategic advice around things, turns out that they are very biased towards surprise, surprise. trendy managerial buzzwords out there in the world. ⁓ So if you are asking about strategies on a specific context, you're not very likely to get good answers, but you will get what seemed like good answers because it'll look super intelligent and it'll look like it's really reasoning around things. And you'll feel like, wow, fuck, this thing is really my boss is gonna love this PowerPoint. It's so smart. Yeah. Yeah. There was an article or a study that came out last year about this, about how to, you can easily identify AI generated things just by the words that it's using, because those words are so ubiquitous. it is a, it's a, ⁓ it's one of the words ubiquitous. It might be. ⁓ It's called fuzz slap. It's the newest thing raging across the AI landscape. It's such a. Yeah, because it's a self-reinforcing loop that the more that stuff gets out there, the more that it's retrained on the stuff and then it's like, it's just garbage in, garbage out over and over and over. And this is really interesting. I think that part of the reason of the Translop is, or part of the reason it's going to get worse and worse is that for your average person, I'm very average. You're using these AI tools and because they're anthropomorphic. You have this false sense of this thing is actually reasoning. You can still benefit from kind of helping your strategy or helping your reason through things, but your framing matters a lot. you get much better outcomes and output from this if you do the strategic thinking yourself, but you ask for thinking support. So you're asking for things like, hey, I'm thinking about taking this strategy. tell me all the things that are wrong with that idea. That's using it in a different way that you're still benefiting from it being trained on a large corpus of data, and now its mission is not to pick amongst all these things and it's leaning towards the most common thing that's always said. It has a specific thing to work against, and now it's leaning towards all of the arguments against whatever you've said. And that's a different way to use it where you can still benefit from some thinking assistance, but in a way that's not. so biased, that's not so polluted by bias. The other thing I was going to say was I had a friend at work who she just got back from a five day in-person course, five days where they learn how these things work. They actually get their own model. They fine tune it. They add rags. They deploy it. They do all that stuff. It's called Deep Atlas. It seemed really cool. really, I would love to do it. It's not cheap. ⁓ Faz, let's do it. Let's do it together. Let's go together. We're going to camp, folks. You heard it here first. Yelling at robots goes to camp. ⁓ it's gonna be just like Jason Voorhees in Friday the 13th. It's like sleepover camp for yelling at robots, except instead of teenagers, it's old men. I think it would really help a lot of people. ⁓ absolutely. understand how they work. Like at a high enough level or a low enough level that you're not, it's not so mystic anymore for you so that you can start to just realize it's a tool and it's technology and it's not fire given to us by the gods. God, but when you say that, man, it's also like, people don't even take, I don't know if we don't make, like people don't understand email or how email works. People don't understand messaging or how messaging works. People definitely don't understand websites. It's like, ⁓ I've got, just, well, but I got a pop-up that said that I needed to install an antivirus. The internet told me, so I think we're kind of on that front. We're fucked, once again. Damn, that was so funny and smart. Up next, the Great Claude Shakedown. Up next, the Great Claude Shakedown, or how I was wronged by the internet. Actually, how I was wronged by the sh**birds of Anthropic. He's still willing to take an interview, though, Anthropic, if you are interested, just drop me a line. know, I am totally comfortable. just renouncing any and all statements I've ever made if it benefits me personally. So just know that. And I'll bring, that's the kind of ethics I'll bring to Anthropic, which I think are gonna line up based on what I've seen. So what happened last night, I had the big presentation this morning that I was working on, and one of the things I've always hated is making fucking slides. Making slides sucks, especially if you have to use a corporate template. And guess what? Guess what Claude has? a slide skill and it can make your slides for you. And I had done all of the work. I'd written this kind of strategic briefing paper and all I needed to do was translate it into slides, make bullet points, those things. It's like the dumbest thing you can think of. The slide skill will also take the corporate template and use it? Yes. And do it well? And not f***ed up? It f***ed up one slide, not, I mean, much, better. one slide out of like 20. Who gives a s***? That gets me to like, you gotta use this thing, man. It's so good. You get f***ing, oh man, it's so good. You never feel so good. It really was, it was a drug deal, man. I got f**king squeezed. The first taste is free and then you gotta pay. So what happened is, you know, and this was a big presentation. I worked late because this was important and I really wanted to do it well. I'd done all this work on a strategy paper and then posted it to the cloud and like, hey, make a slide deck out of this. I'm not concerned about its ability to extract bullet points because that's also how I communicate and kind of like here are the key points, blah, Midway through it's kind of like, ⁓ yeah, cool. All right, cool. This is kind of what I'm planning. Here's the thing. I'm looking at your template. Okay, I've got the template looked at and it's done. You know, like doing its thing and it's like, all right, starting to generate the slide slide one slide to slide three out of tokens. ⁓ you're out of tokens. ⁓ Your tokens were new at 11 PM central time or something. So it's like my tokens are not gonna renew before my presentation and a pop-up that says, would you like to purchase some extra usage credits? Yeah. And I did. I did. It shook me down and I paid. That's it. And that's it. That's what we talked about. You bought yourself some intelligence. I paid. mean, this is Jensen Huang's dream come true. You know, it's like, gotta keep the tap flowing. So I had run out of tokens earlier in the day. I was preparing information for the slide and I just wanted to do some like, Jira calculations, pull some ticks out of Jira and do some s***. You other things that I hate. A couple things happened out of that. I out of tokens early in the day and I'm like, I don't know how to work anymore. Like, I literally- That was what tickled me. This is what I shared with my team because you texted me when I was at work and I just started laughing out loud from looking at my phone because I got a text from Josh and it said, I ran out of like, well first of all, it was like a big rant about f*** these guys. these guys. And then just like a bit of a pause. And then he's like, I think I've also realized I don't know how to do my job. It felt like that. It really felt like that. Like it has become and I consider my AI usage to be pretty pedestrian. Yeah. Like just just reasonable. You know, I'm mindful of the things that we've talked about about skill erosion. being mindful like the strategy paper I wrote, didn't use AI. I didn't use AI at all initially. Like I just wanted to do some like hard thinking and then I'm sure of course I brought AI to blah, blah, But yeah, then when the tokens are completely cut off, it's like, I don't even know how to get to the page I want to get to in Confluence. It's like, I got to open up Confluence and like try to twirl down this garbage. Yeah. So that was a bit of an eye opener. And although in that instance, I was like, I'm not paying for credits. can wait till 6 p.m. But then I got my 6 p.m. fix and like by the way that and I'm sure this is no mistake the PowerPoint skill does burn through your cache of tokens by the way. They where the money is middle management. Absolutely. Just the value destroyers. Yeah. So what what my My takeaway from that was, f**k these guys. This is gonna be like this all the f**king time. The way things will likely evolve will be, yeah, we will still use frontier models, but much less. And you as an employee will get 300 credit budget a month, but actually part of your job is gonna be managing your token spend or managing your intelligence budget. using it for really high value tests, then using local, local models, using cheap models, kind of like, how do you map that out? And you're getting a big smile on your face and tell me why. Because I'm going to keep this as anonymous as possible. I can tell you who I'm talking about later, but I, I spoke to someone who spoke to the to spoke to the, and this is probably true at every company right I spoke to someone who spoke to someone who spoke to someone in a frontier no, I spoke to someone who spoke to the CFO at their company who's looking at a giant fucking number being like, show me the return on investment please. Oh yeah, oh sure, sure. And no one can answer that question. CFOs everywhere are fucking losing their shit. Is it finally happening? yeah. Cause people are like, well, ⁓ so are we shipping more features? Are we like getting more stuff out? And they're like, well, no, not really. The beautiful thing about this is that ain't even the real number. That number that these CFOs are seeing, that's not the real number. That's the insanely subsidized number. Because they're all taking the Uber approach right now, which is just like, we'll eat a lot of this cost for the time being until we capture so much of the market. We can up the price of the brick. Anthrophic! Good anthrophic! Good times. Get your fix when it's cheap. Up next, Dino Maw. Okay, Foz, we've just talked about three topics and not attempted to vibe code our webpage using Big Pickle and a Haskell implementation. Any last words for our viewers slash listeners? Have a great weekend, everyone. well, that's nice. Yes, my last words Even though by the time you listen to this, it won't be the weekend, it will have passed. It will be a weekend. It will be a weekend. And whatever that weekend is, enjoy it. My last thoughts prepare for the coming squeeze my friends because the great Claude shakedown continues They're just the first they're coming for your money. They're coming for your tokens So prepare you build out your bunkers get your h200s ready get your solar powered battery packs because the squeeze is on The price of the brick is going up Pandemic and that's a wrap You just survived another episode of yelling at robots. Better luck next time.