s03 e07 The Rise of Vibe Coding

Welcome back to the Big Cheese AI Podcast. I am your host today. I'm here with Brandon Corbin. Hello everybody. And we got Jacob. Yo. Well, let's get it on. Oh, let's go. That music. All right. We are here today. This is March 6th. We had spring or we had winter. We had fake spring. And we had winter again. We had second winter. So now and we're coming out of second winter here, downtown Indianapolis. And you got next weekend is St. Patrick's Day. We just got through Fat Tuesday, which is a really big deal around here. I mean, maybe kind of past Wednesdays a big deal. I know you got your ashes on Wednesday. Yes, I did. But we are absolutely loaded today. So we're going to get it going. And there's been a ton of stuff going on. Today's episode, we're going to talk about our favorite thing in the whole wide world, five coding. Five coding. Five coding alert. So for those of you guys who have never seen the term, you probably lived under a rock for at least the last two weeks. And that's where actually to try and determine is how many people know what this term is. Because it was coined. And it's now becoming an end. You know, I don't know how that happens. But we have this industry standard terminology that comes into play. Somebody just makes something up. Right. I would be curious if we have any idea of where it actually originated from. It was Andre Korkkopf. Gosh, darn it. Why'd you ask me that? It was a tweet. Oh, OK. It was a tweet. And the guy he worked at, he was the head of AI at Tesla. So there's a podcast that I've been listening to. And his name's Alex. And I think he's called the Cosmic Skeptic. Guys are genius. But he had this guy on that was just talking about etymology, where it's basically that social media is now creating the new terms and new words at such an insane clip. And nobody can kind of keep up with it. But like, mid is one. Mid is this mid? Yeah. Which I'm sure we'll hear more about in the future of the podcast. There's just all these new terms that are coming out and vibe is another one that like literally in a couple weeks, all of a sudden, it's a whole fucking new term that like a huge swath of the population knows. It's just it's wild. Anyway, it's like those high-gurking and Carol Baskins. That bitch. I never watched that. You have it. No. OK, so that's one of those ones. And I think you should just try to watch. At least just it is. It's like watching a train wreck. And you're just, you just can't stop. Did Mandip and Jail or something? Yeah, he's totally in jail. Like, no, it's a shit show, dude. But yeah, the tiger king was actually like one. I was like, oh, yeah, that's perfect. So vibe coding is we'll get into it. But we got a few news articles that we wanted to talk about. And I was really unimpressed with the demo I just got. But I know it's like you guys were very impressed. But we've got this new, like we thought we had voice mode. Like I use chat GBT. So if you have the chat GBT app, I think I don't know if it requires a plus subscription or not, which is $20 a month, it does for the advanced voice. So you can have a conversation with chat GBT. And it does feel natural. In mid conversation, you can tell it to talk in a different accent or there'd be a different gender or different type of voice. But it does feel conversational, but you've got something which is this sesame insane voice mode. And I need to know about this. So sesame, I believe I might have just moved your guys' camera a little bit. Now we're good. All right. So sesame.com is where you need to go. You need to try this on your computer because we just tried to do it on the iPad. If I had my headphones on, it worked fine. But if I didn't have the headphones on, I couldn't do it. But this is her. I mean, this, the voice is... So when you say her, you mean the from the movie? The movie where it was, it was Scarlet Johansson just talking naturally. Maya is the female version and Miles is the male version. And it's absolutely insane. So I use every morning, I use Chatchy PT Advanced Voice and I'll go and I go for my walks and I have a conversation about what I'm trying to do for the day or whatever things I want to work on, maybe some like break out some post ideas or whatever. And it's good, but it's still, I mean, it's still very synthetic. Like you feel that, like as you're talking with it, like especially with me, I have a tendency to rant. And then I end and it has to kind of catch up. It's not immediate. It's not instant. So it still feels very conky with Maya, man. It feels like you're just on the phone with this girl who's really into you. I mean, it's almost like they've purposely put some sex appeal into her voice where when you're listening to it, you feel like you have a, or at least I did, I had a visceral reaction where I could feel like I was like, oh my god. Like this is like, if somebody would talk to you like this, you would know that they're kind of into you. And I think that was absolutely by design. Yeah. And so this is Sess Samarit. And yeah, so the one thing I noticed that was really compelling was the breaths that it's her deposits. I think a lot of voice models are trying to do this, but I think they've done it the best. And it just felt very natural. And I asked her about that. I was like, okay, your breath sound very, very natural. And she like emphasized it when I said that. And I was like, oh, that's pretty, pretty impressive. Ian, apparently what they're going to do is they're going to make it so that it can record. And it's basically like a running two week context window of all of your past chats. So for the past two weeks, now it'll be interesting to see because again, that was kind of something that I did with big cheesy application itself. Is that I pretty much just truncated a chunk of time in it, in it works. It keeps your context window small. I'm sure it's a rolling two weeks. It is a rolling two weeks. And but that's kind of what like I played around with that. And the problem is though is that there were things that you talked about a month ago that still is something that you want to be able to bring up. You want to. Before you have your context window with your chat, GBT voice conversation has been. Is it like new every time? Yeah, it's pretty much new every time. And there's a 45 minute max of where you get the advanced research. And usually I run out of it every day. But I will say that even when that context window gets really long, it does start to not just like, because most of the time like I'm like, hey, I want you to write this content, never use M dashes. Right? And by the time that I'm done with my conversation, that context window is huge. It completely ignores it. And it will just be full of the traditional M dashes that it populates. And I'm like, motherfucker, no M dashes is like, oh, I'm so real for go. Boo. So it's it. Yeah. So but the context window fills up pretty quick. Now I don't use Gemini. I don't. And a lot of people are saying that the voice side of what Gemini is doing, especially with the studio LL or the studio LLM. So what's called? No book LLM. And the voices that they have is in those podcasts are really, really good too. But I still think sesame might take the lead here. Yeah, I mean, I agree. Google's voice models are really good. But Sesame's was the most and I agree too. It was very central. I to start talking to her about some pretty risky stuff. I was asking about Elon and racism and stuff like that. And you know, typical LLM fashion, it was very politically correct. And it was like, everything's nuanced. And you have to consider context. And I was like, no, he just he did this. What do you think that means? And she was like, well, the truth is in your own opinion. And I was like, OK, but it was willing to have the conversation. Whereas Google, I'll ask it about something pretty benign. But if it's like construed as medical advice at all, it would be like, remember, I am an LLM, not a doctor, every single response. And so it feels very not human, right? It's not I'm not having a conversation. It's very clear I'm having a conversation with the computer at that point. And this felt very much like even when there was a pushback on, can it answer that or not? It always it prefaces with, well, you know, just keep in mind that things are nuanced. He didn't say, hey, I am LLM, I will not answer furthermore. Yeah. Yeah. So I've been very impressed now that I will give props to X AI. I know I should on on Elon as much as I can, rightfully so. And because he deserves it. Are you transforming my voice? I guess so as messy with you. Sorry. That's why we don't give Sean the controls of the of the buttons. But but they rock three is coming out. A lot of people are very impressed with it. It is seeming to be like a new foundational model. They do have sexy mode. They do have like get rails off kind of mode where you're about to get railed mode. Yeah. What is this? Well, it sounds like the sexy mode will get railed mode. Like I mean, it will go hard. Yeah. So it's like completely unsensitly. Yeah. Did you mean to say it will go? I know. No, but talk about though. Did you guys did you guys hear about though? What? So when it first came out and a lot of people were like, hey, well, you posted one of the things and there were somebody asked is Elon racist. Yeah. And and Grock went on a diretry about how he's racist. And then a couple days later, everybody started noticing that it wouldn't talk about about Elon or Trump. And then all of a sudden comes out. Somebody gets their system prompted or to get the Grock to output its system prompt. And they basically just hard coded in the system prompt. You don't speak ill basically about King Trump or VP Trump. Or King Elon and VP Trump. Yeah. That stuff goes without even mentioning at this point that that stuff's going on. But okay. So check out Sesame. Insane voice mode. I think that what we're seeing is that the game is being leveled up. I've been using Whisperflow, which is a way to do voice dictation. So literally replacing all your inputs on your computer. And what it's doing is it's doing a really good job of picking up exactly what you're saying, removing pauses. You don't have to say like period, question mark, you know, things like that. And I'm literally just prompting AI with this. I mean, the voice stuff is not going anywhere. Well, I think you hit the nail on the head. Like the thing that makes Whisperflow and Sesame's model really, really good is how fast they are. That was awesome. Yeah, we had super fast. And you're right. I was talking to it. And I was, you know, whenever you first are interacting with these voice models, you're a little clunky because it's not used to talking to a thing. So you're trying to talk to like a computer almost at first. And it did a really good job of naturalizing what I was trying to say. It was a text message and it took out all the pauses and all the weird awkwardness that is that happens. And yeah, it was great. So voice, hey, you know, you get, you get mentioned on the big cheese AI podcast. You're doing something right. So we got next, what the future may be app free. And what we're talking about is now, and I'll get into the Apple stock take on this one in a little bit. But we have a new set of tech that's coming out. I think maybe we're talking about post rabbit. We're talking about post, what was the humane pin? Oh, did you hear about the humane pin sidebar? So humane pin got bought by HP. Did we talk about how I got hijacked? Oh, you didn't. So humane pin gets bought by HP. Well, guess what? People wanted to keep using it. And they have there's a discord community around humane pin. Well, guess what? One of the internal employees gave them the essentially the certificates, the way to get or access to its core API. So there's a whole bunch of people hacking the humane pin now and building on top of it. Okay. So this could be the most interesting thing about the humane pin. So again, roll with me here because I'm completely serving at this point. But all of a sudden, you could have a complete open source movement that could be revolutionary because they just said, all right, fuck it. Just release it to the wild and let people go nuts with it. That would be so people are hacking the humane pin and they got some internal person gave max to everything and they started doing cool stuff with it. So it shows you. Yeah. You know, it shows you and I've I've run into this a little bit at work this week. People don't understand to understand development scale very well. Like I got compared our company got compared to Google in a conversation with a client today or two days ago. And I'm like, okay, we have 15 employees. We aren't Google. News flash, right? And neither he was humane pin. And also humane pin was a product company that was started by the Apple people, right? They're their device company, right? And obviously we know what Apple's doing with the software side of AI. But it's like you give this stuff. You give you give a bunch of software guys a piece of hardware. They're going to do something. Yeah. Now, so that's that's fascinating. I had no idea that that was going on. Yeah. So anyways, check that out. But a back to it, the AI phone, the concept of building a phone that's not built by what Google Samsung or Apple, right? What do we got here? So this is the Duchess, Dweech's telecom is working on an AI phone. And that's basically the whole concept they're trying to move away from like this app, the app mentality of what mobile phones are. Now it's still not very clear like exactly what their technology is, but they're basically talking about they're going to incorporate numerous standalone apps, including a unified AI assistant, capable of like booking flights. So a lot of the things that we saw like with the rabbit are one and their large action model, which is bullshit. You know, it's they're partnering with Google Cloud, which is kind of interesting. 11 labs for voice, pick a art for a lot of partnerships, right? Doing a real time object recognition. So they're really trying to bake everything into AI. I can't help, but feel like this is crazy in a way because again, the technology like, yes, AI is, it's a technology and you build it into things and solve problems with it. So what are they going to do that's going to be any different than what Samsung's going to do or what Google's going to do or what Apple's going to do when it comes to the phone? And you know, like if there's anybody who I think can truly revolution, I don't know if it can be a telecom like a big telecom out of Germany that's going to really revolutionize the next play. It still feels like somebody who's going to have a design leader like Johnny I is partnering with like a, you know, with open AI and doing something like from a radically designed differently than what currently exists. So I'm kind of curious what they're doing, but at least they're gunning for it. I'm ready. I'm ready. So as a, as a Apple investor in Apple fanboy, right, you look at what's going on. They've spent so many years and years and years perfecting their UI, right? You're not fooling anybody over there, buddy. It was a lot louder than that. Every single thing that that is, I mean, all these companies, Apple, Google, they all have, they're all their paradigms are built on user interface technology, right? The amount of time that Steve Jobs spent saying you have to, you have to prioritize the UI thread. When I press that button, it needs to put, what if there is no button, right? Like Apple is maintaining the very disruption that Apple created when they invented the iPhone is the very disruption someone else is going to do to them. So you were talking about using a specific voice thing for interfacing with, was it cursor? Were you actually using, were you using the chat for cursor? So there's a, so on Twitter, there's a viral tweet that happened, which is a vibe, good and tweet. And it's a guy who cloned Airbnb in 11 minutes. Right. And with the back end, by the way, and it looks just like it. And he had really good cursor rules, cursor prompting. So he had pre set up, he used his web app starter kit. He had a really good baseline to, to, to do this. But he did it for, he went screenshot to code and set up a database and everything. So one of the, one of the tricks he had for going really quickly is using a, it's a SaaS product. Yeah. Whisper flow. Whisper flow. So WISPR flow, it's, it's, there's no H after the W, it took me forever to find it. Because he didn't link to it or anything in the tweet. He was like, I use Whisper flow and literally you just press the function button and start talking on your computer and it immediately picks up everything you're doing. And any active text input in the entire OS is input with what you said. And you want to hear something funny when I had my surgery, I couldn't type. So I turned on accessibility mode on my laptop and I was trying to do this exact same thing. But the biggest pain points was like, getting it to actually listen to me work. It did. Listen to me when I really wanted it to do it. And then also just the, the filler stuff. Like it was listening to everything I was saying. And there was just no, there was no processing my words. So everything I saw my off. It's the same thing that people talk about with Microsoft. They're like, Microsoft just built this cool new AI thing and then everyone in the propelizes like, but you can't, they like, show a screenshot of Outlook. And they're like, they're search results of Outlook and search results on Outlook are still terrible. Same thing with Apple, right? Like Apple has this huge, huge ecosystem and all these things that, I mean, even there, I bet they're the change management associated with their accessibility mode is like really difficult, right? Because they're, they're tied down. So you have something like this new paradigm, which is a UI list world. We've talked about this two years ago and maybe it's coming. But that is a disruption risk for all the major players in the mobile phone space. The voice modality, which is going to be that, which I do believe will really be kind of that next mode that we're going to be interacting with computers does change everything. But again, it, and I've always said it that the best UI is know UI at all. Like as long as I can achieve whatever I'm trying to the problem I'm trying to solve, if I don't have to point and click, that makes it easier. So now all of a sudden we're going to be designing voice experiences and you're going to have, and Sesame's prime for it, you know, these people who are putting that focus on there, I would be like trying to figure out how could you get in some investments into those companies. And just a quick preview. So next week, we're going to be talking about MCP and some of the ways that you can interface AI with actual tools and technologies. And you start to see the stars are aligning in this space. But we got, what do we have? Oh yeah, we got, we got, we got, we got, we got it the most important thing that we'll ever talk about in this podcast, which is the absolute revolution. Have you guys been seeing Twitter? I mean, we talked about it all week on our Slack. So we have our own Slack, the big cheese Slack, and we just constantly throwing at ideas at each other for the pod. Like this, this, this concept called vibe coding, which Brandon is a vibe coder. Like I'm sitting across from a guy who literally does this all day, every day. And to a certain extent, we are an Artemis as well, but, but, but, but maybe not, maybe not to the same extent where you're literally vibe coding. So let's just back up. Vibed coding is essentially where you, where you take off the handcuffs. You, you, you, you, you stop worrying about what is even being output. And you just let the LLM write the code. Yeah. You copy and paste an error. Yep. Into the, into the prompt, let it fix it. And just keep churning and churning and churning into, and see how far you can go down the rabbit hole. Right. Right. And, and it was funny. Like you were the first one who mentioned vibe coding, like the term vibe coding. And it does. It just fits completely. It, it, it, it fits perfectly. Now I would say the majority of, of my time vibe coding. And if you follow me on LinkedIn, you've definitely now seen me bitching about it constantly. Um, is that I yell like, I mean, I'm literally yelling at the LLM just like, because it won't do the shit that I need, right? And, and so like, I've got this little tiny house in developer town, which is in the small broad, broad, robust place in Indianapolis. And I guarantee, I put sound dampening wool things up because I'm so loud. I'm like, I'm in my room that people think I'm like, and they're murdering somebody. Because like, for whatever reason, while I'm vibe coding, I have to verbalize my anger. I can't just type like, hey, dumbass. This is wrong. I'm like, you gotta get whisper flow. Yeah. Yeah. So yeah, I've got, I've got a lot of, I've got a lot of anger that I take out on the AI. But no, it's, it's absolutely amazing. It's helping me become a better coder too, which is another piece that I don't think people understand where I get into a lot of debates thinking that, you know, AI's going to replace developers. I'm like, no, AI developers are going to replace non-AI developers. That's the whole thing that you got to pay attention to, right? That's going to happen. Yeah. I totally agree. Sidebar 72% of people talk to LOMs politely. So I don't know what that says about you. I'm a sociopath. Yeah, probably. Yeah. So I, for me, I take vibe coding totally differently, which is, it's, it's more from a, okay, I, oh, let's say I have a company that has 10 developers and now all of them are, are laced with cursor, okay? Or armed with cursor, laced. Maybe, maybe laced. Laced, yeah. Right, they're laced with cursor and, and, and our, we can't just push out whatever the LOM says. Yeah. But doesn't mean that you can't still vibe code and check their work. 100%. And so all there's so, if just go to Twitter right now, go to Twitter right now on search vibe coding. You will see so much stuff right now. You'll see tweets from the CEO of, white combinator, the biggest VC in Silicon Valley. You'll see tweets with videos of people vibe coding entire applications in no time. You, but you'll then you'll see tweets of people saying vibe coding is, is a mess and I ruin my whole day and I, and I, I kind of laugh when I see those because I go, you know what, you're just not being good at it. And, and I think it's those initial pain points, right? Like you, you've experienced this and I did this the other day where I built a little polling feature on crafted MBA in a matter of like an hour. Right. So it was like real time updating database interactions UI. It's just a little widget in the lower right corner. It pulls in from a CMS question, answer, boom, hit it on another browser. You get the real time results if you've already voted. Pretty impressive. Especially since I rolled up, you know, using Postgres and all that. And I'm like, okay, this is great. But I went to ask it to change something and it broke something else. They like just decided it had written the wrong code earlier. So I was like, I just undid it. And then I went and asked it something else. And then it was like, Hey, I noticed that you changed my code. I'm going to go and change that back. And I was like, I don't read every single outline because it's a little too verbose. Like the step by step of what it did, I just, a cursory view of it. I'm like, okay, cool. I'm like, I'm going to do what you did. And I saw like, oh no, you undid the thing that I already read it. Why did you do that? So let me back check. When I, my reaction is you're not going to get at it. But the reality is, is that at any moment, the LLM can go trash your entire product. Yeah, I'm 100%. And you'll have, so you absolutely, to do this, you have to have a level of patience and determination. And you need to do modern software development stuff like, like commit a working version. Right. I mean, cursors try to fix this by like having a restore point. They have, they have restore checkpoint and all that. But the, the idea is that you need to get really good at understanding how to keep this thing on the rails. And once you get to a point where you have, because, okay, let's pretend that in the past, you, where you could do something and scaffold out without, let's say we're a, a senior dev and you had medium, you know, you had, you had some libraries that you're used to, but you, you know, most of the time you're, you're starting from, pretty much from framework baseline scaffolding, right? You don't have a bunch of, you don't have your own starter kit. You're, you're, you're going to get in two weeks where this thing is going to get you in 10 minutes, right? Yep. So value those, value that. Yeah. Take, take value in that. And then get out of your own head and ego because here's another thing that happened to me anyway was like, I like having control and lots of developers who are really, really good developers fall into this. You'll see here them all the time. It's not AI. It used to be frameworks. Oh, I don't want to, I don't want to start with that because I could just build on my own. And then I'll have ultimate control, right? There are obviously tradeoffs when you're using these. And when you're saying that someone's not good, it's, they aren't good at leveraging those tools because they most likely can't get out of their own way. Like, and there's a ramp up period where you, developers using tools like cursor and AI or whatever are going to be slower at first. And then it will, but the payoff, I've never seen anything with the payoff as fast as this. Like, remember, react and, and view. And that took us months to figure out. And then we could build way better UIs, right? So I, I actually tried to, to verbalize this in a presentation, verbalize in a presentation, visualize this in a presentation. And I couldn't quite get it. But I noticed that the, I did the Google trends and vibe coding so early that's not even on there. Unbelievable. By the way, and by the way, everyone's trying to SEO the shit out of that term. I already tried. Oh, yeah. Oh, yeah. There's, dude, people are jumping out that niche keyword. Don't take my ideas. But the react JS took, took off in the fall of 2014. So, but I, I think that vibe coding is the biggest thing to happen in my, my career and software development since the advent of view and react. 100% agree. Which is those things, I mean, trying to, trying to build like modern, like, not call it modern, like enterprise professional level web apps that had routing, that had data flows that had state management, that had authentication, that had, you know, unified experiences, like really building applications on the web, you know, you would have had to use something like a, like a flex or, you know, or, you know, or, you know, I mean, like it just wasn't happening. Yeah. You know, Facebook did it in 2004, you know, but they were doing it with PHP. They were doing it with, you know, and see, they, they, like doing it in the client side. Boom. It's crazy. You started going to the moon. Crazy. But now you have vibe coding and vibe coding. So the scope of vibe coding is so broad. People are building video games with it. People are building web apps with it. People are building SaaS startups with it. People are building websites with it. But the question on everybody's mind is, and you know, I went, we have to answer is, is this stuff just mid? Is it mid? Is it mid? I mean, we're just making a bunch of mid stuff. So, well, it absolutely can be met. And I think the people who create it, you're going to get to a point where your product's working and you're like, okay, do I stop here or do I take it to the next level? Yep. And if you're not taking it to the next level and you're just letting it be like, hey, that feature works, then it's probably going to end up being met. So for me, like, now I have the opportunity, I'm working on all this stuff that's just pure native iOS or either native Mac OS or iOS. And I've always been morally opposed to doing this for 10 years. Which is a sidebar, an absolutely great example of Y-Vive coding so cool is that you are coding native Swift and you've never written in that language. Never written in that language. But back to your story, sorry. Yeah, so I never written in that language. But I now get it. I now understand it. I'm starting to see it. But one of the things I was building, which is this private meeting recorder, right, like to take on an author, right, where they just steal all of our information and use to train their own models. I want to have one that's completely private. But I have the ability to kind of, I think it was a Gonzo that said, be the change in your world that you want to see, right? And that's what I do with these Y-Vive coding projects is that like, I'm now making it where you bring your own AI. If you want to bring your Olauma, cool. If you want to do your own Open AI key, cool. If you want to bring Open Router, great. So it just allows you to like really lean in and add some really badass features that are going to truly make your products like really stand out. So it's not mid. But yeah, we're still going to have people that are just like, kill him. I did this and it's going to be fine. Yeah. Well, the other really cool thing about Y-Vive coding is a product validation. Like I had this conversation a year and a half ago with another product team in Indianapolis and they had a really successful startup 15 years ago. They made a bunch of money. It's still their number one revenue generator. But they have been banging their head against the wall for 15 years of how do they get that next win, right? And if you think about it when you're a startup and you're trying to make something work, you put a lot of hours into a thing, right? And sometimes you get lucky. Sometimes there's traction. You pick the right idea or whatever. But the reason I think startups work oftentimes is just the sheer amount of effort you're willing to put into it and iterate and figure out what do the customer want. Well, when you've had success, I think you fall into this trap of like, well, you forget how long that might have taken. Or you have this other thing that's making you money so you're less likely to rob Peter to pay Paul, right? With vibe coding, you could create a prototype in a weekend or a couple days that you put out into the marketplace, get validation, do people like it? Can we iterate on this over the next week or two? And then throw it away? Yep. How much better would you sleep at night if you, I've spent six months building this thing? And now I have to throw it away or I spent six days building this thing and now I have to throw it away. Exactly. I've built a new product every six days and throw it away if it didn't work. Then the six months you're going to have some concoction drum all day and you're never, you're never going to want to hit that, that delete. Yeah, that's a really good point. That's a very good point. And that's ultimately, I think what's going to happen here is that you're just going to be able to iterate at an insane clip. For me, I'm looking at these products that I built. I'm like, well, why can't I just take the core and I'll just niche this out? Like I have one that's tiny label, which I use because I can't see. My vision's so bad that I take instructions, manuals, whatever menus I take photos of it and it transforms it into just a webinar face or not into an app interface. But I use it now just for menus. Like that's where the price, so I'm like, why don't I just create a core one that says, this is now tiny menu. This is a good important public service announcement that Brandon's probably created through five three billion dollar companies. Brandon, can you do the tiny label? I'm sorry, this is a pain point I have with menus is when they put all those fancy ass ingredients. So I have to like Google it or I'll take a picture with like chat GPT. I want a dummies version. So that's exactly what I'm thinking because like so before with tiny label, I basically, my prompt was, hey, someone's going to be taking an image of a menu of an instruction manual or whatever, you need to break this down into sections and tabs and all this kind of stuff. But then I realize when I'm doing the menu, if I would go in there and I make that menu very specific, be like, you're taking a photo of a menu, make sure you capture the price, make sure you capture the description, all of these things that you could absolutely have this killer. And I did it last night when we were at HC tavern because there are menus so small I can't see shit. But now I end, I've now got a menu, there are menu at all the time and then my wife was like, you know what would be cool is if you could actually just like tap, tap, tap on the things that everybody's buying and you would have known that you just spent $700 for dinner at HC tavern before I got the shock of the third. Hey, Brandon, did you know that a twin peaks only has the QR code menu and you could probably come a legend over there? I've never been there. I've never been there. Have you, I hear it. Yes, good food. I hear it's good food. The place that was before there is where I had rattlesnake and I think it was like that they had the airplane. Yeah, what was that all about? I don't know. It was like, it was kind of like trying to be that they have that one like your rainforest cafe. It was trying to be like the like the Indian version of rainforest cafe before that. But anyway. So I've got a raid. I've got a, I've got two raids, a rage things to talk about though with vibe coding. Rage away. Rage away. Okay. 3.7. So, so Claude 3.7 came out now. Here's what I was saying. 3.7 drill quick before you get two technical is a model from Anthropic, which is the developer model of choice in terms of their suite of models. 3.5 is the de facto standard right now for writing code. Anthropic is a Silicon Valley startup heavily funded. Their competitors open AI. Yes. And you had an experience with 3.7. So 3.7 comes out and 3.7, 3.7 is amazing. Now you've, you've used 3.7. You've had a lot of luck with like a lot of people have had a lot of luck with 3.7 doing amazing things. The only problem with 3.7 is it will not shut the fuck up. It just keeps going. So like you're like, hey, fix this thing. So like, all right, I'm going to fix it. It's like, I'm going to search, search, search it, like it calls tools like 30, 40 different times to try to figure out what it's doing. Then it's going to fix it. And then all of a sudden it's like, oh, hey, I noticed this thing over here is wrong. I'm going to go fix this. And you, and it was like, I'm in my case with with belly land. It was like changing the Google models because it doesn't know that 4.0 is a model. So it like sets it to like, you know, 4 turbo preview or whatever it was like a year ago. And finally, like, so I've ended up re re re going back to 3.5. Call it 3.5. I'm going to go back to where we have 5. I mean, so I now I hop back and forth between them, finding that like sometimes when 3.5 can't figure it out, that 3.7 can. But for the most part, man, 3.7 is just, I end up just screaming at it and be like, shut the fk up, just stop, just stop. I did. Yeah, I was trying to get it to write some tests for me yesterday. And first of all, I didn't realize it had added like six files that it didn't need to add. And then it wasn't doing something right. And I was like, no, that's, that's not right. And then it wrote me a fking book about how it was going to fix it. And then it did it. And it didn't fix it. So I was like, that didn't work. And they're like, oh, shoot, you're right. And then another book. And I'm just like, okay. So I think like when it gets stuck, it's just very annoying. But I've had better success in the app itself in cloud. So like Ian doing more high level tasks, like try to, I was talking about converting an app from layer available to X or something. And it gave me a really good roadmap of how I might do that as opposed to like, I didn't, are you doing that within the, in, in cladclaw.com? Well, I downloaded the desktop app, but yeah, okay. Okay, but still says that. Okay, okay. So, so that's, and they don't have a voice model though, do they? Actually, I don't know. I only use that chat GPT and gym and I would be very curious. Like one, clad has a voice model. I would, I would love to use it. But the other problem though that I'm running into is specifically with cursor. So cursor now has like all these new rules and things. Originally, we just had the cursor rule file and that's all you did. Now they've got like documentation where you can go and you can add dynamic documentation. You put your URL in and then it goes and it tries to index all of it. One thing that I'm running into as a, now as a quote, unquote, Swift developer, Swift Vivecoder is, so, if you're a Swift developer, let me just throw this out there, SwiftPackageIndex.com, which is pretty much where everybody goes and they put up their documentation is on this thing. Do not use it. It won't index shit. It cannot index like no cursor. Nothing's going to index it. You and I were talking about it and I said, I'm writing, literally I'm writing a bunch script to go download these websites and try to do it. They've got it all behind Cloudflare and they need human validation. So they do everything possible to not make it so you can actually scrape this documentation. Why? Because they're fucking stupid. I don't know. Like that application is dog shit. Don't use it. And if you do use it, then for the sake of everybody make one single doc, one single markdown document that has every single page of your documentation in it. And it could be five megabytes for all I care. I don't care. One document that basically represents all of your documentation. So I can take that, drop it into my project and I'm going to bring that context in every time I need to work on it. I have that for SwiftWhisper. I've got a whole and markdown document that has all of its things in it because I can't actually have cursor index the websites. It's wild. Go ahead, sir. No, I mean, I had a similar experience with certain docs and I didn't even look into it far enough to see if it was an indexing issue. That's exactly what it is. Yeah, but bullshit. Yeah, I think eventually these docs sites are going to have to provide a way for the AI access. Yeah, because like that is a common, we've talked about this problem so many times. Or it's like I'm writing something, but it's a newer version. Yep. And it doesn't work anymore. Right. So one thing I was I was thinking about this week with vibe coding is, and this is where cloud could become really like get a moat. Okay. So we've talked about the thirst for models to find more content and if stack overflows traffic's down like to the say 90%. Right. Right. It's done. You're done. Sorry. Sorry. No one's going to post GitHub issues. Don't matter. Nothing matters. Right. Well, they're going to they ran out of shit to index. Well, guess what though? They have everything to index. They have the conversations that we're having with them every single day. And all the problems that we're solving uniquely within them. So every time we go, great job. They go, okay, next time someone asked me about, you know, how to create a robot TXT file in an XJS project and make sure that they as you know, there's SEO friendly title. You know what I mean? I have no doubt that I am probably somewhere on some conversation where this guy keeps saying that this thing is like every ex-splitive in the English language. But doesn't that doesn't that actually there's two things that that's to take away is the internet is now becoming less and less important way less important. The concept of content being created on the internet is no longer relevant because all the contents being created inside conversations with LLMs. So they're solving their own, they're solving their their their their inventory problem by themselves. Yes. And I think you just light bulb went off of my head. I don't know if this is what you're saying, but I think it is. So imagine when you go to Stack Overflow and you ask it a specific question about reactor whatever some issue you're having. You are normally 99 out of 100 times just a lurker, right? Right. So you went to and you found a problem that was similar to yours and you maybe solved it. They don't know if that was the right answer because only one in whatever actually answered that. But what is it, what do you think the data from 3.7 came from? It probably came from them finding all the popular frameworks sucking in their docs, but it probably came from conversations of 3.5. Yeah. There's no doubt. We need to look at what their privacy policy is because like, you know, the free, free user's. But that's where you'd almost want people to be like, you know what? you're better off saying share my information. Right. Well, if you're a free user, that's basically what you are. Like, if you're a free user, your information is being trained. Yeah, I'm not paying cloud. Yeah. OK. So then, yeah, then your data is going to be used to try. I have no problem. Now, again, this is how Chatchy BT modeled it, that the pro users who are paying their stuff is not going to be used for the training. Bullshit. Well, but again, there's going all the way back to like two years ago, we were talking about the Samsung thing, right? Like then if all of a sudden, Samsung people are paying, and they're saying that in their privacy policy, that they aren't training it and all of a sudden, my chip instructions start showing up and the thing, then you could just get sued into absolute fucking oblivion. But then again, everybody's suing them anyway. Wait, hold on. Hold on. Sam Altman's online. Oh, hey, hey, Sam. Hey, I just wanted the world to know that I'm doing this for the good of the world. What a guy. What a guy. For the love of humanity. I'm doing this because I love humanity. It's not for the money. It's not for the money. I do not do that. OK, Sam. OK, man. See you later. Hey. Hi. Yeah, so those are my two rants. 3.7 is a Chattie bitch. And documentation has to absolutely change for this to improve. And I'd be honest with you, Brandon, for a Brandon rant, that was pretty, pretty nice. Thank you. It's because I hadn't had so much to drink today. Well, hey, I think that for, let's just go around real quick. So people, I think it's not as an approachable topic for some of the folks that aren't developers. But as we move forward here on the pod, we're going to get more and more focused. And so I'll start with you, Jacob. But if I wanted to, let's say I'm a developer, a aspiring developer, let's say I have installed VS Code. I've built some node. I've built some Python. Like, what would you, what advice would you give to someone that wanted to say, you know what? I kind of understand and grasp what you're saying with this vibe coding thing. But where do I, what should I do? Where should I start? Great question. So I think my advice to early developers is the same it's always been. It's just better than ever to be an early stage developer is build as much shit and throw as much stuff away. Like, go try to build Twitter, clone, go try to build a thing and throw it away as fast as you can. Because that was one mistake I made early on in my career was I watched countless hours of tutorials because I wanted to know exactly how to do it the perfect way. And it's like, no, I could have just tried to build off like 10 times. And then that's where I eventually got to. But you have the best tool set that's ever been invented to build fast and break in, throw it away. What about you, Brianna? All right, so step one, go to cursor.com and install cursor. Step two, find a problem that you have. Like, find your problem, the thing that's annoying you. Like, it could be something that you're tired of moving fucking files around on your desktop. I don't care. Whatever the thing that's annoying you, I want you to go into cursor, set up cursor, and just be like, hey, here's my problem. I want to solve this. Ask me follow up questions and let it ask you questions and then just start kind of going through it. But I think solving your own problem is the best way to kind of keep you engaged with a product. You know, I mean, I spent eight years doing Nomi because I needed, I just wanted a private way to track all the shit that I was doing in my life. And I learned so much. Like, I learned React Native that way. I learned native development that way. Like I learned PWA stuff that way. Like, so you could just solve your own problems. And what you're going to realize is there's a ton of other people out there that have that same problem. And if you solve it, well, you might have a product that people will actually use. Yeah, I think those are great. I think I'll save something more and be more academic, which is if you're a developer and you want a vibe code and you want to maybe ship something that people are going to use and you think you're going to actually do this. I would say figure out where you are from a fundamentals perspective. I think it's important for all developers. If you're really going to get good at this stuff, even if you're vibe coding, unless let's say you're unless you're a founder that's just trying to chalk some stuff up and get it out there, right? And you know it's going to be thrown away. But if you really want to be like next level, if you want to be the, what I was I saying the other day, I said vibe coding, a great engineer with vibe coding is a goaded engineer. Right. Yeah, yeah. And to me, it's like go do all the things you guys are talking about. But if you're focused on, you know what, I'm going to build beautiful websites. Get good at CSS. Learn CSS. Learn the fun of learning. Learn, learn, learn, learn front-end JavaScript. All right, if you want to be a data scientist and do and do analysis, learn Python, learn data structures, right? Learn those things. So let me argue against this a bit. Sure, yeah, I think you should. So I think that everybody needs to understand the fundamentals. You need to understand how memory works. You need to understand how a CPU works at a high level. I don't know if you need to worry about the nuances of syntax of specific language. Yeah, I wouldn't, I wouldn't, I would say that it definitely, no, I would say you do at some level. Because you need to be able to debug problems when they arise, when shit hits the fan. And you need to build an incremental understanding of your code base along the way. But do you? Yes. So no, no, so I agree that you need to have an understanding of your code base. Yes. I don't know though that if I, if, if, if, right now, if something goes horribly wrong with some of the more complicated algorithms that I'm working with and, and Swift, I cannot debug them. And so instead, I'm going to spend four hours screaming at, at, at, at, at Claude. But, but I know that memory management is important. See, but if you were, if you were doing something, let's say I was a funded startup, you would need at some point to have a Swift developer on the team. 100% yeah, yeah, yeah, yeah, yeah, yeah. But the, but your point is well taken, which is like, I think that it doesn't necessarily mean you have your cursor project and your vibe code, and then you go and you get the PHP, or JavaScript book from the library. I think it's implement this stuff. Then go instead of pressing command K, press command L and go, can you explain this code to me? Right. Can you explain this function? Is it what is this? Yeah, I think the journey, so if, to, to, um, make my point, match yours, because I agree, is when you're building something new for the first time, or you, like iteration is the best, right? But as you're iterating, well, now you've got this thing that, so it's like, okay, when I build this app, I want to use an edge. It's like on the job train. Explain to me why you would use next off here, and what, what, it's going to accomplish, and how it's going to accomplish it. And, and that's way better than, than just like, copying and pasting some code. Right. And, I'll give an example. So I cannot tell you how many times cursor has gone down the round path from right off the top. Like, I'll be working on an expo project, and the first thing it does is bring in a bunch of react native binary libraries, which is basically like saying, I'm using framework A, but I'm, I, I think that I need to use framework B, because it's part of a larger um, bro. This is very specific knowledge. And it will literally send you down the absolute wrong path. And, you know what I mean? And so I fear for people that are like, I think there's a lot of reasons why people are like, I can't get this to work. I think a lot of people that are vibe coding are trying to do the whole thing. They're trying to go full stack, right? They're trying to build the entire thing. And, and I'm thinking, you know what? You probably don't do that at your job. You probably aren't the, the, the judge during executioner. Right. And so when you put those, you put those, those pants on, you know, you, you better be ready for the repercussions, because there is a path to absolute domination here. Yes. Right. But there's an absolute path to, to shooting yourself in the dick and, and having a horrible time. Yeah. Again. So, so, so, BeliLint real quick was, um, as I'm working through it, I decided I better go look at the, I, I better go monitor the performance of the application, right? And again, because I know how memory management works, and I realize every time I listen to a past meeting that I recorded, uh, 200 megabytes of RAM increases. I just never cleaned up. And by the end of it, I, if you listen to six recordings, you're at seven gig of memory being used. And I'm like, what the fuck did you do? And then all of a sudden it goes in. And it's like, oh my god, I completely, I'm not releasing Whisper at all. So every time that you're hitting play, I'm loading Whisper, even though we're not doing the transcription, I'm just loading it every single time. And, and, and so eventually I finally figured it out and we got that cleaned up. But like those are the types of things where I think these new kinds of like, if they're not going to be developers at the code level, they're still going to have to understand the basic principles of how software works. You just switched your position to Sean's. But, but I win. No, I think, I think you're, he's just jamming all of it the same. I think that you're both right. And you're saying similar things differently. But like, yeah, I mean, yeah, fundamentals are good. And there's a line. You don't have to know assembly anymore to write. I have never been happier that I went to college and sat in a basement for a year and learned web development. Me too, 100%. Yeah, yeah, yeah, yeah. It makes it so I can vibe code with a thousand times faster. Vibed coding makes a founder be able to look cool. Right. And vibe coding makes an engineer go to it. Right. So one of the things that we're going to, we're kicking around the idea of here. We're going to, I want to tease it just a little bit. All right, fine. We're going to, we want to bring out a vibe coding competition in Indianapolis. And so we're going to be making an effort here to make this thing happen. And we've got the domain, which we got inspired by the tequila next door. What's that place called? Bakersfield. Bakersfield, which is phenomenal tacos, too, if you're ever in downtown Indianapolis. Amazing tacos, amazing tequila. But we're going to absolutely do a vibe competition here in Indianapolis. Yeah, I'm pitching it to some big wigs tomorrow. Totally vibe coded the presentation. If you will. I love it. I want to, so I think that the idea for the competition is important to me. Hey, I think I'd be good at it. Right. We, us three would be us. I mean, we could all be on separate teams. We could have in the same, didn't even matter. We'd be, it'd be tough to beat. Right. There's not that many people that, that I think that could do this under pressure. Right. You know what I mean? Yeah. Yeah. That would be awesome. I think it's almost like, you think of like Karate Kid. You show up and you're looking around and you're like, who's my competition, right? I think that if you stick some founders with some ideas, you stick some people that have vibe coded. Not people, they need to expose so many different aspects of developers and problems because every developer is so different. I think it would be absolutely amazing to see what would happen. But I think what's important is the rules. Like for me, I've been thinking about what the rules would be. And like for me, it's you can't bring, you have to have approved, like you can't come in with an app built and just like start customizing it. You'd have to start from a fresh repo. OK. You could bring in, I think the rules that you bring to the prompter would have to be approved. Really? Yes, I would want to approve there. I would want to approve. No. Yep, I would want to I would not let any low code, no code, solutions. So they wouldn't use bubble. OK. Well, let me ask, can bubble, can you export code from bubble? I don't know. So I think the minimum requirement is that you have to have the code. We have to be able to see that. It has to be that the minimum requirement is that the code has to be published to one or more repose. OK. Right? Yep. So you have to be able to get to the code. Yep. You have to be able to get to the code. And if there's some sort of infrastructure in place that was done, it would need to be there would need to be some guard rails around that. OK. And to some extent, I think you'd be flexible there. But what you don't want is someone coming in with a starting point that's way it's like, you know what I mean? It's like, it's like, it's like you could come with an idea. Don't come with a repo. No, and they wouldn't be able to know what they're building until the competition started. They would find out. It's like, it's like, it's like, what's this show on, on HGTV or the Food Network? Chopped. It would be like Chopped. So they would come out and you would tell them, you are building X now. I think you would, I think what we would probably do is we would go, we would go, SaaS, Native app, and video game category. Oh my God. So you would be able to join your category potentially. Yeah, yeah, yeah. Right. So I like that idea a lot. And we might have a data science category too. Right, right. Right. Yeah. A pure play data science, you know, category. Yeah, that could be really cool. That could be really cool. Yeah, not not. So I hate the idea of coming in and not knowing what you're building. Everybody would create such a, it would create such anticipation. Exactly. And then the, so the other piece is where the sponsors could be the people like again, you go and you find some of these bigger companies here who would be willing to sponsor the event, give us your problems. Yeah, they put you very well could like, they all they're going to do is give you the problem. Right, right. All they're going to do is give you the problem. So not only do you have to come up with and vibe code and app, you're going to actually have to figure out how to solve the problem. Right. Exactly, exactly. So that's it. That's what we're going to do. So vibe coding is real. Hey, I want to hear if anyone in the comments don't have to be a good engineer. Try it. Send us something. Go download cursor, you know, figure out what you want to ask it how to start a project. You want to build a web app? You want a little website. Utilize, you know, the internet to figure out, you know, how to use it. But try to build something. I don't care if it's just a hello world. Yeah, share it with us. Share it with us. We'll totally talk about it. We'll talk about it on the pod and, uh, yeah, we'll see how that where this, where this journey takes us. So yeah, with that said, this has been another episode of the big cheese podcast. We'll see you next week.