s04e01 Cursor just pulled the biggest rug on developers – why’s nobody talking about it?

Speaker 2 (00:01.398) Okay, good.

Welcome back to the Big Cheese AI Podcast. I'm your host, Sean Hyson. I'm here with Brandon. Hello everybody. And Jacob. And today we're going to get started as quick as possible because we've been having audio problems. Welcome to season four and so much has changed.

Welcome to season four.

So much has changed.

I just can't even get it. It's like we're living in it and I just every day, every single day.

Speaker 3 (00:34.766) So today's podcast, we're going to talk about the rug pulls. The rug pulls, and there's been plenty of them. There's rug pulls from Cursor, rug pulling us with their pricing model. There's ChatGBT, or I'm sorry, OpenAI, who rug pulled us with their models when they released ChatGBT5. And then there's one more rug pull that we had on our list, and I can't think of it because we're using my iPad that I had my list for. Was there another rug pull?

has clod increased there?

The overall pricing is going to be a very interesting kind of discussion. But where do we want to start?

Let's start with, let's start with chat. GPT five, the big one, big.

This is the big one.

Speaker 3 (01:20.748) Are you guys using it?

I don't know.

That's okay, so that

That's literally my thoughts on chat GPT-5. I mean, I do know.

That's a very good point is that you don't necessarily know. what's what's happened with chat GPT five is this rollout. They basically said the idea of having people be able to choose their models is antiquated and that what we should do is we should have kind of a intelligent router that when you ask a question, it's going to determine should you use judge GP five mini? Should you use reasoning? Should you use this? And it's basically going to pick for you which model you ultimately want to go with and then it's going to go and it's going to run it.

Speaker 3 (02:00.362) But during that process, they ended up removing the list of all of the different models that everybody tended to use. I should say not everybody. Again, what I think we're seeing here is that OpenAI doesn't actually understand how their users use their platform. And I think that that's what we're kind of seeing here is that there were a lot of people, my sister's included, she basically was like, what the hell happened? Because like, chat GPT was very...

personable. was like my companion. It had this certain voice. It had all this stuff. And then all of a sudden now it's just like cold, non feeling kind of thing. And she's like, I and so I had to try to explain to her like what happens when they kind of rule out these models and whatnot. But then I realized that there's a ton of people online who are complaining about the same thing that the companionship of four Oh,

very specifically for was something that I don't think OpenAI understood how much people actually used this companion where Sam when he was on some.

Podcast is basically talking about well. It was just too friendly to people it just agreed with people too much It just did this and it yes, it might have caused some psychosis here and there But the reality was people fucking loved that and then they just took it all away And so there's a very interesting thing that we're kind of experiencing here is that as these AI companies are building? They're gonna have to start thinking about the actual personality that they're unleashing onto the world because those personalities people are gonna form relationships with them and then if they just like rip them out it's almost like if you have a

favorite CEO and then all of sudden he's gone. And you're like, well, wait a minute, like what the hell? I liked that person, I related to that person, I had that. So it's a very interesting position and I think from what I heard is that they're gonna be rolling that back. 4.0 is gonna be coming back.

Speaker 1 (03:43.598) I it's already back, or at least the switcher's back.

So you can so that actually that was one is that you can you can go to your settings and you can enable show legacy models and then you can get to four back. But like I think it's that's the only one that they did is that for like three three is completely gone right like all the other options but four is still there that you can kind of get to but I don't know. I've been playing with it. I'm very impressed with it. But again, I'm approaching it from a nerd.

point of view and not from like this is my buddy. But some of the things that I've had to do the accuracy that I'm seeing out of it is really, really, really good. Like taking huge documents generating huge JSON files that match my criteria and all that without the errors that I was getting with, you know, for many. So I've been very impressed with the model itself. But man, they've just shut the bed when it came to this rollout.

Yeah. So first, anytime there's a change in a system, everyone's going to bitch. And then they don't know what's good for them longterm. Um, but, I don't know if that's the situation here, but

I know, I think it absolutely is. now, so so chat GPT or open AI when they they didn't realize that they just did the Instagram logo swap, right? And everybody hated that logo. Now, everybody's going to adjust and everybody will kind of get on board and there's an argument to say that the decision to say no, we're gonna have one model, that model is going to be intelligent enough to determine what sub models it needs to use to answer these questions. And that this is just a pain point of our

Speaker 3 (05:18.022) of our mistake before and everybody's just going to fucking deal with it. And eventually everybody will get on board and we have this single model. I don't think that feels right to me that like this this idea that you can have one model to rule them all.

I mean is chat GPT-5 just a chat GPT wrapper?

They're very welcome.

I mean, so, so my impressions of chat GPT five and I actually shared an article about this, on our Slack today was chat GPT five is purposely less knowledgeable. Okay. So it has less knowledge built into it. It is much more intelligent about getting in deciphering information, which is probably one of the reasons why I did a good job with your docs. So it's not bloated down with the entire weighted history of the world and they're just out now. Just

browsing the web for you and just ripping content left and right.

Speaker 3 (06:11.212) That's interesting. you're thinking that the model itself is potentially smaller.

The model itself from a, let's pack in, let's pack in all of the matrices of data that's ever existed in this model. And that's not what they're doing.

That's okay. So that would be, that's a very interesting shift where you then say, okay, we're actually going to rely on using the web, using the different tools, using the MCPs to basically

What's the, cause what's the reason why they put all the knowledge in the model in the first place is cause they were like, well, we got to rip off all that. We got to train these models, but we've got to rip all this stuff into the model. Cause no one's going to ever just let us browse the internet and just steal all this content.

I almost wonder if like the the because I could get

Speaker 1 (06:53.038) They're normalizing stealing content.

So I did. I track probe with perplexity. Have you guys like what perplexity has been doing? I was watching an interview with CloudFlare CEO and he's like perplexity is doing is tantamount to what the North Koreans have done in terms of like cybersecurity, getting around a bunch of your stuff, trying to make it even when you say don't don't go and scrape my website, they're doing it anyway. And so like CloudFlare has built that entire fucking like.

And we don't care

Speaker 3 (07:25.364) Model where you can basically say yes, LLMs can can use my website and I'm to get paid for it Because of what perplexity is done, which I thought was kind of crazy And I want to talk about the perplexity comet browser at some point today,

Are we going to touch on perplexity trying to buy open AI for a trillion bajillion? Chrome.

Chrome

yes, no, so I saw a headline. didn't know about it. Did you know anything about it beyond the headline?

than it's it's probably like a publicity stunt they don't have well

Speaker 3 (07:53.87) They are but but it aren't isn't Google getting pressured Because of the mic the antitrust saying to be a silly spin off the Chrome browser, but

Yeah, but I mean, the price that they put out there is like two or three times their value or their valuation. And it's like, they're already overvalued. how could they?

I mean how much what's the value of the chrome user base? They don't have the money, but what's the value of the chrome user base?

37 billion according to Replex.

I wouldn't doubt it.

Speaker 3 (08:24.258) Yeah, it's got to be something just ridiculous. Speaking of that.

I fucking left Chrome.

Well, okay, so you're a safari. Yeah. And what are you?

It depends what device I'm using. On my Mac, mostly Safari. But when I'm at home, not really working, it's exclusively Safari. If I'm doing any web development, I still use Chrome.

computer.

Speaker 3 (08:51.256) So you guys never got on the Arc bandwagon. And then so that's it. That was the other rug hole is that, I don't know, you pointed something like that. I'm yelling again. Sorry. We're working on the audio. The, the Arc, basically Arc was like, yeah, this isn't the right path. We're going to completely pivot.

He did.

you did. And then it got right pulled.

Speaker 1 (09:02.646) clipping because you're yelling.

Speaker 1 (09:13.204) Isn't that the browser company?

Was the browser company and now they're doing dia, D-I-A, which is supposed to be basically it's Chrome with like the chat, you know, sidebar. And that's their new direction. And all of the great things that they designed and the user experience and stuff that they came up with in the ARC experience is just going away completely.

I fell in love with Arc. So Arc was my daily driver. Like I love the side panel. I loved all the controls and all that. And so once they decided that they're going to basically dip on it because again, they were struggling. They basically made a browser too advanced. And so normal people just didn't understand it is the argument. And so now that's why they're going to go with the and just have the normal top tabs and all that bullshit.

But so I've been on a journey to try to find what my next daily driver browser is going to be. And I have now sold on Zen browser. so Zen browser is heavily inspired by Arc from a user experience standpoint, but it's actually using Gecko.

the engine from Firefox. And so you get all the Firefox extensions you get. I have not used Firefox in like literally 10 years like I used to be a you know, like I would only use Firefox or whatever.

Speaker 1 (10:31.054) I mean when Firefox first came out that really changed web browsing forever. I mean tabs, hello.

Yeah, I mean it was like Firefox when it came out it was absolutely revolutionary and in their their stance on obviously on privacy aligns with me just as a person and all that but I always struggle with the rendering engine was just kind of always shit like their fonts were gross like the scrolling was chunky like all of just the subtle the obsessive attention to the unobvious that like Apple obsesses about like all of those things just seem to die when you go to an open source project.

But man, I'll tell you what Zen browser if you're looking for a new browser, Zen browser is really impressive. There's a lot of there's still a lot of the gank that you just get with like some of these open source projects where they don't have it. But my god, I've I've now completely moved over to it. Last three days have been just using it exclusively and I have not had any problems whatsoever. And it's been great. So Bitwarden

What do you use for password manager though? Okay. Yeah. You're, you're good. I tied mine officially to I password or whatever. I prefer that because then when I'm on my phone or wherever, mean, I'm sure you can do this. I made the mistake of putting everything in Chrome and then when I switched to arc, not everything transferred over properly. And like I was constantly like having a recent password.

Yeah, right.

Speaker 3 (11:55.112) not here's the here's the thing I beg of any of you do not use a browser's password manager as your password manager right use use iCloud if it's available there's there is an iCloud extension for Chrome and I don't know I'm sure someone's probably ported it for Firefox to be able to use that whatever tool but use a modern-day password manager that's absolutely critical right it's going to give you two-factor authentication a lot of them are going to do the passkey stuff right out of the gate so we can just stop

worrying about passwords altogether. But having a central place where all of your passwords are that's not tied to a browser, I think is absolutely critical. You can make an argument not to use Apple because it ties you to an operating system. But realistically, most people like if you're going to be a Mac user, you're probably going to be a Mac user for the rest of your life. You know, there's like me who I constantly think, you know, I might go to Linux again, right? And I might just dip out of everything. So I want to be free from like, Apple locking me into those things.

But realistically, I can tell my wife, I'm just like, yeah, just use Apple stuff because it's like their password app that they just released is actually a really solid application and makes it super fucking easy.

Yeah, it's super nice. So one thing I still get frustrated with is a AWS accounts where, um, it just refuses to understand what a sub domain is. And I'm like, like, all right. So I have to like, remember what was my username for this account.

that's interesting. So it doesn't do a very good job of accounting for subdomains.

Speaker 1 (13:19.15) What about localhost 3000? He fucking has... I'm like, nope, that's 20 apps ago, dickhead.

I hope

Speaker 3 (13:26.784) Now, I will say for all my local host ones, pretty much do use like my brand, my here's my email and the same password that I

Brandon almost gave away his global username and password.

totally just about gave you what my email was gonna be. I wouldn't have told you the password because I make my passwords that I remember, ones that I would never feel comfortable saying out loud without getting just skewered by the world. So all of my passwords are the most horrific things you can possibly think of that if anybody would, like I wouldn't even, like if I had a gun pointed to my head, I'd be like, no, I probably shouldn't tell you what my password is.

That does not spray.

Anyway, yeah, so we're chatty with you five Are you using it no you said you don't know are you using it Jacob?

Speaker 1 (14:12.631) Elvis.

I have dabbled. It did something cool the other day. I uploaded an image of our wedding event doc or image. Yep. And it what it's really good at now is taking images within the or things within the image you upload and then transposing that to the new image. So all I wanted to do was like their layout. I wanted to make it a winding path dotted line through the thing. It didn't do that well, but it copied the underlying layout.

Perfectly. Okay, so I didn't even turn on it because I'd already fired off the email to the design team. I was like, yeah, they'll do that.

I think this is a good time to bring up the and we should someone should bring this up on a and actually read through it. But it's that that paper that Apple came out with about AI. Yeah.

the reasoning.

Speaker 2 (15:03.354) they're basically like, reasoning is dumb.

And, and so like, think there's a, basically a logarithmic scale, okay. To the, to these models, right? And what happens is, is that there's only, it can't, it's, it's, keep trying to push it further and further and further, but it's not getting there. Right. And Apple's paper was really interesting because it came out and basically said, guys, there's no reasoning here. There's no intelligence here.

All right.

Speaker 1 (15:33.24) This is really, really, really, really good pattern match.

Well, and I thought it was interesting because the way that they kind of proved that, if I'm understanding, was they gave it novel problems. They basically were like, here's a problem it can solve, given that this lives on the internet and it has been solved. And they're like, let's change everything up so it doesn't match their words. And to your point, it's all...

an illusion right now. It's just a very, very good, people keep saying it's like a very smart, magical, autocomplete. it is like, sure, which is very valuable, but we need to understand that it's limitations.

So I don't think and you know, my stances is we will not get to AGI with a large language model. We just won't. You know, the large language model, they're amazing, super powerful. They're there. They appear to be intelligent. Again, if you throw enough parameters at something, all of a sudden some like, you know, some intelligence. It's not no. Right. All the novel.

like.

Speaker 1 (16:27.31) The intelligence isn't just tech. And that's problems. I mean, that's why I think what's more, what's, what's really cool is, is understanding how much our vision, how much data we consume through vision. Like we consume more. Somebody said this, I don't know if, I don't know if it was, we talked about in the podcast, but like a four year old child has consumed like

Pajillion petabytes of data not from hearing anything. Yeah, but from seeing things. Yeah

And even even tone and context of. Yeah, yeah, we're incredibly right. Good at processing that.

It's them all together.

Speaker 3 (17:07.201) And until so like right now as we've talked about right you go ask it you know why don't tell me a joke and it's going to tell us that why don't scientists like atoms because they make up everything right it's going to answer that question and it is going to have taken you know a small little villages amount of electricity and answer that right where I can ask you that and it takes literally a quarter

And I I might say something that's never been said

until we get there. So I think, again, that we're at the beginning of this, but to pretend that we're gonna get to AGI just by more parameters, more energy, more of this, I think is a fool's errand. And it's gonna be a new architecture that somebody's gonna come up with. There's this dude that I'm following on LinkedIn.

And he just goes ham on these LLM people saying that LLMs are stupid as shit, that they've been building their new AI model and that they're basically building it in a way that you raise a child, that you go over time, you're teaching it, you're starting, you're starting with small things, you're starting with interactions, and that you're building up actual intelligence based on experiential growth using the same models that we go through as our development stuff.

That guy could be either batshit crazy or he's absolutely onto something. And I'm not sure which it is, but I absolutely love his stuff because he goes hard on people and they just want to battle him.

Speaker 1 (18:33.966) I agree with him. I mean, I think that the concept of actually solving novel problems is proof in the pudding, but I think it also, how do you apply this? So, right, if you're using, a lot of us are using AI every day with success. So you would say, oh, well, it's not intelligent. But you have to understand the use case of AI now. You are the master. are the conductor of the symphony.

You're the intelligent.

You're the intelligence. They're not. So what can you do? Plan your work, right? Break up your work into manageable chunks where the patterns are easily recognizable. Right? Right. Don't throw the book at it because it's not going to be able to figure it out. It won't. It won't understand all the nuance, just like a person can't. I mean, well, so even okay, sorry, let me rephrase that. An intelligent person can't do that. An LLM is not going to be able to either.

Right? You need to give specific instructions. You need to plan your work. You need to have it broken up. And you know what? If you're a programmer, you want to give it, you want it to be able to solve problems that have already been solved really quickly. Give it your, give it your, your work that, that it can, that it can actually do. But when it comes to building that new algorithm to do that, that new thing that no one's ever done before, it's going to be able to give you some ideas and it's going to be doing some.

implement it maybe right like because it knows but like again you're bringing you're bringing the you're bringing the true novelty to the equation it cannot do novel

Speaker 1 (20:05.527) No, even if you went down that path and you tried, it would just churn on its reasoning model and its thinking over and over over over and over and next thing we know we just burn down half the rainforest.

Yes. And also I think something that's going to be very apparent very soon is, um, sure. It's not good for the environment and some people don't care or some people do whatever. Um, I care, but, but what people will care about is the amount of power that it is requiring will have a direct impact on our utility bills. We are going to be paying 30 % more.

It's already it's already happening. Yeah. Number one most popular topic on my next door app is people complaining about their AES bill and it just must I don't even pay attention to mine, but it's going up. And the other thing that's happening I think is with that is that I think there's a bubble coming in the data center market because they're building too many of them. Well, and they don't know how to build them. They don't know where to build them. They don't have enough power. They don't have enough water. They're sucking all the resources dry because they're it's a gold mine. It's a gold rush.

There's a playing field, is it playing field or Greenwood I think is.

All of the above. Any place where there's a field and a river.

Speaker 2 (21:18.53) Well Lebanon...

That's exactly what it is, a field and a river. So they can just suck all this mon...

A local government that's willing to give them a bunch of

We're an Indian, and they do not care at this point, you know, that is so don't know if any of you guys are following the Indiana policy. got Mike Braun, who's literally just like ass chap to Trump, like whatever Trump wants, Braun's going to friggin do it's it's been wild. So there's no doubt in my mind that Indiana will absolutely sell out anything environmental for the the spoils of the AI.

It's just our future. That's the of us to deal with.

Speaker 3 (21:54.254) And it's it's and it's and it's fucking dystopian. And it's yeah, it's it's it's sad. But it's almost to a point where like, like, I don't feel like there's anything I can do about it. There's nothing I can do about it. And so like, what am I all I'm going to do is I worry about shit that I can do something about. as these people are going and just trying to do all this crazy shit. And it's just like, God damn, what do you do? I don't even know.

So, we we've gone through cycles cause we've been talking about this stuff for two years. And it's like, where are we at in the hype cycle? Because I think when we, I think we're, if I can remember correctly, when chat GPT first came out and we were like, my God, my God. There was a moment where we were like, well, I can't generate an avatar of us. Right. Remember that? We tried to do that. We tried to do that many times. And then I think there were moments where we like, well, it's really not that good at this. And then I think there's a, there's a whole.

a phase where we're like, this is amazing when cursor came out. And I still think it is. But like, as time goes by and it continues to go by, we're seeing that there is a finite limit that it's going that until it fundamentally changes, it's not going to get there. So I'm no longer a buyer in what that weird looking guy from opening either left.

he's like, yes, she's going to kill everyone. You know, we're all going to die. Right. well he invented it. So, you know, he wants to seem important, but I, I'm just, I'm, I'm sitting here going, you know what? It's in its spot right now and it's really useful and it's going to be, you know, it's going to change the way you work in order to has, but I'm not sitting here going.

yeah, the super Ilya.

Speaker 1 (23:40.92) there's going to be a robot that can navigate around your house, having multiple conversations with people and you know what I mean? And get every, like it's just not, there's not enough, there's too much data from a vision perspective. No one's gonna be, no one can process that amount of data without burning down the entire data center. Right? And then the ability for it to reason through novelties of just like, I mean I could picture a robot walking in here right now trying to shut the door, right?

Yeah, exactly. Do you know what I mean? What are you fucking doing?

What do you what do you do? First of all, why are you showing your there's a table in the way like it's going to be I need to shut door.

You're completely, you're totally right.

You're absolutely right. Yeah. I put that on LinkedIn and I don't get, I don't get viral and that didn't really go viral, but I do get a lot of people saying, yeah, anytime you hear your AI say you're absolutely right. It just means it's glazing you and doesn't actually agree with you at all. Okay. Or even know how what agree means.

Speaker 3 (24:38.744) Okay, so I want to use that as my opportunity to go shit all over cursor now. Because cursor, I've avoided the update cursor button.

As Jacob is literally cloud coding right now while he's on the podcast.

We're going get to the cloud code thing, but with cursor I've avoided the update button and I don't hit the update button. I finally hit the update button and there's like hey, do you want to try out chat GBD? And I'm like, I guess I don't fucking know. And so anyway, I start going through some work and I'm like hey, go update this endpoint to make sure that we're passing this new variable that I added to super base the table. So like you're passing it through and it's like alright cool and it starts doing it and it's like oh, I see this problem.

And then it's like, ooh, I see this problem. Let me go fix that. Ooh, I see this problem. I'm going to go fix it. I see. It literally just kept going. And eventually I'm like, what the fuck are you doing? And I'm losing my shit at this point. And to a point where I actually like it got really bad at Monon works yesterday. Somebody opened my door as soon as I screamed at it and it like echoed through the whole place. And like people came out of their houses thinking that somebody was getting into a fight. It's a whole different conversation. But it completely.

completely refused to stop and just focus on one problem. So I'm like, what model are you? Like what model is working right now? And it's like, I'm cursor AI.

Speaker 3 (25:57.006) And I'm like, no, I get your cursor AI, but what model are you? Because I want to make sure that I don't use you anymore. And it's like, I can't tell you that. I'm like, well, of course you can fucking tell me this. You're an AI model. And so either you've been told explicitly not to explain what model you are or you don't know. And it's like, well, I know what model I am. I'm just not allowed to tell you.

said that.

And so then I'm like, well, that's dangerous as fuck. And it's like, I totally agree. This is sketchy that I'm not allowed to tell you what it is because the fact is, if as a developer, as I'm using cursor, if I'm not allowed to know what technology is actually implementing the code that I'm in, I can't trust it. You just can't trust it. And so yeah, they're basically masking and making it so you cannot see what model is actually being used unless you go and you select it specifically.

But when you're in that agent mode, it's going to pick whatever model it wants, and it will not tell you what model it is. And that's

Do you trust me, bro?

Speaker 1 (26:53.194) Do you think that some of the, what's going on with the dynamic agent model switching has less to do with picking the right model and more to do with like queue resource allocation?

It's absolutely both. no doubt about it. like, ChatGBT5 apparently, trying to follow any of these companies' pricing models of like their usage and all that stuff is like impossible to fucking follow. But apparently with ChatGBT5, you're going to get whatever it is, know, X amount per week. And then once you hit that criteria, you're going to automatically drop to the mini model. So there is no doubt that they're all trying to figure out how do they stop burning money, like at the levels that they're currently burning money.

And so yeah, there's there's there's I think there's a fine dance between saying that it's yes for optimal model selection, but it's also for ensuring that you know, as a billion people all of sudden go ask the same damn question that we can kind of allocate resources appropriately.

Have you, there's an Nvidia like local LLM chip competitor. Have you seen that one? I forget the name of it, but it's got 128 gigabytes of they're calling it unified Ram, but it's basically takes a copy and shares it with the GPU and, with the CPU. But my question on those is like, cool. I can run it locally and I'm like, I'm a home lab kind of tweaker guy, but my frustration is always.

that the cloud service, open AIs, the Geminis have such a better interface. And I think Sean shares my frustration with this of like, okay, I can set up an Olamma to run these. These are just ways to run large language models locally, but they're just not polished. what's your opinion?

Speaker 3 (28:41.55) Open Web UI is a phenomenal open source application that you run that connects to Olamma and it basically gives you that chat, you BT kind of experience. Phenomenal phenomenal open source project. actually all built on Svelte too. Funny enough, yes felt and.

Nobody at Chachibjee5 will know about Swalt.

And it does. It actually does a way better job with Svelte 5. You don't have to keep reminding it that, dumbass, we're using Svelte 5. So that thing is really, really impressive. But Olamo also just launched their own application. So now if you have Olamo installed on your computer, you now get a desktop application that's this kind of bare bones right now functionality of kind of a chat interface where you can select your models, you can have all your chats. But Open Web UI, mean, that thing

really capable platform that you can use and I use it for all sorts of shit, but we'll use this to parlay into the GPT OSS. So OpenAI just released their first open source model that's now you can get it on Olamma, you can get it on the Hugging Face and others, all the weights for GPT OSS and they released them in two versions, one's like a 28 billion and 128 billion or something like that.

And once finally, like for like four or five days, I tried to pull that fucker down using Olamas. It's like, you need a new version of Olamas. And I was like, I'm on the latest version of Olamas and it would not, it just wouldn't work. Finally got it to work.

Speaker 3 (30:15.374) And it's actually a very capable model. Like, and what I'm the most excited about, I started doing some of my, because I want my LLMs to output JSON, right, following my JSON schema. That's just the only way that you can guarantee kind of deterministic outputs is by saying, hey, here's my data, here's my input, I want you to output it as JSON following this JSON schema, get your output validated against the JSON schema, and then you're like, yep, that's right, right, or it's not right, and go do it again until it's right.

That thing actually did a really good job and that excites me because all of the open source models that have been out there have just been dog shit when it comes to like taking data and actually output it into a schema that it could follow. And this is the first time that I'm like, ooh, I might go through every single email that I've got, right? Like I might go through every file that.

Like when you're talking, you're talking about something you wouldn't send to the class.

Exactly. You just let it go through everything and just be like, here's like a huge and basically kind of do what prompt privacy did, but just build like a little piece of software that's just running on your local machine. That's just basically ingesting all of the stuff and giving you like a true vector representation of your data. That could be really cool. So I'm excited about that. But yeah, so open web UI, it's a little hard to set up. Like if you're, you know, you got to be kind of nerdy because you got to set up Docker. You know, you got to have that kind of shit. But once that's there, it's you know, it's great. I love it.

Total sidebar, but do you see Apple's containerization? No, cause it's pretty beta right now and it doesn't support a lot of things. So I'm going to get wait a little bit.

Speaker 3 (31:38.55) Have you done anything with that

Speaker 3 (31:44.75) People are saying it's a really cool approach the way they're doing it. I don't really know much about it, but I think it sounds like it's very similar to what Windows did with the Linux subsystem. That they're basically making it so that they've got their own little Linux subsystem that you can just run these little containers on. think that's great because personally I hate Docker. Like anytime I have to deal with Docker I'm just like, mean like, Docker, Docker.

Yeah.

Speaker 3 (32:09.494) I don't want to I don't want to poo poo on it because like they they really did change the game in a lot of ways like they brought like some crazy shit to the to the space but I'm glad that other people are basically taking the ideas and

Well, the other thing is that there those new Swift libraries to access models on them. right. Right. So Apple is doing what they always do and their stocks reflecting it, which is their their start every the hype cycles going in there and they're they're sitting there releasing their paper doing their thing and they're going to come out with something.

Yeah, yeah, yeah.

Speaker 3 (32:44.969) Aye, aye, aye.

I hope so. I think one of the things that... Because Jinmoji is a shit. One of the things I think that's important, if you think about the... If you relate it back to the ChatGPT 5 thing, where it's like, you know, don't really have to have the entire world's knowledge base built into the model. You just need to be... It just needs to be intelligent about what to do with information and where to find it. You know, I think that a decentralized AI model

Dogs.

Speaker 1 (33:15.426) works better than a centralized AI model.

I think that that's absolutely where it has to go. Absolutely.

Yeah. And that and what we're saying by that for those that might not understand or get it is like your MacBook is the AI, not servers running in your pollutant rivers. Right. Right. Right. It's all pollutant China's rivers. God damn it.

No, have to get there where, again, these models need to get more efficient. Again, our brain can run on four watts, right? We surely are gonna be able to get these models to be able to run on a phone that's gonna be super intelligent, that can do it. And again, this might be five, six, seven, eight years from now. But once we have that and that we're no longer sending, because again, if you really think, I get panic attacks every time I really think about all of the things that are going to chat GPT.

right

Speaker 3 (34:07.08) Like if you think about everything that people are uploading through the app, everything that's going through the API.

Well, and do you trust Sam Altman to have all the securities like he's gun they're gonna get

no, no, no, not only are they going to get hacked, but there's no legal recourse. You have no privacy whatsoever. So if somebody comes right now and tries to sue you, they can basically be like, hey, you are using chat GBT. So we're going to go and we're going to we're going to subpoena chat GBT and they're going to give you every single chat, every single deleted chat, every single private chat, because again, they don't actually delete anything. Right.

All of that information is available and is out there. And we're gonna I think we're gonna we're gonna there's gonna be some bad shit that's gonna come out from this. So the company let let chat GBT and grok and all these people go and build out these huge models, figure out all the pain points let

There's gonna be some I told you so moments. Yeah. Yeah.

Speaker 2 (35:01.11) Yeah. Lots of things to consider there. But one thing I'm excited about is the

advantages or, or gains we're getting in like hardware design because we have to, right? Like that. we're getting lower, like Mac recently. So one of the things that they've done that is, or Apple that's completely underrated is that Mac mini it's a freaking powerhouse at what? 500 bucks. Yeah. Yeah. Yeah. Amazing. It consumes very little, power and it's super, super efficient.

You can run a ton of models on it. I mean, it's crazy.

I have one and that's our my family's primary computer and it's just it's just great.

Dude, so I was watching this guy who's he decided he was going to have his he's going to use his iPad as his daily driver and then he's going have a little Mac mini on the side and then he did some hacking on the Mac mini to basically make it when the Mac mini wakes up and he just ends up using the laptop as a secondary monitor using sidecar. Nice. And and and he was just like, yeah, this is.

Speaker 1 (36:01.996) Everything you just said went, woo.

Speaker 1 (36:06.652) BUBUBUBUB

And so I've been looking a lot into Unifi. That's what we have here for our wifi. But, so anytime I try to do shared screen, I have a little, an old Mac book that's a running plex and a couple of other Docker containers. And I try to do shared screen with it or whatever it's called screen sharing. And it's super choppy. And that's because my, Deco.

Um, wifi, it's just, doesn't have that throughput or the reliability. And there's like a door in between me or whatever. Um, but like that would be my dream is like having almost like a thin client sort of situation because I want the Mac mini cause it's a better bang for your buck. But I just don't know if the internet is ever going to be fast enough and like rely, you have to be hardwired and like on the same network and have 10 gigabit.

apparently the sidecar what he was saying is so you can do it there's there's two different ways that can do it you can either use sidecar or you can use on like an HDMI adapter that like an HDMI out from your thing to your iPad or whatever he's like but the the the visual his whole thing was all about the color accuracy and he's like sidecar cannot be beat he's like it's it's absolutely amazing the quality that you get in the color representation on sidecar which is going over the wire over the air versus the HDMI is like it's not

even a he's like it's not

Speaker 2 (37:25.782) That must be using like, he's got it.

Do you know a crazy story of sidecar stuff? my kids have, we bought the newest Apple TV.

Okay. do you do the karaoke?

no, but so, so the, Xbox is in the basement. Okay. And, and at eight 30, they have to come upstairs. Right. And they figured out that not only can you Bluetooth connect a Xbox out of the box with an Apple TV, you can remote down into the Xbox and, and play Xbox remotely on your Apple TV. And you can do the same. And.

New Apple TV? wait. That's actually-

Speaker 2 (38:07.118) Thanks for being

guess what even more they take the iPad pro and they can play on that too. So when they get kicked off that they switched to the iPad pro and they're just they just sit there and play platform games or whatever on the on on the iPad and just streaming from the Xbox downstairs. I didn't even know they were doing it. They figured it out on their own.

That's what I'm so excited about, like having these kids, because kids can get just naturally curious, right? They're gonna go and they're gonna try things, they're gonna figure out things that we've never even considered, and we're gonna basically be handing them this arsenal of technology. The same thing, so my daughter, at some point we were at a, my son's like cross-country meeting or whatever, and all of a sudden I get a call from my daughter, who is at home. My daughter doesn't have a phone.

And we didn't have a home phone either. I'm getting, I'm like, how the hell is she even calling me? So I answer, I'm like, hello? She's like, hey. And I'm like, how are you even doing? She's like, so I figured out if I use, I could go to Google Voice. I signed up for Google Voice and now I've got a Google Voice number and I'm able to call you and I'm calling you through my computer. I'm like, what the fuck? This is insane.

So like the fact that they figured all this out, right? Like, that is that's where I have so much hope for the younger kids. They they might not be able to hold down jobs, they might get fired a lot, right? They might not like having bosses, but they're going to be in just just just at a different level of connection to the technology that's

Speaker 1 (39:33.004) You know they have all have one thing in common. They all fucking hate Chromebooks.

Get a lot of them at school.

Yeah, that's what they use and that's what most of them use in schools. The other thing that they all have in common is they never teach him how to type and maybe they do it at different schools. They all have a different way of typing and they're all decent at it. But they are all, none of them are home row and I'm always like, where's the home row?

think saw that somewhere that like typing per minute has gone way down. A lot of it probably has to do with we're mostly on our phones now. like I did Mavis. I had a typing class in high school. Hey well, hell yeah.

Completely unacceptable. yeah. So did I.

Speaker 3 (40:08.482) as AOL.

You had an AOL class?

I know I had an AOI. Yes, I did have about the life of hard knocks on a well back in the day when you're chatting with people and all. Exactly ASL baby and so chatting for me like I had I had a typing class and it's fine but like I had been typing a long time before I even took those typing classes but AOL instant messenger and AOL it was before instant messenger really. But that was what made it so I could learn to fucking.

Next location, please.

Speaker 2 (40:38.764) Why you know what really got me into typing with Starcraft because like you're there trying to see you know, why the games were like no rush 20 minutes, but you're trying to communicate to your partners. Yeah, I'm like getting up in the chat and I'm like, okay, I'm gonna do that and then like you gotta be totally or else you're gonna get those

Those are the best ways to learn, right? Like in my opinion, it's like those are the greatest ways because it's like everybody can learn that way. Even like the people who just are like book smart can still learn by having fun. But the kids who are not book smart like me, I wasn't book smart, right? Like I couldn't learn from just reading a book. I had to learn experientially. Yeah. And that is it.

You also get, you also have the endorphins and like the, the added pressure and anxiety of like, you're trying to talk to a cute girl and you're trying to get your message in as fast as possible.

Yeah, so there's a tall.

I gotta get in this Pamela Anderson pic first. Cheep cheep cheep cheep cheep cheep.

Speaker 3 (41:34.505) That was one thing I just posted not too long ago on LinkedIn talking about because like I was waiting for some image to generate on. Exact thing I'm just waiting as this image generates on chat, GVT and I'm just like, do you do you do? I'm like this is exactly as.

It's the same thing.

Speaker 1 (41:48.106) That's one thing I'm always like, are you really progressively generating this image? Like, what do you do?

It's kind of interesting because like when you watch it, it like it I think it is I think it is really doing the progressive because what it basically does is I think it takes each layer and then it just spreads that layer down across the full coverage of and it just blurs the shit out of it because then like you'll you'll notice as it's loading the bottom blur changes a bit and it starts to kind of form a little bit so I do I think they're like slicing it somehow or another I will say that that GPT for image

I kind of sucks. It kind of sucks. The text is great. Like it does good text. Everything just feels like this plasticky kind of like texture unless you're very specifically telling it to like not make it look plasticky. Everything. People look plastic products with plastic. Everything looks plastic.

In other news...

Music on the other hand has has taken a leap to do it. Sue knows new model. What was the one that I was sharing the other day? That was really cool.

Speaker 1 (42:57.55) No, you made a whole album of it. couldn't tell that you had... If I didn't know you had a drinking problem, I do now.

Dude, I couldn't stop. Yeah. So, so I upgraded to Suno to get, cause like I, I well, all of them. So I basically took a lot of great songs like Hurt from Johnny Cash, Nine Inch Nails. So there's, so, so I took Hurt from Nine Inch Nails and it's like, you know, it's Painful Pine or whatever.

What is it the point of what?

Speaker 1 (43:18.126) Okay, I can tell some similarities.

Speaker 1 (43:25.25) Yeah, the painful pain.

And then I did Punishment from Biohazard and that's all about pints and that was like the punishment pint and then something else. So so but what was funny is one of my buddies reached out. He's like, how are you getting Suno to write these lyrics because they don't actually like it. It stops me every time I try to do anything that's copywritten. I'm like, all right, here's how I did it. So the first thing is, is I come up with my persona. So I'm like, I need you to generate a like a description of Logging Molly and Dropkick Murphy's. So they're the two Irish punk bands that I fucking love.

And so then I take that, create the persona in Suno. And now any song that I create using that persona will be in that same style, right? And then I go to chat GBT and I'm like, hey, using these lyrics that I copied from rap genius or whatever, go rewrite this so it's about a pint. And then it just basically rewrites it. I copy that, I put that into the lyrics and using that new 4.5 mod.

It's amazing. The quality that's coming out of that. Suno 4.5 is insane because like the three version was like super bitty and like you kind of it's it all sounded like it was kind of coming through like a 16 bit graphic, you know, like video game back in the day. But these new ones, man, I mean, it literally sounds like dropkick Murphy's and flogging Molly and and they're good. The songs, the breakdowns are great. mean, it was just like, holy shit. I ended up publishing like seven or eight songs.

They're good. yeah, 11 labs. So 11 labs as a as a music. They do. That's their name.

Speaker 3 (44:52.99) Music one now. Yeah. That's that's right. You were doing a video game. That's right. We did the So so I would be curious now because that you did the 11 labs music for the big game. I then took it to Suno and that was using the three models that was before I upgraded. I'd be curious to see if I give it now the the new one. But the one that 11 labs generated was really on point.

That was really good for that. Yeah.

Speaker 1 (45:16.374) It was great. Yeah. Yeah. My kids, kids, my kids were loved that. No, we haven't been working out lately. Yeah. We decided to build a video game, but one thing that's cool about, about doing the video games, we're doing it all with AI or trying to. and so we, we built all of the characters, you know, like we thought about what they are, but then we built all the characters, like what they look like and stuff with chat, generate with generated images from chat. Gbt. And chat Gbt does a good job of generating images in like a

I wanted in a 64 bit, needed in a 32 bit, I needed in whatever.

really cool. and so it's so what game engine is it actually using?

Oh, it's called Ka-play.

Okay, cool. yeah, yeah, you shared that. So go bot. Okay, so I just learned about this go bot is what Tesla uses for its mapping. They're now switching to unreal.

Speaker 1 (46:01.924) It's just it's I don't know if that's

Speaker 1 (46:15.894) Yes. For mapping?

It's insane. what like imagine the mapping now, but it's using the Unreal Engine, and that the cars are like legit, like it's not it's no longer just like a top down map with the fuzzy little arrows. It's now literally like, here's your car. It's the it's the red three, you know, and the semi trucks are like, like, like video game quality. I saw that I'm like, fuck, I don't like Elon, but man, that guy got to give the guy props that he's willing to go down that path, man. And that's cool as shit.

Yeah, I mean, think about how long that library has been worked on. yeah. and iterated.

Forever.

Yeah, I just created an intro for us on 11 labs. You guys want to hear it? Yeah, let's hear it. Okay, so we'll do it live. Oh wait, it's still generating. updated the names. We'll see what the first... This is 11 labs, yeah.

Speaker 3 (47:01.878) Is that with the 11 Labs one? is. And they just did they.

bought it by the way. Yeah, it's five bucks or the one that I bought was five bucks a month.

So I did the same thing I paid for the year for Suno and like I Know I'm like I need to come up with a bunch of songs now

They literally like these people are just can use this once.

Eleven Labs voice mimic or cloner is really good. Right. Yeah. All right. we go. Ready? Yep.

Speaker 3 (47:24.852) for cloning the voice?

Speaker 2 (47:47.47) All right.

that was it? it. Okay. That was heavily layered. What was the prompt?

it.

Speaker 2 (47:53.998) where did my prompt go? I said, I said, create an intro for our podcast. It's about AI. We are all loud and opinionated. We're also quirky and fun too. And then I gave it a little bit more context. said his name, big cheese and, Sean, Jacob and Brandon. it gave me like a spoken intro, which I don't.

That's weird that that UI looks completely different than the one I they must like just updated other UI.

So I'd be curious to take the the intro that we're using for the off this device here or whatever the fuck this thing's called the dude dude dude dude dude dude dude Yeah, that one right if we would take that feed it to chat GPT that audio and be like I need you to describe this

Right. I'm gonna say make it more techy and punchy. Yeah.

But like trying to get somehow trying to get it to be inspired by that intro to make it to make the new one because that's what I'm finding with like these music ones is you have to give it the inspiration. Yeah. And and and even though they try to cock block you on the copyright, you can absolutely tell the moment that you're like, it's Celtic Irish pop.

Speaker 1 (48:58.666) Yeah, like when I generated my intro, said, need it to be 90s video game. Think Mega Man meets Mario. Exactly. Exactly. And it really nailed it. Yeah.

And so like, even if so if they do block copyright, which that one didn't, because it allowed you to put Mario, you could put Nario. And it's like, okay, past the copyright limit. But as the LLM, the LLM knows that you just misspelled Mario. So you can kind of sneak around it a bit on some of those and basically get to a point where you can generate, you know, copyrighted material. Just saying.

Well, we, we know that we're going to get rug pulled by all these AI platforms are going to be paying twice as much as it that we're paying next year.

So real quick, we're going to talk about this is that so so I have not switched to Claude code yet. yeah, you haven't. Are you still

No, but everyone for the vibe coding competition most of the teams are using clogged code

Speaker 3 (49:56.11) I'm gonna have to probably, I should probably sign up for it, but Jacob, you've been using it. I have. And so then I came in and I asked you how much you're paying for it, and you're like, I don't think I'm paying it, because I thought to use Claude Codes, you're paying 200 bucks a month, no matter what, and I'm too fucking cheap for that.

I don't really understand what I did. just installed the CLI and authenticated to my account. Now it works.

Okay, so but we have not been able to confirm because you can't actually verify that it's tied to the account or whatever So I think I'll probably sign up for it and see if how much that ultimately comes out to but I also I also like the integrated chat interface So I don't like having it in my terminal That's another reason that I've kind of held off on on doing it But you posted a thing where it's us like again, you're you know, like as this new shit comes out You're gonna have to continue to evolve

Yeah, like we're already dinosaurs. We're starting to AI dinosaurs.

I'm using cursor. What you fuckin old man?

Speaker 2 (50:47.136) And I thought I would hate the terminal as well, but like to me, it feels like almost the exact same experience. Like it's just the, the, it's the context in the terminal and set up in my sidebar. So it's actually not that bad.

Yeah, which I have a problem with right now where it's like left sidebar terminal, which kind of talks wants to talk back and forth to each other. Yeah. And then the thing and then my code barely.

Yeah

You know, I guess you could I could take the terminal and put it up into my left side panel there under your right side panel and just have it be on there too. But yeah, no, I feel but yeah, so anyway, Kurt, Claude, curse or Claude code is something that I need to I guess try apparently. the last thing. So today, one of my newer clients, which love working with these guys are there 10 year old startup, but funded and the CEO is very passionate about AI and how AI can leverage it. But he's

also not like crazy about it. Like he understands that you need to have a human loop and all this. So it's just been wonderful to work with him. But he, he's like, let me show you guys this thing that I, UI that I built and I'm thinking lovable is going to bring up lovable and kind of go through it. He brought up chat GPT five and he clicked on it and it said bloop loading up dependencies.

Speaker 1 (52:04.366) really?

It rendered a whole react application exactly like you would with lovable exactly like it. So I'm just I literally and I this was this was low 1030 today that I had this call that I've never seen anybody build a full lovable like user experience with chat GPT and it installed the dependencies. It was a react application. It did everything and it rendered it out in the side panel and it looked just like lovable dude and that.

should be making everybody shit their pants who's in that space because they just like they literally just did it. just they I think they're I think Chachi Buti is absolutely gunning for lovable for bolt for all these people.

Yeah, why the fuck can't you buy Claude code? that's just Claude.

Yeah, but with the pro, so I just, it has a helper function where you can be like, Hey, what's my, uh, cost. And it was like, you have this included in your $20 a month plan. You can do it on the $20 a month. But most people that use it, I don't, I don't use it that much. Most people that use it a lot. Yeah, that is weird. And why can't you do max for a team?

Speaker 1 (53:12.056) But their team's pricing model is weird.

Speaker 1 (53:17.632) Exactly. You can't do Max for T. You have to do Enterprise.

All the pricing models of all these places just die.

shit. You know what it is. It's they're like, let's build a SaaS product. And everybody's going to be one customer. They're like, wait, but we have to teams. And they're like, OK, add another table to the database.

Link these people together!

Well, we had this issue where we have a Claude project and we both wanted to use it and we're like, well, we don't want to sign up for five seats at 20.

Speaker 1 (53:45.486) Yeah, can only the minimum you can buy is five seats

And you can't share projects between teammates as far as I understand it. So it's like, right, so what are we going to do? Set up our own individual projects and we need to have a shared context.

Drop it. Figure out your teams. OK, Google Workspace model. I mean, that's done.

Well, you know, the only problem with the Google Workspace stuff is that

Is Google Drive's worst product ever built?

Speaker 3 (54:08.622) Well, so I can't I can't on my eye Corbin account, is all which is grandfathered in free Google domains shit, right? So like I'm ancient. This thing's like 15 years old at this point, but I can't actually log into VO. I can't have any of the AI stuff. I have to log into my my my Brandon Corbin account. That's just Gmail, which I never use. But yeah, so like that kind of sucks. I'm like, why I've enabled it all in my admin and I still cannot

get anybody on

Okay, second last point though, I was thinking about this the other day is what is What people are building the entire platforms on these platforms? What happens when like? The the rug pulls and these platforms just shut down And you have your whole infrastructure built on this shit.

yeah, no, it's gonna be it's gonna be bad. It's not only that, but but I think if this model of like, well, I'm gonna I'm gonna remove this model. I'm gonna remove this model. You don't have access to the model. I built an entire application that uses 04 many. And now all of a sudden you rug pull me and you shift it up to another one that's not doing it the same thing. The the they're just leaning into the indeterministic.

Yeah, it's expecting that output.

Speaker 1 (55:24.598) Well, right now I'm dealing with the cursor. I keep thinking all the time, well, everyone's seen life with cursor. You can't go back. Right, You can't go back and they know it.

Yeah. No, there's.

People are going to be deciding between do I, do I have to fire two people to keep cursor when they, when they start charging $10,000. I mean, cause at some point don't tell anyone this, but I would pay $30,000 a month per per user.

really good point. Yeah, it's a very

Speaker 3 (55:58.914) know yeah I I would once

I would pay I would pay I 10 grand a month for the company

for the company. want a bunch of people to be able to use it. And it needs to get to a point where I stop losing my shit with it. Because like, when I lose my shit with it, that the executive function that goes into it, the amount of energy that goes into it, until they can address it where I'm not losing my shit, then I'll pay that much. But until that point, I'll fucking.

don't think there's ever going to be a point where you don't.

So a great question to ask, Chachibiti. What is my curse without any explanation? Okay, what is my curse without any... No, mine was relentless dissatisfaction.

Speaker 1 (56:41.182) I'm done with chat gpt therapy

Speaker 1 (56:50.722) Oh, with that said, this has been the big cheese AI podcast. We'll see you guys in a couple of weeks.

Hahaha

Speaker 1 (57:09.214) Cool.

Stop recording.