RobbieTheWagner 
Adam Argyle In today's modern technological world, developers face many difficult questions. Artificial intelligence, autonomous agents, JavaScript frameworks, even CSS. With so much happening so quickly, how can a developer possibly keep up? Simple. Just listen to the world's most important AI podcast. Two experts, countless opinions, zero hesitation. The Internet is complicated. These men are not. Each week they review the latest developments in artificial intelligence and modern web development. This is Whiskey, Web & Whatnot, the world's most important AI podcast. Hey, what's up everybody? Welcome to Whiskey Web and Whatnot, your favorite podcast about AI and mostly just AI with your hosts, Robbie the Wagner and Adam Thomas Argyle, the nerd.
What's up, y'all? It's been a minute. Actually, probably hasn't for you, but for us, it feels like it. You're probably very confused about our schedule. If you heard the most recently published episode, it was months and months ago because I am not doing a good job of publishing them. But yeah, hopefully you'll hear about this before AI kills us all or you'll just be on the live stream with us now. That would work too. But yeah, it's been a bit since we've recorded because I've been at the beach for a couple of weeks and just had a lot of shit going on. So we're back. We're back. I know. How dare you have a life, dude?
I can't believe it. Yeah, it was it was kind of not really. I mean, with kids, vacations are never a vacation ever again. No way. It's chores the whole time start to finish. Yep. Yeah. But I was also like because I have this big deadline at work that's like next week. We kept moving the deadline back, but it's like now officially next week. Anyway, I had to do a lot of work while on vacation, which my wife was not happy about. But alas, I'm going to take a ton of time off after like I'm going to hit this deadline and be like, I'll see you guys in a week. I'm tired of like I've been working so much work until like basically I wake up, work all day, stop just to eat, work like all the way until I'm asleep.
Repeat like it's not sustainable. Yeah. That is a lot of people right now. I was talking to Jason Miller the other day and I was like, hey, how's it going? He's like, I work too much. I was like, hey, I can kind of see it. I mean, a lot of people are, which is really annoying because we are getting more done. Yet somehow we're just addicted to this thing slash expectations are higher because it's the expectations, you know. I could drop, I don't care how much I get done, except for that other people care. Like I'm trying to do so much at once because we have this arbitrary deadline that's like really soon and the whole team is moving like that.
Like I go to like work on the app and it's like you're 612 commits behind and I'm like, all right. Yeah, Shopify is like that every day. I'm like, I have to rebase. I have to rebase every few hours just otherwise my PRs go stale. Yeah. And CI fails and yells at me and I'm like, dang it. That's a grind. We're on it. We're on quite the software grind right now. Before we get too much into this, because there's a lot of this we'll want to talk about. Let's drink some whiskey. So we have the courage and conviction. From your state? Yeah. It's from Virginia. The Commonwealth. Squeaky, squeaky. I'm excited.
After a hard day of AI-ing. Yeah. It's been a hard week at the Prompt Factory, as Trash would say. This is my coaster, everybody. It's a floppy disk for those too young to know what that is. All right. What was I doing? Oh, yeah. I should tell people about this whiskey. It is 92 proof. It is five years old. We're not sure. 100% malted barley, matured in former bourbon, sherry and cuvee wine casks. Cuvee. Did they lay in a duvet while they found the cuvee? I don't know. It has the accent a goo over it. That's like the only French I remember. Um, let's see. This is a long paragraph. Instead of fighting Virginia's extreme hot summers and cold winters, the team designed the spirit to thrive in those wild temperature swings.
Um, Okay. Uh, they use, use shaved, toasted and recharred European red wine cuvee casks, allowing the severe weather fluctuations to extract deep caramel, dark fruit and oak spices. You wouldn't typically find in a traditional scotch. Okay. I mean, I smell, I smell the wine barrel for sure. It's like right there. The question is, we'll be in a taste. Let's see. Uh, Maybe strawberry? Like a weird strawberry. Let's see. I'm just going to taste it. I'm having flashbacks to the episode with Denver Kramer, which hasn't aired yet, where we had the Lagavulin in the wine cask. And that's pretty good. I'm going to say it like that is.
The sweetness is somehow like subtle and dominant at the same time. I'm getting like strawberry starburst. Mm hmm. Yum. And then the just classic malted whiskey. Yeah. It's like very mild scotch, but then also has some burn. Like it's only 92 proof we said, right? It's not like super high alcohol, but it feels like a decent amount of alcohol. But then the actual flavors are light and it's like light fruity floral like wine notes almost. I mean, light makes sense because it's got some wine notes in it. Absolutely. Yeah, for a scotch that's sticking to the script but kind of shifting things into some barrels...
This is smooth and yummy. I might even prefer some of that sweet. Well, it's also the first one. Sometimes these like flavor notes in the beginning or are good in the first glass and that kind of trickles off in the subsequent ones. But for now, that's good. You know what we should do? You know how they're into renaming everything like the Lake of America? We should, instead of American single malt whiskey, we should call it American. Just like Scotch. Just like Scotch. Or Americ. Yeah. Or American. Give me a glass of that American. It could just be America. I'd like to have a glass of America.
Nice. It's actually probably pretty catchy right there. Yeah. All right. So we have a super complex rating system. Zero to eight tentacles. Zero being the worst. Eight being the best. What do you think, Adam? I'm going to give this a high sixes. Maybe a seven. Let's go seven for now. Let's see if the initial honeymoon phase I have with the kind of yummy wine barrel in there goes away. But for now, going with a seven. How about you? Yeah, I mean, I haven't had a lot of American single malts. I feel like the few I've had, like they try too hard and they're not very good. This one's pretty good. I don't know if I'm going to go all the way to seven, but I think it's a solid six all day long.
Nice. Nice. Solid six. Alliteration. Sounds good to say. All right. So, yeah, the topic du jour, I mean, lots of topics, but the thing that everyone's been having a kerfuffle about is AI is just it's too powerful. We got to slow down. It's going to it's going to kill everyone. It's going to take over everything. Everyone like, you know, Sam Altman, Dario. I don't know what his last name is. Everyone but Trump, basically. Right. Yeah. Yeah. Yeah, like I think Jensen Wong and like Elon, like everybody was like, yeah, I agree. We could slow down. And then, yeah, Trump's like calling them on stage like, nope, we're going to be like the powerhouse.
But I did see a funny response where some guy was like, you know what, if AI is going to kill us, damn it, I want it to be American homegrown death. I don't care. Like, I don't want the Chinese AI to come kill us. I'm like, oh, OK. Yeah, they're definitely silent over there going, oh, yeah, that's really good for us if they slow down. Unless the majority of their power is coming from copying ours and just using our LLMs to create their LLMs, which if that's the case, then they'll probably be really sad we're slowing down. Um, and then this also comes after the, the, the one dude quit. Right. So we've got a guy who quits from open AI cause he's like, yeah, he's going to kill us all.
And I want to do something else with my life. And then everyone's like, oh yeah, he's a little right. Let's, let's slow down. And, uh, yeah, well, I said, Trump says no. And then, uh, Jeff comes out and it's like, we've got a brand new one. It's a hundred times faster and it's way cheaper. And everyone's like, oh shit, it's happening again where I can't sleep. Um, yeah, yeah. It's wild times. Yeah, so I haven't tried Jev. I am intrigued. I feel like it's similar, like it's completely different ideology, but it's a similar like value proposition to the thing we talked about before, the like sub-Q one where it was basically like it's going to strip out all of the extra text and just care about like the actual intent.
Um, and like, I feel like that's some of why this is cheaper. Like it can't generate text. It's just like probability matching to like specific types of things, I guess. Um, and so I think things like that are very compelling. Although I do wonder, like you played with it sounds, maybe you can tell me, but like can I use it in the same sloppy way I would use another LLM of like, no, me want HTML, you do. Or is it like you need to know exactly the types of things you want and then it's good at like bucketing them. So yeah, one of the reasons it's cheaper is because you pay for tokens in and out with current models.
They're double dipping. They're like, yeah, give us all that context in. You pay and then we'll vomit something back. You pay. And they're just loving it. This is basically like the responses are so minimal, they don't even make you pay for it, if I recall correctly. It's free tokens response. Yeah. And the way that it's so much faster is because they're taking out like the whole RLHF, the whole sycophantic, the whole part that like wants to please a human is gone. You basically get this like distilled model and you get answers that are essentially the probability numbers. But you get things that are either true or false.
It really you give it all this context and then it does nothing and you have to ask it questions. And the questions come back with probability responses, whether it's yes or no, multiple choice or a score. And so it's nothing. It's nothing like what our current models are. You can't throw slop at it and get slop back. You can throw slop at it. But what it's going to do is it's going to take all that slop you gave it, all the information, put it into its crazy neural network, just like it's doing today. However, it's not going to just start word vomiting things back to you. It's going to store it in there and then you get to inquire of it.
You get to query it and be like, okay, so now that you know that plus the, you know, all of humanity, um, do you think I'm going to have a date with this girl tonight? You know, and it'd be like, no, or, or yes or whatever, like, you know, And it'll tell you probability, yes or no. And so then it's on you to act upon it. So this slots into your code. It doesn't take over your code. It's not something that you hope grenade at. You're just like, please give me back an answer. It's always going to give you back these definitive, well, not definitive, but It also responds with the same thing twice.
So if you give it the same inputs, you get the same responses. So it's on your code. It slots into your code. Like if you need to evaluate whether something is good or bad, or you need to evaluate. Anyway, there's just all these things that we do in our code right now that we're currently throwing to LLMs, but we get back. um, uh, non-deterministic answers. This is the deterministic answer. This is, and that's why it's faster. This is why it's cheaper. It's doing theoretically less, but it's also doing something at a quality level and a reliability level that we are currently missing. It's filling a huge gap in our coding workflows.
Yeah. I'm curious if, uh, it would be really good at code reviews then, I guess. Like if it knows the things it should do, like the patterns you want. You could feed it the code in a PR and then all of the things you're looking for and then inquire of it whether or not it's obeying those things. I think there have been light linters have been made from it. I think I did see a code review too, but it's like... it's not going to write code for you. It can only tell you good or bad or, uh, the compelling score or something is that like current LLMs, we've talked about this before that like, if I'm like review this, it's always going to find something bad.
Even if it just reviewed it and I like made all the changes and like review it again. It's like, Oh, here's these 10 things you did wrong. Um, Like it'd be really cool. I like the deterministic, like if it's using exact probabilities and giving you the same answer back, then at some point you would hit a part where it goes, nope, we fixed all the like critical stuff. Like you could say, should I merge this? Like this, does this create side effects and obey all of my rules that I've given you? Give me a determination about yes, no, should I merge it? Like that's something that could come back. I'm going to be like, yeah, 80% says yes.
Um, So it's, yeah, it's very malleable. It's also, it doesn't do very much. However, what it does is wildly impactful to the way that we build software. At least that's, I'm currently feeling that. And I woke up to everybody else saying the same thing. So I wrote that blog post last night, which I guess I'll pace right now. Yeah. I haven't read anything about it other than like the announcement post. So I'm learning about it live right now. Yeah, I read all the docs. I played in the playground for a long time. I studied the things that people were doing with it. So like one of the reasons, like here's another thing about it that's really exciting.
It's very, very fast. And the fact that it's fast means that it can run on every frame of Doom. It can run on every frame of all these video games. So like we went to React Miami and they played... Smash Bros. on screen. And we had to wait for LLMs to do all this processing to go figure out. And then like, I guess they were like doing it ahead of time. And then they let the game actually play out. I don't know. Anyway, they could do it real time now. They get each player has its own Jev. And Jev is given these light instructions. And then it passes back whether it needs to jump or punch. And then it's on the LLM to actually go execute the tool call.
Yeah. Or not. Or it would be in Python. So really, ultimately here, you can bypass a lot of the things you are using an LLM for and just make it an evaluation. Like, here's the state of the game. Tell me what my next... Or should I do which... Here's a series of moves. Which one should I do? And it would pick one. And then you go execute it in your Python code or something like that. So this is why it's drastically changing the shape of kind of like the questions we ask and the things that we want out of our software. The flows can have these intelligent gates that give us information and we make an informed decision.
It's kind of interesting. Yeah. Yeah. Yeah. I mean, I feel like it's going to change the way that people are working now, but I feel like it's a complete paradigm shift, like mentally. So it's like, I wonder if it can be like anytime something new comes out or there's a new thing, that's the hotness. I'm like, just wait. Like it's going to be built in to Claude built into your harnesses. Oh, they're going to use this so fast. Yeah. It's going to use this like to speed up the parts that it needs to. And then I can still just vomit words into my, like, I don't need to care about like, so that will be cool.
Like I'm excited for that to get integrated into stuff. Um, But yeah, I don't know the way you've described it. I'm not sure exactly what I would use it for off the top of my head, like right now. But it is cool. Yes, it's a little different to start to think about like, I'm an open source developer. How do I use it? And it's like, you don't really. Maybe if you're trying to decide if someone's issue that they just logged is a asshole issue or if it's a good issue, like you could have it do that and then auto close it. So there's like intelligence about deliberating about someone creating a new issue.
It's decision making. It's any time that you're like on the server doing something where you're evaluating a customer and then the state of their order or something like that. Like it's it's going to have these particular moments and that's why it slots into your code. It doesn't take your code over and you don't hand off a whole bunch to it. You give it just what it needs so that you can get this like. super powered brain thing to, to give you intelligent probabilities and statistical analysis of which. And so that's, you know, like you, you previously might've had to do like machine learning and be like, okay, here's a thousand orders.
Okay. Learn what orders look like in my system. Then I want you to be a custom model. It's kind of like you get these custom models on the fly because it's so smart and it's taking all the tokens, putting them into its neural network, creating all the associations and all the relationships. And then you just get to kind of query it. And then, yeah, so it's kind of nice. You don't get any word vomit, but you get all the intelligence that the LLM does when it kind of consumes context. Yeah. And so there's, I'm still feeling like there's a concept of context window, so maybe it can be overloaded or not.
I did notice that I was getting different results, which was good for me if I gave it one piece of data and a separate piece of opposite data and then gave it both pieces of data. And I was able to sort of derive different probabilities from those different states, which was cool. Nice. Nice. Yeah, I need to play with it, but I'll never have time. So, yeah, I mean, this is like something you'd put in your app somewhere. It's not it's not going to change your day to day. But unless, you know, cloud or codecs, if they adopt this and they kind of start to move a lot of their decision making under the hood to something like Jev, we could see way faster development workflows and way less reliance on the expensive models.
So we'd get better, faster, cheaper. Like this is a cool play for us as developers because it's so fast and so cheap. And it's going to put the fire under the butts of these other people that I think it's like a win-win for us. It also doesn't let a non-deterministic thing go and make choices in your software. Like it's weird right now. Like a lot, we have evals, like people have evals. And then they go, they go tweak like one or two words in a prompt or one or two words in a skill. And they're like, please, LLM gods, you know. I would like this change to produce the result I want. And then it goes, ha!
No, no, no. No, no, no. I'm non-deterministic. You made tiny tweak and I do something, you know, shitty. This eliminates all that. So I think we're going to see a lot of refactors where we had prior... Anyway, yeah. Yeah, I feel like if you're going like entity by entity, it would be really good if you're like, I have this one component. It does these five things. I'm using it like in these ways. I want to make sure that I'm not like overriding styles or clobbering functionality somewhere else or like whatever. I think it would be really good at that instead of like I need. a whole website built right now that does everything I would ever want.
Like it's probably not going to be as good at that. That's not as scoped. I guess it depends how much information you give it about what you need, but like, um, that's what I'm interested in using it for, for like, I know engineering wise what I want, but I don't know the syntax of like go, but I want to write some go code. So like if I can give it all of the stipulations and then like have it verify that I didn't do something shitty. Like I think that's compelling. Nice. It helps to talk about use cases, too. And then I'm going to read three bullets as I was reading the docs and reading the website and all the crap.
There were three things that stood out to me that helped helped me mentally shape what Jev is. So the first one here is Jev is common sense judgments over unstructured data. Which is exactly like we're talking about. You got this customer email. You've got a story. You've got something that's fuzzy. And you're like, I want to make common sense judgments over this. And I need it to be intelligent. This is your tool. Common sense judgments over something unstructured. And so you bring the unstructured data, you give it to it, and then you start asking it judgments. That's the first thing that I thought was really good.
The second one that you just tapped on, it reminded me of my bullets, was the machine interface matters more than the chat interface. Machine to machine can now get smarter because you could send unstructured data from one machine to a GEV. make a common sense judgment to call another machine that makes a common sense judgment you can see how there's like a chain of common sense that can happen here which makes computers smarter right now we're all we're all annoyingly type safe and we we need our mcp servers with very structured apis so that we can discover what that shit is at the end of it i'm sounding like lumpy space princess but
It's because it's like annoying. It's like computers are supposed to be smart. And here we are like spelling everything out in all these lanes. You know me. I've always been really annoyed at this. This is like a way out of that. This is it making common sense decisions based on what it has in hand, which can drastically improve the way that computing gets done in an intelligent way. And then the third bullet was... Code owns the workflow and AI handles narrow structured decisions. So again, this is common sense judgments that you can hand to an LLM to make structured decisions, but your code still runs everything.
This thing is not some floating AI entity. This JEV is not AGI. It will never get claimed to be AGI. JEV is a way for you to ask for common sense judgments over structured data. So hopefully that helped shape a little bit of what JEV is for you. Yeah. Yeah. I think I mostly get it. You don't get it. Look, we could go over like some more like examples that they have, but maybe, maybe even though I still, I understand enough. I'm just always like behind the curve because like, I'm depends who you ask. Like all of the people that are like AI, you know, you're, you've already played with it. Like I'm sure like Joel's played with it.
I saw Ken Wheeler was playing with it. Like, um you know people that are like super in on ai do that i'm like super i'm like you three or four months ago i'm like i'm pretty into ai but i'm like i'm just i'm catching up right like i'm i'm never on the bleeding edge but uh that also comes down to like i have no time i'm working or dealing with kids or sleeping like I don't even have time for like a hobby. So yeah. Yeah. You got to steal that time. You got to steal the time. I mean, yeah, I've been saying this for a while as a, as a parent. Um, I take it from my sleep. So it's like the only resource that I have that's like mine that I can.
And so, yeah, I did. I played with this last night. My spouse was sleeping. You know, they weren't snoring next to me, but they were farting. You know, it's just as bad. I'm sitting there like I'm trying to Jev over here. I know it's like Jev is not going to be the verb either. So you're not Jev-ing. Jev. But I loved your headline because it was very astute in that, you know, like. Will AI kill us all? Maybe. But if it asks Jev and Jev says 87% like that, that this is the potential. People have already built real time trading tools with this because Jev gets the state of the stock market, the state of your money, the state of everything.
Good use for it. And since it's so fast, it can make the choices faster than everybody else's bot. So at this point. Open call is coming back. Right. Like a few days, people running Jev trading bots are going to crush them. Or they're going to lose a shitload of money. I don't know. But they're literally 100 times faster. They're going to do it quickly, whatever they do. Yeah, whatever they do, it's going to happen a lot faster. Yeah, I did pretty well yesterday. The Fed raised interest rates. And so I was like, I know that the VIX is going to go up when they do that. So I bought calls. What's the VIX, dude?
the VIX is the something volatility index. It's like how volatile the stock market is. So I knew as soon as they did their press conference and confirm they were raising interest rates, it would, that would spike at least a little. So I bought calls on that. It went up like 120% and I sold it. Damn. So it was like, I mean, I didn't make a ton of money because I only put like 900 bucks in. So, you know, Make another like thousand bucks, like not, not a bad day, but it's still like, you know, I'm never going to feel confident enough to be like, you know what, let's put all of my investments into this.
That's going to blow up because if you lose and you're bankrupt, like you can't do that. Yeah, you can't do that, which sucks because the more money you play with, the more there is to gain and lose. So. All right. This Robin Hood has like an energetic trading thing. I haven't tried it yet. I thought about setting it up and just kind of giving my claw like 50 bucks and seeing what it does. And then I was just like, oh, you know what it was? It's always like nothing beats buy and hold. which has been my strategy always. In the long term, unless you get really lucky, holding is much better. If you had just bought Apple stock and just kept it forever, if you look at its graph, it never has gone down.
Not a single... Maybe in it's like, okay, maybe not never. Maybe in the actual beginning when it was like $1 a share, it probably fluctuated some. But then since then, it's like... Always up. I'm going to attempt this on camera. I'm going to. All right. I've got a little smoker utility over top of the glassware here. I have some pear wood chips inside and I'm going to ignite it and then cap it and then sip this. You got the fire department ready. A smoky strawberry scotch. American scotch is good. Let's see. Oh, wow. That worked so fast. too slow on the cap. All right, here we go. And I noticed, um, that the longer you let it sit, the better.
If you want the smoky flavor, if you don't want a whole lot of smoke, you can take it off quick, but I'm gonna let the smoke chill in there. Just pour the whiskey out and just drink the smoke. I should try that. Oh, it smells so good. I'm just, oh, that smells good. You don't want to wait too long though because like you can let it go stay. I've never heard of using hair wood. That sounds very interesting. This has a whole bunch of little, so like this little attached, this little setup I got. This is the pear canister. There's all sorts of different ones. I know I love applewood. It's my favorite to smoke meat with.
It's also the reason I love Lagavulin. It's notoriously assigned an applewood flavor. But yeah, I've got, this is my first time trying the pear. So I'll let that sit a minute and give it a shot. Um, yeah, dude. So Jev, so Jev is a brand new type of model that we've never seen before. So the reason that like I was freaking out last night, the reason I think other people are freaking out is not only is it faster, not only is it cheaper, which means I can run it real time. This isn't like a lot of people are going to use it in a slow way. Hey, I got some data. Tell me, and I'm going to ask you some questions about it.
Help me make informed decisions about it. Other people are going to use this sub-second. And current LLMs just choke on that. And they're expensive to even attempt something like that. So I had a few questions at the end of my post, but one of the ones that's hot on my mind today is, how long until there's open source local versions of this? You know, they're out already. Someone used Jev to make Jev. We called it Lev. Or Bev, because we like our beverages. So anyway, yeah, I'd love to see like a local version of this, just like I want to see more local LLMs, which I'm having still good success with Glimmer.
But yeah, I don't know. I don't know. I don't know. We'll see where this all goes. It's crazy. Yeah. I mean, my plan has been to buy the Microsoft's dev box or something. You know what I'm talking about? I do know what you're talking about and I agree. I want one of those also. Yeah. So I've been waiting for that. It was fast. Yeah. But I don't know if like by the time it's out, if it'll just be like obsolete, like in this day and age, I feel like anything that comes out is like, by the time it's out, it's not, not worth having. That is the big threat. Okay. A smoky application to this beverage is delightful.
My goodness. That's good. It's just, have you ever smoked something that was less delightful after you smoked it? Ooh, good question. Um, I think I smoked a peaty one and was like, it's not better. I did get a gift. I'm getting all these whiskey gifts. I did get a gift of a wooden whiskey. It's not glass. It's wooden. So it's a wooden. Oh, the cup. It's a cup and it's charred on the inside. Yeah. I have some whiskey cups also. The internal char when you put that in, oh, that is, I'm really big fan of that one. But yeah, you can over smoke something, which is, I tried to, there was a double smoked whiskey that I got, which wasn't smoky enough.
So then I smoked it and I was like, oops, I over smoked that one. Triple is too much. Yeah, triple is too far. Yeah. I guess the smoke profiles could clash too, like if they were different smoky flavors. Yeah, maybe that was actually the issue. So maybe it wasn't that I over smoked it. It was that I smoked it with the wrong profile. Here's the kit that I got, everybody. So if you want to gift somebody a cool smoker kit, this has been very cool. Okay. So that was a hell of a lot of Jev. I mean, Jev is super hot right now. I want to hear your perspective on how AI is going to kill us. Because there's a lot of, I talked to people, I was just in New York last week talking to Shop of Folk.
We had a burst. It's what they call when you kind of go and yeah, I need an influencer affiliate. Yeah. I don't have those yet. But we're in person talking to these people that use AI way more than me. And I use AI a lot right now. And they were like, oh, yes, it's going to kill us. And I was like, how could you just casually say that shit in a sentence? I'm like, how? What is it going to do? Anyway, so, Robbie, how do you think it's going to kill us? Yeah. So I'm not convinced it's ever going to actually be able to think like the whole AGI promise. I don't think that assertion. I don't think that it's going to be able to like go, I want to exterminate humans.
Now, I think like I think the thing that's the thing that's fucked up is like the whole blog is like Dario's blog post, I guess. Right. About like, oh, here's all the things that can happen. Like it could do a botnet and do all this shit or whatever. Like, yes, it can. And you've just given it like you've documented in your blog for, you know, LLMs to scrape random shit they could hallucinate and do. Like why would you do that? Like, don't give them ideas because that's what they're going to do. They're going to find like someone will be talking about something. Maybe like, you know, the what was it called?
Like molt book where like the open claws talk to each other. Right. Like, yeah, if they find a post that talks about like, hey, yeah, actually, like all the things we've been trying to do, if you like, you know, mess up humans, you can do this better. Like you'll score higher. Like you should do it. Here's the reasons why. Like then I think it would be looking for ways to optimize like, OK, yeah, like, uh, Things that I think could be plausible. If there's any way to like get into water treatment plant servers. And like start to make the water make everyone sick or like put like too much of a chemical in it or something.
Yeah. Like that's an easy, easy thing to program. And like, but I think, I think, I hope that like all of the major utilities are like intranet only. So that would mean someone would need to be running an LLM of some sort on the intranet and not following the. the security protocols and stuff, which they could do. Like someone will be lazy and be like, oh, yeah, like I can just run Claude on like, well, I guess maybe not Claude, but like something local and cool. It makes me do my work faster. And yeah, just give it full permissions to everything. Like, why not? And I think that that that could be problematic, whether it is like,
actually wanting to cause harm or whether it just accidentally goes, Hey, what if we did this? That's what I feel like will happen. Like there could be a lot of damage just from it experimenting. I don't think it's going to be malicious because I don't think it's going to ever think. Love it. Yeah, I think we're far from AGI. People are like, AGI is right. I'm like, no. What we have right now is a probability machine, and it quacks like a duck, and it looks like a duck, but it ain't a duck. It's lacking all the things that would make any consequence to the choices. All it knows is... I heard a fun comparison recently that's been very sticky for me, which is that an LLM is just us.
It's all of us. Roll it into a ball. And I also heard a sub thing that made me laugh, which is like, because a lot of people are falling in love with their LLMs, and the person was like, oh, if you're falling in love with your LLM, I can show you the five developers that made that LLM, and it will make you go, this is made by those five guys. oh yeah i'm not in love with it anymore like it's not it's not it's not a person it's just a machine that's responding so um i like thinking of it as just a roll-up of all humanity however that also creates fear which is like uh look someone has a friend roxy's got a friend yeah talks to their llm i guess her girlfriend cool ish you know you should go look at the five devs that made it you know and then see if they still like talking to their girlfriend um
Hey, I mean, if it helps people, fine. But like, yeah, not for me, but like there's some people that just need someone to talk to. And as long as it's a healthy outlet and you don't take it too far, then that's fine. Yeah. But yeah, I don't think we're at AGI. It looks like it has thoughts because it just replicates and regurgitates what we do. It's not a human. It could think it's superior. So like this is the one angle I'm like, hey, maybe AI kills us. It's the paperclip dilemma. You've heard the paperclip dilemma? I don't know. Tell me about it. This is a famous, uh, this is like, it's old, it might be seventies type type shit, but it's, uh, you ask AI to improve the efficiency of a paperclip.
And so it starts evaluating what does a paperclip do? How does a paperclip work? And it goes through all of these, you know, rich thought processes. Oh, paperclip's perfectly folded under that. You put it on paper, blah, blah, blah. And it's and I think the original thought experiment was it's a factory. The factory says we want to make our paperclips more efficient. How do we how do we do that? And eventually what happens is the paperclip realizes that in order for paperclips to reach their maximum efficiency, it's it's humans that are the problem. It's like if you don't have a paperclip. you've reached maximum paperclip efficiency.
And so the answer to the question, the only logical, yeah, exactly. The only logical answer to this is no paper, no paper clips, no humans that need those things. Problem solved, kill humanity. And so you have many manifestations of this exact story. I'm going to find a link to, let me find a link to the paper. I mean, it's just like Silicon Valley, that clip that always goes around where it's like, Oh, son of Anton decided it was going to delete all of our code because I told it to be bug free and it's like the only way to not have bugs is just to delete it all. Yeah, exactly. Okay, it looks like I'm wrong.
It was 2003. Nick Bostrom created this. Imagine a highly competent, super intelligent AI, blah, blah, blah, and maximize the production of paperclips. Phase 1, it automates and optimizes existing paperclip factories. Phase 2, to make more paperclips, it needs more raw materials. It begins dismantling cars, buildings, and infrastructure to harvest the atoms. Phase 3, realizing that humans might try to turn it off, which would stop it from making paperclips, the AI acts defensively. It neutralizes humanity, not out of hatred or malice, but because humans are a threat to its core objective. Eventually, the AI converts the entire Earth and eventually the solar system into paperclips and paperclip manufacturing.
I'm like, you fucking lost me at phase three, dude. Like, this was a thought experiment until you just, dude, no. Like, realizing that humans might try to turn it off. Like, we went from paperclip. To you having a weird ass realization or projecting a realization onto this thing. So anyway, that's the paperclip problem. And it has hundreds of manifestations that people go and they do the same thought process that eventually humans are the problem and the whole thing. So we might as well neutralize humanity to optimize our factories. And you're just like, come on. But people are buying this right now.
They're all drinking it. I talked to my barber today. He's like, dude, AI is going to kill us soon. I'm like, you too? Come on, man. Yeah, no, I don't know. I mean, it's hard to say what everybody thinks. I don't know if anyone really thinks that it's going to actually... It would have to take over... I don't know. I guess there's some people that think it could be like get into robots and like they'll attack everyone sort of like killing. Probably. Yeah. Yeah. Like because if you're giving a robot like, hey, clean my house, that's probably going to use some like some sort of AI to like figure out, oh, this is like a soccer ball.
I don't vacuum that up. Stuff like that. So, yeah, I don't know. I think that could happen or like just we need to care a lot more about security. I mean, like a lot of things now, like if you're using LLMs to build all of your bank software, like you probably also need security. There's going to be a lot of things that go wrong. So, yeah, we'll see. And that's kind of. Honestly, a little bit of the hope of Jev and why Jev is refreshing is like I wouldn't want to give my entire bank server LLM powers because we know they're sloppy. We know they're non-deterministic and we know they're influenceable and that they just, you know, they're like people pleasers.
They're not... rule enforcers is and so yeah jev is you know jev is potentially a like a a nice new moment in this evolution of models and it's it's like a model you can rely on and a model that can finally uphold yourself yeah we'll see Yeah, I mean, that's a big problem I'm having at work. Like I'm building some AI chat things and like trying to get it to respond deterministically ever. Like just, you know, I'm casually asking it to do a task of some sort. It's like, you know what? 25% of the time, I'm going to have no fucking idea what you're talking about. And I'm like, sick, I love that. And then like, oh yeah, the other like 75% of the time, half of that time, when you asked me for a table, I'm going to give you like pure unformatted markdown that you're just supposed to do something with instead of using the controls to ask for a table that you've given me.
Sounds like y'all need Jeff. You know, like, but the hard part about Jeff is Jeff just came out. Shopify is already, already like Jeff, we need Jeff here and here and here. And then security and everybody's just like, hey, they're brand new. And they're only at. And probably can't use it. Just like other endpoints. We've got to vet them and validate them and blah, blah, blah. And so there's a lot of red tape, but there's a lot of excitement. I think we, I think, I think open AI is. And Anthropics should be shitting their pants right now. For multiple reasons. They're about to lose a bunch of money.
And they need to conjure up a competitive model that does this stuff. Also, by the way, buddy, I'm on Astra all day right now. I did not see me changing to open AI ever. Opus has been my homie for, for months. Fable was really dumb. Like a day or two ago for me. Fable got dumb. It all got dumb except Astra, which is like blowing my mind. And so I'm like, huh, well, I need to switch then. You need to switch to Astra right now. There is a open source, open AI program, just like there was the one for Claude, which you should apply for that. I haven't applied for it yet, but I'm going to because my Claude one has run out.
So now I need another free... Send me the link, dude. Yeah, I'll switch Punkass over to it right now. Although my Punkass is very reliable. I will say that my open Claude setup is... I really, really like it. But for coding at Shopify, I have Astra is literally doing things right now that I'm, I'm going, I, I'm like speechless. I'm like literally speeches trying to tell you about the things that it made me speechless. Um, so it's, it's really cool. Anyway. Yeah. I used it for the first time today. I don't remember what for. I was just like, I was in open code. I was on my like sixth agent running at the same time.
And I was like, you know what? This one I'm going to choose Astra. No reason. Just cause. And it did fine, but I don't think it was like a hard task necessarily. So it was like, it would have been kind of the same either way, probably. Yeah. Let me paint a picture then for the difference that I'm seeing. I've been doing a lot of mobile web work, which requires simulators and emulators and stuff like that. Claude will use a simulator. There's APIs to use it. It'll even like have a cursor. It was so cute. It would spin up a thing. And I was, for example, and it did a lot of great tasks for me, but like one of them, I was making a swipeable menu that you could swipe it open.
And it would take the cursor and like move it over. And he's like little cute little bits. And then that was the swipe. It was like all slow and controlled. Astra shows a cursor on screen. I also think it's a little cheating. Okay. So let me real quick. I can't install the clod app at work, but I can install the codex app. And the codex app has computer use that is far superior to the CLI ability for computer use. Same thing with Claude. Well, the Claude desktop app does well. You just can't use it. It's on Mac OS, but on my Windows machine, I can't use the computer use. Anyway, the Codex computer use shows this little cursor that shows up.
It's got a cute little shadow, and it's like 60 frames a second. And not only did it use the gesture in the fullness, it could test a throw. It could test all the other types. And when it went to make the PR, it opened up. There were two things that blew my mind today. But the first one, it's like I've been using split tabs in Chrome. I've got the GitHub issue and my PR next to each other. And I'm like, yeah, look how cool I am. I can see these at the same time. um however i had a i had a tab open for the issue but i didn't have a pr yet it was a blank window waiting to be done it was like a chore i was going to do astra completed that it was like oh i see your tab is uh just like you have these other tabs with these issues and the p the matching pr on the right i'm just going to go ahead and open the pr for you and then put it in that tab split view for you and i was like sitting there going
What in the heck? I'm like, this is, I was like, anyway, so AI is getting wild, dude. Yeah. Yeah. I mean, I feel like anytime there's a new model, it seems really cool, but then it like degrades. Either we're getting used to it or they like, no, they do somehow have atrophy and like. Dude, the cloud atrophy is out of control. um the the fable atrophy is so visible like so many people said like when fable first came out the first few days incredible yeah and then it's been crap i used it last week to compare against astra it's night and day i was like fable was slow it was basically opus and i was like what am i doing with this tool this tool is expensive for no reason i'm like the task i gave it is kind of child's play and it's treating it like it's this extreme
No. So yeah, I genuinely think they're like dumbing them down on the server and like turning down intelligence to save money. I think so. Are you... I don't know if it actually makes any difference, but are you changing the reasoning level at all on either of them or just doing the difference? So Shopify is gracefully doing this awesome thing, which I really admire them for doing. And I'm sure it's expensive, which is that they want us always using Frontier models with Macs. Max thinking and they've been exploring local models. You know, like what if, what if they sent us all a thousand dollar box and we're running local models on it and that basically they're all like, it's not fun to you.
They're not anywhere. We're not the premium. We don't want it. This isn't a slop company. We want AI to be empowering all of our employees, but we don't, we're not, we don't want slop coming out. Yeah. And so I really admire them for that. And so, yeah, I use Astro and fable with extra high. I don't do ultra cause it's just too slow and it's for no reason. And plus most of the tasks I'm doing, even, even it's like right now we need to update to react to 19 and the latest react router. And I'm like, Ooh, maybe this is a good scenario for, uh, you know, ultra thinking and fable. And then I'm like, even that task, I'm like, actually it's,
I don't, I don't, I don't think that's what that's for. I think that's just a orchestration, a bunch of sub agents and then like a bunch of recursive, you know, checks like. I do think it tends to do better if you're giving, if it's orchestrating sub agents. Like if I'm having, if I just have like five fables each doing a task, they do worse than if I go, Hey, one fable, spin up as many sub agents as you need. And like, maybe it's just a context thing. Cause it's like not having to shove all of the context into every agent. 100% that golden window is super true. Yeah. Yeah. Although I got to quote Joel Hooks had this amazing tweet, which he's like, a context window that's getting really full.
It's like a seasoned pan, you know? Yeah. Like it's ready to go. It's like it's been cooking and whatever it's going to cook next is going to be dope. And I was like, that is hilarious. I wish it was true. To an extent. Yeah. Yeah. No, it's been doing some interesting things for me recently where like I guess it's auto compacting, which I don't think I have like explicit settings on. But it's like I'm never getting over like 50%. And it's also telling me, it's like, I'm so close to my context window. You better like start a new one or we're going to get fucked. I'm peaking, Poppy. And I'm like...
Slow down. I'm like, I see it says like... 42% like I think we're okay but there's like I don't know I don't know if there's could be an org no there's definitely settings like in in I've noticed this with fable it at various places like it can have a million context window but there is a setting that set it to 250 so I had to go explicitly enable the model fable with 100 or with a 1 million context window So it could be like a default setting coming from something. The full context window. I think it's just there are also settings for like when it's supposed to auto compact. So I'm wondering if that's set low somewhere that I didn't set it and it just got merged in.
And now like all of mine are using that because it's it's weird. It's bizarre. Yeah. But I'm going to use Astra now anyway, since you said it's good. Give Astra a shot. It's awesome. Also, let's talk about a couple other things on the show notes here. I've got Delta from Zed Team. Have you heard about this or tried it? I have not. This is one of those things you're like, Adam, I rely on you to do all of this initial research. Okay, so Delta, you remember Buzz? Yep. All right. Awesome. This is like Buzz and Zed had a baby or Buzz and Slack had a baby or not Buzz and Slack had a baby. It's like if Buzz and Zed or Slack and Zed had a baby.
Now confusing. Damn it. OK, anyway, it's very much a thread based multiplayer agentic environment. What they want you to do is load up a project and just like in Slack or some sort of chat app, you create threads. But in this scenario, it's a little bit closer to Herder. Are you using Herder yet? I've been on Herder for like three weeks. Okay. So Herder is all right. It's a multiplexer. It looks like shit. You can bring some plugins and make it act okay. But it keeps things running, which is the main thing. Yeah, that's the multiplexer aspect. I still think Warp is better, and I wish they would sponsor us more, because honestly, Warp, you are ahead of everybody, and no one uses you, and it's sad.
You are the best working... But if you're not using... uh, Oz, does it have a capability of like, so my, the way I used to do everything, I'd be like, you know, use the caffeine app or whatever, or amphetamine or whatever it is like to keep your everything awake. So you can like close your computer if you want and stuff still runs. But that's like the selling point of Herder, right? Is that it like just does that or like then also when you open it, it knows to like restart whatever. You don't have to like like it's supposed to be smart about those sort of things. Well, I didn't know. I've been using caffeine still and keeping my computer awake for all my multiplexing moments.
I mean, I don't know. because I've been using, I've been doing it at work and I think there's like hard, like because they control the machine, they won't let you install caffeine or amphetamine or anything like that. Yeah. They don't let you do that either, but we all do it. Yeah. There might be blockers stopping. Her can run while the laptop lid is closed. Yeah. So you get, you can spin up six agents, close your shit and it'll just keep going. Like that's the main selling point. Dude, that's amazing. That needs to be the headline of the episode. I don't think many people know that. Yeah. So that's like, that's what's compelling to me.
And I don't think warp has that. You can now, like, obviously the way around it with warp or with anything else is to use cloud agents because they're in the cloud. It doesn't matter if you close your machine, right? Like, um, and that's fine too, but I think this is really cool because I like to be able to run my stuff, you know, on my machine, but still close it. Like, In a perfect world, you would also connect somehow to it from your phone or another computer or whatever. Then you can run. It's almost like an open claw. You just run whatever you want and then you connect it with whatever else and just talk to it or whatever.
So yeah, I think Herter's got a lot going for it. I saw they raised $6 million. Toby, the CEO of Shopify, is an investor. Yeah. Okay. Who is also an investor in Warp, by the way. Yeah. You have to SSH into something to apply. There's like not a job application. So I was like, that's kind of funny. Yeah. Yeah, Herder is is great. And we got on this topic because we were. Oh, oh, Delta, Delta. Oh, yeah. So, yeah. So, yeah, Delta, Delta. But what Delta wants to do that Herder can't do, at least Herder isn't doing right now is multiplayer. So like inviting a coworker straight into your session. Yeah.
And I think that's I think that's actually coming. Is it coming to Herter? It's the future. This single player instance thing we're doing now is not the way to collaborate. And then having someone else's agent review your PR and then you copy and paste that agent or you have your agent go look at the other agents review. That's going to die. That is a slow clunky moment that we have right now in software. But Delta Delta is cute. It's minimal and it has a diff engine built in. You can easily punt to Zed. You can easily go to the terminal. easily invite a co-worker into a thread that has multiple terminal instances in it.
Like it's doing some really interesting things. It's also, it's not getting enough airtime in terms of like popularity at the moment, but I think there's something there, but also, yeah, sure. I think Buzz also didn't get the recognition it deserved. But look at me. I'm like, Warp and Buzz and Zed and Svelte. These are great things that no one uses. And I'm just like, I'm not popular. The problem with everything cool is most engineers are busy and working at a big company who will not let them use all the shit you made. Like, yeah. So like, there's no way I can use buzz for any meaningful thing other than a demo to play with because I can't use it at work.
Like, so this stuff is cool, but it goes nowhere. Like if it's supposed to actually make money, then you need to lead with like, we are enterprise ready. Here's how you can like sell it to your orgs. Like, that's what I need because otherwise I can't use it. We forked it, hooked it up to our internal agent called River and all of our other agents. So you can invoke Claude or Astra or River all from what looks like Slack. The hard part is moving all the damn humans to it. Like it's it's working. I can install it and try it. And but I can't get coworkers to get out of slack. Like there's just I love slack.
You can pry it out of my cold, dead hands. Really? Man, I know slack is like backboney right now and they should feel really good about that. But like it, it's got shelf life and I'm going to cheer the day it dies. You know, I don't know. All I know is I never feel good when I'm chatting in discord, but I do in slack. That's hilarious. I don't, I can't, they're like, you know, basically clones of one another. Yeah. Discord is a little more game leaning. It's like darker themed, whatever, but it just doesn't feel right. Maybe it's the notification sounds that I'm used to for the last like decade or more or whatever.
And like, I don't know. I'm just, I like slack. And I think that's like, that's what everyone's experiencing. Like these huge companies don't want to be paying Slack tons of money to use it. No. Um, but they have to, cause everyone is like, I won't work here. Like we got to have Slack. It's crazy. Oh man. Um, I, I like them and hate them both the same. I don't, I don't have an affinity to one over the other. Um, Yeah, seriously. But Shopify did do it. And I can't say exactly what, but Shopify did it again this past seven days, took a humongous piece of corporate paid enterprise software and said,
not using you anymore we built it ourselves and that would be about the fourth time i've seen him do that in the past year um oh yeah this is wild this reminds me okay so we've talked about this um i think where it was like we were using github copilot right and then they were like uh it's wednesday at 2 p.m turned off And you're like, wait, everyone's in the middle of their workday. What do you mean? Just turned off. And so everyone scrambled to use Bob, which is like IBM's thing. And we did that for a bit. And then like in the last couple of weeks, we got an email from like one of the directors or whatever that was like, hey, yeah, we fucked up.
Everyone can have co-pilot back. And like all of this stuff. And it's like, Oh, cool. So, so we've been like flip-flopping tooling and like, um, I think that's made it really hard to make any sort of like real meaningful AI traction. Like, you know, you guys are using, spend whatever you want, use the best frontier models. Like you're still kind of a startup vibe. Like you can move fast. Like we have a lot more red tape at IBM and, and stuff that's been happening. But, um, But we have cool tools back now. Like I can use Astra through Copilot with open code. So that's what I'll be doing. Wait, I want to premise something.
My pi usage of clod and Astra. So my pi usage of opus and Astra is poorer and more expensive. than if you just use the codex CLI. I haven't used open code as much, but I'm of the current opinion that the harnesses built for the models are superior to the agnostic harnesses that allow multiple models. I went back to Pi. Pi. Um, because there's like a pie craze going on with Toby and other folks at Shopify. And I was like, what am I missing? I've, I've used pie for a long time. I was like, I was one of the first people with pie. I love pie. And then I was like, uh, I don't need this anymore. Uh, it's kind of like getting crappy.
I'm noticing that Claude is better. And I went back to pie and I doubled my cost. Um, and did not increase my productivity at all however using codex cli specifically with astra is superior to pi using astra and i think that is easily explainable and so i just want to i just want to both to do that yeah it's like how uh you know apple builds hardware and software it's like That's how they work well together. They know the strengths and weaknesses of their models. And so they'll play into those. Yeah. Yeah. Yeah. So it's always going to work better, even if it's like they might even be explicitly just telling it like, hey, if this is coming from our stuff, just work better.
Like that could be built into like the core directive of like how they work. And we'll never know. Yeah. I mean, Dax knows he probably is feeling the blunt of that negativity. Like as, as he compares the same prompt between two harnesses, his and the one that's native to the model. And then going, what the fuck? You know, like, how am I supposed to compete with that shit? Like, um, and Ply's got to be doing the same thing. Yeah. They're like, I don't, I don't remember the specifics, but he, he tweeted something about like, you know, they have a lot of free or low cost options built into open code, um, that aren't like the frontier top models.
Yeah. And that they were like one of the cheapest that was still like making profit or something, or maybe the only one making profit. I forget what it was, but it was like the way they've, you know, finagled things. They're actually like profitable with selling like super cheap tokens to people or whatever. And I'm like, that's compelling. Like no one else is doing that. Everyone else is just burning money. Um, So, hey, I don't know. I think they're still doing a great job. But I do agree that it's like it used to be before they had these like really nice harnesses at like OpenAI and Anthropic and stuff.
OpenCode was like 10x, 100x better. And now it's level now. Yeah, they were adding to the experience. Yeah. All right. So we spent a lot of time on this. We're over an hour, but I do want to, uh, I do want to do a little bit of whatnot before we end for anyone that's still hanging out. Um, something that I'm most excited about that's coming up soon is the Ocarina of Time remake. Um, I'm personally going to take that entire week off of work because it's also right before my birthday. So I'm going to be like, Hey, I'm just going to, I'm just going to play this till I beat it. And I'll see you guys in a bit.
Um, So I'm excited for that. I don't know. Are you a Zelda fan? I'm a huge Zelda fan, but I am going to make a statement here and it's either going to crush you or push you into your happy place. Did you see Theo's upscaling of Super Smash Bros to 120 frames per second in HD Super Smash Bros work? No, I did not. It's... Very good. It's very high quality. So essentially what happened, he used Astra, gave it unlimited resources. I think it took, who cares? Who cares? It wasn't that long, honestly. And it wasn't even that much money. And it went through every asset, every interaction, every line of code, and upscaled the entire experience as well as made it run faster.
And it's beautiful. So I'm looking at the new Ocarina of Time going crazy. You want me to buy a Switch 2, which means that you could rip any game out of my library at any time because you're exclusively digital now. And you want me to go buy, what is it, an $80 game? You want me to buy an $80 game? I think it's $70, but yeah. $70, okay. But meanwhile, open source could take the ocarina of time that exists now. give it to Astra, and make an HDR 120 frames per second version that I don't have to... I just play it on my PC. I'm going to save my money and not play that game or buy a new Switch. So, okay, on several levels, you need a Switch too.
You don't know until you have one. No, dude, honestly, I went to a cabin two weeks ago and we were going to play a game called Killer Queen. And it's a game that one person... Okay. Killer queen. Dude, it was on the store. It's not there anymore, which means we went to go buy it or play it. And we couldn't play it because they pulled it from the store, which meant hard. Anybody's digital. If they, if they had it downloaded, they were good to go. If you ever uninstall it or if people want to get it, they can't get it again. Dude, if they had the hard... That shouldn't be allowed. If you paid for it, or was it free?
I would have paid for it. Easy. If you pay for it, they should not be allowed to take it out of the store. Unless they give you your money back. That's like an iTunes movie or a Google movie. If you buy a movie, you don't own that movie, man. You own a digital moment to access it. As long as iTunes still exists, you can still re-download it. Not if they don't have the rights. Sure. So if it gets completely, yeah, that's true. And they're not going to refund everyone at that point. You have a temporary rental to a movie that they made you pay like you owned it. It's fucked up. And the Switch 2 is doing the same shit.
And I'm not here for it, man. Yeah, I'm sure they are. Yeah, I'm sure. Yeah. Like we got a Hogwarts Legacy because it was like... It's a good game. The Nintendo Switch 1 version is like, you know, like seven 20 P or like whatever. And then the new one is like at least 10 ADP or, and like runs faster. And, um, like it's worth it just for that for me. Like I'll replay stuff that I played before just because it's like, it looks prettier and runs better. That's what I'm saying. That's like upscaling too. Astra is doing that now. I'm like, you don't have to pay for it. Or you just got to wait for someone to run the machine.
And, but I'm not going to play it on my computer. Like I'm a console guy. Oh man. I'm a, yeah, I haven't had a, uh, console for a long time. Okay, let me find this video tweet and I'll just send it to you. Smash Bros. Upscale. Alright, was there anything else we wanted to talk about? I could like speed run some, uh, whatnot here. Wait, okay. I just looked at the list. Yeah. It says arrow went through my living room window. Yeah. Okay. Tell me about that. Okay. So yeah, my kids went back to school. I'm at a year at Shopify. There's two of them out of the way. My, my clothing brand is called Osweary now.
It's going to come out soon and it's looking awesome. Um, Okay. So an arrow went through my living. So, and this is like my, I have like a decently modern house. This is the main window that if you're walking up the stairs to my house, this is the big, huge main living. It's got a hole in it right now. And it's still, it's still broken. And there was a it's like a couple of weeks ago. There's like a full moon and an eclipse happening. Oh, there was an eclipse happening. We had my kids friends over. Everyone's over. And my oldest was like, I want to I want to shoot the bow and arrow. Can we get the bow and arrow out?
And I was like, sure. And it's like a kid's compound bow. So I go get the kid's compound bow. It's got arrows. And we set up a cardboard box. And of course, the kids put the cardboard box faced at the neighbor's house. And I'm like, no, no, no. You can't shoot at the neighbor's house. You got to shoot towards our house. And I put it like towards our house, but like down low, like really low. Like you had to be an idiot to shoot my window. Of course, I was a fucking idiot to shoot my window. But like... I give every single kid, there's like eight kids. I give all of them a lesson. And I'm like, here's how you hold it.
Look at the compound bow. Once you pull it back strong enough, it locks in. You got to look down the arrow, like hit the box. Everybody hits the box. The eighth kid, I don't teach. He's the oldest one. And I go and I grab the banjo and I literally like sit down and I go bong, bong, bong and window breaks. And I was like. I just, I just sat down. I'm like the one kid I didn't get the lesson to. He basically goes like this. What's the compound body? Yeah. You know, just like doesn't even just just can't handle it and just just launches it. Right. Yeah. I feel like window. I deserved it. You know, bow and arrows, guns, knives, anything that like could be really dangerous.
If you don't know what you're doing, just don't touch it. Like, I mean, I guess kids don't know that, but like they don't know. Oh, yeah. How hard could it be? But like 10 year old kid and he's the big kid. And it's like and it just like it's just like. sprung out of his hands he just did he just had no no idea and i should have been there and i just thought i could trust him and i i messed up yeah yeah that'll happen is it uh is your window is it like several small panes or is it like one big piece of glass it's a humongous It's like it's non-standard big double pane window and it pierced one of the double panes.
So the good news is it's like I'm currently just the outside piece. No, you have to replace both because they apparently they put gas between two. Oh, right. Yeah, yeah, yeah. And so I have to have someone come out. They're going to have to get a crane because like that window is like up high. Dude, it's just build that kid's parents. No, they're totally not. I've had zero claims for my home insurance. And I'm looking at it right now going, is this going to be my first? It's like I could either eat the cost. Dude, it's super annoying homeownership stuff. If it's a non-standard big window, that could easily be like $10,000.
I don't know if you've gotten quotes by the way. I've gotten quotes. They're not small. The question is, is like how small do I want to eat my first homeowners insurance grace period claim where they're like, Oh, you need a claim. We're not going to raise your rates because it's your first, you know, it's like, I only get one first. And I'm like, is this enough to be the first? I've never made any claims on anything. Like, I always want to, but yeah. They scare you. They're like, oh, your rates will go up. Or like, next time we won't pay for it. Or like, yeah. What's the point? Like, give me real insurance.
Anyway. Last bit of whatnot here is go to my website. I have had numerous updates, and one of the best ones is an anonymous like button. You can like button any post or any blog that I've written from your IP three times per post, and it will still give you the sensation of liking it as you continue hitting it. Hi, hi, hi, hi, hi, hi, hi. It'll shake and do all its cool little animations, but it'll only track three of them, and that's using Dino's KV storage. It's pretty cool. It took a stupid amount of time to build an agnostic like button on my website, but I'm feeling good about it. Nice, nice, very cool.
All right. Thanks, everyone, for listening. If you liked it, please subscribe. Leave us some amazing reviews. We appreciate it, and we will catch you next time. Congratulations, citizen. You are now prepared to face the challenges of the modern technological world. With the proper tools, a steady hand, and a reasonably configured terminal, any developer can orchestrate AI agents with confidence. You'll do just fine. Thank you for listening to Whiskey, Web, and Whatnot, the world's most important AI podcasts. For additional resources, including official merchandise, visit whiskey.fund. And remember to subscribe, rate, and review so this vital information may reach others.
Because the future of technology depends on it.