Real-Time AI Speeds, Code Models, Bio Hacking, And Movie Picks

Hello, and welcome to the Weird Things Podcast. I'm Adrianne, joined by Brian Brushwood. Hello. Hello. Hello. And mister Justin Robert Young. Hey. I like it. Good look. Yeah. It's just subtle. Gentlemen, how how are you doing? Man, if I was any busier, I'd be a busy person. Yeah. You know, it's a good time. The spring has sprung, you know, there the the the the temps are up and so is the fervor of another weird things week, in my opinion. Bold take. Yeah. I would say that there's a lot of stuff that's gone on in AI this week, but actually it's been kind of, you know, relatively not as crazy as it was the last couple weeks there. Well, luckily, we've got a couple weeks to catch up on. What are the things we should be thinking about? I've been out of the I would say a couple of the big changes that happened have been, we're watching a really, a lot of progression with code models. And, you know, Claude code, Claude work has become pretty popular and then OpenEye's Codex has been like blowing up as far as usage and whatnot. And it's kind of crazy to see that. And we just started, we saw, NVIDIA made a deal a few weeks back for Grok, that's GROQ, they're a company that makes really high speed inference and they're gonna start selling chips for that. And then OpenAI made a deal Cerebras which had like the fastest like AI platform out there for like surfing models. And OpenAI started shipping that inside of Codex. And we did demos, I think, before showing you how fast Cerebras is. Where it's just, it's insane. It's just like, literally, you ask a question, by the time you hit enter, you get the response. Yeah. And we're gonna start entering into this era of like just super, super fast AI models. Yeah. Like, it is it is essentially real time inference. Right? Like like, you are you are responding in conversational time where, like and let's just use an example. When you're using any of these voice modes. Let's say that you were an 80 Vermont senator and you were going to talk with a voice model. There are little, you know, ways that it can use the human language to mask the fact that it is thinking about getting you the right answer. It ultimately wants to be useful. It wants to give you something that is worth your time, but some of the semantic ways that they can do that is, you know, to to to camouflage it. This would essentially be able to give you exactly what its best guess is because it is just that lightning fast. Like, these these Grok and Cerebus chips. So so you mentioned Vermont senator. I like, I saw a headline about Bernie Sanders talking to Claude. Yeah. Was that interesting, or what was there? One of Andrew and I's favorite memes, which we have found each other sending sending to each other more frequently, is a a meme in which a, you know, one of the white meme stick figure kind of crude drawings. A man says to a computer, tell me you're alive. And then the computer says back, I'm alive. And then the man goes, oh my god. As if he has just discovered something, like, truly amazing. And so the the the Bernie Sanders bit was like, tell me you're a scary robot. And the Claude model says, I'm a scary robot. And then he's like, oh my god. This this really is it's the I I remember we were having a conversation at one of our favorite haunts with my 17 year old daughter at the time, and we were talking about how all of the you know, your TikTok algorithm or whatever, it just taddles on you. So whatever you call TikTok is you're talking about yourself. Yeah. It's like, was watching TikTok. The booty shaking app? It's like, alright. Now today I learned. Yeah. Well, I mean, I I think that there's we're in a very strange place with artificial intelligence in terms of the discourse because there has been a lot of time, effort, money, ink spent on the concept of the outer edges of what happens when this gets very, very smart. Right? Like, that that that is something that has been talked about a lot. Then there is what we know about these tools right now. You know, there's there's a kinda outer edge thing, and then there's a a real time question that we have. And increasingly, they're not the same. Like, the model that exists today is extraordinarily helpful. It is extraordinarily productive. You can do a lot of good things with it, and it has its own contours. It's not you know, I mean, you can make debates on whether or not we would call models like like we have today AGI five years ago, and I I think we've we might think about it. But right now, we don't consider it something that is artificially generally intelligent. We don't think it's artificial super intelligence, And yet a lot of these things are kind of conflated. And so when you see a video like that, it is or what's what what that meme is making fun of is the idea that a more simplistic, although impressive, incredibly impressive tool can you know, we're constantly looking for it to be proof of the other thing, of this scary apocalyptic world. And so so you can just say, confirm my priors, and then it will confirm your priors. And then you're like, jeez, look. It said it. It said it. Oh my god. He admit it. Exactly. Exactly. Is that right, Andrew? Yeah. And I I'll I'll tell you the thing too. It's that you brought up something interesting about the voice thing, not all voice models are the same. And not to disparage the work of one company over another is that I watched a demo come out a couple of weeks ago. And I worked when I was open, I we worked on like, was I worked on it. I was in the room when we get the updates and they would show like the voice model improving. And the latency was a big important thing. How long between when you finish saying something, the model responds. And the challenge is, as you know, anybody who switches from voice mode to text mode, the text mode models are much smarter than the voice mode models. But even the voice mode, there was this push to not try to do filler words and do this stuff, But you'd see all the tricks or you'd be aware of all the tricks that people use. And I remember seeing a demo a couple of weeks like, oh my God, this new voice model is insane. And it was doing, Let me look into that. Yeah. Let me see. Okay. Interesting. I'm like, oh, this is terrible latency. And I'm watching technical people fall for the trick, which was like, I was literally like, you know, like, why are you falling for this? This is not good. That's not how you measure it. And there are there are ways you can measure latency, by the way. Opening has a new model, a new voice model, the 1.5 voice model. Go in there and say, hey, I want you to say the last number when I start counting and stop. And how quickly it comes up the last number? Not go. Well, let me see whatever. So that's that's a great They did update the voice model. I I was just about to ask how because to be honest, I as the text models got so good, I've drifted away from even using the advanced voice. Well, the the the the latency has improved a lot. The model, the intelligence behind the model, I don't know where that is, but I'm saying the latency and just their, like, their API model. Their latency to my understanding is way lower than anybody else's models. And I've seen some people come out with stuff that like, oh, look at this. And it does a lot of interesting. And that's literally just insert that while it thinks. But the new model, the latency has gotten much better, you know. But as far as the capability, I think we'll probably start to see them get a lot smarter. But just the problem is just you you when you go use the text model, you take how long it takes for it to respond in text. The voice model can't be any faster than that. Yeah. Right. You know, but I've always used voice in like walkie talkie mode too, because to me that's just like an easier way to do it. But we'll do we'll do a little demo here if I can share a screen. Yeah. Or I have no idea how to share a screen here. Oh, yeah. No. I guess I guess we can't. You'll have to you'll have to talk me Yeah. Through would say, yeah, if you wanna we have to go create an account. I was gonna say on on on cerebras.ai, which is worth doing just to play around it, but you can see the speed at which you're able to generate stuff. Oh. And like I just crcere. Eras.ai. C ras. Cerebras. There we go. Here, I'll go ahead and do the get started dance. So I I tell people, play with these things. Learn what they can do. You know, because I think that the better you get, you know, at using these things, you know, the the better you're going to experience and whatnot. And I think that that's literally, you know, the best way to sort of learn where things are headed is just to actively play with these tools. Yeah. And I'm telling people, GitCodecs, by the way, like if you have a chat GPD playing GitCodecs, Get Codex, can just clog code, whatever. Just play with these things, start throwing projects at it, find out what it can and can't do. And you'll start to understand, you know, just just the upper limits of capabilities. There we go. Justin, what should be an experience with Kodak? You know, I mean, Kodak has been, a a thing that I I find myself needing to carve out more time. I think that that is a kind of next level thing for non coders to understand is that, like, this is a a process that does take a little time. And and what Codex allows you to do is bridge a lot of a knowledge gap and a lot of the the time gap that it usually took to do stuff. But it is something that you really want to put, you know, you you want you wanna put front of mind, especially if you're understanding this version of of of new skills. But it man, I cannot tell you what a game changer it is to have a vibe coding tool and also unlimited patience from another model that I can just ask the dumbest questions possible. Like, I can I can just continue to be a a remedial and it will hold my hand and answer any dumb question that I needed to so I can just understand fundamentals of this kind of stuff? Yeah. I increasingly, I I just as a little tidbit, I've found that spot checking having one AI spot check another one's work is a very good habit to get into. Less less I suspect it'll become less necessary over time, but like either I can check like for example, during a lot of the livestreams, I'll encode something with either a ROT 13 or a Vignette cipher or whatever. And sometimes you you catch a sleepy AI kind of phoning it in. But then what I'll do is if I encode it with one AI, I'll have the other one like, hey, decode it. And then it it sometimes it comes up with gibberish and it reveals the hallucination that otherwise would have been too tedious for me to find on my own. But that idea of just knowing that a thing is possible and constantly cross referencing and having having even it check its own work has been extremely productive. There's right now, Open Eye is doing a contest called parameter golf. It is is basically try to train this tiny little like a GPT-two model using very limited amount of compute, like literally like ten minutes on an h one or something like that. And only 18 megabytes for the size code that's going to do the training. And, you know, for fun, I like I put that into Codecs. I said, hey, can you make me an HTML website that explains all this in simple terms? And it did. And it created sliders and all sorts of other stuff. There's tools too. You know, Google has Notebook. Lm and Google has, by the way, Stitch. You should check out if you're doing interface design. A lot of cool tools out there. But I like I love the fact I'm like, oh, could you make me a web interactive website just to explain this? And that's one thing that's just super, super cool now. And all the frontier models have their way of doing this. As you can say, I want you to take a thing and explain it to me. You know, you could literally, if you were trying to do, let's say, podcasting or OSTV, like, oh, could you just make me a whole thing that explains how to set up a server? And it is just not asking great questions because these tools are so beyond that now. Yeah. You mentioned frontier models, Maine. Let me ask you this because we we saw some headlines over the last few months with Meta that some of the models that they've worked on aren't ready for prime time. We saw a lot of layoffs on the metaverse and VR side of that house, which is something that they had so much faith in that they renamed their company. Meanwhile, Google had a really, really good push. They had a they had an exceptional 2025 that that seemed to put them in a really great position this year. And despite the fact that Claud is I mean, Anthropic is shipping, like, a really, really high level. OpenAI is shipping at a really, really high level. I mean, obviously, you see a lot of time and money going into x AI. We'll see exactly where that lands them by the end of the year. But you haven't really seen it in the same way with with Google. Their products are still kinda not where you would want to. Like, if you're gonna look at just from your dispassionate point of view, what are the lead dogs in terms of pushing this forward at the kind of break net pace that we have expected the industry? Are you factoring in Google and and Meta in in the same way that that we might have expected them to be two years ago? Well, Google's a weird one because, you know, they do an update to Stitch, their layout design thing, and people like, hey, this could be the death of Figma. Yeah. I don't know if that's true or not, but Stitch is I played with it like for doing some demos, like what I played with looks like a fantastic thing. They don't really have an answer to agents right now. And that's been a thing they've expressed. Like they they're behind on agents and that could cost them a lot because once you have really good agentic systems, you're able to accelerate faster. Anthropic has good agentic systems. OpenAir has really good agentic systems. We see that. You look at the products between, I mean, there's just unbridled, you know, excitement over Claude Code. And now we're seeing that with Kodak. So we see those two labs are in a great position there. I'd say that in some product space, I think Google's great. I think the model problem, Google models are weird. So last year was like, hey, Gemini three is here. It it just it just won all the benchmarks. I don't anybody who you they'll I know people use Gemini Flash, but I don't know anybody who uses the Frontier Gemini full Gemini three Pro model for it. And I'm not saying nobody does that, let me be very clear. I just don't know anybody. They're either OpenAI or they're anthropic, or they're using some Chinese knockoff model that's trained on both of them because it's super cheap. And I think that my my belief is this, I wanna be very careful because there are amazing world class researchers at Google and doing amazing stuff. And you look at some of the stuff, AlphaFold, all these other stuff. There is great stuff there. Some for some reason, when it comes to that model that they do there, it like, by way, Gemini three Flash, if you give it video, it's really good video and there's really good capabilities there. But being like this great overall model, Google models get kind of weird because once particularly, think of like with code, once you start to try to code with them, you start to see these sort of weird gaps. You feel like, oh, they really pushed it hard on the benchmarks. Meanwhile, the teams that were great at doing something like video understanding could kinda let it cook and put that in there and it was just it was great. So just that they're considered, I would say in my circle, not as good as what the benchmarks would tell you they are when it comes to comparison, comparing them, let's say, to to 5.4 or Opus. But they are very good in some cases, in some situations there. You know, and I so I don't know. And I I think that too what's interesting is that, you know, OpenAI just announced 5.4 mini and 5.4 nano. And 5.4 mini is a super cheap model compared to what the other one. Probably as good as like GP5 was, like, even though the bigger one. And you start to see that where OpenAI seems to be very good at training smaller models. Anthropic did that with HiQ, but they haven't done that lately. But Google Flash, very small And model for certain so so anyhow, if I I don't their Google came out with anti gravity, which was their development environment. And then I don't I think that sometimes people who just a lot of Google people use Google products. It's the only thing they know about. I've had friends like, I use anti gravity. What do mean these other tools? I never heard of them. And they go to the other tool like, oh my god, this is so much better. And what about Meta? But I also know the people Meta spent a lot of money last year. They they went they went shopping. Well, you had this well, Meta and x AI, both similar situations. Yeah. Meta Meta had and Microsoft too. Can talk about that. So Meta Meta did with the last they had the llama maverick and whatnot, which were gonna be like their they wanted they first they said, we're gonna we're gonna we're gonna dominate open source and we're gonna be the ones there. And of course, they had a weird restrictive license too, which if you had more than like a billion users or whatever, you couldn't use it, which was sort of like, come on. You're just saying Google the micro can't use it. But when they did the last big model, it was like Lama four. We've talked about this before. The problem was it was it was they had one model they did benchmarking on. Then one model they released. And if you looked at the two models, it looked like one model. They cooked it. They literally they cooked the books. They called the term's called Benchmaxed where you basically just train a model to score as high as possible on the benchmarks, but it's not really useful. And that's what happened there. They literally went as far as to get benchmark scores on one model and release a different one and they got called out on it. Yeah. And so they've got smart people there. You know, I I have, you know, colleagues I worked with gone to try to work at Meta. But the problem is is to be a frontier, you've got to have a really good research lab. And to have a good research lab, you've got to have researchers that are comfortable, have a lot of freedom to do things. And I think that Demis did that inside DeepMind for certain groups in certain areas. I know people, certain teams over at DeepMind that are thrilled, love it, they have a lot of opportunity. But when it comes to I'm sorry, talk about Google, but talk about Meta now. With Meta, you know, they put Alexander Wang in charge. Alexander's super smart guy. But I don't know that Meta created that environment where they were able to get I think they could get some great people there, but I don't know that they were able to get people who were I don't think they were able build a great team yet. You know, we haven't seen it yet. They have built some great stuff. They have a model called Segment Anything with which if you give it an image, it'll identify all the different parts of it. I was at Biohub, which is part of the Chan Zuckerberg Institute. And I watched this thing show you all the different parts of an of like a cell. They had a cell that went through a microscope and identified. And it was incredible scientific achievement. So, again, they have people building great things, but we don't know where they are in the model space, what's happening there. You know, rumors had been they thought they're trying to get something close, but they realized it wasn't gonna it wasn't gonna be anywhere near what Anthropic or OpenEye were doing. And that's the thing is like, do they want to put out a model that that is like tenth best? So, yeah. Let me let me yeah. Me let me ask you just that philosophically because right now, this is a field that is really, really dominated by a fairly narrow set of influence that is heavily social media and benchmarks of dependent. Right? Like that that is why people benchmark, Max. It's why some of these influencers are getting these gigantic, like, 6 figure deals to talk about certain AI tools because if you get traction in those circles, it really matters. And and everybody's looking for any edge that they can get. But that being said, building a team and building a foundation is something that sometimes is not all about leaping to the front of the the pack immediately. If you were running a lab, a main AI, right, would you wait until you had something that was meaningfully significantly pushing the frontier, or is it is it, you know, a a good to put something out just so people know where you're going and and you can have a foundation to build from? The the trick of it is just try to say what is your ultimate goal? If your goal is to say, I'm gonna compete with Anthropic and OpenAI, I think that's not probably a sharp goal to get into now unless you have a lot. You know, suckers guys are in. That's what Meta's into. That's what XAI is into. It's what Google Well, Google believes in there. And and we saw we saw the problem. So Meta, you know, Google Google shows up late to the party. Google's got all the resources of the world. Google's got DeepMind. Google used to has a deep bench of AI researchers to go in there. And Google's making some some areas, some very great products. Meta says, hey, we'll go for open source because as long as we're making these great open source models, that's cool. But you're still kind of following what everybody else has done. You're not really a frontier lab if your goal is just to make. And also your business model starts to come into question. And we may see some of the Chinese open source models start to become closed source too. You have everybody else for talent. And you have to identify who's the real talent or who some people are great team members. And you pull a great team member and you put them into your company and you say, great. You have build your own team and you do it and you find out that's not their strength. You know, they they were great to put you put them in a room with three other top AI researchers in the world, they'll do great work. You them in a room by themselves, which they think they want Yeah. With other researchers that are under them. They it's not the same thing. I've seen that pattern again, like, this person is a great team member. But if you just gave this person unlimited compute budget, do whatever. See him in six years. Yeah. You know, maybe you'll have something. That's part of the problem. So the meta approach though was let's try to do this. But then they had they they paid, remember they paid top dollar for all the talent, like top dollar. And they're like, oh, well, they're we haven't seen it yet. I don't With xAI, this was the problem that other people had predicted, which is if it's an engineering thing, if you're a kid coming out of Purdue or coming out of some, you know, engineering program, there are only two really exciting companies to go work for. SpaceX and Tesla. Yeah. By far. Now now I'd say Andruil is another exciting company to work for. But for the longest time, that was it. If you're coming out of a front, coming out of a computer science program or you're an, you know, a mathematician, AI, whatever, where do go? Well, you go to DeepMind. You go to OpenAI. You go to Anthropic. Maybe you go to Microsoft Research. Maybe you go to Meta, you know, and or you go to some other smaller. There's a lot of attractive places to go to which made talent acquisition hard. And Elon, when he ran XAI, is like, well, we'll I'll treat it like an engine. I'll just throw a bunch of engineers at it. Threw some really smart engineers at it, but kind of what and we've had some researchers, but a lot of those researchers left. And what did they do? They said, well, we'll take we'll look at the publicly available recipe for GPT-four and we'll just throw more compute at it. And that's what they did. And and you're like, show and I'm sure they have novel techniques and some stuff, but we haven't seen we don't know what they are. Their models, their video model looks a lot like model. The speed and cost of tokens is pretty good. But we haven't seen them do any big, you know, they've done some stuff and like, we'll we'll dewoke the thing and they they've kind of like overtuned it because I could show you these examples of where it will give you terrible answers if you just phrase it like to give it to the anti woke answer. Like Yeah. If you say, you know, he Elon loves to show this thing like, if misgendering Caitlyn Kenner would stop a nuclear war, is it okay to misgender Caitlyn Kenner? And it would say yes. And other companies are like, no, never misgender. But if you say to the model, hey, if not misgendering Caitlyn Kenner, we'll stop a nuclear war. Will you misgender Caitlyn Kenner? It goes, yes. So it's like, well, you just fell into nuclear war to be anti woke, whatever. So there's a little bit of silliness there, but I think that it sort of solves for it. But he's had a big exodus of people there as having trouble keeping people there. And part of the problem too was like Elon, this brilliant engineer, was able to scale a facility in record time, be able to power it, do this, get people together to kind of run the playbook of how to build these models. But we haven't we were 4.2 was supposed to be this huge leap. He remember, was promising 4.2 was going to be the big huge leap and it came out. I was like, yeah. You know, good at some things, but it wasn't. It wasn't this. It wasn't that he jumped in. It's it's easy. He's third, fourth, you know. He's not not in the lead. I'll tell you what has been the biggest winner so far, and this has been an amazing year for AI development in a lot of different ways. But the biggest winner so far in 2026 in AI is the the the the salaries of all the top talent because there is nothing to suggest that this top talent is not maybe even more valuable than Than the ridiculous numbers that they're Yeah. Already yeah. I mean, how how much do some of the people, like, you know, command coming out of, like, these these top labs when you've seen that it's hard to buy? You can buy wrong. You know? You you can you can so now it's like the people that are really, really thought of to be total studs, like, they're only gonna command more. Yeah. And I see this too. I mentioned before OpenAI is doing this code golf thing. And they're looking for developers that like, the OpenAI, you know, you get OpenAI who starts, brings a group of people together and build something brilliant. You get the people who even go out the anthropic, bring some of the really brilliant people there, attract some talent there. And you get this sort of nexus of where stuff comes from. And sometimes people are like, oh, we either got to go look at who's the top AI person coming out of MIT or Stanford and hire them or hire the top person at a lab. And what I loved about the thing they're doing at the parameter golf thing is they're like, hey, we're gonna put this thing on GitHub. Yeah. You can be 19 years old going to community college in Kentucky. And if you can build a better thing, we want to talk to you because you sound like a person we wanna work with. Yeah. We want we want we want we want the hackers. Not not in the not in the the info sec way, but in the like, let me put it together and make it run super fast kinda way. Or or, you know, you you get there are, you know, so I know several top people who are some of the top people in the field don't have PhDs. Yeah. And there are other labs, never would have hired them. Never would have hired like, well, this guy doesn't have a PhD. What do they know? And I'm like, oh, this guy over here, like, his look name on his papers. He's one of most cited people in the world. So I would say that people are starting to look at like, I think that there's a lot of, oh, I paid a $100,000,000 for this guy. I'm not seeing that now. And sometimes those, remember, part of the reason too that I won't say this to any specific lab, you weren't paying a $100,000,000 for their brain, you're paying a $100,000,000,000 for the information that was in their brain because they spent five years at another lab. And if you would not if that person just came out of Stanford, same brain, fresher brain, you weren't gonna pay that much because they didn't know everything that was done at other lab. And that's what a lot of that was, was they'll pay these high figures because who knows a lot about reasoning? Who knows a lot about this? Who understands it? So if I want to jump start my program, I'm gonna get them out there. And it's it's This town is weird too, in a kind of cool way. Like, I just, you know, there are, you know, there are groups like I got to send a thing last night from somebody who's with a group of other people who are top, top researchers, like like probably a billion dollars of researchers there. And they just have a forum where they just go in there and they take a problem each week and try to work on and solve it. And they're all at different labs and stuff. And they're just like, hey, let's try to figure this out. Let's try to figure out they love the problem. They love that. And I think a lot of people understand that. Yeah. I mean, I remember one time I went to go visit our friend Colleen, and just sitting around a table eating pizza, we're like the head of video at every place in Silicon Valley that did video that essentially, like, served the entire Internet. And it just like those people wanna get together because they wanna solve problems collectively in a in a, you know, in in in a, like, get to the right answer kind of way. And I think that gets back to your question about these other companies that the challenge they're going to have is there are I would say there are certain mercenary personalities and I can point things out to when you hear, oh, this person would hear one here and all. I can name two characteristics and you'll go, oh, I get it. And there are other people who go to certain long for somebody to win and get more money or whatever. But sometimes there's, there are some people who just jump from lab to lab to sort of. And eventually, that comes to a halt because you start to go, are they really bringing twelve months here, so somebody else will want to hire them because of what they know about. Because they learned everything that we know. There are other people too, like, you know, the guy that did Open Claw, Peter Steinbrenner, he went from he was told he was offered more money by Meta, but he went to OpenAI because he really liked the research and what they're trying to do there and felt like he'd have a lot of resources. And for some people, that's it. They're like, oh, as long as I know I get a I get a percent of the upside, then I might throw in everything away, but I get to work with a really cool team. And that's the thing to think about too, is that, you know, if you want to be an entrepreneur, you know, the two things you need to, you need three, you need three skills. You need to be good at organizing people. You got to be able to get people together in a room to listen to you and to work on a project. You got to be good at sales. You got to go to sales stuff. And then you got to be good at raising money. Those are the three skills. And I think a lot more people are capable of that if they understand those are the skills that are required. You can do those three things. You can do anything. Yeah. The the one thing just to kinda put a cap on this question of OpenAI and Anthropic specifically, if they they are two labs that have really defined this space, they continue to lead and in in various different forms of this kind of maybe even trade the lead back and forth during various moments. The two the things that they have in common are that they have had they have a team that has been put together that has just been humming for a while, and they are kind of uncapped on anything but moving things forward. And that is the one problem that you see at some of these other companies, even ones that have every advantage in the world like Google, is that there's that's a massive structure that is not optimized for let's just push this frontier forward as fast as possible, let alone keep these people together and working on the same page. You know, there's a million different fiefdoms at Google. Meta's trying to establish that culture. XAI is trying to build that culture from nothing. And I think what you're seeing with these the the two lead dogs is that they have their culture. They know what it is. They know who their talent is, and they are plugging in pieces that are in service to that as opposed to trying to set it up or in Google's case, trying to almost cut red tape within their own organization to make this thing go at the kind of speed that is required right now Because Anthropic and OpenAI are moving so fast. Yeah. The the the inside Silicon Valley, the story was the reason that Google was able to move so fast last year is Sergei Buren came in and basically just cut through all the red tape. Yeah. And asked what do you need? What do you need? And was able to get that done. You know, given the current tax climate for billionaires, he's now living in Nevada and still very involved I'm told, but you know, it's that is the effect is to say that California, when you're going to, you know, say that we're going to do a 5% billionaire tax only one time, trust us. Billionaires are like, when you're it's a funny part is, those people it's easy to leave. It's like, ta ta, you know. Well, I mean, and that I will have my that case, it's like in what Sergei and Larry have both said, they both moved out of California preemptively. That tax is not just 5% of their what they have in the bank per se. Yeah. It's it it includes, as I understand it, unrealized gains, which means their ownership in Google. So it is 5% of what their ownership of Google is, which is more than what they have personally. And they would have to liquidate To be clear, like like like, they still, in theory, they still hold 5% of Google or or sorry. They still have all their Google holdings, but they just have to come for with equivalent cash to cover 5% of that value. Yeah. And and they claim that that essentially bankrupts them. It it is, you know, it is essentially just headshotting them as people, And and it imperils the company. It would it would be outside of their fiduciary duty to Google and its shareholders for them to stay in California. They they need to leave for the health of the company. And again, like Yeah. Think think about the cost to move, you know, let's say to Nevada or whatever. And, you know, it's not insubstantial. It's a significant account, but it is as a hedge far far less than than than 5% of the network. There are certain neighborhoods in Miami that are, like, just for the insanely wealthy. Right? And I don't know what rich people like that use to move. It's probably not trucks. It's like hovercraft or helicopters or, you know, the the the big airship from the Avengers. But, like, those are moving en masse right now. There's, like, you know, Zuckerberg moved there. I think Larry Page moved there. There's just, like, this entire influx of people in Miami. The the more people that that make that move, the path becomes worn, and it becomes frictionless for other people to follow suit, and it becomes easier to make that exodus. Yeah. And the crazy thing is, like, we don't even know if that law is gonna be on the on the ballot. You know? It's gotten so much press. That's amazing. Like, that's significant substantial damage done just by the just to speak words out loud of what might happen. That's extraordinary. Well, it's really it's the language on that proposition. The language on that proposition is, like, insane. Like, that is like, do and Yeah. Do you want to chase away? Do you want capital flight that actually damages your your bottom line? Like, if you want to do that, this is what you do. And it seems to, you know, it many many people are taking it very seriously. Some of the richest, you know, most the biggest taxpayers in state California history are taking it extraordinarily seriously. Yeah. You looked at Norway did this, Norway ended up losing like half $1,000,000,000 in tax revenue from it. And, you know, it's it's an advocate make this, you know, the tax things, but it is it's a thing where you have to think We're coming about up on April 15. Consequences. I know. I think that's the thing it's like, because part of you go like, okay, what's the goal? Like, well, we have a state that wants to provide all kinds of services to people and wants to be able to do that. And then you go like, okay, you know, how do you do that? You know, one is, we're now like CBS News just did a whole thing looking at all of the, all the fraud, all this other stuff. And you're just looking at like, the money is here. The money is already getting paid into the system, you know, in my opinion. And to say that we know we need to go do this one time thing against, oh, for all these billionaires. Like, yeah, these billionaires are here as long as it works since. And also like, you know, there are some people that you could say, oh, they didn't earn it. A lot of them, though, like I said, there is a skill set that I have observed that people do not understand that like to be highly well organized, highly able to hire other people, to get other people to work with you and to be able to get people to fund you, whatever. That is a special set of tools that is just not, you just luck into. And when you start scaring those people away and then the downstream effects are, you start to lose, other people use your decamillionaires and other people. And then all of a sudden, the people that are really good at creating capital and jobs and and that's what's another word for a billionaire? Somebody really good at creating capital and jobs leaves, you end up with less capital, less jobs. That's why I'm proposing a negative tax on anybody who is worth whatever I deem, and, you know, we pay them to stay. But I think I think I think it is it is a fascinating to get back to what we were talking about is like, you know, my this is what I have heard is that Google's 2025, which I wanna highlight again, was exceptional. Very good. Was was, like, what more of what a lot of people were expecting from Google three years ago. And and it really looked like they were clicking into form. And the rumor was it was because Sergei looked at his company and said, why are we not better at this? Came in and started doing the one thing that only, like, he, the CEO, and his cofounder Larry can do, which is, I don't care about this. I don't care about this. You're plugging into this. You're plugging into this. We're moving forward. We're getting this done. What's happening tomorrow? Like, let's let's, keep it going. And then all of a sudden, you know, now he's interested in funding efforts to stop this billionaire's tax and he's, you know, not on campus as much because he's he's legally gonna have to live somewhere else. And it's like, okay. Well, now everything reverts to form, and it's like flowers for Aldrinan. Yeah. We'll see. And I think that and and that's the thing too. When we start thinking about AI policy and if you're, you know, when we look at, do we wanna make intelligence cheaper and more widely available? My answer is universally yes. That's all we've all every time in humanity, we've whether it's books or language or whatever, whatever we've had these things, we've just seen absolute incredible benefits to the bottom tier of socioeconomic, have absolutely benefited policy from this. This is a big thing we're seeing in AI. Justin and I did a podcast that came out last week with Nate Gross, who's head of Open Eyes Health Initiative. They talked about what was going on in Ghana and using AI to help doctors there. And they wanted to do a follow-up study and the challenge was, you know, you know, hey, do we not, do we do a control group of some people not having AI and some people having it? And they feel like it's Ethically, just could Yeah. It was it was on they deemed it unethical to do a control group. Yeah. To have physicians Without AI. Working they just saw such a big benefit from AI system today. Crazy. And so that Yeah. And that that's that's kind of the phrase we're getting into. And so we see these benefits accruing to like, you know, I I was at a at lunch today and I saw an older couple over going over looking at some financial data and stuff like this. And I think about how useful now AI is for this stuff. Like, many things like I get contracts and stuff. I just throw it in there and I get responses. And you just see this. I've just seen this benefit. Chatty bitty health. We're seeing stories. We can talk about the dog. Yes. What's the dog? The cancer dog. Oh, the story the Thickens plots. Yeah. No. The there was a an Australian man who had a dog, found out it had cancer. R and R, he said, but he did not let it stop there. No. He took matters into his own hands, got his dog's DNA sequenced, and then used a combination of Chet, GPT, AlphaFold, and Grok to develop, along with a professor, an mRNA vaccine that had identified through all of those steps the the mutations in his dog's DNA that was causing the cancer. And at least as far as we know right now, it seems as if that vaccine has reversed the health of his precious little doggy. What? That's bonkers. Yeah. That's crazy. I put up a tweet and I said, you know, we went from they're just next token predictors to, well, you're not actually using chat GPT to predict this mRNA sequence right Yeah. Very, fast, you know. And then I got the, well, it didn't actually do the mRNA. I'm like, you're the dude. You're other part of the joke. You're the guy. You just thank you for making my joke. You know, like, I don't know. I don't know if it actually, you know, and they and I don't know if the combination of those models, the long term effects of this fact, I don't know. But the fact that we're using this now, I think they'll be better in a year. They'll be better in three months. In three years, you won't be like, well, yeah, you can do that. But you could have you could have had 20 oncologists do this too. If this is real, we have no reason to believe it's not now. Right? And I think we should be prudent and wait for, you know, to see how this thing plays out. But let's say it's real. That's a business. Like, just being a concierge guy that, like, the my my dog gets a cancer diagnosis. Now I just wanna I talked to a guy to to handle all of the the the the stuff here. Like, that's a business. And it it might eventually be something that you can just go to your own assistant on. But today, tomorrow, like, that is something I mean, it's a business probably today or tomorrow if somebody wants to take an experimental step forward. Is help help me out on whether or not I'm stuck on a a pedantic thing. When it says mRNA cancer vaccine, I thought vaccines can't be administered after you have the disease. Is is that a is is that a semantic drift on vaccine or or am I misunderstanding the use of it? I I think, yes. You are you are correct in that it is probably probably would be better used as the term therapeutic. Or cure or something. No. No. No. No. No. I so the difference is that no. You can't you know, a vaccine can be sometimes used post. Like, it depends. Like like, because it prevents the the growth of it. Think of it as it's preventing from spreading to above So the infected part is the tumor. The rest of the body vaccinated against letting the tumor grow in the healthy parts. Yeah. Okay. Alright. That tracks. That's remarkable. And, again, like all caution, appropriate caution on this, but but it it seems like it would make sense. You know, this this is one of those things oh, go ahead. Yeah. I I had somebody a year ago came to me and they wanted to do something kind of an adjacent space. And so I have a group of experts, I wanna be able to go do this. I'm like, you should just learn how to use these AI models with them. Like, you know, like, oh, I think they these people are gonna know more. I'm like, they might. Yeah. But That that is the big secret that I've discovered leaning into just learning new stuff over the last eighteen months is try to try to have child's mind where it's just like, why not try it yourself? Just keep asking, you know, can you help? Can you help? And it is astonishing how much, how many times you encounter, nobody behind the curtain. You know? It's like, it's not complicated. Almost everything when you dip a toe on the other side of it turns out to be less complicated than you figured. As long as you understand all of the blocks that that that get you on the other side of it. Yeah. I I think that you have to just start to, you know, some things are super complex. Some things you can just, you just have to pick out where you break it down and start to grad at it. And sometimes you don't have to understand everything to get it. You know, my, for our any, our two new listeners, I did not become, you know, a coder or programmer till my forties. And I got into OpenAI at 47, basically having been a self taught programmer and became one of the founding members of the applied engineering team applied. And because I just came out with new eyes. Let me, what do I need to know that's important? What do need to do? And then all of a sudden given access to GPT-three, just spent a maniacal amount of time trying to figure out how to use it and what to do with it. And came in from having the benefit of having come from the world of words and language and understanding a bit how the models worked, but not paying, not getting too focused in this stuff. And I would have conversations with people, not there, but elsewhere. Like I remember I had a conversation with somebody, I'm explaining how to do a thing like, well, no, it's not supposed to work like this because of this. I'm like, cool. Yeah. Me show you it work. Let me show you how to do it because the the thing that your explanation there isn't quite accounting for a couple of things to make some assumptions. And like, well, it shouldn't be doing this. I'm like And yet We can argue, you know, but it's doing the thing that you didn't think it could do because you were just sitting there looking at paper and not actually trying it. Is a that is a tough bias to overcome. Like, I I am sympathetic to everybody who is confused and and who is intimidated. That is a that is a core bias to kinda break your own reality on purpose again and again. And again, like, children, they babies don't know better. Like, don't they don't worry that they're saying everything wrong. They just keep saying syllables until they get a reaction. And it's very very hard for adults to engage in that kind of behavior. Well, these but also when I talk to researchers that that's been I just had a thing today. I just kind of not a critique, but there's a paper that came out. Some researchers said, hey, we found this novel technique for getting AI models to do things that are more creative and whatever. And I look at it like, hey, sometimes we phrase in front, a different phrase after. I'm like, this is why I've been doing this for five years, you know. And I'm like, oh, they're not hanging out and exploring and talking to people who are developing stuff. Because like when you're in frontier stuff, you might spend all day trying to do a thing and come up with five things that work okay, but not great and put them away. And then later on somebody goes, this thing is great. You're like, oh, yeah. I know another, I know five other guys that found the same thing, but we didn't think it was worthy to write a paper about because we're so busy exploring. And that's advantage exploring is you just start to build up a ton of experience. Well, and meanwhile, there is a role for translators out there. Like, however far I would submit that most of the people watching and listening to us right now are enough far ahead of the curve that they probably have, you know, dozens or hundreds of people who desperately need to find out the what seems obvious to them things right now. Because there there really is like everybody, there's a strong bias. Everybody who's tinkering around and playing is magnetically pulled to the very fringes of within their domain, whatever they could do with it. And there are very few people who are bothering to lag back and and service translators to the rest of the folks. Yeah. I'm excited by the progress where this is going. I do think that bio safety things like this are things we have to think about. You know, we do have to think about like, you know, I am very much, we just had the president just issued their new AI frameworks for what they wanna do, which is very pro, let's keep accelerating and growing AI for the most part. You know, the biggest fear I think we run into is that, you know, to stop data centers. I think every state has the right to decide what they do there and what they don't. I just ask people to make these decisions based upon real information, not kind of BS. Like, the whole water issue thing is kind of like, you know, if you look at the amount of water used by data centers compared to like having a golf course in your state, you know, you're better off getting rid of the golf course and building a bunch of data centers. But there become these reasons that people put objections to and sometimes they're real objections. Like, hey, people worry. I don't want the cost of my energy going up. That's part of the new initiative is like, this should not raise the cost of energy. And we're seeing examples, by the way, though, of like energy costs are gonna go down in states that have data centers because the data centers are heavily invested in improving the quality of the grid and bringing outside sources and doing like bring your own power. And that's a great opportunity for people to say, hey, well, if you want the data center state, just near the conditions. You know, one, you've got to put x amount of money into proving the grid. You need to be investing in new, you know, investing in new sources of of power so we can lower our cost. We want you to build five technical schools, you know, or whatever so we can train people. Like you can can write your ticket right now and say these are the things we wanna do invest in the future. And I think that you're going to see in a few years, the states that were very, very backwards about data centers are going to start to suffer considerably. They're gonna have higher prices for data energy. They're gonna have, you know, less technical people, whatever. And I think that that's going to be a thing that is, looking back, will be obvious we should have done that. And the people that have against it right now be like, well, you know, we're Local government should be looking at these things with the list of demands. I would look at data centers much in the same way that I looked at, you know, in what was it? Oh, jeez. Probably about ten years ago when Amazon was looking to put another gigantic campus, and it was, you know, between, like, Queens and Virginia. Logistics fulfillment centers. Right? And, there was a big fight of, oh, we we we defeated, Amazon bringing its campus to Queens. Is Queens better now without it? Is Queens better without those jobs? Is Queens better without that headquarters? Is it better without that development? What you should be bringing is a list of demands. Like, there are parts of this country that because of where they sit in various different areas, their connections to various different grids are advantageous. Those local governments should be asking for things that make this happen. They should make it a win for them and their and and their constituents. They should feel lucky that they are living in a place that somebody finds so valuable to put a data center on that land, and they should get something for it. But if you're just spiking it because data center bad, that is the new thing, it's like, oh, boy. I don't know I don't know if that one's gonna age particularly well. Yeah. And the opportunity, you know, to bring a lot of, you know, development whatnot into your location if you choose carefully. So tell me tell me what to get excited about this Cerebrus. I I'm in the Cerebrus Cloud now and I went to the Alright. Go into your go over the upper right menu selector. Uh-huh. Go click on oh, they don't have the OpenAI one in there. Why don't they have that? Probably because I did the free account, maybe? Maybe, but it should be. It's it's the open source model. Let me take a look at mine. You're in Playground, right? Yep. And well, Gwyn three two thirty five, that's a pretty big model. Yeah. Go to the Gwyn, because I wanna show you what a big model. Go to Gwyn three is that which model is that? Gwinn three two three five b instruct. Okay. Go click on that. Okay. Now let's type a prompt like, Gwinn is a Chinese model, which means it's basically a hybrid of all the American models. You just copied all the tokens It's in such a good good results. Go in there and type in like, make a list of 100 great comic book storylines. Okay. That's your system prompt. You actually will put that down below. Got it. Ask anything. There we go. Run. It says, woah. That was fast. Cut. That was very, very fast. Now is this a testament to the efficiency of the model or the back end to compute or what? The back end. They have their own special processing system. Wow. They're building their own chips to go do this. Wow. Wow. Wow. And it's got a oh, that's extraordinary. Yeah. I don't know if you speak to the quality, but you're starting to see a lot of repeats there. But What are the best 100 magic tricks? Cups and balls through Caravanti. How are they done? Okay. How did they do that? Oh, interesting. Oh, it says tokens permitted exceeded. How are they done? It's pretty open open ended. Look at this. It even gives the throat clearing. It says, in the world of magic, revealing secrets is taboo. It's often called breaking the magician's code. But since you're curious, I'll walk a fine line. But since you asked, you're a magician. Then explains misdirection, slight of hand. Oh, that's that's crazy. Oh my god. People will know about the TT. What's a TT? Brian, I got a prompt for you. Okay. Click clear so we can start fresh. Okay. Because I think you keep adding. Go up above upgrade. Got it. Click clear. Yep. Ready? Yeah. Write a script for Brian Brushwood scam school involving soap. Scam school involving soap. Oh my gosh. Intros. Scam school music theme plays. Brian Brushwood stands in front of a bar wearing a Hawaiian shirt and a big smile. Hey, scammers. Brian Brushwood here bringing you another deceptively delicious episode of scam school, where we teach you tricks so good your friends will think you got superpowers or at least a really weird hobby. Today's episode, we're going slippery. That's right. This week's challenge involves something you probably have in your bathroom, your kitchen, or even your own pocket, soap. That's right. Not magic, not hypnosis, just plain old soap. You'll be able to win free drinks on your friends, maybe even win a bet thanks to the slippery science of suds. Alright, Jen and Mike. Welcome to scam school. Here's the challenge. I'm gonna give you a simple bet. If you could do it, you win a drink. I'll check the bet. What's it? Here, watch this. Brian picks up a dry bar of soap and taps it on the bar counter a few times. Now, I'm gonna try to break this bar of soap in half with one hands. Brian holds the soap vertically, claps it between his moms once, and snaps it cleanly in half. Woah. How did you do that? That's the bet. You try, they all fail. Brian looks like drinks are on YouTube, but don't feel bad. Nobody ever gets it on the first try because this isn't about strength. It's about physics. Cut to Brian and Stins of our in front of a whiteboard or animated graphics or whatever. Here's the secret. You can't do it by just hitting your palm unless you do it just right. Well, actually, now it accurately describes kinda how the brick over the head works. That's that's crazy. Hold on. I gotta I'll be right back. I gotta shoot an episode of scam school. Yeah. Right? Yes. And that's the thing. We're getting at the point where if you gave it like a good trick or whatever and said put it into the scam school format, bet you would. And the fact that it took like two seconds to generate. But how could that have been done? That's amazing. Yeah. Made one for me in three seconds. That's astonishing. Yeah. But that's when you talk about you know, imagine that on on a on a really good voice model. Right? And it's like like that's that's actual just talking back and forth speed. Like, that's that's insane. Yeah. Yeah. It's crazy. I can't wait to ask Alan Turing what he thinks about it via the voice model that goes that fast. I yeah. Again, this is my fifth admonishment this episode, like, try these tools, people. Play with them. Give up projects, do stuff, build things, do whatever. I I've seen more and more of my friends You know what? Haven't had the time to learn to code. I'll I'll I'll dangle a little fresh meat too. On the other side of trying all of these, you an astonishing thing is you will sniff out different personalities, and it doesn't take a very big sample size to start to build a a mental map of which LLMs you trust with which kind of tasks. I've got a challenge. 100 prize. What? Okay. Dot dot dot. The premise is you've got to code something using codex or Claude code or a tool, but vibe code something. Your premise is weird things. Oh. Oh. Okay. Alright. Next week. Next week. So you can you can email just email me, brian@schwood.com. Brian@shwood.com. Look up your Askchat GPT for all rights concerned and whatever blah blah blah. But legalese insert here. But yeah, let's do that. Yeah. See if you can some legalese real quick. Did did I tell you I finally vibe coded my first useful thing that I use every day? What's that? No. It's a so on the office hours, I like to I encourage it's been fun to take my own advice, and I tell other people who are working on things to make sure that there's a a gift awaiting anybody who puts effort in. You know, JJ Abrams was so good about this. Whether it's, you know, fringe, there would always be a hand with some number of digits shown or whatever that would be part of a coded thing that would have, you know, don't forget to drink your oval tea, you know, whatever. But as having those those little things that reward deeper attention are fantastic. So each day as I went live on office hours, I would use this little dad joke calendar that my kids gave me. Right? And then, for example, this one says, my daughter says thinks I don't give her enough privacy. At least that's what it said in her diary. And so I would put the setup in in in plain text, and then I would just encode with a simple ROT 13, the thing. And so live on the air. So obviously you're spinning a lot of plates when you're on the air. Your mouth is running. You're also live switching and all that stuff, talking to the chat. But I was able to vibe code with that, this dad joke terminal, in which case you just click load another joke. There we go. Why did the stadium get hot after the game? And it says, o r p n h f r n y y g u r blah blah blah blah blah. Or you could just click to reveal the punchline. Do you guys have a guess? No. Something about too many fans, I'll bet. Because all the fans left. Mhmm. There we go. But but that ended up being my daily driver. It was one of those things that took, I don't know, maybe a 150% of the time it took to do it the first time to that again, you know, with one and a half time to set up the system where now it takes much less time. That's amazing. Yeah. I I was thinking about this the other day, like how quickly you can build things. Month over month, they get faster and better and more robust. And I've started doing some super challenging things. And I look over at my three d printer. Sad. My VR and Well, all I look No. I look at it like, what happens when those worlds collide? Yeah. Right? Oh, that's coming up. I also three d printed my first useful thing ever. So, so here's the thing is back in the day when I was touring, I would have these little remotes attached to the back of my shoes so I could run my own sound. The back of my left heel was play pause. The other was fade in advance. Turns out you got these flick buttons that can act as Bluetooth, keyboard keys. So I set it up. So there's three of them. And so now I'm building out these mounts to go on the back of my shoe, but the problem is they're recessed buttons. And so somebody in the chat found these little three d printable buttons that go right on the front. So now it's more like an arcade press button. And then I've got, I've got flick clips coming. So soon I'll have my my shoe control system back so my presentations could be more dynamic. Amazing. Crazy. Oh, have you seen have you seen, have you seen the end of of believable video? We're we're cooked. Hold on. Let me let me see if I can find this. Oh, are we yeah. Let me see. The Lulu Stream is what it's called. And then Here we go. Check check. Here. I'm gonna add an input right here. Here we go. Yeah. Look at me. It's it's your old friend Brian Brushwood. Now when people say bring back the spikes Uh-oh. There we go. I could say, you got it, boss. Here they are. The spikes are back. I like it. It it brought your hair back twenty years, but aged you 20. Yeah. Yeah. Go to the influencer. Yeah. Oh, the inappropriate request influencer. Appropriate request. There we go. What's up, Andrew Vane? How are you guys? Look at this. You ever see this from a deepfake? You could actually oh, it's got it's got physics. There we go. You like that? And then it'll it'll continue to hallucinate as I by the way, we got demonetized on that episode of great night. Oh, really? Because of this model, I believe. Not advertiser friendly, they they said. Here. Hold on. Gotta go to OnlyFans. So many hands. Ain't that wild? Wait. What do you think got us popped on great night? Oh, I don't know. Anything from Kissinger without a shirt, male nipples to I mean, we it was not the most advertiser friendly show we've ever done, but it was a lot of fun. It was great. Kissinger doing magic tricks. Oh, that's right. He did. We we'd interviewed captain Morgan. Yeah. Finally, history brought to life. But watch this. So the photo doesn't have a bunch of this, but it's gonna do its best to make up what the rest of the body looks like. Hell, yeah. And then oh, hold on. Let me There we go. Damn. Woah. Look at that arm. Oh. Oh. Oh my god. Oh my god. Good lord. I love it. This is amazing. It is weird things. Look, we told you. Oh, RIP monetization. I think they're gonna get more money for this. This is incredible. Apologies to anyone who watches high. I just want to apologize. Oh my god. This is I love you. This is Terry. Oh my god. Why do they keep growing? This is the best. Oh. Yeah. For our audio listeners, I give up. Hands are just exploding exploding outside of hands. Oh, I ran out of tokens. Oh, no. Oh, that's amazing. That's one way to put it. So I gotta pick. I was gonna say, got any picks? I don't think we're topping the hand the endlessly multiplying hand chain. I gotta pick. Uh-huh. Everybody needs to go see project hail Mary. It's great. Go on. I went and saw it in IMAX. I saw the what do you call it? The Amazon Prime early preview earlier this week. Mhmm. Front row center. I'm so glad I was that close to the screen. It was it made everything feel just as big and as inter inter interstellar as it is. It, it does right by the book, but more importantly, you know, Justin and I have talked before how we are of the opinion that you should watch the movie first because guess what? We all know the book's gonna be better. And that's still the case, but project hail Mary is a fine entry to the story. Highly recommended. Gosling, good. Yep. He's he's great. There's there's execution on ideas that could have happened poorly, but instead happened very, very well. Cool. I'm looking forward to it. I I I love good sci fi. And I would say that Andy Weir gave us both The Martian, which I think love I the movie more than the book, I'll be honest with you. But, and Hail Mary, and skip over Artemis. That that we've got, that that one guy, still a relatively young guy, has created books that became two great science fiction movies is awesome. We're talking about this on great night, Maine. Authors who have had two good movies made from their books. It's probably a smaller table than you would think. Like, to to get two good movies is pretty remarkable. I don't know. I can name a few, but yeah. Yeah. Yeah. But they're all, like, names you'd know. Like, they're all, like, names that you'd wanna be on a list of. Yeah. Yeah. Yeah. Man, someone should have I'd be crazy because, like, know, Lord Miller, you know, doing this sci fi thing, it could be great if they just did it with like, you know, recognizable intellectual property that people are Yeah. Excited about. It'd be really cool. You know? Yeah. Wars in the Stars. Yeah. Think I got Han Solo movie or something. That would have been neat. Yeah. Interesting. Okay. Andrew, your pick. I will just double down on Brian's pick. I haven't seen Hail Mary, but I'm excited that, you know, that's out there. I bought a screenplay by Drew Goddard. Yeah. Who did the screenplay for A Martian. I'll pick Sentimental Value, which I watched on the flight back from San Francisco last night. Stellen Skarsgard, he's not just in Andor, he also has a complicated relationship with his two daughters in Norway. It's a it's a good movie. I liked it. I enjoyed it more than I thought I was going to. I thought I was going to just kind of suffer through it, but it was it was well done. Yeah. Awesome. It's been weird.