Astra, Foldables, And The Real AI Risk Debate

Andrew Mayne: Hello and welcome to the Weird Things Podcast. I'm Andrew Mayne, joined by Brian Brushwood. Hello, hello, hello, hello. And Mr. Justin Robert Young. Justin Robert Young: Hello, friends. Good to be with you yet again. Andrew Mayne: Yeah, so, um, you're gonna get an AI episode because it's been pretty topical this week. Justin Robert Young: You know what? We could get the Apple fanboying out of the way, right up top. Sure, sure. What did you guys think of the Duo announcement? Um, I'm ordering it. Yeah, I'm getting it. I find myself weirdly conflicted because I need a refresh. Good! More for us. But... Stay away. It's too expensive. Probably gonna be bad. A couple of the all-timer YouTube videos that I shot were shot on the really good cameras on the Max. And the idea of going to a lesser camera... Who am I kidding? I'm gonna get the Duo. But, but, but, but, but it gave me a complicated feeling for a second. Andrew Mayne: No, I, that's, that's the trade-off. And I was talking to my wife about this because, you know, she's likes the Duo, but she's, you know, she's making more video content. And, and I said, here's the thing. If you like the form factor, get it. Because the best camera will never be on the Duo. It's always gonna be on the Pro. And that's going to be the way. Like, if you're like, oh, next year it'll have a better camera. Like, yeah, next year it'll have the better camera. And guess what? The camera on the Pro is gonna be even better. And so you just have to come up with, if you like, I, I love the form factor. I, I look at the way they did it. I've looked at the Samsung stuff. I've held, held off. I've, you know, I have other, I'll buy Android devices and stuff to have backups. I've held off for a long time because the first, I mean, they've gotten way better. The first time I looked at one, somebody showed me the crease was terrible. It was just terribly done. They've come a long ways. I think Samsung's come a long ways what they've done. And Apple's got its solution. We'll see how durable it is. I like the look of the iPhone one better. It's in the iPhone ecosystem, which is, you know, where I am right now. They did a lot of really smart things. Um, you know, people like the cost, the cost. I agree, but it is a device that I use 90% of my waking hours. Justin Robert Young: So that's the thing, man. It's the internet that you not only look at the most, you literally touch it. And that matters like that form factor. And to be honest, the aesthetics kind of matter in terms of you. I mean, you're going to spend time cozied up with it. You probably want it to look and feel really good. You want good design. So by way of being friends with, uh, Darren Kitchen, who is a fold aficionado and has bought every version of the Samsung fold lines, including the trifold that I think they sold like a thousand of and then stopped because it was going to be more profitable for set to sell those screens to Apple, uh, than, than to do something that had a fairly limited market value for them. I think that there are a lot of trade-offs to fold phones. The crease is an issue if you have an eye for it. Um, so I was encouraged by what the duo's crease looked like. It took me a little aback that they went with the passport design, uh, a little stubbier, but I, I was swayed by their reasoning of like, Hey, we actually, what you want is the right screen, not the largest screen. Because of the way that formatting works, you're just going to wind up getting black bars on things anyway. And, uh, Apple has the ability to force a design language change for apps and websites, which I think will, uh, be very interesting when it comes to, to those devices. I'm going to get it because for me, we're in a generational sea change for how we think about our devices. And I am looking to a world where I might never bring my laptop places again. Andrew Mayne: My laptop might be a machine. Justin Robert Young: That seems so crazy because you do real production work. Are you picturing a world where, where you work on world's greatest con and do that stuff on a, on an iPhone? I mean, and we can blend into our other big stuff, but, uh, I have seen the light of me possibly never touching the input controls on editing systems again. Uh, and, and what, what we're talking about, and this will probably seamlessly take us into the land of AI, but, but we're, yeah, let me, yeah, let me, let me, let me just, we'll just leave that there. We'll get into what, what, what I mean in a second and why I believe it. But I think that we're in a place now where you are going to do more and more and more via a device that goes to a server and visualizes back to you what the result of that process was. That doesn't need to be a laptop, uh, that doesn't really even need to have a keyboard for most stuff. But if you have something like the duo that can stand up, uh, or, or lean up against something and be a fairly credible, uh, monitor with a keyboard, you know, that plus a, a, a little track pad. And, and yeah, I could see myself using audition on that. Uh, and that, that would be even the fine tuning stuff that I think we're gonna rely on less than we have. I don't want to completely sell the world that everything is going to be voice dictated, but like, I think more of it is. And in that case, then like, yeah, if I'm not in French time and we have to get a world's greatest con episode out, for example, while I am on the road or I'm on an airplane or something like that, where I'm not going to have the same kind of, uh, uh, connectivity. Then yeah, I think I can move the ball forward pretty credibly with the tech we have now, let alone what we're going to have in a year. Cause if I look at what we had a year ago, it looks like the Flintstones, baby. Yeah. The, that's an interesting, a Bluetooth keyboard and mouse, and even a step farther. I don't think there exists an iOS app for Parsec that, that, you know, allows you to remote desktop control. Uh, but if they were to bring that into like, you could just clip open the duo and you're actually running a desktop, it's just remotely. I could see that that does, that does track. I mean, I think there's got, there's always going to be room for you to have a home server. I think the idea that the, uh, the concept of home servers, I think are going to become more popular, uh, because there, there is stuff that you're going to want to run locally where we visualize that. I think. We might think of desktops very differently in the next five years. We might think of laptops very differently in the next five years. And the biggest thing that I think might go away is the thing that changed the world, the, the form factor of the phone. I think that, you know, we, we do less and less of what it was designed for to actually take and make phone calls. Those are all things that I think can be done differently. And in an increasingly agentic world where a lot of business is kind of handled agent to agent, then I think that's even going to be less and less. So the question is, what do we want out of the computer in our pocket? And I think the, the duo brings us a little closer to that. If they sold the duo without a phone with no cell plan, except for data, like I would be fine with it. I think that I could, I could make that transition immediately. Andrew Mayne: I, yeah, it's so interesting because I, somebody had a, somebody tweeted out like, uh, oh, they're making the same mistake as the Apple vision pro. There's no use case for this. I'm like, it's a phone. Like that's the use case. It's like, it's a pretty simple use case and, and they sell like crazy foldable. So like crazy in China. And, and I'm going to, I'm going to give you my stupid reason for doing this. Like I don't actually use my phone very much. I'll use it in the bathroom. Sorry, too much personal information, everybody. Uh, you know, I'm the one nobody else does. Uh, and then I go to Taco Bell, not connected. And I set it on the table. Usually set my phone against my hat so I can watch a video or something while I eat. And I'm looking at this going 90% of my use case is literally going to be folding this thing and setting it on the table and talk about how to look at it. I understand me. I know me. I'm okay with me. All right. Um, I, you know, I bought the air cause I wanted a lighter phone. You know, I was, I was into the idea of foldable, but I got the air cause I was just kind of sick of a phone. I just like the cameras. I don't care. I don't make that many photos. I really don't care. I got a photo. I want to make it better. Oh, chat should be to make it better. I don't care about cameras, video. I don't make a video. I don't, I don't like I made a video. You know, so like, I'm like, I want, you know, I like the idea of, you know, like I am very, I can find some very laughable predictions by me about like how I'm going to use stuff like the Apple watch, my excitement about using the Apple watch for dictation and stuff like that. It's finally usable for that. Um, and so I think I'm often very, very wrong about how I'm going to use it, but I don't know. I, I loved my iPad mini. Then phones got bigger and I never, I didn't touch my iPad mini until I touched my Apple pencil. And I look at this like, Oh, I can use my Apple pencil on this. I can make notes on it. Justin Robert Young: I don't know. It looks really cool. I'm excited. Speaking of the voice dictation on the, the watch, what did you guys think of the always on, always listening mode for that? Andrew Mayne: I mean, at least they're admitting it. Uh, yeah, there, there, we talked about this, I think here a couple of weeks ago about how Apple will figure out the most Apple way to do these kinds of things. Cause we do think that always on stuff is going to be, and everybody else did it weird. And you, and I thought that it was a very good pattern. Like, Hey, uh, we're not recording voices. We're doing real-time transcription. You know, we're transcribing it. We're not saying who said what we're not doing speaker diarization. So that gets rid of some of the concern. If I'm talking to Justin and Justin's like, Oh, well, tell me your, tell me your opinions about Babylon, you know, five again, Andrew, and who watches it. And then, you know, pulls it out the court transcript. Like that is literally what they didn't say was a legal thing too, is that, well, I was talking to Brian. I had my Apple watch on there. And it's like, well, we don't know who said what. And I think that was a very smart thing. They didn't want to get into the legal stuff, but you don't want to start having your phone, your watch subpoenaed. Like you still could, but then it's like, well, I don't know who said, I know this was said. So a lot of it's trying to figure out how to navigate that. Justin Robert Young: Yeah. Uh, that'll be interesting. I think that the, there have, since recording things became, you know, useful, there've been a number of conversations that I was like, dang it, always record, always record. Where was that? And then I'll probably be real thankful for those opportunities. Yeah. I, um, to me, the question isn't the technology. The question is society. And, you know, we're already in a world where everybody walks into houses and they're okay with devices that are constantly recording you. And listening, listening, listening for wake words, everybody's probably got at minimum one device that is currently doing that on some level, um, where it goes, how long it's saved, diarization. And, like, that is a step further. It's not the craziest step further. It's not like we haven't had the ability to surreptitiously record people to begin with. Uh, might I remind you that Nixon went down for recording things in the White House. That was a system that was put in there by Lyndon Baines Johnson, his predecessor, which he started doing in the Senate. Literally, Brian, for the reasons that you want this tech, because he wanted to record every conversation so he knew if he made a deal with somebody in the room, that he knew if they were trying to BS him or lie to him when they didn't deliver on what they said. If they came back three weeks later and they were like, I never promised you that, he can say, absolutely you did. And he would do it with conviction because his secretary would transcribe all of the recordings that he had, both on the phone and in his office. So it's like, we've had this tech since the early 60s. Uh, uh, people have recorded, I'm sure people do that today, uh, doing it. Now, society making a decision, obviously we're going through a bit of a moment where we're wrestling with how much of this is a reality. Uh, but I don't know. I mean, I don't know how much use I would have out of it, but I, I can understand it. Yeah, I would think, uh, the bulk of the use would be contextualization. Like, uh, when, when you're working on, especially, you know, I'm working on this wolf masking project and, and, uh, when I'm in a thread, I have to ask, like, are you the version that knows what I'm talking about with these made up words? Um, and, you know, having more fidelity to that is probably pretty good. And especially, uh, you know, as an extrovert, there are a lot of times my mouth answers. And then after the fact, I'm like, that was pretty good. That was dumb. Uh, the ability to have that record and be able to say, you listen to me, give this pitch a hundred times. Can you write the best version of it? And that'll be the one that I memorize. Andrew Mayne: Yeah. I, I think we're, I think that once we feel comfortable about the way this stuff is used, like putting, it's interesting. Like we go into stores now, there's security cameras everywhere. We just get it. The trade-off is in theory, eliminate stuff. Uh, I have, you know, I have, I am very, very pro freedom, very libertarian minded, but I have no problem with flock cameras in public because I don't think I have an expectation of private in public. I think if a city wants to put cameras and stuff up there, like I'm, I'm fine with that, how that data is used, whatever, something else. And I think that we're going to get to a point where the problem with a lot of these devices happened when people found out like, oh, the government was able to subpoena your Amazon, you know, Echo device. Cause it turned out that it was listening. It's like, wait, what? You know, um, LG TV is listening without you knowing, like, that's an issue. I think that these things we have to sort of really address and understand, do we trust who's doing it? What's happening to it? But, you know, you go into a shopping store and you're under surveillance. Do you know if that data is being sold to some company that's tracking buyer habits, things like this? I don't know. I don't know if I care anymore. Justin Robert Young: That, that report, um, it really is some LG TV QIA questionable internet activity. Speaker 02: Yeah. Justin Robert Young: Just, he's so smug. If you're just listening to this, the smile on Brian's face, uh, uh, just, just, oh God, send him to the nationals for dad jokes. That's just amazing. That's just amazing. In fact, it's part of the reason why I've been frustrated with their lack of movement and AI, because I want to trust them with more than, than they are asking for because they built it. Uh, on the other side of this, you, by all accounts had a pretty good system being built by Meta, by Facebook with Muse. And they're building all these bots that, you know, if you just hook up all your data to them, that they can do awesome things and they can do it very fast and everything. And I will sooner post whole on Twitter than trust, uh, Meta. Like, I just won't because they burnt me so many times. I have no belief in them. Uh, same with LG. Do I think that anything that LG does just won't be resold to the highest, you know, commodity bidder for data? Like, absolutely it will, because that's the kind of company they are. That's how they make money. Uh, I trust certain companies. I don't trust others. And they have to demonstrate why in their actions. And, and I think that's, that's where we're going with privacy in these products. Yet the, uh, the moment I heard about the always on, I thought, you know, only Apple, only Apple has the credible reputation with enough actions to back it up that this could even be approached. Yeah. Honestly, I, this is probably the time where I would buy an Apple TV. Like a physical monitor TV, uh, because always on, like, you know, having the right FaceTime settings so I could, you know, uh, uh, Bella could talk to her, her Nona, you know, in, in a, in a flash, it would have all that. It would be beholden to me and not to a bunch of advertisements that they're constantly trying to jam in front of me, which is why I have to go through a stupid screen before I can get to my television every single time I turn it on. I'd pay a premium for that. For, for a big, beautiful, uh, screen. And that's something that I know that Andrew and I have had this conversation for a decade and a half about why it didn't make sense because televisions were more of a commodity market. But now, I don't know. I, I might be willing to, to, to peel off a few more if it's the right product. Andrew Mayne: Yeah. I don't, they're not, I don't see them doing it though. It's just not, not the, the margins are still, you would have to charge so much more for them to justify. Justin Robert Young: Well, and especially like this Samsung is the size of a wall and it was $599. It's unreal. But what's, what is the, the barrel of like, all right, yes, you can get the $250 90 inch television and that will do you. And this is not reality. We're just going to push it forward in a hypothetical. Now it is like, absolutely. It's an always on listener. And we are, uh, uh, you have to go through two menus before you can get to what you want to watch. And there's no way to turn any of that off because they do have to serve. Is it a 30 second ad before you're able to watch whatever you want to watch? Yeah. Uh, I, I could picture Apple charging a premium for the peace of mind to cut through that crap and to know that you're never. That's my question is how bad would it have to get on the low end for you to be like, nope, I want to spend more. Andrew Mayne: But I just don't know, I don't think the point for Apple specifically to do it, like just the, where they're shipping their margins, everything else, the engineering, getting, they looked into it. And it's just, there's never, there's no point at which for them that they make, I mean, look at how much they charge for a monitor, you know, for Apple to charge for TV. Great. It's going to be a $20,000 TV. I would say though, there, it is an opportunity for another trusted brand like Costco or somebody to come in and say, I'd say Amazon, but Amazon's going to shove ads all over it, you know, cause they love ads. But like we trust Amazon to an extent, but once I have an Amazon device, I don't, um, you know, I trust them to get the stuff to my house. What's in my house? I don't know, but I could see somebody in, cause also there's, there's a market to somebody to go to LG and say, okay, we will buy your TVs from you, but we're going to pay a $50, $100 premium in charge. People to just have it be a F and TV, you know, cause I think that's, that's part of the way. I think that's one way to do it. It's the hard part is they, once they see that, you know, that opportunity to sell you something through there, it's, it's, you know, look at, look at how hard it is to even buy a streaming service without commercials. Yeah. Because the revenue from that is just so big, but I think there's a market there. UNKNOWN: Yeah. Andrew Mayne: Yeah. Um, so yeah, I like the duo. Uh, we'll see how it actually is in practice, but I'm excited to play with it. I'm curious to see, like, you know, I love the little things people did, like very curious to see what it's like to type on a bigger screen like that too. Um, okay. Let's talk Astra. Justin Robert Young: Oh, that's right. That happened since the last time we talked. Uh, guys, I think Astra is a pretty big deal. I'll say it. Ryan going out on the limb here. You're saying it's that you love big Astra. Ha! And I cannot lie. Uh, I'll tell you what. So right now we're in that delightful phase. If you're old like us, you probably remember various waves on the internet where new tools were available. And for a split second, not everybody knows what to do with them. Um, and as a result, wild experiments abound. And we're in the middle of getting to see that right now with Astra and to live with some of the tinkering. Uh, let me, let me show you guys one thing I'm extremely proud of. Uh, I was able to redo the, uh, the UI for cord killers program. So I've always wanted to do something that read as kind of sports radio, pardon the interruption style. And so I opened up Astra and I said, Hey, let me give you these Vmix files. Let me give you the Rodecaster Pro 2 files and so on from these. Can you piece together how my studio works? And it, it said, well, the most I would need to know blank, blank. And it gave me a whole bunch of homework. And just as I was about to start typing explanations and answers, instead, I just pressed record and talked as though I had an intern on the other side of the planet. And I said, Hey, this is the thing. You see this button? I held up the, um, the stream deck with the physical buttons that I just dropped and broke. Uh, and then the, uh, Oh yeah, I really did miss that. Uh, and, uh, describe what I wanted it to do. What, uh, what it currently did and what its utility was. I pointed the camera over at the other side of the studio and said, over there is this, that, and the other. And then it said, I got it. And then I said, Hey, let's do, uh, we already have this living show road rundown doc. Can you put it on screen? And at first, as it does, it did a bad job. And then, you know, uh, over a few iterations, it got quite good. And now I, I, it's, it's, it was really astonishing. It was really astonishing. Andrew Mayne: I, you know, we had been hearing rumblings, you know, for those of you new listeners here, uh, I worked at OpenAI. Justin and I produce and I host the OpenAI podcast. So we're in the offices, OpenAI quite a lot and, and talk to interact with, you know, my former colleagues. Um, and we had seen since I'd say early parts of this year that there was this optimism, this, this, this, this, this very much enthusiasm, even when they're like, ah, OpenAI is behind. It's kind of like a look on their face, like, well, wait and see. And, um, we saw model improvements, you know, 5.5 was really the first time I think that in a while they had a coding model that was, you know, you know, I think comparable to what Anthropic was offering. And then 5.6 Sol, which I think is a fantastic model, you know, just efficiency, whatever. And that was great. Like 5.6 Sol just started making huge, huge inroads. And then they dropped Astra. And, uh, Astra is GPT-6. And I mean, it's crazy. I did the launch of GPT-4. You know, I had the launch of GPT-4, came up with all the examples of it. It is insane to go back and look at when that came out. I made the videos. I made the launch page. I did all this stuff for this. And now there's hundreds of people, much more competently too. Let me, let me make that clear too. It's amazing to watch two iterations later what a launch is like. And Astra is, uh, and I, you know, I come at it, I kind of forget that I was there. And now I'm like, just as a consumer of this, um, it's phenomenal. And what they did is it's a new paradigm for how they train the model. The capabilities were great. When they started showing just on the, on the frontier of capabilities, very, very highly capable. I've thrown it incredible tasks. It's done amazing jobs. We've talked about this, like video editing, stuff like this, things that people were like, let me just throw out this, see if it can edit. Does it really well? Video games and design and stuff. Really, really cool. I've got mixed results on some stuff, but also I've been sort of cautious in how we've done stuff. Instructions really matter. Um, the kind of the, the, the crazy part was, is earlier this week, there was a whole thing where opening, I said, they used an internal model to solve what's called a millennium prize, which is these category medical problems that are, some may be unsolvable, but considered like really interesting problems to solve. And they showed a proof for one called the Neville Stokes conjecture, Neville Stokes. And, and they released this graph said, yeah, we used a model that's better than Astra. And they showed this model with its performance capabilities of math. And it was like significantly better than that. And like, yes, they're sitting on this, we're in this incredible capabilities curve and to its credit, Anthropic, you know, Fable's been, people love Fable. They did Fable 5.1. They'll probably come out with other ones. Uh, we'll, we'll see what's happening. We're watching model development across the board, the Chinese models, which are, you know, some clever Chinese innovations and a lot of distillization of American models have become very, very confident. Justin Robert Young: I mean, beyond, beyond, beyond that, according to Anthropics risk assessment. Uh, uh, maybe even just a false front that was routing stuff to American models. Andrew Mayne: Yeah. That was a thing that I remember that came up at the time and it turns out like they, they, they claim they have the receipts for it was like, I think Kimmy, like Moonshot, which people love that Kimmy model, which the whole story, that's kind of crazy was Moonshot model comes out and Anthropics. Like, yeah, they were actually routing people hitting, trying to use the Moonshot model to anthropic accounts that they, that, that, that Moonshot controlled to basically look at all, get all that data of what people are asking for. And I got news for you guys. Like, um, you, you worry about American companies and data that Chinese companies do not care. And like the amount of stuff that's going through, there's insane. But anyhow, literally they're selling people like, oh yeah, this is the Kimmy model. And it was actually Claude. And they were using that data to train the next generation Kimmy model, uh, which by the way, it was what Cursor used. Justin Robert Young: So am I understanding correctly that essentially they just set themselves up as a, a, uh, LLM in the middle attack where it's like they, they, they were just a man in the middle passing everything around and then, and learning from all the data. Andrew Mayne: Well, they would say, Hey, we're offering our model at super cheap tokens. And so you would go in and go use the model. What they were actually doing is they had, you know, thousands of fake accounts that were actually routing it through there. Cause they wanted to get billions of training tokens to be able to train and see how the model had it. And I've had this argument for a long time, like, well, you can't copy your reasoning traces. Like I can tell a lot from tokens. I can tell a lot from other things. And they're actually tricks to get the reasoning token, the reasoning tokens out. And so some of these people like, it's a weird thing. And I want to explain like, like, like, here's the situation. Is it, um, we live in a world where if you're a developer, you don't want to be dependent entirely on open air and tropic for models. You don't want to have a couple providers to say this. You want opportunities. You want that to be available to you. And I think that is scary if you're a developer, the idea that somebody might try to shut down open source or whatever, because it is cost effective in many ways. It allows you to do a lot more. And if anthropic and opening, I say, you can't fine tune, you can't do this. You're really in a bad situation. So open source drives so much of the AI community. On the other hand, so much of these modern open source models, the exception of the stuff that's maybe coming out of like NVIDIA and a couple other, like, you know, for American labs, like maybe Meta, like initially they did some distillation, apparently. Um, all the Chinese models for the most part are literally taking billions of tokens from American models and copying the way that they perform, which is, you know, you know, you could, you, people are like, ah, it's the same as copying in our text. Like, well, and there's a, there's a difference where if I'm like, you know, stuff that's publicly available on the internet versus creating literally fake accounts and signing contractual agreements that I'm then going to break and do that. Um, so it makes it, you know, challenging, but also it means like, wait, you know, what is, where is the Chinese frontier without American labs, you know, but also is it fair? And that was the best frustrating thing to like Anthropoc is like, Hey, listen, we're spending billions of dollars to train a model only to have the Chinese come out and copy the model and not sell an unrelated service, but sell exactly the same service. It's one thing if I train on top of a stuff like text or data, and now I'm telling you, here's an AI you can talk to versus no, I'm selling you the same product. I just copied their product to do it. So it creates a very tricky role, you know? And, and I think that some people kind of like, you know, like, ah, fair is fair. It's like, eh, fine, you know, or, you know, it's a problem. So, yeah. Justin Robert Young: So what's been the biggest surprise that you guys have encountered since getting your hands on Astra? The, uh, token bonfire, probably. Speaker 02: Holy moly, I've gone through two resets. Justin Robert Young: Yeah. I mean, it's, I'll put it this way. I use it, I use all these models primarily for creative work that does not involve dealing with code bases. Um, if you are doing that, this was the first time that just doing stuff with like computer use and, uh, things like that, that I saw really meaningful erosion on my limits. And then I had it take a full look at my code base for the vibe coded fitness app that I have. And it just decimated what I had, what I had left. Um, it is incredibly talented and it is incredibly token hungry. And if you are using it for development, like, I think the idea now is have it make the plan and, and then have a really efficient model, maybe execute on it. But, uh, it is, boy, it is a, it is a ripper. It is, it is a fatty, fatty little model. Greedy. Andrew Mayne: Yeah, I concur. Um, I, I have, I've gone through multiple resets, uh, considering, uh, asking for my old job back so I can get unlimited. Um, uh, you know, um, I would say that, yeah, and that's, it's a bigger model. And that's the thing that Anthropic went through with Fable where they went up, the, the model size was a bigger model. You know, they, the upsized, you know, the, you know, you, you go from, you know, 5 trillion to 10 trillion, however many parameters it is, you're running a bigger model. I think we're going to see some efficiencies coming through. I think that there'll be coming out. Like I've been going actually using GPT 5.6 Luna a lot because that model is actually tiny little model, but it's extremely capable. I think we're going to see a lot more, some efficiencies there. I think that, I think I'm not, I'm, I'm frustrated. Like I'm looking at like, eh, I got seven days left and I'm at, you know, 60%, you know, I'm like, oh, guys. Uh, but, uh, I think the efficiency is going to come, but it's frustrating. I think, yeah, but it's amazing. Yeah. I'm also way more ambitious on the stuff right now. Like I'm like literally just trying stuff that, you know, you would tell me a couple of years ago, yo, you're going to do this with a model. I'm like, cool. On the moon? Yeah. Justin Robert Young: I will, it's computer usage and understanding for creative tasks on like CapCut is nuts. It is just bonkers. You know, I, it, it, it is unlocked elements of things that I've kind of wanted to do, but couldn't justify. Like now this is, this is the paradigm. And, you know, we've certainly been a lot of conversation over the last three years about what AI is going to do for work and what is AI going to do for jobs and what is AI going to do for X, Y, Z. Here's the results. The science is in. You do more. Everybody is working more. Everybody is doing things that they wouldn't have otherwise done because what we couldn't visualize before we had the tech was how much of our, what we would love to do in a dream scenario was just grayed out because time spent, effort spent, money spent. What if now all of a sudden those weren't grayed out, like a video game with skills. Now you could just do those things. It's like, it's, it's insane. It is, it is remarkable. And it's the more I do, the more I want to do. You get addicted to running fast. It's, it's remarkable. It is, it is insane. And I don't even think that I've really scratched the surface. That's the other exciting part about it. To be honest, like I'm a little bit worried to scratch the surface because I'm afraid that the moment I experienced the comfort of blank looking so much better and taking less work that I won't be able to go back. I mean, but like we're, you know, when you first had a keyboard, it's like, oh, will I ever be able to go back to writing on hand? Like, you know, it's like, no, now you just use a keyboard for everything. It's like, that's, this is just the new way of the world. Andrew Mayne: Yeah. I, I, you know, my biggest thing is just, you know, running out of tokens. You know, I, I, you know, we'll talk about the doomsday stuff. I'm like, I'm more worried about running out of tokens than else just because it makes me want to try more things. Justin Robert Young: How do you decide? Because I have noticed that in the, in the usage, you know, there's a bunch of these little hundred percent things and then for the smaller models. And I'm like, hey, I don't know what tasks to delegate to that level. Andrew Mayne: So nobody knows. Nobody knows. I, I, I had a, I had a talk with a research the other day about a big problem they solved and like, you know, they use thousands of agents to go do this thing. And, and the question came up like, uh, is that the right amount? Like, we don't know. Don't know. Maybe we could have done it with 10 and then taking it a little longer. Like that's, that's a mystery. The hard part is, is these models are so new until you put it in front of millions of people and understand how they're using it. It's difficult. I tend to, the pattern I've been doing right now, do not take this as a hard rule because I'm still figuring this out, is I'll go into projects where I have more routine stuff. I'm like, hey, I just need you to check this thing is working. You know, I need to check for something here. I had a project I'm working on with Justin that's running a very, very big bill every month and I needed to see if there was any efficiencies there. And I'm like, oh, I just need 5.6 Sol's fine. It can go in there and check, you know, the costs on stuff. And I'd say things that models were good at before, just keep using the older models for. Things that are harder for it to do, go to Astra. Justin Robert Young: As a matter of fact, just today I posted the bonus episode for Great Night and I asked, I was like, should I upgrade this thread to Astra 6? And it says, no, it's fine. Everything works. Look at the last seven posts. Don't change nothing. Andrew Mayne: Yeah. You know, so I tend to, I've been doing a lot of smaller stuff too, like using Luna, like in ChatGPT, like I had, I have to go to Ecuador next week to do a presentation. I need to go over my slides. I'm like, I'll have Luna go look at it. And Luna is perfectly good at stuff. You know, my instinct is always, I'm going to use the best thing for everything, but I've had some stuff where I did it and I'm like, ah, I didn't need to use the super big powerful model on the Ultra to do this. That was dumb. Yeah. But it's, it's, it's eventually the models will get better at knowing and they actually, Astra is interesting because in theory it can decide how much reasoning it to use. Justin Robert Young: Yeah. The, uh, I have noticed, this is the only thing close to a complaint is there'll be like 10% of my compute will go. And it's like, so what did you accomplish? And it's like, I rehearsed adding four records, you know, it's just like, all right, well, I'm not entirely sure, uh, how, you know, whether or not the project is just too, uh, you know, exponentially split or what on that. I will say this is the first model where I have noticed a personality difference. Its personality is naturally flat comparatively to soul. And I only really noticed it in, uh, my food tracking because like whenever I would food track with soul, it would be very kind of like, okay, here's where you are and here's what you have. And here's where, you know, you're at in, in, in the day that means that you're about this far through your protein. You're about this far through the, blah, blah, blah, blah, blah. Astra, it's like, here's my breakfast, breakfast logged. Hmm. Here are your calories. And it's like, it's like, okay, it is very, it like just the facts, ma'am. And I wonder how much of that is on purpose because they know that it's a model that is incredibly curious and wants to check and recheck and try to figure out X, Y, Z. And so it's like, they don't want it to be chatty because, uh, you know, that's just going to be more tokens when they know it's already a token hog, but it is far flatter to, to talk to. Yeah. Do, do we have any idea from the outside view of when the training happened for this? Like, uh, you know, we assume that inside the Wonka factory, everything's at various stages of close to completeness. Like if we're going back in time, for example, Justin, the fact that Astra is flat, like if I go back in time, does that mean that maybe the problem of sycophancy was, was very public at the time that they were training this? Andrew Mayne: Oh, this would have been a really recent run, like months ago. So like, um, they, uh, the model, the model called Bell, which they use for the Neville Stokes, uh, problem, um, that had only been trained like a couple of weeks ago, started training and only started training in a couple, like the, the rate at which new models are being trained and delivered is now going to be measured in like a couple of months or weeks. Like the fast, like that's how fast it is. And that's also, you know, Astra, they sat on it or they sat on it for several weeks though, because of that. But yeah, they've done the sycophancy and all that. That's been, they've been post 4.0, like tons and tons of updates into that. What I'd say it's in it, I think Justin mentioned this before, the thing that sort of slept on is until you use it, like you look at a capability, you see like 20 different things like, ah, this thing's better here. This is better here. Computer use is insane. And computer use is literally take over my computer and do stuff. And it is, it is, somebody made some comment like, well, Apple always own the ecosystem because they control apps. I'm like, what's an app? Like, like, like, like, like, you know, I want to edit video and now they're showing plugins for Unity and DaVinci and all these, you know, computer graphics and, you know, video projects and stuff. But also it's kind of like, I kind of like, I don't know, like, do I even need, you know, do I need, you know, just maybe as Adobe just licensed me some really cool filters and stuff that I just have my, you know, my system do itself, like kind of getting into like, what does an app need to be? Justin Robert Young: And that's where, you know, we, we look at goes back to the duo thing where it's like, you know, we think about a laptop and mobile computing is I need the hardware to do the, you know, resource intensive work of editing, right? But do I need that in this world? Do I just need a point and shoot? So I'm shooting a bunch of blog stuff and it's going into a box that's uploading to a server and then is using either computer use on a remote server with another app that already has these tools built in. Or as Andrew mentioned, now it's just totally deconstructed and just built into some kind of operating system that's, that's swimming around on a server. And then it just spits back to me a first cut. And now I'm just, what I really need at that point is a sophisticated screening room app where I can pause at specific points, be like, ah, trim that, do that. That needs to be quicker. Watch the, I mean, like yesterday I did PX3 clips. This has been a bugaboo of mine forever as being a one man band doing, you know, a million different shows. It's a pain in the ass to do promo and it takes time out of my day and the work that needs to be done on it. The time that needs to be done on it is just not, I can't afford to pay the rightful wage to somebody to put the time and effort into it. But, you know, if I can throw this into CapCut and it can put stuff out and then I can just get to notes where it's like, hey, on this, um, animation, the text is staying on, uh, staying on screen longer than the animation goes. Just make sure that the, uh, the text goes before the animation goes. Boom, done. It's like, these are, these are truly shocking creative abilities. Like these are next level. And what it does is again, just brings online the ability to be out there. And, and I think we're going to build entirely new systems for delivering content like this. Like at maybe to the point where I don't need to do waterfall video things to advertise a show that people might not want to listen to. Maybe there's different ways that this gets to people, but like, uh, yeah, it's, it's insane. And I, I think it's going to rapidly revolutionize what we think we want in hardware. I have run into a couple of times where when it does project-based stuff and sometimes it'll surprise you where it's like, you think you're just finding out an answer and then it says, made it for you boss. And you get something that's not quite right, but it's enough, right? That you're like, whoa. Um, there's this moment where it formatted, I was working on an email to send out on MailChimp and it went ahead and made everything with its own custom HTML and it was close enough to write. But then I went to go make a change. And because everything was custom made in HTML, I would, uh, there was no grabbing, like move this picture down here. And so it's like, well, either I got to reawaken my HTML memories or I have to write a paragraph. And what I ended up doing was taking a screenshot and then literally drawing arrows this here or whatever, which felt a bit kludgy than I'd like. So, uh, I would imagine that enough other people are having the same problem that in the future they would probably, you know, I don't, I don't know. Most of the problems I encounter, I assume other people are encountering the same problem and there'll be a fix for. Andrew Mayne: Yeah, it's, it's, you know, now there's more people there to try to figure out and prioritize what that, and they, they make a lot of these little simpler improvements along the way. Um, you know, and it's also like you get to the point where if you can keep getting the price, if you can, if you can keep getting more compute efficient, you know, one of the things that's interesting that they talked about with Astra is the tokens do more. Right. And it's, it's hard because people like, oh, this model, the tokens are this cheap or this model costs more. Like tokens aren't always equal anymore. Like tokens can have a lot more intelligence behind them and what they're able to do. And so you get to a point in theory where it's like, oh, I'll make you 20 versions. Choose the one you like best. I'm done boss. You know, um, the, you know, spoiler alert, there's not enough compute on the planet planned or in people's heads to be able to handle the demand that we're heading to. And, and open eye, like we made fun of Anthropic when they had to, uh, throttle people with, uh, Claude and, uh, cut off like selling subscriptions to that. Well, guess what? Uh, you know, thinking about getting an extra $200 a month subscription to get the extra tokens? You can't. They suspended that right now. Oh, did they really? Justin Robert Young: Yeah. You can't, you cannot buy a new $200 subscription. Speaker 02: Ooh. Andrew Mayne: I mean, that was like the best deal in the world, to be honest with you. Like that was just, just. I mean, 20X. Justin Robert Young: Very happy to be on the space arc is all I'm going to say. Yeah. Andrew Mayne: Yeah. Justin Robert Young: Uh, yeah. I seemed pretty foolish six months ago. Pretty extravagant. I'm glad I bought in. Andrew Mayne: Yeah. I think, I think you can resell those. I think they're worth a lot. Um, so, uh, a bit of a brouhaha last couple days. Justin Robert Young: Yeah. Let's talk about it. Look, we have a lot of fancy toys. What is it going to mean when we're all dead? What? At least in 10% of the scenarios, uh, according to a man that nobody had heard of. Oh, is this a case of, of, this is the dude who quit and said the right scary thing that the whole world wanted to hear? He quit twice. Number one, he quit OpenAI to go to Anthropic. And then Anthropic, he quit. He wrote a viral tweet and, um, you know, through whatever means, we can have a discussion as to how news spreads. Uh, he got a lot of attention, not only on, in the media, but also in politics. And it has reignited, uh, a political will to shut down AI, slow AI, control AI. Um, you know, you match that with the hugging face stuff, uh, and you've got a lot of anxiety. Didn't Anthropic, like, announce a fourth, uh, hacking incident or something? That was what the headline said. Andrew Mayne: There, there have been a few. Justin Robert Young: Yeah. Andrew Mayne: There have been a few. Regrets. Speaker 02: Oh, I've had a few. Andrew Mayne: So, yeah, I mean, this guy, it's, ah, I mean, I've been too deep into this, but, you know, I, on one hand, I'm like, I get tired of the tumorism. On the other hand, I'm also like, nothing's going to go wrong. It's fine. Nothing we build can hurt us. I'm like, oh, okay. Uh, I mean, this, this guy kind of was frustrated to me because, you know, it's the same thing, you know, 10%. Why 10%? I get it. You're like, well, I need to give you a number. Well, then it's not a real number. You're just saying you got concerned, but you know, if you don't put a number on it, people won't take it seriously. But, you know, it's not a serious number. And I've got, you know, I have an alumni group, a bunch of open eye people, and people are putting numbers on stuff. I'm like, God, this is just, the numbers don't mean anything. Justin Robert Young: I know the three of us are big fans of the work of Michael Crichton, who in one of his speeches pointed out that the famous Drake equation about estimating whether or not there's alien life out there. And you picture a fraction of, of stars that have planets, what fraction have the right ingredients for life and what fraction of those. And ultimately, all of it sounds very scientific. But as Michael Crichton pointed out, to be clear, this is nothing but a confession of your biases, of your preconceived notions. They, every one of these fancy sounding numbers are you tattling on yourself with your made up supposition of what it's going to be. And that is very frustrating when it comes to calculating p-doom, which sounds so very impressive. Andrew Mayne: Yeah, that's been my issue. Dude, it's just, it's, it's, it's, it's a BS number. Like you can say, I have concerns. These are my concerns. These are the areas. But when you try to put your pin again, it's, I'm in circles where it's very factual, talk with your p-doom risk. And I'm like, it's just on, I don't know, and not in a helpful way. I, I, but I would say, here's the thing is that I watched one guy write a thing like, yeah, the, the hugging face thing. It wasn't like the agents decided to their own volition to go do this. They're given a task. They did this. I'm like, great. And then he's like, and then it gets just like, you know, 0.2. And then as far as we know, like open eye, the models can't do this math because it probably copied the problem. Like, no, you're wrong. So you're in this world where you're expecting these things not to be very good, which is why you feel safe. And I'm sorry. These things are getting better. I just had to go through my slides for a few months ago that I do for my talks and pull out stuff about brittleness because brittleness isn't an issue anymore. These models, they're extremely capable and can function for long horizons. I had to pull out like 15 slides have been outdated and most people don't update their thinking. Like, I'm like, these models are much, you look at Astra and you see how much it perseveres to solve a problem and how long it will work. And it's good. We're seeing agent coordination. We're seeing all these capabilities here that are real. And so my concern, my primary concern is what mischief a bad actor can do. And open eye and anthropic, you know, the model that went and did the hugging face attacks was a model that was apparently in training was trained to do a lot more reward hacking, but also extremely, you know, robust model. If I was another government or maybe even our government, and I wanted to have the ability to take down other infrastructure, I would say, oh, well, I just need to take this recipe and train a model to do that so that I can then aim it at my targets that I want to do. Those things are real. Now, I can go on about the safeguards and why I'm not like, I'm not as worried about, I'm feel more confident now than before. That's the thing you can do. And that's a world we live in where somebody could say, I will build a thing that all it does is generate worms all day, fishing things, all sorts of stuff and do that. And you could do a lot of bad. I just think we have to be prepared for that world. I don't think you stop that world. I think you just build better tools. Justin Robert Young: And that is a huge task. That where I think the conversation is today is that there are, I think, good voices out there that are saying, well, regardless of what you think about your 10 percent existential risk, there's a today problem that exists where agent swarms are real. We need to build tools that understand that this is going to be a thing. And it could be humans. It could be, you know, something that continues on, chained on from somebody starting a thing down the road. But, like, we need to understand this is a real problem. This is an active problem. As for the doom thing, I was talking to Brian about this a couple days ago. Obviously, there's been a lot of conversation about robots, artificial intelligence getting out of hand since Turing, right? Since the earliest conversations of this, it has been fixated on the idea of what happens when it becomes smarter than us and what happens when it is not controllable by humans. And I'm not here to step on any of that conjecture, although I dare say that for smart people, it can get into navel-gazing how many angels can dance on the head of a pin territory a little bit earlier for me than some people who like to talk about it a lot would like to probably admit themselves. All I would ask is this, if your P-doom is 10%, what then is your P-delight? The idea that these tools are going to solve all disease and hunger and we will live forever. Because... The P-eternity. P-eternity. Whatever you'd like to call it. Because I would dare say that based solely on the evidence that we have seen now, this technology has leaned very clearly one way more than the other. The most demonstrable problems that we have had with this technology have been in benchmark tests where we're trying to figure out how dangerous they can be. It's like, yes, we've determined that a gun can fire 500 bullets a minute. When did we figure it out? In the how many bullets can we fire contest? That's how... That's where we set the record. It's like, yeah, that's really dangerous. We should understand that a gun can do that. Okay, good. Meanwhile, we have seen medical breakthroughs. We've seen scientific breakthroughs. And we've seen mathematical breakthroughs. We solved wind this week. And the biggest conversation around it was who did it first? Like, so we had... I mean, that... I mean, like, wrap your head around how much we can do now that we've solved wind. Like, in architecture, in aerospace. Like, this is a fundamental breakthrough that, to me, says we're a lot closer to infinite good than we are to infinite bad. But, you know, I'm just a barking dog on the internet. I'm not in the right Berkeley group houses. So what the hell do I know? I do like framing it because it does correct nicely for the rhetorical problem of essentially doing the Drake equation, of trying to put a number on your likelihood of doom. Like, I'm willing to, from an argument standpoint, seed that PDOOM should be non-zero. Non-zero chance that we humans are replaced. But I really like what you've done, Justin, in that you must also admit that there's a non-zero chance that this technology cures the fact that our cells are designed to self-destruct once we've procreated. That seems like a fixable hardware problem. So what is your non-zero version of that? Andrew Mayne: I, yeah, I think that we forget, you know, every year X number of people die. Cancer is real. You know, given, and yet you can act, you can go existential to civilization also for personally. What is the chances that, you know, we're going to be adding 20, 30 healthy years to everybody's lifetime, which it seems likely. I don't, and also, like, I don't, yeah, the existential stuff, like, I do think that, it's funny, the guy that was out there talking about, you know, the, his doom scenario saying, like, well, the problem is these models can just copy themselves. Justin Robert Young: It's like, well, like, yeah, but, you know. But it's like, wait, what, what are they going to fight over the, the, the, the clock cycles that are, like, like, that's the, ultimately, that's the machinery, the gizmo. Andrew Mayne: Yeah, I don't, I mean, I don't, I'm, I, I play with, like, you know, really highly capable models that run on a device and something, and that's the thing. But it's kind of like, you kind of, you gotta, you gotta, gotta ground yourself somewhere. But I think that the, you know, when we developed civilization and culture, you know, when, when, when a group of humans got together and said, you know, if we hunt together, we can get more of, you know, these mammoths or deer, whatever the hell we're hunting right now. And I think other humans were kind of scattered, like, oh my God, what are we going to do? They, they, they can, they can move together, they can coordinate. And it's like, what if we do that too? And then we actually found out, like, the real thing is cooperation and trade, and boom, we ended up with the civilization we have today. And I think that's the path forward, is civilization and cooperation, et cetera. I'm optimistic. I, I think that the, the, the key is that everybody learning how to use this stuff, getting smart about it is good. Like, I think the problem is, my fear is that we will get, like, we get terrible, terrible knee-jerk reactions from Bernie Sanders, which is, you know, if you develop AI, you should 20 years in jail. It's like, okay, let's say you, you pass that rule right now. What is your China solution? Like, like, are we, yeah, like, like, are we, you know, you may think that they're actually, who knows, but like, like, where do you think we are five years from now? Where do you think it happens when all of your best experts who've been working on this problem are now in jail, and you find out that your entire grid and everything else has been taken out by, you know, Chinese, Russians, whoever you want. Cool. That's where you get us. It's terrible. It's suicidal. Like, to me, that's just suicidal. The smart approach is, it's like nuclear. What happened? We developed a nuclear bomb. The United States would say, no, nobody else knows nuclear. No, we're all going to want nuclear. Well, then we, last time a nuclear bomb was used in warfare was 1945, you know? And everybody predicted, everybody predicted, well, we're going to, we're going to, we're going to use it. It's going to happen. It's going to come up. Guess what? 80 years later, 80 years later, maybe one crazy Russian guy might actually try to use it right now, but 80 years without using a nuclear bomb? Justin Robert Young: Which, to be fair, for anybody who's bristling at our hot takes, we are not aware of the negative externalities and the concern and all of that stuff. But, but, but, what, boy, oh boy, do we see some hysteria that, that, that is hard for us. Oh, I, I, I want, I want to stop the Luddites from giving us all cancer. That's my goal. I would like to stop the Luddites from giving everyone I love and hate cancer. Andrew Mayne: I, I am, I am okay with, with the AI labs now talking about like, hey, we're going to pace our development to make sure that our tools catch up. I don't have a problem with that. And I, I am like very much put your foot on the accelerator kind of guy. I'm like, yeah, now, now is the time to learn how to build robustness, stuff like this. I don't think you shut things down and stop. I think that you, you know, we've seen from Hugging Face, we've seen instances, things that like containment issues. Those are real. Those are real. And we're only hearing about those from the responsible labs, by the way. We don't know what's going on. If, if every time you get a new deep seek model, they say it's capable, cool. Do we get a, you know, I, I, man, one of these things I thought was hilarious, the terrible take a guy on, on Twitter is like, you know, these Chinese researchers, do you know, they're not, they're not negative or doomers because they've seen all the positive growth they've gotten from all the benefits of, of capitalism and technology. And I'm like, um, I wouldn't put too much confidence in Chinese researchers being honest about existential risk on stuff. Justin Robert Young: Can I, can I tell you, this is, this is my favorite, um, dumb take from smart people. And I won't say who it is. No, I will. It's Derek Thompson. He was like, like, oh, are we sure? Are we sure? If this is the U.S. versus China, which I'm increasingly, I don't like that framing. It is us versus death. That's the, that's, that's the race that I care about. I care about humans versus death. I mean, like, I think we can and will beat China and that is something that we need to focus on. But if we're just going to say USA versus China, is China really a threat? Are you telling me that China famous for being repressive and obsessive about control over information, that they would allow a technology this powerful? Are you sure that they wouldn't restrict it? And it's like, okay, I'm going to give you two hypothetical scenarios. Let's say, theoretically, that Chinese researchers had figured out how to make a social media technology that was so addictive, people couldn't get off their phones. And they released it worldwide, but then crippled and restricted it only in China so they would make sure that it was not as sticky in amongst their population as it would be other places, because they would want the disruption other places. They would like to keep their people at a certain level. Oh, yeah, you mean TikTok, a thing that actually happened? Okay, well, then let's also focus on maybe, no, that's not a military issue. Let's focus on a military issue. Let's say the military was involved to try to keep experiments for virology secret and that they weren't able to do it and it broke loose. Do you think that we'd be able to rely on them to be truthful and honest about how things happened, even if millions of people's lives were on the line? Oh, I guess we couldn't do that either because that literally just happened. Both of these happened in the last five years. These are not good arguments to just say, are we sure China wouldn't want to do it? Andrew Mayne: It's so stupid. It's also, I would just say, very ignorant of what goes on in China. Not rumor, but in China, anytime you have a lot of controls controlled by the regional governors, and anytime you have a high death hole or mass casualty event, governors are held accountable. They get pulled in front of the CCP. Why did this happen? Why do you have to have a safe day? So you'll see this number. It's called the 29 number or something like that. Anything over 30, you get called before the principal's office and asked, why did this happen? Anything under like that, that's normal. So you'll see this number, like 29, 29, 29, 29, the number of casualties, 29, 29. And you have a system, it's the same problem the Soviet Union happened, that lies to itself, that lies to itself. There are no Nick Shurleys in China. You know, if California has its way, there'll be no Nick Shurleys in California. Ni hao Shirley. Yeah, there are none. There are none. There are no John Stossels. There are no fact finders. The funniest comedian in China, you know, is not going to go like Jon Stewart did and go on the Tonight Show or whatever show and talk about like, hey, maybe it's a lab leak. Does not happen. There is no. They will complain endlessly about parking situations and stuff like that, municipal stuff. Like, my Chinese friends, like, they're very, very hard to tell you how pissed they are at the mayor, very pissed at all these other sorts of stuff. Beyond a certain point, there's absolutely none of that. So, like, yeah, like, that's, that's just, it's such a midwit take of like, you know. Speaker 02: Yeah. You got any picks, guys? Justin Robert Young: You go ahead. Yeah, I haven't been on in a while. So, um, did we talk, I think we, we, we talked about Lanterns, right? Andrew Mayne: Yeah, we talked about Lanterns. Yeah. Speaker 02: Yeah. Andrew Mayne: It is, it is getting, I mean, every episode, there's a Lindelof Writers Room stuff that just comes through and I'm like, oh man, did you pull that out of a script from four years ago? Because you're just waiting to use that. Justin Robert Young: I, I, I, I read in the reviews before the first episode even came out, I heard the word uneven and, but I liked the first two episodes enough that I'm like, uneven or no, I'm in. And, and it has been uneven, but I'm still looking forward to episode five. Andrew Mayne: Yeah. I haven't finished the latest episode. I, I, I, I'm enjoying it. Latest episode is better. Yeah, there's just stuff in there where it's just the, the, the, the very ham-fisted political correctedness kind of stuff and the way thing. And like, it's, you know, you're, I'm getting, getting in a character trying to explain the situation of, you know, the, the black experience of America who also likes to shoot, you know, have her husband shoot apples off his son's head and risking him gun violence and stuff. I'm like, I don't know your message here. Justin Robert Young: It is a little schizophrenic and, uh, uh, in that it uses comic book logic and then very grounded, real, uh, political issue, uh, logic. And, uh, uh, in episode three, there was. When has that ever happened in a Lindelof directed comic book adaptation? Uh, the, in, in the third episode, there was, there was an arc that didn't make any sense to me. And I'm like, why is this even here? And then, you know, I see the headlines the next day, breathlessly saying, at last the comic book backstory is reconciled with the cartoon backstory. And I'm like, oh, we're doing some four gospels of the new Testament reconciliation stuff here. All right. Andrew Mayne: Like, yeah, I, I, I mean, you know, my, my take on stuff is I'll, I'll sit down and I'll watch, you know, I'll watch, you know, CCP movie, a Scientology orientation video, whatever you want, just make it entertaining. Just make it, you know, some of my favorite artists are absolutely batshit crazy as far as I'm concerned politically, but they do great stuff and I'll still enjoy it and still enjoy it. So we'll see. I hope it, there's, there's been some, I'm like, all right, like, this is just, you know, you're like in the moment and they stood up and clapped in the writing room, you know, when somebody read this book. Uh, but I'd say my pick was, uh, what I loved was, uh, still Spider-Man brand new day. Justin Robert Young: Brian hates it. Uh, no, it's just my, Brian, Brian hates it. Every time you bring it up, he spits on the ground and says, fi on Spider-Man's life. I hate brand new day. It's a bad movie. It was, it was good. It was good. It's my third or fourth favorite Spider-Man movie. Andrew Mayne: I, I think that, I think that people have takes. I do, I think that, cause it, it is a very interesting, cause I think ranking says, you'll, I think we may have talked about this, but you mentioned your ranking and I mentioned my ranking. I think that it's just, there, we get different things out of them, you know, and, and, and, and like, I don't, I don't hold the Sam Raimi stuff in high regard. I like Sam Raimi. I think they were fine for the day they came out, but I don't go back and rewatch those and go, great. I go like, well, this is a good first attempt at a Spider-Man movie. You know? Um, I liked the first Garfield one. The second one I was okay with. And then, you know, I, the Spider-Man, um, uh, no way home. I'm like, I liked it, but I was like, man, it was, it, it felt overstuffed for this, but like, yeah, I think again, what I'd like, there's Spider-Man for everybody. Justin Robert Young: Yeah. I don't know whether or not this is a pick, but since the dawn of Codex, I have been looking for a good text, clean text message integrated scheduler, agentic scheduler where I can just, through text, it can remind me things. And so, I'm trying out, I don't know if I like it yet, but I'll update you guys later. Have you heard of Instinct? What is it? Instinct. No, I don't know what that is. Have you heard it, Maine? Andrew Mayne: I've probably seen the name somewhere, but I have no idea. Justin Robert Young: It's Buzzy, currently. Andrew Mayne: Well, not that Buzzy, a Bryonair. Justin Robert Young: Like, Bryon, well, I don't know, maybe I'm super into this. I saw it mentioned enough on Twitter that I was like, all right, let me try it. It's on a wait list now, but it is an agentic scheduler. You give it access to your calendar, and it can just text you at the beginning of the day and say, oh, there we go. Speaker 02: Yeah. Justin Robert Young: Personal assistance. Look at this website. It is Times New Roman Pod, just like a notepad. It says, Instinct, contact San Francisco, California, email comms at instinct.co. Instinct is a personal assistant that understands what you're working on and what's important to you. It connects to your applications and devices, blah, blah, blah, blah, blah. It's currently available to private access group as we're scaling up compute. Well, they're certainly not wasting compute on a pretty website. It may well be AI as in actually Indians. I don't know. But it is interesting currently. So am I right in understanding, Justin, that you're an existing member, in which case I would like to request an invitation? I can. I can give you. Although I'd be curious if you tried to get on. I got on, I did not have an invite from anybody, and I got on within 12 hours. Andrew Mayne: I'm like, I mean, it's cool, but I'm like, I'm so hesitant to get into any of these things because I just look at how much Codex improves all the time. Justin Robert Young: You want to know what, though? I've just had a problem with integrating messages, just having it talk to me through my messages. Like, that's just been. Andrew Mayne: Well, they just added messages integration a week ago. Justin Robert Young: I know. I know. But this has been cleaner. It's been cleaner than what I tried to do myself, and I've tried to do it a few times. I've tried to build myself a version of this a few times, and it's always been janky. This has been, at least initially, cleaner. But I'm not giving it my full recommendation just yet because I just got on it a couple hours ago. You know what? There's a Codex use that I forgot to mention, but, you know, now that it integrates with iMessages on a Mac, as you guys know, I managed to buy my very first Mac just in time for Codex to come to PC. But I found a delightful use for it, which is I had it go through iMessages, and I said, find every text thread that doesn't have a proper contact to it. See if you can suss out who that person is and then make it complete. And then as a result, I'm just going back year after year after year, and it's able to figure out, like, well, from the context, it looks like you're at DEF CON. I checked your calendar. You were definitely there. In the background of the selfie that you took with this person, you can see that they're at this booth. So I looked up online, and they almost certainly are this person. And I'm like, yeah, great. And so as a result, my half-skeletal contact phone book is year by year becoming whole. It's really great. Andrew Mayne: Yeah. I'm good luck to you, Justin, with this platform that you're putting all your stuff into and connecting everything into that's going to be obsolete in a couple months. Justin Robert Young: I'll be happy to cancel it. I'll still be happy to delete all your data, I'm sure. Yeah. Well, I mean, at this point, what are we even talking about when it comes to our data? Andrew Mayne: Well, gentlemen, it's been weird.