mergeconflict275 === [00:00:00] James: Frank, Frank, Frank, Frank on the road again, how sweet. [00:00:14] Frank: Ooh, it is sweet. Well, actually I got detoured again. I keep getting on the road and then falling off the road, but I'll be back on the road tomorrow. It is sweet, but hot. It turns out the south part of the United States is hot newsflash. There's a reason I live in the north looking forward to getting back up to the north. [00:00:32] James: I believe that they're, they're closer to the equator. [00:00:35] Frank: Oh, is that how it works? I thought the sun was closer. I don't know. [00:00:40] James: I think that's because they're closer to the equator. I don't know. I don't know how the, I dunno, right? Yeah, you're [00:00:45] Frank: right. You're right. I'm just trying to make terrible jokes. It's been a, it's been a good driver. It turns out I actually really liked driving. I think I've said that already on this podcast, but I'll be happy to get back to my iMac pro and a regular schedule. [00:00:59] James: There you go. Yeah. I don't mind a good six, seven hour drive, eight hour drive max, but then doing that for about a week, that's quite a lot, to be honest. Yeah, [00:01:08] Frank: I've been, I limit myself to 600 miles. Cause after that I start getting cranky. So 600 and I find a hotel or something to go to in with a lot of fun during the COVID COVID times. So I just, you know, I kinda wrapped myself in a cocoon don't touch anything and just asleep while standing. That's easy. [00:01:29] James: There you go. Well, I mean, that's the fun part about the cross-country road travel, right? Is that you get, you can, you can do that. If you can afford to take the breather and just kind of stop along the way and make it a little adventure. I think that that's fun. And, um, I hope that you have the safest of travels and that, uh, and that you do return to that. Beautiful. I iMac pro [00:01:46] Frank: oh yeah, I miss it. I hope it misses me. [00:01:49] James: It does definitely. It definitely does. I, I got a topic for you this week, Frank and. Squid game. Do you mind? Squid game, squid game. Good game. [00:01:59] Frank: Okay. Uh, you kind of have me at a loss here, so this is going to be a good interview show James, what in the world is squid game. Okay. [00:02:07] James: So, ah, squid game. It's a Netflix show. Do you know that you know about it now? [00:02:12] Frank: Yeah, I, I believe I see a logo in my head, so this is interesting. Is it a game because Netflix is movies, but they did that one game. They did it. Yeah. [00:02:23] James: Great question. So when I saw it squid game, uh, on Netflix and it's being promoted to everybody, I was like, oh, interesting. Um, I was like, this must be a new game show. No, it is not a game show. It is a survival drama television series out of South Korea. [00:02:43] Frank: Okay. I, I have no idea why we're talking about this, but I am here for this. I am a survivor lover. So please tell me it's anything like survivor. [00:02:52] James: Okay, so I'm just, so I'm going to give some public available information here. So if you haven't watched any of squid game there, won't be spoilers about the. Um, however, uh, we're going to talk about the first episode, um, specifically red light green light. Um, so if you haven't seen it, I'm not going to spoil what happens. We're going to talk about the technology behind this specific game, right? So here's, here's the see the series synopsis on Wikipedia? No spoiler alert. If you watch the trailer, this is what you would see. High level the series centers around a bunch of contestants, four or 500 contestants that are sort of drawing. To this, you know, gain from all different walks of life, they're all kind of deeply in debt and they are set to play a set of children's game with deadly penalties for losing basically. Okay. Think of it as a cross between hunger games, deadliest game and survivor, right. Except for, yeah. Like hunger games, people die. Right. So like same thing. So like there's Def as a. [00:03:58] Frank: I was worried. You were going to say it's like saw, but these are people playing children's games and death is a penalty. That's terrible. [00:04:06] James: What are we doing? That is correct. So it's not it. It's, it's intense. Uh, and it's very good, but it's also sad and emotional because it's drama. And you think about yourself, but let me set up the first game here. The first game is actually what you see all the time on Netflix. Did you see this huge. Uh, like ceramic doll. And then you see all these people behind the doll, and it's kind of like a bloody mess at this point, but that's, this is what you would see if you go into Netflix and it's a big blank banner. And what they have to do is they're told that they have, um, they're playing red light green light. If you, you know, red light. [00:04:47] Frank: I feel like I used to probably when I was four years old, what are the [00:04:50] James: rules? Yeah, it's very simple. So even say at the game is when the person who's in charge of red light, green light, um, is not looking so they have the back to you. That's Greenlight and you can run. If they turn around, they scream red light and then you have to stop. And if you move, then you're out of the. [00:05:14] Frank: Okay. Yeah, maybe I was six. Okay, got it. Yes. Children's [00:05:18] James: game six years old. You were in we're in the mind space, right? You're in, you're out there. You're over there. You're going. It's great. And you're playing in the field and red light green light. You get caught. You're out of the game. You move off to the side and now you got to go sit in the corner, just like, you know, in Dodge ball, if you get hit, you got to go set off. Right? So red light green. So here's the technology that I want to talk about here in, if you're playing red light green light in real life, Frank, there is no technology, right? You are. There's like a person that would be the gatekeeper of did that person move? [00:05:51] Frank: Yeah, we could call them a Semafore to turn this into computer science. [00:05:56] James: That is correct. There are the Semafore. Yes. Um, per perfect. I like how you bring this bag. [00:06:03] Frank: I'm trying here. I have no idea what we're talking about. [00:06:06] James: Okay. So, so, so that would be the distinguishing part of it is, and there's also a truth system to that, to that game as well. Right. Because at this point, Um, you know, if you move, if someone sees you move, they'll call you out on it. Or are you of yourself. If you fall over, you're going to pick yourself up and move now. And it's a great game though, is a fantastic series so far, very intense. Uh we're we're not even done yet, so I don't even know the conclusion. So in this red light green light, they have, I think it's five minutes to get from the start, cross the finish line, and basically think of it as a huge robot. Okay. There's a huge robot who is. Uh, Semafore in this case, who's basically turning around and flipping around. So the, the robot had is away from you and it's Greenlight and you move. And then it turns around its head red light and you got to stop. Okay. Now at this point, Since it is the squid game and they're in this really intense elimination match. If somebody moves, they get shot and the die, their sniper rifles, right. This sad. Okay. Okay. So it's intense, right? So this is what's [00:07:18] Frank: happening in the shooting robots, a bad reputation here. [00:07:22] James: Robot is not doing the shooting, but here is the thing about the robot that I want to talk about today in the robot. It's eyes are motion detectors from. Oh, it's like a huge IOT robot. Okay. And this big robot is scanning the room for, for all sorts of different motion for basically motion, right. For any type of motion that's going on. And what I want to talk about today is the accuracy of this potential huge robot that has people's lives. And it's hands, right? So it's scanning the room and it's going crazy. It's looking for emotion and it's pinpointing people, right? So it has to not only diff detect and find the person in the movement, but it has to be able to read the number of them to understand, Hey, number 400 or 300 or whatever. Is going on now? You haven't seen this at all, but are you picturing this in your mind? [00:08:19] Frank: I suppose. So I still don't know what shape to give this robot. So I'm giving it like a really 1950s kind of scifi look, it's a really cute, like 10 robot with scanners all over it. But you do have me thinking already. Of course, I'm thinking lasers, like connect style lasers. I'm thinking. Basic AI, I'm thinking you just diff some frames. I'm thinking what are the, what happens when the humans cover their numbers? Because humans are great at defeating stupid AIS. So w where are we going with this? [00:08:52] James: All of that. I'm going to send you a link here. This is a photo of a robot, um, which is a big life size, like. Schoolgirl basically. Well, that's [00:09:02] Frank: freakish. You just had to ruin it. My robot was way cuter, everyone. If I could describe this, think of a creepy 1950s doll, not 1950s robot, 1950s style and acute dress with a, a deadly, deadly staring face looking at you. And it's way too big for a [00:09:19] James: doll. It's it's ginormous, right? It is ginormous. Yes. So she's standing there. She has to scan the room. Motion detection. We'll, you've done a lot in the space of motion detection, people, detection, all these different things that are in it. Right. We have all these things, right. I have like smart cameras and they know if an animal mood or a human mood or, or anything moved, but it's not a hundred percent accurate. Like the actual there's a threshold there. So what I want to talk about is what kind of. In today's world is doing motion detection and how accurate is that thing? [00:09:53] Frank: Oh, okay. So do you want to talk about like consumer tech or state-of-the-art because I think that there's quite a Gulf between them. I have a friend who loves buying surveillance cameras, and so I get to see a lot of the funny, different softwares out there. And it's surprising that most of the. Image detection software. I see out there is kind of stuck at 2014 levels of sophistication, but, uh, that does not mean the state of the art is still stuck at 2014. So it's interesting to see this stuff distributed. Uh, what do you think the show is? Do you think they're using high tech stuff or do you think they're using consumer grade? [00:10:31] James: Well, I have to imagine. So what we see from the, the visuals in the episode is you do get to see. The inside of the computer, right. They do the first person view outside of the computers are outside of the robots eyes. And you see her again because in the game, in the game, right. She's just this little school doll. So, so it, the robot and you see you're scanning constantly. Like her eyes are moving and it's scanning up and down and doing all this stuff. And, uh, it looks like it's scanning with both eyeballs. So both of them. We'll have two independent cameras in it. And I have to imagine that they have to be really high definition and see really far. Right. I'm thinking, you know, those little $30 cameras that you buy in your house. The motion detection, especially at, at night, this is during the day, but it's only for so far into the distance. Sometimes that it can do it or detect a full Mo for movement, for example, of, of a, of an entity of a box, basically. And it's drawing these boxes around and it's looking at different heatmaps. So the first question would really be at this level, they have to be extremely high end. Video cameras to be able to see this huge field right there running for a long time across this field, maybe it's a football field. Let's say maybe a half a football field, half a football field. Let's say that's all a long distance to, to, to, to, to analyze and see the numbers and see the human beings on. [00:12:07] Frank: Yes. Yes. Uh, so a few comments on that. Number one, the two eyes moving in sequence that could just be for dramatic effect because certainly we expect the eyes to move. There's nothing saying that you'd necessarily need, um, stereoscopic vision for this problem. Depth is nice, but it's not always required. Uh, even if you're tracking people in a field, but the big benefit here is that this. Ghastly doll is giant. This, I don't want to call it a robot because it's an evil robot, but, uh, you could put like a full frame camera in each eye. You could put a gorgeous camera with a really nice lens. So you want to catch a large playfield. You figured. How big that playfield is how tall this robot is, and you can probably buy the exact right lens for the exact sensor on the cameras that you bought. Uh, the show looks like it has high production quality, so I'm sure that's probably the route they went and they probably just have some beautiful Canon cameras for the eyes. [00:13:10] James: So, so, yeah. So, okay. So it has really super big. You know, camera's in there, which makes sense. Cause it is big. So, you know, when we think about those little cameras, they're there, they're tiny sensors, right? So the image quality coming off of that is relatively low to, to read you, you know, you're buying a camera for 20 bucks. You're not going to be able to probably read a hundred percent texts from far away. Right. And then the numbers on the numbers on people's shirts or jackets are relatively small. So it's not like they had like. The huge numbers everywhere. [00:13:45] Frank: Well, and importantly for that are things like motion blur, because the way a camera works, if you have a rolling shutter, I think that's what it's called. I'm terrible at camera terms, but that means that they're sampling, uh, the buffer over time. And so the buffer doesn't capture an incident in time and you can get. Bendy affects if something is rotating or if the, the number is rotating on their clothing, things like that. Clothing is always a bear in these kinds of problems or, uh, the bigger issues are also exposure times. So all the tricks that an iPhone has to do, you know, you take a picture on an iPhone, it takes nine pictures at nine different, uh, exposure levels, and then does crazy math to integrate all those. Picture that you find we get in the end deep fusion. I think they like to call it so necessarily they're going to have to do something like that. The interesting part of this, and I don't want to skip ahead, but if you are actually going to analyze this, it almost doesn't matter if you have the highest quality camera with the most megapixels on the whole planet with super resolution, if your neural networks and things like that, can't process. That amount of data. So the optics are a tricky thing, but they're almost easy. You get a decent camera that does a full frame capture. You take a few exposures, you do a little bit of integration. That's kind of easy, but fitting it into the algorithm can be hard. [00:15:14] James: Okay. Two things I want to hit on first. Let's go back just a little bit, which is the eye movement, right? So there's no, none of my cameras move in my house or anything like that. And as long as they can see something, they would be able to see it. So if this thing can already see the whole play field, there's really no reason for it to be moving its eyeballs, besides the dramatic effect. [00:15:36] Frank: Yeah. Yeah. If it actually had to move its cameras, then you can have the humans have another advantage. The humans can distract it while they run a flanking maneuver or something like that. So it's just curious, it's almost a game rules. Can the robot see the whole field, but at the same time, I'm curious what it actually can do, uh, when they visualize it on the screen, do they show the entire field or did they just show like zoom and enhance? [00:16:01] James: It looks like it's like zooming in and enhancing on different parts of movement. That's why it's moving its eyes because it's like, it's move. It's like zooming in and then focusing in on, on each individual to basically try to pinpoint. [00:16:15] Frank: Do people ever win against this robot? [00:16:18] James: Um, okay. I don't want a spoiler alert. It says no spoiler alert. Um, and it, but you know, if you follow the rules, you're there. So let's get back to that, right? Because this is the first game and there's 456 individuals. So let's talk about that algorithm of processing it because you know, when, when it, when my little camera detects me, That's sitting there exactly. Then I forget to turn it off. It detects a person or maybe two people, but it's not processing 456, you know, things on the screen. And let's say there's a bunch of movement. What does that load like on, on a system? Right? Because it has to be analyzing every single frame. Can it just look at a frame? And there's 456 objects and can it in that frame be able to map and do all of that in real time? You know what I mean? Like that seems complicated. [00:17:14] Frank: Uh, in fact, there are neuro networks that do this. Um, I 450 sounds like a big number, so I don't want don't hold me to the, whether they can actually do that quantity because what I think they're doing is probably different. But I'll talk about this first. Uh, it turns out if you try to. Tele neuro network to just find one person in a picture. And there happens to be another person's arm. And the picture is something they're going to have to learn that people come in different forms and sometimes they're obstructed and things like that. And so it's actually more natural for the neuro networks to learn multipurpose. Things. And so instead of like taking an image and doing bounding boxes around objects, like in the Terminator, what you actually do is for every pixel say, what is the probability of being this object or that object. And in that way, it almost doesn't matter how many people are in the scene because you're just labeling pixels. You're not labeling bounding boxes or anything like that. So as long as it's in the screen, it's there that said that's probably not the technique they're using, because even once you get that blob, uh, you still have to do it with some kind of secondary analysis. [00:18:24] James: And in that analysis, I mean, granted, this is a huge robot thing, so they could be putting supercomputers inside of it. But let's say it is highly technical. It doesn't seem like it's military grade stuff, but like, Is it feasible that, that a system like this could exist? To process that data a hundred [00:18:48] Frank: percent, no problem. You get yourself, those high resolution cameras. And the neat thing is that you can process the frames of those cameras in parallel, by breaking the frame up into different chunks or different tiles. So put a grid over the frame and make that grid as many computers as you're willing to spare. And so it's really comes down to computational power and you. Identify all the objects and just that little grid square instead of in the entire image. And you can keep doing that. Keep doing that. The theoretical limit is, you know, down to one pixel, just identify the object in that one pixel so you can subdivide the problem. That's, what's kind of really nice about, uh, real time imagery. Most of these algorithms, if you tried to run that on your iPhone, you're, we're, we're not talking. We have to also talk about what's real time and what's not real time, but, um, an iPhone could run that. Five or 10, uh, uh, frames per second. Who knows? Is the show running at five or 10 frames? A second. Is it running at 30 frames per second, but either way you can keep subdividing the problem. There is a last integration step. Obviously you have to take the results of all those sub-analyses and feed them in. But I feel like that's going to happen anyway, because you're going to throw out all the grid tiles that haven't changed or that the algorithm decides, oh, the wind is just blowing. That's just the wheat in the field blowing or something like that. So first you run it through an algorithm of, are there humans in this grid, tile? And if so, then you pass a higher resolution version of that grid tile up to another process. Then you try to identify the individuals in that frame and then pass that all up to a final integration process. This is all yeah. Very doable with today's technical. [00:20:35] James: Got it. It's kind of, kind of, kind of cool to hear you step through the process of, of building, you know, if we were to build a red light green light analyzer, right. Uh, you know, friendly version of it. Uh, that's what you'd want to do is, is, is, is feasibly there's. There's a bunch of people like moving through a field and doing that. And that becomes the other point of reference that I want to talk about is. The PO potential error. Right. And you talked about, is it wind moving? Is it, uh, uh, blades of grass? Is it, uh, uh, dust coming or what if someone's behind somebody and moves their arm? Right. How do you detect that in this, you know, in this scenario and is that something that actually can be accomplished if you and I are moving in parallel and I move my hand behind you? Ease is the computer. Like a physical person could detect that, but could a computer, which is what the show is set up, analyzing these things was. [00:21:36] Frank: Yes. Yes I can. And I'm so excited because this is actually a semi recent development. So we're talking state of the art here in 2021, but, uh, those kind of look for the human in the image algorithms. They worked with images, uh, flat 2d representations, and we're doing kind of pixel labeling or bounding box labeling kind of, as I was describing. But when I came down to what's called human segmentation. So you want to identify the hands, the arms, the shoulders, the torso, the hips you want the whole human. Do you want to know individual parts? Uh, the algorithms would generate some real grotesque beings. You know, things that aren't quite human, because as he said, if an arm is behind another person, it just has a hard time figuring out exactly. Um, projecting, it's having a hard time using its imagination of where that arm could be. So what's been done what the actual state of the art is, is looking at that 2d image, try to fit three dimensional human shaped models into that image with the constraint. It has to be humanoid. It has to have joints where humans have joints. It has to have, uh, rough proportions that are within limits of a human and the algorithms. Instead of identifying pixels or identifying bounding boxes, it's taking this kind of dummy human template, human, and just kind of morphing it around to fit it into the picture. And it'll do that for multiple humans. And once you have that, then you can start. It's naturally solving the, uh, it's called the penetration problem with this terrible phrasing, but it's the occlusion problem of someone's behind something, or just think about someone twisting their arms all around it. Humans can get themselves into funny shapes. It turns out. And so the best way to do that is to take this. Three-dimensional model it's even an anatomic model because it knows the limits on all the joints, that standard limits. Of course, people are crazy and they change their limits, but, uh, and it'll try to fit that to the picture. So the modern stuff actually can, uh, obviously it can't tell you what color is their clothing, if it's obscured, but it can say with 98% probability that arm is there behind that other person. [00:23:59] James: Wow. Okay. So that is, uh, A great sort of analysis because as I'm hearing you feed it in and sort of makes a lot of sense, right? Because you're talking about like a three-dimensional model. So, so really it's these cameras have that depth, depth of field is really what we'd be looking at then. [00:24:19] Frank: Well, right. So that is almost a whole separate question. These 3d models are used just to make sure that the results make a logical sense. They're making sure that we don't get like three armed humans or, you know, two headed humans that are just things that are not probable, uh, There is the whole question of if it has actually two cameras, it could technically be doing stereoscopic vision and actually creating a depth map of the field. The benefit of the depth map is you can measure distances between objects or a goal or something like that. It's. Creating a Cartesian plane. You're saying this object as this X, Y, Z coordinate, and you are actually putting them into three-dimensional space. It's debatable whether you actually need that for this kind of problem. I haven't seen the show, so I don't know what kind of problems arise, but I'll say even for my robots, I'm starting to look into three-dimensional stuff instead of just 2d because you know, it's there it's good data. It's doable. The math is somewhat known at this point. So they could definitely be creating depth maps, whether they're useful or not. I don't know. I'd have to see the, [00:25:32] James: yeah, I don't think they got that deep of, I wish you could find a clever and you can watch the first episode. I mean, I think that that all happens in the first episode, so it gets pretty intense pretty quick. The thing about the show is that it is obviously viral. And bloody, but it's not to Russo them. So it's not to the saw extent. I mean, it's all one and two were kind of like, you know, fine. Um, but like saw three and four or five, seven, 8,000. Like they all just got super way until I stopped watching them as too much for me. Um, I was younger and then the day, I guess I had a highly higher, um, tolerance for that stuff, but I don't anymore, but there there's a great story backend here that that's really in depth. It kind of reminds me of, you know, parasite, the movies you watch Paris. [00:26:15] Frank: Okay. I feel embarrassed because I think parasite is right up my league, but I haven't seen it. Yeah. Okay. [00:26:21] James: Anyways, so you got it. You got a little bit of research to do on your road trip, so, all right. So it sounds like so far, what I'm hearing from you is that the technology in this show is reasonable. There are some potential, um, things that they're doing just for cinematic effects, like moving the eyes, just to do some stuff. Not necessarily, you could be helpful, but not necessarily. [00:26:47] Frank: Well, we we've left off one part. Um, once you do object recognition, object tracking is kind of a whole separate thing. What you're trying to establish is what is the movement of an object over time? And that's not exactly predicting a human humans are unpredictable. They can do whatever they want anytime, but it's just more like if you're rolling down a hill or if you're running at a decent clip, chances are you're going to keep going in the same direction. And so you can do predictive analysis things like. But it also solves it's also helps solve the occlusion problem of if the object disappears for a second and then reappears, is that a new object or is it the same object that disappeared for a second? Uh, that can get into philosophy too, but, um, it's a hard problem. Object tracking, uh, which is separate from object detection. So that's a whole different level that I don't think people give quite enough credit to. Uh, if you don't do that, then you get those jumpy systems where people can pop in and out of existence. And that's [00:27:46] James: bad. Got it. Yeah. And you, and that, that is something to talk about because as this robot turns around, it's just it's head. That turns around by the way, even creepier. Um, but it analyze it, the field and it has to take a snapshot and then analyze that motion. And I really couldn't figure out what the threshold of movement to detect movement is. Right. Because as human. I mean, I can stand pretty still, but is it blinking? Is it, is it, is it a pixel? What is a pixel? How, how granular is that? Especially with these cameras are so good. You know, what is the tuning of that system to detect actual motion and moving. [00:28:33] Frank: Well, I'll tell you a consumer grade stuff is terrible. I was just looking at some cameras that do all this object tracking and everything, and their false positive rate is terrible. Especially if it is a spider decides to climb out and the lens or something like that. Um, it's. It's one of those problems that it's really easy to give an 80% solution to. And it's really hard, nearly impossible to get about 99% or a hundred percent solution to, so these cameras that. You know, or detecting is someone walking by your house? Are they nefarious or not? You know, are they gonna come bashed in your door or is it just someone walking a dog? And in that case, maybe you don't want to turn all the spotlights on and blind them in their eyes. But as it is, you know, that threshold for movement. That's almost like, gosh, I hope none of these things even work that way. That should not be how they work. They should be recognizing what is the object and what is the intent of that object given its motions. That's how these systems should work. And. So does it actually work that way? This is a Netflix production. This could all be smoke and mirrors. It could be a bunch of poor interns back there looking at a thousand camera feeds and trying to target them or something like that. But, uh, chances are, if they actually got this thing working and in production, then they would be using technology from just a couple years ago or three years ago. Nothing, uh, too old. Nothing's state-of-the-art so. I almost think they probably already doing, I've just talked myself all the way around in a giant circle to back to you're probably right. They are probably just thresholding everything. Yeah. [00:30:16] James: And there's, there's gotta be a certain threshold that that's there for it. Cause that's the complicated part is, is if you watch through it, there's, there's questionable moments where like, uh, trying to understand what that threshold is in the episode. And I'm like, okay, well what is the movement threshold? Because is it, you know, Yeah, like I said, is it blinking? Is that movement right? It pixels have changed on the screen. That's what I'm trying to say. You know, pixels, you blink your eyes, pixels have changed. Is it a pixel change? Is it a pixels have changed so much? And, and, and, and is the errors that did the bounding box move? You know what I mean? What is it, [00:30:54] Frank: if they do it that fancy way where you're actually registering 3d models to it, then you can simply say through the hips move than the hand. Both. Yeah. You don't even have to zero in on the eyes because ma you know, that's just more pixels to deal with. You can just give it more general parameters of has the mannequin shifted at all. [00:31:15] James: There you go. [00:31:15] Frank: That's what I would do it. Wow. You know, if this actually is, um, a computer running, all that stuff, then hats off to the engineers also creepy 1984 vibes, but mostly hats off to the engineers because it's easy for me to sit back here and say, oh, this is state-of-the-art this isn't state-of-the-art. But getting a functioning system, working at all, like this stuff, it's really difficult. And so [00:31:40] James: good for them. Yeah. That's cause you know, my, the cameras that you buy, these little consumer ones, even the ones that have the little UV cameras there during the day, they do their best to detect human motion and report back. But at night, as soon as dusk hits, it's like, just add anything that moves. We, I always report everything. We, you know, we don't, we don't know. Right. Because the threshold level of them being able to analyze it is. It was just too, too low. They just there's anything, any movement we're in, you know what I mean? And that's car movement, that's Deere movement, that's human movement. Or sometimes that's just a spider moving over to camera. There's like, blah, blah, blah, blah, blah. Right. I of kind of find it, think about, yeah. You know, I, we, we haven't really done a diagnostics of shows or movies that are supposed to be set in present day and break down the tack. And we've talked around a lot around AI and different things like. It's kind of fun as I was watching this to be like, is, is this actually real? Is this real right? Because we've talked about machine learning models and tracking cars and Tesla and, and what's next for, for driverless vehicles and that's all motion and mapping detection. And that stuff is rubbing there's cars that are going to be driving around. Right. That, that do this stuff. But. Trying to think about in the real world of, okay. Like how long is that taking them to come up with that? And if this is real world to be able to detect this, is this even something humanly possible. So it's kind of fun to talk about through, through technology, even though it's a. It's an odd one because the show is relatively gruesome and like not, uh, not a, not a great thing for all things, but if everyone's watching the show, definitely let us know right. Emerge conflict that FM. And, um, let me know what you think of as good game. If you want more squid game analysis, the other games that I've seen so far, I have not been tech-related at all. So in this regard, I thought this would be kind of a, just a fun, silly episode. [00:33:40] Frank: Yeah, it was, and I love Saifai. So I I'm totally here for at least for that episode. I'll let you tell me if I should watch any more episodes, but yeah. Uh, I just want to say that if you're at all interested in the subject and want to see a fun technical, deep dive for an actual real system Tesla and the. I don't know what they call the Tesla day. It was like a month or two ago. They did a real nice technical, deep dive on how they do, uh, all their map generation and high level, how they do their navigation with their automobiles. But it's real nice to see a practical system built and they actually talk about all the troubles they had integrating all the data. Doing things at different resolutions, building a cohesive map, making sure that it agreed with reality, all those important things that, you know, it, you don't, you can ignore those for a neat tech demo. But when it comes to working in the real world are really important. So I recommend the Tesla tech day thing. Yeah, [00:34:42] James: I'll have to go watch that. That sounds really cool. [00:34:45] Frank: Yeah. It's I mean, I don't know why Iwan allowed it. They gave away a lot of nerdy stuff, but I guess they must've decided that the whole industry has at least roughly caught up with them. So it was okay to talk about it. [00:34:58] James: Yeah. Interesting. All right. Well that's good game. Um, yeah, I'm so far liking it. Um, it's good. It's [00:35:06] Frank: intense. It's good. I get nervous when he keeps saying it's intense. I, I love horror films, but at the same time, I get a little bored from horror films. And so this is not sounding like a horror film. No. So I'm a curious how it's going to feel when I actually watch it. So maybe, maybe we'll do some followup. Yeah. [00:35:26] James: I'll do some the follow up. It's a good schizophrenia. Yeah. I, I, yeah, so far I'm enjoying it because there is a lot of story you're starting to get like, You know, connected some of these characters. And again, we'll just see how, how, uh, how well it does and it's all dubbed, right? So it's all in Korean. So it, you know, you got to want to read some words on the, on the screen and really pay attention. So, but it's good. Um, yeah. Anyway, uh, if people enjoyed this sort of analysis, let us know because, uh, you know, Frank and I are nerds, uh, we love analyzing technology. Uh, I think, you know, uh, one of my early things in technology. That I always thought about and we've talked about is Ironman, right? And, uh, and just everything is happening only in the suit, but also just in his physical space in his world that he's created. Uh, I don't know. That's one of Frank's hot topics. If you're ever interested, give us your recommendations of, is that tackle really possible or when will it be right into the show at merge conflict that FM there's a contact button. Your head is up on. Well, that's going to do it for this squid game. So until next time James wants mag now. And I'm Frank [00:36:33] Frank: Krueger. Thanks for listening.