The Return of Claude Fable 5 | First Impressions

0:00 Anthropic's best model Fable is back, and today I want to do a slightly different video to what we normally do. Normally, we look at the models, compare them, try to deduce where my maybe models are not very good, maybe some other downsides, but today I want to do a very indulgent video. I just want to savor the moment what kind of model we have access to, especially for a few days while it's still in the part of the cloud code subscription.

0:28 And I really just want to show you guys some of the cool generations that I've done. I've done I think about 60 old generations, and I want to show you some of the best ones that I got. So, today will be more kind of imagine like a delicious cake. You get a cake, you don't want to ruin it, you don't want to think about the calories, you don't you don't want to think about that you paid $18 for the piece of cake.

0:53 but the fact that it is just a delicious cake. So, no government regulation, no blocks, no rate limits, just what we can do. Fable here on the arena on the agent arena got the highest score ever, and we keep testing it now to get the more up-to-date score as well. And I'm very excited to see how it's going to work for all of you guys on the arena. but today I've got a lot of different tasks that I've done. They're going to all be these kind of 3D tasks where I'm giving it very difficult prompt, typically that very long prompt, sometimes a bit shorter. And then it goes off for a while, and then it comes up with something. A lot of them are one short, but a bunch of them maybe I've done a little tweaks as well, got it to make some improvements.

1:44 So, let's get into it. Let's see what cool stuff it's done. So, this is the first one, and man, this is completely insane. All right, so you can see how difficult the prompt is, but hopefully you recognize it, right? It's Manhattan, right? With so much detail. I don't know Manhattan that well. Like I don't know how accurate it is, but like superficially to me, apart from the buildings being underwater, this looks to me like pretty accurate, right? So, we've got like Central Park here in so much detail. We've got kind of the taller buildings, the like you can see like this is like completely insane.

2:28 moon dusk. Yeah. Like I've not seen any model come anywhere near close doing something like that with so much complexity. And like we can take a look at the code here. Is it just No, that can't be right. Yeah. So, oh yeah. So, yeah. Actually, could it be Yeah, 16 100 lines of code. Anyway, so, yeah, very very impressive. You might have seen that one before. I posted this before, but this is a voxel generation of Rome, which is like by far the best. So, in terms of the detail, in terms of the kind of the coherence of different elements, like completely insane.

3:17 if you look at some others, so you can see here I did have a follow-up prompt here. I asked it to just make the flow a little bit nicer, but what we are seeing here is that Yeah, let's see. So, what we are seeing here is the seven wonders of the world. So, lighthouse of the Alexandria, and it goes through the different steps. You kind of you are sailing past them, and it shows what what they are.

3:50 and difficulty here that how much of the world that it needs to create. All right, this is like the complexity of the space and how much it would be able to like dedicate to each one. And then we've got actually like complexity of the light and yeah, the Hanging Gardens of Babylon. And each one like wouldn't be to be honest with us for each individual one to be done in a high level of detail. It would be even more impressive. But here the this kind of combination of different ones, that's what I'm impressed by here.

4:28 If we keep looking, so there were a couple of iterations here. Let's have a look at this one. Oh yeah, so that one was quite cool. So, let's play this and I'll tell you a bit more about it. So, this one is a view of London over a couple of thousand years. So, London is a old city with a lot of history. so, there's Great Fire of London here. So, it started off in Londinium, which is like before the current era, like 2,000 years ago, when it was a Roman town. And then it goes through the town the times.

5:06 And it's showing me the kind of the different key steps, like there was Great Fire of London, there's the Blitz during the Second World War, and now this is the modern times. And again, the if if I was just to ask it to generate the modern London, it will do a better job than this. Like we can't like I mean, I said I'm not going to do this, but you know, it's not like perfect. But the fact that it had to had to kind of create all of those scenes and kind of how they evolve, that's like really, really impressive. And to be like like it is pointing to historical accuracies and the shape of London and all of this.

5:46 This is really cool. Like this is like this is very, very impressive generation. And yeah, you can see how the whole scene is changing and so on. This is just so cool. let's keep looking. Oh yeah, so this is Paris. The This was a one-shot generation, so I didn't ask for any iterations. I think the the There's more opportunity to make it better, but this is quite cool. Like nice nice sweet generation. let's not spend too much time on it, but the Golden Gate one, I was actually trying to do it last time, but it never quite like worked out for me. It was like timing out or breaking or something like that. So now I put the effort to make sure that we do get the Golden Gate. I think I've got like two or three of them.

6:34 the the reason why I do want it is that we have quite a lot of comparison examples. I'm not going to show them here just for brevity, but it kind of gives me an intuition of where it is. And in terms of the quality of the actual bridge, like the there's like water here, there's like the ships going, like traffic going. I want to have like variety of traffic. Like all of this is completely like top top notch. Right, the the shape of the bridge is not the kind of thing that you actually see models like generate well. Oh yeah, the comment here like this the reflections you see. This is like really, really nice. Oh yeah, you can even drive. I Did they I can't remember if this was in the prompt. I don't think other models like did that.

7:22 So like this is really, really excellent. There's still some weirdness. Like why is there no road here? Why there's like it kind of goes through the mountain weirdly. So like you can nitpick, but like this is re- really, really excellent. I think I was trying to really make sure that I do get one at least. So I had like a three of them, I So, you can see a little bit different. I think the first one was better. I can't remember if I did any follow-ups.

7:50 No, that was actually one shot. Oh, no, not quite. Oh, yeah, yeah, you see I did That was a one thing that I did ask it to do like a few times is to So, this was V1. I'm not sure like V2 is that much better. Like maybe a little bit. but yeah, sometimes I was asking it to do try and be more ambitious and that was definitely a theme. I would say if I was to nitpick more kind of a failure mode that I was experiencing is that I was feeling that the model can do so much more than it was generating. And sometimes it really felt like it was kind of holding back almost and like not doing a great job.

8:31 so yeah, I did kind of kick it a few times. So, maybe like if you're not quite getting as good generations as this, maybe try and encourage it to like be more ambitious. And this was one shot, I think. Yeah. And this is actually probably the best one out of those three because look at the road. This is the first one where it was actually the road is actually going through like it is more plausibly. and in here, yeah, the water, look at that. I mean, this is crazy.

9:04 Like this is so good. yeah, excellent excellent generation. Like I I think this is like pretty clearly especially for one shot. Like this is pretty clearly the top top generation. I did get to some really good ones with like I think GPT-5 I can't quite remember the vision pro model, but it took like 10 iterations to get to something good, like really good. But here, this is one shot. I mean, I can't get enough of this water. This is crazy.

9:36 Yeah, and let's see the the comet time of day. So, if I go to the night, I've got the comet button here. And not the absolute best one, but yeah, pretty cool. Anyway, let's move on from the bridges. We've got a lot more to look at. so, this one is a historic Istanbul. which is a capital city of Turkey, if you're not sure. previously Constantinople, and rich history, and it kind of straddles the the Europe here on on my left, and then Asia on my right. So, a little education for you. Hopefully, you knew all of that anyway.

10:26 so, yeah, like brilliant brilliant generation. Like, look at this water. Look at the reflection. And you can nitpick. I mean, I'm pretty sure it's not floating like midair like this. but this is just completely insane. So, you've got the the Blue Mosque, the Hagia I'm going to get the name right. Sophia. Hagia Sophia here as well. yeah, this is just just brilliant in terms of like I'm I'm sure like people who from Turkey would look at this and say, "Oh, this is all wrong." But like in terms of the outlines, this is looking like completely exceptional.

11:08 Like, look at this. How many elements of complexity needs to get right to for this to come together is pretty insane. And that's why I'm excited about these tests is not that you particularly care. Like, not many of us actually care about the 3D generations, but how smart does the model need to be that I can give it a prompt that I'm pretty sure it's not like trained on generating, Istanbul. But, the fact that they can come and and like arrange the city accurately, it needs to know that. It needs to know how to render the water, how to coordinate different elements. This is impressive.

11:48 So, that that's why like I think that's quite a nice test. that things like dynamism, right? Where we can see the different elements moving. All right, how does it create like this kind of glistening icy here? Like this is pretty pretty cool. All right, this kind of diversity, the fact that it doesn't just like mode collapse into like one narrow space. So, hopefully that gives you a feel for like what's the range of the of this model that that it can do.

12:21 so, here the the Winter Palace in Beijing. >> >> I think it's like maybe slightly missing some of those, but actually maybe not. Yeah, no, maybe to me this looks like amazing. And yeah, time of day maybe Oh, yeah. Yeah. I think this was one shot. So, yeah, maybe I would have given it a few pointers, but like the quality of this is just exquisite. I I I hope you agree.

12:52 So, I I know not spending a lot of time on these, but just I have so many I want to show you guys so many. So, this is Phi Phi Phi Phi Islands in Thailand. And this is just this kind of magical place. so, you can see the the lagoon and the boats and this kind of this kind of really beautiful clear water. Like all of this kind of all together. Yeah, just the water. I'm blown away by how well it does the water.

13:24 And the the little boats moving through they're all also very coherent, very coordinated. Like that's what I'm impressed by by good models is when they're properly coordinated the different elements are properly coordinated with each other. What we see sometimes with not so good models is that they kind of they kind of try something and then other things like don't fit at all and they kind of break away. I think it's like a reasonable proxy for us to be able to kind of see like how good a model's like it doing more complicated tasks, right? If it's like does 80% of it correctly and the last 20% is all over the place, even if it's like cheaper faster model, I do do want to then be debugging the last 20%. And I think it's not like I'm not actually suggesting that the right answer is always to go for the fanciest model, like maybe not. but it's just something that you need to be calibrated on that if you are doing these kind of complicated tasks, it could be taking you so much more work to untangle like the last details. And I've certainly had a lot of experience of like having to do that myself when it's like not quite getting this right.

14:44 You are getting to the 95% in one shot and then the last 5% you are like tinkering for 2 hours. Like I I have a lot of personal experience and anxieties of that. So, I'm just showing you like a few few cool things just while I'm talking. like I don't want to like call out too many specific things just not to overload you. but the general pattern is like super clear, right? These are I don't say cherry-pick. Like I've selected probably I I probably did like call it I don't know, 90, maybe like 70, 80 prompts.

15:29 but these are sort of about 65 that I'm showing you. Like oh yeah, 63 prompts. So this not like cherry-picked exactly, but this is like selected down a little bit. So there were like a few that I haven't haven't shown you guys, but they're mostly like they were kind of weird things such as like the axis wasn't turning and and so on. But these are not like heavily cherry-picked. It's not like I've generated 500 and I'm showing you just like this small sliver. This is more like I've generated maybe 20% more and I've selected down a few that was like some weird bugs or something that I just like didn't have time to to clear out. So this is like almost completely unfiltered.

16:21 and sometimes it's completely one shot, but sometimes I would I would get it to do maybe one or two iterations. And when I was doing iterations, couple of times it was maybe some kind of slight visual bugs. Like I maybe I could have gotten it to fix this style here just kind of going through that. or sometimes it was just like I thought a little bit lazy, so it could have pushed it itself a bit harder.

16:46 so let me just keep showing you some more cool things. so Grand Budapest Hotel, again pretty impressive stuff. this is what I was getting it to do is to just like tidy up a little bit the the kind of coordination of different elements. But what it is doing here, the idea here is that it's showing the ancient Egypt build out. And it's like how the different elements appearing, how the pyramid is being built, and this is just so so cool that models like can do that at all.

17:32 Like look at this, the little people moving. the workers are moving. the pyramids are being built out. Like this is just like completely awesome, the fact that they can do that at all. And I appreciate, you know, yeah, there are definitely some issues you can like debate. But the fact that this like at all works like I think is really really worth the recognizing that and I had no idea this was at all possible. So I'm I'm very impressed.

18:11 more pyramids already built. Yeah, this is very beautiful. I would say that kind of stuff it's already kind of almost tapped out the quality cuz I I did get like previously before when I was using this specific prompt, it was quite similar like I think Opus 4.8 was kind of generating that kind of quality already. So that's why I'm moving on to like harder prompts that that I want to show. This one is meant to be showing like the Roman Empire and how it gets built out.

18:46 Actually, let me show it. This was V1. Let me show you V2. I think it would be in this just a small more complicated one. Yeah, so I got it to tidy things up a little bit. I wasn't like quite perfectly happy with it, and I think it's just my standards keep going up, you know, moving goal posts in a crazy way. But it's meant to be kind of showing the development of the Roman Empire. And I thought this is like slightly like a bit too cartoonish almost like the elements like don't quite fit nicely, but I don't know. Maybe I need to think about my prompt for this one. Like what what do I actually have in mind?

19:28 but yeah, and you know, they this kind of complexity and even like to build a map. I bet you give this to a weaker model, it wouldn't get to this map even. But the fact that it can tell the history of the Roman Empire and how it developed, like this is pretty pretty insane. and I think what Fable is particularly good at is actually this kind of educational content about like showing how how things were and kind of explaining different elements. I think it has kind of improved on the on the kind of theory of mind side of things like quite a bit, which is definitely an issue that a lot of the models had.

20:19 So, this kind of imagining what the user is thinking and and trying to address that. I think that's like a really nice improvement. So, it's much better at explanations. It's much better at this kind of scientific data presentation telling me like what how to like look at this data set, that kind of thing. That's really good. >> >> So, this one is Pompeii. And again, this is kind of telling the history of Pompeii a little bit and it's kind of you can see the time I'm going through and it is showing how the kind of the the Oh, look at that. There's a kind of the ash or and the people escaping.

21:02 I know it sounds excited about I mean, it's been a while. Yeah, so it's telling the whole story of it of Pompeii, the eruption. and how like the whole world has changed. So, the the fact that it can tell this in such a immersive way and coordinating different elements, like it's crazy, right? Like this is so so good.

21:36 Yeah, I had no idea on this. Had no idea this is even possible to be honest. Yeah, we've kind of gone through the cycle here. Right, let's keep looking. I hope you guys not getting bored of this. Like I love this stuff. Like I don't even Yeah, Minoan festival. so again, the ancient ancient culture. I think it was in Greek islands kind of area, I think. I think it's in Crete. where they were kind of showing me this tradition, traditional ceremony. I I don't know enough about this to judge how how realistic that is, but looks pretty cool.

22:23 So, oh yeah, this is another ancient festival here. And >> >> All right. Oh yeah, it's showing the whole procession here. And like I think for educational content, that's something I didn't predict at all. The fact that it could be so so good for this kind of like take something that I don't know enough about and just explain to me in this kind of visual way what it looks like. I don't know if any of you are in in education and you are maybe teaching some particular topic.

23:02 Like maybe try this. I I don't know. Maybe that could be like quite a cool thing or get your students to do this. Maybe it's quite a cool way to learn about any like particular events. and these are kind of slightly like semi-random ones which I vaguely remember like from school. but yeah, maybe any like specific topics. Pretty sure I've seen some of this in the British Museum.

23:32 at least like the the mock-ups. But yeah, that that could be interesting. Petra here. yeah, a lot of a lot of you can see like the the quality of generation. So yeah, worth worth exploring. yeah, this one I like the Saharan Caravan as a kind of test of ability to coordinate different elements. And this is the this is really really nice. I mean, I wish they were like a little bit more grounded.

24:07 So there's still a little bit room of for improvement. But the fact that each camel is like nicely nicely designed. You can see it like this is each individual one. This is pretty cool. Cappadocia here. We had like a bunch of these generations. So what I find normally is that it's a lot of the models kind of do it in a very kind of almost like schematic way. But here it actually created these kind of valleys that are realistic like the houses. The yeah, it looks like much much higher level of realism that it attains to. Not just kind of plopping a few things and balloons which is like vast majority of the other generations as well.

24:58 And you can see like pretty much whatever I pick just looks kind of incredible and well set out. Think yeah, Stone Canyon I think much more on the realistic side and some of the others that that I've seen. So that's is definitely like an improvement forward. So oh yeah, what was that? Yeah, I think it's an another festival. Yeah, maybe I mean looks cool but let's let's keep looking.

25:31 Yeah, so another ancient one and I think what I was trying to get at here is this kind of Yeah, I think you can see the fidelity here is not quite as high as some of the others. So there's definitely like a downside and I think if I if I had asked it to maybe put more effort into that, maybe it could have done that. But I think what we're trying to gauge here is this kind of overall ability to to create the the whole world. So Yeah, I like yeah, the stone forest kind of floating through that. Not sure how realistic this is. I don't think so but it's a kind of a nice Yeah, nice kind of also world that is created and this ability to kind of go through that.

26:25 All right, I I'm going to keep going and see how many of you watch until the end. I've got a few more tabs. So this is a game. I know people had like incredible experiences building out games. I haven't quite got it to this point and this was a couple of iterations if I remember correctly. I was getting it to like improve the gameplay a bit more. So I guess it's like it's Look, I mean, it is cool generation, but I wouldn't say to me there's like a super impressive game. So, the one that I like better is actually this one. Again, took a couple of generations, but I wanted to like have a game. Ooh, come on. Oh, no, it missed it. have a Oh, yeah, I can lower the hook.

27:17 Yeah, well, I can destroy things. Look at that. Oh, no. Let's hide this. this is Look at everything shaking. That's kind of cool. Like this is not I wouldn't call this like game game. but look at that. It's definitely like much more within the direction of games, right? Where you want to have like things moving around and the whole world kind of being created.

27:52 But yeah, maybe I can work on my game in prompts a little bit more. I think like I feel like some other examples that I've seen are a little bit more impressive than this. but that's kind of cool. Anyway, another another cool thing that turns out you can do, which I didn't know. the flight simulator. I think I was kind of hoping it will look a little like a little bit better. Think actually there was a V2, no? Yeah.

28:26 Yeah, the cockpit was kind of like looks a little bit weird. And I I didn't quite get to the point that that I wanted. but it's still it's still like to to get to this point to even have such a render of the world, this is pretty pretty impressive. Oh, yeah. This one is like I think it's not like completely perfect what I imagined, but I think this is far more complicated than I imagine that it has to do. So, this is again I believe this should be 3GS as well, right? Yeah, so and it's generating this kind of 3D world, but it's putting it in this kind of a 2D view, which I think is like kind of maybe unfair thing to ask. But the fact that it like creates this slice of the city, you've got the underground you've got the station, you've got like the different pipes, you've got the street level.

29:24 The each individual like Wow, look at the level of detail. Like you can probably I don't know maybe there's a different way to do this. Maybe my prompt was like a little bit too like specific about like all of these elements, but the fact that it did something like this like again, I had no idea you could do this. this looks like very very impressive to me. I'll show you a few more. There's like London stuff. I have tried this on other models and again the coherence of the different elements how they're placed and that they're not overlapping. This is like so much better than so many models. A lot of models were kind of maybe things that like I don't know the London Eye would be in the middle of the river like for example like something like that. And the fact that here and it's not to scale but like the fact that different elements are placed like pretty much as good as you can imagine like in terms of in relation to each other they're all placed accurately.

30:24 This is like super cool. like super super cool. yeah, you can see more more London. You can see I've got the London prompt collection. a few more as well. Yeah, anyway, it's so anything I show you is just looks like insane. so this is kind of a little game and this was a version two. This was actually I think one iteration. And I was getting into just up the fidelity and the realism and the quality. Maybe let me show you. I think there was a V1 in here. You can see that's a V1 but it's like if you look at the house, if you look at the river, it looks like at the canal, it kind of looks a little bit basic versus here like look at the quality of the water rendering here. So that was one iteration for it to do that.

31:23 But it was like gameplay. I I don't know. It's like it it's pretty basic. I don't know. Maybe I need to learn more about how to make games and what to prompt for but it's not like great game but the fact that actually controls are like probably probably the best that I've seen models do. Controller is one thing models just not good at. so oh yeah, finish. Look at that. Perfect. Let's see Oh yeah, did this work?

31:57 Oh yeah. This was yeah, another like couple of iterations. Again, probably I know I said at the beginning I'm not going to like criticize the models too much but this is kind of cool but I'm still not quite getting to the point that I've seen some other generations about like the the games. So maybe I'll I'll keep trying some some better ones. I think Yeah, they were like kind of fun but like I don't know. I I don't feel like I want to play them.

32:25 Like feels like 30 seconds and I and I'm bored. but the controls are like really really high quality. and by the way, this is just cuz I opened too many tabs. It's not like there's some problem with the model. that's why I just my my Chrome is not opening things properly. Oh, yeah, this one. This one I think actually I got Fable to generate few prompts for me.

32:56 So, this one and I was trying to get it to go a little bit wild and it created here what would it be like being being the firework? Like inside the firework. So, it needs like a little bit of a focusing to understand what's going on here. So, you can see there's a city below and you're kind of being blown to smithereens here as the firework. Which is like I guess it's like an interesting idea. So, that's what we were up there. So, yeah.

33:31 Fable also creating some fun prompts for itself. few others. I'm going to I'm going to try and get through all of these. I know maybe it's like a little bit too much, but bear bear with me. This is the creation of Michelangelo. Kind of just trying to push it like as much as I can. Like just do anything possible. yeah, it kind of goes through the motions and yeah, I I just can't get enough of like how cool this stuff is. I mean, not quite Michelangelo rendered there there end, but like this stuff is cool.

34:12 we're going into this art section now. So, this is Klimt's like famous the painting The Kiss and I was trying to get it to like create some kind of 3D world where you can actually go into it and you can like flow around it. And I don't know what I was expecting, but like this looks kind of cool. All right, you see you can flow around, and you can like see the flowers. Like, yeah.

34:42 And like that's completely unreasonable thing for me to like expect the models to do. but I also want to see like how they going to do it. Like, look at this. So, this is Starry Night by Van Gogh. And in here, if I asked it to like create the world of it, right? But like how would you do that in 3js? Because or like any 3D generation approach. Because the way like if you look at the painting, it's like this kind of smudges of paint.

35:15 Like, so how would you even go about creating this? And the fact that it like did this kind of weird like lines, like individual lines that come together into the painting, like how cool is that? Like, this is just completely mind-blowing. All right, this is cool cool stuff. All right, I think I'm lost. Anyway, like this is just super cool. Oh.

35:45 All right, maybe I'm going to play with this later. >> >> Oh, no. Okay, I Just this world is massive. Look at like the stars here with the Starry Night. I mean, I know it's weird, but I like it. let's let's keep looking. What else do we have? Oh, yeah. Another one. So, this is Monet's lilies here. And this is a real place, but there like numerous paintings.

36:16 And in here again, this kind of technique that it used here to like create this kind of impressionist style 3GS generations. And I don't know, like maybe tell me if I'm being overly impressed by this, but I think this is insane. Like how could you even like I don't know. Is this like in the training data? Am I missing something? I've never seen anything like this. So, I don't know.

36:49 If it's not in the training data, then it's like real levels of of creativity there by by Fable. And that's another one. So, this is like I hope you guys recognize this. This is like the wave by Japanese artist. Actually, I I I don't know if I know the name. yeah, and there's like Mount Fuji in the distance. And I haven't checked like this is 3GS, right? And I'm I'm not 3GS expert, but I didn't know you can do that kind of thing. Like this is really really cool. So, it kind of created this kind of almost like paper-like feel of of the generations.

37:29 And I think that idea that you could even do that and like how would you would do that, this kind of very papery feel which creates the this whole experience of the of the wave. I even haven't tried that. Yeah, you can see it move as well. like this is Yeah, I'm I'm impressed. This is cool stuff. more stuff.

38:00 the Tower of Babylon. Very cool. Love it. yeah, there's probably more stuff I can I can show you inside. Let's see. Oh yeah, that one was interesting. So, this is Pollock. So, Jason Pollock Jackson Pollock rather. If you know his art, he does this kind of, I don't know, super post-impressionist this kind of paintings where the paint is flowing everywhere.

38:33 And what we want to do here is to like again, get the model. It kind of looks like that, right? This is accurate. We want to get the model to create this kind of world where you can go and and float around it. And then it created this kind of 3D view of Jackson Pollock's painting. And I don't know if it resembles any specific painting. I'm not such a big expert in on his art, but so this might not be exactly accurate, but like the fact that it even thought to do that kind of thing, like that is cool.

39:11 Like this is really, really cool. Right. Let's Let me not get stuck on this. more cool stuff. so this is I think this is another prompt that I got Fable to like just like go wild and come up with some stuff. And this is like a raindrop experience of a rainstorm in a garden on 1 mm tall kind of ant height.

39:42 and what I like about this is that it's kind of making the objects in the distance more blurry. I see it's understanding like what what it feel like to be in that kind of environment. And I think this kind of this kind of experience that it can create, I can imagine the different elements, like what would things look like. And this is this is nuts. Like this is really, really cool stuff. And I hope you guys like hope you're not thinking about this as like, oh, you know, a bunch of silly stuff, but it's impressive.

40:18 I think it just shows you that you can I imagine you've got lots of different other tasks, but what you should do is really to think about what else can you do that is maybe different to what you would have expected the models to do before. Right? I think this is the time when we've got such a overhang of the capabilities that there is no way we would have like tried to do this 6 months ago.

40:53 Right? So, if you kind of built up your routines or what you know about this 6 months ago, then this is like completely different world. >> >> this is the crossing of the Red Sea. Jesus. and you know, you can go and you can have a look around and fly around. insane, right? Yeah, and you can go across. yeah. Crazy crazy world crazy times we live in.

41:27 Yeah, go around. Oh, yeah, I think that was V1 and I gave it some feedback that was like a little bit too like floaty and so on. So, oh yeah, the viewing platform was a bit weird. So, I got it to do like more stuff. So, yeah. Look at Look at this. Like I think I was just throwing more stuff at it and just see like what else it it can do. And like it just kept going. Like I I actually like the ones that you kind of failed at that I have not included. It was quite often not because it was like, oh, it's too hard. It it like couldn't do it or something. It's more that it was like some generations were excellent, but I just like cuz I've got so many that I didn't want to like tinker with each one for too long.

42:17 >> >> and I was I was basically just generating really really high quality outputs almost like no matter what I asked. So, these are like the hardest problems I could possibly think of. Like, okay, maybe I'm not asking to like create GTA 6. Like, I'm sure it it wouldn't work but like any kind of reasonable things I can expect it to do as a like HTML file, like I can't think of anything particularly hard that I can ask that what I'm already asking. So, I think it's like on a lot of these things it like tapping out already. Like, look at the bears fishing on salmon.

43:04 is it actually going to eat something? I mean, it's pretty fat. Oh my gosh, look at that. It got the salmon. >> >> I'm going off on a walk to eat this. I am Do they do that? Do they just eat it? Oh, look, there's another one. And how how crazy is that? Like, this is just You know, it just makes me happy for a little bit. I know we're not going to like all of our subscriptions and so on going to go to hell and then it's going to be super expensive to do it. But, you know, use AG mode, then you might get it on the on the arena. And yeah, you might get this kind of stuff. Like, this is this is cool stuff.

43:48 So, in conclusion, I'm going to finish showing you like a few, but I want to conclude just by saying like, look, like AI industry is developing very fast, right? These are the kinds of things that like it's very hard to keep up with. So, I want to show you what are the kinds of things you can do. And my goal with this is to help you maybe think about like just broaden your mindset a little bit about the kind of things that are possible. Like these things are were not possible like even a month ago before Fable, right? You couldn't do this.

44:24 Like you maybe could do this with a lot of iteration. So the fact that you can do this now is like just the coolest thing. And I'm sure you don't care about 3D generations like yourself, but you do have something that you would care about and whether it's like I know your little app that you're building on the side or your you've got I don't know a small business you're running or you're just a hobbyist developer or whatever it is or you're a mathematician. Like there is probably something that you haven't thought about so far that previous models couldn't do that now you could do. So this is like a kind of ink yeah like Japanese ink world where it's like painting the brushes and you can see it look at the ink in a lot of detail. Like how like who knows? No one is there, but it's like it's so cool.

45:24 And I really encourage you to try and push the models and and try and make sure that you really put yourself in a position where you can benefit from these models. Don't get stuck on what the model could have been doing 6 months ago, right? Don't worry about, you know, it's like cost too much money. That's a bummer. user arena trying get Fable on the agent mode.

45:54 but they're like really miss blah blah blah. But this is something that's going to get cheaper. It's going to get more accessible. And hopefully, you guys will get to a point that you if you do try these things, you can get ahead, you know, of everyone. You can come up with new things to do. And the coolest thing about this is that because it's so new, even people who work in these labs, they don't understand yet everything that these models can do. So, I would really encourage you to go and like try try things out. Try and build This is, by the way, the space elevator that that we're looking at.

46:37 like build what's the space elevator in your world, right? Try and try and do that. No one's tried this before. whether you're using Fable or something else, these models will keep getting better. As you can see, I'm quite excited about this. It's quite a lighthearted episode. So, hope you learn something new and I'll see you in the next one.