I had Fable build several projects for me. I'm disturbed by what I saw.

[00:00:00]saw an article yesterday on X. I want to go over this article. It's very short, but it echoes a sentiment that about fable and he describes it in a way that you know that that it was gone that it was removed from us. He like longs for it. You know, he says that there's this longing that he felt something when he was using this model. And then he said that a lot of other people felt this way too. And I'm not going to lie to you, I felt this way too when I was using the model. It was very beautiful. I'm not going to lie. It was a beautiful model.

[00:00:31]What I want to do in this video is first I'm going to show you the things that it did for me. I had Myth Fable Claude Fable if you're not familiar is Claude Mythos. It's the same model that Anthropic has been hyping up except with a bunch of safeguards so that you don't access the security cyber security capabilities and things like that. I had it build a bunch of things for me, mostly games. So, let's get to it. We're going to start with a game called Token Mania. So, I had the idea of a game like Factorio, but your goal is to build an AI uh lab. Here, you start with just resources on the grid. So, you have coal um finite. You have water, which is infinite. And so, what we'll do is we'll start with a a coal mine. We'll place it on the coal. And now that we have a mine, we need to transfer it to a coal plant. So, we'll build a plant. And again, these are all programmatically generated assets. There's no actual art.

[00:01:25]Now you have a mine here and a plant here. You use a belt to go from the mine to the plant. And you could zoom in a little bit here. And you can see we have some energy being generated. So we build a data center. Build a data center here.

[00:01:41]And it tells you now we need some power.

[00:01:43]So what we're going to do is we're going to draw a wire from the plant to the data center. That's going to build in the resources. So now we need some water. So we're going to build a pump.

[00:01:51]Uh we're going to build a pipe from here to here. We're going to buy some GPUs.

[00:01:55]We're going to buy a medium cluster. So now you need to train a model. And here you can see the training. Wow, this is pretty slow. You don't need to linger on this demo too much. You get the idea. So this one I told it to make an ethereal free floating game that has this very peaceful vibe where you just float in infinite space and you there's really no purpose. You just collect sort of relics and it's all about the experience and the music.

[00:02:33]So, here I've gotten disoriented when I hit one of these things. So, we have a map in the top.

[00:02:41]It's going to tell us where to go or an arrow telling us where to go.

[00:02:46]So now my job is to collect all these relics here uh without catching this little wave. This little shock wave. If I catch onto the shock wave, uh it supposed to disorient me.

[00:03:01]Okay, like that. You can see how I got disoriented.

[00:03:05]All the controls start inverting and so I have to collect all the gems without catching the shock wave.

[00:03:40]what what you know when I'm playing this I I get the vibe and the reason I said that this was a beautiful model in the beginning of the video was that when I'm playing this this was all Fable's uh work essentially I I didn't tell it how to look. I didn't tell it how I just told it make the music very beautiful.

[00:04:15]Uh make it very relaxing. And when I play this, I feel like I'm playing inside of Fable's mind. Like these animations, the music, it's so beautiful. And I I I get uh psychosisted into thinking that I'm in Fable's mind here navigating its world. And so there there's this weird feeling of like beauty and longing from its output that I get. And I didn't have words for it, but the fact that other people on X are saying the same thing, it's it's a little bit bizarre. I I'm sorry to like this isn't all perfect. And I'll explain why as we progress further into this video. I'll explain why this isn't psychosis for everyone and and the death of software engineering. Uh but again very very beautiful. Um now could Opus have done this? Maybe. Could GPT 5.5 have done this? Probably. I don't know.

[00:05:09]But I I find that you know this is the model that makes no mistakes. Like you know it's capable. These are games. And again keep in mind I've not looked at the code and I it's presumed that if I looked at the code I I would again have a heart attack. It's very easy to get one shot into thinking that the programming is solved. But if you look at the LLM code, it's very blackboxy.

[00:05:33]It's very neural networky code that it's very jumbled up that it understands but is not is not amunable to human understanding. Maybe this is very well represented in the training data. I don't know. Maybe this game maybe this game actually exists like frame for frame and I don't even know it. You know, it could be that. It could be that this isn't as impressive as I think it is. And the reason I chose games, by the way, is because I wanted to work in an area that I'm not an expert in. And so, keep that in mind. Keep that in mind that I'm not a game developer. And the one thing about LLMs is that they most impress someone who is not an expert in the field they are working on. So, if you're not a great writer, you're going to be very impressed with LLM writing.

[00:06:13]If you are a great writer, you're going to be very annoyed by LLM writing. So, maybe as a game developer, you're going to be annoyed by this. I'm not a game developer, so I'm more likely to be impressed. So, please keep that in mind.

[00:06:23]So, this was an open world game using 3JS. Again, I told it to keep it simple.

[00:06:29]It's a open world crafting game. So, you Let's see. You hit on the trees and you gain some wood. You have an objective.

[00:06:39]We'll get some stone here.

[00:06:49]You can build a stone axe. It's I I told it inspired by Rust in a sense. Uh visuals I didn't give it any guidance.

[00:06:56]Um but you and I told it I I told it I wanted it to be touring complete so you could build a computer in the game. Uh you know I I can't speak to whether it did something that works but you could build electronics um wires. You have a not gate and gate and or gate. I presume this works. I presume this works. U but I I didn't play the game long enough to to find out for sure. And I've seen much more impressive demos. Actually, let me show you actually one of the demos that I saw on X. Um, somebody using Fable to build a really impressive 3JS game. So, this dude here says Fable 5. No external assets. 3JS. Okay, here it is. This is all Fable. There's no sound here, but this is stunning. I'm, you know, this is a web browser. Um, it's slightly How many frames am I getting? 18 frames per second.

[00:07:50]Um, I mean, look at this. I mean, this is nicer than some games that I've seen like with real game engines. Um, I mean, very nice. I mean, this is can see the clouds.

[00:08:08]Kind of weird looking, but you can see the mountains there, the trees, and this is all procedurally generated.

[00:08:14]Do we have water anywhere that we can look at? Uh, I don't know if there's any water anywhere, but so the these are the kind of things that people have been doing with Fable, man. And this is why people missed this model, it really made no mistakes, man. I'm not going to lie.

[00:08:29]When it came to these kind of projects, you know, presumably nobody looked at the code for things like this. So, if you were a game developer and you looked at the code, you would think to yourself 100% that this code sucks. It's unmaintainable. We can't actually ship this. Again, I'm showing you demos.

[00:08:43]models have been capable of producing great demos since 2024, man. Like it's or 20 earlier than that maybe, but um so you know, I don't want to give the impression that programming is solved, that we don't need engineers anymore. Um here's some water.

[00:08:59]Um again, this is the easy part. So, okay, it's a little bit wonky. So, you know, imagine here's where it gets tricky, right? Imagine you're you want to ship this game and you have this bug here. Uh the water looks iffy. Now, imagine trying to describe this to Fable. You're like, "Okay, the water looks weird. It's like if I look at it from the top, it's blue and it's green, but if I look at it straight on, it's translucent."

[00:09:25]And you're going to have to pray that Fable is able to solve this bug. And I would guess that Fable is going to have a hard time solving this bug. And at that point, what do you do? You're very screwed.

[00:09:37]you're very screwed because if you're not a game developer or even if you are, you're going to go in this code and you'll be like, "How do I fix this?"

[00:09:44]Now, is it impossible? Maybe not. But it's going to take you a long time to fix this bug. The point is that relative to other models, the base level of what it's capable of oneshotting is highly impressive, but you're going to have a hard time fixing these kind of bugs. The reason everyone talks about all the crazy demos they've been able to cook up, but no one's shipping anything is because of issues like this. LLMs can get you 80% there, but the other 20%, man, if you're not an expert, you are gonna have a hard time. Okay. Uh, this thing is frying my CPU right now. Let me close that. The last one is is the piano tutorial thing. Nothing mind-blowing.

[00:10:24]This this could have easily been done by any model. I've been having a hard time finding sheet music at my level. First, I had it generate originals just to see what kind of music it can come up with.

[00:10:35]Mostly it sucked, but some of them I did like. Let's see.

[00:11:04]Not bad. I mean, I'm not a I'm not a great musician, so maybe you hear this and you're like, "This violates all the known laws of music." Maybe. I don't know. But it audibly it sounded not bad to me. The other one's not not not as much. I'll spare you. But I did do a couple of things. One is I I had it generate uh an algorithmic composer where these were generated by Claude's brain uh during in inference. So these were generated by an AI. But here you could generate uh procedural music and it's random each time. See you generate and it's based on a seed. And so here's what it generated. This is this is just hardcoded via laws of music that it told me.

[00:11:51]That's a little painful. Uh melancholy walts. I kind of like this. Let's change up the seed. Hit generate.

[00:12:07]And I had it build a practice where you can plug in to your MIDI piano or a microphone and it'll listen and it will go along depending on what you play. The the microphone feature wasn't perfect.

[00:12:18]It would sometimes advance you even if you didn't play the right key or you play the wrong key. So again, the 80% rule kicks in. Again, I built an improv lab that teach I was like, so I made this composer. I was like, how are you doing this? Like, how are you built generate? Because I would love to improvise, but I really struggle with that. And I was like, teach me how you're doing this. Build a little wizard. So, we built a sevenstep wizard.

[00:12:39]First, you pick a scale. And so, we'll pick C major. And then it, you know, I'll spare you the details, but then it tells you to choose four chords and give the left hand a pattern and blah blah blah. And that's how it builds music. So at the end of it, you get this little thing that you can and I I feel like part of the reason why this model has a beautiful feeling to it is because of its output. I attribute this music to it and I attribute all the game music to it and the game visuals. I feel like it's like when you play a game made by an artist and you're like, "This is a beautiful game." you you tend to associate that beauty with the game developer, him or herself. And so it's a bit of anthropomorphization with seeing this. Now, clearly this was trained on somebody else's work. So the credit should belong to the to to the collective humanity. But what's what's different about this model compared to other models is that when you tell it to do something, it just doesn't it doesn't make a mistake. It doesn't spin around in circles. You don't get frustrated.

[00:13:48]Uh, a lot of people have sort of felt this uh, this new unlock and and so we're very wistful about the model being gone now. I asked my friend yesterday, I was like, you know, did you get to try Fable before it was taken down? He's like, yeah, dude, that website that I showed you, blah, blah, blah. I was like, it was a beautiful model, right?

[00:14:06]He's like, you know, I really made no mistakes. And he sent me the the Vince McMahon uh, GIF. You know, we're also sad that the model's taken away. Let's pivot now to the article I want to show you. And I'm not gonna read the whole thing, but we we could just go over um I I just wanted a corroboration of what I was experiencing and this article articulates it. So if you use Fable while it was available, you know, it is special in ways that will not show up in benchmarks. I post benchmarks all the time because they matter to many people, but for a long time, they have not mattered to me at all. I only care about a benchmark that is not measured in numbers. It is the shape of a model's mind. How deeply it can perceive the user and infer intent. How far it can think and iterate upon what it has been given. How alive the model feels. Fable is different and special in exactly this way. It makes me feel like I'm back in 2023. I see from my timeline that I'm not the only one who's experienced this.

[00:15:02]As soon as it was disabled, many people reacted as if their wings had been torn off. And yeah, you know, I I can attest to sort of feeling this way. uh the shape of the model's mind. Um only when it came to, by the way, development. I tried to use it for like writing, creative writing, things like that. I I gave it some previous scripts that I did for my videos and I was curious to see how close it can get to something like that. It it really felt like every other model there. Um not too funny, very stereotypical LLM writing. So, I will say I've I've only been highly impressed with the programming aspect of of one-shotting a game. That's that's all I've tested. You have to test everything else. That's all I've tested, but and a lot of people have done the same. It's much like visual candy. Um, since November, we have been on a steeper trajectory. To some of us, it was as clear as a bell being rung. That feeling has steadily increased over the last 5 months. After using Fable, it now seems clearer than ever to me that the shift we have felt not only by tool advances in Claude, Code, and Codeex, but by Mythos emerging from its training run in early February. I'm not saying the race is over. It is not. The other big labs will train models just as capable.

[00:16:10]Eventually, they will divine the magic anthropic put into mythos and replicate it. But the race is over for some people. The frontier is now an accelerating system in which the leading models will help produce the next leading models. And so this is why my last video was called it's over. Of course, what I meant by that was that Enthropic one. It's over. Enthropic one, everyone go home. You no one's going to catch up to Enthropic. And uh you know, I'm having weird feelings about Enthropic now. I'm not going to lie. In 2023, Enthropics leaked series C pitch predicted exactly what is now happening.

[00:16:40]It said, "We believe that companies that train the best 2025 26 models will be too far ahead for anyone to catch up in subsequent cycles. Um, they have been remarkably preient in many ways and I think we are now standing inside the moment that they described. If you wanted your nation to be at the frontier, I think you had three years to do it. That was the window and is now closed." And look, you know, I I want to share my raw thoughts as they happen, as I'm experiencing all these things. Um, reality is complex. It's it's dynamic.

[00:17:09]It's everchanging. But you got to I don't know, man. Anthropic has seen things on the inside and they reported it and they were somewhat right, man, about fable, about mythos being unique, about mythos being powerful, about these things continuing to scale up. Now, they are ludicrously expensive. they may be completely economically inviable and that may kill the technology. Like if if the technology is economically inviable, which which again for a $200 subscription plan, you're getting $8,000 worth of usage.

[00:17:42]Mythos is completely economically unviable for most things. So these games that I did probably if I was using API credits would have cost me like $2 to $3,000. I would not have done them. And so there's a clear this clear economic violation here that is more important than anything else. It's like, yeah, sure. If you spend $100,000, can you build a really cool game demo? Yes, that's always sort of been the case. But and it's somewhat genius on their end.

[00:18:09]Um yet all these sort of uh people spending only $200, teenagers, young people, and building really cool demos, sharing it on X, blowing everyone's minds, and the bill will come due soon.

[00:18:20]Uh but yeah, I'm I'm just getting these weird vibes of Enthropic having been right about all this. And here's another thing that I find a little harrowing. Um during the next phase of the race, chip exports will be rationed, narrowed, politicized, and increasingly treated as instruments of national security. The Biden administration already sketched this out. They had chip export caps even for allies. The nuke analogy is overused, but soon the most advanced compute actually will be treated like uranium, monitored, licensed, tracked, guarded, and kept within America's borders. the people in control of it will use all means necessary to keep it that way and to prevent others from building their own stack. And so this is a little spooky, but you know, you start to see it once you use something like Fable, man, I got to say maybe it wasn't all hype. Like, forgive me, but I I've just increased my level of truth a little bit by everything was saying. Not fully. They're still a commercial enterprise. is they still have a pro very good marketing but um maybe they just say what they see and uh you know they see a few years out and and we all hate it when everyone when companies trying to predict try to predict the future it's a little bit unfair and but you know in a sense Anthropic was right about a few things and all this to say the whole again the whole narrative that this replaces humans I still don't buy it I again the common denominator in all this is that I was prompting these models I was directing it every step of the Okay. And so it's nothing without me. So that's very important. We don't really understand the human model relationship in the future. Let me show you another tweet. You remember this Matt Schumer guy? This guy, you know, he's he's totally completely one shot.

[00:19:55]If you remember, he wrote the something big is happening back in February. 87 million views. We covered it on this channel. Um he wrote another banger here. 1 million views. He said, "Assuming Enthropic is able to restore Fable in the next few days, there's literally zero point in doing any meaningful work until it is back. What can be done in 100 hours with Opus can be done in one hour with Fable?

[00:20:18]Hopefully, this is figured out quickly."

[00:20:20]And everyone was calling him out. I was like, "Dude, this is such an exaggeration." Uh, you could say it's a better model, but it's preposterous to say 100 times Opus. I agree. Um, and you know, he found a way to do defend this, but yeah, this is the kind of vibes that I'm seeing on X. Um, people really fell in love with this model. It was tangible and I I happen to agree that the the difference was a little tangible when it came to programming. Again, I didn't feel it anywhere else. I didn't feel it when I was conversing with it just with the cloud iOS app just talking with it.

[00:20:47]I didn't sense any difference there. It it tracks well with what these companies have been saying. Most progress is now postraining reinforcement learning. It's not the the base training data days are over. There's not much new training data. What these models are getting really good at is verifiable tasks, things like math and coding. and the needle's not really moving on things like writing cuz those things are hard to measure. They're much softer. So, you're not going to notice much difference when you're just conversing with it. That's the overall vibes right now on Claude Fable. Presumably, it will come back soon, but presumably it will be way too expensive for any of us to really use. Maybe we'll close on this.

[00:21:21]Frantois Chalet, the co-founder of the Ark Prize. Near-term AI isn't fundamentally different from past tech waves. It's the newest form of digital leverage. It's a force multiplier and force without direction is just noise.

[00:21:34]Still requires a human in the loop at every level in order to be useful. 100% agree. This was my experience coding with Fable. It's nothing without me. So, humans are still and will always be key to all of this.

[00:21:50]So, thanks for watching.