从厌恶到真香：Fable 5.2 如何改变了我对 Claude 的看法

Surprisingly, this started when they released Fable. I remember Fable showing up and it was such a huge leap in intelligence and such a stark reminder that intelligence does not make you intelligible, which meant I hated working with Fable. Fable was so hard to work with, so annoying to talk to, so not easy to understand. Same with Opus 5. Just not an enjoyable model to work with, even though it maybe got the job done.

At the same time, those models had come out a couple months ago, I felt like Codeex really outpaced Claude Code in some nice to have experience pieces in terms of their desktop app. So, Codeex became my favorite place to work and code. Not only because it was a nice place as a software engineer, but things like browser use and computer use were just light years ahead of Claude Code.

Once we all have access to this model, I'm going to be a true menace to society. I'm going to build so much stuff. I think this might be the model that unlocks consumer AI because there is just so much cool that you can do with it. I love Fable 52 and today we're going to do a just in time mini episode telling you how I tested this model, what I think it's great at, where it needs a little bit of work, and why I think I'm a claw boy. Let's get to it.

There are going to be so many headlines about Fable 52. My favorite is that in science, it solved two open problems in the physics of black holes. This is the little tidbit I told my kids this morning at the kitchen table. I am really excited about the frontier in terms of life and hard sciences.

But let's focus a little bit about coding. What they say is it's really good at stubborn bugs, performance optimization, architecture, and long running tasks. Um, it can do huge migrations. It can build things like operating systems. And I will tell you my experience is it's very fast. You're going to see a lot of amazing benchmarks. It's basically outpacing all the Frontier models at each model's highest effort level. Frontier code, Frontier Suite, Automation Bench, all of them. It is doing really well. I am going to run it against the full how I AI bench. We're going to see a preview of that today, but I want to do the full end to end against all the newest models. And today is just a mini app on my impressions of Fable 52.

So, there are still cyber and bio guard rails on this model, of course, unless you're in the Mythos program. And there's also a lot of anti-distillation safeguards in this model. We'll talk about what it is, what it costs a little bit later.

It's also really good at creative work and at being creative with one exception, which I will talk about. Um, it's great at video generation and detailed video generation. You're going to see a lot of fun videos in this episode because I am super excited about what I was able to build. Um, 3D and gaming. I think it's a step up. I'm not going to do one of those like AI YouTube bro race car games, but I did make a game and I thought it was pretty fun. It is really great at coding. It has helped me with some architecture rearchitecture work, longunning engineering tasks, things that previous models, Opus 55, Astra, etc. have not been able to really get through and it's not annoying.

What are the cons? Well, it's not good at front-end and web design. You will kind of see that in um some of the evals and things that I show you. Like it's really good. It's really tasteful when you give it a good brand or style guide and it's good at following it, but left to its own devices, it has incredibly concerning taste. Like she just does not have good taste. And then I think this comes from a lack of image gen model really shows because when a model is really creative, you want it to be creative along a lot of aspects. And I found myself a lot of times invoking OpenAI's image gen 25 to create images. And then you know Fable wasn't very good at that in particular. So I, you know, high level I am going to use this model all the time. I really love it. But I want to show a couple things that it was really good at and then again later we'll do the full how I AI bench.

And so the example I want to show you is I had this tweet about software factory. So, everybody's been talking about software factories. I had this tweet about software factories. And I gave Fable that tweet plus our CXO.dev brand guidelines and said, "Make this tweet a video." Now, I will say I have to go back and forth and I want to show one sub loop that I did that was really impressive with Fable that made this amazing. But look at what it generated for me.

There's a story I keep seeing inside engineering orgs. It starts with one sentence. We need a software factory and everyone falls in love with the idea. So one team gets a quarter to build it. Plan, code, discovery, validation, review, release. On a slide, it's beautiful. Then they turn it on. The repo isn't ready. The checks are slow. Nobody's been trained. And what they've really built is a worse version of Claude Code.

So I won't make you watch the whole thing, but you can just see there. This is a full animated little Space Invader bot full video and it keeps going. There's like a bug factory. It's done all these amazing visualizations. It ends with this like billboard. It's not 100% perfect, but it's way better than it was before. It is certainly not like an animated slide. And I was just so impressed. I did not give it any creative guidance other than our brand book. I did not storyboard this. This is really Fable being independent and building this by itself.

In fact, I have built this kind of invader um animation into the new Howi AI eval bench. And so, if you see here, animate the invader bots will now be in Eval as part of the Howi AI bench. I'll test different models against it. But look, you can see he walks in, he looks like a little Claude code claude. Um, he asked if you can ship it. They get excited and jump up and down. Again, this is something that 100% it looks super simple. I could not get any other model to pull off. And it was fast. It was fast. And so I just think about how it has a good sense of geometry, how it has a good sense of movement. And I've been really impressed with Fable 52 in terms of animation.

If you've ever done complex IKEA or flatpack furniture, you know that the order in which you put stuff together is really important. And it put really good. And I know cuz I followed these instructions with the um the guy that came to my house to assemble it. It did it in the right order. And this was something again that no other model was able to do. I tried. I tried. So I know this is a true benchmark. And it did it truly from pretty low res images of the flat pack items on my floor with Chinese language labels on them. And again, zero instructions. Um good news to you all. This piece of furniture is now in my house and feeling really nice.

So, I made this game about not running a bug factory. Um, you are an engineer that has to stop bugs before they get to um production. You get to pick your challenges. And so, if you have more agents, you have more agents stopping bugs, but then if they miss a bug, more bad stuff gets through production. You get protected time, therefore you have more time to catch bugs or you have tests. Again, you can see the tokens are getting through. Oh, I got some free I get some free credits. I get these look good to me. Oh no, vibe coding. So many bugs.

Okay, so this is my CXO um game. And what I really loved about this game was again I just said make a game about fighting bugs using our space invaders from our brand. And it came up with all these concepts. It tied it to business things that we thought were awesome. It made a trailer for us that I thought was really funny. I'm not going to play the sound because it's going to be really annoying for people watch listening on the podcast, but it says basically this quarter the team had one job then they added more agents. They shipped faster. they call Finance, but Finance saw the token bill. So, they get downgraded to the cheap model and can only solve easy bugs and it um stars the monor repo, which is this giant space invader, as the final boss. I think it's really funny. I played it a couple times and actually giggled. I'm going to try to get this live today.

But again, like remember how annoying Opus 5 and Fable were when they came out. So annoying. Not a good hang. not enjoyable to be around. And now we have this like funny model that I like working with.

And so I asked Fable 52 just to make the pipeline more efficient, easier to maintain, just do its greatest product thinking. And I will say I'm like very happy with the work here. It's quite sensibly organized and in fact matches my upcoming kind of like in my mind road map of things that I needed to do. The other thing that was really nice on working with Fable 52 on this is it stopped at the right moments but was pretty autonomous until it needed a PR. And so I found that it's longunning, that it's got good architectural thinking, that it does code nicely, and that it's not annoying to work with um on this fairly complex kind of like data pipeline product that has I will say like very messy code in it from 6 months of iteration. So I'm about halfway through this migration. I will get through it all and let you all know if it works, but so far so good and I have been really happy with it.

It's bad at web design. It's like real bad at web design. And so, I've been trying to make AI that um helps me slow down, websites that help me slow down. So, I asked you to design a new kind of like browser homepage for me. This is so so ugly. Such much slop. So, so terrible. And so, when I've left Fable 52 to its own devices, it's made pretty bad designs. Same with this story book. I've been working on a kind of like ghostly story book. Um, same with this story book. Like, it's okay, but it's pretty ugly. So, I might pair Opus 55 with Fable on the back end for some of the design stuff, but I really don't love the front end design of Fable. So, that's a kind of like thumbs down on the model.

And I would like to introduce you to Claire's typewriter. So this is a fully functioning 3D rendered typewriter. You can type on it by clicking the keys or you can type on it by using your keyboard. You can zoom in and get a closeup of how it works if you backspace and type over things. You can do that. Welcome to how I AI. This is my beautiful typewriter.

Okay, so this is you can see I have a a literal typo in this. You can clean it up. It will go and put uh wide out. It doesn't quite get it right, but it will do. Wide out and type over. But that's not the only thing. You can make it real viby. So, let's turn on some music. Let's turn on some rain. We need some nice rain. Um, you can start a writing timer here by turning one of these sandlasses that starts a 15-minute timer. The big one's 30 minutes. Um, and then you also have these beautiful prompts. So, if you want to write about waiting, don't say what it's for. You can get a writing prompt. You have this nice cup of coffee. I have actually integrated GMO's record player over here. You can change different radio stations. So, we can have a jazzy radio station. Isn't this nice? Um, and this is like I am totally going to write this way.

And then my favorite one, I can do a Y2K computer and I can change it between a green and amber screen. And if you go back and I say, I love this thing. I can print the page and it will print it on my doc matrix printer. Keep it in a stack. I can open up these stacks, read them. I can save them. I can copy the text. Um, and I am gonna deploy this and you all can use it. I am obsessed with this thing. It is a singleshot.

I'm going to turn off the music for a minute. Actually, it's pretty relaxing. I'm going to keep it on. I like it. Um, it is so relaxing. And I just think I'm going to come in here and do full screen writing on my beautiful typewriter. And listen, you can even hear the static when it plays on the record player. And this truly one shot by Fable the keyboard. I've done some tweaks. I've kept giving it new features. Um, but I'm going to deploy this and see if you all want to type on my beautiful, beautiful typewriter.

This has been my AGI moment. I lost it when I saw this. I think it is so cool. It's helped me be so creative and it's just something I could never imagine being able to do before and I was able to do with Fable 52.

Now again, I have done the H full howi AI bench here. You can see here it is very fast. So um even something as complex as animating the invader bots only took 16 minutes. A lot of things are in sub 1 minute. Um there's some complex tasks. So I will do the full how I AI benchmark against a couple other models um maybe in a week or two just so we can see all of this against each other. But I really love the model.

Um, again, let's just close with how do you get it? So, it's going to be available in the API and on the cloud platform as well as AWS, Google Cloud, and Microsoft Azure. It's going to be in the cloud apps and in cloud code. I've been using it in cloud code. There's going to be a fast mode. So, even though it's already fast, they're going to make a new fast version. And subscribers, you're going to get fable resets. So, use this thing. I just it like sparks joy and I just think back to how much I truly hated truly hated the original fable and now I just I I feel warm fuzzy feelings about this model. So good job Anthropic. You hooked me back. I never thought it was possible but you did a really good job.

That has been how I AI hot takes on Fable 52, my new favorite little creative model. I'm going to get back to my typewriter. I'm going to get back to my creative projects and you all are about to see a lot of video and a lot of consumer apps out of your friend Claire. Thanks for joining how I AI.

Thanks so much for watching. If you enjoyed this show, please like and subscribe here on YouTube or even better leave us a comment with your thoughts. You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review which will help others find the show. You can see all our episodes and learn more about the show at howiipod.com. See you next time.