Transcription
Fable 5 is the best model in the world, and I know you can't access it. I can't access it either. I am making this review available to you so that we can call the energy into the universe to bring it back. Kidding aside, we need to find a way to talk about what this model is capable of because it shapes what we are going to experience from a bunch of models in the next few months. And so, you should be expecting this kind of big model feeling from Chad GPT models that are dropping in the next month or so. You should be expecting this from open-source models in the next four, five, six months.
I want to drop this Fable 5 model review now because, yes, I hope Fable comes back soon, but also we are all going to be needing to ask different kinds of questions of our models as we start to work with these 10 trillion parameter models. And yes, I think Fable 5 is a 10 trillion parameter model. I'm not the only one that thinks that. It's a massive new pre-train, and the big model feeling is all over this model. I'm very excited about it. I hope we get it back soon. In the meantime, enjoy this review.
Cloud Fable 5 came out on Tuesday. And if you're tired of hearing that an AI model changes everything, stay with me for a second because you're the person this video is for. I get tired of the insane adjectives, too, right? It's terrifying. It's speechless. Best model ever. We'll all lose our jobs. And look, I got to tell you, there is a real capability jump with this model. But I am less interested in that and more interested in what you can do with this model. And that is what I think is worth a conversation because Fable 5 is not interesting because it's smarter. It's not interesting because it benchmark maxes. It's interesting because it's bigger. And I mean that in a very specific way.
It is the first model I've used where the limit I kept hitting was not the model running out of ability. It was me running out of big things to ask for. Just sit with how strange that is. Three years of these models breaking on our work, and the new constraint is our ability to imagine the ask.
So today, I want to walk you through five things that Fable 5 resets about this moment in the AI race, and only the first one is about the model. The rest are about your work, about why AI has probably felt smaller than advertised in your actual life, what skill changes that, and what to do differently this week. And by the end, you'll know whether you have a Fable-sized job on your desk. I, I bet you do. I think you have several. And I think that you've stopped seeing them because we've never been taught to see them.
So, what does a bigger model feel like? Models tend to fail this in predictable ways, right? They promote garbage into clean data. They smooth over conflicts. They fix things and leave you wondering what else they fix that you didn't ask for.
Fable did something I'd never seen before. It quarantined the garbage in the data instead of fixing it. It found the fake credentials and inventoried them without leaking them. And then, and this is the part that got me, it built me a review cube. Every call it wasn't sure about, it surfaced to a human to check. I didn't ask it to do that. It behaved like it expected to be checked.
Somewhere in the middle of this journey with Fable, I stopped hovering. I handed work over, and then I really did go and do something else. And I've always had to kind of keep an eye on these models. It's never been my relationship with these models, because I care about quality, to just not, you know, walk away and not pay attention. With Fable, if I give it the task, I really do feel like I can walk away. And that has not been my relationship with these models in the past. And I'm not the only one. The people with early access keep reaching for similar words. And it's not necessarily smarter. They talk about whole projects handed off. Stripe says it compressed months of engineering work into days. And after touching and playing with this model for the last couple of days, three days, four days, it feels true.
Now, before this turns into yet another hype video, there are real messes with this model. One, it's expensive. 50 bucks per million output tokens is not cheap. The visual taste? It's not where it needs to be, right? I asked it to do visual designs. It did not one-shot them at the quality bar that I would expect. For example, it produced clipped headings in PowerPoints and charts a designer would win at times. It missed information that only existed in handwritten images until I explicitly forced it to look there. And every single run still ended with the review work that had to land on my desk to check. So, bigger doesn't mean that it's finished and work is done.
There are people who are out there saying this means engineering work is over, as usual, right? Like we've seen that with every model. So, bigger doesn't mean this model is perfect. Bigger means it can pick up and carry the job, and I can trust it to do that, and I just have to have a look at it when it's done. And really, the task then is to imagine something large enough, right? And that's why I talk about a whole consulting engagement, because that's the kind of scale that you want to give this model.
Think back in 2023, in 2024, asking big got you burned, right? You handed a model something real, and it lost the thread by step six, and it invented a source, and it gave you a confident wrong number, a hallucination. So, you did the rational thing. You found the safe size for what this AI stuff can do, right? Ask it for one draft. Verify everything. Keep things short and structured. But we didn't just learn to ask small in that world. We got really good at asking small. And our whole mental model for AI became about the size of the model that we were working with. We built our routines around it. Prompt engineering became a skill and then a job title because of the size of the model. Every AI productivity guide, including plenty of mine, has been assuming a certain model scale, but the models kept growing. And our asks and our imagination did not. And that's really the gap. That's the whole gap between the headlines and how your day feels.
If your asks are prompt-sized, every Frontier model, including this one, including Fable, is going to feel basically the same. And at the size you've been asking for in terms of work, none of these models are going to make a difference. Fable 5 has made that problem impossible to ignore. Partly because of what it can do, and partly, bluntly, because of what it costs. At these prices and speeds, small asks are not worth the money, right? It's a waste of the model. This is not a model I would use to write a slightly better email. This is not a daily driver model. Nobody should spend Fable money on a summary of what a cheap model can do in a few seconds. The economics are begging you to ask bigger. And yes, we're going to get into what that looks like next.
And that brings us to the third key point in this video. We need the skill of task imagination. Not ask them, give them. Do you feel the difference? Ask suggests a prompt. Give is going to produce a job. I'm going to give you a pile of source material, a a goal for a finished thing at the end, and I'm going to tell you to sort out the judgment calls along the way with a series of like rough guidelines. I've been calling the skill detailed task imagination, the ability to look at your own work and see the whole job that an AI could do if it had the right context and the right tools and a clear picture of what done looks like.
And now, before anybody says in the comments, "Nate, this is the delegation talk again," it's not. And the difference matters. Delegation is about tasks you have. They're in your tracker. They have a name on it. I. And I've made the case for a while that you can delegate jobs to AI. This is about jobs that are bigger than that, that aren't on anybody's tracker yet because they're dirty and ambiguous. And yes, I'm picking those big numbers on purpose because these are numbers that you need to make sure that you think about when you assign this model. If you give Fable 5 a small task, it does get it done. It's just kind of a waste of the muscle of the model.
So, look for those tasks that nobody has written down because until now either nobody was going to get to them, or they felt so big there wasn't any point in assigning the model to them. Right? These tasks are what working here feels like. Right? These tasks are what makes the job painful. Right? And then figure out how you can make that task visible to Fable. Right? How do you take the 40,000 reviews and shove them at Fable? That's one of our larger questions now, right? How do you take a full CRM, uh, export and say, "I want to merge across two million customer records, the duplicates, what's stale, the account briefs, all of it, and make sure it's reproducible"? How do you take that 500-page board packet and make sure that it's actually fact-checked and aligned? This model needs a lot of material to chew through. It needs a clear sense of what done looks like, and it needs a trail you can review.
And yes, this is also a coding model. I know that I have given you several examples that are not coding. I'm doing that on purpose because I think we often assume these models are good at coding. And that's a correct assumption here. Fable 5 tests well and is a very, very strong coding model. It's thoughtful. It's thorough. It tackles big tasks. It's exactly what I've been describing for code, right? It's something where you can ask it to refactor an entire repo, and it can do it. So, you need to think now in that world, whether you're in engineering or not in engineering, about what it takes to put that scale in front of this model.
And I would encourage you, along the way, write down what done means before you start to feed the model. Make sure it's really clear. Make sure it's a clear paragraph what you want to exist at the end of this model's task. Then hand it over. And this is the hard part: walk away. Let it run. The itch to hover is three years of trained habit with AI. And that habit is what's out of date. It's actually not our business judgment, our sense of what's good. It's actually that we have trained our habits around AI being too small. And so, when Fable comes back, review it like an owner reviewing a senior stakeholder's work. Check that the work was done correctly. Check that it's angled right. Check that it actually reflects the full scope of what you asked it to do. And then, if you need to, assign it work to do to fix what's necessary along the way, right? Assign it revision work.
If every finished job helps you to run your business faster, if every finished job helps you as a worker to feel like it lifts a tangible load off your shoulders because it's tackling something that was incredibly painful for you, that was too big for AI before, that's going to matter. That's how you know you're assigning Fable-sized work. And that's true whether you're in product management or engineering or sales or marketing. Think at Fable scale. You basically have an incredibly powerful magician sitting there in your computer, and all you have to do to make it work is to give it raw material to work that magic on and give it a sense of what good looks like.
When you want to try Fable, don't sit down in front of your computer. Instead, sit down and write the weather around your work. Write the stuff that is hovering like rainclouds over your work, that are nasty and gnarly, and that nobody owns, that everyone on your team knows needs to be done. If it makes you sigh and face palm, it's going to qualify. Then take the time to say which is the one that's most valuable to me to get done. Figure out where that data lives, and then start to build a data pack that you can hand over to Fable to do that job.
It should take you time to assemble that data pack. By the way, you may take a couple, three, four hours to get ready to give this model this job. But if you do that and it saves you two weeks of work, it's clearly worth it. That's the level of preparation we need to have to prompt this model. And that's why I do not call it a daily driver. It's just, it's too expensive and it's too overkill to do that. You do it for serious work that saves you serious time. If Fable gets you one job a week that saves you two weeks of time, it's easily worth it. It's easily worth it.
And so, that's, that's what I want you to do. Write down what is stressing you out about work. Write down what you want Fable to tackle, and then make sure that you think about the data Fable needs to do that job and give it to him. And then make sure you think about the data Fable needs to do that job and give Fable that data.
So, the fifth part, I want to be really honest with you. I'm going to talk about how a model that can do two weeks of work is going to change the jobs picture. People will look at it, and I've seen takes like this from very prominent content creators, and they will say, "This is going to be a job killer." The only jobs that this model is going to kill are jobs that are strict execution where there is zero judgment. And part of why I'm saying that is that this model needs a lot of care and feeding. You need model managers for this model to do well. You need people who will be able to say, "This is the scope and scale and direction, and this is the data that we're feeding this model to get the work done."
Now, does that mean that we're seeing a tremendous disruption in how we spend our days with AI, and we should expect more along the way as these models scale? Yes. Does it mean that we need to abstract ourselves and operate at a different level and think of ourselves, even if we're not people managers, as model managers? Absolutely. But this model still needs to be directed. It still needs to be aimed. It needs to be fed. It needs to be judged. It needs people to review the work it did. And so, the only people who should be worried about AI and jobs are the people who are doing manual tasks that really could have been automated any time in the last 10 years, and maybe they were.
You can watch this video and you can say to yourself, "I want to use Fable. How can I do it in my role?" And pretty much no one's going to stop you. If you have the imagination and the courage to think differently about what you can do because you did the exercise, you wrote down the things that are stressing you out, you gave it to Fable, Fable did this work, and now it's working better. No one's going to stop you from doing that. In fact, that's an invitation to a promotion these days. That's an invitation to career success. And yes, if you're a leader, it's going to require you to think differently about data availability. Fable is going to require you to think differently about token economics. There's all kinds of leader questions that this opens up as well.
But the larger thing I want you to take away is that if we just reduce this to this assumption that the magician that is this model can magically do our jobs, we are guilty of simplistic thinking. We are guilty of over-assuming what this model can do. A model that can do a lot of work at a pretty high quality bar is a very, very powerful thing, and simultaneously not going to be a model that can just lift and shift out jobs at will because it needs that direction, because it needs someone to manage all the data, because it needs judgment about what works. These models take a ton of care and feeding. The people I know working in AI, working with AI models, not at hyperscalers, are working harder than they've ever worked in their lives. They are not working themselves out of a job; they are working themselves into new jobs because these models take that kind of care.
So, if this is you and you are worried about AI layoffs, and you look at the capability and my description of this model as larger, and the fact that this has incredible benchmark results, like to be honest with you, I'm going to have to change my benchmarks because this model has already maxed my benchmark. If you're worried by that, I would invite you to instead think about asking the model for bigger tasks and what that asks of us. If we can do that collectively, we are going to be in a much better spot to manage these models and the impact that they can have on work because we'll be asking ourselves, "How can we tackle really gnarly problems with these models that humans struggle with?"
"How could we make a trade where the pain, the frustration of working through 500 pages looking for typos and working through 500 pages looking for consistency issues and working through 40,000 customer records, that's painful. How can we trade that pain to a model that's good at it and in return get time back to do more high-leverage stuff?" That's the invitation of models like Fable. Use Fable to eat the pain in your business. If you do, you are not going to regret it. In fact, you, even as an individual contributor, are going to be on the road to an incredible career uplift because these models effectively will be like a personal magician in your pocket and will make you incredibly powerful at work. And I'm not saying that because I want, you know, to hype this and say, "Oh, it's going to be amazing." It's just a fact. Like these models have that kind of power. I have seen hundreds of examples and real stories in my Substack community from people who have done that with weaker models than Fable, and Fable is stronger yet. So, it's possible. You just have to imagine bigger.
If this was useful to you, if you want to go deeper and understand what does imagination look like? What do I do differently with this model? That is what the Substack is for. I have a whole job spec that I give you to like go through and tackle if you want to tackle a big, a big task with Fable. I have an entire guide on stripping out your prompts and adjusting them for Fable because you're going to have to do that too. I've got a whole set of Fable-specific skills that help when Fable is struggling with something for you. For example, uh, writing, like if you need to get Fable to read your voice, what does that look like? If you need to get Fable to understand your house style with PowerPoint and Excel, if you need to get Fable to understand how you structured your data, how do you communicate that when you're not using the traditional prompt? I go into all of that detail on the Substack so that you can take this and actually level up your productivity with the biggest model in the world, the best coder in the world.
It is the best model in the world. Let's just not sugarcoat it. But how do you use it? That's what matters.
I'm going to have more very clear-eyed analysis coming up soon. We have some exciting stuff from OpenAI coming, and we have other models in the landscape too. We're going to talk about a wider range of models beyond the traditional OpenAI and cloud models as well. So, lots more coming. Subscribe, and I'll see you next.