Episode Transcript
[00:00:05] Welcome to Prompting Curiosity, a podcast for the AI curious. No coding background required. I'm your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what these AI tools actually are. Really, though, are the files in the computer, how to use them, and what they might mean for how we think, work, create, and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it.
[00:00:38] Hello, hello, hello, my curious people, and welcome to episode 48 of Prompting Curiosity. I'm your grateful host, the Maestro, and today we are talking about Anthropic's latest model release, Fable 5. So right off the bat, full disclosure, I only used Fable 5 once. I used it to, uh, outline this episode, which is honestly not at all what it is intended for, four.
[00:01:00] But I wanted to at least try it before talking about it. So here we are. Left off the bat, this model, Fable 5, probably isn't for you.
[00:01:10] I have no idea of everyone who listens to this podcast or who reads the newsletter, but with how expensive the model is and what it's actually built for, it's very likely not for you. But we are curious people here, and, uh, like, uh, uh, I know. I like keeping y' all informed. So I figured I'd do a brief episode on Fable 5 so you're at least aware of it and you have some information about it, and then you can go and. And do with it what you'd like. So Fable 5, let's get into it first. Uh, off the 30, 000 foot view. So Fable 5 is the first publicly available version of Anthropic's Mythos model, which is their most powerful model to date. So I spoke about Mythos quite a few episodes ago when it was causing a stir in the news, and it was, you know, it had Security World in a tizzy, and that's still the case. Uh, you can think of Fable as the new top shelf drink right now. New top. Not new top shelf, top tier.
[00:02:09] Uh, you know, lady on the block. So the lineup used to put Opus at the top, right? It topped out at Opus, that, um, monologue. All these names, um, these. These tech people seemingly are notoriously bad at naming things. Um, but the top of the line before, you know, the smartest. I hate saying that. The most capable model was Opus, and now it is.
[00:02:34] The Mythos class sits above that. Um, and you perhaps have noticed that I'm saying two different names here. Mythos and Fable.
[00:02:43] So let's decode this a little bit. Mythos is. And Mythos 5 is the unrestricted version, uh, for just a. A subset of vetted partners. Not everyone has access to mythos M5.
[00:02:56] Fable 5 is the same model with safeguards applied, and that is what has been released to the rest of us.
[00:03:04] The five is because they are Anthropic's fifth generation model. Right? So we're on Opus four Point something and Sonnet four Point something. And so this indicates, like, a new jump, a new tier, and that's why they have the five. So Mythos, the big bad boy, only for certain people, it's Mythos 5. And then, uh, Fable 5, what they released to everybody else. So just a little bit of backstory.
[00:03:31] Uh, Anthropic unveiled Mythos, um, in April, and shortly after that, I did the. The episode about it. Um, and it was a limited rollout because it had, like, incredibly advanced cyber security capabilities. Like, it was a problem, and basically it was like, this is too powerful. So at the same time, Anthropic also rolled out what they called Project GlassW, and that was the countermeasure. So Project GlassWing gave select organizations early access to Mythos, and they were able to use it, and they could use it to fix software vulnerabilities in critical infrastructure before, you know, Anthropic released it to the public. Right. So they were like, hey, Bill, here's a team, you guys, a bunch of, uh, cybersecurity folks. We're gonna give you the model, use the model to go and look at all cybersecurity and all this stuff and banks and financial stu.
[00:04:30] And use Mythos on that because it's really good at picking up bugs. And this way, when we release it to the public, the public can't use it against, you know, in a bad way because you've already patched the bugs. That's essentially what Project glasswing was apparently.
[00:04:46] Just not releasing it at all was not an option. So how is it safe to use now? How is it safe to release it now?
[00:04:56] I. I don't know necessarily that it is, to be completely honest, but, um, what they did was they put guardrails in place, right? So they didn't release Mythos 5 to everybody that only a certain number of people, a certain, you know, class of group of businesses or whatever, have, uh, access what, um, they released to the public. And what you and I have access to is Fable 5.
[00:05:17] Uh, and Fable 5 uses classifiers that will flag any high risk requests. Um, in three specific areas. Cybersecurity, biology, chemistry and distillation. So if you're like me and you have no idea what distillation is, uh, distillation is when a company trains a cheaper, smaller model by feeding it the outputs of a more powerful one. So the smaller model learns to mimic the responses of the big model without needing the big model's compute or architecture. Right. This is how you would get 80% of the capability for 10% of the cost. Which is a nice little hat tip throwback call back to that episode that I did before. Uh, will chat GPT get old navied. Um, and so basically, you know, anthropic's like we don't want companies China they, you know, they're saying without saying to use this to build a cheaper model. So if there are queries that relate to cyber security biology or chemistry or distillation, uh, the model will automatically route those, those prompts to the older Opus 4.8 model which doesn't have the same capability as, as Fable 5.
[00:06:30] So apparently over 90% of sessions are unaffected.
[00:06:35] Um, and by this, like most people are not using it apparently. So what they've done so far and the testing, I don't know. But threads of course had some people complaining that their tasks were wrongfully rerouted. Um, Boris, he's the, the founder of the creator, one of the creators of cloud code and he's very active on threads and he responds to people and you know, some of what he was saying is like hey, it's the new model and there's going to be some, some, I don't want to use the word bugs because that's like what it's like designed to not have. But um, there's going to be some issues with things and uh, issues that people were bringing up. He's actually very quick to respond to and an offer suggestions and be like, hey, run this thing to see what the output is to see what's going on.
[00:07:15] Um, but apparently you know, so those are the safeguards. That is why they released it. Apparently, you know, only 5% of conversations and sessions would were affected and should be affected. Uh, but uh, plain English here.
[00:07:32] If you ask it something sketchy, a different less capable model will answer instead. Answer, answer. Wow. Enter the bouncer analogy that I use for the title of this episode. Right. Uh, when the most powerful model comes with the bouncer. So how it's actually better. I don't have that in this episode because they don't say, they don't say what's under the hood. They don't talk about parameters or anything like that. They're just like, this is the most, the newest, most capable model. Uh, so what is it actually good at though? Uh, the headline for this isn't smarter answers to your questions, right? It is, it's stamina. That's literally it. Fable 5 can work autonomously for longer than any prior Claude model without, without going off the rails. And it outperforms opus. Remember that was their previously their top tier model. It outperforms OPUS on longer, more complex tasks. So the longer and more complex the task, the larger its lead. Fables 5's lead over the other models. So strong areas, um, that, areas that it, that it's very good at in theory, coding and software engineering, long running autonomous tasks, scientific research, financial and analysis and legal document work. So maybe you fall into that catego.
[00:08:43] I don't know. Again, I don't know everyone that listens to this. Um, but I definitely struggled with a use case for it. You know, maybe at some point I would have one, but right as of right now, definitely not. So who actually needs this? Well, if you're vibe coding big projects or you're running long agentic workflows, or you're doing deep multi step research, this is the model for you if you are, you know, drafting emails, which I hope you're not just having it right for you, but if you have, if you're drafting emails, you're brainstorming with it, uh, you're thinking with it, uh, you know, going back and forth with it, uh, asking questions. No need for this, no need for Fable. Just stick to the everyday models. My daily driver continues to be Sonnet 4.6. I actually do all of my building, all my vibe coding with Sonic 4.6. Um, but that is more, is more than capable, right? So once again we, we see AI with their backwards marketing as it relates to the masses and the general public where they're like, hey, here's a solution, go find a problem.
[00:09:39] Um, and of course people on threads, you know, I'm, I'm in that vibe coding, I'm in that AI world on there. And so people are using it for what? I, I honestly don't know what these people do, what they build, they never really show it. They're like, I have a million agents going at once and I'm using all a zillion tokens and I'm like, what are you making now? What is the output of this? I have no fucking clue. Right? I think it is just volume for the Sake of volume. But, uh, I, again, I am not currently building anything or doing any vibe coding, so I literally had no problem for it to solve. So I used use it to help me with this episode and didn't notice the difference. But, uh, that, that was all I could try. So let's talk a little bit about who has access to it and the money side of this. So this powerhouse, it's basically just available for as a preview, little test trial. So Fable 5 is currently included in the paid tiers Pro Max team, um, and See face Enterprise plans, but only until June 22.
[00:10:37] On June 23 it is getting pulled and from there will require usage credits, uh, until Anthropic, you know, maybe has the capacity to fold it back into to subscriptions. But we see it is yet again the Uber playbook. But at least they're being transparent with it. And they're like, hey y', all, you got access to this thing till 23rd, 22nd, and then you're gonna have to pay. So at least they're being upfront. I don't know. You know, the day that this episode drops, it'll be June 18th, so you'll have four more days, um, to play with it, uh, get a trial window. If you're curious, go play with it before then because then it's going to be expensive. So speaking of that, the money side, this model is expensive. It is token hungry. It draws down usage two times as fast as Opus, and that's again their next most capable model. Actually, when you, when you go and flip it, uh, if you're using or trying to use Fable, you can access it. I do everything on the Claude, uh, desktop app. And when you go to select which model you want to use, uh, as soon as you go and flip on like, select Fable, Fable 5. It says it, it says it, it says, what is it like? I'm gonna go to it right now. It says, uh, Fable is the most capable model and draws down usage two times faster than Opus. Right, And Opus is already very notoriously expensive and token hungry. So keep that in mind. You're going to build with it, you're going to use it. It will eat up your usage very, very quickly. All right, so last thing before we get into how I used Claude this week, the, uh, something that's very, very, very much worth noting is that this launch, this release of Mythos 5 and, and Fable 5 came only days after Anthropic publicly urged all major AI labs to establish a coordinated brake pedal on Frontier development.
[00:12:39] And while they're also prepping for an IPO, uh, if you don't know, IPO's initial public offering. So they're trying to go public, uh, be on stock market, let everybody buy up a piece of the company there.
[00:12:53] This is crazy, right? They're literally urging. They're like, hey, hey, y', all, we, we all should pump the brakes. This stuff is getting too good too fast. We should slow it down.
[00:13:04] They're not slowing it down. They're not stopping.
[00:13:06] Right? And then, you know, obviously the argument would be, well, if we stop, other people aren't stopping. So at least, like, you know, I don't know, maybe they're thinking we're the good guys, and so we have to keep going.
[00:13:17] Are we headed towards Skynet? Friends, I'm not here to say either way, no, no doomerism from. From me. Uh, but I will leave that answer up to you. All right, last things last. How I used AI this week. So if you're new here, welcome. Each episode I share a quick example of how I use AI this week. That week. This week, I'm giving it super meta. Because like I said earlier, my use case for AI was simply asking Fable 5 to explain itself to me.
[00:13:46] And then I asked it to help me with the outline asset to go and pull the research, the launch coverage and help me with this.
[00:13:54] With the outline for this episode. Uh, it felt approximately 0% different than when I've used clot in the past.
[00:14:02] And, uh, you know, at the end of the day, I suppose that is part of being curious, right? All right. That, friends, is all for the today. Hopefully you found this episode helpful if you did consider leaving a little rating or a little review. I haven't checked it in a minute. I should go check it. But I haven't checked it, uh, in a little bit. Um, but either way, I love hearing from you. Don't forget. Also, I have a companion newsletter and blog, the Curious Companion, that drops every Thursday right along with this podcast episode. And it's basically, and by basically, I mean exactly, the podcast episode in text format. So if you prefer to read or you just want a written record, you can join the newsletter, or you can just go and read the bloggity blog you head to prompting curiosity.com, forward slash newsletter, or prompting curiosity.com blog. Or you can keep it simple and just check out the link in the show notes.
[00:14:55] As always, endlessly, endlessly, endlessly appreciative for every single one of you. Until we chat again next Thursday, stay curious.
