S2. Ep23 - Fable 5: The Best AI Model You Can't Use Anymore

Katie (00:26)

Hi, welcome back to another episode. Hi, hello, I'm Katie. And as always, I've got Noel here with me. Hi, Noel. How are you doing this week?

Noel (00:34)

I'm doing absolutely fabulous this week. How are you doing?

Katie (00:37)

Yeah, I'm good. Thank you. So there is quite a bit of news. We always start the podcast with what's new, what are the updates, what are the headlines in the AI and automation world. But this whole podcast is going to be on news and an update.

Noel (01:04)

Yes. It's quite a big one.

Katie (01:06)

Okay, so Noel, do you want to explain a little bit more about what we'll be discussing this week?

Noel (01:13)

Yeah, so I thought we'd go through the drama around Fable 5. So it's had quite a week.

Katie (01:20)

Do you just want to clarify what Fable 5 is, in case someone hasn't heard of it? Because it is very new, it only came out about a week ago.

Noel (01:32)

Yeah. So Fable 5 is a brand new model from Anthropic. Normally Anthropic have three model levels. They've got Haiku, Sonnet, and Opus. Opus is usually their most powerful. And last week they released Fable 5, which sits above Opus. So it's far more powerful. And looking at the benchmarks, completely agree, it's way more powerful than anything they've released before.

But they've also got an extra layer on top of that called Mythos, which is only available to certain big tech companies around the world. Not everyone's allowed that one because it is that little bit better than Fable. And a little while ago, about a couple of months ago, when they first talked about Mythos, they were quite worried about people maybe doing stuff that they shouldn't be doing with the AI models. So they were worried about cyber security and things like that. They were putting in loads of guardrails to make sure that anybody using it could find security issues on their own software for big tech companies. But they didn't want everyone having access to it because they could go off and find different workarounds and hacks.

So Fable was brought out and sits in between Opus and Mythos. Last week it got released. I had loads of time with it. I had loads of fun using it, got loads of usage, and then the US government turned it off.

Katie (03:15)

Yeah, but they didn't turn it off just for the US. They've turned it off for everyone, for users in whatever country you are in.

Noel (03:26)

It's kind of weirdly worded. I think that is the case, but there was some weird wording where US citizens in the US could still access it. So that was raising question marks like, well, what about Anthropic employees that aren't from the US? They could be based in the UK, for example, and wouldn't be able to access it. So it's kind of messy. But I would say no one has access to either Fable 5 or Mythos currently.

It's kind of worrying, actually, how quickly it was turned off. The reason it was turned off is because the US government spotted something in there that they didn't quite like. I think they managed to break one of the guardrails and get round certain security things and do stuff that they shouldn't be able to do. And because it's super powerful, they were like, get rid of it. It's not safe.

But the way Anthropic have come back to it going, well, the same issues are in Gemini, the same issues are in OpenAI models, they're in Opus, they're in Sonnet. This isn't a brand new vulnerability, it's something that's always been there. They haven't publicised what it is, but they have made it clear that it's not just them.

Katie (04:53)

Okay, so what does this actually mean for users? Not just in terms of Fable, but in terms of being able to use these platforms and having some reliance on them to run some form of your business, whether it's just using AI or if it's part of a bigger automation.

Noel (05:20)

Yeah, so this is the first time this has happened. And I think it's been talked about and was kind of like, well, these companies like Anthropic and OpenAI, Gemini, they're all gonna mitigate those risks because they want people using it, because they're then earning revenue from the usage. So it's in their best interest that they toe the line and don't really do anything they shouldn't.

But the speed that this was turned off has highlighted quite a few risks for businesses going forward. Normally what I would do is assess a new model. If it's any good or fits my use case, I would then swap it out and put it into my automations. So now going forward, I'm probably not gonna do that. I'm probably gonna wait a couple of weeks. I'll test it, but then I'll wait and see if there's any backlash. Because having that turned off, there were a few press releases about it, but it wasn't that well known. And you wouldn't know about it until your automations started failing and you start getting emails from Zapier or Make going, this is all broken.

Katie (06:30)

Yeah, because I didn't know that it had been turned off. I think I said to you, it must have been Sunday, I can't use Fable. And you were like, it's been switched off. And I was like, what do you mean it's been switched off? I was thoroughly enjoying using Fable.

Noel (06:55)

Yeah, it's an incredible model. It really is.

Katie (07:03)

So going forward for you, Noel, you're not going to jump straight in to switch into a particular model as soon as it's released, even though it's going to be better and have newer features. Just because now you're thinking, well, actually if I jumped too soon and changed everything over, it can be pulled or it can have some sort of issue. And you'd be spending time swapping it over to a new model, and then by Saturday, Sunday evening UK time, you'd have been swapping it all back over.

Noel (08:00)

Yes. It depends on how many automations and apps you've got. That could be quite painful. Worst case, we could have been on holiday and then having my automations just not working and I wouldn't be able to get in and change them. It's quite a risk.

Katie (08:20)

Or for people who don't work from home, maybe they're in an office. They might not have noticed until they went in Monday morning. And actually Monday morning they're putting out fires rather than just cracking on with their day. They would have spent the day working out how to change this back and spending what could be hours swapping everything back over. And of course that costs you time, it could cost you clients, cost you money. So you're going forward with caution.

Noel (09:00)

Yes. I would still use it within chat and things like that. That's fine. I use Claude Code a hell of a lot on the desktop app, so I'm absolutely gonna hammer it once it comes back. It was brilliant for coding. So I would still use the latest models for that. But for automations or within apps, I'd probably not go straight in.

Although with Fable as an example, I wouldn't have put it in my automations anyway, just because of the cost. It was double the cost of Opus. So if you've got lots of things going through AI in your automations, that was gonna cost you tens or hundreds of dollars just to run it, where Sonnet could have just done it for about twenty. The price difference between Sonnet, which is still a very capable model, and Fable is huge. So for me, it's only ever gonna be used for coding stuff and building things, or just chatting with.

Katie (10:24)

So Noel, is there anything else that we can do to protect ourselves when it comes to our AI and automation access?

Noel (10:33)

Yes, there is. There's loads of open-source models available on the market. You've got your DeepSeek, you've got your Kimi K2.5, all your Chinese models, you've got Mistral from France. There's loads of lightweight and open-source models that you can download, you can host them. If you've got a good enough computer, you could host them locally on your computer. You could host them in the cloud.

You can even train them as well. You could do your own fine-tuning if you so desire. But what that means is once you have that model, you can attach it into your automations and that model will always be there, providing your server is up and running and it has access to it.

So it saves you from all of this. If the US government goes, right, that's it, I'm turning off OpenAI, Anthropic, and Gemini, all three are going, you could sit in your office with a smiley face going, well, it's alright, I've got whatever model on the VPS somewhere which is continually working. So there are those options available.

But there is a trade-off in terms of quality of output. A lot of the open-source models aren't as good as the big models from the big three providers. You'll soon spot that. I would have them running automations which are fairly basic and not too heavy on the need for AI. Maybe it's categorising stuff, maybe it's writing summaries. That sort of thing, it'll do perfectly fine. But complex stuff like writing full documents, maybe not.

They're all different. They're all upgrading just as quick as the main models. So there's lots of options out there. And from a business perspective, you own that model as well, technically. Only your data is in there, only you access it. It's not going off to the US or China. It could be hosted within your country, which could be a requirement for your business and your clients.

Katie (13:24)

In what sense do you say it's been a wake-up call? Why are you saying that? Can you explain to whoever's listening why you feel that way?

Noel (13:37)

I think the biggest thing from it was the speed and the total loss of access to a model. I was using it all last week. It was fabulous. I thoroughly enjoyed it. It was doing things that Opus 4.8 could only dream of when it came to coding, which was brilliant to see.

Especially for people that are outside of the US, or say China is the other big country with loads of AI models. It makes us a bit more wary. These countries can just turn it off without any prior notice. It can be there one minute and gone the next. That's the wake-up call.

Hopefully one day in the UK we'll have our own frontier model. That would be very handy. But going forward, I think lots of other countries will probably look at that as well. Some things coming out in the UK news were like medical laboratories using Fable 5 for research purposes and now all their research has stopped. They were enjoying that great model and then all of a sudden it's gone and Opus 4.8 is probably just not up to snuff for what they wanted. So it does have big real-world impacts.

Katie (15:28)

Do you think that Fable is going to get switched back on? Are they just ironing out some problems? What's going on with it? What do you know?

Noel (15:35)

They've been very quiet officially since it got turned off. But apparently some of their engineers are now in Washington having a chat with the US government to try and iron out this problem.

There's also a lot of chatter about it never coming back because it's maybe too powerful and maybe too much too soon for the consumer market. I kind of get it, it can be a bit disruptive when you have such a leap like that. But looking at it from a business perspective, having a model that's twice the price of your most expensive model currently, having that turn back on is only gonna be a good thing when it comes to making money. So I would have thought Anthropic would be doing absolutely everything to make sure this model can be turned back on so they can start making more money off people using it.

Hopefully it's gonna come back soon. But as of yet, no word just yet.

Katie (16:55)

Okay, so there have been many interesting and thought-provoking takeaways from this week's episode. Thanks, Noel, for sharing your knowledge, as always. And thank you so much for listening. We hope you've really enjoyed this episode, and we will catch you next time for another one very soon.