**SPEAKER_1** (0:06)
This is Marketing Over Coffee with Christopher Penn and John Wall.
**John Wall** (0:15)
Good morning, welcome to Marketing Over Coffee. I'm John Wall.
**Christopher Penn** (0:18)
I'm Christopher Penn.
**John Wall** (0:19)
So we got crushed by this heat early in end of May, but now it's just been fantastic. It's been like 70 degrees and nice, cool at night. It's been perfect. So yeah, we'll see how long this holds out though. Of course, that's the ultimate thing. I have completed this year's air conditioner power lifting challenge of getting seven air conditioners out of the attic and did not die. So that's a big win. And most of them still work. So that's fantastic. But yeah, it's been crazy on the tech front.
Like this is a surprise. But I think it's more insane than normal. The big story I want to hear the inside view on is fable. So this new model rolls out and then suddenly the government takes it away. What's going on with that?
**Christopher Penn** (1:05)
There's so much drama here. Okay. So for those who have not heard, Anthropic released its fifth generation model family. Fable is the standard version and the unrestricted version that can do things like hack into stuff. It's called Mythos. So if you think back to the models, the original model was called Claude, way back when. And then they came out and started splitting out in multiple generations. So you have Haiku, which is the small model, Sonnet, which is the mid-level model, and Opus, which is their large model. Fable is their largest model and was released on June...
was it 9th? I think it was. It was basically last Tuesday as the day we were recording this. By Friday, Secretary of Commerce Howard Lucknick ordered Anthropic to ensure that Fable was not usable by anyone who was a foreign national. Whether or not that person was located inside of the outside of the United States, which effectively meant they had to turn it off for everyone, because there's no way you can prove who is and is not a foreign national. This was done at the behest of Amazon CEO Andy Jassy, who called his buddy, longtime Trump friend David Sacks, to say, Hey, I think we can make Fable do bad things. You should turn it off, which is confusing, because Amazon rents a lot of compute to Anthropic, Anthropic buys compute from Google, from XAI, from Amazon, from anybody they can get compute from. And so as of Friday, Anthropic had to pull the model and say, until we can figure out how to work around, you know, this directive or get more clarity on the structure, we can't allow people to use this. So as of today, as the time of recording, it's not available to the public. However, already some folks in the Open Models community, things have been looking at the reasoning traces that they got out of Fable and have successfully replicated some of its capabilities in other models, like using Open AIs GPT 5.5, using the Haiku AIs brand new GLM 5.2. So Fable, you know, we'll see when it comes back. I got a chance to try it out. We had three days and running that thing, boy, does it vacuum up your usage quotas.
But that's the story so far.
**John Wall** (3:29)
Right. And so I guess first, though, so no other additional story as far as it's just like, it can do dangerous things. That's really all we know.
**Christopher Penn** (3:38)
That's the thing is that Jassy sent that to his buddy and provided no proof, no evidence. They just said, I'll take my word for it. And of course, the CEO of Amazon telling a person in the federal government who is not particularly knowledgeable or skilled at AI, just sort of nodded and said, okay, we'll do it.
**John Wall** (3:57)
Yeah. Well, so you make an interesting point that they're a customer and yet they still raised a hand. So that does give you the impression that there is definitely something there. They would not have gone against their best interest to shut it off if that was not the case. All right. Well, and then the token thing, were you finding that the results were insanely better? Or like, why is this chewing up so much more resources? Or is this just a business thing? They're trying to actually turn things profitable for a change?
**Christopher Penn** (4:21)
It's, when it's running, it has a tendency to think more, which uses basically a talk to itself more. And it tends to spawn dynamic workflows, which is essentially make copies of itself to work, to decompose a problem and work on each piece independently.
20 more minutes of transcript below
Try it now — copy, paste, done:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090
Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000773347602