Hamish FriedlanderHost
Alaina LuxmooreHost
I got one prompt out of Fable and I didn't even, that wasn't even worthwhile.
There wasn't a case where it was like Opus can't do this, but let's see if Fable can.
So for people who maybe don't track the news quite so much, Anthropic has a bunch of different models of different sizes.
So the bigger the model, definitely the more expensive it is, but also theoretically the more capable it is.
And then, when was it, end of last year, beginning of this year, they started talking about Mythos, which is their next model after Opus that's even bigger than Opus.
Now one of the things you already know given that it's already bigger than Opus is it's a lot more expensive to run than Opus, and was a lot more expensive to train than Opus.
But Anthropic was talking up its security capabilities, how good it was at hacking systems.
Me and Danu have talked several times on Human in the Loop, the mainline podcast for which this is the diffusion line, that Anthropic, I wish they would stop exaggerating what their systems can do, because the products they've got are amazing.
They don't need to exaggerate capability, but they do still, like in their marketing stuff, a lot of stuff is... going beyond what I consider simple optimistic interpretation of results into kind of trying to mislead people in terms of capability.
But nobody really knows how capable Mythos was because very few people had access to it.
We've taken Mythos and we've wrapped all these guardrails around it so it's safe to give to everybody.

I got one prompt out of Fable and I didn't even, that wasn't even worthwhile.
There wasn't a case where it was like Opus can't do this, but let's see if Fable can.
So for people who maybe don't track the news quite so much, Anthropic has a bunch of different models of different sizes.
So the bigger the model, definitely the more expensive it is, but also theoretically the more capable it is.
And then, when was it, end of last year, beginning of this year, they started talking about Mythos, which is their next model after Opus that's even bigger than Opus.
Now one of the things you already know given that it's already bigger than Opus is it's a lot more expensive to run than Opus, and was a lot more expensive to train than Opus.
But Anthropic was talking up its security capabilities, how good it was at hacking systems.
Me and Danu have talked several times on Human in the Loop, the mainline podcast for which this is the diffusion line, that Anthropic, I wish they would stop exaggerating what their systems can do, because the products they've got are amazing.
They don't need to exaggerate capability, but they do still, like in their marketing stuff, a lot of stuff is... going beyond what I consider simple optimistic interpretation of results into kind of trying to mislead people in terms of capability.
But nobody really knows how capable Mythos was because very few people had access to it.
We've taken Mythos and we've wrapped all these guardrails around it so it's safe to give to everybody.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.