Cory

And it's been really interesting to watch efficiency start to play a role in those kind of builds where like,

Not every step.

I don't need a frontier model to classify a thing, maybe.

I need the frontier model to analyze the thing so we can tick through and really dial in where you're spending your money and your time.

you know as as models have gotten better like across the industry and uh the bottom end of what isn't impressive now is way higher than it used to be like like like the base level it just keeps rising and and we're reaching this point where

I think that is where most normal tasks are is in that range that's really dialed in now too.

And it's funny because like, you know, before we come on, we're always sitting here and we're throwing prompts at whatever model we're working on.

And our questions are feeling a little more lame than they used to and a little more dated.

It's like, it's almost more,

about looking at how it answers the question and its behavior for me with some of the simple questions more than it is about can it do it.

And it's trickier to find questions that stretch it and that we understand as people.

That's a strategy we lean on really heavily when working on shows and newsletters and websites and all of the things is like,

Give me a bunch of options.

Don't, I don't ask for one of anything anymore.

Like it's very like, if I'm trying to come up with a headline and I'm just stuck, I'll be like, you know, give me 25 and let me, or a hundred, you know, it really doesn't matter.

It will give you however many you want.

That's the thing.

And it's fascinating how that works.

And that when, when a new model like a Gemini three flash comes out,

is that the first instinct of, of like the AI Twitter sphere, for example, is like, okay, how do we break it?