Hi, I'm Joe Wiley. And I'm Soie Ball. And we're here to tell you about a brand new podcast, Digit.

Digit is all about the things that Joe and I love. From the kids to the parents and all the stuff in between.

There's a lot going on when you get to this age. Listen wherever you get your podcasts and watch every episode on Spotify. Dig it. Every week. Every week.

For the rest of time. Yes. Okay. Good.

I'm John Batscher speaking with my colleague Rachel Lamasky, who is the chief data scientist at Flux.

That is a firm that concerns itself with a responsible AI.

And one of the ways to do that is to concern itself with people who are using AI, not as a crutch, not as a helpmate, but it's the thing itself.

Rachel, one more experiment accidentally came up, and I'm reminded of it in your

article. I was very carefully presenting headlines similar to the one I've written now, and I wrote it

and put it up, and Claude wanted to correct it, so he asked permission. I said, fine.

And the version he came back with, I gave him a word limit.

Well, I said 50 words, now do it in 30 words, now do it in 25.

Each time, Claude failed to be less wordy.

He couldn't do it.

He was limited in his ability to generate the right words so that it didn't have to take three or four to get it done. And I was struck by this again. Claude has no

experience in the world. Claude has actually never had a conversation. Claude is a summary of

conversations that may or may not have the same context,

the same culture, the same background, the same demography. That suggests to me that Claude is

random us. Is that a way to think of Claude? Yeah, I mean, I think that's right. It's, again,

it's missing the context, right? It's training on all of this data, right?

You don't, for example, writing a headline and wanted to sound like a movie script, right?

But it doesn't know that, right?

...