This is an I-Heart podcast.

Guaranteed human.

Quarzo Media.

Hello and welcome to Better Offline.

I'm your host, Ed Zichron.

Better Offline. And we are recording here in beautiful New York City.

And I have a wonderful pair of guests today.

I've had, of course, Alison Morrow from CNN Nightcap newsletter and Edward

Anguoso Jr.

The Hater himself from the Tech Bubble newsletter.

Thank you so much for joining me today.

Great to be here as always.

Thanks for happening.

So I think we should start with exactly what we were just talking about.

The Open AI claims that they have worked out what causes hallucinations.

Alison, do you want to go over this?

I should have read the paper a bit more carefully, but I can, you know, the highlights were getting digested yesterday on X and Blue Sky. And it seems like it's kind of the test taking problem of when

you encourage students, when they're taking standardized tests, and you don't know the answer,

you guess. And that's exactly how these models are trained. And they don't, you don't get a point if you say,

I don't know. Right. So you come up with something. And the models are meant to keep guessing until they

get something close to right. Right. So that's why you get a lot of nonsense and hallucinations. And Open AI, at least in their reading of it, says, oh, this is a simple solution.

We'll just encourage the models to understand better when it's like a binary question.

...