Beth Lyons

You gotta read the chat while it's happening over mental days and weeks.

I don't.

But to Brian's question, if it were that easy to reverse engineer, we would see this a lot.

Right.

Yeah.

So, yeah.

I'm going to name two things.

They both happen on the show.

And in my head, it does not mean that they're related, but it does mean they're not related.

We also saw a story that OpenAI has an internal model to identify leakers, right?

Oh, that's full.

And I don't, that I assumed was like, who's writing that on X, right?

Who's the anonymous source of blah, blah, blah, blah, blah.

But if something was purloined, that may be actually another reason to,

run that internal model to identify things.

And Jeff is saying that maybe actually it's just knowledge distillation.

Take a bunch of input output pairs that come out of an open A model and then train your models to emulate that.

In which case, maybe you don't have to actually steal the weights, but you get close enough.

I don't know.

We know that Grok did it because Grok said that it was open AI at one point when Grok first came out.