Brian Roemmele

Love it.

Isn't that the best group?

Little Philadelphia soul, little Curtis Mayfield there.

I love it.

I'm doing wonderful and excited.

Well, thank you, Erica.

And wonderful to be here, guys.

I really appreciate it.

Love you guys so much.

All right.

To train AI, you need to have the widest breadth of data possible.

And you're training AI not just simply on the content of the data, like who, what, where, when, and why, and how, but you're training it on the context

and the situations of how words follow each other.

Large language models are next word predictors.

That's why some people call it smart spelling correctors.

And some truth to that, but not really, because so are we.

I mean, we are next word predictors to a certain level, but we operate in a non-word environment, right hemisphere, our concepts.

So we're really next concept predictors, which is the next AI, by the way, I just gave you a secret on where this is all going, next concept.

But don't wanna digress.

I'm gonna try to stay really tight on