Beth Lyons

Basically what it does is it can create, it can understand the physics in a game, whether the physics are reality-based or special in some way.

It can understand worlds generated on the fly, 3D worlds generated in the fly.

And then it achieves significantly better generalization across games than SEMA 1.

It can successfully work in unfamiliar virtual worlds.

They've tested it on several games.

It's a current research preview.

And what it does, because I was asking, like, okay, are we at the Borg level?

Like, did one gen go through the experience and now everybody knows?

No, but it did help me understand a little bit about how the...

The upgrades within a single model are happening because what it does is it reweights itself after it goes through an experience where it's learned something.

So it shifts the weights a little bit and then the weights are put back into the SEMA model, right?

So it's not something that...

automatically every other SEMA instance knows what it is like a software update.

And it doesn't touch Gemini because Gemini is the core model powering everything.

So this is like a little layer that sits on top of it.

And I hadn't really clocked that, right?

Like that wasn't something that was clear to me until I had that conversation.

It seems like I want the Borg reality.

I don't.

It is not something that I am trying for, but it is a benchmark for me that I keep sort of looking at because I think we are going to get there.