Beth Lyons

with the companies.

They're beholden to it in terms of they may not be able to use it anymore.

Those accounts got shut down, I'm sure.

But if the speed with which new models gets created is faster, the distinction between the new models, while big on benchmarks, is not super big from a user interaction standpoint.

And China can do this better.

cheaper uh right if if the path to frontier ai capability is steal the outputs and smuggle the chips and it's working then how does that affect hundred billion dollar bets that labs in the west are making

And on building safe AI, like there's a lot of spokes from that question that come out.

Yesterday, we had sort of the funny takeaway that is, well, technically, it's better for the environment if we don't.

build a new data center and and try to create it like if this is the better environmental way to create models um but it's not the better financial bet for the people who are investing in it in the west well that definitely is a much larger story and uh you're right gareth uh

Right.

Well, because the other context of that, and I swear I will let you weave off this to Notebook LM if that's where you want to go.

But the other piece of this is the context in which these stories are taking place.

we've just had the release of two really good models that had versions of those models help train and help make the new models better with the advances that they have.

So if you've stolen enough, if you've stolen enough, now you also have a model that can train its own next version.

Not as far as I know.

Usually chat tells us what's launched while we're online.

And you're charging for the knowledge of what it feels like when a wrench is turning something that is going to turn eventually easily or is turning something that's going to break the pipe.

Yeah.

Right.

Like that's really what you're paying for.