Grant Harvey

And maybe there'll be something when we hit the right number of sort of node equivalents, like the number of neurons in the human brain and suddenly there'll be magic.

I can't predict that, but I...

I don't know.

I'm not betting on it as much.

There'll need to be other structural advances that I'm very much not qualified to even understand, let alone talk about, to get to, like, AGI.

Like, it's revealing that the people talking about AGI are the ones very, very incentivized to talk about AGI.

So I think we're going to, but we are going to see more of a data need because people are going to try and make incremental improvements for competition.

And also we're going to start to see more like second level users, like not the foundation model builders, but the people who are building complex products in the healthcare space, in the financial space, whatever, using agents, using models who still have a need for trajectory data, for evaluative data, for data for reward modeling, for improving whatever.

be it multilingual.

I mean, think about it.

Lots of these models are very, very good at English because they're trained on a predominantly English data set.

And then there isn't at other languages, but there are a lot of other languages.

I mean, there are seven or eight languages in India with 100 million speakers.

That's a lot of potential consumers.

So there's a huge demand in these different places.

With their own literature, with their own quirks and things.

There's a huge amount of stuff that can be done there.

And also, as training costs go down, maybe, then you can start doing... Just because of... I don't know.

This isn't exactly an example of Jeben's paradox, but just because something becomes easier doesn't mean people use less of it.

Oh, it's massively, massively gone up.