Sholto Douglas

I mean, paperclip.

Okay, some rapid fire.

What is the bus factor for Gemini?

I think there are, yeah.

Yeah.

Although I think it's an interesting note about the field that people like you can get in and in a year or so you're making important contributions.

Um, and I, especially with anthropic, but many different labs have specialized in hiring like total outsiders, physicists or whatever.

And you just like get them up to speed and they're making important contributions.

I don't know.

I feel like you couldn't do this in like a bio lab or something.

It's like an interesting note on the state of the field.

Who are you paying attention to now in terms of there's a lot of things coming down the pike of multimodality, long contacts, maybe agents, extra reliability?

Who is thinking well about what that implies?

I mean, that was one of the things in the Will Scaling post, I was referring to something you said to me, which is, you know, I miss the undergrad habit of just reading a bunch of papers.

Because now there's nothing worth reading is published.

Yeah, it is kind of striking.

All right, cool.

Okay, I'm trying to think, what is a good last question?

I mean, the one I'm getting those thinking of is like, do you think models enjoy next token prediction?

We have this sense of things that are rewarded in our accessor environment.