Sholto Douglas

That's such an interesting idea.

I'm still confused on what will be the bottleneck for these, what would have to be true of an agent such that it's like sped up your research.

So in the Alec Radford example you gave where he apparently already has the equivalent of like co-pilot for his Jupiter notebook experiments.

Is it just that if he had enough of those, he would be a dramatically faster researcher and so you just need Alec Radford.

So it's like you're not automating the humans, you're just making the most effective researchers who have great taste.

more effective and like running the experiments for them and so forth.

Or like, um, like you're still working at the point at which the intelligence explosion is happening.

You know what I mean?

Like, is that what you're saying?

Right.

Um,

So compute and taste, that's interesting to think about because at least the compute part is not bottlenecked on more intelligence.

It just bottlenecked on, say, I'm seven trillion or whatever, right?

So if I gave you 10x the H100s to run your experiments, how much more effective are researchers?

TPS, please.

How much more effective a researcher are you?

So you have some fixed size of compute and some of it goes to inference and also to clients of GCP.

Yep.

Some of it goes to... Huh?

Some of it goes to training.