Sholto Douglas

And there...

I guess as a fraction of it, some of it goes to running the experiments for the full model.

Yeah, that's right.

Shouldn't then the fraction goes to experiments be higher given that you would just be like, if like the bottleneck is research and research is bottlenecked by compute.

Okay, so then tell me what it looks like to be in the world where AI has significantly sped up AI research.

Because from this, it doesn't really sound like the AIs are going off and writing the code from scratch and that's leading to faster output.

It sounds like they're really augmenting the top researchers in some way.

Yeah, tell me concretely, are they doing the experiments?

Are they coming up with the ideas?

Are they just evaluating the outputs of the experiments?

What's happening?

Although in that one, somebody was mentioning to me, like the problem is that

when a human is trying to do a pull request, they'll like type something out and they'll like run it and see if it works.

And if it doesn't, they'll rewrite it.

None of this was part of the, the, um, the opportunities that the LLM was given when run, run on this, like it just like output it.

And if it runs and like checks all the boxes, then, you know, it passed.

Right.

So maybe it might've been an unfair test in that way.

There's so many threads.

Yeah.