Sholto Douglas

You don't have to use 100% of your brain all the time.

MARK MANDELBACH- Right.

Welcome to my world.

FRANCESC CAMPOY- And so you should be able to run that faster and this kind of stuff, basically.

So I think it's net-net, I think, typically the same model.

Because you want to be able to scale the understanding as the complexity and difficulty.

You want to be able to do that dynamically.

So, yeah.

The residual screamers is like this operating RAM.

You're doing stuff to it.

Right.

It's like the mental model I think one takes away from interpretability work.

There's a surprisingly strong bias so far towards token syntax.

It seems to work very well.

There already is some amount of neuralese, right?

If you think about the residual stream for each token is like neuralese to some degree.

And so now we're just trading off axes, like how much neuralese are you doing versus how much actually is read out to tokens all the time.

Interpretability becomes dramatically more important as you shift in this direction of neuralese.

And there'll be like some selective pressure against it so long as the agents are working with humans because they'll want to sort of cooperate.

But then like as agents begin to work more and more with each other, then that selective pressure like changes the other direction basically.