Mark Zuckerberg

then you need to make the design and cost trade-offs that they made that lead you to making a $3,500 device.

So I think that there is a use case for that for sure, but I just think that the companies, we've basically made different design trade-offs to get to the use cases that we're trying to serve.

Yeah, I'm excited to talk about that too.

Anytime where there's a number of serious people who are raising a concern that is that existential about something that you're involved with, I think you have to think about it, right?

So I've spent quite a bit of time thinking about it from that perspective.

The thing where I basically have come out on this for now is I do think that there are

Over time, I think that we need to think about this even more as we approach something that could be closer to superintelligence.

I just think it's pretty clear to anyone working on these projects today that we're not there.

And one of my concerns is that we spent a fair amount of time on this before, but there are more –

I don't know if mundane is the right word, but there's concerns that already exist about people using AI tools to do harmful things of the type that we're already aware, whether we talked about fraud or scams or different things like that.

And that's going to be a pretty big set of challenges that the companies working on this are going to need to grapple with.

Regardless of whether there is an existential concern as well at some point down the road.

So I do worry that to some degree people can get a little too focused on

on some of the tail risk and then not do as good of a job as we need to on the things that you can be almost certain are going to come down the pipe as real risks that kind of manifest themselves in the near term.

So for me, I've spent most of my time on that once I kind of made

The realization that the size of models that we're talking about now in terms of what we're building are quite far from the superintelligence type concerns that people raise.

But I think once we get a couple steps closer to that, I know as we do get closer, I think that there are going to be some novel risks and issues about how we make sure that the systems are safe, for sure.

I guess here, just to take the conversation in a somewhat different direction,

I think in some of these debates around safety, I think the concepts of intelligence and autonomy or the being of the thing, as an analogy, they get kind of conflated together.

And I think it very well could be the case that you can make something and scale intelligence quite far, but that...