Harlan Stewart

We might not even know when we've crossed it.

Yeah.

Or do we know?

Yeah.

Well, yeah.

So there's, um, AI agents are kind of this other category where it's, you know, what if you took this thing that you give a prompt that answers a question and you gave it some tools and like one of those tools was it could output some texts that calls a function that looks something up on the internet.

And then, you know, once you give it another tool where one of the functions that could run, one of the things that could output is to prompt itself to say something again, then you've got this loop and it can keep running on its own.

And that's one way to, um,

get it to be able to go off and do things like, you know, make a delivery order for you or order your groceries.

I tend to think that a lot of the people who have very confident predictions about what the timelines will be for this stuff are overconfident.

And I think that it's really risky to be overconfident about this stuff.

So I hesitate to say anything other than that.

We just don't know.

We might have only one or two years left until superhumanly powerful systems are something we have to contend with.

And it might be that we have 10 years.

But either way, we're unprepared.