Caspar Eliot

Now there's caveats in there.

But I think one of the biggest things will be public opinion on Waymo's because you'll read stories where people are like, oh, I don't like that they idle in front of my driveway or I don't like that they make this loud beeping noise.

Yeah.

There was a study that just came out about this.

A Harvard business study that basically said people blame robots more.

Like when robots are at fault, people really don't like it.

Like they're way more likely to call them.

That's right.

Totally.

Well, I mean, from a data perspective or like training perspective, I mean, is it harder to train?

It must be harder to train robots and self-driving cars and the physical side of things, right?

So, I mean, if you're a person who plays League of Legends and you're really good at it and or you are a parent of a kid who plays League of Legends, like, get your kids an AI.

Yes.

Is there a magic number of once we get it past a certain threshold in terms of our evaluation metrics where it's okay to deploy it?

Does your team have some sort of internal way?

Is it different for every company?

Yeah, because we keep hearing about evals and certain people are saying, well, you have to have a systematic way to evaluate your outcomes.

Otherwise, it's just going to fail.

There was that study that's like 95% of enterprise deployments don't have an ROI because maybe this or that reason.

So as someone who builds these tools and trains them, meaning you know them pretty well, what is in your personal AI stack right now?