Jamie Bartlett

You need to train them on that stuff.

very, very strict safety rules about them and sort of rebuild them, if you like, for this specific purpose.

And if you can do that, I think there's a world in which people use lots of different small language models for very particular tasks, which will be less likely to hallucinate.

They'll probably be safer.

We wouldn't get fooled quite so often by them.

We wouldn't get sucked in so often by them.

Well, because it just sounded a bit like my name.

That was it, really.

Look, I was trying to do something a bit unusual for BBC audiences, which they're often quite sensitive on things like deception and using AI and stuff.

But I wanted to show to the listeners that we're entering into a world where you might never be quite sure whether what you're hearing is machine generated or by a real person.

And what better way to do it by just building a kind of a bot based on me.

Yeah.

And seeing what it came up with and showing how easy it is to be fooled by that.

Right.

Really, that was it.

All right, I'll stop dodging and actually pick one.

No hedging, no vibes-based cluster.

Favourite episode, Is Britain Broken?

Early 2024.

This one really sticks with me.