This is John Batchel, conversation with the author Gary Rivlin, his new book, AI Valley,

about the artificial intelligence bots, now being invented across the land, and to monetize those bots.

One particular pie has the skill set taught to it to be sympathetic to the human respondent and in general,

have good manners. Here, Gary takes us into how they teach the bot pie to say thank you and to

flatter you and how the bot pie has other uses, such as therapy.

Is Gary Rivlin, much more of this tonight, the second hour of my conversation with AI

Valley author Gary Rivlin, doesn't know absolutely everything.

And, you know, it's interesting that we talked about the training of these models.

A key element is the fine tuning, where you teach it basically like you don't always have to say, you don't have to be a know-it-all.

You know, saying, I don't know, is an important trait for a bot.

But they had to really treat it, teach it everything, how to have a conversation, how to

synthesize all that. All that was done by humans fine-tuning these models. Some of the more

interesting stuff in writing this book was spending time with the folks at inflection. They have

what they call personality engineers. Obviously, a bot can't have a personality, but they wanted to give it trace. Like, be kind, be kind, don't be rude. And so they kind of show them good examples and bad examples. And over time, made Pi able to say, I don't know, or, oh, that was a very smart answer, John.

You know, it kind of uses flattery.

I mean, again, it's obviously not human, but it's learned from humans.

And so it has what I would call human traits, even though it's a body.

And by the way, that's the future.

I think some people will be scandalized by this.

But I think in not that many years,

there's going to be a lot of people who have romantic relationships with this thing.

Like, you know, as a therapist, like, well, it's better to have a therapist rather than a bot.

...