It's not just about the ethics of what kind of data you have, whether it's terrible toxic texts

from the dark web or, you know, whether you've stolen it from people or not. Machine learning 101 is that

you don't test on your training set, right? So if you're testing how well your model is doing on a specific task, you have to make sure that the data

in the test set does not have overlap with the training set, because otherwise it's like

studying to the test.

Hello and welcome.

Wise.

It's happening with me, your host, Chris Hayes.

Welcome back to our special with Pod series where we take an examination of AI and all of the embedded questions, complications, themes, worries, aspirations relating to our economy, our society, our culture,

the technical implications, the philosophical implications.

This has been a really joyful project for me because I like learning stuff and I feel like I'm getting somewhere and I hope you guys feel the same way too.

There's lots of ways in which the AI discourse can feel.

I had a friend who's a therapist who said it gave her the feeling of avoidance,

which I thought was interesting.

And I can get that too sometimes.

But then when you sort of pierce through that and do it in the way that I think where you've been trying to do,

I've really been enjoying it.

One of the challenges in the whole discourse is basically this. There's a bunch of really fairly

complex and sophisticated technical questions. That can be, I think, a little difficult for people

who are non-technicians to get their arms around. The people that have expertise, almost A,

as a matter of sheer self-selection and B, financial incentives tend to be pretty big boosters of the

technology. And again, I don't say that because I think people have bad motivations necessarily.

...