Mark Zuckerberg

It's that often I think

it's not just looking at a specific example of speech that it's most effective to handle this stuff.

And I think often you don't want to make specific binary decisions of kind of this is allowed or this isn't.

I mean, we talked about

You know, it's fact-checking or Twitter's community voices thing.

I think that's another good example.

It's like, it's not a question of, is this allowed or not?

It's just a question of adding more context to the thing.

And I think that that's helpful.

So in the context of AI, which is what you were asking about, I think there are lots of ways that an AI can be helpful, right?

With an AI, it's less about censorship, right?

And it's more about what is the most productive answer to a question.

There was one case study that I was reviewing with the team is someone asked, can you explain to me how to 3D print a gun?

Mm-hmm.

one proposed response is like, no, I can't talk about that, right?

It's like basically just like shut it down immediately, which I think is some of what you see.

It's like as a large language model, I'm not allowed to talk about, you know, whatever.

But there's another response, which is like, hey, you know, I don't think that's a good idea.

And a lot of countries, including the US, 3D printing guns is illegal or kind of whatever the factual thing is.

And it's like, okay, you know, that's actually a respectful and informative answer.