We really don't have great ways to test if a new technology is going to destroy civilization.

From American public media, this is Marketplace Tech.

I'm Megan McCarty-Karino.

Earlier this month, OpenAI released its newest, most powerful chatbot, GPT-4.

And along with it, a technical paper that summarizes the testing the company did to ensure

its product is safe, like asking it how to build weapons of mass destruction or to engage

in anti-Semitic attacks.

It's a process known in the cyber security world as red teaming.

You look for vulnerabilities, security gaps, anything that could go wrong before a product

launches.

Aviv Ovadia was part of the red team for GPT-4.

He's now at Harvard's Berkman Klein Center.

And he says, this kind of testing needs to go further.

With traditional red teaming, you're mostly trying to protect the system itself.

But here these new capabilities can affect so many other parts of our lives.

And the extent of that impact on public institutions, even things like trust, more diffuse public

goods, are just enormous, and we don't have the level of resilience that we necessarily

need to do this.

And there's a lot of calls to slow down development of systems like this, but the economic incentives

aren't there.

And so given that, we have to think about how can we make things resilient and do it as

quickly as possible?

...