Sam Altman

When designing something that can change the world, there always has to be an acknowledgement of the fact that it can change the world in the worst way or for the worst.

With each leap of technology, there's been an outsized ability for one person to do more damage.

Is it possible, the first part, to make AI completely safe?

And then the second part of it is, what is your nightmare scenario?

What is the thing that you think of that would make you press a red button that shuts open AI and all AI down?

When you go, you know what, if this can happen, we have to shut it all down.

What are you afraid of?

And so the first one is, can you make it safe?

And the second part is, what is your nightmare scenario?

So the way I think about – so first of all, I think the insight that you started with, which is the number of people that can cause catastrophic harm goes down every decade or roughly every decade.

That seems to me to be like a deeply true thing that we as a society have to confront.

Yeah.

Second, about making a system safe, I don't think of it as like quite a binary thing.

Like we say airplanes are safe, but airplanes do still crash very infrequently, like amazingly infrequently to me.

We say that drugs are safe, but people don't.

The FDA will still certify a drug that can cause some people to die sometimes.

And so safety is not like – it's like society deciding something is acceptably safe given the risk-reward tradeoffs.

And that I think we can get to.

But it doesn't mean things aren't going to go really wrong.

I think things will go really wrong with AI.