Dario Amodei: belief

Transcript context

…Is there any ideas about sandboxing it? How difficult is the sandboxing task? Yeah, we sandbox during training. So for example, during training we didn’t expose the model to the internet. I think that’s probably a bad idea during training because the model can be changing its policy, it can be changing what it’s doing and it’s having an effect in the real world. In terms of actually deploying the model, it kind of depends on the application. Sometimes you want the model to do something in the real world. But of course, you can always put guard, you can always put guard rails on the outside. You can say, “Okay, well, this model’s not going to move data from my, the model’s not going to move any files from my computer or my web server to anywhere else.” Now, when you talk about sandboxing, again, when we get to ASL-4, none of these precautions are going to make sense there. When you talk about ASL-4, you’re then, the model is being, there’s theoretical worry the model could be smart enough to kind of break it to out of any box. And so there, we need to think about mechanistic interpretability. If we’re going to have a sandbox, it would need to be a mathematically provable. That’s a whole different world than what we’re dealing with with the models today. Yeah, the science of building a box from which ASL-4 AI system cannot escape.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.