Demis Hassabis: recommendation

Transcript context

…Some of these other labs have specialized in different things relative to safety, Anthropic for example with interpretability. Do you have some sense of where you guys might have an edge? Now that you have the frontier model, where are you guys going to be able to put out the best frontier research on safety? I think we helped pioneer RLHF and other things like that which can obviously be used for performance but also for safety. I think that a lot of the self-play ideas and these kinds of things could also be used to auto-test a lot of the boundary conditions that you have with the new systems. Part of the issue is that with these very general systems, there’s so much surface area to cover about how these systems behave. So I think we are going to need some automated testing. Again, with things like simulations and games, very realistic virtual environments, I think we have a long history of using those kinds of systems and making use of them for building AI algorithms. I think we can leverage all of that history. And then around Google, we’re very lucky to have some of the world’s best cybersecurity experts, hardware designers. I think we can bring that to bear for security and safety as well. Let’s talk about Gemini. So now you guys have the best model in the world. I’m curious. The default way to interact with these systems has been through chat so far. Now that we have multimodal and all these new capabilities, how do you anticipate that changing? Do you think that’ll still be the case?…

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