AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Ai Safety Grid Worlds
One approach is a kind of list, just like millions of things about the world, different things that you can measure. And then i guess the's people wo've ben trying to develop these systems that kind of learn caution. So anything they do that creates irreversibility, where they can't reverse their step, is negatively punished in yet, right away. The progress in these toy environments is encouraging, but scaling it up to a real world setting is far from obvious.