AGI Can Be Safe
Data Skeptic
00:00
The Future of AI Systems
The European AI Act defines that you should be putting safety layers in your AI system. A safety layer might also be human oversight so it might be a person sitting next to the AI who is going to look at what the thing is doing and then as the hands above the emergency stop button and then they'll hit it. I think current language models as used as world models they make too many mistakes. They have too many blind spots. If you imagine that future language models might be much better world models which do not make mistakes anymore yeah then they might have James Bond level competence in a way in your commands. So at that point you definitely want to insert some safety layers that I designed or that
Transcript
Play full episode
