Does anyone get the feeling that this might be a bad idea? Even if LLMs are far away from real intelligence (which I think is probably true), increasing their power and letting them interact doesn't sound so great. What happens if their collaboration has new emergent phenomena that makes them want to do destructive things? What happens when someone, just for fun, puts one of them in a device that can physically manipulate the world, like maybe a drone or something.

Regardless of the hype, I think there could be quite a few unintended consequences from this sort of behaviour, which seems quite reckless to me. To me it seems that those who are pushing AI forward have completely lost their sanity.