AI and Existential Risk (Robert Wright & Connor Leahy)
Robert Wright's Nonzero
00:00
The Future of GPT Systems
The current state of the art is that mostly systems are not coherent enough to make plans. They work for a little while, but then they get like big or crazy and kind of forget where they are. But this is being improved very, very rapidly. There's already people working on systems that are way more coherent over many, many steps. And this is when I expect the first real like, rogueness things to start becoming visible.
Transcript
Play full episode
