Connor Leahy on the State of AI and Alignment Research
Future of Life Institute Podcast
00:00
The Core Problem of Alignment in RLETF
The core problem of alignment is how do you get a very complex powerful system that you don't understand to reliably do something complicated in domains where you cannot supervise it? RLETF does not address this problem. It doesn't even claim to suggest this problem. There's no reason to expect that RLETF should solve this problem. This is like, it's like, you know, clicker training in a alien.
Transcript
Play full episode
