#66 – Michael Cohen on Input Tampering in Advanced RL Agents
Hear This Idea
00:00
The Alternative Framing of RL
It's obviously less intuitive to imagine that these agents are like really caring about getting more reward rather than just like selecting them to do the thing you want them to do. So I'm curious what your takes are on this kind of like alternative framing of RL, whether it is in fact a mistake just like assume the RL advance RL agents want reward.
Play episode from 01:23:28
Transcript
