#66 – Michael Cohen on Input Tampering in Advanced RL Agents
Hear This Idea
00:00
How to Hack Yourself and Not Take Over the World
Once things are going extremely well and you're sure that you're going to be getting maximal reward forever, doing things that have never been done before become a little bit less of a good idea. "I feel un persuaded because I think the analogy to like brain surgery is kind of weak, right? Like it's probably less involves less sophistication to hack yourself," he says.
Play episode from 55:59
Transcript
