#66 – Michael Cohen on Input Tampering in Advanced RL Agents
Hear This Idea
00:00
How to Train a Dog to Do Tricks
In theory, you probably could train a myopic dog to do tricks. So I want to get a sense of when I come across a particular problem to know whether this is a job for reinforcement learning versus something else. What are some like real world examples of reinforcement learning? There's an agent called Gatto that DeepMind came out with recently where they just trained it to do well in a bunch of kind of random tasks. Well, you'll probably have heard of AlphaGo or AlphaZero, um, which are Go and chess playing agents. It might be better to call them optimal control.
Play episode from 16:42
Transcript
