Reward and Restraint

Optimize a three-answer policy and discover why more reward can mean less truth.

Try it · toy model

12 min

Read the concepts behind it