Full course coming soon

Reward Design for AI Systems

7 chapters · 30 lessons

1. 1. From Business Goals to Behavioral Signals

   4 lessons

   Learn how to translate abstract objectives into concrete signals that guide AI system behavior.

2. 2. Reward Hacking and Misaligned Incentives

   4 lessons

   Identify how narrow objectives create shortcuts, gaming, and unintended behavior.

3. 3. Shaping Rewards for Multi-Objective Systems

   5 lessons

   Balance competing goals like quality, diversity, safety, and user satisfaction.

4. 4. Credit Assignment in Multi-Step Workflows

   4 lessons

   Learn to attribute success or failure across sequences of decisions and actions.

5. 5. Exploration, Exploitation, and Learning Dynamics

   4 lessons

   Manage the tension between trying new strategies and using what already works.

6. 6. Human Feedback and Preference Modeling

   5 lessons

   Incorporate human judgment into reward design through preference data and RLHF.

7. 7. Evaluating and Iterating on Reward Design

   4 lessons

   Test whether your reward structure produces the behavior you actually want.