From raw model to assistantLesson 3 of 10
Learning from preferences, and reward hacking
A reward model that likes politeness scores a padded wrong answer 1.56. After ten updates the engine flags the hack: reward up, correctness down.
Locked
Unlock the rest of Deep Learning: from one neuron to a transformer.
- Every lesson, every resource, unlocked instantly.
- Track progress and pick up where you left off.
- Free preview lessons stay readable from the outline.