Lab 13 · Quiz
Post-training and Alignment Quiz
SFT
RLHF & DPO
Reward hacking
Evaluation
Work in progress This quiz is still being reviewed: questions may change, and some parts may still contain mistakes.
Before you start
Quiz for Lab 13: Post-training and Alignment. Answer every task, then press Check answers at the bottom: each answer is marked, the explanations appear, and you get a score per skill.
Assumptions unless stated otherwise
- The real goal is the behaviour we actually care about; a proxy is a measurable quantity used to stand in for it.
- A reward-model output is a learned estimate, not objective truth.
- Preference labels depend on context and may contain disagreement or noise.
How answers are scored
- Numbers: type the value as a decimal, e.g. 0.82 (a probability may also be written as 82%).
- Select all that apply: scores only when the selection is exactly right.
- Rows: one choice per row; each correct row earns its share of the points.
- Your answers are saved in this browser only, so you can leave and come back. Nothing is sent anywhere.