RLHF Overview Quiz

5 questions Pass: 70% +25 pts (50 with Pro)

Quiz covering Fine-Tuning and Customization

What you'll be asked

  1. What is the primary goal of Reinforcement Learning from Human Feedback (RLHF) in the context of Large Language Models?
  2. In the standard RLHF pipeline, what is the role of the 'Reward Model'?
  3. Which of the following describes the 'Supervised Fine-Tuning' (SFT) phase that typically precedes RLHF?
  4. Why is a KL-divergence penalty typically applied during the Reinforcement Learning phase of RLHF?
  5. In a scenario where a reward model is trained on human pairwise comparisons, what is the mathematical objective during the optimization phase?

Answers and explanations come after you submit.