RL & Alignment

Reinforcement Learning for LLMs

Ten questions on the policy gradient theorem, PPO, reward modelling, RLHF, KL penalties, and the RL concepts that underpin modern LLM alignment and fine-tuning.

• 10 questions · single best answer
• Explanation revealed after each answer
• Score shown at the end — no data stored