Reinforcement Learning Quiz (Easy)

In Reinforcement Learning, what is a “Policy”?

What is the primary method by which a Reinforcement Learning (RL) agent learns?

What occurs during the phenomenon known as “Reward Hacking”?

Why is Proximal Policy Optimization (PPO) popular in the industry?

What is the purpose of the “clipping” mechanism in PPO?

Which method is usually more stable and resource-efficient for LLM alignment?

How does Direct Preference Optimization (DPO) differ from PPO?

In the context of LLMs, how does PPO typically receive its reward signal?

What kind of training data is specifically required for DPO?

Why is DPO often described as an “Offline” method?

What is the fundamental link between a robot learning to walk and an LLM learning to chat?

RL agents don’t just look for immediate points; they look for the “Expected Return.” What does this term mean?