RLHF and Prompt Engineering

Reinforcement Learning from Human Feedback

• RLHF uses signals from human evaluators (e.g., upvotes/downvotes) to improve AI model performance.
• Feedback helps the model adjust decisions to align with human expectations.
• Useful when defining a clear objective function is difficult.
• Example: social media moderation AI uses RLHF to better identify offensive or harmful content and limit its spread.

Prompt Engineering and the New Interaction Model

• Prompt engineering is a valuable tool with broad applications; it's a new interaction model.
• It changes how we talk to AI and shape its behavior.
• Goal: learn how to talk to AI.