For Outlier, I participated in reinforcement learning from human feedback (RLHF). This entailed crafting questions with certain parameters to elicit incorrect responses from the AI model, identifying and pointing out errors, correcting them, and rating and critiquing the model\'s instruction-following and truthfulness.