Answer: Decrease temperature

## Explanation **Temperature** is a parameter that controls the randomness of the model's responses: - **Lower temperature** (e.g., 0.1-0.3): Makes the model more deterministic and focused on the most likely responses, resulting in more factual, consistent, and on-topic answers - **Higher temperature** (e.g., 0.7-1.0): Increases creativity and randomness, which can lead to more imaginative but potentially off-topic responses **Why other options are incorrect:** - **B. Increase top-p**: Top-p (nucleus sampling) controls the diversity of responses by limiting to the top probability mass. Increasing top-p would actually allow more diverse responses, potentially making the problem worse. - **C. Raise max-tokens**: This controls the maximum length of the response, not its factual accuracy or relevance. - **D. Disable stop-sequences**: Stop sequences are used to control when the model stops generating text. Disabling them wouldn't address the issue of imaginative, off-topic responses. **Best Practice**: For factual, HR-related applications where accuracy and relevance are critical, use a lower temperature setting (0.1-0.3) to ensure the model stays focused on providing accurate, on-topic information.

Author: Ritesh Yadav