An accounting firm is deploying an LLM-based document processing system and wants to follow responsible AI practices. Which two actions should the firm take during development and deployment? (Choose two.)

Including fairness metrics for model evaluation is crucial because it allows the firm to systematically identify and measure potential biases in the LLM's outputs, ensuring equitable treatment across different demographic groups. Modifying the training data to mitigate bias directly addresses the root cause of many fairness issues in LLMs. By curating or augmenting the data to be more representative and less skewed, the firm can proactively reduce the model's propensity to generate biased or discriminatory results. Adjusting the temperature parameter primarily affects the creativity and randomness of the output, not inherent fairness. Avoiding overfitting is a general machine learning best practice for model generalization, but doesn't directly address responsible AI concerns like bias. Applying prompt engineering techniques can influence output but doesn't fundamentally alter the model's underlying biases or provide a systematic evaluation of fairness.