Multiple creativity assessment tasks and human preference ratings are provided for evaluating model-generated creative content. The data includes human-generated responses and qualitative scores used in the Creative Preference Optimization framework.
Use Cases
- Train a reward model using the human preference ratings to rank creative outputs
- Evaluate model performance on divergent thinking using the creativity assessment prompts
- Fine-tune models using preference optimization techniques based on the human responses and ratings
Strengths
- Includes human preference ratings for creative response pairs
- Features multiple creativity assessment tasks for multi-task evaluation
- Contains human-generated responses to standardized creativity prompts