Preference Labeling
I evaluated two or more videos or images that were generated by AI in response to a user prompt. I wrote extensive feedback on my choice after evaluating the models on specific criteria including image quality, instruction accuracy, and the degree of AI-generation appearance. Most videos were under 5 minutes in length and most prompts were several paragraphs. Subject matter often included science, mathematical diagrams, academia, documentary style photography and even the relationship between stylized art and photography.