Exploring LLM capabilities to create content for Microlearning : Evaluation of AI vs. Human-generated content

This study explores the use of Large Language models (LLMs) to automate content generationof microlearning content, and compares its quality to corresponding human generated content (HGC). The AI-generated content (AIGC) consists of multiple choice quizzes based on a prede-fined set of YouTube videos, as this is a common form of content in microlearning. A Python-based AI-workflow was implemented to automate the content generation using OpenAI’s gpt-4o-mini model. A blind user evaluation was conducted, using three quality dimensions derived from existing research. These were Resemblance (how well a quiz corresponds to the video it wasbased on), Clarity, and Perceived reliability. Results from 64 evaluations revealed no statistically significant differences between AIGC and HGC across any quality dimension. Additionally, participants were generally unable to distinguish between the two types of content. These findings suggest that with only moderate prompt refinement, and the use of a small model, LLMs can produce educational micro-content of comparable quality to that created by humans, offering a promising path for improving content creation efficiency without compromising quality.