Large Language Models (LLMs) can help students learn effectively. However, their use can introduce several issues, including ethical breaches and environmental harms. While ethical breaches have been extensively addressed, the environmental harms receive little attention. LLMs require substantial computational power, harming the environment. One LLM prompt results in a carbon footprint four times that of a Google search. There are a few studies that focus on informing LLMs’ environmental harms in education. However, many of those are designed as course materials or general guidance. Students might not have time to learn the course materials and may find the guidance limited in relevance to their own context.
We present an online platform that bootstraps LLMs, promoting their sustainable use through mindful prompting and gamification. Mindful prompting is encouraged by applying course-relevant pre-prompts, setting minimum prompt lengths, limiting one prompt per minute, and recommending the reuse of similar prompt responses. The first two aim for effective prompt interactions, while the last two aim to minimise environmental harm. Gamification is applied by reducing the incentive for students using LLMs excessively. For each assessment, students will get 100 game points if their use of LLMs (measured in the total number of words in prompts and responses) is no more than 25% higher than the average. Otherwise, they will have game points deducted based on a proportion of excessive use.
The platform is being evaluated in an informatics course (60 students) via a quasi-experiment. Our platform will be evaluated based on students’ awareness of the sustainable use of LLMs, the number of reused prompt responses, the number of words in prompts and responses, and their perception of the platform. Other indirect metrics will also be discussed, including academic performance, prompt analysis and response similarity.