Learning to Prompt: Improving Student Engagement with Adaptive LLM-based High-School Tutoring
Researchers have developed an adaptive LLM-based tutoring system that personalizes education by extracting 14 pedagogical features from raw transcripts to inform prompt selection. This system demonstrated improved instructional efficiency and reduced interaction turns in simulations and real-world A/B testing with high-school students. While a greedy router achieved similar exercise conversion rates to static baselines, a stochastic router significantly increased conversion rates. AI
IMPACT This research demonstrates a method for improving LLM-based educational tools through adaptive prompting, potentially leading to more efficient and personalized learning experiences.