Learning to Prompt: Improving Student Engagement with Adaptive LLM-based High-School Tutoring

2026-06-18 · Source: Artificial Intelligence · Field: Education & Learning — Educational Technology (EdTech), Online Learning & Digital Education, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A novel system for adaptive LLM-based high-school tutoring has been developed and tested, featuring subject-aware prompting guided by 14 pedagogical features like tutor scaffolding and student understanding. The system trains a prompt routing model in a simulation environment before deploying it for online adaptation with actual high-school students. Simulation benchmarks demonstrated the router's superior performance over two static baselines, achieving \$0.694$ compared to \$0.647$ and \$0.64$ ($p<0.001$). A/B testing with $N=656$ conversations from 359 students confirmed sim-to-real transfer, showing the model's ability to switch learning strategies. This adaptive mechanism improved instructional efficiency, maintained pedagogical quality, and reduced interactions by approximately 3 turns ($p=0.007$). Notably, a stochastic router achieved a higher exercise conversion rate of \$28.1\%$ compared to a greedy router's \$19.1\%$ and the baseline's \$19.6\%$.

Key takeaway

For AI Scientists and Machine Learning Engineers designing adaptive educational systems, this research demonstrates that dynamically adjusting LLM prompts based on pedagogical features significantly improves instructional efficiency and student engagement. You should explore incorporating a prompt routing model, trained in simulation and validated with A/B testing, to achieve sim-to-real transfer. Prioritize stochastic routing strategies over greedy approaches to potentially boost exercise conversion rates to \$28.1\%$.

Key insights

Adaptive LLM prompting, guided by pedagogical features, significantly enhances high-school tutoring efficiency and student engagement.

Principles

• Pedagogical features enable dynamic LLM tutoring adaptation.
• Sim-to-real transfer is achievable for adaptive prompting.
• Stochastic strategy sampling can outperform greedy routing.

Method

Train a prompt routing model in simulation using 14 pedagogical features, then deploy for online adaptation to dynamically select learning strategies.

In practice

• Implement subject-aware prompting for LLM tutors.
• Utilize pedagogical features for dynamic strategy selection.
• Consider stochastic routing for improved conversion rates.

Topics

• LLM Tutoring
• Adaptive Prompting
• Pedagogical AI
• Student Engagement
• Sim-to-Real Transfer
• Stochastic Routing

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• RECAP: Regression Evaluation for Continual Adaptation of Prompts — Existing prompt adaptation methods are structurally inadequate for proactive, real-time constraint evolution in agentic systems.
• The System Prompting Guide for The Business Engineer — Effective LLM prompting requires a systems thinking mental model, viewing prompts as interventions in a probabilistic distribution.
• Learning to Route LLMs from Implicit Cost-Performance Preferences via Meta-Learning — Meta-learning enables LLM routers to adapt to diverse user cost-performance preferences with minimal interaction.
• Embedding by Elicitation: Dynamic Representations for Bayesian Optimization of System Prompts — ReElicit uses an LLM to dynamically create a semantic feature space for Bayesian optimization of system prompts with aggregate feedback.
• Skills on the Fly: Test-Time Adaptive Skill Synthesis for LLM Agents — Dynamically synthesizing task-specific skills from retrieved trajectories improves LLM agent performance without parameter updates.
• From Heuristic Selection to Automated Algorithm Design: LLMs Benefit from Strong Priors — Integrating benchmark code examples into prompts significantly enhances LLM-driven algorithm optimization performance and robustness.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.