A new system uses subject-aware prompting to adapt tutoring strategies based on student performance and discipline. A/B testing with 656 student conversations shows the model reduces interactions by 3 turns and increases learning strategy conversion from 19.1% to 28.1% with a stochastic router.
arxiv
arXiv cs.LG
·
6d ago
·
research
Adaptive LLM Tutoring Improves Engagement and Efficiency
from English
Benchmarks
| Benchmark | Model | Score |
|---|---|---|
| AIME 2025 | stochastic router | 28.1% |
| AIME 2025 | baseline | 19.6% |
| AIME 2025 | greedy router | 19.1% |
| SWE-bench Verified | router | 0.69null |
| SWE-bench Verified | baseline | 0.65null |