Research map / Large language models / Language, safety and interpretability
Can Language Models Teach? Teacher Explanations Improve Student Performance via Personalization
Swarnadeep Saha, Peter Hase, Mohit Bansal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| random intervention with a larger teacher's explanations | Flan-T5-Large student · StrategyQA | accuracy: 58.51 -> 62.96 | Improved |
| random intervention with a smaller teacher's explanations | Flan-T5-XL student · StrategyQA | accuracy: 68.12 -> 61.86 | Hurt |
| human teacher explanations at 100% intervention | Flan-T5-Large student · StrategyQA | accuracy: 58.51 -> 81.95 | Improved |
| stronger LLaMA-65B teacher explanations | LLaMA-7B student · StrategyQA | accuracy: 61.13 -> 69.43 | Improved |
| weaker teacher explanations for stronger student | LLaMA-65B student · StrategyQA | accuracy: 77.58 -> 69.14 | Hurt |
| human explanations for already strong student | Flan-T5-XL student · CommonsenseQA | accuracy: 92.38 -> 92.21 | No clear effect |
| Flan-T5-XL teacher explanations | Flan-T5-Large student · CommonsenseQA | accuracy: 84.78 -> 89.72 | Improved |
| LLaMA-13B teacher explanations | LLaMA-7B student · GSM8k | accuracy: 9.62 -> 12.25 | Improved |
| Expected Utility intervention instead of random at low budget | Flan-T5-XL teacher, Flan-T5-Large student · StrategyQA | accuracy: up to 7 points at 20% budget | Improved |
| increase intervention beyond 60% with Expected Utility | Flan-T5-XL teacher, Flan-T5-Large student · StrategyQA | accuracy: 69% -> 63% | Hurt |
| rank by true utility using student confidence | Expected Utility teacher · StrategyQA | accuracy: 76% at 20% cost; 81% overall | Improved |
| Expected Utility computed with teacher answers instead of gold labels | Flan-T5-XL teacher · StrategyQA | accuracy: up to 2 points at 20% budget | Improved |
Closest papers by meaning
- DataEnvGym: Data Generation Agents in Teacher Environments with Student Feedback (ICLR 2025 · similarity 0.88)
- MASTER: Enhancing Large Language Model via Multi-Agent Simulated Teaching (NeurIPS 2025 · similarity 0.87)
- SocraticLM: Exploring Socratic Personalized Teaching with Large Language Models (NeurIPS 2024 · similarity 0.86)
- Few-Shot Knowledge Distillation of LLMs With Counterfactual Explanations (NeurIPS 2025 · similarity 0.86)
- Unveiling the Learning Mind of Language Models: A Cognitive Framework and Empirical Study (NeurIPS 2025 · similarity 0.86)
- The Signal is in the Steps: Local Scoring for Reasoning Data Selection (ICML 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
