AI Paper Atlas

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

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations

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 changedWhereMeasuredEffect
random intervention with a larger teacher's explanationsFlan-T5-Large student · StrategyQAaccuracy: 58.51 -> 62.96Improved
random intervention with a smaller teacher's explanationsFlan-T5-XL student · StrategyQAaccuracy: 68.12 -> 61.86Hurt
human teacher explanations at 100% interventionFlan-T5-Large student · StrategyQAaccuracy: 58.51 -> 81.95Improved
stronger LLaMA-65B teacher explanationsLLaMA-7B student · StrategyQAaccuracy: 61.13 -> 69.43Improved
weaker teacher explanations for stronger studentLLaMA-65B student · StrategyQAaccuracy: 77.58 -> 69.14Hurt
human explanations for already strong studentFlan-T5-XL student · CommonsenseQAaccuracy: 92.38 -> 92.21No clear effect
Flan-T5-XL teacher explanationsFlan-T5-Large student · CommonsenseQAaccuracy: 84.78 -> 89.72Improved
LLaMA-13B teacher explanationsLLaMA-7B student · GSM8kaccuracy: 9.62 -> 12.25Improved
Expected Utility intervention instead of random at low budgetFlan-T5-XL teacher, Flan-T5-Large student · StrategyQAaccuracy: up to 7 points at 20% budgetImproved
increase intervention beyond 60% with Expected UtilityFlan-T5-XL teacher, Flan-T5-Large student · StrategyQAaccuracy: 69% -> 63%Hurt
rank by true utility using student confidenceExpected Utility teacher · StrategyQAaccuracy: 76% at 20% cost; 81% overallImproved
Expected Utility computed with teacher answers instead of gold labelsFlan-T5-XL teacher · StrategyQAaccuracy: up to 2 points at 20% budgetImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy