Sherpa: Teaching LLMs to Teach Adaptively
Sherpa trains LLM teachers with reinforcement learning to adapt instruction to simulated student archetypes.
Sherpa is a multi-turn reinforcement learning framework that trains a teacher language model against several LLM student archetypes with distinct learning preferences. The teacher is optimized directly for those students' learning outcomes rather than fixed pedagogical rules. Trained teachers raise student performance across all archetypes by an average of 20.5 percentage points, and MathTutorBench's overall pedagogy score moves from 52.5% to 79.2%. Human raters preferred the trained teacher over the base model in 79.6% of pairwise comparisons.
- Sherpa optimizes a teacher LLM against simulated students with different preferences.
- Instructed students improve by 20.5 percentage points on average across archetypes.
- MathTutorBench pedagogy score rises from 52.5% to 79.2%.
- Humans prefer the trained teacher in 79.6% of pairwise comparisons.
- Products
- MathTutorBench
- AI models
- Sherpa
From the original article
1 min read at arxiv.orgLarge language models (LLMs) have become increasingly capable problem solvers, but being able to solve a problem is not the same as being able to teach it. Existing approaches to training LLMs as teachers rely on demonstrations, preference data, or predefined pedagogical criteria that specify what good teaching looks like.
By Weixian Xu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.