From Problem-Solving to Teaching Problem-Solving: Aligning LLMs with Pedagogy using Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dinucu-Jianu, David, Macina, Jakub, Daheim, Nico, Hakimi, Ido, Gurevych, Iryna, Sachan, Mrinmaya
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912643756851200
author Dinucu-Jianu, David
Macina, Jakub
Daheim, Nico
Hakimi, Ido
Gurevych, Iryna
Sachan, Mrinmaya
author_facet Dinucu-Jianu, David
Macina, Jakub
Daheim, Nico
Hakimi, Ido
Gurevych, Iryna
Sachan, Mrinmaya
contents Large language models (LLMs) can transform education, but their optimization for direct question-answering often undermines effective pedagogy which requires strategically withholding answers. To mitigate this, we propose an online reinforcement learning (RL)-based alignment framework that can quickly adapt LLMs into effective tutors using simulated student-tutor interactions by emphasizing pedagogical quality and guided problem-solving over simply giving away answers. We use our method to train a 7B parameter tutor model without human annotations which reaches similar performance to larger proprietary models like LearnLM. We introduce a controllable reward weighting to balance pedagogical support and student solving accuracy, allowing us to trace the Pareto frontier between these two objectives. Our models better preserve reasoning capabilities than single-turn SFT baselines and can optionally enhance interpretability through thinking tags that expose the model's instructional planning.
format Preprint
id arxiv_https___arxiv_org_abs_2505_15607
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle From Problem-Solving to Teaching Problem-Solving: Aligning LLMs with Pedagogy using Reinforcement Learning
Dinucu-Jianu, David
Macina, Jakub
Daheim, Nico
Hakimi, Ido
Gurevych, Iryna
Sachan, Mrinmaya
Computation and Language
Artificial Intelligence
Large language models (LLMs) can transform education, but their optimization for direct question-answering often undermines effective pedagogy which requires strategically withholding answers. To mitigate this, we propose an online reinforcement learning (RL)-based alignment framework that can quickly adapt LLMs into effective tutors using simulated student-tutor interactions by emphasizing pedagogical quality and guided problem-solving over simply giving away answers. We use our method to train a 7B parameter tutor model without human annotations which reaches similar performance to larger proprietary models like LearnLM. We introduce a controllable reward weighting to balance pedagogical support and student solving accuracy, allowing us to trace the Pareto frontier between these two objectives. Our models better preserve reasoning capabilities than single-turn SFT baselines and can optionally enhance interpretability through thinking tags that expose the model's instructional planning.
title From Problem-Solving to Teaching Problem-Solving: Aligning LLMs with Pedagogy using Reinforcement Learning
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2505.15607