Research Summary: INSIDE the Student's Mind: Jointly Modeling Latent Reasoning and Action in LLM Student Simulators

A new framework, INSIDE, enhances Large Language Model (LLM)-based simulators by enabling them to model not only observable actions but also the underlying reasoning processes. This development, particularly relevant for educational applications, addresses a critical gap where prior simulators failed to capture the diverse motivations behind identical student outputs. By fine-tuning LLMs with paired 'think traces' and actions, grounded in Bloom's Taxonomy, the framework provides a more comprehensive and nuanced understanding of simulated student behavior.

Why it matters

This advancement in LLM simulation represents a significant step towards more sophisticated and accurate modeling of human cognitive processes. It has the potential to enhance the efficacy of complex adaptive systems, particularly in domains requiring deep understanding of user or stakeholder behavior beyond mere observable outputs, thereby improving system design and evaluation methodologies.

Key insights

• Traditional LLM-based simulators often reproduce actions without accurately capturing the underlying reasoning.
• This limitation is particularly significant in educational settings where student simulation is used for evaluating tutoring systems.
• The INSIDE framework fine-tunes LLMs to simulate both student actions and their internal thought processes.
• INSIDE generates 'internal dialogue' grounded in Bloom's Taxonomy, spanning cognitive, affective, and action dimensions.
• The methodology involves fine-tuning models using paired data of 'think traces' and corresponding actions.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.10492