Toward a Benchmark for Controllable Simulation of Imperfect Students with Large Language Models
Omri Sason · Alexander Apartsin · Yehudit Aperstein
Can a language model be steered to retain some skills while suppressing others? We introduce a benchmark-oriented framework for controllable learner simulation, representing a student as an explicit skill vector and evaluating selective partial mastery in a structured mathematics setting.
Teacher education requires deliberate practice with learners who exhibit identifiable strengths, weaknesses, and partial mastery. This project investigates whether prompted language models can simulate such students in a controllable, measurable way.
The framework represents a simulated student as an explicit skill vector. Prompt-based control specifies which competencies are retained and which are suppressed. Results show that selective partial mastery can be induced and measured in a structured mathematics setting, though the degree of controllability remains model-dependent.
ImperfectStudent/
paper_chapters/ Paper source (Markdown per section)
figures/ Figures used in the paper
paper_v5_final.html Self-contained paper (HTML)
reviews/ Reviewer notes
@article{sason2026imperfectstudent,
title = {Toward a Benchmark for Controllable Simulation of Imperfect Students
with Large Language Models},
author = {Sason, Omri and Apartsin, Alexander and Aperstein, Yehudit},
year = {2026},
url = {https://apartsinprojects.github.io/ImperfectStudent/}
}