Can LLMs Estimate Student Struggles? Human-AI Difficulty Alignment with Proficiency Simulation for Item Difficulty Prediction
This addresses the cold start problem in educational assessment for educators and learners, but the findings are incremental as they highlight limitations of current models rather than proposing a new solution.
The study investigated whether Large Language Models can accurately estimate item difficulty for educational assessment, finding a systematic misalignment where models converge to a machine consensus rather than aligning with human cognitive struggles, and high performance often impedes accurate estimation.
Accurate estimation of item (question or task) difficulty is critical for educational assessment but suffers from the cold start problem. While Large Language Models demonstrate superhuman problem-solving capabilities, it remains an open question whether they can perceive the cognitive struggles of human learners. In this work, we present a large-scale empirical analysis of Human-AI Difficulty Alignment for over 20 models across diverse domains such as medical knowledge and mathematical reasoning. Our findings reveal a systematic misalignment where scaling up model size is not reliably helpful; instead of aligning with humans, models converge toward a shared machine consensus. We observe that high performance often impedes accurate difficulty estimation, as models struggle to simulate the capability limitations of students even when being explicitly prompted to adopt specific proficiency levels. Furthermore, we identify a critical lack of introspection, as models fail to predict their own limitations. These results suggest that general problem-solving capability does not imply an understanding of human cognitive struggles, highlighting the challenge of using current models for automated difficulty prediction.