arXiv 2501.15581

Error Classification of Large Language Models on Math Word Problems: A Dynamically Adaptive Framework

By Yuhong Sun, Zhangyue Yin, et al.

Published 2025-01-26

Mindmap

Browse the paper's core ideas, clusters, and relationships in a structured outline.

Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains. Math Word Problems (MWPs) serve as a crucial benchmark for evaluating LLMs' reasoning abilities. While most research primarily focuses on improving accuracy, it often neglects understanding and addressing the underlying patterns of errors. Current error classification methods rely on static and predefined categories, which…

View the original paper on arXiv