The Missing Primitive: Diagnosing and Repairing Mathematical Reasoning in Large Language Models
While Large Language Models (LLMs) have demonstrated striking capabilities on frontier mathematical problems, it remains unclear whether they possess the structural mathematical understanding underlying their solutions. In this paper, we take a first step toward systematically studying mathematical understanding in LLMs, from diagnosing its distinct capabilities to leveraging t
Reading in Plain English mode — toggle below to switch
While Large Language Models (LLMs) have demonstrated striking capabilities on frontier mathematical problems, it remains unclear whether they possess the structural mathematical understanding underlying their solutions. In this paper, we take a first step toward systematically studying mathematical understanding in LLMs, from diagnosing its distinct capabilities to leveraging t
This paper introduces a novel approach to systematically evaluate and improve mathematical reasoning in LLMs, moving beyond simple accuracy metrics. Understanding how to diagnose and repair specific reasoning capabilities is crucial for developing more robust and reliable AI systems, especially in domains requiring precise mathematical understanding.