Benchmarking Reasoning Robustness in Large Language Models

2025-03-06

AI Summary

Large language models (LLMs) exhibit insufficient reasoning robustness, often relying on memorized patterns rather than systematic logical inference. This work identifies four fundamental bottlenecks: positional bias, instruction sensitivity, numerical fragility, and memory dependency. Method: We introduce Math-RoB, the first benchmark explicitly designed to evaluate reasoning robustness. It employs instruction-driven synthesis of diverse mathematical problems with intentional information omissions, augmented via multi-dimensional perturbations—including positional shuffling, instruction fine-tuning, numeric substitution, and critical information masking—to systematically induce hallucinations and expose reasoning flaws. Contribution/Results: Evaluated on state-of-the-art models including GPT-4o, Qwen2.5, and DeepSeek-V3, Math-RoB reveals up to 15 percentage points of performance degradation under perturbation, effectively distinguishing genuine reasoning capability from spurious memorization-based fitting.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Despite the recent success of large language models (LLMs) in reasoning such as DeepSeek, we for the first time identify a key dilemma in reasoning robustness and generalization: significant performance degradation on novel or incomplete data, suggesting a reliance on memorized patterns rather than systematic reasoning. Our closer examination reveals four key unique limitations underlying this issue:(1) Positional bias--models favor earlier queries in multi-query inputs but answering the wrong one in the latter (e.g., GPT-4o's accuracy drops from 75.8 percent to 72.8 percent); (2) Instruction sensitivity--performance declines by 5.0 to 7.5 percent in the Qwen2.5 Series and by 5.0 percent in DeepSeek-V3 with auxiliary guidance; (3) Numerical fragility--value substitution sharply reduces accuracy (e.g., GPT-4o drops from 97.5 percent to 82.5 percent, GPT-o1-mini drops from 97.5 percent to 92.5 percent); and (4) Memory dependence--models resort to guesswork when missing critical data. These findings further highlight the reliance on heuristic recall over rigorous logical inference, demonstrating challenges in reasoning robustness. To comprehensively investigate these robustness challenges, this paper introduces a novel benchmark, termed as Math-RoB, that exploits hallucinations triggered by missing information to expose reasoning gaps. This is achieved by an instruction-based approach to generate diverse datasets that closely resemble training distributions, facilitating a holistic robustness assessment and advancing the development of more robust reasoning frameworks. Bad character(s) in field Abstract.

Problem

• Identifies performance degradation in LLMs on novel data
• Highlights four key limitations affecting reasoning robustness
• Introduces Math-RoB benchmark to assess reasoning gaps

Innovation

• Introduces Math-RoB benchmark for robustness testing
• Uses instruction-based approach for dataset generation
• Assesses reasoning gaps via missing information hallucinations