Hide and Seek with LLMs: An Adversarial Game for Sneaky Error Generation and Self-Improving Diagnosis

2025-08-05

AI Summary

Large language models (LLMs) exhibit limited capability in diagnosing deep, subtle errors—particularly in mathematical reasoning—due to training objectives prioritizing correct outputs over explicit exposure to and learning from complex, concealed mistakes. To address this, we propose HSG, an adversarial framework featuring co-evolving Sneaky (stealthy error generation) and Diagnosis (adaptive error identification) agents. HSG dynamically synthesizes highly deceptive errors via reasoning-chain perturbation, inter-LLM adversarial interaction, and self-feedback learning, while simultaneously strengthening diagnostic robustness. Evaluated on multiple mathematical reasoning benchmarks, HSG achieves a 16.8%–31.4% absolute improvement in diagnosis accuracy over GPT-4o, significantly enhancing fine-grained error understanding and localization. Furthermore, we release the first high-quality, expert-annotated mathematical diagnosis dataset containing stealthy, semantically plausible erroneous reasoning traces.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Language Models (LLMs) excel in reasoning and generation across domains, but still struggle with identifying and diagnosing complex errors. This stems mainly from training objectives that prioritize correct answers, limiting exposure to and learning from errors. While recent studies have begun to address this by introducing error signals, most rely on shallow, static errors, restricting improvement in deep diagnostic ability. To overcome this, we propose Hide and Seek Game (HSG), a dynamic adversarial framework for error generation and diagnosis, and evaluate it on mathematical problem-solving. HSG involves two adversarial roles: Sneaky, which "hides" by generating subtle, deceptive reasoning errors, and Diagnosis, which "seeks" to accurately detect them. Through adversarial co-evolution, both error stealth and diagnostic precision are enhanced. Experiments on several math reasoning tasks show that HSG significantly boosts error diagnosis, achieving 16.8%--31.4% higher accuracy than baselines like GPT-4o. We also release a challenging dataset of deceptive errors and diagnostic annotations as a benchmark for future research.

Problem

• LLMs struggle with identifying complex reasoning errors
• Training lacks exposure to subtle, deceptive errors
• Need dynamic adversarial framework for error diagnosis

Innovation

• Adversarial framework for error generation
• Dynamic co-evolution of error stealth
• Boosts diagnostic precision significantly