arXiv:2606.05704v1 Announce Type: cross 
Abstract: Recent Large Language Models (LLMs) have shown impressive reasoning abilities; but they are still susceptible to hallucinations, intermediate reasoning mistakes, and unreliable reasoning results in complex mathematical reasoning problems. In this study, we introduce a critic-based heterogeneous multi-agent approach to improve the dependability of mathematical reasoning. This framework incorporates several LLM agents of different specialties and employs a critic-driven adaptive learning system to assess and guide the reasoning process based on intermediate feedback. The system adopts a generator-validator framework, with the validator not only determining correctness but also offering critiques to guide regeneration of solutions. This allows for adaptive error correction and prevents error cascading. Our experiments on the GSM8K benchmark show that the proposed method achieves up to 13% accuracy improvement over single-shot and non-critic models. Additionally, findings suggest that heterogeneity and critique reduce the need for large models, allowing smaller models to perform on par. Ablation studies reveal the main performance gains are due to the critic-based feedback loop and not model size. In summary, the proposed approach showcases the benefits of combining heterogeneous multi-agent collaboration and critique to obtain reliable and interpretable reasoning systems.

أدت التطورات الأخيرة في نماذج اللغة الكبيرة (LLMs) إلى تقديم إطار عمل هجين متعدد الوكلاء مدعوم بنظام نقدي، يهدف إلى تحسين موثوقية حل المشكلات الرياضية. تستخدم هذه الطريقة عدة وكلاء LLM ذوي مهارات متخصصة ونظام تعلم تكيفي مدفوع بالنقد لتحسين دقة التفكير، محققة زيادة تصل إلى 13% في الأداء على معيار GSM8K.

Los avances recientes en Modelos de Lenguaje de Gran Escala (LLMs) han llevado a la introducción de un marco heterogéneo de múltiples agentes guiado por un crítico, destinado a mejorar la fiabilidad en la resolución de problemas matemáticos. Este enfoque utiliza varios agentes LLM con habilidades especializadas y un sistema de aprendizaje adaptativo impulsado por un crítico para mejorar la precisión del razonamiento, logrando hasta un 13% de aumento en el rendimiento en el benchmark GSM8K.

Des avancées récentes dans les modèles de langage de grande taille (LLMs) ont conduit à l'introduction d'un cadre hétérogène multi-agents guidé par un critique, visant à améliorer la fiabilité de la résolution de problèmes mathématiques. Cette approche utilise plusieurs agents LLM avec des compétences spécialisées et un système d'apprentissage adaptatif dirigé par un critique pour améliorer la précision du raisonnement, atteignant jusqu'à 13 % d'augmentation des performances sur le benchmark GSM8K.

Recent advancements in Large Language Models (LLMs) have led to the introduction of a critic-guided heterogeneous multi-agent framework aimed at enhancing the reliability of mathematical problem-solving. This approach utilizes multiple LLM agents with specialized skills and a critic-driven adaptive learning system to improve reasoning accuracy, achieving up to a 13% increase in performance on the GSM8K benchmark.