This research presents an ensemble framework that effectively reduces hallucinations in large language models by intelligently combining outputs from multiple models while accounting for their uncertainties.

• Leverages the wisdom of crowds approach but enhances it by incorporating uncertainty measurements
• Offers a deployment-friendly solution without requiring additional training data
• Demonstrates significant improvements in factual accuracy compared to traditional ensemble methods
• Provides practical security benefits by reducing harmful misinformation in AI outputs

For security professionals, this framework represents an important advancement in creating more trustworthy AI systems that can be safely deployed in sensitive environments where factual accuracy is critical.

Uncertainty-Aware Fusion: An Ensemble Framework for Mitigating Hallucinations in Large Language Models