This research introduces a novel collaborative learning framework where large and small language models progressively improve each other's fake news detection capabilities.

• Combines LLMs' zero-shot reasoning with SLMs' efficiency
• Uses an iterative knowledge distillation process between models
• Achieves superior performance compared to conventional methods
• Demonstrates adaptability to evolving misinformation tactics

This approach addresses critical security concerns by providing more robust defenses against misinformation campaigns that threaten public discourse and social stability, while requiring less manual labeling of training data.

Collaborative Evolution: Multi-Round Learning Between Large and Small Language Models for Emergent Fake News Detection