A Method for Enhancing the Safety of Large Model Generation Based on Multi-dimensional Attack and Defense

Keke Zhai

Currently, large models are prone to generating harmful content when faced with complex attack instructions, significantly reducing their defensive capabilities. To address this is…