Core Concepts
Multimodal large language model SNIFFER detects and explains out-of-context misinformation effectively.
Abstract
SNIFFER is a novel multimodal large language model designed for detecting out-of-context misinformation. It employs two-stage instruction tuning on InstructBLIP, integrating external tools and retrieval methods. The model surpasses state-of-the-art methods in detection accuracy and provides accurate explanations validated by quantitative and human evaluations. Experiments show that SNIFFER can detect misinformation early with limited training data and generalize well across different datasets.
Stats
No metrics or figures provided in the content.
Quotes
No striking quotes found in the content.