Listen to the Layers: Mitigating Hallucinations with Inter-Layer Disagreement

Koduvayur Subbalakshmi, Sabbir Hossain Ujjal, Venkata Krishna Teja Mangichetty +1

Pretrained Large Language Models (LLMs) are prone to generating fluent yet factually incorrect text-a phenomenon known as hallucinations, undermining their reliability and utility…