LLM Microscope: What Model Internals Reveal About Answer Correctness and Context Utilization

Jiarui Liu, Jivitesh Jain, Mona Diab +1

Although large language models (LLMs) have tremendous utility, trustworthiness is still a chief concern: models often generate incorrect information with high confidence. While con…