Lifecycle-Based Survey of LLM Hallucinations

This survey categorizes hallucinations into data-related, training-related, and inference-related stages. It provides a framework for identifying, detecting, and mitigating errors throughout the entire model development lifecycle to improve reliability in high-stakes domains.