#factual consistency

Learn about the Hughes Hallucination Evaluation Model (HHEM-2.1) and its role in measuring hallucination rates in large language models. This regularly updated leaderboard offers insights into factual consistency and summarization accuracy, presenting detailed rankings and methodologies for performance evaluation. A resourceful tool for those interested in advances in reducing hallucinations in LLMs and improving factual summaries. Comprehensive datasets and prior research are also available for further exploration of LLM evaluation.

This survey examines the issue of hallucination in large language models, focusing on how generated content can diverge from user inputs or factual information. It reviews evaluation methods, explanation frameworks, and mitigation strategies, addressing input-conflicting, context-conflicting, and fact-conflicting hallucinations. The article highlights studies and benchmarks across domains such as machine translation and summarization, providing insights into research that seeks to enhance the factual reliability of these models.