The Hallucinations Leaderboard, an Open Effort to Measure Hallucinations in Large Language Models

Hugging Face · 29 January 2024, 00:00

Researchers created the Hallucinations Leaderboard, an open platform that evaluates large language models on their tendency to generate false or inaccurate content through a series of standardized benchmarks. The leaderboard tests models across 12 categories including factual question-answering, summarization, reading comprehension, and hallucination detection tasks, with scores normalized on a 0–1 scale. The platform enables researchers and developers to compare model reliability and identify which models are most prone to producing misinformation or contradicting user instructions.