HillZhang1999/llm-hallucination-survey

Organizes hallucination research across three taxonomies—input-conflicting, context-conflicting, and fact-conflicting—with curated papers covering evaluation metrics, root causes, and mitigation strategies. Covers diverse NLG tasks including machine translation, summarization, dialogue, and QA, providing systematic categorization of how models generate plausible but incorrect outputs. Functions as a structured bibliography linking to arxiv papers and benchmarks rather than implementing evaluation tools directly.

1,078 stars. No commits in the last 6 months.