Reproduce “A Survey of Large Language Models Attribution” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Dongfang Li, Zetian Sun, Xinshuo Hu, Zhenyu Liu, Ziyang Chen, Baotian Hu, Aiguo Wu, Min Zhang
This survey from Harbin Institute of Technology (Shenzhen) systematically reviews the field of Large Language Model attribution, categorizing existing approaches, datasets, and evaluation methods. It highlights that attribution is a pragmatic solution to mitigate LLM hallucinations and enhance the verifiability and trustworthiness of generated content.
Key insights
- Attribution offers a pragmatic approach to reducing LLM hallucinations by focusing on verifiability, bypassing the direct determination of truthfulness.
- While LLMs can sometimes self-generate attributions, these are often incorrect or entirely hallucinated, indicating a need for more robust, external grounding mechanisms.
- A systematic taxonomy is crucial for structuring the nascent field of LLM attribution, identifying common errors (granularity, mistaken synthesis, hallucinated generation), and guiding future research.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.