What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models

A first systematic study of epistemic diversity in large language models, testing 27 LLMs on 155 topics across 12 countries and generating 1.7M responses and 70M individual claims, found that epistemic diversity has increased substantially over the past three years but that every system remains less diverse than a search baseline, according to the arXiv paper "What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models" (arXiv:2510.04226v7, revised 31 Aug 2026). The study reports that retrieval-augmented generation can improve diversity, that larger models are counterintuitively less diverse than smaller ones, and that LLM parametric knowledge systematically reflects English over local-language knowledge for country-specific topics.

Computer Science Computation and Language Submitted on 5 Oct 2025 v1 https://arxiv.org/abs/2510.04226v1 , last revised 31 Aug 2026 this version, v7 Title:What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models View PDF https://arxiv.org/pdf/2510.04226 HTML experimental https://arxiv.org/html/2510.04226v7 Abstract:Large language models LLMs are increasingly used as primary knowledge sources, yet their epistemic diversity - defined as the diversity of real-world claims in their outputs - has never been measured. Low epistemic diversity would pose a risk of knowledge collapse as homogeneous LLMs mediate a shrinking in the range of accessible information over time. The dominant paradigm is that overall LLM diversity is low, but this is always with respect to a single point in time, with no reference baseline or consideration for variation across countries. We address this gap in knowledge by performing the first systematic study of epistemic diversity in LLMs across time and cultural context, testing 27 LLMs on 155 topics covering 12 countries, resulting in 1.7M responses and 70M individual claims. We find that epistemic diversity has increased substantially over the past three years, a positive counter to recent diversity pessimism. However, despite progress, we find that every system is less diverse than a search baseline. This gap is not uniform: RAG can improve diversity, while large models are counterintuitively less diverse than smaller ones. Moreover, LLM parametric knowledge systematically reflects English over local-language knowledge for country specific topics. Together, these results demonstrate that while progress on epistemic diversity is tangible, it is insufficient and unevenly distributed. Submission history From: Dustin Wright view email https://arxiv.org/show-email/0c2d6b8d/2510.04226 Sun, 5 Oct 2025 14:29:15 UTC 305 KB \ v1\ https://arxiv.org/abs/2510.04226v1 Tue, 7 Oct 2025 16:07:31 UTC 807 KB \ v2\ https://arxiv.org/abs/2510.04226v2 Wed, 8 Oct 2025 07:35:57 UTC 807 KB \ v3\ https://arxiv.org/abs/2510.04226v3 Thu, 30 Oct 2025 14:52:48 UTC 807 KB \ v4\ https://arxiv.org/abs/2510.04226v4 Tue, 11 Nov 2025 18:13:57 UTC 820 KB \ v5\ https://arxiv.org/abs/2510.04226v5 Wed, 28 Jan 2026 13:27:36 UTC 826 KB \ v6\ https://arxiv.org/abs/2510.04226v6 v7 Mon, 31 Aug 2026 12:37:31 UTC 226 KB Current browse context: cs.CL References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs https://info.arxiv.org/labs/index.html .