CopeNLU

Our paper on measuring epistemic diversity in LLMs is accepted to #EMNLP2026! We find that diversity has improved, though LLM output is much less diverse than Web search, especially for non-English. Paper: arxiv.org/pdf/2510.04226 Code/data: github.com/dwright37/ll... #NLProc …

arxiv.org View on Bluesky →

Our paper on measuring epistemic diversity in LLMs is accepted to #EMNLP2026! We find that diversity has improved, though LLM output is much less diverse than Web search, especially for non-English. Paper: arxiv.org/pdf/2510.04226 Code/data: github.com/dwright37/ll... #NLProc …

GitHub - dwright37/llm-knowledge: A package for extracting knowledge and measuring epistemic diversity in LLMs. Associated with the paper "Epistemic Diversity and Knowledge Collapse in Large Language Models" github.com

AI Weekly's analysis →

• A study of 27 LLMs across 155 topics and 12 countries found every system was less epistemically diverse than a basic web search baseline.
• Larger models were counterintuitively less diverse than smaller ones, while retrieval-augmented generation improved diversity across the systems tested.
• For country-specific topics, LLM parametric knowledge systematically reflected English sources over local-language ones, the authors report.

Read full analysis →

View on Bluesky →

This paper addresses how LLMs encode their own knowledge and external context when explaining different knowledge-interaction scenarios, and how these interaction dynamics change over longer sequences in Natural Language Explanation, CoT reasoning, and hallucination generation.

arxiv.org View on Bluesky →

📄 Read the full preprint on arXiv: arxiv.org/abs/2605.12515 Sincere thanks to my supervisors Anna Korhonen & @iaugenstein.bsky.social 🤗 #NLP #LLMs #Interpretability #MultilingualNLP @cambridgeltl.bsky.social @copenlu.bsky.social

Mitigating Cross-Lingual Cultural Inconsistencies in LLMs via Consensus-Driven Preference Optimisation arxiv.org View on Bluesky →