Story
arxiv_llm_reliability · Aug 28, 2026 · paper
Source brief
Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge
arxiv.orgAug 28, 2026
original source linked
In brief
Factual question answering (QA) typically assumes a single canonical answer, obscuring whether large language models (LLMs) retain divergent accounts of long-tail facts. To address this gap, we introduce ElephantBench...
Feed lens
eval