# 盲人摸象：探测大语言模型在长尾分歧知识下的认知盲区

- 来源：HuggingFace Daily Papers（社区热门论文）
- 发布时间：2026-08-28 08:00
- AIHOT 分数：41
- AIHOT 链接：https://aihot.virxact.com/items/cmtgmdkag0306roch4maqen8r
- 原文链接：https://arxiv.org/abs/2608.28478

## AI 摘要

ElephantBench 闭卷知识探针含 1,094 道基于可审计图流程生成的多方叙述问答。测试 32 个模型，最强模型也仅在 52.4% 的问题上同时召回两种叙述，其余大多只记得一种。扩大模型规模和推理时计算能提升召回率但无法消除不完整性，代码已开源。

## 正文

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, a closed-book knowledge probe comprising 1,094 questions generated through an auditable graph-based pipeline. The pipeline retrieves related documents from a low-exposure web corpus, identifies naturally occurring disagreements, and converts them into multi-account QA records. Each answer is verified against the originating documents and authoritative public web sources and is then reviewed by human annotators. Across 32 models, even the strongest model recovers both accounts on only 52.4% of questions, while on nearly all remaining questions it recalls one account but omits the other. Scaling model size and inference-time reasoning improve recall but do not eliminate this incompleteness. Corpus analysis further shows that exposure imbalance favors the dominant account, whereas greater minority-side exposure is associated with more complete recall. These findings establish ElephantBench as a reproducible knowledge probe for diagnosing epistemic myopia in parametric memory. More broadly, our graph-based benchmark construction pipeline provides an efficient and scalable way to turn long-tail corpora into source-traceable knowledge probes, supporting efforts to evaluate and advance the epistemic rigour of next-generation LLMs. Code is available at https://github.com/Tencent/ElephantBench.