English(EN) Where's Waldo? Query-language Preference under Cross-lingual Knowledge Disparities

新的基准测试Waldo揭示LLM偏好具有冲突信息的查询语言

作者 PulseAugur 编辑部 · [1 个来源] · 2026-10-02 04:00

研究人员开发了一个名为Waldo的新基准测试，用于研究大型语言模型（LLM）如何处理跨语言知识差异。该基准测试由维基百科构建，包含12,000个问答对，这些问答对要么突显一种语言中信息的缺失，要么突显跨语言信息的冲突。当事实缺失时，LLM通常会利用可用语言的证据，但当事实冲突时，模型倾向于偏好查询语言中的来源，导致根据用户的语言得出不同的答案。研究还探讨了缓解策略，包括消融注意力头和使用基于LoRA的训练，这些策略将查询语言偏好差距减少了高达61.5%。 AI

影响突显了LLM中可能存在的偏见，这些偏见可能会影响不同语言的信息访问和准确性。

排序理由该集群包含一篇详细介绍新基准测试和研究结果的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 13 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍新基准测试和研究结果的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, safety; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Dayeon Ki, Ruochen Zhang, Silviu Cucerzan, Ryen W. White, Ning Gao · 2026-10-02 04:00

沃尔도在哪里？跨语言知识差异下的查询语言偏好

arXiv:2610.00606v1 Announce Type: cross Abstract: Large Language Models increasingly serve as interfaces for knowledge-intensive information seeking tasks across languages by synthesizing multilingual evidence. Prior work has shown that they often exhibit query-language preferenc…

本文如何被排名

报道来源 [1]

沃尔도在哪里？跨语言知识差异下的查询语言偏好

相关实体

相关话题