PulseAugur
EN
LIVE 09:03:15

New benchmark Waldo reveals LLMs favor query language with conflicting info

By PulseAugur Editorial · [1 sources] ·

Researchers have developed a new benchmark called Waldo to study how large language models (LLMs) handle cross-lingual knowledge disparities. The benchmark, constructed from Wikipedia, includes 12,000 question-answering pairs that highlight either missing information in one language or conflicting information across languages. When facts are missing, LLMs generally use evidence from available languages, but when facts conflict, models tend to favor sources in the query language, leading to different answers based on the user's language. The study also explored mitigation strategies, including ablating attention heads and using LoRA-based training, which reduced the query-language preference gap by up to 61.5%. AI

IMPACT Highlights potential biases in LLMs that could affect information access and accuracy across different languages.

RANK_REASON The cluster contains an academic paper detailing a new benchmark and research findings. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

New benchmark Waldo reveals LLMs favor query language with conflicting info

How we ranked this

Signal score
15 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
The cluster contains an academic paper detailing a new benchmark and research findings. [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.

Full methodology in our editorial standards.

COVERAGE [1]

  1. arXiv cs.AI TIER_1 English(EN) · Dayeon Ki, Ruochen Zhang, Silviu Cucerzan, Ryen W. White, Ning Gao ·

    Where's Waldo? Query-language Preference under Cross-lingual Knowledge Disparities

    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…

RELATED ENTITIES

RELATED TOPICS