Large language models are increasingly serving as interfaces for knowledge-intensive information seeking, synthesising evidence drawn from multiple languages. A new paper, Where's Waldo? Query-language Preference under Cross-lingual Knowledge Disparities (arXiv:2610.00606), takes up this setting to examine a known behavioural quirk.
Prior work has shown that LLMs often exhibit a query-language preference: the language in which a query is posed tends to shape the model's output, even when multilingual evidence is available. The paper appears to investigate this preference specifically under conditions of cross-lingual knowledge disparities, where relevant information is unevenly distributed across languages.
The released abstract is cut off before presenting the paper's own findings, so this digest can only report the stated motivation and the prior results it builds on. No experimental conclusions are available from the source as provided.