no-crosslingual-improvement-with-scale
IN premise — summaries/2026/08/24/xu-2024-knowledge-conflicts-survey-s4-intra-memory-conflict.md
Created 2026-08-25T02:59:03+00:00
Increasing model size does not improve cross-lingual knowledge consistency, as measured by the RankC metric (Qi et al., 2023).
Summary
Making a model larger does not automatically make it more consistent about what it knows across different languages. This means that simply scaling up parameters won't solve the problem of a system giving different answers to the same question depending on the language used, so multilingual consistency requires targeted fixes beyond brute-force scale.