Same question, different history: language, national identity, and credit in large language models

The article presents an analysis of eleven large language models (LLMs) in relation to historical credit for inventions across different languages, revealing that query language significantly influences which claimants are recognized. The study evaluated 75,896 responses to 21 disputed inventions and found that lower-status claimants are more likely to be acknowledged in their native languages, while dominant Anglophone figures remain consistent. This highlights the role of LLMs as systems of cultural memory, where language shapes historical narratives, which is crucial for practitioners to understand the implications of bias and representation in AI outputs.