Cultural attitudes and social norms evolve, but most benchmarks for large language models (LLMs) treat culture as a fixed attribute of a country or region. A new preprint on arXiv introduces ChronoCulture, a benchmark designed to address this gap by evaluating cultural awareness across time.

The benchmark consists of 10,360 items, according to the abstract. While the preprint does not yet provide results or detailed methodology, its stated purpose is to test whether LLMs can track how cultural attitudes and social norms change over time, rather than assuming they remain static.

ChronoCulture highlights a growing recognition that temporal context matters for cultural understanding in AI systems. The abstract does not specify which models were evaluated or how they performed, but the benchmark itself offers a new direction for assessing LLM cultural competence.