Original title: Accurate in space, unreliable in time: how LLMs represent national cultural change
Yalda Daryani, Miranda Bogen, Madeleine I. G. Daepp · September 2026
Most tests of artificial intelligence check if models understand a culture right now. But human cultures do not stay frozen; their values change in different directions and speeds over time. The researchers tested if large language models understand how societies change across time, not just where they stand today. They used over two decades of global survey data from 40 countries and compared real historical shifts to the answers given by four top models. They found that models struggle with history. The models place countries close to recent positions, but they lag several years behind. They catch only part of the real size of cultural shifts. They also invent movement where little happened, and they rarely show when a country reversed its direction. Testing AI on a single moment hides these mistakes.
You have a friend you haven't seen in years. You remember them as they were, but they have changed—new hobbies, different opinions. If you only rely on that old memory, you might misunderstand them today. Here, the researchers show that language models treat countries the same way: they remember the past well but miss how cultures evolve over time.
If LLMs are used to assess cultural alignment, their temporal lag could cause them to reinforce outdated stereotypes or miss recent shifts. This matters for anyone deploying these models in culturally sensitive contexts, from customer service to public policy. It also challenges the current practice of evaluating cultural awareness with a single snapshot, which can hide these temporal gaps.
The abstract doesn't explicitly list limitations, but the study focuses on 40 countries and four specific LLMs, so results may not generalize to other models or regions. It uses only the Inglehart-Welzel cultural map and does not examine other dimensions of culture. The paper identifies the problem but does not propose solutions to fix the temporal lag.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Accurate in space, unreliable in time: how LLMs represent national cultural change — Yalda Daryani, Miranda Bogen, Madeleine I. G. Daepp
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.