LLMs Infer Cultural Context but Fail to Apply It When Responding

In three linesLLMs can infer cultural context but fail to apply it in responses. A new CAPRI dataset shows models recognize cultural conventions (measurement units, time interpretation) but don't spontaneously use them unless explicitly instructed. Biases remain aligned with the model's country of origin.

Summary generated by Claude — human-verified