The term "out-of-distribution" is a machine learning jargon that describes situations when the data being processed is completely unknown to the system. In these cases, the algorithm struggles to make accurate predictions, as it lacks data to draw from.