To me the weights of a large language model absolutely seem like aggregate data. The weights are numbers derived from the input data as a whole - in aggregate

To me the weights of a large language model absolutely seem like aggregate data. The weights are numbers derived from the input data as a whole - in aggregate - and not from individual bits of code.