And the subjectivity is bidirectional.

People judge models on their outputs, but how you like to prompt has a tremendous impact on those outputs and explains why people have wildly different experiences with the same model.