A team is developing a system to learn from human feedback, where annotators provide their preferences over several generated text responses. The team is considering two different strategies for training a model to predict these preferences: a 'pairwise' approach that compares two responses at a time (e.g., 'Response A is better than Response B'), and a 'listwise' approach that considers an entire ranked list of responses simultaneously (e.g., 'Response A is best, followed by C, then B'). Analyze the fundamental differences between these two strategies. In your analysis, discuss the potential advantages and disadvantages of each approach in terms of data requirements, computational complexity, and the granularity of the preference information they can capture.