Consider two scenarios:

Scenario A: A team is building a system to classify legal documents into one of 500 highly specific, proprietary categories. They have a massive, well-labeled dataset of 10 million documents.

Scenario B: A team is building a chatbot to answer questions about a new software product. They have a small, curated dataset of only 500 question-and-answer pairs.

Which scenario would benefit more from starting with a large, general-purpose pre-trained model? Justify your answer by explaining the relationship between the amount of available task-specific data and the primary advantage of the pre-training approach.