Researchers at Stanford University and the University of Washington have created a dataset of responses to Reddit posts that received more support than earlier responses. The dataset contains 385,000 posts in 18 domains, such as culinary or legal advice, that contain either a question or instructions inviting helpful responses. It also contains a pair of responses to each post, where one response received more support than the other despite occurring at a later point in time. Because the later response received more support than the earlier response, despite having less visibility, researchers can infer that the forum community preferred the later response. Researchers can use the dataset to train models to predict which responses people will prefer to questions or instructions.