How Evaluation Guides AI Research: The Message Still Counts More than the Medium