What happened
Researchers analyzed what climate information people actually seek from AI systems versus what current AI benchmarks (standardized tests) measure, finding a significant mismatch. This matters because if we're testing AI systems on the wrong questions, we're deploying them to answer climate questions while blind to whether they actually help people make decisions or understand the problem.
Why it matters
As AI becomes the primary way millions of people access climate information, we're discovering that our standard tests for AI accuracy don't reflect real-world usefulness — meaning an AI could ace its tests while failing to give people what they actually need to understand or act on climate change.