Knowledge without Wisdom: Measuring Misalignment between LLMs and Intended Impact

Michael Hardy, Yunsung Kim

LLMs increasingly excel on AI benchmarks, but doing so does not guarantee validity for downstream tasks. This study contrasts LLM alignment on benchmarks, downstream tasks, and, im…