Measuring what Matters: Construct Validity in Large Language Model Benchmarks
The review identifies recurring gaps between constructs, operational tasks and resulting claims, and proposes a practical checklist for benchmark design and interpretation.
Verify byline in the original ↗