Measuring What Matters: Construct Validity in Large Language Model Benchmarks

A very large review of AI benchmarks that reveals a worrying trend in their effectiveness and scientific rigor

Also Register picked it:

https://www.theregister.com/2025/11/07/measuring_ai_models_h...