Measuring what matters: construct validity in large language model benchmarks

Conference item

Evaluating large language models (LLMs) is crucial for both assessing their capabilities and identifying safety or robustness issues prior to deployment. Reliably measuring abstract and complex phenomena such as ‘safety’ and ‘robustness’ requires strong construct validity, that is, having measures that represent what matters to the phenomenon. With a team of 29 expert reviewers, we conduct a systematic review of 445 LLM benchmarks from leading conferences in natural language processing and machine learning. Across the reviewed articles, we find patterns related to the measured phenomena, tasks, and scoring metrics which undermine the validity of the resulting claims. To address these shortcomings, we provide eight key recommendations and detailed actionable guidance to researchers and practitioners in developing LLM benchmarks.

Authors: Bean, A; Kearns, RO; Romanou. A; Hafner, F; Mayne, H

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: NeurIPS
• Publication date: 2025-12-04
• Acceptance date: 2025-09-18
• Event title: 39th Annual Conference on Neural Information Processing Systems (NeurIPS 2025)
• Event location: San Diego, CA, USA and New Mexico, Mexico
• Event website: https://neurips.cc/
• Event start date: 2025-12-02
• Event end date: 2025-12-07
• Language: English