article
Holistic Evaluation of Language Models
Bommasani, Rishi, Percy Liang, and Tony Lee. "Holistic Evaluation of Language Models." Annals of the New York Academy of Sciences 1525, no. 1 (2023): 140–146. https://doi.org/10.1111/nyas.15007.
Why this matters here
Bommasani, Liang, and Lee’s Holistic Evaluation of Language Models (HELM) argue for multi-metric benchmarking and document limits of context-free model comparison across scenarios and metrics. Coupling uses that evaluation stance to insist benchmarks and model cards help procurement but do not replace testing on one’s own data, tools, and failure history. The paper establishes multi-metric evaluation practice; the book asks whether evaluation reconnects systems to consequence before production finds the failure first.
