Efficientmulti-promptevaluationofLLMs
–Neural Information Processing Systems
Most popular benchmarks for comparing LLMs rely on alimited set ofprompt templates, which may not fully capture the LLMs' abilities and can affect the reproducibility ofresults onleaderboards. Manyrecent worksempirically verify prompt sensitivity and advocate for changes in LLM evaluation.
Neural Information Processing Systems
Feb-9-2026, 19:31:56 GMT
- Country:
- Europe > Denmark > Capital Region > Copenhagen (0.04)
- Genre:
- Research Report > New Finding (0.46)
- Industry:
- Education (0.46)
- Technology: