研究人员推出了一项名为QUIET的新基准，旨在评估大型语言模型的创意生成能力。与依赖多项选择格式或主观人工评分的现有基准不同，QUIET采用多空白级联故事填空方法，具有明确的内容约束和空白间的依赖关系。这种方法允许基于“校准惊喜”框架进行客观、自动化的评分，该框架奖励符合约束且富有创意的响应。 AI
影响 提供了一种更客观、自动化的方法来评估大型语言模型的创造力，可能推动生成式AI的改进。
排序理由 该集群描述了一篇提出新基准以评估大型语言模型能力的新学术论文。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2605.25955v1 Announce Type: cross Abstract: Large language models (LLMs) face a dual challenge in creative capability evaluation: existing benchmarks (e.g., Story Cloze Test, HellaSwag) measure models' discriminative ability over narrative continuation using multiple-choice…
Large language models (LLMs) face a dual challenge in creative capability evaluation: existing benchmarks (e.g., Story Cloze Test, HellaSwag) measure models' discriminative ability over narrative continuation using multiple-choice recognition paradigms, rather than directly measu…
Large language models (LLMs) face a dual challenge in creative capability evaluation: existing benchmarks (e.g., Story Cloze Test, HellaSwag) measure models' discriminative ability over narrative continuation using multiple-choice recognition paradigms, rather than directly measu…