一篇新的研究论文探讨了多模态大语言模型（MLLMs）生成美学评论的能力，并将其输出与人类评估进行比较。该研究利用Reddit照片评论数据集，在各种提示条件下评估了五个开放权重MLLMs。研究结果表明，虽然MLLMs可以生成全面的评论，但其风格与人类评估存在显著差异，通常更冗长、选择性更差，这对当前的评估指标提出了挑战。 AI
影响 凸显了当前MLLM在开放式生成评估方法上的局限性，表明需要新的训练和评估策略。
排序理由 在arXiv上发表的研究论文，评估MLLM能力。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2606.29689v1 Announce Type: new Abstract: Open-ended aesthetic critique is a challenge for multimodal large language models (MLLMs): unlike multiple-choice aesthetic benchmarks, it has no single correct answer, and most aesthetic evaluation has measured models against numer…
Open-ended aesthetic critique is a challenge for multimodal large language models (MLLMs): unlike multiple-choice aesthetic benchmarks, it has no single correct answer, and most aesthetic evaluation has measured models against numeric scores rather than the written critiques peop…