📄 中文摘要
研究表明，询问25个不同的语言模型关于时间的隐喻时，得到的1250个响应几乎完全集中在两个隐喻上：'时间是一条河'和'时间是一位织工'。这一发现来自于《人工蜂群》一文，该文被华盛顿大学、卡内基梅隆大学、斯坦福大学和AI2的研究人员在NeurIPS 2025上作为口头报告接受。这一结果表明，尽管模型由不同公司构建，训练于不同数据，采用不同架构，但它们在隐喻表达上却趋同，反映出当前大型语言模型在创造性表达方面的局限性。
NeurIPS 2025 证明了：每个大型语言模型都说了相同的话——解决方案在这里
出处: NeurIPS 2025 Proved It: Every LLM Says the Same Thing — Here's the Fix
发布: 2026年3月19日
📄 English Summary
NeurIPS 2025 Proved It: Every LLM Says the Same Thing — Here's the Fix
The research reveals that when asking 25 different language models for metaphors about time, the resulting 1,250 responses converge almost entirely on two metaphors: 'time is a river' and 'time is a weaver.' This finding comes from the paper titled 'Artificial Hivemind,' which was accepted as an oral presentation at NeurIPS 2025 by researchers from the University of Washington, CMU, Stanford, and AI2. This result indicates that despite being built by different companies, trained on diverse datasets, and utilizing various architectures, these models exhibit a convergence in metaphorical expression, highlighting the limitations of current large language models in creative articulation.