研究人员制定了一项正式定律，用于量化使用多样化的大型语言模型（LLM）集成所获得的性能提升。该定律将集成提升分解为“挽救”和“损害”两个组成部分，并提供了一种基于准确率调整的正确率相关性（$\phi_{\mathrm{adj}}$）等指标来预测性能的启发式方法。所提出的启发式方法在十个开源模型和多个基准测试的超过767,000次推理中进行了测试，显示出强大的预测能力，并能有效地迁移到未见过的数据集。 AI
影响 通过理解和利用多样性，为优化LLM集成性能提供了框架。
排序理由 该集群包含两篇相同的arXiv论文，详细介绍了新的研究发现和方法论。
在 arXiv cs.MA (Multiagent) 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2607.17384v1 Announce Type: new Abstract: This paper provides an experimentally verified formal law for calculating the uplift that diversity of thought provides in Large Language Model (LLM) ensembles. From first principles, we derive an exact decomposition of LLM ensemble…
This paper provides an experimentally verified formal law for calculating the uplift that diversity of thought provides in Large Language Model (LLM) ensembles. From first principles, we derive an exact decomposition of LLM ensemble lift into rescue and damage masses, which yield…
This paper provides an experimentally verified formal law for calculating the uplift that diversity of thought provides in Large Language Model (LLM) ensembles. From first principles, we derive an exact decomposition of LLM ensemble lift into rescue and damage masses, which yield…