Inheritance as an Admission Test for LLM Prompt Merge Operators

📄 arXiv: 2610.29023v1 📥 PDF

作者: Zan-Kai Chong, Hiroyuki Ohsaki, Bryan Ng

分类: cs.AI

发布日期: 2026-10-21

💡 一句话要点

提出基于继承的LLM合并操作的入学测试方法

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 大型语言模型 合并操作 性能评估 入学测试 自然语言处理 智能代理

📋 核心要点

  1. 现有方法在验证器饱和后无法继续提升LLM代理的性能，导致训练效率低下。
  2. 本文提出通过比较当前种群的性能来评估代理，并引入合并操作的入学测试机制。
  3. 实验显示，某些合并操作在开放循环中表现优异，而自由重写操作未能超越初代成员的表现。

📝 摘要（中文）

大型语言模型（LLM）代理通常通过迭代直到固定验证器完成来进行改进。一旦验证器饱和，循环便无进一步提升的空间。本文提出了一种新的评分机制，通过与当前种群进行比较，记录是否有更好的代理出现。合并操作的入学测试要求子代在向捐赠者移动时超越零剂量下相同操作产生的子代，并且该移动的置信下界达到请求剂量的一半。实验结果表明，某些操作在开放循环中被接受，而在关闭循环中未能超越最佳初代成员。

🔬 方法详解

问题定义：本文旨在解决大型语言模型（LLM）代理在验证器饱和后无法继续提升的问题，现有方法缺乏有效的性能评估机制。

核心思路：通过引入一种新的入学测试机制，评估合并操作的有效性，确保子代在性能上超越零剂量下的同类操作。

技术框架：整体流程包括性能评分、合并操作的入学测试和结果记录。主要模块包括当前种群的性能评估、合并操作的执行和结果的比较。

关键创新：最重要的创新在于引入了基于性能比较的入学测试机制，使得合并操作的有效性得以量化，区别于传统的固定验证器方法。

关键设计：在实验中，使用了gemini-3.1-flash-lite进行性能测量，设置了置信下界和剂量请求等关键参数，以确保合并操作的有效性和可靠性。实验中还对不同操作的表现进行了详细记录和分析。

📊 实验亮点

实验结果表明，所有策略附加和模板附加的运行均超越了最佳初代成员，而自由重写操作未能达到相同的效果。开放循环中的约束重写操作被接受，但在关闭循环中未进行测试，显示出不同操作在不同环境下的表现差异。

🎯 应用场景

该研究的潜在应用领域包括自然语言处理、对话系统和智能助手等。通过优化LLM的合并操作，可以提高模型的训练效率和性能，推动智能系统的进一步发展。

📄 摘要（原文）

Large language model (LLM) agents are often improved by iterating until a fixed verifier is done. Once that verifier saturates, the loop has nothing left to climb. We instead score agents against the current population, and a frozen panel records whether anyone got better. The panel mean can rise inside the seeds by elimination alone. We therefore count a run as training only when its final mean beats the best generation-zero member. A merge is admitted only when the child moves toward the donor, past the child the same operator produces at zero dose, and the lower confidence bound on that movement reaches half the requested dose. That is the admission test. A detectable shortfall rejects the operator, and the test is otherwise undetermined. We keep realised fitness φas the selector and reference performance θas the unused ruler. Every measurement uses gemini-3.1-flash-lite on one negotiation task, and no fixed-oracle arm was run. In the open loop, tactic append grafts short donor rules and is admitted. Free rewrite moves only a third of the requested dose and is rejected. Template append pastes raw donor replies into the child. Its movement clears the bound on the first probe but not on the re-probe, and it is left undetermined. A held-out re-probe on the five situations the merge never saw confirms that both appends move, and it does not rescue free rewrite. Neither append rises with the dose. Constrained rewrite is admitted in the open loop and was not run in the closed loop, which changes only the operator. The population, the scores and the task stay fixed. No free-rewrite run beats the best generation-zero member. Every tactic-append run and every template-append run does. Sampled loop children keep that split. Free rewrite moves no further toward the donor, and both appends still clear the bound, by less than in the open loop.