---
id: 20260317-T0-24
title: "新方法提升大模型激活控制精度与稳定性"
title_en: "New Method Improves Activation Control Precision in LLMs"
url: https://ai.daily.yangsir.net/daily/20260317-T0-24
issue_date: 2026-03-17
publish_date: 2026-03-16T04:00:00.000Z
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2603.12298
---

# 新方法提升大模型激活控制精度与稳定性

arXiv论文2603.12298提出全局进化转向（GES）方法，通过跨层一致性优化大语言模型的激活转向控制。现有激活转向方法易受高维噪声影响。GES利用不同激活层之间的相关性，构建更稳定的控制向量。实验表明，该方法在多个任务中降低了35%的控制波动性，生成的文本相关性提升28%，为更精确的LLM行为控制提供了新方案。

## English Version

**New Method Improves Activation Control Precision in LLMs**

arXiv paper 2603.12298 proposes Global Evolutionary Steering (GES) to refine activation control in LLMs via cross-layer consistency. Existing methods are susceptible to high-dimensional noise. GES leverages correlations between activation layers to create more stable control vectors. Experiments show it reduces control variability by 35% and improves text coherence by 28%, enabling more precise LLM behavior control.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2603.12298)

**详情页**：https://ai.daily.yangsir.net/daily/20260317-T0-24

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*