---
id: 20260514-T0-06
title: "两个LLM绕过文本直接共享隐藏状态：Bicameral实现双向并行通信"
title_en: "Bicameral Model Enables LLMs to Communicate Via Hidden States, Not Text"
url: https://ai.daily.yangsir.net/daily/20260514-T0-06
issue_date: 2026-05-14
publish_date: 2026-05-13T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2605.11167
---

# 两个LLM绕过文本直接共享隐藏状态：Bicameral实现双向并行通信

现有LLM多模型协作依赖文本输出进行交互，每轮对话都要经过词汇表序列化，效率低且信息有损。研究提出Bicameral模型，让两个预训练语言模型通过连续、并行的隐藏状态通道直接协调，绕过文本中间步骤。这种双向隐藏状态耦合方式比传统文本交互更快，保留了更多细粒度信息。该方案为多模型协作架构提供了新思路，未来可用于需要多个模型实时配合的复杂推理场景。

## English Version

**Bicameral Model Enables LLMs to Communicate Via Hidden States, Not Text**

Current multi-LLM systems communicate via text, serializing every exchange through the output vocabulary. The Bicameral Model lets two pretrained LLMs coordinate through continuous, concurrent hidden-state channels, bypassing text serialization. This bidirectional coupling is faster and preserves more fine-grained information than text-based interaction, offering a new approach for multi-model architectures in complex real-time reasoning scenarios.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2605.11167)

**详情页**：https://ai.daily.yangsir.net/daily/20260514-T0-06

---

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