Beyond the Hivemind: Escaping LLM Homogeneity via Meta-Persona Anchoring and Sequential Temperature Scaling

超越蜂巢思维：通过元人格锚定与序列温度缩放摆脱大语言模型的同质化

Abstract: Recent studies have identified an “Artificial Hivemind” effect in Large Language Models (LLMs) causing models to converge on a narrow, homogenized consensus even for open questions. This semantic collapse limits the diversity of AI, resulting in high inter-response similarity ($\approx 0.80-0.90$) even under high-temperature sampling.

摘要： 近期研究发现，大语言模型（LLMs）中存在一种“人工蜂巢思维”（Artificial Hivemind）效应，导致模型即使在面对开放性问题时，也会收敛于狭隘、同质化的共识。这种语义崩溃限制了人工智能的多样性，即便在采用高温度采样（high-temperature sampling）的情况下，响应之间的相似度依然很高（$\approx 0.80-0.90$）。

In this paper, we propose a novel mitigation framework to increase diversity: Meta-Persona Anchoring combined with Filtered Temperature Scaling (FTS). Our approach utilizes a two-stage generation process: first, the model is prompted to self-select a unique, idiosyncratic persona to anchor its starting point; second, we apply a dual-stage sampling sieve, utilizing Top-$p$ filtering to preserve grammatical validity followed by extreme temperature scaling ($T \ge 4.0$) on the surviving candidates to explore the broadened probability distribution.

在本文中，我们提出了一种旨在增加多样性的新型缓解框架：元人格锚定（Meta-Persona Anchoring）结合过滤温度缩放（Filtered Temperature Scaling, FTS）。我们的方法采用两阶段生成过程：首先，提示模型自主选择一个独特且具有特质的人格作为其起始锚点；其次，我们应用双阶段采样筛选机制，利用 Top-$p$ 过滤来保持语法有效性，随后对留存的候选结果进行极端温度缩放（$T \ge 4.0$），以探索更广阔的概率分布。

We evaluate our method using the INFINITY-CHAT dataset on state-of-the-art open weight models under $\sim$20B parameters. Our results demonstrate a significant reduction in semantic convergence, with average pairwise cosine similarity dropping from ($\approx 0.85$) to ($\approx 0.65$). Our scheme achieves a majority of questions below the 0.7 threshold, effectively reducing the gap between artificial mode collapse and human-level typological diversity. We provide our implementation as an open-source framework to enable more diverse and creative AI deployments.

我们使用 INFINITY-CHAT 数据集，在参数量约为 20B 的最先进开源权重模型上评估了该方法。结果表明，语义收敛现象得到了显著改善，平均成对余弦相似度从 ($\approx 0.85$) 降至 ($\approx 0.65$)。我们的方案使大多数问题的相似度低于 0.7 的阈值，有效缩小了人工智能模式崩溃与人类水平类型多样性之间的差距。我们将该实现作为开源框架提供，旨在支持更多样化和更具创造性的人工智能部署。