Exploring and Controlling Diversity in LLM-Agent Conversation

TL;DR

This paper investigates how prompt design influences dialogue diversity in long-term LLM-agent interactions, introducing a novel adaptive prompt pruning method to control diversity effectively.

Contribution

It proposes Adaptive Prompt Pruning (APP), a new technique that dynamically adjusts prompt content to modulate diversity in LLM-agent conversations.

Findings

Reducing contextual information increases output diversity.

APP effectively controls diversity with a single parameter.

Memory components significantly influence diversity constraints.

Abstract

Controlling diversity in LLM-agent simulations is essential for balancing stability in structured tasks with variability in open-ended interactions. However, we observe that dialogue diversity tends to degrade over long-term simulations. To explore the role of prompt design in this phenomenon, we modularized the utterance generation prompt and found that reducing contextual information leads to more diverse outputs. Based on this insight, we propose Adaptive Prompt Pruning (APP), a novel method that allows users to control diversity via a single parameter, lambda. APP dynamically prunes prompt segments based on attention scores and is compatible with existing diversity control methods. We demonstrate that APP effectively modulates diversity through extensive experiments and propose a method to balance the control trade-offs. Our analysis reveals that all prompt components impose constraints on diversity, with the Memory being the most influential. Additionally, high-attention contents consistently suppress output diversity.