Reducing Conversational Escalation in Large Language Model Dialogue with Nonviolent Communication Constraints

Short summary

Researchers show LLM dialogue can be improved with Nonviolent Communication (NVC) prompt constraints, which reduce conversational escalation in emotionally charged situations. The approach was tested across multiple models and user resistance levels using dual-agent simulation. Results suggest simple communication guidelines can meaningfully improve LLM trustworthiness in conflict-prone settings.

• •NVC-derived prompt constraints reduce conversational escalation in LLMs
• •Tested across multiple instruction-tuned models and varying user resistance levels
• •Lightweight approach improves dialogue trustworthiness in conflict-prone settings

Generated with AI, which can make mistakes.