DiscussLLM: Teaching Large Language Models When to Speak

2025-08-25

AI Summary

Current large language models (LLMs) engage passively in multi-turn dialogues, lacking proactive intervention awareness, which severely limits their collaborative capability. Method: This paper proposes the first LLM-based framework for modeling proactive speaking timing in dynamic dialogues. It introduces a large-scale annotated dialogue dataset with five AI intervention categories and a “silence token” mechanism to explicitly represent non-intervention decisions. A two-stage data generation pipeline is designed—combining end-to-end modeling with a decoupled classifier-generator architecture—and optimized for low-latency inference. Contribution/Results: Experiments demonstrate that our method accurately identifies high-value intervention opportunities, significantly improving dialogue quality and contextual awareness across diverse collaborative scenarios. To our knowledge, this is the first systematic solution to the problem of timely and effective proactive participation of LLMs in human dialogues.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities in understanding and generating human-like text, yet they largely operate as reactive agents, responding only when directly prompted. This passivity creates an "awareness gap," limiting their potential as truly collaborative partners in dynamic human discussions. We introduce $ extit{DiscussLLM}$, a framework designed to bridge this gap by training models to proactively decide not just $ extit{what}$ to say, but critically, $ extit{when}$ to speak. Our primary contribution is a scalable two-stage data generation pipeline that synthesizes a large-scale dataset of realistic multi-turn human discussions. Each discussion is annotated with one of five intervention types (e.g., Factual Correction, Concept Definition) and contains an explicit conversational trigger where an AI intervention adds value. By training models to predict a special silent token when no intervention is needed, they learn to remain quiet until a helpful contribution can be made. We explore two architectural baselines: an integrated end-to-end model and a decoupled classifier-generator system optimized for low-latency inference. We evaluate these models on their ability to accurately time interventions and generate helpful responses, paving the way for more situationally aware and proactive conversational AI.

Problem

• Teaching LLMs to decide when to speak proactively
• Bridging the awareness gap in reactive AI dialogue systems
• Enabling AI interventions at valuable conversational triggers

Innovation

• Two-stage data generation pipeline
• Predict silent token when not needed
• Integrated end-to-end and decoupled architectures