CLEAT: Learning Value-Guided Clarification for Interactive Agents
LLM agents can execute increasingly complex tasks, but effective assistance also requires aligning their actions with goals and constraints that users may leave unstated. We study this problem as interactive intent alignment through clarification.
This article introduces CLEAT, a framework for improving how LLM agents handle ambiguous user intentions by learning value-guided clarification through co-training.
What happened
The article presents CLEAT, a framework that trains LLM agents to ask clarifying questions by co-training a clarification-value controller and task policy. It uses residual query value to evaluate the utility of clarification, with paired rollouts from interaction history to guide training. The system adapts as the agent learns, improving task completion accuracy.
Why it matters
For AI developers and professionals, this research enhances agent systems' ability to handle ambiguous user inputs, leading to more effective task execution and better user interactions in real-world applications.
- Clarification is critical for aligning LLM agents with user intent
- Co-training improves task policy performance through adaptive guidance
- Residual query value provides a measurable metric for clarification effectiveness
- The framework maintains benefits even without the controller at inference
- Dynamic adaptation to evolving policies enhances long-term performance
The bigger picture
The framework demonstrates significant performance improvements in benchmarks, offering a practical approach to enhance LLM agents' ability to handle underspecified tasks through dynamic clarification strategies.
CLEAT advances interactive agent systems by introducing a value-guided clarification framework that improves task execution through co-training and adaptive learning, achieving state-of-the-art results across multiple benchmarks.
Technical details
- Clarification-value controller
- Task policy co-training
- Residual query value (measures task utility from clarification)
- Paired rollouts from interaction history
- Adaptive clarification guidance during agent learning
FAQ
What is CLEAT's primary purpose?
CLEAT aims to improve LLM agents' ability to handle ambiguous user intentions by learning when and how to ask clarifying questions through co-training.
How does CLEAT measure clarification value?
It uses residual query value, calculated through paired rollouts from interaction history, to quantify the task utility gained by asking clarification questions.
What are the key benefits of co-training in CLEAT?
Co-training enhances task policy performance by allowing the system to adapt clarification strategies as the agent learns, even when the controller is removed at inference.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by arXiv - cs.AI on Oct 12, 2026.
Comments
Sign in to join the discussion.