Tag
The paper introduces CLEAT, a framework that co-trains a clarification-value controller with an LLM task policy using paired rollouts to estimate residual query value, enabling agents to ask useful clarifying questions that adapt as the policy improves, achieving top scores across four benchmarks with Qwen3-4B.
本文提出 MIRA（Musical Intent Refinement Agent），一种在测试时通过将用户意图细化为可验证 rubric 并进行轨迹感知树搜索来改进文本到音乐生成对齐的智能体，并发布了音乐专家标注的基准 MuRA-Bench，使开源生成器达到接近 Suno、Mureka 等商业系统的性能。
ByteDance and University of Notre Dame researchers introduce Drift-Bench++, a benchmark pipeline for evaluating LLM agents' interactive intent alignment under miscommunication and evolving user intent, along with the GRIP evaluation protocol showing agents remain far from oracle performance.