Learning Steerable Clarification Policies with Collaborative Self-play
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
The Flow has not summarised this story yet — read it at arXiv Machine Learning.
CLEAT (CLarification through valuE-guided Agent co-Training) is a framework that jointly trains a clarification-value controller and a task policy to improve interactive intent alignment in large language model agents. It introduces a residual query value metric, estimated via paired rollouts, to assess the utility of asking clarification questions and uses this to guide both controller predictions and policy decisions. Experiments with Qwen3-4B show CLEAT outperforming baselines across multiple benchmarks, with significant gains in user‑centric tasks and demonstrating that co‑training benefits the task policy even when the controller is omitted at inference.
ReferClaw is an agentic framework designed to improve egocentric AI assistants by treating question answering as an interactive process. It uses three cooperating agents to generate candidate interpretations, ask clarifying questions, and produce answers grounded in resolved intent and visual evidence. The system balances information gain against user effort, learns from interaction, and shows higher accuracy and joint answer‑grounding performance on augmented egocentric QA benchmarks, with zero‑shot transfer across datasets.
arXiv:2609.37588v1 Announce Type: new Abstract: Users of language-based assistive agents often make ambiguous requests. In response, an assistant can either directly act on its interpretation of the...
Users of language-based assistive agents often make ambiguous requests. In response, an assistant can either directly act on its interpretation of the request --- risking misalignment with the user --...
ReferClaw is an agentic framework designed to improve egocentric AI assistants by treating question answering as an interactive process. It uses three cooperating agents to generate candidate interpretations, ask clarifying questions, and produce answers grounded in resolved intent and visual evidence. The system balances information gain against user effort, learns from interaction, and shows better accuracy and grounding on augmented egocentric QA benchmarks, including zero‑shot transfer across datasets.
UnifiedPlayers is a cooperative framework that jointly adapts planning, execution, and evaluation for tool-integrated reinforcement learning agents. It consists of a Planning Player that generates tasks, an Execution Player that creates multi-turn trajectories with Python tool calls, and an Evaluation Player that builds executable verifiers, all coordinated by role‑specific rewards under GRPO. The approach outperforms prior baselines on mathematical and general reasoning benchmarks and yields a verifier with high adversarial detection accuracy and more discriminative reward signals.