A single multimodal large language model (MLLM) struggles to excel simultaneously at detection, localization, description, and reasoning in multimodal industrial anomaly understanding (MM-IAU). We show that in-domain training does not close this gap. On MMAD, a widely adopted MM-IAU benchmark, trained specialists reach...
Xing-Wu Zhang, Duan-Yang Du, Hui-Ling Zhu et al.· 0 citations
The main findings are that diverse, high-quality SFT establishes a strong capability floor and difficulty filtering keeps RL prompts within a productive learning range, and reward reliability provides a practical principle for ordering stages.
Chia-yuan Chang, Ren-Yuan Cheng, Rui Feng et al.· 0 citations
Search agents are usually trained under a single harness. But once an agent is deployed in a real application, its harness is frequently updated (e.g., a rewritten system prompt) to fit production needs. This exposes a fragility of post-trained agents: because a learned behavior is entangled with its training harness,...
Xin-Lu Zhang, Ying-Chun Lin, Zhi-Han Zhang et al.· 0 citations
This work demonstrates that a general-purpose agent can directly drive a physical robot throughout task execution without any task-specific or environment-specific training and introduces Agent as Policy (AGP), which places task planning and execution under the agent's control.
Meng-Zhao Jia, Yang Lin, Xi-Xin Zhang et al.· 15 citations· ⚡1
QUBRIC, a framework that co-designs queries and rubrics can make rubric-based RL a practical complement to RLVR beyond strictly verifiable tasks, provides evidence that co-designing queries and rubrics can make rubrics a practical complement to RLVR beyond strictly verifiable tasks.
