Domain experts trained from a shared checkpoint can transfer their specialized capabilities to a single student through on-policy distillation (OPD). Existing research primarily focuses on improving this merging process, while the algorithms used to train the experts have received limited systematic comparison. We inve...
Jing-Yuan Huang, Zu-Ming Huang, Yu-Cheng Shi et al.· 0 citations
SkillAny is introduced, an online skill learning framework that enables language agents to learn, retrieve, and invoke reusable skills at any intermediate state during task execution, and shows that SkillAny consistently outperforms strong baselines.
Jiaxi Li, Ke Deng, Yun Wang et al.· 9 citations· ⚡3
RiseHarness is a framework for iteratively improving agent harnesses with model weights fixed that treats a specific model as the unit of harness adaptation, studying how changes to shared interaction mechanisms can make existing capabilities more reliable across tasks.
Ru-Han Wang, Yu-Cheng Shi, Zong-Xia Li et al.· 0 citations
Results indicate that existing scientific software can provide scalable, behaviorally verified supervision for terminal agents, and introduces software-in-the-loop reconstruction, a self-supervised framework that obtains reference outputs and verification targets from existing software workflows, executable programs ma...
Zhong-Zhi Li, Yu-Cheng Shi, Zong-Xia Li et al.· 2 citations
Successful trajectories on difficult tasks provide valuable supervision for model improvement, but specialized harnesses introduce interventions that may be unavailable during deployment. We propose Recursive Self-Rewrite (RSR), a framework that uses one base model, Qwen-3.8-27B, to discover successful solutions under...
Zong-Xia Li, Yu-Cheng Shi, Zhong-Zhi Li et al.· 3 citations
The Code-with-Image Bench (CwI-Bench), thirty task families induced by hidden visual computations with disjoint learning and evaluation splits, even GPT-5.6-luna stays below 30% with tool-free chain of thought; given a bare interpreter it reaches 43%, and with skills evolved through its own executable reflection, 67%.
Tianze Yang, Liang Wu, Rui Sun et al.· arXiv.org· 2 citations
AnyJev is described, which reads a typed decision from one prefill of a pretrained instruction-tuned language model, which divides out a label prior estimated from unlabelled inputs, and it averages log-probabilities over the K cyclic rotations of the option list.
Jia-Mu Zhang, Tian-Ze Yang, Yu-Cheng Shi et al.· 6 citations
Domain experts trained from a shared checkpoint can transfer their specialized capabilities to a single student through on-policy distillation (OPD). Existing research primarily focuses on improving this merging process, while the algorithms used to train the experts have received limited systematic comparison. We inve...
Jing-Yuan Huang, Zu-Ming Huang, Yu-Cheng Shi et al.· 0 citations
Tencent WorkBuddy Bench is introduced, a multi-domain evaluation suite for coding agents, and its construction methodology, scoring protocol, and a cross-model leaderboard are documents, with a cross-model leaderboard across several model families.
T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool-call turns per task, rewarded by executing each task's own verifier by executing each task's own verifier.
Jun-Yao Yang, Yu-Cheng Shi, Zhong-Zhi Li et al.· 9 citations
RadFabric is presented, an agentic AI system that orchestrates fourteen specialized open-source CXR analytics models and two Vision-Language Models through a modular protocol that enables explainable, robust diagnoses across common and rare pathologies while facilitating extensibility through additional agents.
Wenting Chen, Yi Dong, Zhaojun Ding et al.· npj Digital Medicine· 3 citations
Recursive Synthetic Terminal Tasks (RST) is presented, a recursive verified synthesis framework for constructing long-horizon terminal-agent tasks at scale and shows no ceiling, indicating that the process can continue well beyond the scale reported here.
Zhongzhi Li, Yucheng Shi, Zongxia Li et al.· 5 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.