The list as it stood that day: the papers researchers were upvoting on Hugging Face's daily papers. Papers stay on arXiv.
We propose Wave Forcing (WF), which co-designs generation, training, and execution for continuous video output with multi-step denoising.
Yijie Jin, Yuxuan Liu, Jiajun Xu and 7 moreWe introduce CoworkOS, a general-purpose harness for long-horizon work that separates task strategy from execution management: the model decides actions, delegation, and acceptance, while the runtime maintains state and feedback.
Jiajie Jin, Guanting Dong, Chengjun Pan and 37 moreTo address these challenges, we introduce CUA-Master, a nested double-loop recursive improvement framework.
Xinlei Yu, Gen Li, Ji Zhao and 15 moreComputer graphics is built on code. Large language model (LLM) agents can now generate graphics code and iteratively refine it through various feedback.
Hengyu Liu, Jihyeon Je, Ruiqi Wang and 11 moreWe study their intersection through the agent harness: an executable runtime that constructs context, exposes tools, maintains state, routes feedback, and manages budgets around video models.
Liaoyuan Fan, Yuhao Zheng, Dongxing Mao and 18 moreTo democratize RUA research, we introduce uni-robot-agent, a unified harness for competitive open-weight robot use that supports further training.
Jiarui Yuan, Yuxin Zuo, Xiaoyu Huang and 25 moreWe tackle this problem by developing SimpleLoRIF, a scalable gradient embedding method that identifies fine-tuning examples whose gradients align with unsafe behavior.
Saket Reddy, Amro Abbas, Anshuman Suri and 3 moreWe present NorthCUA, an integrated system for turbocharging computer-use agents through experience at scale.
Chaoyun Zhang, Bingxuan Li, Pinxue Guo and 27 moreTo address this gap, we present ReTrace, a scalable evaluation framework for systematically studying recovery behavior in computer-use agents.
Byeonghun Bang, Sangwoo Park, Jinheon Baek and 3 moreCoding agents solve repository issues through long sequences of exploration, editing, and validation.
Ziming Li, Lu Yu, Jiatan Huang and 3 moreWe present PhotoGuider, a measure-then-frame method that answers both questions with compact learned scorers over one shared set of composition-aware features.
Yuan-Kang Lee, Kuan-Lin Chen, Chia-Che Chang and 1 moreWe argue that not all visual evidence deserves equal attention, as the importance of visual information depends on the content of each video.
Tengfei Liu, Yang Shi, Yuran Wang and 12 moreWe argue that this fragmentation obscures a deeper convergence. Driven by contrastive learning, Transformer architectures, and a granularity shift, embedding research across modalities is converging toward a common set of principles, even as modality-specific challenges remain…
Ziyin Zhang, Yuyu Guo, Ying Zhang and 15 moreWe present an evaluation framework that separates what frozen representations contain, how they organize appearance and physical differences, and what changes a model's own prediction error.
Anton Hristov, Aleksandar Slavov, Vasil VasilevRather than further increasing complexity within individual repositories, we introduce CrossSWE, a benchmark for cross-repository joint development comprising 171 real-world instances spanning 165 repositories, 20 technical domains, and 13 programming languages.
Silin Chen, Yuling Shi, Heng Lian and 4 moreWe argue that progress should be measured by how effectively agents amplify human agency: enabling people to pursue their goals with less routine supervision while retaining control over consequential decisions.
Zheng Wu, Xinbei Ma, Yao Yao and 8 moreWe argue that such failures, commonly attributed to perception, reveal a reference gap: the failure to bind language to the relevant visual entities, regions, or locations.
Guizhen Chen, Weiwen Xu, Wotao Yin and 2 moreTo bridge this gap, we introduce DecisionRank, a decision model trained through item-space on-policy distillation (OPD).
Yanran Wu, Dian Yu, Chongyang Bai and 10 moreWe present FLAMIA, a family of attention kernels for Blackwell with MXFP8 and mixed NVFP4/FP8 training paths.
Fedor Velikonivtsev, Mikhail Koltakov, Denis KuznedelevRetrieval-augmented generation (RAG) lets a large language model (LLM) answer questions by retrieving relevant passages from a large corpus.
Dingyi Kang, Bingzhe Li