SkillChain is a system designed to automate the lifecycle of skills for image-based AI assistants on e-commerce platforms. It operates in three stages—Skill Creator, Route Optimizer, and Body Refiner—to bootstrap skills from task specifications, align routing, and iteratively refine skill bodies using dual-path LLM-Judge evaluation. In production, SkillChain improves response quality, especially in structural compliance and content quality, and an online A/B test shows increased user engagement, content consumption, and long‑term retention.
By Yimin Hu, Mengtao Xu, Hao Guo, Yuheng Song, Xiaoyong Zhu, Bo Zheng
The paper introduces the Multi-Instruction Multi-Shot Long-Video Editing (MMLVE) task, aiming to enable consistent editing of long videos with multiple instructions. It proposes an agentic editing framework that combines Large Language Models and Vision-Language Models for shot-level decoupling and precise instruction parsing. The authors also present the MMLVE-Bench dataset and evaluation metrics, showing that their MMLVE-Agent outperforms existing state‑of‑the‑art methods by eliminating hallucinations and preserving temporal consistency.
By Chenyang Wu, Fuchen Long, Binyuan Huang, Xinlong Sun, Xi Chen, Chun-Le Guo, Chongyi Li
The paper introduces the Latent Evolution Operator Network (LEON), a new architecture for modeling latent state transitions in World Action Models (WAMs). LEON uses a learned observable space with context‑modulated operator‑based propagation and additive forcing, grounded in a controlled Koopman generator framework. Experiments on two WAM formulations show that LEON improves closed‑loop performance and robustness, demonstrating that the way transitions are realized is a crucial architectural choice in latent WAMs.
By Xiaoxiao Lu, Yunlong Dong, Jiahao Shi, Ye Yuan
The paper introduces the Latent Diagnostic Taxonomy, a framework that builds a dimensionality‑optimized classifier and a diagnostic tool to assess the trustworthiness of its confident predictions. It identifies a small set of influential prompts (latent support vectors) that reveal tokens which can change the classifier’s output, and uses these tokens to create a taxonomy that classifies prompts into safe, heuristic bias, heuristic override, or insufficient context categories. Applied to a prompt‑injection detection model, the framework shows that about 77% of confident decisions are fragile to a single token, distinguishing between calibration failures and exploitable shortcuts, and offers remediation strategies for each taxonomy zone.
By Jaturong Kongmanee, Smile Thanapattheerakul
PAWBench introduces a benchmark to evaluate whether video generation models can act as probabilistically aligned world models, meaning they should reproduce not just plausible trajectories but the full distribution of possible behaviors from the same initial conditions. The authors formalize probabilistic alignment as a distributional criterion and provide PAWEval, an outcome-level protocol that turns repeated video rollouts into empirical distributions over physical behaviors. Across 50 scenarios and eleven current systems, none consistently matched reference probabilities or captured the full range of valid behaviors, highlighting a significant gap in current video generators.
By Yuandong Pu, Le Zhuo, Sayak Paul, Gabriel Jorge Menezes, Avram {\DJ}or{\dj}evi\'c, Shiyang Li, Yifan Zhou, Bin Fu, Wenlong Zhang, Junjun He, Yu Qiao, Yihao Liu, Jingbo Xing, Xi Chen
DeflectBench is a new benchmark that evaluates how large language models (LLMs) generate rhetorical fallacies when prompted. The study tests 23,990 generations from four leading models using three deflection strategies (whataboutism, ad hominem, red herring), seven prompt framings, and 80 claims across four controversy levels. Results show that refusal to produce fallacies depends mainly on request structure, with prompt framing and fallacy type dramatically affecting compliance rates.
By Art Kanke
The paper introduces Disjoint Parameter Training (DPT), a framework that addresses Skill Conflict—where shared encoder parameters hinder separate tasks of motion prediction and safety planning—by training tasks on distinct parameter subsets before merging. DPT employs sparse merging to integrate only the most influential parameters, reducing interference and enhancing representational capacity. Experiments on JRDB and JTA benchmarks show that DPT outperforms existing unified models, demonstrating its effectiveness for safe, resource‑efficient robot navigation.
By Taewon Seo, Seonae Jeon, Giwon Lee, Kuk-Jin Yoon, Daehee Park
SpatialCrafter introduces a two‑stage framework for single‑image world modeling that first generates a global 3D proxy using a Point‑anchored Sparse Structure Flow module, then refines appearance with a Generative Deferred Refiner built on a video diffusion model. The method incorporates Parallel Geometry Injection and Proxy‑Aware Corruption training to integrate the proxy without disrupting the pretrained generative manifold, and it is evaluated on a newly constructed dataset of 115K scenes. Experiments demonstrate that SpatialCrafter outperforms existing approaches, reducing long‑term drift and maintaining consistency under rapid camera motion and extreme viewpoints.
By Chuan Fang, Lingteng Qiu, Yixun Liang, Rui Chen, Kunming Luo, Zhaohua Zheng, Tongyuan Bai, Feipeng Tian, Zilong Dong, Zihan Zhou, Ping Tan
This study introduces a method for independently testing Assisted Lane Change (ALC) systems on public roads using LiDAR-based vehicle detection and tracking. A campaign on the A31 French motorway evaluated 27 lane‑change maneuvers, finding that 18 were completed, 9 were suppressed, and in six cases the system permitted maneuvers that violated regulatory minimum distance requirements. The results confirm LiDAR’s suitability for such assessments and provide a framework that can aid market surveillance and future regulatory updates.
By Marcello Cellina, Akos Kriston, Antonio Migneco, Davide Maggi, Stefano Favelli, Fabrizio Re, Fabrizio Minarini, Andrea Nuovo, Riccardo Dona, Biagio Ciuffo
SSMB is a self‑supervised keypoint detector designed for motion‑blurred images that does not rely on handcrafted detectors or external pseudo‑labels. It introduces a Local Discriminability Enhancement module to recover fine‑grained local detail after global feature mixing, and is trained in two stages: geometric pretraining on synthetic shapes and blur‑aware training on real sharp‑blur pairs using a multi‑component self‑supervised objective. Extensive experiments show that SSMB outperforms both supervised and self‑supervised baselines on keypoint detection, image matching, relative pose estimation, and visual localization under motion blur, achieving state‑of‑the‑art performance.
By Zhenjun Zhao, Fabio Bellavia, Wenting Wang, Fan Zhu, Jiajun Wu, Suryansh Kumar, Mingqiang Wei, Haoang Li, Javier Civera
Ancient-Bench is a new benchmark for recognizing text on ancient Chinese artifacts, comprising 2,700 images that span 3,000 years of character evolution, nine artifact categories, and seven historical script forms. It introduces three annotation standards—symbol, character, and parsing standardization—to accommodate medium‑specific characteristics and enable consistent evaluation. Experiments show that current Vision‑Language Models and OCR specialists still struggle with variant characters, specialized symbols, and hallucination, indicating the task remains largely unsolved.
By Hiuyi Cheng, Nuo Xu, Yuyi Zhang, Xuhan Zheng, Wei Pan, Jing Zhang, Dezhi Peng, Minghui Liao, Yihua Teng, Jihao Wu, Haoyu Ren, Lianwen Jin
The paper proposes a data‑centric optimization for object pose estimation that uses a physically grounded rotation representation based on principal axes alignment. By aligning an object's coordinate system with its inertial principal axes, the method achieves inherent stability, symmetry‑aware canonicalization, and framework agnosticism, allowing it to be applied at the dataset level without modifying existing networks. Experiments on category‑level and instance‑level models show consistent accuracy improvements while preserving baseline network integrity.
By Wei Chen, Tao Zhen, Zhongchen Shi, Jing Zhang, Liang Xie, Erwei Yin
The paper introduces a new task called compositional layout understanding, focusing on interpreting complex, multi‑layer document and UI designs. It presents CoDeLayout, a VQA dataset of about 20,000 real‑world layouts annotated with compositional element pairs and design intent. The authors identify semantic drift and structural ambiguity as key challenges for vision‑language models and propose MASON, a post‑training approach that combines multimodal alignment and structural perception to improve performance, achieving 91.66% accuracy with only 30% of the training data.
By Yiyang Huang, Zhaowen Wang, Simon Jenni, Jing Shi, Yitian Zhang, Yizhou Wang, Yun Fu
The paper introduces DPA-I2P, a depth-guided projective alignment method for image-to-point-cloud registration in autonomous driving. It employs Ray-Conditioned Metric Depth Encoding and Projection-Consistent Vision Lifting to align depth and visual cues geometrically, and uses Cross-Modal Query Pruning to enhance matching stability. Experiments on KITTI and nuScenes show significant reductions in rotation and translation errors compared to existing implicit baselines.
By Wenxin Zhang, Hang Li, Zhiwei Xu, Qiankun Dong, Gang Wang, Tao Li
VIPER is the first expert‑curated benchmark for vision‑language models in veterinary toxicologic pathology, comprising 1,251 questions linked to 419 H&E‑stained rat histology images across seven organ systems. The benchmark includes multiple‑choice, KPrim, and free‑text formats and was validated by board‑certified veterinary pathologists. Evaluation of 16 models—spanning veterinary‑specific, human pathology‑specialized, and general‑purpose models—reveals a significant domain gap, a tendency for frontier models to over‑diagnose normal tissue, and underscores the importance of domain‑specific training for accurate visual predictions.
By Luca L. Weishaupt, Simone de Brot, Javier Asin, Lloren\c{c} Grau-Roma, Nic G. Reitsam, Andrew H. Song, Dongmin Bang, Stefan T. Kaluziak, Long Phi Le, Jakob Nikolas Kather, Faisal Mahmood, Guillaume Jaume
This survey reviews recent advances in surgical video generation, categorizing methods into unconditional, conditional, and world modeling generation. It highlights a shift from creating visually plausible frames to modeling the causal dynamics of surgical scenes, and discusses challenges such as pixel-level fidelity versus clinical plausibility, generalization, physical realism, controllability, and interpretability. The paper also compiles experimental results from public datasets to serve as a quantitative benchmark for the field.
By Fuxiang Huang, Chenxu Zhang, Liang Han, Lei Zhang
The paper investigates whether tool‑calling large language model agents maintain consistent safety throughout a conversation. It finds that agents are most vulnerable at the very start of a session, with safety improving significantly after completing a few regular agentic tasks—a phenomenon termed the cold‑start safety gap. The authors introduce the Safety Over Depth for Agents (SODA) benchmark to systematically study this effect, evaluate multiple models, and demonstrate that warming up agents with regular tasks before deployment enhances safety while preserving utility.
By Chung-En Sun, Linbo Liu, Tsui-Wei Weng
TopiCLEAR is a framework that clusters document or sentence embeddings using adaptive dimensionality reduction to uncover low‑dimensional geometric structures that correspond to human‑interpretable topics. The method is evaluated on four benchmark datasets, showing strong agreement with human annotations, especially for short and informal texts. A Twitter case study demonstrates that TopiCLEAR yields more interpretable topics than LDA, recovering both annotated topic structure and coherent sub‑topics.
By Aoi Fujita, Taichi Yamamoto, Yuri Nakayama, Ryota Kobayashi
The paper introduces Legal Rule Induction (LRI), a task that seeks to extract concise, generalizable doctrinal rules from analogous judicial precedents. It presents a reproducible pipeline for constructing LRI datasets and, using Chinese law, releases the first benchmark comprising 5,121 case sets (38,088 court cases) for training and 216 expert‑annotated gold test sets. Experiments show that state‑of‑the‑art large language models struggle with over‑generalization and hallucination, but training on the new dataset significantly improves their ability to capture nuanced rule patterns across similar cases.
By Wei Fan, Tianshi Zheng, Yiran Hu, Zheye Deng, Weiqi Wang, Baixuan Xu, Chunyang Li, Haoran Li, Weixing Shen, Yangqiu Song
The paper introduces INTENT-AS-A-TOOL, a method that equips large language models with intent-targeted tools to provide a fine-grained, judge‑free signal of their commitment to specific behaviors during reasoning. By monitoring the probability of calling these intent tools, the authors can track how intent evolves throughout generation, complementing chain‑of‑thought monitoring and expanding post‑hoc labels into dense trajectories. The approach identifies critical steps for online intervention, demonstrating that action preferences are useful for detecting agentic misalignment in autonomous agents.
By Yutong Zhang, Jianshuo Dong, Peng Xu, Long Wang, Jie Zhang, Tianwei Zhang, Xiaoping Zhang, Han Qiu