Wenlong Hou

Yuxiang Feng, Juncheng Wang, Chao Xu et al.

Video generation models produce visually compelling results but systematically violate physical commonsense -- on VideoPhy-2, the best model achieves only 32.6% joint accuracy. We identify a specification bottleneck: text prompts are lossy compression of the physical world, omitting the parameters that fully determine dynamics, and no amount of model scaling can recover what was never specified. From this diagnosis we derive three properties that physics conditioning must satisfy -- sufficiency, dynamism, and verifiability -- and show that no existing approach satisfies all three. We present NEWTON, in which video generation is demoted from the system output to one action inside an agent's toolbox: a learned planner orchestrates physics-aware tools (keyframe generation, scientific computation, prompt refinement) to construct rich conditioning, and a verifier closes the loop for iterative re-planning. The planner is the sole trainable component, optimized on-policy via Flow-GRPO inside the live multi-turn loop. On VideoPhy-2, NEWTON improves joint accuracy from 21.4% to 29.7% on LTX-Video and from 30.7% to 37.4% on Veo-3.1, without modifying either generator. Our project page: \href{https://Newton026.github.io/newton}{https://Newton026.github.io/newton}

Ziqi Xie, Xiao Lai, Weidong Zhao et al.

Current image stitching methods often produce noticeable seams in challenging scenarios such as uneven hue and large parallax. To tackle this problem, we propose the Reference-Driven Inpainting Stitcher (RDIStitcher), which reformulates the image fusion and rectangling as a reference-based inpainting model, incorporating a larger modification fusion area and stronger modification intensity than previous methods. Furthermore, we introduce a self-supervised model training method, which enables the implementation of RDIStitcher without requiring labeled data by fine-tuning a Text-to-Image (T2I) diffusion model. Recognizing difficulties in assessing the quality of stitched images, we present the Multimodal Large Language Models (MLLMs)-based metrics, offering a new perspective on evaluating stitched image quality. Compared to the state-of-the-art (SOTA) method, extensive experiments demonstrate that our method significantly enhances content coherence and seamless transitions in the stitched images. Especially in the zero-shot experiments, our method exhibits strong generalization capabilities. Code: https://github.com/yayoyo66/RDIStitcher

Junyu Liu, Zirui Li, Qian Niu et al.

As Large Language Models (LLMs) are increasingly deployed in healthcare field, it becomes essential to carefully evaluate their medical safety before clinical use. However, existing safety benchmarks remain predominantly English-centric, and test with only single-turn prompts despite multi-turn clinical consultations. To address these gaps, we introduce JMedEthicBench, the first multi-turn conversational benchmark for evaluating medical safety of LLMs for Japanese healthcare. Our benchmark is based on 67 guidelines from the Japan Medical Association and contains over 50,000 adversarial conversations generated using seven automatically discovered jailbreak strategies. Using a dual-LLM scoring protocol, we evaluate 27 models and find that commercial models maintain robust safety while medical-specialized models exhibit increased vulnerability. Furthermore, safety scores decline significantly across conversation turns (median: 9.5 to 5.0, $p < 0.001$). Cross-lingual evaluation on both Japanese and English versions of our benchmark reveals that medical model vulnerabilities persist across languages, indicating inherent alignment limitations rather than language-specific factors. These findings suggest that domain-specific fine-tuning may accidentally weaken safety mechanisms and that multi-turn interactions represent a distinct threat surface requiring dedicated alignment strategies.