Beyond Scalar Scores: Exploring LLM-based Metrics for Clinical Significance Evaluation in Radiology Reports

Reliable evaluation of generated radiology reports requires strict clinical accuracy, as omitted critical findings or mischaracterized radiographic observations can directly affect patient care. Existing metrics obscure this requirement by reducing report quality to a medically ungrounded scalar. Although Large Language Models (LLMs) possess rich medical knowledge, they likewise struggle to draw a reliable boundary between clinically significant errors and harmless variation. We study this boundary using ReEvalMed benchmark as testbed and evaluate metric-level clinical significance from detecting true clinical errors ("Discrimination") and tolerating insignificant variations ("Robustness"). Across 8 LLM evaluators under one-pass and two-pass settings, we identify a widespread discrimination bias: models effectively detect errors but also over-penalize harmless rephrasings. To mitigate this, we synthesize 4k report pairs and train lightweight interpretable metrics on Qwen3-8B and MedGemma-4B. Our trained metric sharpens the clinical significance boundary, surpassing 32B-scale medical LLMs and remaining competitive with proprietary models. Crucially, the more costly two-pass setting fails to consistently improve overall performance and mainly trades discrimination for robustness. These findings suggest one-pass trained metrics as the practical choice for cost-sensitive deployment, with two-pass inference reserved for settings where D-R balance is critical. We will release the dataset and metric.

Attribution-Guided and Coverage-Maximized Pruning for Structural MoE Compression

Mixture-of-Experts (MoE) models scale compute efficiently, yet remain expensive to deploy due to their substantial memory footprint and inference overhead. Prior compression methods mainly operate at the expert level, either removing entire experts or ranking experts by coarse-grained importance scores. However, such expert-wise decisions are often too coarse to capture fine-grained redundancy, leading to misallocated pruning budgets and limited compression. To address this problem, we observe that information within MoE experts is highly concentrated in a small subset of channels, leaving substantial redundancy even in experts deemed important. Based on this observation, we propose a structural pruning framework tailored for MoE models. Our method reformulates prune-ratio allocation as a channel-score coverage maximization problem and solves it efficiently using an attribution-based approximation. Experiments on DeepSeek and Qwen MoE models show that our method preserves model accuracy under 50% or 25% structured pruning when combined with 4-bit quantization. On Qwen3-30B-A3B, our approach reduces memory footprint by 5.27$\times$ and consistently outperforms state-of-the-art baselines across diverse benchmarks.

Kairos: A Native World Model Stack for Physical AI

World models are transitioning from passive visual generators to foundational, operational infrastructure for Physical AI: they must natively acquire world knowledge from heterogeneous experience, maintain persistent states over long horizons, and execute efficiently within real deployment constraints. We introduce Kairos, a native world model stack designed around these requirements. (1) Kairos learns the world by pioneering a Native Pre-training Paradigm governed by a Cross-Embodiment Data Curriculum, which organizes open-world videos, human behavioral data, and robot interactions into a progressive developmental pathway. (2) Kairos maintains the world by unified world understanding, generation, and prediction within a Native Unified Architecture equipped with Hybrid Linear Temporal Attention, where sliding-window attention captures local dynamics, dilated sliding windows capture mid-range dependencies, and gated linear attention maintains persistent global memory. We establish formal theoretical bounds demonstrating that this temporal factorization strictly limits error accumulation, mathematically guaranteeing state propagation across extended horizons. (3) Kairos runs the world by incorporating a Deployment-Aware System Co-Design to support low-latency rollout generation on server and consumer-grade hardware for real-world observation-action-feedback loops. Experiments on embodied world-model, long-horizon, and action-policy benchmarks show that Kairos achieves top level performance while offering a strong efficiency-capability trade-off. Together, these results position Kairos as a cohesive operational foundation for future self-evolving physical intelligence.

Who Pays the Price? Stakeholder-Centric Prompt Injection Benchmarking for Real-world Web Agents

Web agents driven by large language models (LLMs) are increasingly deployed in real-world environments, where they operate over untrusted web content and execute actions with direct consequences. This makes them vulnerable to prompt-injection attacks, in which seemingly benign content embeds adversarial instructions that manipulate agent behaviour. Existing security benchmarks adopt an \textit{attack-centric} perspective, focusing on the technical feasibility of injections while overlooking the nuanced distribution of resulting harms. In practice, however, prompt-injection risk is victim-dependent: a single exploit can produce asymmetric consequences for different stakeholders, and the same attack pattern may exhibit substantially different effectiveness depending on whom it targets. To capture these properties, we introduce \textbf{\sysname}, a \textit{stakeholder-centric} benchmark to systematically categorize and attribute harm in real-world web agent systems. It distinguishes between affected entities (e.g., user, seller, platform), decomposes the attacks into concrete objectives, and evaluates each case with complementary outcome- and process-level metrics. Our results reveal substantial and heterogeneous vulnerabilities: not a single attack objective is reliably resisted by current agents, and failures distribute across qualitatively distinct modes ranging from \emph{stealthy parasitism} (attack succeeds without disrupting the user's delegated task) to \emph{misaligned disruption} (task disrupted without attack success) and \emph{compounded failure} (both adversarial objective and task integrity simultaneously violated). These patterns are missed by conventional evaluation, highlighting the need for stakeholder-aware assessment of LLM-based agents in real-world deployments. Benchmark is available at https://github.com/StakeBench/SBC.

CineDance: Towards Next-Generation Multi-Shot Long-Form Cinematic Audio-Video Generation

The fidelity and structural diversity of training datasets fundamentally determine the capabilities of video generation models. While commercial systems showremarkableabilitytogeneratecinematicnarratives, the progress of open-source models remains limited by the scarcity of high-quality training data. To bridge this gap, we introduce CineDance-1M, a large-scale, open research Text-to-Audio-Video (T2AV) dataset designed specifically for multi-shot, long-form joint audio-video generation. Averaging 92.8 seconds and 24.2 continuous shots per video, it provides configurable, structured annotations for both audio and video modalities. This exceptional quality is achieved through a rigorous three-stage curation pipeline: i) diverse sourcing and comprehensive cleansing, ii) film-theory-inspired narrative parsing, and iii) hierarchical dual-modal captioning. For a comprehensive assessment, we propose CineBench, featuring a diverse prompt suite and a six-dimensional, human-aligned metric system tailored for complex narrative audio-video evaluation. Furthermore, we adapt LTX-2.3 into CineDance, which demonstrates exceptional single-modality quality alongside precise audio-video alignment and robust subject and environment consistency, effectively validating our curation strategy and the high quality of CineDance-1M. We anticipate that this work will serve as a solid foundation for accelerating future research in multi-shot, long-form joint audio-video generation. Our project page is available at https://aliothchen.github.io/projects/CineDance/.

Closed-Form Spectral Regularization for Multi-Task Model Merging

Model merging combines several independently fine-tuned experts into a single multi-task model without any training data, reducing the storage, serving, and decentralized-development costs of large foundation models. State-of-the-art merging methods formulate merging as a layer-wise quadratic interference minimization problem. Although this problem admits an exact closed-form pseudoinverse solution, that solution underperforms hundreds of iterations of gradient descent in practice. The iterative loop dominates the cost of the pipeline, yet its effectiveness has remained unexplained. We revisit this regime and show that the iterative solver does not primarily act as an optimizer; rather, it serves as an implicit spectral regularizer for an ill-posed normal equation, where small-eigenvalue directions of the per-layer interference operator amplify proxy noise. Building on this finding, we formalize multi-task model merging as a noisy linear inverse problem and propose a spectral filtering estimator parameterized by a per-direction filter. We instantiate this estimator with SWUDI, a closed-form method that combines a soft exponential filter, which matches the gradient-flow trajectory of iterative descent, with a hard top-K truncation that suppresses noise-amplifying small-eigenvalue directions. Furthermore, we propose SWUDI-A, an adaptive variant that replaces the global rank hyperparameter with per-layer rank rules, further improving robustness across architectures. Both variants share a single symmetric eigendecomposition per linear layer and require no training data or optimizer state. Across four general benchmarks and a multimodal merging benchmark spanning VQA, Geometry, Chart, OCR, Grounding, and modality merging, our proposed spectral solvers match or outperform state-of-the-art merging methods. Crucially, they reduce wall-clock time by 28-72x and peak GPU memory by up to 50%.

MetaWorld: Scaling Multi-Agent Video World Model from Single-view Video Data

Video world models are a foundational generative technology for embodied AI and the Metaverse, yet existing approaches are inherently limited to a single agent observing from a single perspective. Extending these models to multi-agent settings introduces two critical challenges: data scarcity (coordinated multi-view recordings are prohibitively expensive to collect for general open-domain scenarios) and world state alignment (independently generated video streams cannot ensure that shared physical environments and events evolve consistently across views). To address these challenges, we propose MetaWorld, a novel framework that scales multi-agent video world models to open-domain environments directly from single-view videos. First, we introduce Monocular World-State Unrolling (MWSU) to explicitly decompose monocular footage into the camera operator's ego-motion and the visible subject's spatial trajectory. This camera-trajectory decomposition naturally extracts synchronized multi-agent motion data within a shared 3D space, completely bypassing the need for multi-camera setups. Second, for precise visual control, we develop the Subject-Aware World Generator to enable appearance-driven simulation conditioned on per-agent identity images. Finally, to ensure both views are grounded in the identical physical reality, we propose World-State Alignment, a per-frame inter-branch cross-attention mechanism inserted at every transformer layer of the video DiT. By jointly synchronizing the denoising process, WSA enforces both static geometric consistency and dynamic motion consistency, encouraging that the shared 3D environment and physical events remain well-aligned across both egocentric views. Extensive experiments demonstrate that MetaWorld achieves superior cross-view consistency and identity fidelity, establishing a highly scalable, physics-driven paradigm for multi-agent video world modeling.

Cert-LAS: Toward Certified Model Ownership Verification for Text-to-Image Diffusion Models via Layer-Adaptive Smoothing

Large-scale text-to-image (T2I) diffusion models have enabled unprecedented creative applications, but their unauthorized use has raised serious intellectual property concerns, making model ownership verification (MOV) increasingly critical. We find that existing backdoor-based diffusion watermarking methods often (implicitly) assume a "faithful" verification process, namely, that the verifier can query a suspicious model and obtain the faithful watermark response to complete MOV. However, in practice, adversaries may intentionally or unintentionally damage potential watermark signals, significantly degrading verification reliability. To address this issue, we propose Cert-LAS, the first certified MOV method for T2I models based on layer-adaptive smoothing. In general, Cert-LAS embeds specified watermarks using diffusion classifiers and an LFS-guided layer-adaptive noise, and verifies ownership by examining whether the suspected model exhibits significantly stronger watermark responses compared to unwatermarked references through hypothesis testing. We further prove that, under certain conditions, our Cert-LAS can still achieve reliable verification even in the presence of malicious removal attacks. Extensive experiments validate the effectiveness of Cert-LAS and its resistance to adaptive attacks. Our code is available at https://github.com/Leyi-Qi/Cert-LAS.

Factorize to Generalize: Retrieval-Guided Invariant-Dynamic Decomposition for Time Series Forecasting

Time series foundation models (TSFMs) have recently achieved strong zero-shot forecasting performance through large-scale pretraining and retrieval-augmented prediction. However, our empirical analysis reveals a non-trivial limitation of retrieval-based forecasting: retrieval tends to induce more oscillatory predictions, improving performance on highly fluctuating series while degrading accuracy on smoother, trend-dominated ones. This suggests that retrieved information may be fused into prediction without explicitly distinguishing stable temporal structure from instance-specific variations, which can reduce robustness under distribution shifts. We propose a Retrieval-guided Invariant-Dynamic DEcomposition framework for time series forecasting. Rather than using retrieval as auxiliary predictive context, we leverage retrieved sequences as implicit samples from related environments to guide representation decomposition. Specifically, we first construct a retrieval-aware representation via attention-based aggregation, and then introduce a retrieval-guided routing mechanism to decompose it into an invariant component capturing stable shared structure and a dynamic component modeling context-dependent variations. These two components are forecast separately and fused for final prediction, enabling the model to preserve transferable patterns while remaining adaptive to evolving dynamics. We further design training objectives that encourage invariant learning and disentanglement, and provide theoretical insight showing that retrieval aggregation reduces variance and approximates invariant representation learning without explicit environment supervision. Extensive experiments demonstrate that our method consistently improves robustness under distribution shifts and outperforms existing TSFMs and retrieval-based baselines in zero-shot forecasting settings.

Better, Faster: Harnessing Self-Improvement in Large Reasoning Models

Self-improvement training enables the large reasoning models (LRMs) to improve themselves by self-generating reasoning trajectories as training data without external supervision. However, we find that this method often falls short in complex reasoning tasks and even leads to model collapse. Through a series of preliminary analyses, we reveal two problems: (1) data imbalance, where most training samples are simple, but the challenging yet crucial samples are scarce; (2) overthinking, where many undesired samples with redundant reasoning steps are used for self-training. To this end, we propose HSIR, which effectively Harnesses Self-Improvement in large Reasoning models via two simple-yet-effective approaches. Specifically, HSIR introduces a verify-then-exit sampling strategy to mitigate data imbalance by efficiently collecting more accurate solutions for difficult queries, and designs an Intrinsic Diversity score to quantify overthinking and filter out the undesired solutions. We apply HSIR to various post-training paradigms, among which we further propose H-GRPO, an enhanced GRPO algorithm that leverages the intrinsic diversity as an external reward to encourage concise and diverse reasoning via reinforcement learning. Extensive results show that HSIR not only effectively enhances the reasoning performance, i.e., bringing up to +10.9% average performance gains, but also significantly improves the reasoning efficiency by reducing up to 42.4% relative inference overhead.