Recent progress in vision-language segmentation has significantly advanced grounded visual understanding. However, these models often exhibit hallucinations by producing segmentation masks for objects not grounded in the image content or by incorrectly labeling irrelevant regions. Existing evaluation protocols for segmentation hallucination primarily focus on label or textual hallucinations without manipulating the visual context, limiting their capacity to diagnose critical failures. In response, we introduce HalluSegBench, the first benchmark specifically designed to evaluate hallucinations in visual grounding through the lens of counterfactual visual reasoning. Our benchmark consists of a novel dataset of 1340 counterfactual instance pairs spanning 281 unique object classes, and a set of newly introduced metrics that quantify hallucination sensitivity under visually coherent scene edits. Experiments on HalluSegBench with state-of-the-art vision-language segmentation models reveal that vision-driven hallucinations are significantly more prevalent than label-driven ones, with models often persisting in false segmentation, highlighting the need for counterfactual reasoning to diagnose grounding fidelity. less

We train models to Predict Ego-centric Video from human Actions (PEVA), given the past video and an action represented by the relative 3D body pose. By conditioning on kinematic pose trajectories, structured by the joint hierarchy of the body, our model learns to simulate how physical human actions shape the environment from a first-person point of view. We train an auto-regressive conditional diffusion transformer on Nymeria, a large-scale dataset of real-world egocentric video and body pose capture. We further design a hierarchical evaluation protocol with increasingly challenging tasks, enabling a comprehensive analysis of the model's embodied prediction and control abilities. Our work represents an initial attempt to tackle the challenges of modeling complex real-world environments and embodied agent behaviors with video prediction from the perspective of a human. less

With the rapid development of digital technology, AI-driven psychological counseling has gradually become an important research direction in the field of mental health. However, existing models still have deficiencies in dialogue safety, detailed scenario handling, and lightweight deployment. To address these issues, this study proposes PsyLite, a lightweight psychological counseling large language model agent developed based on the base model InternLM2.5-7B-chat. Through a two-stage training strategy (hybrid distillation data fine-tuning and ORPO preference optimization), PsyLite enhances the model's deep-reasoning ability, psychological counseling ability, and safe dialogue ability. After deployment using Ollama and Open WebUI, a custom workflow is created with Pipelines. An innovative conditional RAG is designed to introduce crosstalk humor elements at appropriate times during psychological counseling to enhance user experience and decline dangerous requests to strengthen dialogue safety. Evaluations show that PsyLite outperforms the baseline models in the Chinese general evaluation (CEval), psychological counseling professional evaluation (CPsyCounE), and dialogue safety evaluation (SafeDialBench), particularly in psychological counseling professionalism (CPsyCounE score improvement of 47.6\%) and dialogue safety (\safe{} score improvement of 2.4\%). Additionally, the model uses quantization technology (GGUF q4\_k\_m) to achieve low hardware deployment (5GB memory is sufficient for operation), providing a feasible solution for psychological counseling applications in resource-constrained environments. less

Large language models (LLMs) are regularly evaluated using benchmark datasets. But what justifies making inferences about an LLM's capabilities based on its answers to a curated set of questions? This paper first introduces a formal framework to address this question. The key is to note that the benchmarks used to test LLMs -- such as AP exams -- are also those used to test people. However, this raises an implication: these benchmarks are only valid tests if LLMs misunderstand concepts in ways that mirror human misunderstandings. Otherwise, success on benchmarks only demonstrates potemkin understanding: the illusion of understanding driven by answers irreconcilable with how any human would interpret a concept. We present two procedures for quantifying the existence of potemkins: one using a specially designed benchmark in three domains, the other using a general procedure that provides a lower-bound on their prevalence. We find that potemkins are ubiquitous across models, tasks, and domains. We also find that these failures reflect not just incorrect understanding, but deeper internal incoherence in concept representations. less

Extended Stability Runge-Kutta (ESRK) methods are crucial for solving large-scale computational problems in science and engineering, including weather forecasting, aerodynamic analysis, and complex biological modelling. However, balancing accuracy, stability, and computational efficiency remains challenging, particularly for high-order, low-storage schemes. This study introduces a hybrid Genetic Algorithm (GA) and Reinforcement Learning (RL) approach for automated heuristic discovery, optimising low-storage ESRK methods. Unlike traditional approaches that rely on manually designed heuristics or exhaustive numerical searches, our method leverages GA-driven mutations for search-space exploration and an RL-inspired state transition mechanism to refine heuristic selection dynamically. This enables systematic parameter reduction, preserving fourth-order accuracy while significantly improving computational efficiency.The proposed GA-RL heuristic optimisation framework is validated through rigorous testing on benchmark problems, including the 1D and 2D Brusselator systems and the steady-state Navier-Stokes equations. The best-performing heuristic achieves a 25\% reduction in IPOPT runtime compared to traditional ESRK optimisation processes while maintaining numerical stability and accuracy. These findings demonstrate the potential of adaptive heuristic discovery to improve resource efficiency in high-fidelity simulations and broaden the applicability of low-storage Runge-Kutta methods in real-world computational fluid dynamics, physics simulations, and other demanding fields. This work establishes a new paradigm in heuristic optimisation for numerical methods, opening pathways for further exploration using Deep RL and AutoML-based heuristic search less

Causal reasoning capability is critical in advancing large language models (LLMs) toward strong artificial intelligence. While versatile LLMs appear to have demonstrated capabilities in understanding contextual causality and providing responses that obey the laws of causality, it remains unclear whether they perform genuine causal reasoning akin to humans. However, current evidence indicates the contrary. Specifically, LLMs are only capable of performing shallow (level-1) causal reasoning, primarily attributed to the causal knowledge embedded in their parameters, but they lack the capacity for genuine human-like (level-2) causal reasoning. To support this hypothesis, methodologically, we delve into the autoregression mechanism of transformer-based LLMs, revealing that it is not inherently causal. Empirically, we introduce a new causal Q&A benchmark called CausalProbe-2024, whose corpora are fresh and nearly unseen for the studied LLMs. The LLMs exhibit a significant performance drop on CausalProbe-2024 compared to earlier benchmarks, indicating the fact that they primarily engage in level-1 causal reasoning. To bridge the gap towards level-2 causal reasoning, we draw inspiration from the fact that human reasoning is usually facilitated by general knowledge and intended goals. We propose G^2-Reasoner, a method that incorporates general knowledge and goal-oriented prompts into LLMs' causal reasoning processes. Experiments demonstrate that G^2-Reasoner significantly enhances LLMs' causal reasoning capability, particularly in fresh and counterfactual contexts. This work sheds light on a new path for LLMs to advance towards genuine causal reasoning, going beyond level-1 and making strides towards level-2. less

Grokking, i.e., test performance keeps improving long after training loss converged, has been recently witnessed in neural network training, making the mechanism of generalization and other emerging capabilities such as reasoning mysterious. While prior studies usually train small models on a few toy or highly-specific tasks for thousands of epochs, we conduct the first study of grokking on checkpoints during one-pass pretraining of a 7B large language model (LLM), i.e., OLMoE. We compute the training loss and evaluate generalization on diverse benchmark tasks, including math reasoning, code generation, and commonsense/domain-specific knowledge retrieval tasks. Our study, for the first time, verifies that grokking still happens in the pretraining of large-scale foundation models, though different data may enter grokking stages asynchronously. We further demystify grokking's "emergence of generalization" by investigating LLM internal dynamics. Specifically, we find that training samples' pathways (i.e., expert choices across layers) evolve from random, instance-specific to more structured and shareable between samples during grokking. Also, the complexity of a sample's pathway reduces despite the converged loss. These indicate a memorization-to-generalization conversion, providing a mechanistic explanation of delayed generalization. In the study, we develop two novel metrics to quantify pathway distance and the complexity of a single pathway. We show their ability to predict the generalization improvement on diverse downstream tasks. They are efficient, simple to compute and solely dependent on training data. Hence, they have practical value for pretraining, enabling us to monitor the generalization performance without finetuning and test. Theoretically, we show that more structured pathways reduce model complexity and improve the generalization bound. less

Multivariate time series anomaly detection (MTS-AD) is critical in domains like healthcare, cybersecurity, and industrial monitoring, yet remains challenging due to complex inter-variable dependencies, temporal dynamics, and sparse anomaly labels. We introduce mTSBench, the largest benchmark to date for MTS-AD and unsupervised model selection, spanning 344 labeled time series across 19 datasets and 12 diverse application domains. mTSBench evaluates 24 anomaly detection methods, including large language model (LLM)-based detectors for multivariate time series, and systematically benchmarks unsupervised model selection techniques under standardized conditions. Consistent with prior findings, our results confirm that no single detector excels across datasets, underscoring the importance of model selection. However, even state-of-the-art selection methods remain far from optimal, revealing critical gaps. mTSBench provides a unified evaluation suite to enable rigorous, reproducible comparisons and catalyze future advances in adaptive anomaly detection and robust model selection. less

Multimodal understanding of tables in real-world contexts is challenging due to the complexity of structure, symbolic density, and visual degradation (blur, skew, watermarking, incomplete structures or fonts, multi-span or hierarchically nested layouts). Existing multimodal large language models (MLLMs) struggle with such WildStruct conditions, resulting in limited performance and poor generalization. To address these challenges, we propose TableMoE, a neuro-symbolic Mixture-of-Connector-Experts (MoCE) architecture specifically designed for robust, structured reasoning over multimodal table data. TableMoE features an innovative Neuro-Symbolic Routing mechanism, which predicts latent semantic token roles (e.g., header, data cell, axis, formula) and dynamically routes table elements to specialized experts (Table-to-HTML, Table-to-JSON, Table-to-Code) using a confidence-aware gating strategy informed by symbolic reasoning graphs. To facilitate effective alignment-driven pretraining, we introduce the large-scale TableMoE-Align dataset, consisting of 1.2M table-HTML-JSON-code quadruples across finance, science, biomedicine and industry, utilized exclusively for model pretraining. For evaluation, we curate and release four challenging WildStruct benchmarks: WMMFinQA, WMMTatQA, WMMTabDialog, and WMMFinanceMath, designed specifically to stress-test models under real-world multimodal degradation and structural complexity. Experimental results demonstrate that TableMoE significantly surpasses existing state-of-the-art models. Extensive ablation studies validate each core component, emphasizing the critical role of Neuro-Symbolic Routing and structured expert alignment. Through qualitative analyses, we further showcase TableMoE's interpretability and enhanced robustness, underscoring the effectiveness of integrating neuro-symbolic reasoning for multimodal table understanding. less

Can Vision Language Models (VLMs) imagine the full scene from just a few views, like humans do? Humans form spatial mental models, internal representations of unseen space, to reason about layout, perspective, and motion. Our new MindCube benchmark with 21,154 questions across 3,268 images exposes this critical gap, where existing VLMs exhibit near-random performance. Using MindCube, we systematically evaluate how well VLMs build robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for "what-if" movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps. The significant improvement comes from a synergistic approach, "map-then-reason", that jointly trains the model to first generate a cognitive map and then reason upon it. By training models to reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of unobservable space. less

Agentic search such as Deep Research systems, where large language models autonomously browse the web, synthesize information, and return comprehensive citation-backed answers, represents a major shift in how users interact with web-scale information. While promising greater efficiency and cognitive offloading, the growing complexity and open-endedness of agentic search have outpaced existing evaluation benchmarks and methodologies, which largely assume short search horizons and static answers. In this paper, we introduce Mind2Web 2, a benchmark of 130 realistic, high-quality, and long-horizon tasks that require real-time web browsing and extensive information synthesis, constructed with over 1,000 hours of human labor. To address the challenge of evaluating time-varying and complex answers, we propose a novel Agent-as-a-Judge framework. Our method constructs task-specific judge agents based on a tree-structured rubric design to automatically assess both answer correctness and source attribution. We conduct a comprehensive evaluation of nine frontier agentic search systems and human performance, along with a detailed error analysis to draw insights for future development. The best-performing system, OpenAI Deep Research, can already achieve 50-70% of human performance while spending half the time, showing a great potential. Altogether, Mind2Web 2 provides a rigorous foundation for developing and benchmarking the next generation of agentic search systems. less

Large Language Models (LLMs) have become indispensable in real-world applications. However, their widespread adoption raises significant safety concerns, particularly in responding to socially harmful questions. Despite substantial efforts to improve model safety through alignment, aligned models can still have their safety protections undermined by subsequent fine-tuning - even when the additional training data appears benign. In this paper, we empirically demonstrate that this vulnerability stems from the sensitivity of safety-critical low-rank subspaces in LLM parameters to fine-tuning. Building on this insight, we propose a novel training-free method, termed Low-Rank Extrapolation (LoX), to enhance safety robustness by extrapolating the safety subspace of an aligned LLM. Our experimental results confirm the effectiveness of LoX, demonstrating significant improvements in robustness against both benign and malicious fine-tuning attacks while preserving the model's adaptability to new tasks. For instance, LoX leads to 11% to 54% absolute reductions in attack success rates (ASR) facing benign or malicious fine-tuning attacks. By investigating the ASR landscape of parameters, we attribute the success of LoX to that the extrapolation moves LLM parameters to a flatter zone, thereby less sensitive to perturbations. The code is available at github.com/VITA-Group/LoX. less