MLLMs Get It Right, Then Get It Wrong: Tracing and Correcting Late-Layer Textual Bias

When vision contradicts text, multimodal large language models (MLLMs) consistently favor text, even when images provide clear evidence otherwise. This bias poses risks for applications requiring visual grounding, yet its cause remains unclear. In this paper, we uncover a surprising finding: models often get it right initially, forming correct vision-based predictions in their intermediate layers, before changing their minds and favoring text in the final output. We call this "late-layer textual override". The visual information is encoded, it simply does not survive to the output. More intriguingly, we find that how predictions change reveals whether they're correct: 85% of failures shift toward text, while 89% of successes shift toward vision. This directional signature enables a simple but powerful intervention: when we detect a confident visual prediction being suppressed, we restore it. We propose CALRD (Conflict-Aware Layer Reference Decoding), a training-free method that recovers overridden predictions at inference time. Experiments across five MLLMs of varying architectures demonstrate up to 9.4% absolute improvements on conflict benchmarks while largely preserving standard performance, without training or external knowledge. It recovers what the model already knew but failed to preserve.

StemBind: When MLLMs Get Lost Between Rules and Instances in Abstract Visual Reasoning

Multimodal large language models (MLLMs) often know the rule but pick the wrong answer: on abstract visual reasoning (AVR) tasks, a model can describe what it sees and name the underlying pattern, yet still fail to choose the matching candidate. Existing AVR benchmarks cannot detect this because they collapse perception, rule induction, and answer selection into a single right-or-wrong signal. We introduce StemBind, a shared-stem diagnostic benchmark that probes the same visual stem with three aligned questions: Perception (what is in the image), Rule (what pattern governs it), and Full (which option completes it), so a final-answer error can be attributed to a specific sub-step on the same evidence. StemBind contains 2,298 curated knowledge-light stems across nine auditable visual operations, totaling 19,533 P/R/F tasks, with each full item annotated by Sternberg's four reasoning stages (S1 Encode, S2 Infer, S3 Map, S4 Apply). Evaluating 24 frontier MLLM configurations yields four findings. (i) The R-F chasm: rule accuracy exceeds full-item accuracy on 22 of 24 models, so most failures happen after the rule is identified. (ii) A persistent binding gap: even when P and R are both correct on the same stem, models still answer F incorrectly 51.2% of the time. (iii) The bottleneck is S3: process diagnostics and Stage-wise Stimulus Augmentation localize the dominant failure to rule-to-instance mapping. (iv) Scaling and thinking do not help: neither larger models nor explicit thinking mode reliably closes the gap, and thinking even lowers rule and full-item accuracy. StemBind reframes AVR evaluation from final-answer ranking to locating where abstract visual reasoning breaks down, identifying rule-to-instance binding as a concrete next target for vision-grounded reasoning.

Advantage Collapse in Group Relative Policy Optimization: Diagnosis and Mitigation

Group Relative Policy Optimization (GRPO), a prominent algorithm within the Reinforcement Learning from Verifiable Rewards (RLVR) framework, has achieved strong results in improving the reasoning capabilities of large language models (LLMs). However, GRPO is prone to advantage collapse, a failure mode where homogeneous rewards within a group (e.g., all correct or all incorrect answers) yield near-zero advantages and vanishing gradients. To address this, we introduce the Advantage Collapse Rate (ACR), the first diagnostic metric quantifying the proportion of training batches with ineffective gradients. Across models from 0.5B to 14B parameters on mathematical reasoning benchmarks, we show that ACR strongly predicts training stagnation and final performance. We then propose Adaptive Virtual Sample Policy Optimization (AVSPO), a lightweight extension of GRPO that injects virtual reward samples, guided by real-time ACR monitoring, to enable learning from homogeneous groups without additional model rollouts. AVSPO reduces advantage collapse by 58-63% relative to GRPO and yields consistent accuracy gains of 4-6 percentage points across all model scales, while maintaining generalization on the evaluated out-of-domain task. Code and datasets are available at https://qingyonghu.github.io/AVSPO.

ENC-Bench: A Benchmark for Evaluating Multimodal Large Language Models in Electronic Navigational Chart Understanding

Electronic Navigational Charts (ENCs) are the safety-critical backbone of modern maritime navigation, yet it remains unclear whether multimodal large language models (MLLMs) can reliably interpret them. Unlike natural images or conventional charts, ENCs encode regulations, bathymetry, and route constraints via standardized vector symbols, scale-dependent rendering, and precise geometric structure -- requiring specialized maritime expertise for interpretation. We introduce ENC-Bench, the first benchmark dedicated to professional ENC understanding. ENC-Bench contains 20,490 expert-validated samples from 840 authentic National Oceanic and Atmospheric Administration (NOAA) ENCs, organized into a three-level hierarchy: Perception (symbol and feature recognition), Spatial Reasoning (coordinate localization, bearing, distance), and Maritime Decision-Making (route legality, safety assessment, emergency planning under multiple constraints). All samples are generated from raw S-57 data through a calibrated vector-to-image pipeline with automated consistency checks and expert review. We evaluate 10 state-of-the-art MLLMs such as GPT-4o, Gemini 2.5, Qwen3-VL, InternVL-3, and GLM-4.5V, under a unified zero-shot protocol. The best model achieves only 47.88% accuracy, with systematic challenges in symbolic grounding, spatial computation, multi-constraint reasoning, and robustness to lighting and scale variations. By establishing the first rigorous ENC benchmark, we open a new research frontier at the intersection of specialized symbolic reasoning and safety-critical AI, providing essential infrastructure for advancing MLLMs toward professional maritime applications.

Efficient Hallucination Detection: Adaptive Bayesian Estimation of Semantic Entropy with Guided Semantic Exploration

Large language models (LLMs) have achieved remarkable success in various natural language processing tasks, yet they remain prone to generating factually incorrect outputs known as hallucinations. While recent approaches have shown promise for hallucination detection by repeatedly sampling from LLMs and quantifying the semantic inconsistency among the generated responses, they rely on fixed sampling budgets that fail to adapt to query complexity, resulting in computational inefficiency. We propose an Adaptive Bayesian Estimation framework for Semantic Entropy with Guided Semantic Exploration, which dynamically adjusts sampling requirements based on observed uncertainty. Our approach employs a hierarchical Bayesian framework to model the semantic distribution, enabling dynamic control of sampling iterations through variance-based thresholds that terminate generation once sufficient certainty is achieved. We also develop a perturbation-based importance sampling strategy to systematically explore the semantic space. Extensive experiments on four QA datasets demonstrate that our method achieves superior hallucination detection performance with significant efficiency gains. In low-budget scenarios, our approach requires about 50% fewer samples to achieve comparable detection performance to existing methods, while delivers an average AUROC improvement of 12.6% under the same sampling budget.

Re4: Scientific Computing Agent with Rewriting, Resolution, Review and Revision

Large language models (LLMs) serve as an active and promising field of generative artificial intelligence and have demonstrated abilities to perform complex tasks in multiple domains, including mathematical and scientific reasoning. In this work, we construct a novel agent framework for solving representative problems in scientific computing. The proposed agent, incorporating a "rewriting-resolution-review-revision" logical chain via three reasoning LLMs (functioning as the Consultant, Reviewer, and Programmer, respectively), is integrated in a collaborative and interactive manner. The Consultant module endows the agent with knowledge transfer capabilities to link problems to professional domain insights, thereby rewriting problem descriptions through text augmentation. The Programmer module is responsible for generating and executing well-structured code to deliver the problem resolution. The Reviewer module equips the agent with the capacity for self-debugging and self-refinement through interactive feedback with code runtime outputs. By leveraging the end-to-end review mechanism, the executable code provided by the Programmer attains the iterative revision. A comprehensive evaluation is conducted on the performance of the proposed agent framework in solving PDEs, ill-conditioned linear systems, and data-driven physical analysis problems. Compared to single-model, this collaborative framework significantly improves the bug-free code generation rate and reduces the occurrence of non-physical solutions, thereby establishing a highly reliable framework for autonomous code generation based on natural language descriptions. The review mechanism improved the average execution success (bug-free code and non-NaN solutions) rate of the latest reasoning models. In summary, our agent framework establishes automatic code generation and review as a promising scientific computing paradigm.

AxonCallosumEM Dataset: Axon Semantic Segmentation of Whole Corpus Callosum cross section from EM Images

The electron microscope (EM) remains the predominant technique for elucidating intricate details of the animal nervous system at the nanometer scale. However, accurately reconstructing the complex morphology of axons and myelin sheaths poses a significant challenge. Furthermore, the absence of publicly available, large-scale EM datasets encompassing complete cross sections of the corpus callosum, with dense ground truth segmentation for axons and myelin sheaths, hinders the advancement and evaluation of holistic corpus callosum reconstructions. To surmount these obstacles, we introduce the AxonCallosumEM dataset, comprising a 1.83 times 5.76mm EM image captured from the corpus callosum of the Rett Syndrome (RTT) mouse model, which entail extensive axon bundles. We meticulously proofread over 600,000 patches at a resolution of 1024 times 1024, thus providing a comprehensive ground truth for myelinated axons and myelin sheaths. Additionally, we extensively annotated three distinct regions within the dataset for the purposes of training, testing, and validation. Utilizing this dataset, we develop a fine-tuning methodology that adapts Segment Anything Model (SAM) to EM images segmentation tasks, called EM-SAM, enabling outperforms other state-of-the-art methods. Furthermore, we present the evaluation results of EM-SAM as a baseline.