Automated scoring plays a crucial role in education by reducing the reliance on human raters, offering scalable and immediate evaluation of student work. While large language models (LLMs) have shown strong potential in this task, their use as end-to-end raters faces challenges such as low accuracy, prompt sensitivity, limited interpretability, and rubric misalignment. These issues hinder the implementation of LLM-based automated scoring in assessment practice. To address the limitations, we propose AutoSCORE, a multi-agent LLM framework enhancing automated scoring via rubric-aligned Structured COmponent REcognition. With two agents, AutoSCORE first extracts rubric-relevant components from student responses and encodes them into a structured representation (i.e., Scoring Rubric Component Extraction Agent), which is then used to assign final scores (i.e., Scoring Agent). This design ensures that model reasoning follows a human-like grading process, enhancing interpretability and robustness. We evaluate AutoSCORE on four benchmark datasets from the ASAP benchmark, using both proprietary and open-source LLMs (GPT-4o, LLaMA-3.1-8B, and LLaMA-3.1-70B). Across diverse tasks and rubrics, AutoSCORE consistently improves scoring accuracy, human-machine agreement (QWK, correlations), and error metrics (MAE, RMSE) compared to single-agent baselines, with particularly strong benefits on complex, multi-dimensional rubrics, and especially large relative gains on smaller LLMs. These results demonstrate that structured component recognition combined with multi-agent design offers a scalable, reliable, and interpretable solution for automated scoring.

Reward is critical to the evaluation and training of large language models (LLMs). However, existing rule-based or model-based reward methods struggle to generalize to GUI agents, where access to ground-truth trajectories or application databases is often unavailable, and static trajectory-based LLM-as-a-Judge approaches suffer from limited accuracy. To address these challenges, we propose ProRe, a proactive reward system that leverages a general-purpose reasoner and domain-specific evaluator agents (actors). The reasoner schedules targeted state probing tasks, which the evaluator agents then execute by actively interacting with the environment to collect additional observations. This enables the reasoner to assign more accurate and verifiable rewards to GUI agents. Empirical results on over 3K trajectories demonstrate that ProRe improves reward accuracy and F1 score by up to 5.3% and 19.4%, respectively. Furthermore, integrating ProRe with state-of-the-art policy agents yields a success rate improvement of up to 22.4%.

Reinforcement fine-tuning (RFT) often suffers from \emph{reward over-optimization}, where a policy model hacks the reward signals to achieve high scores while producing low-quality outputs. Our theoretical analysis shows that the key lies in reward misspecification at the high-reward tail: the inability to reliably distinguish Excellent responses from merely Great ones. This motivate us to focus on the high-reward region. However, such tail examples are scarce under the base LLM. While off-policy exemplars (e.g. from stronger models or rewrites) are easier to obtain, naively training on them yields a misspecified reward for the policy we aim to align. To address this, we study rubric-based rewards. By design, rubrics can leverage off-policy examples while remaining insensitive to their artifacts. To elicit rubrics that capture the high-reward tail, we highlight the importance of distinguishing among great and diverse responses, and introduce a workflow to implement this idea. We empirically demonstrate that rubric-based rewards substantially mitigate reward over-optimization and deliver effective LLM post-training improvements. Our code can be accessed at https://github.com/Jun-Kai-Zhang/rubrics.git .

Reinforcement learning with verifiable rewards (RLVR) has become a leading approach for improving large language model (LLM) reasoning capabilities. Most current methods follow variants of Group Relative Policy Optimization, which samples multiple reasoning completions, scores them relative to each other, and adjusts the policy accordingly. However, these approaches invariably sample discrete tokens at each reasoning step, discarding the rich distributional information in the model's probability distribution over candidate tokens. While preserving and utilizing this distributional information has proven beneficial in non-RL settings, current RLVR methods seem to be unnecessarily constraining the reasoning search space by not using this information. To address this limitation, we investigate mixture-of-token generation (MoT-G) in RLVR. We present a unified framework that generalizes existing MoT-G approaches, including existing training-free methods that construct mixture embeddings as weighted sums over token embeddings, and extend RLVR to operate directly in this continuous mixture space for generating chain-of-thought. Evaluating two MoT-G variants on Reasoning-Gym, a suite of reasoning-intensive language tasks, we find that MoT--G methods achieve substantial improvements (5--35 \% gains on 7 out of 10 tasks) compared to standard decoding with the Qwen2.5-1.5B model, while reaching comparable accuracy with half the number of trajectories, suggesting improved training efficiency. Through comprehensive hidden-state and token-level analyses, we provide evidence that MoT--G's benefits may stem from its ability to maintain higher hidden-state entropy throughout the reasoning process and promote exploration in token space.

Diabetes mellitus is a major global health challenge, affecting over half a billion adults worldwide with prevalence projected to rise. Although the American Diabetes Association (ADA) provides clear diagnostic thresholds, early recognition remains difficult due to vague symptoms, borderline laboratory values, gestational complexity, and the demands of long-term monitoring. Advances in large language models (LLMs) offer opportunities to enhance decision support through structured, interpretable, and patient-friendly outputs. This study evaluates GPT-5, the latest generative pre-trained transformer, using a simulation framework built entirely on synthetic cases aligned with ADA Standards of Care 2025 and inspired by public datasets including NHANES, Pima Indians, EyePACS, and MIMIC-IV. Five representative scenarios were tested: symptom recognition, laboratory interpretation, gestational diabetes screening, remote monitoring, and multimodal complication detection. For each, GPT-5 classified cases, generated clinical rationales, produced patient explanations, and output structured JSON summaries. Results showed strong alignment with ADA-defined criteria, suggesting GPT-5 may function as a dual-purpose tool for clinicians and patients, while underscoring the importance of reproducible evaluation frameworks for responsibly assessing LLMs in healthcare.

Current molecular generative models primarily focus on improving drug-target binding affinity and specificity, often neglecting the system-level phenotypic effects elicited by compounds. Transcriptional profiles, as molecule-level readouts of drug-induced phenotypic shifts, offer a powerful opportunity to guide molecular design in a phenotype-aware manner. We present PhenoMoler, a phenotype-guided molecular generation framework that integrates a chemistry large language model with expression profiles to enable biologically informed drug design. By conditioning the generation on drug-induced differential expression signatures, PhenoMoler explicitly links transcriptional responses to chemical structure. By selectively masking and reconstructing specific substructures-scaffolds, side chains, or linkers-PhenoMoler supports fine-grained, controllable molecular optimization. Extensive experiments demonstrate that PhenoMoler generates chemically valid, novel, and diverse molecules aligned with desired phenotypic profiles. Compared to FDA-approved drugs, the generated compounds exhibit comparable or enhanced drug-likeness (QED), optimized physicochemical properties, and superior binding affinity to key cancer targets. These findings highlight PhenoMoler's potential for phenotype-guided and structure-controllable molecular optimization.

Large language model (LLM) providers boast big numbers for maximum context window sizes. To test the real world use of context windows, we 1) define a concept of maximum effective context window, 2) formulate a testing method of a context window's effectiveness over various sizes and problem types, and 3) create a standardized way to compare model efficacy for increasingly larger context window sizes to find the point of failure. We collected hundreds of thousands of data points across several models and found significant differences between reported Maximum Context Window (MCW) size and Maximum Effective Context Window (MECW) size. Our findings show that the MECW is, not only, drastically different from the MCW but also shifts based on the problem type. A few top of the line models in our test group failed with as little as 100 tokens in context; most had severe degradation in accuracy by 1000 tokens in context. All models fell far short of their Maximum Context Window by as much as 99 percent. Our data reveals the Maximum Effective Context Window shifts based on the type of problem provided, offering clear and actionable insights into how to improve model accuracy and decrease model hallucination rates.

Text-to-image (T2I) models have been widely applied in generating high-fidelity images across various domains. However, these models may also be abused to produce Not-Safe-for-Work (NSFW) content via jailbreak attacks. Existing jailbreak methods primarily manipulate the textual prompt, leaving potential vulnerabilities in image-based inputs largely unexplored. Moreover, text-based methods face challenges in bypassing the model's safety filters. In response to these limitations, we propose the Multimodal Prompt Decoupling Attack (MPDA), which utilizes image modality to separate the harmful semantic components of the original unsafe prompt. MPDA follows three core steps: firstly, a large language model (LLM) decouples unsafe prompts into pseudo-safe prompts and harmful prompts. The former are seemingly harmless sub-prompts that can bypass filters, while the latter are sub-prompts with unsafe semantics that trigger filters. Subsequently, the LLM rewrites the harmful prompts into natural adversarial prompts to bypass safety filters, which guide the T2I model to modify the base image into an NSFW output. Finally, to ensure semantic consistency between the generated NSFW images and the original unsafe prompts, the visual language model generates image captions, providing a new pathway to guide the LLM in iterative rewriting and refining the generated content.

This study aimed to enhance disease classification accuracy from retinal fundus images by integrating fine-grained image features and global textual context using a novel multimodal deep learning architecture. Existing multimodal large language models (MLLMs) often struggle to capture low-level spatial details critical for diagnosing retinal diseases such as glaucoma, diabetic retinopathy, and retinitis pigmentosa. This model development and validation study was conducted on 1,305 fundus image-text pairs compiled from three public datasets (FIVES, HRF, and StoneRounds), covering acquired and inherited retinal diseases, and evaluated using classification accuracy and F1-score. The MDF-MLLM integrates skip features from four U-Net encoder layers into cross-attention blocks within a LLaMA 3.2 11B MLLM. Vision features are patch-wise projected and fused using scaled cross-attention and FiLM-based U-Net modulation. Baseline MLLM achieved 60% accuracy on the dual-type disease classification task. MDF-MLLM, with both U-Net and MLLM components fully fine-tuned during training, achieved a significantly higher accuracy of 94%, representing a 56% improvement. Recall and F1-scores improved by as much as 67% and 35% over baseline, respectively. Ablation studies confirmed that the multi-depth fusion approach contributed to substantial gains in spatial reasoning and classification, particularly for inherited diseases with rich clinical text. MDF-MLLM presents a generalizable, interpretable, and modular framework for fundus image classification, outperforming traditional MLLM baselines through multi-scale feature fusion. The architecture holds promise for real-world deployment in clinical decision support systems. Future work will explore synchronized training techniques, a larger pool of diseases for more generalizability, and extending the model for segmentation tasks.

With the emergence of large-scale vision language models (VLM), it is now possible to produce realistic-looking radiology reports for chest X-ray images. However, their clinical translation has been hampered by the factual errors and hallucinations in the produced descriptions during inference. In this paper, we present a novel phrase-grounded fact-checking model (FC model) that detects errors in findings and their indicated locations in automatically generated chest radiology reports. Specifically, we simulate the errors in reports through a large synthetic dataset derived by perturbing findings and their locations in ground truth reports to form real and fake findings-location pairs with images. A new multi-label cross-modal contrastive regression network is then trained on this dataset. We present results demonstrating the robustness of our method in terms of accuracy of finding veracity prediction and localization on multiple X-ray datasets. We also show its effectiveness for error detection in reports of SOTA report generators on multiple datasets achieving a concordance correlation coefficient of 0.997 with ground truth-based verification, thus pointing to its utility during clinical inference in radiology workflows.