Objective: To evaluate large language models (LLMs) against supervised baselines for fine-grained, lesion-level detection of incidentalomas requiring follow-up, addressing the limitations of current document-level classification systems. Methods: We utilized a dataset of 400 annotated radiology reports containing 1,623 verified lesion findings. We compared three supervised transformer-based encoders (BioClinicalModernBERT, ModernBERT, Clinical Longformer) against four generative LLM configurations (Llama 3.1-8B, GPT-4o, GPT-OSS-20b). We introduced a novel inference strategy using lesion-tagged inputs and anatomy-aware prompting to ground model reasoning. Performance was evaluated using class-specific F1-scores. Results: The anatomy-informed GPT-OSS-20b model achieved the highest performance, yielding an incidentaloma-positive macro-F1 of 0.79. This surpassed all supervised baselines (maximum macro-F1: 0.70) and closely matched the inter-annotator agreement of 0.76. Explicit anatomical grounding yielded statistically significant performance gains across GPT-based models (p < 0.05), while a majority-vote ensemble of the top systems further improved the macro-F1 to 0.90. Error analysis revealed that anatomy-aware LLMs demonstrated superior contextual reasoning in distinguishing actionable findings from benign lesions. Conclusion: Generative LLMs, when enhanced with structured lesion tagging and anatomical context, significantly outperform traditional supervised encoders and achieve performance comparable to human experts. This approach offers a reliable, interpretable pathway for automated incidental finding surveillance in radiology workflows. Introduction Incidental findings, or incidentalomas, refer to unexpected abnormalities discovered during imaging studies performed for unrelated reasons [1]. Their detection has increased as imaging utilization has grown across healthcare. These findings create a clinical dilemma, since most are benign while some represent early-stage disease that requires intervention.

In recent times, denoising diffusion probabilistic models (DPMs) have proven to show significant success in medical image generation and denoising, while also serving as powerful representation learners for downstream tasks such as segmentation. However, their effectiveness in segmentation is limited by the need for detailed pixel-wise annotations, which are expensive, time-consuming, and require expert knowledge--a significant bottleneck in real-world clinical applications. In order to mitigate this limitation of label-efficiency, we propose a fast and efficient model named FastTextDiff, a diffusion-based segmentation model that integrates medical text annotations to enhance semantic representations. Our approach leverages ModernBERT [3], a transformer-based language model capable of processing long medical text sequences, to establish a strong connection between textual annotations and semantic meaning in medical imaging. ModernBERT can efficiently encode clinical knowledge for directing segmentation tasks since it has been trained on both MIMIC-III [19] and MIMIC-IV. Label-efficient segmentation with enhanced performance is made possible by cross-modal attention processes, which enable smooth interaction between visual and textual modalities. This study validates ModernBERT as a quick and scalable substitute for Clinical BioBERT [4] in diffusion-based segmentation pipelines [2] and demonstrates the promise of multi-modal techniques for medical image analysis.

This study explored whether supervised machine learning and deep learning models can effectively distinguish perceived lower-quality news articles from perceived higher-quality news articles. 3 machine learning classifiers and 3 deep learning models were assessed using a newly created dataset of 1,412,272 English news articles from the Common Crawl over 2018-2024. Expert consensus ratings on 579 source websites were split at the median, creating perceived low and high-quality classes of about 706,000 articles each, with 194 linguistic features per website-level labelled article. Traditional machine learning classifiers such as the Random Forest demonstrated capable performance (0.7355 accuracy, 0.8131 ROC AUC). For deep learning, ModernBERT-large (256 context length) achieved the best performance (0.8744 accuracy; 0.9593 ROC-AUC; 0.8739 F1), followed by DistilBERT-base (512 context length) at 0.8685 accuracy and 0.9554 ROC-AUC. DistilBERT-base (256 context length) reached 0.8478 accuracy and 0.9407 ROC-AUC, while ModernBERT-base (256 context length) attained 0.8569 accuracy and 0.9470 ROC-AUC. These results suggest that the perceived quality of worldwide news articles can be effectively differentiated by traditional CPU-based machine learning classifiers and deep learning classifiers.

Pretrained transformer-encoder models like DeBERTaV3 and ModernBERT introduce architectural advancements aimed at improving efficiency and performance. Although the authors of ModernBERT report improved performance over DeBERTaV3 on several benchmarks, the lack of disclosed training data and the absence of comparisons using a shared dataset make it difficult to determine whether these gains are due to architectural improvements or differences in training data. In this work, we conduct a controlled study by pretraining ModernBERT on the same dataset as CamemBERTaV2, a DeBERTaV3 French model, isolating the effect of model design. Our results show that the previous model generation remains superior in sample efficiency and overall benchmark performance, with ModernBERT's primary advantage being its support for long context, faster training, and inference speed. However, the new proposed model still provides meaningful architectural improvements compared to earlier models such as BERT and RoBERTa. Additionally, we observe that high-quality pre-training data accelerates convergence but does not significantly improve final performance, suggesting potential benchmark saturation. These findings show the importance of disentangling pretraining data from architectural innovations when evaluating transformer models.

Critical thinking represents a fundamental competency in today's education landscape. Developing critical thinking skills through timely assessment and feedback is crucial; however, there has not been extensive work in the learning analytics community on defining, measuring, and supporting critical thinking. In this paper, we investigate the feasibility of measuring core "subskills" that underlie critical thinking. We ground our work in an authentic task where students operationalize critical thinking: student-written argumentative essays. We developed a coding rubric based on an established skills progression and completed human coding for a corpus of student essays. We then evaluated three distinct approaches to automated scoring: zero-shot prompting, few-shot prompting, and supervised fine-tuning, implemented across three large language models (GPT-5, GPT-5-mini, and ModernBERT). GPT-5 with few-shot prompting achieved the strongest results and demonstrated particular strength on subskills with separable, frequent categories, while lower performance was observed for subskills that required detection of subtle distinctions or rare categories. Our results underscore critical trade-offs in automated critical thinking assessment: proprietary models offer superior reliability at higher cost, while open-source alternatives provide practical accuracy with reduced sensitivity to minority categories. Our work represents an initial step toward scalable assessment of higher-order reasoning skills across authentic educational contexts.

Retrieval-Augmented Generation (RAG) is a powerful technique for enriching Large Language Models (LLMs) with external knowledge, allowing for factually grounded responses, a critical requirement in high-stakes domains such as healthcare. However, the efficacy of RAG systems is fundamentally restricted by the performance of their retrieval module, since irrelevant or semantically misaligned documents directly compromise the accuracy of the final generated response. General-purpose dense retrievers can struggle with the nuanced language of specialised domains, while the high accuracy of in-domain models is often achieved at prohibitive computational costs. In this work, we aim to address this trade-off by developing and evaluating a two-stage retrieval architecture that combines a lightweight ModernBERT bidirectional encoder for efficient initial candidate retrieval with a ColBERTv2 late-interaction model for fine-grained re-ranking. We conduct comprehensive evaluations of our retriever module performance and RAG system performance in the biomedical context, fine-tuning the IR module using 10k question-passage pairs from PubMedQA. Our analysis of the retriever module confirmed the positive impact of the ColBERT re-ranker, which improved Recall@3 by up to 4.2 percentage points compared to its retrieve-only counterpart. When integrated into the biomedical RAG, our IR module leads to a state-of-the-art average accuracy of 0.4448 on the five tasks of the MIRAGE question-answering benchmark, outperforming strong baselines such as MedCPT (0.4436). Our ablation studies reveal that this performance is critically dependent on a joint fine-tuning process that aligns the retriever and re-ranker; otherwise, the re-ranker might degrade the performance.

The growth of highly advanced Large Language Models (LLMs) constitutes a huge dual-use problem, making it necessary to create dependable AI-generated text detection systems. Modern detectors are notoriously vulnerable to adversarial attacks, with paraphrasing standing out as an effective evasion technique that foils statistical detection. This paper presents a comparative study of adversarial robustness, first by quantifying the limitations of standard adversarial training and then by introducing a novel, significantly more resilient detection framework: Perturbation-Invariant Feature Engineering (PIFE), a framework that enhances detection by first transforming input text into a standardized form using a multi-stage normalization pipeline, it then quantifies the transformation's magnitude using metrics like Levenshtein distance and semantic similarity, feeding these signals directly to the classifier. We evaluate both a conventionally hardened Transformer and our PIFE-augmented model against a hierarchical taxonomy of character-, word-, and sentence-level attacks. Our findings first confirm that conventional adversarial training, while resilient to syntactic noise, fails against semantic attacks, an effect we term "semantic evasion threshold", where its True Positive Rate at a strict 1% False Positive Rate plummets to 48.8%. In stark contrast, our PIFE model, which explicitly engineers features from the discrepancy between a text and its canonical form, overcomes this limitation. It maintains a remarkable 82.6% TPR under the same conditions, effectively neutralizing the most sophisticated semantic attacks. This superior performance demonstrates that explicitly modeling perturbation artifacts, rather than merely training on them, is a more promising path toward achieving genuine robustness in the adversarial arms race.

Encoder-only languages models are frequently used for a variety of standard machine learning tasks, including classification and retrieval. However, there has been a lack of recent research for encoder models, especially with respect to multilingual models. We introduce mmBERT, an encoder-only language model pretrained on 3T tokens of multilingual text in over 1800 languages. To build mmBERT we introduce several novel elements, including an inverse mask ratio schedule and an inverse temperature sampling ratio. We add over 1700 low-resource languages to the data mix only during the decay phase, showing that it boosts performance dramatically and maximizes the gains from the relatively small amount of training data. Despite only including these low-resource languages in the short decay phase we achieve similar classification performance to models like OpenAI's o3 and Google's Gemini 2.5 Pro. Overall, we show that mmBERT significantly outperforms the previous generation of models on classification and retrieval tasks -- on both high and low-resource languages.

Illegal marketplaces have increasingly shifted to concealed parts of the internet, including the deep and dark web, as well as platforms such as Telegram, Reddit, and Pastebin. These channels enable the anonymous trade of illicit goods including drugs, weapons, and stolen credentials. Detecting and categorizing such content remains challenging due to limited labeled data, the evolving nature of illicit language, and the structural heterogeneity of online sources. This paper presents a hierarchical classification framework that combines fine-tuned language models with a semi-supervised ensemble learning strategy to detect and classify illicit marketplace content across diverse platforms. We extract semantic representations using ModernBERT, a transformer model for long documents, finetuned on domain-specific data from deep and dark web pages, Telegram channels, Subreddits, and Pastebin pastes to capture specialized jargon and ambiguous linguistic patterns. In addition, we incorporate manually engineered features such as document structure, embedded patterns including Bitcoin addresses, emails, and IPs, and metadata, which complement language model embeddings. The classification pipeline operates in two stages. The first stage uses a semi-supervised ensemble of XGBoost, Random Forest, and SVM with entropy-based weighted voting to detect sales-related documents. The second stage further classifies these into drug, weapon, or credential sales. Experiments on three datasets, including our multi-source corpus, DUTA, and CoDA, show that our model outperforms several baselines, including BERT, ModernBERT, DarkBERT, ALBERT, Longformer, and BigBird. The model achieves an accuracy of 0.96489, an F1-score of 0.93467, and a TMCC of 0.95388, demonstrating strong generalization, robustness under limited supervision, and effectiveness in real-world illicit content detection.

Current evaluations of LLMs in the government domain primarily focus on safety considerations in specific scenarios, while the assessment of the models' own core capabilities, particularly domain relevance, remains insufficient. To address this gap, we propose GovRelBench, a benchmark specifically designed for evaluating the core capabilities of LLMs in the government domain. GovRelBench consists of government domain prompts and a dedicated evaluation tool, GovRelBERT. During the training process of GovRelBERT, we introduce the SoftGovScore method: this method trains a model based on the ModernBERT architecture by converting hard labels to soft scores, enabling it to accurately compute the text's government domain relevance score. This work aims to enhance the capability evaluation framework for large models in the government domain, providing an effective tool for relevant research and practice. Our code and dataset are available at https://github.com/pan-xi/GovRelBench.