Search Results (3,676)

This study introduces a hybrid methodological approach for personalised clinical decision support, integrating SHACL-based deterministic constraints with Bayesian probabilistic models. The primary goal is to validate the model and demonstrate the benefits of combining encoded clinical knowledge with probabilistic uncertainties in managing complex therapeutic scenarios. The framework was applied to recurrent urinary tract infections (UTIs) in postmenopausal patients, a clinical context marked by high frequency, treatment challenges, and potential conflicts among therapeutic guidelines. Realistic simulated case studies were developed, encompassing both simple clinical profiles and complex situations, such as patients with antibiotic resistance. Each profile was modelled in RDF/Turtle, enabling semantic representation of clinical features and therapeutic rules. The system automatically calculates success and failure probabilities for different therapeutic scenarios, dynamically adapting them based on follow-up data. This allows clinicians to assess not only the initial therapy choice (Case study no. 1) but also the potential addition of supplementary interventions during treatment (Case study no. 2). Results highlight that the proposed hybrid SHACL–Bayesian framework enables tightly coupled deterministic–probabilistic reasoning, where SHACL constraints define the admissible clinical decisions and Bayesian inference operates within this validated space. Compared to deterministic or probabilistic approaches, the combined framework more effectively handles uncertainty, guideline conflicts, and temporal updates. The scientific contribution lies in showing that this integration enhances decision support for recurrent UTIs in postmenopausal patients, providing clinically consistent, transparent, and adaptive therapeutic recommendations aligned with the patient’s evolving condition. Full article

The integration of artificial intelligence (AI), particularly large language models (LLMs), into education, marks a profound shift in how knowledge is accessed, processed, and applied. These tools offer clear advantages—including improved efficiency, immediate support, and high productivity—but it may simultaneously weaken foundational skills. This Perspective examines the dual impact of AI on education, arguing that over-reliance on AI may displace essential cognitive processes that reinforce professional competence. Emerging evidence points to troubling associations between frequent AI use and diminished critical reasoning. We propose a model of critical alliance, in which AI augments but does not replace core intellectual processes. Unlike existing AI competency or digital literacy, this model centers on preserving human cognitive agency, judgment, reflection, and intellectual ownership, as primary educational outcomes. This framework not only emphasizes cognitive independence, but also equitable access, ethical vigilance, and faculty development as cornerstones of AI literacy. Addressing these questions is essential to safeguard both intellectual growth and educational equity in an AI-augmented era. Unlike existing digital literacy or AI competency frameworks, the critical alliance explicitly centers on preserving human cognitive agency and intellectual ownership as educational priorities, particularly in environments increasingly shaped by high-performing generative systems. Full article

The deployment of large language models (LLMs) in commercial environments depends critically on the availability of robust digital infrastructure, scalable computing resources, and mature cloud architectures. This study examines how macro-level digital infrastructure, in particular cloud computing adoption, conditions the ability of the European retail sector to deploy and benefit from large language models (LLMs). Using a country-year panel of EU member states from 2017 to 2023, we estimate fixed-effects regressions to quantify the association between enterprise cloud use and retail trade volume growth, and implement an event-study design to explore dynamic responses around changes in cloud uptake. The results show that increases in cloud adoption are significantly associated with higher retail trade growth added and productivity, with especially strong effects in emerging Eastern European markets. We identify a digital threshold of around 20% of enterprises using cloud services, above which the marginal impact on retail performance becomes notably larger. These findings highlight cloud infrastructure as a key enabling condition for LLM-enabled retail applications and inform EU digital and industrial policy targeting regional digital disparities. Full article

The rapid adoption of large language models (LLMs) in corporate and governmental systems has raised critical security concerns, particularly prompt injection attacks exploiting LLMs’ inability to differentiate control instructions from untrusted user inputs. This study systematically benchmarks neural network architectures for malicious prompt detection, emphasizing robustness against character-level adversarial perturbations—an aspect that remains comparatively underemphasized in the specific context of prompt-injection detection despite its established significance in general adversarial NLP. Using the Malicious Prompt Detection Dataset (MPDD) containing 39,234 labeled instances, eight architectures—Dense DNN, CNN, BiLSTM, BiGRU, Transformer, ResNet, and character-level variants of CNN and BiLSTM—were evaluated based on standard performance metrics (accuracy, F1-score, and AUC-ROC), adversarial robustness coefficients against spacing and homoglyph perturbations, and inference latency. Results indicate that the word-level 3_Word_BiLSTM achieved the highest performance on clean samples (accuracy = 0.9681, F1 = 0.9681), whereas the Transformer exhibited lower accuracy (0.9190) and significant vulnerability to spacing attacks (adversarial robustness ${\mathsf{\rho }}_{spacing}=0.61$). Conversely, the Character-level BiLSTM demonstrated superior resilience (${\mathsf{\rho }}_{spacing}=1.0$, ${\mathsf{\rho }}_{homoglyph}=0.98$), maintaining high accuracy (0.9599) and generalization on external datasets with only 2–4% performance decay. These findings highlight that character-level representations provide intrinsic robustness against obfuscation attacks, suggesting Char_BiLSTM as a reliable component in defense-in-depth strategies for LLM-integrated systems. Full article

Frailty is a multidomain reduction in physiologic reserve that impacts recovery and can contribute to poor outcomes following trauma beyond what chronological age, comorbidities, or injury severity predicts. In geriatric trauma patients, a large proportion are frail or prefrail on initial encounter in the emergency department, and because there are opportunities for actionable management plans, major trauma guidelines endorse systematic screening integrated into coordinated geriatric trauma care. We reviewed the literature and identified practical instruments used in the acute trauma setting for risk stratification. Additionally, we highlight the feasibility of using these instruments, as some can be completed via patient report, proxy input, or chart review when cognition, language, or caregiver availability limits history-taking. Implementation efforts succeed when shared mental models are leveraged and screening is embedded in the electronic health record system, linked to order sets and trigger-based pathways that offer downstream goal-directed care management, such as early mobility, delirium prevention, nutrition, medication review, and comprehensive geriatric assessment. Additionally, we highlight the importance of initiating early goals-of-care discussions and coordinating care with palliative care services. Resource-limited systems can preserve the same architecture by using nurse-led or allied staff-led screening, tele-geriatric consultation, and virtual interdisciplinary huddles. Lastly, we expand upon opportunities for longitudinal post-discharge follow-up. We describe how targeted initiatives translate research into practice, improve outcomes, and support longitudinal reassessment through in-person and telehealth follow-up visits. Full article

With the widespread adoption of cloud services, ensuring the quality and security of the representational state transfer application programming interface (REST API) has become critical. Among various REST API testing techniques, fuzz testing stands out as a promising approach due to its ability to automatically generate large volumes of random or malformed inputs. To improve test coverage through fuzzing, we developed an enhanced method for generating API sequences and parameter values, building upon the widely used open-source tool RESTler. The approach extends RESTler by incorporating resource-level dependencies between APIs in addition to the existing producer–consumer relationships, enabling the construction of more valid API sequences. It also leverages a large language model to automatically generate parameter values. To further ensure input validity, a feedback loop is introduced to refine invalid inputs using error messages from API responses. Experimental results show that, compared to RESTler, the proposed method increases API coverage and detects more faults on average, demonstrating its effectiveness. Full article

General-purpose vision–language models (VLMs) are increasingly applied to imaging tasks, yet their reliability on medical visual question answering (Med-VQA) remains unclear. We investigate how three state-of-the-art VLMs—ViLT, BLIP, and MiniCPM-V-2—perform on radiology-focused Med-VQA when evaluated in a modality-aware manner. Using SLAKE and OmniMedVQA-Mini, we construct harmonised subsets for computed tomography (CT), magnetic resonance imaging (MRI), and X-ray, standardising schema and answer processing. We first benchmark all models in a strict zero-shot setting, then perform supervised fine-tuning on modality-specific data splits, and finally add a post-hoc semantic option-selection layer that maps free-text predictions to multiple-choice answers. Zero-shot performance is modest (exact match ≈20% for ViLT/BLIP and 0% for MiniCPM-V-2), confirming that off-the-shelf deployment is inadequate. Fine-tuning substantially improves all models, with ViLT reaching ≈80% exact match and BLIP ≈50%, while MiniCPM-V-2 lags behind. When coupled with option selection, ViLT and BLIP achieve 90–93% exact match and F1 across all modalities, corresponding to 95–97% BERTScore-F1. Our novel results show that (i) modality-specific supervision is essential for Med-VQA, and (ii) post-hoc option selection can transform strong but imperfect generative predictions into highly reliable discrete decisions on harmonised radiology benchmarks. The latter is useful for medical VLMs that combine generative responses with option or sentence selection. Full article

Sentiment analysis of Arabic text, particularly on social media platforms, presents a formidable set of unique challenges that stem from the language’s complex morphology, its numerous dialectal variations, and the frequent and nuanced use of emojis to convey emotional context. This paper presents SiAraSent, a hybrid framework that integrates traditional text representations, emoji-aware features, and deep contextual embeddings based on Arabic transformers. Starting from a strong and fully interpretable baseline built on Term Frequency–Inverse Definition Frequency (TF–IDF)-weighted character and word N-grams combined with emoji embeddings, we progressively incorporate SinaTools for linguistically informed preprocessing and AraBERT for contextualized encodings. The framework is evaluated on a large-scale dataset of 58,751 Arabic tweets labeled for sentiment polarity. Our design works within four experimental configurations: (1) a baseline traditional machine learning architecture that employs TF-IDF, N-grams, and emoji features with an Support Vector Machine (SVM) classifier; (2) an Large-language Model (LLM) feature extraction approach that leverages deep contextual embeddings from the pre-trained AraBERT model; (3) a novel hybrid fusion model that concatenates traditional morphological features, AraBERT embeddings, and emoji-based features into a high-dimensional vector; and (4) a fully fine-tuned AraBERT model specifically adapted for the sentiment classification task. Our experiments demonstrate the remarkable efficacy of our proposed framework, with the fine-tuned AraBERT architecture achieving an accuracy of 93.45%, a significant 10.89% improvement over the best traditional baseline. Full article

Vehicle repositioning is essential for improving efficiency and service quality in ride-sharing platforms, yet existing approaches typically optimize proxy rewards that fail to reflect human-centered preferences such as wait time, service coverage, and unnecessary empty travel. We propose the first two-stage Bilevel Reinforcement Learning (RL) from Human Feedback (RLHF) framework for preference-aligned vehicle repositioning. In Stage 1, a value-based Deep Q-Network (DQN)-RLHF warm start learns an initial preference-aligned reward model and stable reference policy, mitigating the reward-model drift and cold-start instability that arise when applying on-policy RLHF directly. In Stage 2, a Kullback–Leibler (KL)-regularized Proximal Policy Optimization (PPO)-RLHF algorithm, equipped with action masking, behavioral-cloning anchoring, and alternating forward–reverse KL, fine-tunes the repositioning policy using either Large Language Model (LLM)-generated or rubric-based preference labels. We develop and compare two coordination schemes, pure alternating (PPO-Alternating) and k-step alternating (PPO-k-step), demonstrating that both yield consistent improvements across all tested arrival scales. Empirically, our framework reduces wait time and empty-mile ratio while improving served rate, without inducing trade-offs or reducing platform profit. These results show that human preference alignment can be stably and effectively incorporated into large-scale ride-sharing repositioning. Full article

Background: Rapid and accurate fracture detection is critical in emergency departments (EDs), where high patient volume and time pressure increase the risk of diagnostic error, particularly in radiographic interpretation. Multimodal large language models (LLMs) with image-recognition capability have recently emerged as general-purpose tools for clinical decision support, but their diagnostic performance within routine emergency department imaging workflows in orthopedic trauma remains unclear. Methods: In this retrospective diagnostic accuracy study, we included 1136 consecutive patients referred from the ED to orthopedics between 1 January and 1 June 2025 at a single tertiary center. Given the single-center, retrospective design, the findings should be interpreted as hypothesis-generating and may not be fully generalizable to other institutions. Emergency radiographs and clinical data were processed by a multimodal LLM (2025 version) via an official API using a standardized, deterministic prompt. The model’s outputs (“Fracture present”, “No fracture”, or “Uncertain”) were compared with final diagnoses established by blinded orthopedic specialists, which served as the reference standard. Diagnostic agreement was analyzed using Cohen’s kappa (κ), sensitivity, specificity, accuracy, and 95% confidence intervals (CIs). False-negative (FN) cases were defined as instances where the LLM reported “no acute fracture” but the specialist identified a fracture. The evaluated system is a general-purpose multimodal LLM and was not trained specifically on orthopedic radiographs. Results: Overall, the LLM showed good diagnostic agreement with orthopedic specialists, with concordant results in 808 of 1136 patients (71.1%; κ = 0.634; 95% CI: 68.4–73.7). The model achieved balanced performance with sensitivity of 76.9% and specificity of 66.8%. The highest agreement was observed in knee trauma (91.7%), followed by wrist (78.8%) and hand (69.6%). False-negative cases accounted for 184 patients (16.2% of the total cohort), representing 32.4% of all LLM-negative assessments. Most FN fractures were non-displaced (82.6%), and 17.4% of FN cases required surgical treatment. Ankle and foot regions showed the highest FN rates (30.4% and 17.4%, respectively), reflecting the anatomical and radiographic complexity of these areas. Positive predictive value (PPV) and negative predictive value (NPV) were 69.4% and 74.5%, respectively, with likelihood ratios indicating moderate shifts in post-test probability. Conclusions: In an emergency department-to-orthopedics consultation cohort reflecting routine clinical workflow, a multimodal LLM demonstrated moderate-to-good diagnostic agreement with orthopedic specialists, broadly within the range reported in prior fracture-detection AI studies; however, these comparisons are indirect because model architectures, training strategies, datasets, and endpoints differ across studies. However, its limited ability to detect non-displaced fractures—especially in anatomically complex regions like the ankle and foot—carries direct patient safety implications and confirms that specialist review remains indispensable. At present, such models may be explored as hypothesis-generating triage or decision-support tools, with mandatory specialist confirmation, rather than as standalone diagnostic systems. Prospective, multi-center studies using high-resolution imaging and anatomically optimized algorithms are needed before routine clinical adoption in emergency care. Full article

Large language models (LLMs) are increasingly used to predict human behavior from plain-text descriptions of experimental tasks that range from judging disease severity to consequential medical decisions. While these methods promise quick insights without complex psychological theories, we reveal a critical flaw: they often latch onto accidental patterns in the data that seem predictive but collapse when faced with novel experimental conditions. Testing across multiple behavioral studies, we show these models can generate wildly inaccurate predictions, sometimes even reversing true relationships, when applied beyond their training context. Standard validation techniques miss this flaw, creating false confidence in their reliability. We introduce a simple diagnostic tool to spot these failures and urge researchers to prioritize theoretical grounding over statistical convenience. Without this, LLM-driven behavioral predictions risk being scientifically meaningless, despite impressive initial results. Full article

Objectives: The industrial production of lanolin alcohol currently employs batch saponification, which suffers from high energy consumption, prolonged processing time, and excessive solid waste generation, rendering it incompatible with green chemistry principles. This study aimed to develop an efficient, sustainable saponification process by addressing these limitations through integrating large language models (LLMs) with microfluidic technology. Methods: An LLM-based intelligent agent called SapoMind (version 1.0) was constructed. SapoMind employs LLMs as its software core and a continuous-flow microreactor as the experimental platform. Its performance was enhanced through supervised fine-tuning. The system enables automated recommendation of saponification process parameters, autonomous experimental design, and automatic execution of experiments. Saponification conditions were automatically optimized considering product quality, energy consumption, material consumption, and time consumption. Results: The optimal continuous-flow saponification conditions were determined as 70 °C reaction temperature and 9 min residence time, producing lanolin alcohol complying with European Pharmacopoeia standards. Compared to batch processing, the optimized process reduced carbon emissions by 53% and solid waste by 37%, with the greenness score increasing from 82 to 93. Conclusions: This study demonstrates the effectiveness of LLM-driven intelligent agents in optimizing green chemical processes. SapoMind offers significant environmental and operational benefits for lanolin alcohol production. Full article

The Thailand 4.0 agenda elevates entrepreneurship education (EE) as a lever to escape the middle-income, inequality, and imbalance traps, yet EE remains weakly embedded in basic education—especially in Thai language. We designed and piloted a community-economy context-based learning model integrating EE (CEC-EE) for Grade 12 Thai via a two-cycle R&D process: needs analysis (surveys and focus groups with teachers and students) and prototype development. The model operationalizes six instructional steps (6Cs: connect, comprehend, clarify, construct, carry over, and conclude) anchored in Mae Chan’s community economy and targets entrepreneurial language skills (ELSs) consisting of analytical reading and creative writing. In a one-group pretest–posttest with Grade 12 students (n = 32), academic achievement and ELSs—analytical reading and creative writing—improved markedly. Posttest means exceeded pretests with very large effect. Experts rated the model appropriate, feasible, and useful; teachers and students reported high perceived value alongside concerns about implementation cost, support capacity, and student readiness. The CEC-EE model offers a context-responsive pathway for embedding EE in Thai-language instruction; future work should employ comparative designs, multi-site samples, and cost-effectiveness analyses to assess scalability and sustained impact. Full article

The advancement of large language models (LLMs) has facilitated the generation of user interface (UI) code from natural language prompts, thereby supporting low-code development paradigms. Despite these capabilities, ensuring brand consistency remains a significant challenge, particularly when style data is unstructured. We investigated the impact of three data formats—plain text, structured cascading style sheets (CSS), and structured natural language (NL) guide—on the effectiveness of retrieval-augmented generation (RAG) in producing brand-consistent UI components, with No-RAG serving as the baseline for comparison. The findings indicate that RAG substantially enhances brand alignment. Although the structured NL guide yielded the highest CSS recall rate, participants expressed a preference for outputs derived from plain text, suggesting that the optimal data format may depend on specific design contexts and evaluative criteria. Full article

Background: Chronic kidney disease (CKD) is a major global health burden. Patient education is a crucial part of CKD management. Large language models (LLMs) such as ChatGPT and Gemini may help patients access medical information, but their reliability in CKD-related contexts is uncertain. Methods: We collected 291 questions from 100 CKD patients and selected and analyzed 123 of them across three categories: medical condition and treatment, nutrition and diet, and symptom management. Responses from ChatGPT and Gemini were assessed by two nephrology specialists using the Quality Assessment of Medical Artificial Intelligence (QAMAI) scale. Results: When all 123 questions were evaluated together, ChatGPT outperformed Gemini in terms of clarity and usefulness. However, when the questions were analyzed by category, Gemini demonstrated relatively stronger performance in the nutrition and symptom management domains. Accuracy and relevance were comparable between the two models. Neither consistently provided adequate citations. Conclusion: ChatGPT and Gemini demonstrate potential as supplementary tools for CKD patient education, with complementary strengths across different domains. Although they cannot replace clinical expertise, their supervised use could enhance information access and reduce clinician burden. Full article