Search Results (112)

Background: Large language models (LLMs) are increasingly accessible to healthcare providers and patients for clinical decision support, yet their ability to perform precise mathematical calculations required for validated risk scores remains unexplored, and errors could compromise patient safety. The EuroSCORE II requires complex multivariable computation that could reveal fundamental limitations in LLM computational capabilities. Methods: We evaluated four publicly available chatbots (ChatGPT (GPT-4o, OpenAI), Claude (Sonnet 4.5, Anthropic), Gemini (2.5 Flash, Google) and Deepseek (V3)) in calculating EuroSCORE II for 105 patients from the CARDIO-FR registry with gold standard heart team calculations. Each model was tested using two approaches: direct calculation from clinical parameters alone and formula-based calculation with the explicit EuroSCORE II algorithm provided. Performance was assessed through mean absolute error (MAE), correlation coefficients, and clinical agreement within ±2% of gold standard values. Results: Direct LLM calculations demonstrated poor accuracy (MAE range: 3.3–6.4%) with the best performer (Gemini) achieving only 50.5% clinical agreement. Formula provision improved performance in three of four models, with ChatGPT formula achieving the lowest MAE (2.9%) and highest clinical agreement (51.4%), followed by Claude formula (MAE 3.2%, agreement 48.6%). Adding the formula to the prompt significantly improved performance and reduced bias. All methods exhibited significant systematic biases (p < 0.05 for 7/8 strategies). Conclusions: Publicly available LLM chatbots cannot reliably calculate EuroSCORE II for clinical use. Adding the formula to the prompt significantly improved performance, but was not sufficient to reach the clinically required threshold of 90% agreement. Clinicians should rely on validated risk calculation tools rather than LLM chatbots for quantitative clinical risk assessments. Full article

(1) Background: The disaggregated and software-defined nature of fifth-generation (5G) core networks and the Open Radio Access Network (O-RAN) architecture increase the attack surface and produce large volumes of heterogeneous evidence that must be analyzed in real time to support incident reconstruction. Open-source 5G stacks (including Open5GS and srsRAN) have become reference platforms in the literature, yet recent research, such as the RANsacked study that reported 119 vulnerabilities and 97 unique CVEs across multiple LTE/5G implementations, have highlighted the pressing need for AI-based detection and forensic capabilities specific to these stacks. (2) Methods: We introduce an experimental framework consisting of a reproducible srsRAN+Open5GS testbed and an AI-driven forensic and detection pipeline. The pipeline receives control-plane (NAS, NGAP, F1AP) and Service-Based Interface (SBI) traffic, extracts protocol- and statistically grounded features and classifies traffic into seven attack types using a hybrid CNN–LSTM model. Integrity-protected and timeline-correlated forensic artifacts (PCAP, logs, memory dumps) assist in reconstructing an incident. (3) Results: The proposed hybrid model achieves a macro F1-score of 0.972 and an AUC-ROC of 0.995 (5-fold CV) and degrades gracefully under load. (4) Conclusions: We show that AI-based detection can be coupled with a scientifically sound evidence chain in open-source 5G stacks deployed as disaggregated mobile networks. Full article

Fake social media profiles increasingly resemble ordinary accounts, combining believable images, fluent biographies, copied posts, and selective engagement, which makes single signal detection unreliable. This study develops and evaluates a forensic natural language processing (NLP) framework that treats profile text as digital evidence rather than raw model input. Profiles were collected from Facebook, Instagram, X, Truth Social, and LinkedIn through network expansion and vocabulary guided scraping, then anonymized, merged, cleaned, and organized into structured text and profile records. Ground truth came from human eye annotation (referred to throughout as human eye review, human review, or manual review) supported by three LLM reviewers (referred to throughout as LLM assisted review), OpenAI GPT-4o, Anthropic Claude 3.5 Sonnet, and Google Gemini 1.5 Pro, with strict agreement logic producing confidence-tiered labels across platform-specific label spaces covering scam, fraud, harmful, and legitimate behavior. Each model received a defined set of label categories, was instructed to return a single label with a one sentence rationale, and worked independently with no cross-model communication. Multidimensional features, including TF-IDF representations, sentence embeddings, scam lexicon counts, sentiment and emotion scores, and behavioral indicators, were extracted through a pipeline built to keep links, hashtags, contact markers, and repeated phrasing as forensic signals rather than noise. Traditional classifiers, transformer models, and ensembles were trained on verified labels, and the best model per platform was applied to the full dataset of over 309,108 records under confidence filtered classification, with predictions below a 0.70 probability threshold flagged for human review. The framework links classification outputs to textual indicators through explainable AI, achieving weighted F1 scores between 0.86 and 0.96 across five platforms. These results show that profile text alone provides strong, production-relevant evidence for detecting fake and harmful accounts, even though the framework treats this as one evidence stream to be weighed alongside image and behavioral signals rather than as a final determination on its own. The accompanying evidence trail means an investigator or a trust and safety team can see why a profile was flagged rather than being handed a bare score, which is what turns a classifier into a tool that can be checked, challenged, and used directly in an investigation. Full article

Background: Family caregivers of children with autism spectrum disorder (ASD) increasingly utilize large language models (LLMs) for health information. This study presents a systematic comparative evaluation of three widely used LLMs as localized ASD health information tools in Saudi Arabia. Methods: Twenty-four clinically validated, caregiver-oriented questions were posed to Google Gemini 1.5 Pro, OpenAI ChatGPT (GPT-4o), and DeepSeek-V3 using a standardized prompt. Three expert raters independently evaluated responses across four dimensions: scientific accuracy, PEMAT-P understandability, PEMAT-P actionability, and neurodiversity (ND)-affirming language. Readability was assessed via Flesch–Kincaid Grade Level (FKGL) and SMOG indices. Non-parametric Kruskal–Wallis tests with post hoc Mann–Whitney U comparisons and one-sample t-tests were applied. Results: Gemini achieved the highest mean accuracy (2.96/3.00), significantly outperforming DeepSeek (p = 0.003, r = −0.37). Accuracy failures across all LLMs clustered on regional epidemiological, genetic risk, and financial inquiries. ChatGPT achieved significantly higher understandability than Gemini (p < 0.001, r = 0.55), while DeepSeek achieved significantly higher actionability than Gemini (p < 0.001, r = 0.59). However, all three LLM scores fell short of the Agency for Healthcare Research and Quality (AHRQ) 80% actionability benchmark (all p < 0.001). All LLMs exceeded patient education readability benchmarks (FKGL ≤ 6, SMOG ≤ 8; all p < 0.001); ChatGPT was the most readable (FKGL = 7.86; SMOG = 9.97) and Gemini the most complex. No model differed significantly on ND-affirming language, defaulting to a mixed medical-affirming register. Conclusions: Evaluated LLMs demonstrated distinct, specialized strengths: Gemini was the most accurate, ChatGPT the most readable, and DeepSeek the most actionable. Importantly, all models failed to meet established consumer education standards for readability and actionability. LLMs require extensive plain-language adaptation and cultural customization. Clinicians must guide families on navigating LLM outputs, particularly concerning country-specific epidemiological, economic, and healthcare service queries. Full article

Background/Objectives: The integration of artificial intelligence (AI) into pharmaceutical development has the potential to accelerate early-stage formulation design. In this study, large language models (ChatGPT (GPT-4o, OpenAI) and DeepSeek (DeepSeek-R1, DeepSeek AI) were evaluated as supportive tools for the design of sustained-release lornoxicam matrix tablets. Using constrained formulation prompts and a predefined excipient space, each model generated candidate formulations intended for direct compression, with the objective of producing sustained-release systems capable of mimicking the dissolution behaviour of a commercial reference product (LOROX OD 16 mg). Methods: The proposed formulations were prepared experimentally and evaluated for physicochemical properties, including weight variation, hardness, friability, and drug content, as well as in vitro dissolution performance over 24 h. Dissolution profiles were compared with the reference product using similarity (f₂) and difference (f₁) factors, and release behaviour was further characterized using kinetic models. Results: All formulations demonstrated sustained-release behaviour without evidence of dose dumping. One ChatGPT-generated formulation (F3C) met the regulatory criteria for dissolution similarity to the reference product (f₁ = 9.66, f₂ = 71.31), while the remaining formulations showed variable release behaviour with f₂ values ranging from 28.61 to 49.70. However, F3C exceeded the pharmacopeial friability limit marginally (1.108%), while DeepSeek formulations F5D and F6D exceeded pharmacopeial assay acceptance limits. Kinetic modelling indicated a range of transport mechanisms from anomalous diffusion to super Case II transport depending on polymer composition. Conclusions: Although both AI systems successfully generated experimentally viable formulations, prediction accuracy analysis showed high trend-level correlations between AI-predicted and experimental dissolution profiles. However, the magnitude of quantitative error was substantial, with RMSE values exceeding 17% and MAPE values ranging from approximately 38% to 60%. These findings indicate that the models captured general release trends but did not provide reliable quantitative dissolution predictions. Full article

When a large language model (LLM) answers a question using retrieved documents, retrieval-augmented generation (RAG) is the standard approach. Retrieving more documents improves answer accuracy but increases cost and response time; retrieving fewer documents saves resources but may miss critical information. Most existing RAG systems sidestep this dilemma by applying the same retrieval setting to every query, regardless of how simple or complex the question is. This wastes budget allocation on easy questions and under-serves hard ones. This paper introduces Cost-Aware RAG (CA-RAG), a routing framework that solves this problem by treating each query individually. For every incoming question, CA-RAG selects the most suitable retrieval strategy from a fixed menu of four options, ranging from no retrieval to fetching the top $k=10$ most-relevant documents. The selection is driven by a scoring formula that balances expected answer quality against predicted cost and response time. The weights in this formula act as dials: adjusting them shifts the system toward speed, savings, or quality without any retraining. CA-RAG is built on Facebook AI Similarity Search (FAISS) for document retrieval, OpenAI gpt-4o-mini for generation, and text-embedding-3-small for dense retrieval embeddings. We evaluate CA-RAG on a benchmark of 28 queries. The router assigns different strategies to different queries, achieving 26% fewer billed tokens compared to always using heavy retrieval and 34% lower response time compared to always answering without retrieval, while maintaining answer-quality parity in both cases. Further analysis shows that most savings come from simpler queries, where heavy retrieval was unnecessary. All results are reproducible from logged comma-separated value (CSV) files. CA-RAG demonstrates that a small but well-designed set of retrieval strategies combined with lightweight per-query routing can meaningfully reduce the cost and latency of LLM deployments without compromising answer quality. Full article

(1) Background: Clinical trial data extraction from registries such as ClinicalTrials.gov remains labor-intensive and error-prone, often missing critical details hidden in unstructured protocol descriptions. Large Language Models (LLMs) offer potential to automate this process, yet systematic multi-model comparisons on real clinical trial data remain scarce. (2) Methods: Four LLMs (OpenAI o4-mini-high, Anthropic Claude-Sonnet-4, Google Gemini 2.5-Pro, and Meta Llama-4-Maverick) extracted brain stimulation parameters from 67 transcranial direct current stimulation (tDCS) trials in Parkinson’s disease via a structured JSON schema. Pairwise inter-model agreement was quantified with Cohen’s Kappa and percentage agreement across binary, categorical, and multi-component task tiers. (3) Results: Under exact-string matching, agreement was near-perfect for binary classifications (non-invasive classification: 100%; brain stimulation presence: 99.3%, $\kappa$ = 0.50) and substantial for categorical extractions (primary stimulation type: 96.4%, $\kappa$ = 0.70), but fell to 48.6% ($\kappa$ = 0.43) for complex anatomical targets. Numeric parameters revealed model-specific strengths: o4-mini-high and Claude-Sonnet-4 achieved perfect duration agreement (r = 1.000, n = 19) while Llama-4-Maverick diverged substantially (r < 0.12). Validation against an expert gold standard (100% inter-annotator agreement on a 20-trial overlap) confirmed high extraction accuracy across all features (mean 93.7–98.9%). Crucially, the low agreement on anatomical targets proved to be an artifact of exact-string scoring: under the same semantic matching used to measure accuracy, inter-model agreement rose to 97.0%, coinciding with the 95.5% expert accuracy. Inter-model agreement therefore tracks accuracy once both are measured on a common basis. (4) Conclusions: Exact-string inter-model agreement decreases with task complexity, but this decline largely reflects interchangeable free-text wording rather than reduced accuracy. Evaluated semantically, agreement and expert accuracy are both high and closely aligned. A residual risk is not low accuracy but the rare error shared across all models, which agreement cannot detect, and which overall accuracy can itself mask when one class dominates. These findings inform hybrid human–AI systematic review pipelines in which targeted expert oversight focuses on shared-error and minority-class detection. Full article

Smart city environmental monitoring depends on sparse air quality sensor networks and analytics services that remain reliable under node additions, outages, and missing streams. We propose an operational deep learning framework for citywide cross-location forecasting from a limited set of sensors, delivering low-latency, real-time concentration heatmaps at unsensed locations by combining temporal prediction with spatial regression. We formulate single-stage spatiotemporal forecasting and benchmark nine recurrent, convolutional, and multilayer architectures against classical baselines. The framework forecasts ${\mathrm{O}}_3$, ${\mathrm{NO}}_2$, ${\mathrm{PM}}_{2.5}$, and ${\mathrm{PM}}_{10}$ over horizons from 1 h to 10 days. Using open monitoring data from Madrid (Spain) and Cali (Colombia), we evaluate generalization by holding out stations, reflecting deployment to new sensor nodes and sparse coverage regimes. We further compare missing data handling strategies and show that common imputation can substantially degrade accuracy, increasing RMSE by up to 74% in some settings. Beyond prediction, the framework provides a basis for guiding sensor network densification; confidence estimates can highlight locations where additional sensors may be most beneficial. These results provide actionable guidance for deploying AI-enabled sensing services with robust performance under realistic sensor reliability constraints while supporting real-time citywide mapping. Full article

Inadequate management of municipal solid waste represents a critical challenge for the sustainability of modern cities, characterized by low citizen participation rates due to the lack of direct incentives. Unlike existing approaches that isolate hardware classification or fleet monitoring, this article presents RENOVA as a socio-technical closed-loop system based on the Internet of Things (IoT) and artificial intelligence (AI). This system integrates an IoT-enabled smart bin, a gamified mobile application for citizens, and an administrative web panel for merchant redemption, all interconnected via a REST API. The system employs computer vision through the GPT-4o (OpenAI, San Francisco, CA, USA) multimodal model for the automatic classification of recyclable materials (PET plastic and Aluminum) and integrates a gamified rewards program to incentivize citizen participation. The methodology follows an applied technological development approach under the agile Scrum framework. Prototype validation demonstrated successful real-time communication between the IoT device and the cloud platform, achieving classification accuracy exceeding 95% under controlled conditions. A diagnostic survey applied to a convenience sample of 51 participants revealed that 94.1% accepted the proposed gamification model, while user experience evaluation (n = 74; consisting primarily of university-affiliated individuals aged 15–24) yielded a mean overall satisfaction score of 4.77/5.0 (SD = 0.48), with 79.7% of participants assigning the maximum rating. These findings reflect stated user acceptance and behavioral intention under prototype conditions rather than observed long-term behavioral change, and should not be generalized to broader urban populations without further validation. The proposed solution directly contributes to Sustainable Development Goals 11 (Sustainable Cities) and 12 (Responsible Consumption), suggesting a potentially scalable framework. Full article

Generative Artificial Intelligence (GenAI) in cardiac surgery refers to the integration of advanced computational models, such as Large Language Models (LLMs), to automate and enhance clinical decision-making, preoperative risk assessment, and surgical education. In the context of surgical training, it functions as a personalized pedagogical tool that supports various learning activities, ranging from information acquisition and clinical inquiry to procedural practice, while requiring rigorous human oversight to ensure patient safety and clinical accuracy. (1) Background: Generative Artificial Intelligence (GenAI) is increasingly integrated into health professions education, offering new opportunities for learning; however, its specific application and pedagogical mapping in high-stakes fields such as cardiac surgery remain underexplored. This systematic review investigates how GenAI is utilized in cardiac surgery and surgical education, aligning these uses with Laurillard’s six learning types. (2) Methods: Following the PRISMA 2020 guidelines, we searched the Web of Science Core Collection for studies on GenAI in cardiac surgery, resulting in 42 studies that met the inclusion criteria. Study quality was appraised using the Medical Education Research Study Quality Instrument (MERSQI). (3) Results: GenAI applications most frequently supported clinical inquiry (93.8%) and practice (68.8%), demonstrating expanding efficiency across commercial and open-source models (including ChatGPT-4o, Gemini AI, and emerging reasoning architectures such as DeepSeek) for knowledge acquisition and medical production. While it significantly improves individualized learning and preoperative assessment workflows, its practical role in Discussion and Collaboration remains heavily underutilized, highlighting a distinct shift toward individualized solo professional workflows. (4) Conclusions: GenAI provides a transformative and scalable approach to cardiac surgical training by offering personalized and accessible knowledge retrieval. However, clinical educators and governance bodies must deliberately balance these immediate productivity benefits with long-term concerns regarding structural “hallucinations,” data verifiability, and the preservation of collaborative competencies within modern multidisciplinary Heart Teams. Full article

Background: Chatbots are becoming increasingly valuable in clinical settings, offering rapid access to medical information, aiding documentation, and improving perioperative patient education. Their adaptability makes them promising tools for personalized perioperative risk stratification (PRS) and anesthesia planning, but their definitive role remains uncertain. We aimed to evaluate chatbot performance in PRS compared to standard clinical judgment and to assess the certainty of the evidence supporting their use. Methods: This systematic review (PROSPERO ID: CRD42025642357) followed PRISMA extended and PRISMA-S guidelines. The population was defined according to the PICO framework: we included adult surgical patients undergoing anesthesia assessment (P), evaluated with LLM-based chatbots for perioperative risk stratification and anesthesia planning (I), compared with traditional clinician assessment (C), and extracted performance metrics (O). Comprehensive searches of PubMed, MEDLINE, Scopus, Embase, Google Scholar, Open Gray, ClinicalTrials.gov, WHO ICTRP, and Cochrane Library Central were conducted through January 2026. Risk of bias and study quality were assessed using PROBAST-AI, RoB-2, and ROBINS-I. Certainty of the evidence was assessed using GRADE system. A random-effects meta-analysis of pooled chatbot accuracy was performed, with subgroup analyses by ASA status and perioperative risk stratification. A sensitivity analysis was performed with a leave-one-out exclusion test. Results: Eleven studies published between 2023 and January 2026 were included (N = 227,059 patients). Five prospective cohorts, two large retrospective cohorts, one randomized non-inferiority trial, and three non-clinical or mixed-methods studies were found. Meta-analysis showed that the pooled accuracy of LLM-based chatbots for AI–clinician concordance in perioperative risk stratification and ASA classification was 0.90 [95% CI: 0.42–0.99; 95% prediction interval 0.03–1.00]. Subgroup analyses indicated that the ASA status prediction subgroup reached a pooled accuracy of 0.91 (95% CI: 0.46 to 0.99), whereas the exploratory perioperative risk stratification subgroup showed an accuracy of 0.73 (95% CI: 0.10 to 0.98). Performance decreased with increasing patient complexity. Evidence is limited by small sample sizes, extreme sample size skew toward a single center, geographic bias, inconsistent outcome definitions and performance metrics, and incomplete reporting of adverse events. Most studies lacked prospective trial registration or robust control for confounding, and publication bias cannot be excluded. Conclusions: LLM-based chatbots show promising performance in routine perioperative risk stratification but remain unreliable in complex cases, with potential safety concerns. Given the overall very low GRADE certainty of evidence, these tools should be used as clinician-supervised decision support aids for routine ASA assessment, and should not be relied upon for autonomous use in complex cases or for general perioperative risk stratification. Other: This research received no external funding. PROSPERO ID: CRD42025642357. Full article

Background/Objectives: Large language models (LLMs) are increasingly consulted for information about cleft lip and palate (CLP), yet the reliability of their outputs across clinical domains has not been evaluated. This study aimed to compare the quality of CLP-related information generated by GPT-4o and Gemini 2.5 Pro across multiple thematic domains using a validated quality instrument and a reliability-first analytic framework. Methods: Fifty-four standardized CLP questions across six domains were submitted to GPT-4o (OpenAI) and Gemini 2.5 Pro (Google DeepMind) on 25 September 2024 via their public interfaces, using new, history-free sessions and default settings, yielding 108 responses. Three independent, CLP-experienced raters scored each response using the Global Quality Score (GQS; 1–5 scale assessing accuracy, completeness, and clinical usefulness). Before comparing models, we applied a reliability-first filter: only domains where all three raters showed substantial agreement (Fleiss’ kappa [κ] ≥ 0.60) were included in statistical comparisons. Domains that failed this threshold were analyzed qualitatively to identify the source of disagreement. A descriptive taxonomy of errors was developed for low-scoring responses. Results: Three domains met the reliability threshold (General Care Information, General Cleft Information, and Pre-Treatment Information; 30 paired questions). Both models performed at a high and practically equivalent level: GPT-4o median GQS 4.33 (IQR 4.00–5.00) versus Gemini 2.5 Pro 5.00 (IQR 4.00–5.00); the difference was not statistically significant (Wilcoxon V = 139.00, p = 0.691; Hodges–Lehmann median difference 0.00, 95% CI −0.33 to 0.67). Three domains were excluded because rater agreement was insufficient; qualitative review showed this reflected genuine clinical practice variation rather than clear model errors. The most common inaccuracies were overgeneralization of outcomes, outdated surgical timing, and omission of multidisciplinary team roles. Conclusions: Both models provided high-quality CLP information in domains supported by clinical consensus, indicating they may serve as useful adjuncts for general patient and family counseling. Clinicians should, however, verify any treatment-specific content against current institutional protocols before relaying it to patients. Future research should assess readability, alignment with health literacy, and patient comprehension of AI-generated CLP information. Full article

Objective: To develop and evaluate an edge-hosted Large Language Model (LLM)-assisted system for automated Neonatal Intensive Care Unit (NICU) discharge summary generation using an evidence-grounded, field-level evaluation framework. Methods: This implementation and evaluation study was conducted in a Level III NICU in India. Longitudinal patient records were constructed from integrated bedside physiologic data (ARCHITECT) and a structured electronic medical record (EMR) platform Although an embedded audio–video module was present, it was not used in this study. Automated discharge summaries were generated by MORPHEUS, an edge-hosted orchestration pipeline running on NVIDIA Jetson AGX Orin hardware with JetPack 6.2. Local orchestration, preprocessing, and workflow execution were performed on the edge device, while language generation inference was performed using the OpenAI gpt-4o-mini API. Documentation quality was assessed with an LLM-based evaluator guided by a clinician-defined rubric comprising 72 fields organized across 14 section contexts and scored on five dimensions: clinical accuracy, completeness, actionability, coherence, and non-hallucination. Paired, field-level comparisons were performed against clinician-authored summaries. Of 549 NICU admissions screened between 1 October 2024 and 3 November 2025, 401 met the inclusion criteria for evaluation. Prompt refinement was performed iteratively using omission-derived feedback without model weight updates. Results: Across 401 evaluated admissions, MORPHEUS-generated summaries demonstrated higher rubric-based scores and lower omission burden than clinician-authored summaries within the structured evaluation framework used in this study, with mean scores of 0.93 versus 0.75 for accuracy, 0.91 versus 0.67 for completeness, 0.93 versus 0.72 for actionability, 0.94 versus 0.74 for coherence, and 0.95 versus 0.78 for non-hallucination, with the largest absolute advantage observed for completeness. Error taxonomy analysis demonstrated fewer omissions, unsupported assertions, and contradictions in AI-generated summaries than in clinician-authored summaries. Iterative prompt refinement was associated with directional improvement across quality dimensions and reduced omission burden, with omission rate per patient decreasing from 2.484 to 1.807 in the later iteration. Conclusions: An edge-hosted LLM-assisted pipeline can generate NICU discharge summaries that meet or exceed clinician-authored documentation quality under a reproducible, clinician-grounded evaluation framework. These findings support the feasibility of deploying edge-orchestrated generative AI systems for high-stakes neonatal clinical documentation using a clinician-grounded field-level evaluation framework. Full article

Large Language Models (LLMs) are increasingly explored for educational tasks such as grading, yet their alignment with human evaluation in real classrooms remains underexamined. In this study, we investigate the feasibility of using OpenAI GPT-4o to evaluate short-answer quizzes and project reports in an undergraduate Computational Linguistics course. We collect responses from approximately 50 students across five quizzes and receive project reports from 14 teams. LLM-generated scores are compared against human evaluations conducted independently by the course teaching assistants (TAs). Our results show that GPT-4o achieves strong correlation with human graders (up to 0.98) and exact score agreement in 55% of quiz cases. For project reports, it also shows strong overall alignment with human grading, while exhibiting some variability in scoring technical, open-ended responses. We release all code and sample data to support further research on LLMs in educational assessment. This work highlights both the potential and limitations of LLM-based grading systems and contributes to advancing automated grading in real-world academic settings. Full article

Ultra-high-speed railway communication systems face several technical challenges due to extremely high mobility, including Doppler-induced channel variations, frequent handovers, and increasing network traffic. These challenges not only degrade communication reliability but also negatively affect the efficiency of network resource utilization. In this paper, we review the key technical challenges in ultra-high-speed railway communication environments and investigate artificial intelligence (AI)-based intelligent network control techniques to address these issues. In particular, we examine mobility management approaches focusing on AI-based predictive handover schemes and intelligent network control architectures based on the Open Radio Access Network (O-RAN). In addition, network resource management strategies are discussed through mobile edge computing (MEC)-enabled traffic offloading and task migration techniques. Through this analysis, we discuss the potential applicability of intelligent network control technologies for improving communication reliability and enhancing network resource utilization efficiency in ultra-high-speed railway communication environments. Full article