Abstract The rapid deployment of Generative Artificial Intelligence (GenAI) and Large Language Models (LLMs) necessitates a careful balance between computational efficiency, factual accuracy, and user satisfaction. While existing research often evaluates these dimensions in isolation, this study proposes... The rapid deployment of Generative Artificial Intelligence (GenAI) and Large Language Models (LLMs) necessitates a careful balance between computational efficiency, factual accuracy, and user satisfaction. While existing research often evaluates these dimensions in isolation, this study proposes an integrated empirical framework to analyze their combined influence on overall GenAI performance. Utilizing a dataset of 1,000 simulated AI usage sessions across six state of the art models (e.g., Claude 3.7, GPT-4o, Llama 3.1) and diverse application domains, we conducted correlation analyses, Ordinary Least Squares (OLS) regression, and statistical validation tests. Results indicate that prompt length is the dominant positive predictor of response latency, significantly outweighing the impact of decoding hyperparameters like temperature and Top-P. Among the evaluated models, Claude 3.7 demonstrated the optimal balance, achieving the highest success rate and user satisfaction alongside a comparatively low hallucination rate. Notably, task-specific analysis revealed that while translation tasks yielded the highest reliability, Retrieval- Augmented Generation (RAG) tasks exhibited unexpectedly high hallucination rates, suggesting that retrieval augmentation alone does not guarantee factual grounding. Furthermore, statistical validation confirmed the superiority of a proposed multi-layered “Double Lock” guardrail framework over baseline regex methods. Ultimately, this study indicated that concise prompt engineering and task specific model selection are more critical for optimizing LLM deployment than model identity alone, providing actionable insights for developing reliable, user centric AI systems. Read More Read Less

Abstract Neurofibromatosis type 1 (NF1) exhibits significant phenotypic heterogeneity, yet comprehensive exploratory analyses comparing familial and sporadic forms remain limited. This study investigated clinical symptom distributions and identified latent phenotypic clusters using the UCI NF1 dataset comprising 331... Neurofibromatosis type 1 (NF1) exhibits significant phenotypic heterogeneity, yet comprehensive exploratory analyses comparing familial and sporadic forms remain limited. This study investigated clinical symptom distributions and identified latent phenotypic clusters using the UCI NF1 dataset comprising 331 probands. We employed descriptive statistics, correlation analyses, symptom co-occurrence networks, UpSet plots, and hierarchical clustering to characterize phenotypic variability. Results revealed that café au lait spots, freckling, and Lisch nodules are the most prevalent and interconnected core manifestations. While familial (n=135) and sporadic (n=161) cases demonstrated substantial phenotypic overlap, familial cases exhibited modestly higher rates of optic glioma, skeletal dysplasia, plexiform neurofibromas, and overall tumor burden. Hierarchical clustering successfully delineated robust, biologically plausible symptom modules, including a core pigmentary cluster, a neuro ophthalmic and tumor related cluster, and a skeletal and developmental cluster. These findings underscore that NF1 manifestations form distinct, highly interconnected phenotypic clusters rather than occurring randomly. Recognizing these latent subtypes can significantly enhance diagnostic stratification, optimize longitudinal surveillance, and facilitate the early identification of highrisk patients. Furthermore, we propose a conceptual multi layer clinical decision support framework integrating machine learning and explainable artificial intelligence to translate these exploratory insights into personalized precision medicine. Future prospective validation incorporating multimodal data is warranted to refine these predictive models and ultimately improve long term clinical outcomes for NF1 patients. Read More Read Less

Abstract Mathematical reasoning poses a significant challenge for Large Language Models (LLMs) due to the necessity for structured symbolic manipulation, multi step deduction, and logically consistent inference. While Chainof Thought (CoT) prompting has substantially advanced reasoning capabilities, existing... Mathematical reasoning poses a significant challenge for Large Language Models (LLMs) due to the necessity for structured symbolic manipulation, multi step deduction, and logically consistent inference. While Chainof Thought (CoT) prompting has substantially advanced reasoning capabilities, existing datasets often lack transparent, verifiable reasoning trajectories and rigorous statistical validation. This study presents a statistically validated, large scale mathematical CoT dataset comprising approximately 100,000 samples spanning six core mathematical domains (Algebra, Geometry, Trigonometry, Matrices, Derivatives, and Integrals) and three predefined difficulty levels. The dataset preserves complete intermediate reasoning traces and final solutions in standardized LaTeX notation. We conducted comprehensive structural, descriptive, and inferential statistical analyses including Chi-square tests, Spearman's rank correlation, two-way ANOVA, and Tukey's HSD post hoc comparisons to validate the dataset's internal consistency. Results demonstrate a perfectly balanced distribution across topics and difficulties, with the mathematical topic emerging as the dominant determinant of reasoning complexity (partial n² = 0.995). Furthermore, the corpus exhibits rich symbolic diversity, authentic mathematical syntax, and structured command cooccurrence networks. This statistically validated corpus provides a reliable and robust benchmark for supervised fine tuning, curriculum learning, process supervision, explainable AI, and evaluating nextgeneration mathematical LLMs. Read More Read Less