SAIT Polytechnic

Recent years have seen tremendous advances in generative AI, making it more difficult to distinguish between artificially generated and real images. In response, deepfake detection has become a crucial field of study, and numerous models are currently performing well on common academic benchmarks. However, a significant drawback is that the majority of these models were tested and trained on datasets with a limited number of forgery techniques, calling into question how well they truly generalize. DF40 addresses this limitation by offering a large-scale benchmark with 40 different deepfake generation algorithms, ranging from GANs to more modern diffusion-based models and commercial tools. However, traditional detection methods have not been thoroughly tested on it. This work assesses three popular detection paradigms - MesoNet, EfficientNet-B4, and F3Net - each pre-trained on FaceForensics++, using the Entire Face Synthesis (EFS) and Face Editing (FE) image subsets of DF40 - the only categories provided in image format, as FS and FR are distributed exclusively as video. Our findings show that EfficientNet-B4 achieves the most consistent performance across both subsets (average AUC of 0.6952 on EFS and 0.6220 on FE), while F3Net degrades significantly on face editing techniques (AUC drop from 0.7149 to 0.5870), and MesoNet fails to generalize, exhibiting near-random AUC scores across most datasets. MesoNet underperforms due to an inherent bias favoring the fake category, whereas F3Net's lower performance on FE indicates that the FAD branch, as implemented in DeepfakeBench, may struggle to generalize to face editing manipulations; however, this observation is based on an evaluation of that single branch only. The results hint at the need for more flexible and reliable detection approaches, suggesting that traditional detectors find it difficult to generalize to the entire range of modern forgery techniques.

Digital transformation and artificial intelligence are reshaping nursing education by changing how students access information, complete academic tasks, and engage with technology-mediated learning. However, evidence on digital competencies, AI-related constructs, mediating mechanisms, and academic performance remains fragmented and methodologically uneven. This systematic review of empirical studies synthesized how digital competencies and AI-related constructs are associated with academic performance and learning-related outcomes in nursing education. Following PRISMA 2020 guidelines adapted to social science research, searches were conducted in Scopus and Web of Science Core Collection in March 2026, covering 2022–2026. Twenty-five empirical studies were included: 18 quantitative, 4 qualitative, and 3 mixed-methods studies. The evidence was concentrated in the Middle East and North Africa, Asia, and Europe. Findings suggest that digital competencies are associated with academic and learning-related outcomes mainly through self-efficacy, academic motivation, cognitive presence, and learning flow. AI-related evidence remains emerging, mixed, and context-dependent. Although some AI-assisted interventions reported favorable outcomes, one experimental study found greater knowledge gains with traditional text-based study than with ChatGPT-assisted learning. Therefore, AI integration should not be considered universally beneficial, but contingent on pedagogical design, task type, teacher guidance, AI literacy, responsible use, and critical verification.

Methylene blue (MB) is among the most extensively discharged cationic dyes in industrial effluents and remains a persistent environmental concern due to its high chemical stability, resistance to biodegradation, and adverse ecological impacts. Although hundreds of adsorbents have been developed for MB removal, the selection of practically applicable materials remains challenging because performance is commonly judged solely by adsorption capacity, while critical operational parameters are often overlooked. To address this limitation, this review proposes a comprehensive operational screening framework for identifying high-performance MB adsorbents suitable for real-world wastewater treatment applications. A total of 83 adsorbents reported in peer-reviewed studies published over the last five years were systematically evaluated across all major material classes. The proposed framework integrates five practical performance indicators: maximum adsorption capacity (Qmax > 100 mg/g), low adsorbent dosage (≤ 1 g/L), rapid equilibration time (≤ 300 min), exothermic adsorption behavior, and high swelling capacity for hydrogels, thereby enabling a multidimensional assessment beyond conventional capacity-based comparisons. Among the evaluated materials, 55 adsorbents satisfied the proposed selection criteria, including 43 activated carbons/biochars and 12 hydrogel-based adsorbents. Hydrogels emerged as the most promising class, exhibiting an average adsorption capacity of 1660 mg/g at an average dosage of only 0.4 g/L, combined with remarkable swelling capacities (90 g/g) and relatively short equilibration times (160 min). Sodium alginate–multi-walled carbon nanotube (SA-MWCNT) and poly(acrylic acid)/laponite hydrogels demonstrated the most balanced operational performance, achieving adsorption capacities of 1370–1600 mg/g, low dosages (0.33–0.75 g/L), rapid adsorption times (30–50 min), and swelling capacities ranging from 23 to 181 g/g. Carbon quantum dots derived from neem biomass were further identified as a promising emerging adsorbent, attaining 714 mg/g within only 30 min. This study introduces, for the first time, a practical decision-making framework that bridges the gap between laboratory-scale adsorption performance and engineering applicability. While the focus was on MB, the proposed performance indicators are adaptable for identifying adsorbents targeting other contaminants, including pesticides, anionic dyes and heavy metals.

This article examines how adult education can foster critical literacy and global competencies amid the ideological pressures of neoliberalism and the dominance of English as a global lingua franca. Drawing on Habermas’s lifeworld–system framework, Critical Discourse Analysis, and intercultural theory, the article identifies persistent structural forces that shape meaning-making, learner agency, and educational equity. In response, I advance the Applied Hermeneutic Criticality Model (AHCM), a conceptual framework that integrates macro-level ideological critique, meso-level institutional analysis, and micro-level reflexive engagement. The model clarifies how adult learners and educators navigate interconnected cultural, discursive, and systemic influences, and offers a theoretically grounded holistic pathway for cultivating critical being and intercultural competence thus contributing a unified approach for examining ideology, literacy, and power for future research on critical and emancipatory adult education.