D. Pamucar

Challenge in Applying ChatGPT in Education: Evaluating Potential Drawbacks through Failure Modes and Effects Analysis and Logarithm Methodology of Additive Weights

Large language models, such as ChatGPT, are transforming higher education by supporting personalized learning, content generation, and academic assistance. However, their widespread adoption also introduces educational risks that remain insufficiently prioritized in the literature. This study develops an integrated LMAW–DNMA–FMEA framework to identify, prioritize, and evaluate the most critical risks associated with ChatGPT adoption in higher education. Expert judgments are used to assess the relative importance and severity of identified risks, enabling a systematic ranking of potential failure modes. The findings indicate that overreliance on AI-generated content, degradation of critical thinking skills, and unreliable referencing represent the most significant challenges, while issues related to creativity and originality also require attention. Based on the results, targeted mitigation strategies are proposed for educators, developers, and policymakers to support the responsible integration of generative AI into educational environments. The proposed framework provides a structured decision-support approach for AI risk assessment in education and contributes to the development of evidence-based strategies for the effective adoption of large language models in higher education.

Dynamic Replenishment Policies for Vendor-Managed Inventory under Stochastic Demand: A Simulation-Based Comparative Study

Vendor-Managed Inventory (VMI) is a pivotal strategy for optimizing supply chain performance, yet it poses a significant challenge in balancing operational costs against service levels under stochastic demand. While dynamic policies are gaining traction, literature lacks a systematic comparison of the underlying trigger logic (reactive vs. proactive). This study addresses this gap by providing a rigorous comparative analysis of static versus dynamic inventory replenishment policies within a VMI framework for a pharmaceutical distribution network. We design and evaluate four distinct policies: a traditional static (s,S) policy and three novel dynamic policies—reactive, proactive, and inertial—that adapt replenishment triggers based on real-time, system-wide demand signals. The novelty of this work lies in the formal design and first systematic comparison of these distinct dynamic trigger mechanisms, particularly the "Inertial" policy, which utilizes a smoothed urgency signal to enhance resilience. A high-fidelity simulation-optimization framework is developed, where policy parameters are optimized via a Genetic Algorithm to ensure each strategy operates at its peak potential. The results, analysed using ANOVA and Tukey’s HSD tests, reveal that while all policies can be optimized to a statistically similar total cost (p = 0.782), they differ significantly in their ability to maintain service levels. The proposed inertial policy, which utilizes a smoothed urgency signal, demonstrates superior performance, significantly reducing stockouts by 21.5% and 29.7% compared to static and reactive policies, respectively, without incurring a statistically significant cost increase.  This demonstrates that integrating anticipatory, smoothed demand signals offers a robust pathway to enhancing service resilience without sacrificing economic efficiency.

An integrated framework for selecting augmented reality software to enhance occupational safety in the electro-energetic industry