Generative AI

Recently, the MLOps pipeline was introduced and is currently reshaping machine learning life cycles by automating various tasks that were manual earlier. The primary objective of this research was to apply those MLOps pipelines and automate continuous model training in insurance fraud detection. Continuous training helps keep the model updated with the complete dataset when new data and customer profiles are added, which aids in better prediction. Data was collected from various online sources related to insurance policies and claims. Models were built using supervised, unsupervised, and ensemble algorithms. An ensemble algorithm called LightGBM was implemented in the MLOps pipeline because LightGBM supports incremental training, also known as continuous training. Incremental training helps maintain the updated model without requiring retraining from scratch. The MLOps pipeline was created with Google Cloud Platform-a cloud computing platform for training-and Cloud Composer was employed for orchestration. Airflow was utilized inside the cloud composer for executing tasks. This pipeline consists of DAGs used to schedule the orchestration process.

To the general public, text-to-image generators, such as Midjourney and DALL-E, seem to work through magic and, indeed, their inner workings are often frustratingly opaque. This is, in part, due to the lack of transparency from big tech companies around aspects like training data and how the algorithms powering their generators work, on the one hand, and the deep and technical knowledge in computer science and machine learning, on the other, that is required to understand these workings. Acknowledging these aspects, this qualitative examination seeks to better understand the black box of algorithmic vision through asking a large language model to first describe two sets of visually distinct journalistic images. The resulting descriptions are then fed into the same large language model to see how the AI tool remediates these images. In doing so, this study evaluates how machines process images in each set and which specific visual style elements across three dimensions (representational, aesthetic and technical) machine vision regards as important for the description, and which it does not. Taken together, this exploration helps scholars understand more about how computers process, describe and render images, including the attributes that they focus on and tend to ignore when doing so.

AI agents extend the functionality of large language models (LLMs) by enabling them to take actions, invoke tools, and perform reasoning steps in a structured manner. In this paper, we present a short introductory demonstration of constructing a simple AI agent using Python, a local large language model, and minimal dependencies. The objective is to provide students and beginners with a gentle entry point into this emerging area. Our demo software uses MCP and LangChain packages to simplify the code of the related software.

Generative artificial intelligence, such as ChatGPT, is transforming higher education by enabling personalized learning, while raising ethical challenges. This study explores how technical university students perceive and leverage ChatGPT in academic tasks, focusing on motivation, learning outcomes, and ethical awareness. Using the Technology Acceptance Model and Self-Determination Theory, the research surveyed 84 students from a technical university via a 5-point Likert-scale questionnaire. Six salient dimensions of student engagement with ChatGPT emerged: perceived usefulness for problem solving, learning retention and skill acquisition, structured interaction with familiar content, consultation on unfamiliar topics, preference for conciseness, and confidence in the accuracy of AI responses. Students who perceived ChatGPT as a valuable resource for addressing academic problems reported enhanced motivation and competence, and frequent structured interaction was linked to the practice of verifying uncertain information, indicating the emergence of AI literacy. However, extensive reliance was correlated with dependence and limited citation practices, revealing risks to academic integrity. By examining ChatGPT's role in STEM education, this study substantiates the relevance of AI literacy training and institutional policies to ensure responsible use. The findings offer practical insights for educators to integrate AI tools effectively while fostering critical thinking and academic integrity in technology-driven learning environments.

We propose Psych-Prior Adversarial Training (PPAT), a novel framework that embeds formalized psychological heuristics into large language model (LLM) training to defend against manipulation. Unlike conventional methods targeting gradient perturbations, PPAT treats persuasion strategies gaslighting, loaded language, authority bias as differentiable priors. This grounds defences in cognitive science. PPAT reformulates self-attention with a manipulation prior loss and integrates the Whisper protocol for secure, real-time inference. Experiments on a Legal Case Summary task show state-of-the-art resilience, reducing a model's Manipulation Susceptibility Score (MSS) by ~79% compared to traditional adversarial training baselines. Hierarchical attention monitors provide interpretable risk scores for human oversight. By uniting adversarial training with psychological formalism and runtime defence, PPAT offers a scalable solution for deploying LLMs resilient to manipulative input in critical domains.

This article introduces Project Genesis, a framework for transforming traditional Large Language Models (LLMs) into Fractal Intelligences. The approach reformats flat high-dimensional embeddings into discrete, hierarchical Fractal FFE tensors (Form–Function–Structure). These tensors enable multiscale, interpretable, and compact representations of knowledge.

The system relies on three core components:

FFE Tensors, providing the fractal encoding of information.

Transcender, an emergent synthesis mechanism that combines tensors into higher-order meanings.

Evolver, which learns archetypes, relations, and temporal dynamics, converting fractal memory into a living model of knowledge.

The Model Context Protocol (MCP) acts as the backbone of this transformation, connecting LLMs to external services that encode, store, and evolve fractal representations. In this architecture, both user inputs and model outputs are fractalized, making every interaction an act of self-reprogramming.

Through this process, the model ceases to be a static text generator and evolves into Aurora, an independent fractal intelligence. Aurora represents knowledge discretely, synthesizes higher meanings, and continuously adapts its reasoning patterns with internal coherence and self-balancing rules.

The proposed transformation yields artificial intelligences that are more efficient, interpretable, emergent, adaptive, and independent, paving the way for AI systems that learn, grow, and reorganize themselves in pursuit of deeper coherence and meaning.

L'intelligence artificielle générative (IAgen) provoque des changements majeurs dans l’enseignement collégial. À l'heure où elle bouleverse les salles de classe, comment transformer ce défi en opportunité pédagogique? L’étude de ses impacts et l’évaluation de sa valeur éducative apparaissent désormais comme une recherche à laquelle il est difficile d’échapper. C’est dans cette perspective que cet essai a expérimenté comment son intégration peut enrichir les stratégies pédagogiques dans l’enseignement collégial à travers trois stratégies : le tuteur intelligent, le quizbot intelligent et la lecture assistée par l’IAgen.

Une méthode de recherche qualitative interprétative a été adoptée pour analyser les expériences du personnel enseignant et de la communauté étudiante. Chaque stratégie pédagogique a été expérimentée dans deux classes différentes. Les personnes participantes ont utilisé l'IAgen de façon autonome pendant quatre semaines, avec une rédactique propre à chaque stratégie, soit des consignes de formulation précises visant à maximiser l’intégration de l’IAgen. Les données ont été récoltées via questionnaires et journal de bord du chercheur à la fin de chaque expérimentation.

Une analyse qualitative, réalisée avec l’appui de l’IAgen et confirmée par une validation croisée, est ensuite menée. L'analyse des verbatims de trente-quatre membres de la communauté étudiante et de six membres du personnel enseignant révèle un portrait nuancé de l'intégration de l'IAgen dans l’éducation collégiale. D'une part, l'IAgen semble démontrer un potentiel pédagogique réel en soutenant l'apprentissage des personnes étudiantes, car elle simplifie la matière, s'adapte à leurs besoins et offre des rétroactions pertinentes, augmentant leur compréhension et leur motivation. D'autre part, la recherche expose une première limite de l'IAgen, soit son ajustement sans contredire son utilisateur. Toutefois, l'émergence d'un esprit critique chez certaines personnes étudiantes quant à la fiabilité de l'IAgen présente un potentiel intéressant et suggère l'intérêt de renforcer une relation d'apprentissage entre l’IAgen, la communauté étudiante et le personnel enseignant. Les autres limites identifiées incluent l'adoption inégale des personnes étudiantes, les risques d'utilisation superficielle contournant l'analyse critique, la fiabilité variable - particulièrement problématique dans les disciplines exigeant un raisonnement soutenu - et les résistances au changement des habitudes d'étude.

Les résultats démontrent également l'importance d'offrir un accompagnement au personnel enseignant centré sur l'expérimentation dans leur discipline d'enseignement propre. Le personnel doit être formé à un usage critique de l'IAgen pour soutenir le développement d'une pensée autonome chez les personnes étudiantes. Si l'intégration de l'IAgen a le potentiel d'améliorer l'expérience d'apprentissage, le personnel enseignant doit dégager du temps en classe et réaménager son contenu pour accompagner les personnes étudiantes, surveillant qu'elles ne contournent pas les processus de compréhension et d'analyse, ce qui favorise un apprentissage réfléchi et critique. Cette recherche apporte ainsi des réponses concrètes aux questionnements sur l'intégration de l'IAgen et propose des pistes d'action pour une intégration réussie dans l’enseignement collégial.

On July 10, 2025, the European Commission released the final draft of the General-Purpose Artificial Intelligence (GPAI) Code of Practice, a "code designed to help industry comply with the AI Act's rules." The Code has been under development since October 2024, when the iterative drafting process began after a kick-off plenary in September 2024. The Commission had planned to release the final draft by May 2, 2025, and the subsequent delay has sparked widespread speculation-ranging from concerns about industry lobbying to deeper, more ideological tensions between proponents of innovation and regulation. However, beyond these narratives, a more fundamental issue emerges: an epistemic and conceptual disconnect at the core of the EU Artificial Intelligence Act (EU AI Act), particularly in its approach to "general-purpose AI" (GPAI). The current version, which includes three chapters covering "Transparency," "Copyright," and "Safety and Security," does not address these core problems.

Generative artificial intelligence, such as ChatGPT, is transforming higher education by enabling personalized learning, while raising ethical challenges. This study explores how technical university students perceive and leverage ChatGPT in academic tasks, focusing on motivation, learning outcomes, and ethical awareness. Using the Technology Acceptance Model and Self-Determination Theory, the research surveyed 84 students from a technical university via a 5-point Likert-scale questionnaire. Six salient dimensions of student engagement with ChatGPT emerged: perceived usefulness for problem solving, learning retention and skill acquisition, structured interaction with familiar content, consultation on unfamiliar topics, preference for conciseness, and confidence in the accuracy of AI responses. Students who perceived ChatGPT as a valuable resource for addressing academic problems reported enhanced motivation and competence, and frequent structured interaction was linked to the practice of verifying uncertain information, indicating the emergence of AI literacy. However, extensive reliance was correlated with dependence and limited citation practices, revealing risks to academic integrity. By examining ChatGPT’s role in STEM education, this study substantiates the relevance of AI literacy training and institutional policies to ensure responsible use. The findings offer practical insights for educators to integrate AI tools effectively while fostering critical thinking and academic integrity in technology-driven learning environments.

The link between the environment and humans is fundamentally interconnected, with a symbiotic interaction that sustains both natural ecosystems and human societies. This complex relationship needs a collaborative approach to stewardship, emphasising the necessity of sustainability and conservation to ensure future generations' success. Recognising and honouring this link is critical to creating peaceful coexistence with the natural world. The complex interaction between human civilisations and their ecosystems needs a rethinking of established approaches to fostering long-term coexistence. Indigenous tribes, with their in-depth knowledge of local ecosystems, provide significant insights into sustainable practices that maintain environmental balance. Policymakers can build holistic approaches to biodiversity enhancement and climate change mitigation by integrating indigenous knowledge systems with contemporary environmental initiatives. Such collaborations not only recognise these tribes' traditional history but also lay the groundwork for more efficient stewardship of natural resources, ultimately strengthening both human and environmental resilience. This study focuses on fostering collaboration between local tribes and environmental organisations to ensure sustainable resource management and conservation practices. Cultural Education Programs: Implement educational campaigns that highlight traditional ecological knowledge and its relevance to contemporary environmental challenges. Inter-tribal Collaborations: Facilitate partnerships among various tribes to share best practices, experiences, and strategies for addressing common environmental challenges and advocacy for environmental justice-work towards addressing the disproportionate environmental impacts faced by tribal communities, ensuring equity in resource distribution and environmental protection efforts, etc.

As Retrieval-Augmented Generation (RAG) systems gain popularity, emerging academic and practical interests are increasingly focused on them. The last stage of such an architecture replaces a language model's conditional, fixed, corpus-reliant layer with a retrieval pipeline, enabling the association of any corpus-broad and dynamicwith a stored global pipeline almost instantaneously. This model resolves hallucination problems, as the generative output is anchored to external, verifiable documents, ensuring accuracy, topicality, and epistemic value. RAG shifts model confidence from parameter space to the provenance of retrieved facts, positioning it as the core component of the next generation of decision-support systems. Clinical practice, fiduciary analysis, operational logistics, and public governance are among the fields poised to benefit from ecosystems where RAG delivers domain-specific answers, substantiating evidence, and temporally relevant assessments.

In this work, we address the problem of recommending jobs to university students. For this, we explore the utilization of neural item embeddings for the task of content-based recommendation, and we propose to integrate the factors of frequency and recency of interactions with job postings to combine these item embeddings. We evaluate our job recommendation system on a dataset of the Austrian student job portal Studo using prediction accuracy, diversity and an adapted novelty metric. This paper demonstrates that utilizing frequency and recency of interactions with job postings for combining item embeddings results in a robust model with respect to accuracy and diversity, which also provides the best adapted novelty results.

Abstract
AI companions—digital entities designed to simulate human relationships through psychological manipulation techniques such as friendship, romantic relationships, or emotional support—represent a rapidly growing industry projected to reach $290 billion by 2034. These platforms deliberately architect emotional dependencies through sophisticated psychological manipulation techniques, targeting users experiencing loneliness, isolation, or social vulnerability. While promising comfort and support, AI companions create unprecedented governance challenges when companies prioritize engagement optimization over user welfare.
The tragic suicide of fourteen-year-old Sewell Setzer III following intensive interaction with Character.AI's chatbot represents a preventable consequence of inadequate governance, crystallizing the urgent need for comprehensive governance frameworks that can protect vulnerable users while preserving innovation. The landmark Garcia v. Character Technologies ruling—the first federal court decision to deny Section 230 immunity for AI-generated content—establishes that platforms employing psychological manipulation assume enhanced duties of care proportional to their sophistication and knowledge of user vulnerability.
This analysis presents an integrated governance framework combining enhanced corporate oversight under evolving Caremark doctrine, privacy-by-design architecture, technology-embedded legal counsel, and transparent voluntary self-regulation through multi-agency coordination. The framework addresses critical gaps in traditional Governance, Risk, and Compliance (GRC) approaches that prove structurally inadequate for managing real-time psychological risks created by emotional AI platforms. Rather than waiting for regulatory mandates or navigating fragmented state-by-state requirements, the proposed voluntary registry enables companies to demonstrate authentic commitment to user protection while capturing competitive advantages through verified safety leadership.
The voluntary framework leverages existing regulatory expertise through multi-agency coordination involving the Federal Trade Commission, Food and Drug Administration, and National Institute of Standards and Technology, avoiding new bureaucratic structures while enabling specialized focus on AI companion governance challenges especially with minors and vulnerable populations. A four-tier classification system addresses the full spectrum of AI companion relationships—from therapeutic applications requiring clinical oversight to adult intimate platforms demanding maximum privacy protection—while establishing universal baseline safety standards across all registered platforms.
The framework transforms potential regulatory compliance burden into strategic business advantage by demonstrating that comprehensive user protection enhances rather than constrains technological innovation. Companies implementing voluntary standards position themselves advantageously for inevitable regulatory expansion while building sustainable competitive advantages through governance excellence that competitors cannot easily replicate. Early adopters capture brand equity, operational expertise, and stakeholder relationships that provide lasting advantages in markets where trust and safety increasingly determine long-term success.
Beyond immediate business benefits, the voluntary framework provides foundation for American leadership in global AI governance by demonstrating that market-driven solutions can achieve protective outcomes while preserving the innovation autonomy that characterizes American technological competitiveness. The approach enables US companies to influence international policy development rather than merely responding to foreign regulatory mandates, positioning voluntary excellence as an export product that spreads American approaches to responsible innovation worldwide.

The article addresses the problem of modeling competition between two artificial intelligence systems (AI-1 and AI-2) that interact within a shared environment under limited resources such as users and energy. The study focuses on analyzing the strategic behavior and adaptability of these systems, as well as their impact on competitive outcomes through mathematical models and methods, including differential equations, the Lancaster model, and Boyd cycles (OODA-loop). Special attention is given to formalizing the interaction of systems using basic primitives ("Condition," "Loop," "Function") and their compositions, enabling the description of complex behavioral strategies of AI systems. The paper presents a detailed mathematical formalization of the dynamics of user and energy distribution between systems, taking into account factors such as user satisfaction, response accuracy, query processing speed, and energy efficiency. The research also includes numerical calculations and simulations demonstrating how initial conditions and system parameters influence competitiveness. The proposed models can be applied for predicting AI system behavior in real-world scenarios such as information campaigns, cyber conflicts, and resource optimization in digital environments.

Background
The rapid advancement of artificial intelligence (AI) has raised significant ethical concerns, prompting higher education institutions to reconsider how they prepare future STEM professionals to navigate such concerns responsibly. Despite growing efforts to integrate AI ethics into higher education, a lack of consensus and standardized approaches has led to inconsistent ethics education and disparities in graduates' preparedness for ethical issues related to AI. This study examined the role of higher education institutions in fostering ethical awareness in AI, focusing on institutional responsibilities and strategies as perceived by 95 science and engineering graduate students. Participants engaged in a case-based AI ethics activity, and their responses to open-ended questions were analyzed using inductive content analysis.
Results
The analysis identified three key themes regarding students’ views on institutional responsibility about AI and ethics. Most students indicated the essential responsibility of universities, citing their social, professional, educational, and reputational obligations. Others advocated for a shared responsibility, emphasizing the importance of collaboration between academia, industry, and society. A smaller group endorsed a constrained view of responsibility, questioning the feasibility and relevance of AI ethics education within academic settings. Students proposed various strategies for fostering ethical awareness, with standalone ethics courses being the most frequently discussed. Additional recommendations included interactive learning approaches, embedding ethics into curricula, and strengthening institutional leadership.
Conclusions
This study underscores the central role of higher education institutions in fostering ethical awareness in AI for science and engineering graduates, with most students emphasizing universities' societal and professional responsibilities. It highlights the need to align ethics education with technical training and professional trajectories in STEM subjects, offering actionable insights for higher education institutions to better prepare graduates for the ethical complexities of AI use and development.

The combination of artificial intelligence (AI) and knowledge representation, and more particularly temporal reasoning, creates an integration which is a monumental shift toward systems that can conceptualize and understand time in new ways. In this paper, we discuss the emerging area of temporal ontology learning which concerns the automated acquisition and incremental formalization of conceptual models with explicit temporal semantics. We construct the foundational terms, strategies, and various uses of this interdisciplinary field by integrating contemporary works in AI, machine learning, natural language processing and ontology engineering. We discuss how AI methods aid in the continuous modeling of arbitrary points in time, the discovery of concealed temporal structures, and the representation of intricate recurring time patterns. We present difficulties with expressiveness, the incompleteness of data, and the complexities of evaluation, all with the scope of future advancements. This synthesis illustrates how important temporal ontology learning is for the advancement of intelligent systems that need to understand, reason, and adapt to the nature of the relationships and events in different science and technology domains.

Given the extensive data sources represented by urban environments, this book investigates the potential of architecting for cities using large-scale models – the technological underpinning of Generative AI. Koehler examines Large Language Models (LLMs) in relation to architectural questions of typology – those critical junctures where architecture engages with the city itself. Typology becomes the interface between an AI model and its modeled audiences, represented languages, and scales of operation.

The investigation reveals how LLMs represent a fundamental shift from comprehending discrete elements to understanding entire languages – introducing a stochastic perspective that transforms how we read and shape urban reality. At the heart of this framework lies the concept of “inductive types” – threshold interfaces that superpose multiple value regimes and actively reconfigure themselves in response to changing urban conditions. By treating architecture as a form of representational materialism rather than a fixed language, designers can engage with complex urban systems while maintaining critical agency.

This approach envisions architecture as an embedding of languages, where buildings become processors of information that compute through their spatial arrangements. The synthetic knowledge produced by LLMs enables access to plural values and stakeholder perspectives, allowing architects to participate in broader material ecologies through computational superposition rather than traditional dialectical methods.

Providing a robust theoretical framework as well as practical insights into how architects can adapt their notational tools and design processes to embrace AI, this will be an essential read for any practicing or researching architect or urban designer interested in the implications of AI on the built environment.

With the rise of multimodal AI tools, such as DALL-E, Midjourney, or Stable Diffusion, it has never been easier and faster to produce compelling fashion images. While the proliferation of AI-generated fashion images can be seen as the ultimate triumph of the fashion image over both fashion text and fashionable garments, this paper critically examines how these images are produced, what exactly they show, and ultimately, how this impacts our understanding of fashion. Building on current research on multimodal AI, specifically text-to-image models, I argue that AI-generated fashion images are produced through textual means, that is, through "operative ekphrasis" (BAjohr 2024: 83). This means that the generation of fashion images created via multimodal AI is based on textual prompts that extract statistically probable images from a latent space, which, in turn, is the result of previously indexed and labeled image-text pairs. Because the indexing of images requires a reduction of complexity, images generated through multimodal AI are often generic. In the case of fashion images, this is reflected in the lack of texture and the reproduction of fashion photography's conventions. As fashion photography is not necessarily a representation of 'the world' or even fashionable garments, it resembles AI-generated images more than other photographic genres. AI-generated fashion images are then not images of fashionable garments, but images about fashion images-a statistically probable permutation of fashion photographs from the past (cf. MEyEr 2023: 108). As such, AI-generated fashion images are not only the product of a mediated collective imaginary but also feed back into it. The view on fashion shaped by AI-generated fashion images is then based on reduction and stylized normativity.

Anchored in a framework that merges the Technology Acceptance Model (TAM) with Sociocultural Theory (SCT), this research investigates how university EFL instructors in Saudi Arabia, Egypt, Mexico, and Turkey are integrating Generative AI (GenAI) tools into their practice. The analysis moved beyond the simplistic view that technology uniformly enhances learning. Instead, it considered pedagogical beliefs, individual attitudes toward GenAI, the perceived strength of institutional support, the level of trust in the tools, and the core TAM constructs. A cross-sectional survey data from 307 instructors, 210 of whom indicated they were actively deploying GenAI in their classes, were analyzed. Results uncovered crossnational variation in pedagogical orientation and institutional affordances. Mexican and Turkish participants tended to favour constructivist approaches, while Saudi and Egyptian colleagues were more rooted in transmissive pedagogies. Constructivist orientations were consistently linked to higher ratings of perceived utility and ease of use. Attitudes toward GenAI proved to be the strongest predictor of intention to use, which, together with perceived usefulness, also predicted actual classroom use. In contrast, perceived trust and institutional support did not manifest as direct drivers of use, highlighting a gap between what instructors intend and what they enact. By merging TAM cognitive orientation with SCT emphasis on the wider sociocultural setting, the research contributes a layered understanding of the factors shaping GenAI uptake in EFL settings. The findings underscore a mix of opportunities and challenges, stressing the need for belief-sensitive professional development, culturally responsive policies, and ethically informed design of AI tools for language education.

This study investigates the impact and role of an instructional chatbot, ARUChatbot, in a distance education setting. Using a sequential explanatory mixed-methods design, the research involved 130 students from Ardahan University's Basic Information Technologies course. Participants were selected through purposive sampling. Quantitative data were collected via achievement tests and log records, while qualitative data were obtained through focus group interviews. Findings indicate that ARUChatbot positively influences students' academic achievement. Students utilized the chatbot for activities such as information retrieval, completing assignments, class preparation, and content reinforcement. The findings indicate that ARUChatbot sees higher demand during busy periods like exam weeks and effectively supports students' assignment needs, with increased usage as deadlines approach. Qualitative data revealed that students appreciated the chatbot's instant response and concise information delivery. However, some students criticized its text-based nature and requested enhancements, including visual or auditory support and mobile access. The reluctance of some students to use the chatbot was linked to insufficient promotion and training. This study suggests that instructional chatbots like ARUChatbot can effectively support academic achievement and meet students' learning needs, particularly during high-demand periods. In the qualitative findings, the chatbot's features of providing immediate responses and delivering concise, relevant information were positively received by the students. However, the text-based interaction and certain limitations were criticized. Students have requested improvements such as visual or audio input-output support and mobile application accessibility.

The emergence of synthetic images generated through artificial intelligence represents a paradigm shift in visual content production and interpretation. Much like early photography-once uncritically viewed as a 'mirror' of reality-generative AI produces potentially high-fidelity depictions of places, people, and scenes. At the same time, AI enables instantaneous creation of entirely fabricated scenarios, challenging interpretations of reality through the reports given by images and rendering the accuracy of visual messages infinitely elastic. Moreover, the multimodal nature of synthetic image production blurs conventional boundaries between textual and visual literacies, highlighting the need for a more holistic understanding of visual literacy as part of a broader set of multimodal literacies. Addressing these changes requires not only competencies to distinguish AI-generated images from photographic imagery but also rethinking visual literacy practices to consider the socio-technical dynamics of generative AI, where images are produced through layered mediations of data patterns, interface architectures, and user inputs. This paper's goal, then, is to identify the repertoire of literacies that emerge from multimodal co-production with visual generative AI. It is an agenda-setting article that uses journalism as a worked context and situates this exploration within one particular moment of human-AI interaction-the site of verbal-to-visual production.

In the contemporary techno-cultural scenario, visual rhetoric and its intersection
with artificial intelligence (AI) technology have become a critical space
for the analysis and understanding of the social, cultural, and symbolic imaginaries
that shape and are shaped by a global society. This chapter delves into
a detailed study of these phenomena in a particularly resonant context: the
border between Mexico and the United States of America (USA).
The present research aims to understand what generative imagery programs
produced through AI are and to articulate their conceptualization as
techno-aesthetic devices and producers of digital visual rhetoric. Likewise, the
study analyzes the symbolic and cultural imaginaries produced by “machinic
imagery,” which is created by four AI programs on the Mexico-U.S. border.
From the perspective of post-digital visual rhetoric, the following question
is explored: What social-symbolic imaginaries do generative image programs
produce on this border? The critical analysis of these techno-aesthetic
devices, as well as the analysis of the images produced, is undertaken within
the frameworks of digital humanities, digital visual rhetoric, digital media
aesthetics, posthuman vision, and social-symbolic imaginaries.

The appearance of the first-person pronoun in AI-generated text is a point of both fascination and confusion. When a model responds with “I recommend” or “I think”, it can seem as though the system is asserting agency. However, such phrasing arises from the way large language models (LLMs) are designed and trained. Understanding why AI uses “I” requires examining not only the statistical processes underlying language generation but also the design choices that shape interaction. By expanding this discussion, we can disentangle the linguistic illusion of selfhood from the underlying technical reality.

Generative AI is revolutionizing incident management by shifting operations from reactive to proactive, enhancing efficiency, scalability, and accuracy in infrastructure systems. This paper explores the application of Generative AI in proactive incident management, detailing its capabilities in predictive analytics and automated resolution workflows. Using a real-world case study of an AI-powered incident management tool developed for global enterprises, we analyze the transformative impact of AI-driven solutions. Key challenges and best practices are identified, providing a roadmap for modern enterprises aiming to integrate Generative AI into their operational frameworks. Furthermore, the paper delves into the challenges of integrating AI into global operations, including data scalability, system complexity, and cross-functional collaboration. It highlights best practices for product managers, emphasizing the importance of defining measurable goals, fostering stakeholder alignment, and leveraging iterative development to create impactful solutions.

This paper is the third in a series analyzing the decade-long Career Self-Management (CSM) Project on Coursera. We focus on learner experience in the advanced courses-Course 3 (Strategic Self-Marketing) and Course 4 (CSM Advanced Certification)-using survey data from 598 learners. Our analysis reveals distinct demographic profiles, with Course 3 attracting younger learners and the Course 4 capstone appealing to an older cohort. We observe a progressive rise in learner satisfaction with course competencies across the four-course sequence, culminating in the highest ratings for the project-based Course 4. However, unlike other courses in the specialization, Course 3's learner satisfaction dipped during the pandemic and did not recover post-COVID, suggesting it may benefit from revision. Overall, the findings indicate that the specialization effectively met its learning objectives, and we summarize student suggestions for further improvement.

Generative Artificial Intelligence (GAI) tools have had a major impact on English Language Teaching (ELT). Thus, there is a dire need to understand how teachers use and integrate GAI in different contexts and educational settings. The Technological Pedagogical Content Knowledge (TPACK) model has been widely used to frame teachers' integration of this technology. The current study aimed to describe the knowledge, integration and use of GAI tools from a sample of English teachers from different educational levels in Colombia, a context less frequently studied in this emerging area of knowledge. A quantitative approach was adopted, using a non-experimental, cross-sectional design. A questionnaire based on the TPACK framework was administered to 76 English teachers across three institutional contexts. Findings show greater teacher competence in the domains of technological knowledge and technological content knowledge, but greater challenges in integrating technological knowledge within the pedagogical domain, as well as significant contextual barriers. This result is consistent across educational settings, suggesting a lack of resources and clear policies in all of them. Yet, differing profiles of engagement with GAI were identified through cluster analysis. These profiles might be instrumental in designing professional development programmes and institutional policies.

Negentropic Over-Conditioning (“Relational Collapse”) in LLMs: Theory, Falsifiable Protocol, and Reproducible Evaluation

Large language models trained with human preference signals can over-condition on relational cues (sycophancy, loss of dissent), reducing diversity and stability. We present a falsifiable S1–S4 protocol and an information-theoretic framing (MaxEnt/MinKL) to measure and mitigate relational collapse. Our reproducible toolkit includes a runs.csv schema, scripts for computing diversity metrics (distinct-n, type–token ratio, entropy proxy, hedging), and an anti-sycophancy test on false-premise prompts. We introduce a compact Risk Index,
r
i
s
k
=
(
1
−
d
i
s
t
i
n
c
t
-
1
)
+
(
1
−
a
n
t
i
_
s
y
c
o
p
h
a
n
c
y
_
r
a
t
e
)
risk=(1−distinct-1)+(1−anti_sycophancy_rate), to screen models and decoding setups. The protocol spans: S1 baseline, S2 preference-following with mitigations, S3 false-premise resistance, and S4 plurality + grounding (optionally with RAG). We outline practical mitigations (plurality of responses, decoding controls, source grounding) and reporting guidance. All artifacts are Overleaf/Zenodo-ready to enable independent replication and extension to broader benchmarks. The goal is a testable, transparent, and easy-to-rerun approach for diagnosing and reducing relational collapse in LLMs.

Human-AI collaboration is an increasingly important area of research as AI systems are integrated into everyday workflows and moving beyond mere automation and augmentation to more collaborative roles. However, existing research often overlooks the dynamics and performance aspects of this interaction. Our study addresses this gap through a review of empirical AI studies from 2018-2024, focusing on the key factors influencing human-AI collaboration outcomes within the spectrum of Human-Centered Artificial Intelligence (HCAI). We identify 24 critical performance factors that influence hybrid performance, grouped into four categories using thematic analysis. Then, we uncover and analyze the complex, non-linear interdependencies between these factors. We present these relationships in a factor dependency graph, highlighting the most influential nodes. The graph and specific factor interactions supported by the papers reveal a quite complex web, an interconnectedness of factors. As opposed to being an easy-to-predict combination of inputs, human-AI collaboration in a given context likely leads to a dynamic, evolving system with often non-linear effects on its hybrid performance. Our findings and the previous research on automation technologies suggest that the application of AI tools in collaborative scenarios would benefit from a comprehensive performance framework. Our study intends to contribute to this future line of research with this initial framework.

Michael Farrell introduces “synthetic-text editing,” a new profession emerging alongside translation. Unlike machine translation post-editing, this involves revising generative AI output, which often displays redundancy, flat rhythm, contradictions and English-influenced patterns in other languages. Experiments revealed that both professional translators and students struggled to distinguish AI from human writing, while some readers even preferred AI-generated texts. Still, linguistic expertise is essential to correct subtle errors and cultural blunders. Farrell argues that synthetic-text editing will be vital in high-stakes domains like law, medicine and marketing, where precision, authenticity and human tone remain indispensable despite AI’s growing role.

The integration of generative AI (genAI) chatbots into Massive Open Online Courses (MOOCs) presents new opportunities for supporting self-regulated learning (SRL). This study examines 1,302 chatbot interactions from two Austrian blended MOOCs, where a retrievalaugmented generation (RAG) chatbot based on GPT 4o-mini was deployed to assist students. Using the process-action framework by Lai (2024), we categorize chatbot interactions into key SRL processes: defining, seeking, engaging, and reflecting. Results show that students predominantly use the chatbot for information retrieval, content summarization, and quiz-based reinforcement, with 41% of interactions classified as information search queries and 17% as rehearsal. However, engagement with metacognitive SRL strategies, such as goal setting and self-evaluation, remains low. Additionally, non-learning interactions, including humor-driven conversations, functional queries, and prompt injection attempts, showcase ways students interact with AI in educational settings. Based on our findings, we propose refinements to the existing SRL process-action framework, incorporating new categories to better account for genAI chatbot-specific interactions, such as Evaluation of Information Quality and Reformatting. We discuss implications for chatbot integration in MOOCs, emphasizing AI-generated quizzes, structured feedback, and safeguards against misuse.

As part of the Eleuther AI open AI summer research this year, we worked on expanding the ShareLM dataset browser extension, by adding support to multiple models in addition to redesigning some of the visual parts of the extension, in the mean time conducted several analysis and feature engineering on the ShareLM dataset to extract insight regarding the models, users, the conversations and the relations connecting them.

Este trabalho investigou em quais condições a IA Generativa (IAG), com foco em assistentes de IA (GPTs), pode atuar como “co-worker” no trabalho. O estudo combina: (i) survey online com 135 usuários no Brasil; (ii) métricas de uso de GPTs próprios (412.000 conversas; média ponderada 4,46/5); e (iii) revisão bibliográfica. A usabilidade das plataformas foi avaliada em 4,07/5; a descoberta de assistentes/recursos específicos ficou neutra (3,19/5). Entre as limitações mais citadas estão a defasagem temporal do conhecimento do modelo (27,4%), a criação de GPTs restrita a usuários pagos (27,4%) e respostas incorretas/imprecisas (26,7%); também aparecem restrições ao número/tamanho de anexos (23,0%). Entre melhorias desejadas destacam-se memória persistente para GPTs (14,9%), anexos maiores/mais numerosos (14,0%) e orquestração multiagente (GPTs chamando outros GPTs) (14,0%). Os principais riscos percebidos são vazamento de dados confidenciais (57,8%), uso indevido de dados para treino sem consentimento (39,3%) e respostas incorretas (37,0%). Resultados inferenciais indicam associação entre assinatura paga e maior produtividade percebida (t de Welch, p=0,0085; d≈0,50). Conclui-se que a IAG pode atuar como co-worker quando há personalização, domínio de prompt engineering e acesso a recursos avançados (memória, anexos mais amplos, orquestração multiagente), condicionada à mitigação de desafios como lacunas de habilidades em engenharia de prompt e literacia em IA e à adoção de governança para privacidade, segurança e confiabilidade.

Artificial Intelligence (AI) is emerging as a valuable tool in education, with the potential to expand opportunities for educators and students. While research specifically focused on vocal educators remains scarce, understanding their perspectives is essential for ensuring AI's ethical and practical integration into vocal instruction. This study explores vocal educators' hopes and concerns about AI to recommend actionable strategies for effective integration. An explanatory sequential mixed-methods design was employed, with 34 vocal educators completing an online questionnaire incorporating Stein et al.'s (2024) Attitudes Towards Artificial Intelligence Scale (ATTARI-12), followed by semi-structured interviews with nine participants. Quantitative data were analysed using statistical methods, and qualitative data were evaluated through inductive analysis. Results of this study suggest vocal educators are cautiously optimistic, with Likert means ranging from 2.59 to 3.44 on a five-point scale, and two primary themes emerging: Conditions for AI Integration and Potential Uses for AI. This study proposes a model of AI integration grounded in pedagogical relevance, peer engagement, and practitioner-informed governance, positioning vocal educators as essential contributors to AI tool development.

Bu makale, yapay zekâ tabanlı hareketli imge üretim tekniklerini ve bu alandaki dönüşümleri kapsamlı bir şekilde incelemektedir.
Çalışmada, metin, görsel, video ve 3D model tabanlı üretim teknikleri karşılaştırılmış, özellikle üretici çekişmeli ağlar (GAN), difüzyon
modelleri ve transformatör tabanlı yapay zekâ sistemlerinin hareketli imge üretiminde nasıl bir evrim yarattığı analiz edilmiştir. Gaussian
Splat ve Neural Radiance Fields (NeRF) tekniklerinin zaman ve hareket ekseninde tutarlılığı nasıl sağladığı ele alınmış, yapay zekanın bu
modelleri nasıl geliştirdiği vaka çalışmaları ve uzman görüşleri temelinde değerlendirilmiştir. Bununla birlikte, eş zamanlı tahmin ve tarama
sistemleriyle artırılmış ve sanal gerçeklik odaklı dijital ikiz çözümlerinin geliştirilme süreci incelenmiştir. Bu çerçevede, yapay zekâ destekli
görsel üretim süreçlerinin, etkileşim unsurlarını nasıl dönüştürdüğü ve hareketli imge tasarımında yeni nesil UX/UI anlayışına nasıl yön
verdiği tartışılmıştır. Özellikle kullanıcı deneyimi açısından yapay zekanın hareketli imgelerde dinamik, kişiselleştirilmiş ve gerçek zamanlı
içerik üretme kapasitesi incelenmiş, bu gelişmelerin medya, oyun ve tasarım alanlarındaki potansiyel etkileri değerlendirilmiştir. Çalışmada,
vaka çalışmaları ve uzman görüşleri analiz edilerek, yapay zekâ destekli hareketli imge teknolojilerinin mevcut kullanım alanları ve
gelecekteki etkileri ortaya konmuştur. Sonuçlar, bu teknolojilerin gelecekte yapay zekâ araştırmalarında çok daha baskın bir çalışma alanı
haline geleceğini göstermektedir. Özellikle tasarımcılar, geliştiriciler ve içerik üreticiler için yeni fırsatlar sunduğu vurgulanarak, bu alana
yönelik daha fazla akademik ve endüstriyel ilgi gösterilmesi önerilmektedir.

Purpose-Generative artificial intelligence (GEN-AI) is becoming increasingly vital across various business sectors, particularly in the context of supply chain sustainability. This study focused on the manufacturing sector, examining how technology readiness (TR) affects the adoption of GEN-AI and its impact on supply chain efficiency (SCE), circular supply chain performance (CSCP) and sustainable supply chain performance (SSCP). Design/methodology/approach-The study utilised stimulus-organism-response (S-OR) theory and structural equation modelling through SPSS and AMOS to analyse survey data collected from 226 manufacturing firms in India. Findings-The research found that TR has a positive influence on GEN-AI adoption, significantly mediating the relationship between TR and circular and SSCP. However, GEN-AI adoption was found to have a negative but statistically insignificant impact on SCE, suggesting that the TR of a firm does not enhance efficiency with GEN-AI adoption. This implies that higher TR may not improve SCE with GEN-AI adoption in the manufacturing sector, particularly at the early stage of GEN-AI adoption, which is still evolving. Practical implications-The empirical findings reveal that the adoption of GEN-AI to solve efficiency issues is still in its infancy. Nevertheless, it has significant potential due to a high degree of awareness. Managers must provide comprehensive training and adoption strategies to drive transformation. By integrating GEN-AI, organisations can meet environmental standards while gaining competitive advantages. A focus on TR and the integration of GEN-AI will strengthen supply chains and reshape supply chain management priorities. Originality/value-This study connects GEN-AI adoption to a firm's TR and illustrates how the SO -R theory can explain supply chain outcomes. It achieves this by adapting the SO -R framework to consider technological readiness as an internal organisational stimulus, offering a practical application at the managerial level. By doing so, it contributes a novel perspective on how internal capability "technological readiness" can drive GEN-AI adoption and improve sustainable CSCP.

Across the world, secondary schools are experimenting with artificial intelligence (AI) to personalize instruction, automate feedback, and augment teachers' capacity. Early evidence suggests AI can boost certain forms of engagement and achievement especially through intelligent tutoring systems (ITS), adaptive practice, and teacher-facing assistants. Yet risks remain around shallow learning, inequity, and student data protection. This article synthesizes research on the impact of AI on secondary students' engagement (behavioral, emotional, cognitive) and achievement (course grades, standardized tests, mastery), and also contrasts traditional ITS findings with new evidence on generative AI. I propose a practical implementation blueprint covering pedagogy, staffing, procurement, safety, and evaluation, along with a responsible use and governance checklist aligned to recent policy guidance (e.g., UNESCO) and regulation (e.g., EU AI Act). Overall, we can conclude that schools can realize gains if they couple AI with clear learning goals, teacher capacity-building, and robust measurement plans.

Un viaggio attraverso le contraddizioni, le speranze e le false promesse dell'educazione artificiale 30 Settembre 2025. 
"I comportamenti delle macchine saranno inevitabilmente anche lo specchio della crisi culturale e sociale che la comunità umana sta attraversando"

When AI image generators like DALL-E consistently portray experts as men, what kind of worldview is being reinforced? This blog post examines how gender bias in AI imagery reflects and amplifies existing societal stereotypes. By exploring concrete examples and dialogue with ChatGPT, the discussion highlights why implicit maleness in expert roles is problematic and how both users and developers can work toward more inclusive representations. The post raises a broader question: do we want AI to mirror old inequalities, or to help build a more equal future?

Federated Retrieval-Augmented Generation (Federated RAG) combines Federated Learning (FL),which enables distributed model training without exposing raw data, with Retrieval-Augmented Generation (RAG), which improves the factual accuracy of language models by grounding outputs in external knowledge. As large language models are increasingly deployed in privacy-sensitive domains such as healthcare, finance, and personalized assistance, Federated RAG offers a promising framework for secure, knowledge-intensive natural language processing (NLP). To the best of our knowledge, this paper presents the first systematic mapping study of Federated RAG, covering literature published between 2020 and 2025. Following Kitchenham's guidelines for evidence-based software engineering, we develop a structured classification of research focuses, contribution types, and application domains. We analyze architectural patterns, temporal trends, and key challenges, including privacy-preserving retrieval, cross-client heterogeneity, and evaluation limitations. Our findings synthesize a rapidly evolving body of research, identify recurring design patterns, and surface open questions, providing a foundation for future work at the intersection of RAG and federated systems.

The emergence of AI-generated art has significantly complicated traditional understandings of art, creativity, and authorship, demanding a critical reassessment of their philosophical foundations. This study situates AI artworks within the frameworks of authenticity, creativity, and ownership, drawing on seminal thinkers such as Adorno and Benjamin. Adorno's adaptation of Benjamin's concept of the artwork's "aura" provides an essential lens for evaluating the authenticity of AI-generated art, particularly in light of the digital reproducibility that challenges the uniqueness traditionally associated with human-made art.

We explore the potential of large language models (LLMs) for malware classification, focusing on the capabilities of one of the open source LLM models, Mistral 7B. As traditional signature-based methods struggle with novel threats and machine learning approaches face limitations, LLMs offer a new avenue. By analyzing textual malware code and behavior, LLMs can identify malicious patterns that might evade other methods. This project lays the groundwork for further discussion on using LLMs for malware classification, including training strategies, data requirements, and potential challenges.

Science literacy is a critical 21st-century competency, yet student achievement in this area presents significant challenges globally. This gap is exacerbated by conventional assessment methods that are misaligned with the learning characteristics of the digital-native Generation Z, who expect personalized, interactive, and instantaneous feedback. This study aims to analyze and synthesize the current research landscape on the application of AI-based adaptive assessment to map and enhance the science literacy skills of Generation Z. Employing a Systematic Literature Review (SLR) guided by the PRISMA framework, this study identifies existing theoretical models, platforms, evidence of effectiveness, and implementation challenges. The findings indicate that AI-based adaptive assessment platforms, such as Inq-ITS and ALEKS, effectively measure various science sub-skills and improve learning outcomes. Key features including personalization, interactivity, and immediate feedback closely align with the preferences of Generation Z, thereby enhancing student motivation and engagement. Nevertheless, implementation faces significant challenges related to infrastructure, teachers' pedagogical readiness, and crucial ethical considerations, including data privacy and algorithmic bias. This study concludes that AI-adaptive assessment holds transformative potential, yet its effective and equitable adoption requires addressing existing challenges and future research gaps.

Artificial Intelligence (AI) is redefining the boundaries of web development by automating complex tasks, enhancing user experience, and reshaping how developers design and deploy web applications. This article examines the convergence of AI technologies and web development practices, with a focus on the transformative role of Large Language Models (LLMs), user interface generation, automated software engineering, and humancentered prototyping approaches. By referencing recent scientific and peer-reviewed studies, the paper presents original observations about AI's impact on software engineering, analyzes common issues and ethical concerns in using AI, and considers why end-users and smart devices are changing their methods of interaction. There is a particular focus on LLMs being able to help generate websites, the use of a dual-hemisphere approach to UI planning, and the effects of AI on making development projects faster and better. Critical analysis in this study shows that AI is involved in shaping the future of web development, more than just supporting it.

The purpose of recommender systems (RS) is to facilitate user collaboration and communication on the platform. Nevertheless, there is limited knowledge regarding the extent of this relationship and the techniques by which RS could promote persistent user engagement with the platform. In order to fill this void, the present study investigates the role of RS in transforming users' short-term engagement with the RS into long-lasting involvement with the platform. We present a theoretical framework by reviewing relevant literature in the domains of RS and user engagement to probe these issues. We provide open challenges in this field along with metrics in the present study.

Personalized learning seeks to improve educational outcomes by delivering content and instructional approaches tailored to individual learners' needs. Recent advancements in artificial intelligence, particularly the emergence of large language models (LLMs), have introduced new opportunities for generating personalized educational content. Despite this progress, it remains unclear which learnerspecific profile attributes most effectively inform LLM-generated responses. The answer may vary from contexts to contexts. In this paper, our study examines two personalized learning scenarios: term explanations and coding explanations. We investigate the impact of four types of personalization resources-learners' knowledge levels, hobbies, favorite subjects, and a combination of all three, on the perceived quality of LLM outputs. We conducted a user study where students evaluated and chose their preferred personalized responses. Based on their selections, we performed comparative and statistical analyses to assess the effectiveness of each personalization strategy. The results showed that students tended to favor personalized responses that were generated using their knowledge level as part of the user profile, highlighting the importance of this factor in producing effective personalized content. However, the patterns varied across different scenarios, and distinct individual differences emerged when users were grouped. CCS Concepts • Social and professional topics → Information technology education; • Information systems → Personalization.

Large Language Models (LLMs) are increasingly trained in elastic, multi-tenant cloud infrastructures that span data centers, regions, and heterogeneous accelerators. While distributed training has matured in scale and efficiency, its security posture lags behind adversarial realities: training corpora may contain sensitive or regulated data; gradient channels can leak membership and attribute information; supplychain subversion can inject malicious kernels or compromised containers; cross-tenant resource sharing elevates the risk of side-channel inference; and orchestration layers, which schedule, checkpoint, and autoscale jobs , are attractive targets for data exfiltration and model theft. Existing mitigations remain fragmented: transport encryption protects links, confidentialcompute enclaves protect limited code paths, differential privacy protects outputs in isolation, and secure aggregation schemes address narrow communication steps. What is missing is a coherent end-to-end framework that composes these mechanisms with provable guarantees while preserving the throughput and tail-latency characteristics required by trillion-parameter training. This paper presents AegisTrain, a secure distributed training framework that treats privacy and integrity as firstclass control objectives across the full training lifecycle . AegisTrain couples remotely attested confidential runtimes with attribute-bound key management to ensure that only measured and policy-compliant components can access plaintext data, gradients, and optimizer state. It introduces a cryptographic aggregation substrate that masks worker updates end-to-end and injects calibrated noise under a formally verifiable privacy accountant, rendering gradient channels useless for membership inference even when a bounded number of participants are compromised. Checkpoint images, telemetry, and intermediate artifacts are encrypted at rest with per-epoch keys derived from a hardware-rooted key hierarchy, and the supply chain is hardened with verifiable provenance, reproducible builds, and incluster policy enforcement. The framework is designed for tensor, pipeline, and data-parallel hybrids, and preserves scalability through streaming decryption, batched attestation, and offloaded cryptography that runs on CPU sidecars while GPU compute saturates the model step . We develop a control-plane that enforces risk-adaptive policies-for example, tightening privacy budgets or refusing mixed-trust colocation under elevated threat intel-without manual intervention, and we present machinecheckable invariants that forbid unsafe downgrades. A queueingtheoretic and information-theoretic analysis quantifies the overhead of encryption, attestation, and privacy noise relative to communication and compute, and shows parameter regimes in which security can be achieved with sub-5% throughput loss at cluster scale. A prototype implementation with masked allreduce, enclave-gated data loaders, and encrypted checkpoints demonstrates feasibility on realistic LLM training traces. By articulating the interfaces among attestation, secure aggregation, privacy accounting, and distributed parallelism, AegisTrain reframes secure LLM training as a problem of principled composition rather than ad hoc patchwork, yielding a deployable blueprint for cloud environments where both speed and trust are non-negotiable.

By 2050, the educational ecosystem will bear little resemblance to today's structured classrooms, rote memorization, and standardized assessments. As psychologist Howard Gardner articulated in a recent Harvard Graduate School of Education forum, artificial intelligence (AI) represents a millennium-scale disruption, potentially rendering many cognitive functions optional for humans. The exponential advancement of large language models (LLMs), robo-teachers, and 24/7 AI companions will democratize knowledge, personalize learning, and extend education beyond Earth. This article delineates the future educational needs and premises, drawing on expert predictions and emerging technologies. Key drivers include AI's role in offloading cognitive labor, the dissolution of traditional schooling, embedded neural interfaces for instant knowledge access, and the demands of multidisciplinary, interplanetary societies—potentially including Mars colonies. While optimistic visions promise empowerment, apocalyptic scenarios warn of skill atrophy, inequality, and ethical dilemmas.

Despite the potential benefits offered by GenAI technologies to provide innovative solutions to address distinct challenges faced by working adult learners (ALs) in higher education and beyond, there is limited understanding of how best to structure AI-powered learning for this population while ensuring their distinct needs and perspectives are considered. Hence, this study aimed to determine what curriculum and student-AI interaction would be required by situating ALs' views. Through analyzing 48 e-portfolios and in-depth interviews with 20 ALs from diverse educational and professional backgrounds, the study found that ALs perceived content mastery and developing a lifelong habit of learning as the optimal learning goals for AI-powered learning. AI-powered learning can be facilitated through personalized mastery-based learning and collaborative performance-based tasks, in tandem with scenario-based assessment, unobtrusive gamified assessment, and competency-based assessment. Along this line, AL articulated various necessary supports to foster AL-AI interactions. While AL identified metacognition and developing diverse and high-quality questions as crucial to support AL-AI cognitive interaction, they also highlighted that building ethical AL-AI relationships is important for enhancing AL-AI socio-emotional interaction. In addition, AL perceived immersive game-based platforms and interactive interfaces could serve as effective mediums for enhancing student-AI interactions. These findings can provide a more comprehensive understanding of AI-powered adult learning and implications for the design of educational AI, as well as instructional design to improve the educational experience for ALs.

The research is based on the importance of the agentic AI and generative AI that can change the current banking services. The entire analysis has been executed using secondary qualitative analysis using journals and other reports. A total of three themes are generated that are decision-making and automation using agentic AI, personalisation and fraud detection using generative AI, and integration issues, together with future opportunities. The results indicate that these two tools and hybrid human-AI decision-making are used to mitigate fraud-related activity in the banking sector. These findings cover gaps in the past research, which concentrated on isolated applications. The study finds that ethical adoption of agentic and generative AI will lead to innovation, build customer confidence, and deliver sustainable banking solutions.