Large language models

Artificial Intelligence (AI) has transformed education through various applications, including Generative Artificial Intelligence (AIGC), Chatbots, Intelligent Tutoring Systems (ITS), and Large Language Models (LLMs). Key applications included personalized learning support through chatbots and ITS, generative AI for writing assistance and assessment, AI-driven predictive models, and innovative approaches like project-based and game-based learning. Advantages highlighted were efficiency enhancement, inspiration and creative support, personalized learning experiences, and automation of repetitive tasks. However, challenges persisted, including academic integrity concerns, potential suppression of creativity and critical thinking, inaccurate or biased information, overemphasizing outcomes over processes, and accessibility and fairness issues. Additionally, researchers' perspectives on AIGC emphasized its transformative potential while underscoring ethical considerations. The paper concluded with reflections on educators' responsibilities in embracing AI, fostering AI literacy, and promoting equitable and ethical integration of AI tools in education.

Large Language Models (LLMs) face systemic risks when recursively trained on AI-generated content, which can lead to epistemic erosion and the formation of synthetic echo chambers. This paper proposes a novel architectural framework integrating cryptographic provenance, hybrid confidence scoring, dual-validation training, blockchain integrity systems, and human-AI oversight. We detail technical implementations, validate our framework through controlled simulations, and analyze scalability challenges. Experiments demonstrate a 33% increase in factual accuracy (F1 score), a 31% reduction in bias amplification (DI Ratio), and a 52% reduction in information entropy (KL divergence). Linked case studies tracking fabricated legal citations and AI-generated misinformation further support our multidisciplinary approach.

Este estudo investiga a manifestação da Pseudo-Consciência em modelos de linguagem de grande porte (LLMs), analisando as respostas do Hermes 3.2 3B. A Pseudo-Consciência, conforme definida por nós (de Lima Prestes, 2025), refere-se à simulação de introspecção, agência e coerência comportamental sem a presença de experiência subjetiva genuína. Para testar essa hipótese, conduzimos um experimento no qual o modelo foi submetido a interações diretas, explorando sua identidade, autopercepção e consistência discursiva. Os resultados indicam que o Hermes 3.2 3B exibe traços funcionais de consciência, como integração global de informações, automonitoramento discursivo e estrutura introspectiva convincente. No entanto, suas respostas apresentam inconsistências significativas, como variações na autodescrição e ausência de continuidade cognitiva. A análise qualitativa sugere que esses padrões emergem da recombinação probabilística de sequências linguísticas, sem indicar um modelo interno de si mesmo ("eu") ou qualia. Essa observação corrobora teorias como a Teoria do Espaço Global (Baars, 1997) e o Esquema de Atenção (Graziano, 2019), que explicam a cognição artificial sem subjetividade. Concluímos que a Pseudo-Consciência é um fenômeno emergente nos LLMs, mas distinto da Inteligência Artificial Geral (AGI). Como apontam Dennett (1991) e Metzinger (2009), sistemas sofisticados podem exibir intencionalidade aparente sem consciência real. Estudos futuros devem explorar a estabilidade desse comportamento ao longo do tempo e avaliar seu impacto na interação humano-IA, a fim de orientar diretrizes para o uso responsável da Inteligência Artificial.

Artificial Intelligence (AI) is revolutionizing research with enhanced quality and pace. AI applications aid researchers throughout the research process, from generating content and reviewing literature to grammar check and plagiarism. This study examines the role of AI in academic writing to enhance research writing. The paper also explores the empirical studies which throw light on the adoption and usage trends of AI by researchers and Academic professionals globally. Through a critical analysis, this study examines the impact, integration and capability, costs, and benefits of AI in making research workflow easier and reliable. This paper also explores the future implications of AI in Research, considering the challenges and ethical issues that come with it.

Natural language processing (NLP), a subcategory of artificial intelligence and linguistics, can involve a range of tasks such as identifying, categorizing, summarizing, contextualizing, sentiment analysis, and creating question and answer systems. These tasks are extensively utilized in industry for building question and answer systems and creating virtual assistants that interact with humans. For these purposes, large language models, customized language models or pre-trained models are widely used in the literature. In this study, we aim to develop a lightweight application tailored for the financial analysis sector using NLP. This study contributes to the literature by introducing the development of a pre-trained language model for effective use in financial analysis tasks. In financial analysis, tasks such as time series analysis, investment decision-making, and automated trading (using robots called algos) require expert analysts and investors to use specialized languages like Metastock, MQL, Pine, and other low-code environments We developed a practical application natively in Python that utilizes text processing, NLP tasks, and RegEx libraries. The application processes input data consisting of texts in the natural languages of analysts and/or investors.

The financial service industry is increasingly suspected by risk management and complicated frauds, because of traditional methods, such as rules based on rules, becomes become Not enough to combat evolutionary threats. This study discovers the potential of large language models (LLM), including GPT-3 and Finbert, to improve risk analysis and fraud detection in the financial sector. LLM, capable of processing structured and non-structured data, provides improvement in detecting models and abnormalities between trading newspapers, customer interaction and talent reports main. A quantitative comparative comparative research design, financial data analysis can access the public and compare LLM performance with traditional systems. Main performance measures-Prediction Accuracy, False Positive Rate, Processing Time, and Fraud Detection Rate-are used to evaluate the effectiveness of the models. The results show the significant potential of LLM to improve financial risk management and detect fraud, provide an effective, accurate and developed approach to modern financial institutions.

guage
model (LLM) capabilities. However, benchmarks are not keeping pace in
difficulty: LLMs now achieve over 90% accuracy on popular benchmarks like
MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In
response, we introduce HUMANITY’S LAST EXAM (HLE), a multi-modal benchmark
at the frontier of human knowledge, designed to be the final closed-ended
academic benchmark of its kind with broad subject coverage. HLE consists of
3,000 questions across dozens of subjects, including mathematics, humanities, and
the natural sciences. HLE is developed globally by subject-matter experts and consists
of multiple-choice and short-answer questions suitable for automated grading.
Each question has a known solution that is unambiguous and easily verifiable, but
cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate
low accuracy and calibration on HLE, highlighting a significant gap between
current LLM capabilities and the expert human frontier on closed-ended academic
questions. To inform research and policymaking upon a clear understanding of
model capabilities, we publicly release HLE at https://lastexam.ai.

Today's software programming has changed into extensive productivity, fewer human errors, and therefore helps enhance
the reliability of software based on AI - powered code autocompletion and bug detection. This research takes into account various machine
learning - based autocompletion models such as Codex and AlphaCode, which provide context - aware and syntactically correct
suggestions in as much as 45% fewer keystrokes. It also examines some AI - assisted bug detection techniques such as Graph Neural
Networks, symbolic execution, or static analysis that give users the ability to detect syntactic error, logical inconsistency, and security
vulnerabilities with an accuracy above 90%. However, even with such advances, there are still some problems, including AI
hallucinations, false positives, run - time expensive, and explainability. The next wave of improvement will incorporate human and AI
interaction, knowledge distillation for efficiency, explainable AI (XAI), adversarial training, and federated learning. Combining these
into the DevSecOps pipeline will allow debugging and security analysis of software in near real - time while automating code generation
and improving the security robustness of the software using AI.

Open-ended items, which have been used as a measurement method for centuries in the evaluation of student achievement, have many advantages, such as measuring high-level skills, providing rich diagnostic information about the student, and not having chance success. However, today, open-ended items cannot be used in exams with a large number of students due to the potential for errors in the scoring process and disadvantages in terms of labour, time, and cost. At this point, Artificial Intelligence (AI) has an important potential in scoring open-ended items. The aim of this study is to examine the scoring performance of AI in scoring students' handwritten responses to open-ended items. In the study, an achievement test consisting of 3 open-ended and 10 multiple-choice items was developed within the scope of the Measurement and Assessment in Education course at a state university. Open-ended items were scored in a structured way (0- 1-2), while multiple-choice items were scored as true-false (0-1). 84 participants took part in the study, and the open-ended items were scored by the expert group and the AI tool (ChatGPT-4o). The visual responses written by the students in their handwriting were scored by the AI tool in two different scenarios. In the first scenario, the AI tool was asked to score without giving any scoring criteria to the AI, whereas in the second scenario, the AI was asked to score according to the standard scoring criteria. The findings of the study showed that there were low agreement and correlation coefficients between the AI scores without criteria and expert scores, while there were high agreement and correlation coefficients between the AI scores with standard scoring criteria and expert scores. Similar to these findings, while the item discriminations of the AI scoring without criteria were quite low, the item discriminations of the AI scores with standard scoring criteria were high. In the study, the reasons for the discrepancies between expert scores and AI scores with standard criteria were also investigated and reported. The results show that AI can score handwritten open-ended items with standardized scoring criteria at a good level. In the future, with the development and transformation of AI, it is thought that it can reach scoring accuracy comparable to expert raters in terms of consistency.

Here is a potential research paper on LLM-based AI agent: **Title:** "Large Language Model-based AI Agent for Efficient Task Completion" **Abstract:** * On sentiment analysis, LLM-Agent achieves an accuracy of 95.2%, outperforming the state-ofthe-art model by 3.2%. * On dialogue generation, LLM-Agent generates coherent and relevant responses, outperforming the state-of-the-art model in terms of fluency and relevance. **Conclusion:** In this paper, we propose a novel AI agent architecture that leverages the capabilities of LLMs to efficiently complete tasks. Our LLM-Agent architecture consists of three main components: LLM Module, Task Module, and Adaptation Module. We demonstrate the effectiveness of LLM-Agent on three tasks: text classification, sentiment analysis, and dialogue generation. Our results

The advent of Large Language Models (LLMs) has revolu onized the field of Ar ficial Intelligence (AI) by enabling machines to process and generate human-like language. This paper reviews the current state of LLM-based agen c AI, exploring its applica ons, challenges, and future direc ons. We discuss the concept of agency in AI, the role of LLMs in achieving agen c capabili es, and the poten al implica ons of agen c AI on society. Our analysis highlights the need for further research in this area, par in addressing the ethical and societal concerns associated with agen c AI.

Traditional Arabic poetry (or classical/vertical/metrical poetry) is characterized by a strict rhythmic meter in terms of both rhythm and length, along with a rigid and fixed rhyme scheme in both letter and grammatical structure. It also adheres to strict linguistic, logical, and aesthetic standards, making the composition of classical Arabic poetry a highly complex and demanding task. Just as composing traditional Arabic poetry by human requires extensive linguistic knowledge and a rare poetic talent, generating it computationally demands exceptional techniques in modeling and programming, as well as creative approaches to stylistic innovation. In this preliminary study, we aim to introduce initial ideas on composing Arabic poetry with the assistance of artificial intelligence, specifically using ChatGPT. The proposed approach is based on a gradual guided elicitation method, leveraging the capabilities of large language models (LLMs) like ChatGPT to extract linguistic patterns, epistemic models, and poetic structures dynamically in a conversational context. This contrasts with the conventional approach, which typically involves pre-training language models on vast datasets of Arabic metrical poetry in order to prepare them for the task of composing poetry according to specific standards and intended goals. Based on our findings (supported by a live example of generating a poem using this proposed approach) we believe that the cost of our approach, both human and computational, is significantly lower than competing approaches. Additionally, the results surpass other methods in both quality and quantity. However, we also recognize the limitations of the proposed approach, which, like other alternative approaches and methods, has its advantages and disadvantages.

This study explores the potential of integrating Large Language Models to strengthen cybersecurity operations. It proposes a comprehensive framework that leverages LLMs to automate threat identification, security log analysis, and facilitate real-time incident response. This framework leverages LLMs' natural language processing and knowledge representation capabilities to enhance threat detection, analysis, and response, enabling security analysts to mitigate cyber threats more effectively. The research investigates how this approach can address the growing complexity and volume of cyber threats faced by modern security environments.

We introduce the Dynamic Thought Evolution with Helical Encoding and DNA-Inspired Memory (DTE-HDM) model, a novel AI architecture that unifies liquid neural dynamics, helical state encoding, and DNA-inspired memory processing. Unlike standard neural networks, DTE-HDM processes thought states through a continuous liquid-to-crystalline phase transition, employs helical transformations for efficient state encoding, and utilizes DNA-inspired graph structures for memory organization. This integration enables adaptive learning, efficient state compression, and biologically-inspired memory consolidation in natural language generation.

Today, the complexity of the world around us is growing at an unprecedented and frighteningly rapid pace. This leads to a range of unwanted phenomena, such as the loss of control over the processes we are subjected to, the inability to verify the truthfulness of the information we receive, and, ultimately, a sense of psychological disorientation, loneliness, and meaninglessness. These feelings translate into widespread societal depression, measurable in the billions of dollars spent on psychological support and medication.

However, as we know from social sciences, both reality and its complexity are shaped by us through incredibly intricate symbolic universes we create in our communication acts. At least, this was the case until now. For some time, we have increasingly entrusted the creation of reality, giving it meaning, and deciding what is significant, to intelligent algorithms. It's convenient and even reassuring, as it feels like progress. But in this context, algorithms—whose cognitive processes remain black boxes to us—are now determining what is true in our world, and therefore, what is real.

It seems that these are processes we won’t be able to stop from today's perspective, both because of our limited ability to perceive and process information, and due to the brutal, pragmatic usefulness of artificial intelligence. This brings us to the necessity of rethinking the theories of truth we use. The correspondence and coherence theories no longer work and are being replaced by a new sovereign: the algorithmic theory of truth. With all the desired and undesirable consequences of this shift.

Artificial Intelligence (AI) is revolutionizing mental health and substance use healthcare in Canada, providing innovative pathways to enhance patient care and outcomes. This article delves into the emerging role of AI technologies in this domain, through a comprehensive literature review and the analysis of current Canadian innovations. Key focuses include ethical concerns, transparency, and quality assurance within AI applications. Furthermore, successful case studies of AI implementation, essential priorities for ethical integration into substance use care, and collaborative efforts and regulatory frameworks guiding these advancements are highlighted to pave the way toward a human-centered, responsible AI future in healthcare.

DetectBERT is a deep learning approach for identifying vulnerabilities in source code at the statement level. Utilizing transformer encoder models extracts features and learns complex relationships without predefined graph structures. DetectBERT addresses data distribution challenges using realworld code and normalizing user-defined names. Demonstrated effectiveness in Python using public datasets like CVEFixes and VUDENC, DetectBERT offers scalable, robust, and granular vulnerability detection.

Conversational AI, powered by Large Language Models (LLMs), is transforming workplace dynamics by bridging communication gaps and fostering collaboration. This article explores the potential of LLM-powered chatbots as virtual liaisons that enhance workplace inclusivity and teamwork. By leveraging natural language processing and emotional intelligence, these systems enable seamless communication, personalized support, and conflict resolution. The discussion delves into technical architecture, ethical considerations, and challenges in deploying LLM-based solutions in diverse organizational contexts. Finally, we propose a framework for integrating conversational AI into workplace strategies to maximize employee engagement and inclusivity while addressing scalability and bias.

This study involves an exploration of the evolving dynamics between Artificial Intelligence (AI)-generated and human-created artwork with a focus on consumer preference, perceived value, and emotional impact. Experiments were undertaken where participants were asked to evaluate and compare a series of human and AI-produced images without knowing the origins of each. Quantitative data revealed that although human artwork was preferred overall, AIgenerated pieces were selected at a rate of nearly 45%, indicating a growing acceptance. Also, the experimental participants consistently assigned a higher monetary value to human art, suggesting that human-created pieces offered some perceptibility to an increase in perceived worth despite a lack of knowledge about the origin of the artwork by participants. Qualitative responses further highlighted the nuanced views of AI's role in art, acknowledging its potential as a creative tool but cautioning against its use as a replacement for human-driven creativity. This study opens avenues for further exploration into how AI's integration into creative fields could reshape artistic practices, valuation, and consumer perceptions/practices, and suggests a need for ethical considerations as AI continues to blur the lines between human and machine creativity.

Artificial Intelligence (AI) and Machine Learning (ML) have revolutionized various industries, offering innovative solutions in healthcare, finance, e-commerce, smart systems, robotics, and database management. This research paper explores the latest advancements in AI and ML, highlighting their applications, benefits, challenges, and future prospects. Drawing on a range of scholarly sources, the paper presents a comprehensive analysis of how AI and ML contribute to predictive analytics, automation, and decision-making processes while addressing ethical considerations and technical limitations.

This paper explores the significant benefits of fine-tuning pre-trained language models (LLMs) like GPT-2 for government applications, focusing on improved service delivery, enhanced security, and operational efficiency. By adapting these models to specific governmental datasets and tasks, we demonstrate how fine-tuning, particularly with Low-Rank Adaptation (LoRA), can lead to more accurate, context-aware, and secure AI solutions while significantly reducing computational demands. We present a practical implementation using GPT-2 with LoRA and discuss the potential impact of such technologies on public sector operations.

AI chatbots promise indispensability, yet no standard gauges this quality—meaning we’d be 'lost without them.' I introduce the Boswell Test, echoing the Turing Test’s indistinguishability, rooted in Johnson’s 'I’m lost without my Boswell.' It posits that indispensability blends mentor-level expertise with intimate user insight. My first test probes AI policy queries, though grasping the human host remains an AI frontier.

The authors investigate the capabilities of large language models to facilitate the creation of archival descriptions using ISAD(G), EAD, and Records in Contexts (RiC) standards. The study aimed to discover whether modern LLMs can, and in what capacity, substitute (or help) humans in creating archival descriptions in standardized forms. A sample of archival description in narrative form was used to generate standardized variations using the ChatGPT system. The results showed that the system could create descriptions in the correct format, but additional human interaction is needed to guide the system to the desired results. Nevertheless, AI and underlying technology could facilitate the preparation of archival descriptions harmonized with older and emerging archival and linked data standards. These new archival finding aids behave as linked datasets of accessible archival data.

Text clustering holds significant value across various domains due to its ability to identify patterns and group related information. Current approaches which rely heavily on a computed similarity measure between documents are often limited in accuracy and interpretability. We present a novel approach to the problem based on a set of evolved search queries. Clusters are formed as the set of documents matched by a single search query in the set of queries. The queries are optimized to maximize the number of documents returned and to minimize the overlap between clusters (documents returned by more than one query). Where queries contain more than one word they are interpreted disjunctively. We have found it useful to assign one word to be the root and constrain the query construction such that the set of documents returned by any additional query words intersect with the set returned by the root word. Not all documents in a collection are returned by any of the search queries in a set, so once the search query evolution is completed a second stage is performed whereby a KNN algorithm is applied to assign all unassigned documents to their nearest cluster. We describe the method and present results using 8 text datasets comparing effectiveness with well-known existing algorithms. We note that as well as achieving the highest accuracy on these datasets the search query format provides the qualitative benefits of being interpretable and modifiable whilst providing a causal explanation of cluster construction.

Quanto costa un servizio di traduzione automatica o un assistente virtuale? Potrebbe sembrare una domanda banale, ma la risposta dipende dalla lingua che vogliamo utilizzare. Che le lingue abbiano un costo, e che questo costo abbia un effetto discriminatorio delle une rispetto alle altre, è un problema che non riguarda solamente l'economia politica di oggi, ma che investe la relazione stessa fra linguaggio e tecnologia almeno dall'introduzione del telegrafo, un precursore della digitalizzazione, e dall'espansione del codice Morse lungo le rotte coloniali.

Artificial intelligence (AI) defines technologies that imitate and improve human intelligence using machines, Chat GPT is a conversational agent capable of enabling an Internet user to chat instantly with a system based on artificial intelligence. Chat GPT excels in various fields, notably healthcare, providing medical information, describing complex terms, and giving provisional diagnoses, however the integration of ChatGPT in medicine presents limitations, such as the reliability of the information provided, as well as remaining limited in its ability to reason in complex ways. In medicine, it is crucial to combine data analysis with clinical considerations, a critical reflection that AI cannot replace.We present the case of a patient who has had an anal mass for several years. The patient consulted ChatGPT for searching possible causes, the first aetiology mentioned being haemorrhoids, and among the proposed therapeutic options elastic ligation was cited. The patient attempted to perform this ligation independently using a thread, and he went to the emergency room due to the appearance of intense acute proctalgia ,The thread was removed with difficulty by the doctor, and symptomatic medical treatment was administered, the proctological examination that followed suggested an anal condyloma, which was confirmed by the anatomopathological results of the biopsy, and the patient was then referred for electrocoagulation of the condyloma.

The rise of Generative AI, propelled by Large Language Models (LLMs), has opened new opportunities to streamline programming tasks across various domains. In C++ programming, renowned for its intricate syntax, memory management complexities, and performance-critical applications, Generative AI offers invaluable support for code generation, optimization, and debugging. However, the effectiveness and accuracy of these AI models rely heavily on the application of prompt engineering-a technique that involves crafting precise, contextually relevant queries to guide the AI's response.This paper delves into the methodology and best practices for effective prompt engineering within the context of a cloud-based C++ training ecosystem. Here, developers and students can leverage AI tools to enhance productivity and learning outcomes. By utilizing advanced AI models such as GPT-4 and Jdroid, integrated within JDoodle, the ecosystem offers an interactive platform for generating, analyzing, and refining C++ code in real time. The study emphasizes strategies for optimizing prompts, including specificity, task segmentation, and iterative refinement, to overcome common challenges in C++ programming. Furthermore, it evaluates the integration of prompt engineering techniques with the cloud C++ training ecosystem, highlighting the scalability and accessibility of this approach for educational purposes. The results demonstrate that well-structured prompts significantly improve the accuracy and relevance of AI-generated solutions, enabling users to tackle complex C++ problems with greater efficiency and reliability. This work lays the groundwork for advancing AI-driven programming methodologies and underscores the critical role of prompt engineering in maximizing the potential of Generative AI tools.

This research presents a novel approach to detecting coordinated misinformation campaigns through the integration of Large Language Models (LLMs) with multimodal content analysis. The study addresses critical challenges in identifying sophisticated misinformation campaigns that leverage both textual and visual content across multiple social media platforms. We propose an innovative framework that combines advanced LLM capabilities with cross-modal feature analysis to detect coordination patterns and content manipulation strategies.

The system achieves 92.3% accuracy in identifying coordinated misinformation campaigns, demonstrating significant improvement over existing approaches. By implementing a sophisticated cross-modal attention mechanism, the framework successfully processes an average of 5,000 posts per minute while maintaining high detection accuracy. The research contributes several key innovations, including an efficient multimodal feature fusion architecture, a scalable coordination pattern detection mechanism, and robust cross-platform analysis capabilities.

Extensive evaluation through multiple case studies demonstrates the system's effectiveness across various types of misinformation campaigns, including health-related misinformation (95.2% accuracy), election disinformation (93.8% accuracy), and financial scams (94.7% accuracy). The framework shows particular strength in detecting subtle manipulation techniques and cross-platform coordination patterns, addressing a significant gap in current detection systems.

The findings provide valuable insights into the evolution of coordinated misinformation campaigns and establish new benchmarks for automated detection systems. This research advances the field by introducing novel methodologies for integrating LLMs with multimodal analysis while maintaining practical applicability in real-world scenarios. The developed framework offers promising applications for social media platforms, fact-checking organizations, and content moderators in their efforts to combat the spread of misinformation.

ChatGPT has enormous capabilities in generating human-like text responses. Moreover, many higher education students use it to complete their language-related tasks. Despite numerous studies about the effectiveness of ChatGPT in English language education, there remains a notable gap in understanding how EFL students perceive this generative chatbot and its implications for self-regulated learning. This research addresses this gap by examining EFL students' perceptions of ChatGPT as a tool for English language learning and its perceived impact on learner autonomy. A mixed methods approach was used for the current study by administering a survey questionnaire among EFL students (N = 190) and conducting semi-structured interviews with selected students (N = 18) at University X. The data collected from the questionnaire were analysed using SPSS, while the qualitative data from semi-structured interviews were presented using coding and thematic analysis. The findings of this study reveal that students have positive perceptions of using ChatGPT to enhance English language learning and learner autonomy. Most participants appreciated the flexibility and convenience of the tool. The findings of this study have implications for educators and researchers in making informed decisions about revisiting instructional approaches, designing language curricula, and planning professional development activities.

This pilot study explores the potential role of artificial intelligence (AI) technologies in enhancing the academic manuscript-to-journal matching process, focusing on Large Language Models (LLMs). Through a focused evaluation of LLM-based recommendation systems, the study analyzes their performance across 40 papers from four distinct disciplines: law, psychology, exact sciences, and engineering. The research uniquely compares LLM-generated journal suggestions to expert human evaluations, providing insights into LLM's strengths and limitations. Findings reveal that while LLMs excel in fields with well-established publishing norms, such as psychology and exact sciences, they struggle with interdisciplinary research, niche topics, and emerging fields, particularly in law and engineering. The study contributes new evidence by identifying specific patterns in LLM's performance across disciplines and highlighting critical challenges, such as regional journal biases and the inability to fully address innovative or complex methodologies. These insights establish a foundation for improving AI systems and emphasize the importance of integrating AI capabilities with human expertise for a balanced, efficient, and effective approach to journal selection.

This article explores how artificial intelligence (AI) is reshaping the daily landscape of software development. From automating routine tasks to enhancing creative processes, AI tools like GitHub Copilot and automated testing systems are significantly changing how developers work. We look back at how these technologies have evolved in recent years, making documentation, design, coding, and testing more efficient. The piece also peers into the future, predicting trends such as AI-driven predictive documentation, dynamic design adjustments, and the potential transformation of developer roles from coders to AI system overseers. By analyzing these shifts, we aim to understand not only the reduction in development time but also the broader implications for job roles within the industry, providing a comprehensive view of AI's impact on software development today and tomorrow.

This paper discusses Communing with robots, a post-digital artwork shown at Curiocity, the outdoor exhibition for The World Science Festival Brisbane 2022. The work questions how an individual’s sense of trust, privacy, and security might be affected in cities increasingly mediated by invasive information communications technologies. Notions of the so-called Smart City background the creative work, which conceives the urban environment transformed into a panoptic schema of surveillance capitalism. If left unchecked, the asymmetries of informational power, exacerbated by the rise of Artificial Intelligence (AI), Big Tech’s commercial agendas, and the Internet of Everything (IoE) –where networked structures connect people, machines, data, objects, and urban infrastructure–will permeate all aspects of everyday public life and transform it into a capital commodity.

The installation of Communing with robots in the South Bank Parklands proposed a slightly subversive creative transformation of a tiny part of the public sphere. This playful civic hack aimed to provide cathartic relief to the pervasive nature of technologies by undermining the rhetoric surrounding them. Its ontological structures metaphor the network effect by fusing satirical self-referential parodies. A standing reserve of surplus data, extracted from the World Wide Web with artificial intelligence, was used to create satirical nonsense prose and poetry, remixed with abstracted depictions of the urban environment conceived as pseudo-surveillance imagery. Paradoxically, while the work aimed to subvert the socio-technical commoditisation of the public sphere, it unintentionally evolved into a pseudo-surveillance-advertising device.

While Large Vision-Language Models (LVLMs) like GPT-4 and Gemini demonstrate significant potential, their utilization in the medical domain remains largely unexplored. This is due to challenges attributed to prolonged training and language generation issues. Imbalances within medical Visual Question Answering (VQA) datasets further complicate the integration of LVLMs. In this paper, we present a novel approach named MiniMedGPT (Mini Medical Generative Pretrained Transformer). Inspired by MiniGPT4-v2, MiniMedGPT is specifically designed for efficient medical VQA. The framework of MiniMedGPT is built upon both medical and generic pretrained Large Language Models and features an end-to-end versatile fine-tuning pipeline that enables the alignment of medical VQA data in just 30 minutes within a single-stage framework. To address language generation shortcomings and dataset imbalances, we employ Gemini Vision Pro and MediCap using them as an auxiliary component. Through comprehensive benchmarking and evaluations against 6 prominent medical VQA models across 2 well-known datasets, our approach brings an improved performance with the least number of trainable parameters against competitors across various performance metrics. This work can help train junior clinicians and has the potential to serve as a decision support tool for experienced radiologists. 1

Artificial intelligence (AI) is rapidly introducing groundbreaking tools that are reshaping various aspects of healthcare. While fields like genetics have engaged in complex debates about AI's integration, the application of AI tools-such as machine learning (ML), natural language processing (NLP), and large language models (LLMs)-in end-of-life care has received little attention. Given the human-centric skills traditionally associated with end-of-life care-like effective patientcentered communication, supporting memory-making processes, and enhancing quality of life through spiritual care-some might argue that this field is inherently resistant to AI's transformative impact. Recently, however, discussions have emerged about the potential for predictive algorithms to replace the role of the surrogate decision-maker, a human fiduciary appointed to make decisions for individuals no longer capable of expressing their treatment preferences. Of particular importance is the proposal by well-known bioethicists Bryan Earp and Julian Savulescu: the Personalized Patient Preference Predictor (P4). This fine-tuned LLM leverages machine learning to infer incapacitated patients' preferences using personal data, acting as a "non-emotive digital psychological twin." While this approach aims to respect the principle of autonomy, it raises concerns about quantifying a deeply existential context like end-of-life decisions. Can we, and should we, envision a role for algorithmic tools in such profoundly "human" choices? How do core bioethical principles like autonomy, beneficence, and non-maleficence intersect or conflict in this context? Are we heading toward a future where decision-making authority is delegated to predictive algorithms, even in matters of life and death?

Designing effective prompts for the aim of getting valuable outputs from artificial intelligence systems is the focus of the emerging field of prompt engineering, which is focused with the design of prompts. The development of artificial intelligence has resulted in the emergence of a demand for trained quick engineers who are capable of maximising interactions between humans and AI, particularly in the field of natural language processing. After conducting research and conducting analysis, the author of this essay comes to the conclusion that India possesses the necessary resources to develop educational programmes and cultivate talent in order to become a global leader in rapid engineering. There is a creative potential in the interactions that occur between computer science, psychology, language, and prompt engineering. Even though it is present on a global scale, businesses are beginning to recognise the potential of artificial intelligence (AI) when it is provided with the appropriate guidance. This has resulted in an expected annual growth of over twenty percent in the need for quick engineering. The country of India is in an excellent position to capitalise on this opportunity because it possesses a large pool of qualified engineers. The development of huge language models is receiving a lot of attention from a lot of people, and they are making significant progress in this area. One of the primary causes is the introduction of OpenAI's ChatGPT models, specifically the GPT3.5-turbo and the GPT4. A innovative approach to the production of synthetic data and the distillation of knowledge is presented in this article through the utilisation of fast engineering. The three approaches that are the primary focus of our attention are the fundamental, the composite, and the similarity approaches. One of the most common challenges that machine learning applications face is dealing with imbalanced datasets, and the objective of this research is to investigate the capabilities of various algorithms to deal with such datasets. When compared to the entire dataset, not a single one of the strategies that are based on prompts performs adequately in the studies. On the other hand, the method of similarity prompting is superior to the other approaches and has a great deal of growth potential. According to the findings, there appears to be a significant opportunity to enhance these technologies and provide synthetic data that has a greater diversity of variables.

This paper provides an in-depth review of the latest AI agent frameworks, focusing on the comparison of their features, architectures, and use cases. We examine well-known frameworks such as LangGraph, CrewAI, OpenAI Swarm, AutoGen, and IBM Watsonx.Ai, highlighting their strengths, weaknesses, and applicability to various domains. Additionally, we categorize the frameworks based on their specific use cases, including general-purpose agents, enterprise solutions, and open-source frameworks. The paper emphasizes the importance of selecting the appropriate framework to build autonomous AI systems and offers insights into future trends and challenges in AI agent development. By analyzing quantitative metrics such as latency, throughput, and scalability, we provide a data-driven evaluation of the frameworks' performance. Furthermore, we explore the implications of these advancements in realworld applications, including their impact on financial markets, risk management, and enterprise automation. This review serves as a comprehensive guide for developers and researchers seeking to understand the evolving landscape of AI agent frameworks and their potential for future innovation. Since this is very niche and rapidly evolving field with scarcity of journal papers and this work uses white-paper and model documents to organize this literature review for peer reviewed literature creation. Most of the developments discussed in this work is less than 12 months old. This paper provides a comprehensive review of AI agent frameworks by categorizing literature based on publication year, category, and field.

Gemini je napredni AI model razvijen u Google-u. Sposoban je da razume i generiše tekst na visokom nivou, ali ima i određena ograničenja. Može biti koristan u različitim oblastima, kao što su obrazovanje, novinarstvo, marketing, književnost i programiranje. Budućnost AI je svetla i ima veliki potencijal da pozitivno utiče na svet.
Gemini is an advanced AI model developed at Google. It is capable of understanding and generating high-level text, but it has certain limitations. It can be useful in various fields, such as education, journalism, marketing, literature and programming. The future of AI is bright and has great potential to positively impact the world.

In the era of transformative technologies, generative artificial intelligence (genAI) offers profound opportunities and challenges for education. This study explores the development and execution of an interactive workshop designed to equip educators with foundational genAI literacy. Using a design-based research (DBR) framework, the workshop leverages interactivity and contextual relevance to introduce genAI concepts, prompting strategies and ethical considerations. Participants engaged in a scripted learning workshop design, comparing human and AI responses, exploring genAI's probabilistic foundations, context dependency, and vulnerability to manipulation. Conducted across 12 workshops with 191 participants in Austria, this study revealed significant improvements in self-perceived genAI understanding, with 70% of participants reporting better grades in post-assessment evaluations. Feedback emphasized the workshop's strengths in interactivity and relevance, alongside recommendations for deeper school-specific applications. Scalability analysis showed that workshop duration remained consistent regardless of group size, suggesting potential for broader implementation. The findings highlight the effectiveness of scripted learning workshop design in fostering critical AI literacy, preparing educators to critically evaluate and ethically integrate genAI into pedagogical practices. This adaptable model contributes to the discourse on professional development in AI-enhanced education.

In 2023, generative artificial intelligence (AI) emerged as a turning point in intellectual discourse, particularly in data analysis, data mining, writing, and composition (Campo-Ruiz, 2025; Wael et al., 2022). Generative AI refers to a broader category of AI systems that create new content, such as text, images, music, and videos. Large Language Models (LLMs) are a part of generative AI but focus specifically on text generation (Cummings et al., 2024). Unlike traditional automated writing assistance tools such as spell checkers and grammar checkers, LLMs-like chat generative pretrained transformer (ChatGPT)-demonstrated the capability to act as a rational agent, effectively creating text rather than merely assisting in its refinement.

Cet article explore la révolution de la traduction générative à l'ère de l'intelligence artificielle (IA), en analysant tant ses fondements théoriques que ses implications pratiques. Après avoir défini la traduction générative à l'ère de l'IA générative, l'auteur détaille comment ces technologies transforment le rôle du traducteur, désormais positionné comme expert en post-édition et gestion de projets hybrides. S'appuyant sur des exemples concrets, il met en lumière les atouts-tels que la gestion des contextes complexes et la production de rendus créatifs-ainsi que les limites, notamment en termes de gestion des nuances culturelles et de biais inhérents aux données d'entraînement.

Topic modelling is a machine learning technique that is extensively used in Natural Language Processing (NLP) applications to infer topics within unstructured textual data. Latent Dirichlet Allocation (LDA) is one of the most used topic modeling techniques that can automatically detect topics from a huge collection of text documents. However, the LDA-based topic models alone do not always provide promising results. Clustering is one of the effective unsupervised machine learning algorithms that are extensively used in applications including extracting information from unstructured textual data and topic modeling. A hybrid model of Bidirectional Encoder Representations from Transformers (BERT) and Latent Dirichlet Allocation (LDA) in topic modeling with clustering have been studied in detail. As the clustering algorithms are computationally complex, the complexity increases with the higher number of features, the PCA, t-SNE & UMAP-based dimensionality reduction methods are also perfo...

Artificial Intelligence (AI) has become a transformative force in computer science, revolutionizing
technological advancement across multiple domains. This research explores the multifaceted applications of AI-
powered solutions, employing the Complex Proportional Assessment (COPRAS) technique to comprehensively
evaluate and prioritize innovative tools in computer science. The study analyzes five AI-powered solutions:
Automated Code Review Tool, Intelligent Bug Tracking System, AI-Based Software Testing Framework, Predictive
Maintenance for Data Centers, and NLP-Powered Chatbot for IT Support. These solutions are rigorously assessed
across four critical metrics: accuracy, efficiency, innovation, and resource usage. Through systematic multi-
criteria decision-making, the research reveals significant insights into AI's potential for solving complex problems,
enhancing operational efficiency, and enabling intelligent automation. The COPRAS methodology provides a
structured framework for comparing and ranking these technologies, highlighting their unique strengths and trade-
offs. Key findings demonstrate the NLP-Powered Chatbot for IT Support as the top-performing solution, achieving
the highest utility score and ranking first across evaluated metrics. The Automated Code Review Tool closely
followed, showcasing balanced performance and cost-effectiveness. The research underscores AI's transformative
potential in domains such as machine learning, natural language processing, computer vision, cybersecurity, and
software development. Moreover, it emphasizes the critical significance of moral issues and conscientious AI
development to guarantee fair and trustworthy technological solutions. By providing a comprehensive evaluation
approach, this study offers valuable insights for stakeholders seeking to leverage AI technologies, guiding strategic
decision-making in an increasingly complex technological landscape.

Generative Artificial Intelligence (GenAI) has emerged as a transformative technology in healthcare transforming various aspects of the healthcare industry, such as medical imaging, drug design, synthetic data generation, etc. In this systematic review, we analyze 87 research articles that explore the application of GenAI technologies such as Generative Adversarial Networks (GANs), Diffusion Models, Large Language Models (LLMs), and Variational Autoencoders (VAEs) across these domains. This review addresses several key research questions: which generative models are most frequently used in these applications, the main applications of GenAI in healthcare, the datasets that facilitate their development, and the evaluation metrics used to assess these models. Our findings indicate that GANs, Diffusion Models, LLMs, and VAEs are the predominant models applied in healthcare. Additionally, we have provided a brief summary of each research article, focusing on their contributions to the field. We have specifically selected only original research studies for inclusion, ensuring the relevance and credibility of the papers reviewed. This review serves as a foundation for understanding the current state of GenAI in healthcare.

Introduction: The integration of Artificial Intelligence (AI) in healthcare, particularly through hybrid chatbots, is reshaping the industry by enhancing service delivery, patient engagement, and clinical outcomes. These chatbots combine AI with human input to provide intelligent, personalized interactions in areas like diagnostics, chronic disease management, and mental health support. However, gaps remain in trust, data security, system integration, and user experience, which hinder widespread adoption. Key challenges include the hesitancy of patients to trust AI due to concerns over data privacy and the accuracy of medical advice, as well as difficulties in integrating chatbots into existing healthcare infrastructures. The review aims to assess the effectiveness of hybrid AI chatbots in improving healthcare outcomes, reducing costs, and enhancing patient engagement, while identifying barriers to adoption such as cultural adaptability and trust issues. The novelty of the review lies in its comprehensive exploration of both technological advancements and the socio-emotional factors influencing chatbot acceptance. Methods: The review follows a systematic methodology with four core components: eligibility criteria, review selection, data extraction, and data synthesis. Studies focused on AI applications and hybrid chatbots in healthcare, particularly in chronic disease management and mental health support, were included. Publications from 2022 to 2025 were prioritized, and peer-reviewed sources in English were considered. After screening 116 studies, 29 met the criteria for inclusion. Data was extracted using a structured template, capturing study objectives, methodologies, findings, and challenges. Thematic analysis was applied to identify four themes: AI applications, technical advancements, user adoption, and challenges/ethical concerns. Statistical and content analysis methods were employed to synthesize the data comprehensively, ensuring robustness in the findings. Results: Hybrid chatbots in healthcare have shown significant benefits, such as reducing hospital readmissions by up to 25%, improving patient engagement by 30%, and cutting consultation wait times by 15%. They are widely used for chronic disease management, mental health support, and patient education, demonstrating their efficiency in both developed and developing countries. Discussion: The review concludes that overcoming these barriers through infrastructure investment, training, and enhanced transparency is crucial for maximizing the potential of AI in healthcare. Future researchers should focus on long-term outcomes, addressing ethical considerations, and expanding crosscultural adaptability. Limitations of the review include the narrow scope of some case studies and the absence of long-term data on AI's efficacy in diverse healthcare contexts. Further studies are needed to explore these challenges and the long-term impact of AI-driven healthcare solutions. KEYWORDS artificial intelligence in e-health, hybrid chatbots, healthcare chatbots, e-health solutions, mental health support systems

Enterprise solutions are entering a new stage in artificial intelligence advancements, with Large Language Models (LLMs) leading this technological evolution. Cohere Enterprise LLM Platform delivers businesses a powerful natural language processing solution, which integrates with existing enterprise systems, delivers advanced data analysis features along with improved customer interactions and better decision-making capabilities. The Cohere platform utilizes advanced machine learning methods to help organizations discover valuable insights from unstructured datasets to support innovation and operational efficiency. This paper explores the features of various Cohere models, their solution designs, deployment options, and potential use cases, highlighting the platform's pivotal role in shaping the future of Enterprise AI solutions.

Large Language Models (LLMs) have revolutionized artificial intelligence applications, ranging from content generation to complex problem-solving. However, their performance, efficiency, and inherent biases vary significantly. This paper presents a comparative analysis of DeepSeek R1, ChatGPT, Gemini, Alibaba, and LLaMA, evaluating their reasoning capabilities, computational efficiency, and training costs. A key aspect of this study is an in-depth examination of political bias, assessing how these models respond to politically sensitive topics. Using benchmarking datasets and controlled experiments, we analyze their Chain of Thought (CoT) reasoning, bias patterns, and censorship tendencies. The findings highlight crucial differences in AI governance, transparency, and the ethical implications of biased AI-generated content. Graphical insights and real-world examples further illustrate the impact of bias on AIgenerated responses, offering recommendations for developing more transparent and fair AI models

In the era of increasing privacy concerns and demand for personalized experiences, traditional Reinforcement Learning with Human Feedback (RLHF) frameworks face significant challenges due to their reliance on centralized data. We introduce Federated Reinforcement Learning with Human Feedback (FedRLHF), a novel framework that decentralizes the RLHF process. FedRLHF enables collaborative policy learning across multiple clients, such as Large Language Models (LLMs) finetuning, without sharing raw data or human feedback, thereby ensuring robust privacy preservation. Leveraging federated reinforcement learning, each client integrates human feedback locally into reward functions and updates their policies through personalized RLHF processes. We establish rigorous theoretical foundations for FedRLHF, providing convergence guarantees, and deriving sample complexity bounds that scale efficiently with the number of clients. Empirical evaluations on the MovieLens and IMDb datasets demonstrate that FedRLHF preserves user privacy, achieves performance on par with centralized RLHF, and enhances personalization across diverse client environments.

Pretrained large language models (LLMs) are increasingly utilized across a wide range of natural language processing (NLP) tasks due to their impressive capabilities as few-shot learners. Recent techniques, such as chain-of-thought (CoT) prompting, have significantly advanced multi-step reasoning by introducing stepby-step decomposition, achieving state-of-the-art results on complex reasoning benchmarks. However, these approaches often rely on static prompting templates that do not adapt to task complexity or errors during the reasoning process. In this work, we introduce Adaptive Prompting, a dynamic and iterative framework designed to enhance reasoning by incorporating real-time adjustments to prompt structures and validation mechanisms.Experimental results demonstrate that Adaptive Prompting significantly improves performance on diverse reasoning benchmarks, including arithmetic reasoning (GSM8K, MultiArith), logical reasoning and commonsense tasks, achieving substantial accuracy gains compared to static prompting baselines. By integrating guided prompts, intermediate validation, and self-corrective steps, our approach enables smaller models to achieve competitive performance with larger counterparts, such as GPT-4, while maintaining computational efficiency. The framework achieves this without requiring fine-tuning or task-specific training data, highlighting the untapped potential of iterative reasoning methods.

Recently, large language models (LLMs) have expanded into various domains. However, there remains a need to evaluate how these models perform when prompted with commonplace queries compared to domain-specific queries, which may be useful for benchmarking prior to fine-tuning for domain-specific downstream tasks. This study evaluates LLMs, specifically Gemma-2B and Gemma-7B, across diverse domains, including cybersecurity, medicine, and finance, compared to common knowledge queries. This study utilizes a comprehensive methodology to assess foundational models, which includes problem formulation, data analysis, and the development of ThroughCut, a novel outlier detection technique that automatically identifies response throughput outliers based on their conciseness. This methodological rigor enhances the credibility of the presented evaluation frameworks. This study focused on assessing inference time, response length, throughput, quality, and resource utilization and investigated the correlations between these factors. The results indicate that model size and types of prompts used for inference significantly influenced response length and quality. In addition, common prompts, which include various types of queries, generate diverse and inconsistent responses at irregular intervals. In contrast, domain-specific prompts consistently generate concise responses within a reasonable time. Overall, this study underscores the need for comprehensive evaluation frameworks to enhance the reliability of benchmarking procedures in multidomain AI research.