Media, media education, GAI and radical uncertainty DOI

Petri Honkanen,

Mats Nylund

Media Education, Journal Year: 2024, Volume and Issue: 15(2), P. 7 - 20

Published: Dec. 30, 2024

The study examines the transformative potential impact of Generative AI (GAI) on society, media, and media education, focusing challenges opportunities these advancements bring. GAI technologies, particularly large language models (LLMs) like GPT-4, are revolutionizing content creation, platforms, interaction within landscape. This radical shift is generating both innovative educational methodologies in maintaining academic integrity quality learning. aims to provide a comprehensive understanding how impacts education by reshaping traditional practices media-related higher education. research delves into three main questions: nature as an innovation, its effect knowledge acquisition, implications for It introduces critical concepts such uncertainty, which refers unpredictable outcomes GAI, making forecasting planning challenging. paper utilizes McLuhan’s tetrad analyze GAI’s role questioning what it enhances or obsoletes, retrieves, reverses when pushed extremes. theoretical approach helps multifaceted influence Overall, underscores dual-edged where presents significant enhancements learning creation while simultaneously posing risks related misinformation, integrity, dilution human-centered practices. calls balanced integrating advocating preparedness against drawbacks leveraging capabilities revolutionize paradigms.

Language: Английский

Improving Training Dataset Balance with ChatGPT Prompt Engineering DOI Open Access
Mateusz Kochanek, Igor Cichecki, Oliwier Kaszyca

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(12), P. 2255 - 2255

Published: June 8, 2024

The rapid evolution of large language models, in particular OpenAI’s GPT-3.5-turbo and GPT-4, indicates a growing interest advanced computational methodologies. This paper proposes novel approach to synthetic data generation knowledge distillation through prompt engineering. potential models (LLMs) is used address the problem unbalanced training datasets for other machine learning models. not only common issue but also crucial determinant final model quality performance. Three prompting strategies have been considered: basic, composite, similarity prompts. Although initial results do match performance comprehensive datasets, prompts method exhibits considerable promise, thus outperforming methods. investigation our rebalancing methods opens pathways future research on leveraging continuously developed LLMs enhanced high-quality data. could an impact many large-scale engineering applications.

Language: Английский

Citations

6
Ethical dimensions of generative AI: a cross-domain analysis using machine learning structural topic modeling DOI
Hassnian Ali, Ahmet Faruk Aysan

International Journal of Ethics and Systems, Journal Year: 2024, Volume and Issue: unknown

Published: Sept. 3, 2024

Purpose The purpose of this study is to comprehensively examine the ethical implications surrounding generative artificial intelligence (AI). Design/methodology/approach Leveraging a novel methodological approach, curates corpus 364 documents from Scopus spanning 2022 2024. Using term frequency-inverse document frequency (TF-IDF) and structural topic modeling (STM), it quantitatively dissects thematic essence discourse in AI across diverse domains, including education, healthcare, businesses scientific research. Findings results reveal range concerns various sectors impacted by AI. In academia, primary focus on issues authenticity intellectual property, highlighting challenges AI-generated content maintaining academic integrity. healthcare sector, emphasis shifts medical decision-making patient privacy, reflecting about reliability security advice. also uncovers significant discussions educational financial settings, demonstrating broad impact societal professional practices. Research limitations/implications This provides foundation for crafting targeted guidelines regulations AI, informed systematic analysis using STM. It highlights need dynamic governance continual monitoring AI’s evolving landscape, offering model future research policymaking fields. Originality/value introduces unique combination TF-IDF STM analyze large corpus, new insights into multiple domains.

Language: Английский

Citations

5
Differences in User Perception of Artificial Intelligence-Driven Chatbots and Traditional Tools in Qualitative Data Analysis DOI Creative Commons
Boštjan Šumak, Maja Pušnik, Ines Kožuh

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(2), P. 631 - 631

Published: Jan. 10, 2025

Qualitative data analysis (QDA) tools are essential for extracting insights from complex datasets. This study investigates researchers’ perceptions of the usability, user experience (UX), mental workload, trust, task complexity, and emotional impact three tools: Taguette 1.4.1 (a traditional QDA tool), ChatGPT (GPT-4, December 2023 version), Gemini (formerly Google Bard, version). Participants (N = 85), Master’s students Faculty Electrical Engineering Computer Science with prior in UX evaluations familiarity AI-based chatbots, performed sentiment annotation tasks using these tools, enabling a comparative evaluation. The results show that AI were associated lower cognitive effort more positive responses compared to Taguette, which caused higher frustration especially during cognitively demanding tasks. Among achieved highest usability score (SUS 79.03) was rated positively engagement. Trust levels varied, preferred accuracy confidence. Despite differences, all consistently identifying qualitative patterns. These findings suggest AI-driven can enhance experiences while emphasizing need align tool selection specific preferences.

Language: Английский

Citations

0
Demonstration-Based and Attention-Enhanced Grid-Tagging Network for Mention Recognition DOI Creative Commons
Haitao Jia, Jing Huang, Zhao Kang

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(2), P. 261 - 261

Published: Jan. 5, 2024

Concepts empower cognitive intelligence. Extracting flat, nested, and discontinuous name entities concept mentions from natural language texts is significant for downstream tasks such as knowledge graphs. Among the algorithms that uniformly detect these types of concepts, Li et al. proposed a novel architecture by modeling unified mention recognition classification word–word relations, named W2NER, achieved state-of-the-art (SOTA) results in 2022. However, there still room improvement. This paper presents three improvements based on W2NER. We enhanced grid-tagging network demonstration learning tag attention feature extraction, so our modified model DTaE. Firstly, addressing issue insufficient semantic information short lack annotated data, inspired GPT-3, searched during training phase according to certain strategy enhance input features improve model’s ability few-shot learning. Secondly, tackle problem W2NER’s subpar accuracy multi-head mechanism employed capture scores different positions grid tagging. Then, tagging are embedded into model. Finally, retain about sequence position, rotary position embedding introduced ensure robustness. selected an authoritative Chinese dictionary adopted five-person annotation method annotate multiple concepts definitions. To validate effectiveness model, experiments were conducted public dataset CADEC dataset: dataset, with slight decrease recall rate, precision improved 2.78%, comprehensive metric F1 increased 0.89%; 2.97%, rate 2.35%, 2.66%.

Language: Английский

Citations

1
Understanding Privacy Concerns in ChatGPT: A Data-Driven Approach with LDA Topic Modeling DOI Creative Commons
Shahad Alkamli, Reham Alabduljabbar

Heliyon, Journal Year: 2024, Volume and Issue: 10(20), P. e39087 - e39087

Published: Oct. 1, 2024

Language: Английский

Citations

1
Enhancing risk and crisis communication with computational methods: A systematic literature review DOI Creative Commons

Madison H. Munro,

Ross Gore, Christopher J. Lynch

et al.

Risk Analysis, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 15, 2024

Abstract Recent developments in risk and crisis communication (RCC) research combine social science theory data tools to construct effective messages efficiently. However, current systematic literature reviews (SLRs) on RCC primarily focus computationally assessing message efficacy as opposed efficiency. We conduct an SLR highlight any computational methods that improve construction found most focuses using theoretical frameworks analyze or classify elements efficacy. For improving efficiency, manual are only used classification. Specifying the is sparse. recommend future apply toward efficiency construction. By messaging would quickly warn better inform affected communities impacted by hazards. Such has potential save many lives possible.

Language: Английский

Citations

1
Narrativas no sexistas para la infancia e inteligencia artificial. Estudio de aplicaciones para una educación inclusiva DOI Open Access
Fernando Azevedo, Carmen Ferreira Boo, Marta Neira Rodríguez

et al.

La Palabra, Journal Year: 2024, Volume and Issue: 48, P. 1 - 18

Published: Dec. 2, 2024

Con el frenético avance de la inteligencia artificial (IA), se ponen manifiesto las múltiples funcionalidades que esta puede tener en distintos sectores, incluida producción literatura infantil. Se pretende analizar cómo IA promueve valores y representaciones género narraciones creadas para infancia. Mediante una metodología exploratoria cualitativa, contrastan narrativas generadas por dos aplicaciones disponibles plataforma Product Hunter, emplean IA, con cuatro modelos Large Language Models, a partir un mismo prompt. Los resultados muestran configura como herramienta poderosa promover no sexistas e inclusivas generar relatos infancia desafíen estereotipos promuevan diversas género. No obstante, concluye hace necesaria colaboración entre desarrolladores especialistas infantil estudiosos formar generación más consciente tolerante diversidad.

Citations

1
Summary-Sentence Level Hierarchical Supervision for Re-Ranking Model of Two-Stage Abstractive Summarization Framework DOI Creative Commons
E. Yoo, Gyunyeop Kim, Sangwoo Kang

et al.

Mathematics, Journal Year: 2024, Volume and Issue: 12(4), P. 521 - 521

Published: Feb. 7, 2024

Fine-tuning a pre-trained sequence-to-sequence-based language model has significantly advanced the field of abstractive summarization. However, early models summarization were limited by gap between training and inference, they did not fully utilize potential model. Recent studies have introduced two-stage framework that allows second-stage to re-rank candidate summary generated first-stage model, resolve these limitations. In this study, we point out supervision method performed in existing re-ranking cannot learn detailed complex information data. addition, present problem positional bias encoder–decoder-based To address two limitations, study proposes hierarchical jointly performs sentence-level supervision. For supervision, designed loss functions: intra- inter-intra-sentence ranking losses. Compared proposed exhibited performance improvement for both CNN/DM XSum datasets. The outperformed baseline under few-shot setting.

Language: Английский

Citations

0
Revealing the Unseen: AI Chain on LLMs for Predicting Implicit Data Flows to Generate Data Flow Graphs in Dynamically-Typed Code DOI Open Access
Qing Huang,

Zhiwen Luo,

Zhenchang Xing

et al.

ACM Transactions on Software Engineering and Methodology, Journal Year: 2024, Volume and Issue: unknown

Published: June 12, 2024

Data flow graphs (DFGs) capture definitions (defs) and uses across program blocks, which is a fundamental representation for analysis, testing maintenance. However, dynamically-typed programming languages like Python present implicit data issues that make it challenging to determine def-use information at compile time. Static analysis methods Soot WALA are inadequate handling these issues, manually enumerating comprehensive heuristic rules impractical. Large pre-trained language models (LLMs) offer potential solution, as they have powerful understanding pattern matching abilities, allowing them predict by analyzing code context relationships between variables, functions, statements in code. We propose leveraging LLMs’ in-context learning ability learn patterns from contextual solve problems. To further enhance the accuracy of LLMs, we design five-step Chain Thought (CoT) break down into an AI chain, with each step corresponding separate unit generate accurate DFGs Our approach’s performance thoroughly assessed, demonstrating effectiveness Chain. Compared static our method achieves 82% higher def coverage 58% use DFG generation on flow. also prove indispensability Overall, approach offers promising direction building software engineering tools utilizing foundation models, eliminating significant maintenance effort, but focusing identifying problems solve.

Language: Английский

Citations

0
MED-ChatGPT CoPilot: a ChatGPT medical assistant for case mining and adjunctive therapy DOI Creative Commons
Wei Liu,

Hongxing Kan,

Yanfei Jiang

et al.

Frontiers in Medicine, Journal Year: 2024, Volume and Issue: 11

Published: Oct. 16, 2024

Background The large-scale language model, GPT-4-1106-preview, supports text of up to 128 k characters, which has enhanced the capability processing vast quantities text. This model can perform efficient and accurate data mining without need for retraining, aided by prompt engineering. Method research approach includes engineering vectorization processing. In this study, is applied assist ChatGPT in mining. Subsequently, mined results are vectorized incorporated into a local knowledge base. After cleansing 306 medical papers, extraction was performed using ChatGPT. Following validation filtering process, 241 case entries were obtained, leading construction Additionally, drawing upon Langchain framework utilizing base conjunction with ChatGPT, we successfully developed fast reliable chatbot. chatbot capable providing recommended diagnostic treatment information various diseases. Results performance designed from base, exceeded that original 7.90% on set questions. Conclusion assisted engineering, demonstrates effective capabilities texts. future, plan incorporate richer array data, expand scale enhance ChatGPT’s field.

Language: Английский

Citations

0