ABSTRACT
In this letter, we present a comprehensive analysis of the security landscape surrounding large language models, leveraging a dataset from 2022 to 2023. We delve into the collaborative and global nature of this field, emphasizing the interdisciplinary approach required to address the evolving challenges. Prominent keywords like ‘language model,’ ‘cybersecurity,’ and ‘data privacy’ underscore central research themes. Beyond analysis, we introduce a cutting-edge deep CNN-based security framework, successfully tested with the KDDCup dataset, resulting in an impressive 94.73% accuracy on the testing dataset. As large language models continue to play a pivotal role in diverse applications, our research underscores the urgency of international cooperation and innovative security measures to ensure their responsible and secure usage in our interconnected world. This letter serves as a foundation for future research, driving the collective effort to safeguard large language models.
1. Introduction
Large language models, such as GPT-3 (Generative Pre-trained Transformer 3) and BERT (Bidirectional Encoder Representations from Transformers), have revolutionised the field of natural language processing (NLP) and artificial intelligence (AI) in recent years (Kumbhojkar and Balakrishna Menon Citation2022; Shen and Kejriwal Citation2023; Zdravković, Panetto, and Weichhart Citation2022). These models are based on transformer neural networks and have demonstrated remarkable capabilities in understanding and generating human-like text.GPT-3, developed by OpenAI, is one of the largest language models to date, with 175 billion parameters (Arcas Citation2022; Fatemidokht et al. Citation2021; Usuga et al. Citation2022). It has been trained on a vast amount of text data from the internet, allowing it to learn patterns, context, and semantic relationships in language. GPT-3 can generate coherent and contextually relevant text, making it useful for a wide range of applications, including language translation, chatbots, content generation, and more.BERT, on the other hand, is a language model developed by Google. It introduced the concept of bidirectional training, allowing the model to consider both the left and right context of a word during training (Gupta, Gaurav, Kumar Panigrahi, et al. Citation2023; Huang et al. Citation2023; Shen and Kejriwal Citation2023). This bidirectional approach enables BERT to capture a deeper understanding of language and context, improving performance in various NLP tasks, such as question answering, sentiment analysis, and named entity recognition.
The development of large language models like GPT-3 and BERT has significantly advanced the field of NLP and AI. These models have achieved state-of-the-art performance on a wide range of benchmark datasets and have been widely adopted by researchers and industry practitioners alike (Gupta et al. Citation2022; Ling and Jia Hao Citation2022; Shen and Kejriwal Citation2023). They have opened up new possibilities for natural language understanding, generation, and interaction, enabling applications that were previously considered challenging or even impossible.
However, along with their remarkable capabilities, large language models also pose challenges and concerns. One major challenge is the computational resources required to train and deploy these models. Training models with billions of parameters demand significant computational power and energy consumption (Gupta, Gaurav, Arya, et al. Citation2023; Hu et al. Citation2022; Shen and Kejriwal Citation2023; Singh and Gupta Citation2022). This raises concerns about the environmental impact and sustainability of large-scale language model development.
Another challenge is the potential for bias and ethical issues in the output of these models. Since these models are trained on vast amounts of internet data, they can inadvertently learn and reproduce biases present in the training data (Almomani et al. Citation2022; Fatemidokht et al. Citation2021; Nhi, Thanh Manh, and The Van Citation2022; Shen and Kejriwal Citation2023). This can lead to biased or discriminatory outputs, perpetuating existing inequalities and injustices (Khanam, Tanweer, and Sibtain Khalid Citation2022; Onyebuchi et al. Citation2022). Addressing bias and ensuring the ethical use of these models is crucial to mitigate potential harm.
However, there are privacy and security issues associated with deploying huge language models. These models need access to massive volumes of data, some of which may be private or otherwise sensitive. Protecting user privacy and preventing data breaches and abuse are crucial to retaining users’ confidence and avoiding negative consequences.In conclusion, huge language models like GPT-3 and BERT have completely reshaped the landscape of NLP and AI, paving the way for remarkable breakthroughs in the areas of both natural language comprehension and creation. These models analyse and interpret sophisticated linguistic patterns using deep learning and transformer architectures. However, difficulties like as computing needs, prejudice, ethical considerations, and data privacy need to be addressed to enable responsible and productive usage of these models.
2. Related work
Large language models (LLMs) have become a significant area of research in recent years. LLMs are based on artificial intelligence models that aim to achieve human-like artificial general intelligence (Arcas Citation2022). Due to their vast capabilities, they have a wide range of applications in various fields.
In the field of medicine, LLMs have shown great potential. For example, the large language model GPT-4, developed by OpenAI, has been used to enhance the competencies of clinicians in using artificial intelligence-based tools (Bair and Norden Citation2023). LLMs can assist in tasks such as diagnosis, treatment planning, and medical research, thereby improving the quality of healthcare.LLMs have also been studied in the context of language development and behaviour. Research has shown that languages that grammatically associate the future and the present can influence future-oriented behaviour (M. Chen Citation2013). Additionally, language skills have been identified as a target for intervention work to address behaviour problems in children (Chow, Ekholm, and Coleman Citation2018). The development of language in the early school years has been found to be influenced by close relationships with teachers (Spilt, Koomen, and Harrison Citation2015).
In the field of economics, the effect of languages on economic behaviour has been explored. It has been found that languages can influence corporate savings behaviour and belief formation in firms (S. Chen et al. Citation2017). The grammatical structure of languages, particularly the way they handle future tense, can have implications for economic decision-making.LLMs also have implications for language assessment and language teaching. Language assessment literacy (LAL) research has focused on developing the assessment skills of language teachers (Larenas and Brunfaut Citation2022). LLMs can assist in language assessment by providing automated scoring and feedback. Furthermore, the use of LLMs in language teaching can enhance the language learning experience for students.
The development and regulation of LLMs present challenges in various areas. The public availability and interest in LLMs have raised concerns about issues such as intellectual property, ownership, privacy, and liability (Minssen, Vayena, and Cohen Citation2023). Regulating the medical use of LLMs, such as OpenAI’s ChatGPT, requires addressing these challenges to ensure safe and ethical use in healthcare. Hence there is a need of models for the protection of LLMs protection against different types of cyber attacks.
3. Methods
The methodology employed for this letter was carefully structured to provide a comprehensive analysis of the security development in this evolving field. The foundation of our study lies in the systematic data collection from the Scopus database, covering the period from 2022 to 2023. We diligently selected 39 diverse sources, including journals, books, conference papers, and more, to ensure a broad representation of research in this domain. Including various document types allowed us to encompass a wide range of research contributions. Crucially, our methodology was underpinned by extensive bibliometric analysis. This entailed a deep dive into the keyword landscape, author collaborations, and the global distribution of research contributions. Through data analysis, we assessed average publication years, citation trends, and references while also scrutinising the collaborative dynamics among authors. A qualitative dataset analysis was employed to categorise security challenges, advancements, and key themes, ultimately yielding valuable insights. This methodological approach ensured a rigorous examination of the current state of security in large language models and provided a solid foundation for drawing meaningful conclusions and implications for future research in the field.
4. Results
The dataset extracted from Scopus, from 2022 to 2023, provides a valuable snapshot of the most recent research in large language models and their security implications, as represented in . Also, presents the highly cited papers. Comprising 49 documents from a diverse range of sources, including journals, books, conference papers, and more, it encompasses a comprehensive view of the current academic landscape. The data’s freshness is evident, with an average publication year of only 0.102, indicating the relevance and currency of the research. Each document in the dataset has been cited an average of 5.51 times, suggesting that these papers are actively contributing to the academic discourse. Furthermore, the average number of citations per year per document (5.439) underscores the high level of interest and engagement within the academic community. With 2002 references, the dataset is well-researched and builds upon existing literature, indicative of a rich academic context. The dataset consists of various document types, including articles, book chapters, conference papers, conference reviews, notes, and reviews, reflecting a diverse array of research contributions. In addition to that, it consists of 457 Keywords Plus and 163 Author’s Keywords, both of which provide insightful information into the primary subjects and issues discussed in the texts. The dataset demonstrates the participation of a substantial number of scholars in the subject since it contains the names of 271 different writers who contributed to 283 appearances. Most documents are authored by multiple individuals (268 authors), with only three single-authored documents. This highlights the collaborative nature of research in this domain, as evidenced by the average number of authors per document (5.53), co-authors per document (5.78), and a Collaboration Index of 6.09. These collaborative efforts contribute to the robustness and depth of the research in the field, making this dataset a valuable resource for in-depth bibliometric analysis and understanding the current state of large language model research and its security implications.
Figure 1. Dataset information.
Table 1. Highly cited Paper.
4.1 Analysis of source
The dataset includes a diverse array of publishing sources, as represented in , shedding light on the broad dissemination of research on large language models and their security aspects. Notably, CEUR Workshop Proceedings, renowned for hosting academic workshops, contains four articles indicating active discussions about security in large language models within workshop settings. Lecture Notes in Computer Science, a respected series in the field of computer science and artificial intelligence, features three articles emphasising the significance of security topics within this academic domain. Additionally, the ACM International Conference Proceeding Series, with two articles, suggests the importance of large language model security within the ACM community. IEEE Access, an open-access journal covering various technology aspects, comprises two articles underlining the interdisciplinary nature of research in language model security. The presence of two articles in the NAACL 2022 conference proceedings, a major event in computational linguistics, highlights the significance of security in this field. Furthermore, the inclusion of two articles in the Proceedings of the Annual Meeting of the Association for Computational Linguistics indicates the relevance of security concerns within this community.
Figure 2. Most relevant source.
4.2 Analysis of authors
represents authors who have published papers related to massive language models and their security implications. Several writers have made significant contributions to this collection. Indicative of their dedication to and mastery of the topic, writers such as Ahmad B, Dolan-Gavitt B, Jana S, Karri R, Kshetri N, Pearce H, Pei K, Ray B, Susnjak T, Tan B, Xuan Z, and Yang J have each contributed to the field on two separate occasions. This may indicate that the authors are working together on a comprehensive study of the security of huge language models. The dataset also includes works by a wider variety of writers, many of whom are credited as sole authors. This eclectic group of writers is indicative of the dynamic and cooperative character of the research environment in this area, as specialists from several disciplines work together to solve the security problems presented by huge linguistic models. Scholars in this subject may use this data to learn about the breadth and depth of the scholarly community dedicated to this topic, as well as to locate suitable partners for future research projects.
Figure 3. Most relevant authors.
4.3 Analysis of countries
presents a detailed picture of the geographical distribution of research contributions in the area of security for big language models. The United States of America (U.S.A.) is a leading contributor in this field of study, with 116 papers in total. After the US, the UK shows a significant academic presence on the topic with 29 papers. China and India, with 21 and 13 articles, respectively, also make significant contributions, highlighting the growing importance of research from these emerging technology hubs. Australia, Italy, and Spain each offer many articles, with 12, 10, and 10 contributions, respectively. Switzerland and Germany contributed 9 and 8 articles, showcasing their active participation in addressing security challenges related to large language models. Other European countries like the Netherlands, France, and Canada are also actively involved, with 6, 5, and 4 articles, respectively. Beyond these, several other countries, including Singapore, South Korea, Hungary, Norway, Poland, Saudi Arabia, and Belgium, each provide a smaller, yet noteworthy, number of research articles. This international diversity reflects the research landscape’s global nature, where many countries are actively engaged in addressing the security implications of large language models. Authors can leverage this information to identify potential collaborators and understand the geographical distribution of research efforts in the field. It also underscores the importance of a collaborative, global approach to effectively tackle the security challenges of large language models.
Figure 4. Country scientific production.
4.4 Analysis of keywords
As represented in , the dataset provides valuable insight into the most important keywords that feature prominently in the research related to large language models and their security. ‘Language model’ emerges as the most frequent keyword, with a frequency of 33, indicating the central focus of many research articles on these models. ‘Large language model’ is also highly significant, with a frequency of 19, reflecting the specific emphasis on security within the context of these substantial language models. ‘Computational linguistics’ and ‘artificial intelligence’ follow closely, with frequencies of 18 and 11, respectively, highlighting the interdisciplinary nature of the research field.
Figure 5. Keywords distribution.
‘Natural languages’ and ‘natural language processing systems’ appear nine and seven times, respectively, underscoring the importance of understanding the intricacies of human language in the context of security. ‘Cybersecurity’ and ‘cyber security’, with eight and seven occurrences, are indicative of the increasing concern for safeguarding these language models in the digital realm. ‘Data privacy’ also features prominently, with a frequency of six, reflecting the growing awareness of the privacy implications surrounding language models.
Moreover, ‘machine learning’ is a recurring keyword, with six instances showcasing the role of machine learning techniques in addressing security challenges. This set of important keywords collectively reflects the multifaceted nature of research on large language models and their security implications. Authors and researchers in this field can utilise this information to gain a deeper understanding of the central themes and terminologies that drive current discussions and can guide them in selecting relevant keywords when publishing their own work. These keywords are pivotal in aiding the discoverability and categorisation of research in this dynamic and critical domain.
5. Major security challenges associated with large language models
Large language models, such as GPT-3 and BERT, have gained considerable attention due to their remarkable text-generation capabilities. However, their widespread adoption is accompanied by various security concerns that necessitate careful consideration.
5.1 Malicious use
One paramount security challenge pertains to the potential for malicious use. These models can craft highly convincing fake news articles, social media posts, or deepfake videos, enabling the spread of misinformation, manipulation of public opinion, or deception (Cohen et al. Citation2022). Their proficiency in generating text indistinguishable from human writing raises information authenticity and reliability concerns.
5.2 Bias and discrimination
Another challenge relates to bias and discrimination in model outputs. Trained on vast internet data, these models may inadvertently learn and perpetuate biases present in the training data, potentially leading to biased or discriminatory outputs (Cohen et al. Citation2022). This can have profound implications, particularly in domains such as hiring, legal decision-making, or content moderation, where biased language models can reinforce inequalities.
5.3 Adversarial attacks
Large language models are also susceptible to adversarial attacks. Adversaries can subtly manipulate inputs, adding imperceptible perturbations or altering context to manipulate model outputs (Cohen et al. Citation2022). Such attacks can deceive the model into generating incorrect or malicious results, with repercussions in applications like content moderation and natural language understanding.
5.4 Privacy and data security
Concerns about privacy and data security arise from the need for substantial training data, including sensitive information. Storage and processing of such datasets raise apprehensions about data breaches and unauthorised access (Cohen et al. Citation2022). Furthermore, deployment in cloud environments or on-edge devices can expose models to security risks.
6. Addressing security challenges
Several approaches are imperative to tackle these security challenges. Robust evaluation and testing of large language models are essential to identify and mitigate potential risks and vulnerabilities (Cohen et al. Citation2022). Evaluation should encompass aspects of bias, fairness, and resistance to adversarial attacks.
Developing techniques and tools to detect and mitigate misinformation is crucial (Cohen et al. Citation2022). This may involve the use of fact-checking algorithms, content verification techniques, or user education campaigns.
Enhancing transparency and explainability is another key step. Clearer documentation and guidelines regarding model functionality, training data, and data handling are needed (Cohen et al. Citation2022). Collaboration among researchers, policymakers, and industry stakeholders is also vital to formulating ethical guidelines and best practices for model development and deployment.
7. Discussions
After analysing the current literature about large language models, we developed a deep CNN-based filter that protects the LLMs from different types of network-based attacks, such as DDoS and DoS attacks.
A deep Convolutional Neural Network (CNN) model is at the core of our framework, as represented in . This model is designed to process incoming traffic in real time and identify potential threats. The deep CNN model comprises three layers, each consisting of a Conv2d operation followed by ReLU activation, another Conv2d layer with ReLU activation, and a subsequent max-pooling and batch normalisation step. After these operations, the data is flattened and passed through a dense layer for further analysis.
Figure 6. Proposed Model.
7.1 Anomaly detection
The deep CNN model operates in real-time, analysing the traffic data as it arrives. During the first time window, incoming traffic is subjected to this deep CNN-based filter. The model is trained to recognise patterns and anomalies in the data, emphasising identifying malicious IP addresses.
7.2 Blacklisting and traffic blocking
Any IP addresses flagged as malicious by the deep CNN model during this initial analysis are added to a blacklist. Subsequently, in the next time window, all incoming traffic from the blacklisted IP addresses is automatically blocked. This proactive approach is a security measure to prevent potential threats from accessing the large language model.
7.3 Model structure
We depict the architecture and structure of our deep CNN model in a visual representation, a figure that encapsulates its components and the flow of data through the network. This figure aids in providing a clear and concise understanding of how our framework operates and how the deep CNN model fits within the broader security strategy.
Our proposed framework leverages deep learning and real-time analysis to protect large language models from anomaly attacks. It operates as an intelligent gatekeeper, filtering incoming traffic, identifying malicious IP addresses, and taking decisive action to safeguard the model. This innovative approach aligns with the evolving landscape of cybersecurity, ensuring that large language models remain secure and reliable in various applications.
7.4 Model evaluation
In our study, we conducted a comprehensive evaluation of our proposed deep Convolutional Neural Network (CNN) model within the context of securing large language models from anomaly attacks, utilising the KDDCup dataset. The accuracy and loss metrics from our training and testing phases reveal our model’s significant progress in learning and adaptation. At the outset, during Epoch 0, our model exhibited relatively low accuracy, but as training progressed, accuracy improved markedly. By Epoch 9, we achieved an impressive accuracy of 94.09% in the training dataset and 94.73% in the testing dataset, underlining the robustness and effectiveness of our approach. highlights the steady growth in accuracy over epochs for both training and testing data, demonstrating the model’s continuous improvement in making correct predictions. Also, illustrates the progressive reduction in the loss function, indicating the optimisation of our model’s parameters. These figures are valuable visual aids, offering a quick and insightful overview of our model’s performance. In conclusion, our deep CNN-based framework demonstrates its efficacy in protecting large language models from security threats, as evidenced by the substantial improvements in accuracy and loss reduction throughout the training process.
Figure 7. Accuracy and loss function.
8. Conclusive remarks
Our comprehensive analysis of the security landscape surrounding large language models, based on data from 2022 to 2023, provides critical insights into the ever-evolving field. Collaborative efforts across diverse sources and countries underscore the global commitment to addressing security challenges. Prominent keywords like ’language model’, ’cybersecurity’, and ’data privacy’ emphasise central research themes. Our contribution extends beyond observation by introducing a groundbreaking deep CNN-based security framework tested with the KDDCup dataset. Results showcase remarkable progress, with our model achieving a remarkable 94.73% accuracy on the testing dataset. As large language models continue to shape various applications, this research underscores the need for interdisciplinary approaches and innovative security measures. It is a foundation for future research, emphasising international collaboration in securing these models in our interconnected world.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Additional information
Funding
References
- Akbar, K. A., S. Md Halim, Y. Hu, A. Singhal, L. Khan, and B. Thuraisingham. 2022. “Knowledge Mining in Cybersecurity: From Attack to Defense.” Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 13383 LNCS, 427–441. https://doi.org/10.1007/978-3-031-10684-2_7.
- Almomani, A., M. Alauthman, M. Taib Shatnawi, M. Alweshah, A. Alrosan, W. Alomoush, and B. B. Gupta. 2022. “Phishing Website Detection with Semantic Features Based on Machine Learning Classifiers: A Comparative Study.” International Journal on Semantic Web and Information Systems (IJSWIS) 18 (1): 1–24. https://doi.org/10.4018/IJSWIS.297032.
- Arcas, B. 2022. “Do Large Language Models Understand Us?” Daedalus 151 (2): 183–197. https://doi.org/10.1162/daed_a_01909.
- Bair, H., and J. Norden. 2023. “Large Language Models and Their Implications on Medical Education.” Academic Medicine 98 (8): 869–870. https://doi.org/10.1097/acm.0000000000005265.
- Chen, M. 2013. “The Effect of Language on Economic Behavior: Evidence from Savings Rates, Health Behaviors, and Retirement Assets.” The American Economic Review 103 (2): 690–731. https://doi.org/10.1257/aer.103.2.690.
- Chen, S., H. Cronqvist, S. Ni, and F. Zhang. 2017. “Languages and Corporate Savings Behavior.” Journal of Corporate Finance 46:320–341. https://doi.org/10.1016/j.jcorpfin.2017.07.009.
- Chow, J., E. Ekholm, and H. Coleman. 2018. “Does Oral Language Underpin the Development of Later Behavior Problems? A Longitudinal Meta-Analysis.” School Psychology Quarterly 33 (3): 337–349. https://doi.org/10.1037/spq0000255.
- Cohen, J., J. Wright-Berryman, L. Rohlfs, D. Trocinski, L. Daniel, and T. Klatt. 2022. “Integration and Validation of a Natural Language Processing Machine Learning Suicide Risk Prediction Model Based on Open-Ended Interview Language in the Emergency Department.” Frontiers in Digital Health 4. https://doi.org/10.3389/fdgth.2022.818705.
- Crothers, E. N., N. Japkowicz, and H. L. Viktor. 2023. “Machine-Generated Text: A Comprehensive Survey of Threat Models and Detection Methods.” Institute of Electrical and Electronics Engineers Access 11:70977–71002. https://doi.org/10.1109/ACCESS.2023.3294090.
- Dwivedi, Y. K., N. Kshetri, L. Hughes, E. L. Slade, A. Jeyaraj, A. K. Kar, A. M. Baabdullah, et al. 2023. “Opinion Paper: “So What if ChatGPT Wrote it?” Multidisciplinary Perspectives on Opportunities, Challenges and Implications of Generative Conversational AI for Research, Practice and Policy.” International Journal of Information Management 71:102642. https://doi.org/10.1016/j.ijinfomgt.2023.102642.
- Eggmann, F., R. Weiger, N. U. Zitzmann, and M. B. Blatz. 2023. “Implications of Large Language Models Such as ChatGPT for Dental Medicine.” Journal of Esthetic and Restorative Dentistry 35 (7): 1098–1102. https://doi.org/10.1111/jerd.13046.
- Fatemidokht, H., M. Kuchaki Rafsanjani, B. B. Gupta, and C.-H. Hsu. 2021. “Efficient and Secure Routing Protocol Based on Artificial Intelligence Algorithms with UAV-Assisted for Vehicular Ad Hoc Networks in Intelligent Transportation Systems.” IEEE Transactions on Intelligent Transportation Systems 22 (7): 4757–4769. https://doi.org/10.1109/TITS.2020.3041746.
- Gupta, B. B., A. Gaurav, V. Arya, and K. Tai Chui. 2023. “Recent Research on Artificial Intelligence, Machine Learning, and Deep Learning in Consumer Electronics Technology.” In 2023 IEEE 12th Global Conference on Consumer Electronics, Nara, Japan, (GCCE), 1011–1013. IEEE. https://doi.org/10.1109/GCCE59613.2023.10315325.
- Gupta, B. B., A. Gaurav, P. Kumar Panigrahi, and V. Arya. 2023. “Analysis of Artificial Intelligence-Based Technologies and Approaches on Sustainable Entrepreneurship.” Technological Forecasting & Social Change 186:122152. https://doi.org/10.1016/j.techfore.2022.122152.
- Gupta, B. B., A. Tewari, I. Cvitić, D. Peraković, and X. Chang. 2022. “Artificial Intelligence Empowered Emails Classifier for Internet of Things Based Systems in Industry 4.0.” Wireless Networks 28 (1): 493–503. https://doi.org/10.1007/s11276-021-02619-w.
- Huang, Y., S. Lv, K.-K. Tseng, P.-J. Tseng, X. Xie, and R. Fang-Ying Lin. 2023. “Recent Advances in Artificial Intelligence for Video Production System.” Enterprise Information Systems 17 (11): 2246188. https://doi.org/10.1080/17517575.2023.2246188.
- Hu, B., A. Gaurav, C. Choi, and A. Almomani. 2022. “Evaluation and Comparative Analysis of Semantic Web-Based Strategies for Enhancing Educational System Development.” International Journal on Semantic Web and Information Systems (IJSWIS) 18 (1): 1–14. https://doi.org/10.4018/IJSWIS.302895.
- Jouppi, N. P., 2023. Proceedings - International Symposium on Computer Architecture https://doi.org/10.1145/3579371.3589350
- Khanam, S., S. Tanweer, and S. Sibtain Khalid. 2022. “Future of Internet of Things: Enhancing Cloud-Based IoT Using Artificial Intelligence.” International Journal of Cloud Applications and Computing (IJCAC) 12 (1): 1–23. https://doi.org/10.4018/IJCAC.297094.
- Kshetri, N. 2023. “Cybercrime and Privacy Threats of Large Language Models.” IT Professional 25 (3): 9–13. https://doi.org/10.1109/MITP.2023.3275489.
- Kumbhojkar, N. R., and A. Balakrishna Menon. 2022. “Integrated Predictive Experience Management Framework (IPEMF) for Improving Customer Experience: In the Era of Digital Transformation.” International Journal of Cloud Applications and Computing (IJCAC) 12 (1): 1–13. https://doi.org/10.4018/IJCAC.2022010107.
- Larenas, S., and T. Brunfaut. 2022. “But Who Trains the Language Teacher Educator Who Trains the Language Teacher? An Empirical Investigation of Chilean Efl Teacher Educators’ Language Assessment Literacy.” Language Testing 40 (3): 463–492. https://doi.org/10.1177/02655322221134218.
- Ling, Z., and Z. Jia Hao. 2022. “An Intrusion Detection System Based on Normalized Mutual Information Antibodies Feature Selection and Adaptive Quantum Artificial Immune System.” International Journal on Semantic Web and Information Systems (IJSWIS) 18 (1): 1–25. https://doi.org/10.4018/IJSWIS.308469.
- Li, H., Y. Song, and L. Fan. 2022. “You Don’t Know My Favorite Color: Preventing Dialogue Representations from Revealing Speakers’ Private Personas.” NAACL 2022 - 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Proceedings of the Conference.
- Maddigan, P., and T. Susnjak. 2023. “Chat2VIS: Generating Data Visualizations via Natural Language Using ChatGPT, Codex and GPT-3 Large Language Models.” Institute of Electrical and Electronics Engineers Access 11:45181–45193. https://doi.org/10.1109/ACCESS.2023.3274199.
- Minssen, T., E. Vayena, and I. Glenn Cohen. 2023. “The Challenges for Regulating Medical Use of ChatGPT and Other Large Language Models.” JAMA 330 (4): 315–316. https://doi.org/10.1001/jama.2023.9651.
- Nhi, N. T. U., L. Thanh Manh, and T. The Van. 2022. “A Model of Semantic-Based Image Retrieval Using C-Tree and Neighbor Graph.” International Journal on Semantic Web and Information Systems (IJSWIS) 18 (1): 1–23. https://doi.org/10.4018/IJSWIS.295551.
- Onyebuchi, A., U. O. Matthew, J. S. Kazaure, N. U. Okafor, O. Daniel Okey, P. I. Okochi, J. Folasade Taiwo, and A. Okechukwu Matthew. 2022. “Business Demand for a Cloud Enterprise Data Warehouse in Electronic Healthcare Computing: Issues and Developments in E-Healthcare Cloud Computing.” International Journal of Cloud Applications and Computing (IJCAC) 12 (1): 1–22. https://doi.org/10.4018/IJCAC.297098.
- Pearce, H., B. Tan, B. Ahmad, R. Karri, and B. Dolan-Gavitt. 2023. “Examining Zero-Shot Vulnerability Repair with Large Language Models.” Proceedings - IEEE Symposium on Security and Privacy. https://doi.org/10.1109/SP46215.2023.10179324.
- Pei, K., Z. Xuan, J. Yang, S. Jana, and B. Ray. 2023. “Learning Approximate Execution Semantics from Traces for Binary Function Similarity.” IEEE Transactions on Software Engineering 49 (4): 2776–2790. https://doi.org/10.1109/TSE.2022.3231621.
- Sharma, P., and S. Parasa. 2023. “ChatGPT and Large Language Models in Gastroenterology.” Nature Reviews Gastroenterology & Hepatology 20 (8): 481–482. https://doi.org/10.1038/s41575-023-00799-8.
- Shen, K., and M. Kejriwal. 2023. “An Experimental Study Measuring the Generalization of Fine‐Tuned Language Representation Models Across Commonsense Reasoning Benchmarks.” International Journal of Offshore and Polar Engineering 40 (5). https://doi.org/10.1111/exsy.13243.
- Singh, A., and B. B. Gupta. 2022. “Distributed Denial-Of-Service (DdoS) Attacks and Defense Mechanisms in Various Web-Enabled Computing Platforms: Issues, Challenges, and Future Research Directions.” International Journal on Semantic Web and Information Systems (IJSWIS) 18 (1): 1–43. https://doi.org/10.4018/IJSWIS.297143.
- Spilt, J., H. Koomen, and L. Harrison. 2015. “Language Development in the Early School Years: The Importance of Close Relationships with Teachers.” Developmental Psychology 51 (2): 185–196. https://doi.org/10.1037/a0038540.
- Tan, T. F., J. P. C. Arun James Thirunavukarasu, P. A. Keane, L. R. Pasquale, M. D. Abramoff, J. Kalpathy-Cramer, J. Kalpathy-Cramer, et al. 2023. “Generative Artificial Intelligence Through ChatGPT and Other Large Language Models in Ophthalmology: Clinical Applications and Challenges.” Ophthalmology Science 3 (4): 100394. https://doi.org/10.1016/j.xops.2023.100394.
- Usuga, C., J. Pablo, B. Grabot, S. Lamouri, and R. Pellerin. 2022. “Valuing Free-Form Text Data from Maintenance Logs Through Transfer Learning with Camembert.” Enterprise Information Systems 16 (6): 1790043. https://doi.org/10.1080/17517575.2020.1790043.
- Venerito, V., D. Puttaswamy, F. Iannone, and L. Gupta. 2023. “Large Language Models and Rheumatology: A Comparative Evaluation.” The Lancet Rheumatology 5 (10): e574–78. https://doi.org/10.1016/S2665-9913(23)00216-3.
- Zdravković, M., H. Panetto, and G. Weichhart. 2022. “AI-Enabled Enterprise Information Systems for Manufacturing.” Enterprise Information Systems 16 (4): 668–720. https://doi.org/10.1080/17517575.2021.1941275.
- Zhao, R., Y. Huang, W. Ma, W. Dong, G. Xian, and T. Sun. 2023. “Insights and Reflections of the Impact of ChatGPT on Intelligent Knowledge Services in Libraries; [ChatGPT 对图书馆智能知识服务的启示与思考].” Journal of Library and Information Science in Agriculture 35 (1): 29–38. https://doi.org/10.13998/j.cnki.issn1002-1248.23-0116.
- Zhao, X., L. Li, and Y.-X. Wang. 2022. “Provably Confidential Language Modelling.” NAACL 2022 - 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Proceedings of the Conference.