General Language Research Articles

GPT-4 Vision: Multi-Modal Evolution of ChatGPT and Potential Role in Radiology.

GPT-4 Vision (GPT-4V) represents a significant advancement in multimodal artificial intelligence, enabling text generation from images without specialized training. This marks the transformation of ChatGPT as a large language model (LLM) into GPT-4's promised large multimodal model (LMM). As these AI models continue to advance, they may enhance radiology workflow and aid with decision support. This technical note explores potential GPT-4V applications in radiology and evaluates performance for sample tasks.GPT-4V capabilities were tested using images from the web, personal and institutional teaching files, and hand-drawn sketches. Prompts evaluated scientific figure analysis, radiologic image reporting, image comparison, handwriting interpretation, sketch-to-code, and artistic expression. In this limited demonstration of GPT-4V's capabilities, it showed promise in classifying images, counting entities, comparing images, and deciphering handwriting and sketches. However, it exhibited limitations in detecting some fractures, discerning a change in size of lesions, accurately interpreting complex diagrams, and consistently characterizing radiologic findings. Artistic expression responses were coherent. WhileGPT-4V may eventually assist with tasks related to radiology, current reliability gaps highlight the need for continued training and improvement before consideration for any medical use by the general public and ultimately clinical integration. Future iterations could enable a virtual assistant to discuss findings, improve reports, extract data from images, provide decision support based on guidelines, white papers, and appropriateness criteria. Human expertise remain essential for safe practice and partnerships between physicians, researchers, and technology leaders are necessary to safeguard against risks like bias and privacy concerns.

FET-LM: Flow-Enhanced Variational Autoencoder for Topic-Guided Language Modeling.

Variational autoencoder (VAE) is widely used in tasks of unsupervised text generation due to its potential of deriving meaningful latent spaces, which, however, often assumes that the distribution of texts follows a common yet poor-expressed isotropic Gaussian. In real-life scenarios, sentences with different semantics may not follow simple isotropic Gaussian. Instead, they are very likely to follow a more intricate and diverse distribution due to the inconformity of different topics in texts. Considering this, we propose a flow-enhanced VAE for topic-guided language modeling (FET-LM). The proposed FET-LM models topic and sequence latent separately, and it adopts a normalized flow composed of householder transformations for sequence posterior modeling, which can better approximate complex text distributions. FET-LM further leverages a neural latent topic component by considering learned sequence knowledge, which not only eases the burden of learning topic without supervision but also guides the sequence component to coalesce topic information during training. To make the generated texts more correlative to topics, we additionally assign the topic encoder to play the role of a discriminator. Encouraging results on abundant automatic metrics and three generation tasks demonstrate that the FET-LM not only learns interpretable sequence and topic representations but also is fully capable of generating high-quality paragraphs that are semantically consistent.

Stimulus-Response Patterns: The Key to Giving Generalizability to Text-Based Depression Detection Models.

Text content analysis for depression detection using machine learning techniques has become a prominent area of research. However, previous studies focused mainly on analyzing the textual content, neglecting the fundamental factors driving text generation. Consequently, existing models face the challenge of poor generalization to out-of-domain data as they struggle to capture the crucial features of depression. To address this, we propose a novel computational perspective of "stimulus-response patterns" that brings us closer to the essence of clinical diagnosis of depression. Adopting this computational perspective allows us to conceptually unify diverse datasets and generalize this perspective to common datasets in the field. We introduce the Stimulus-Response Patterns-aware Network (SRP-Net) as an exemplary approach within this computational perspective. To assess the performance of the SRP-Net, we constructed a multi-stimulus dataset and conducted experimental evaluations, demonstrating its exceptional cross-stimulus generalizability. Furthermore, we demonstrated the promising performance of SPR-Net in real medical scenarios and conducted an interpretability analysis of the stimulus-response patterns. Our research investigates the critical role of stimulus-response patterns in enhancing the generalizability of text-based depression detection models, which can potentially facilitate data-driven depression detection to approach the diagnostic accuracy of psychiatrists.

Utilizing Artificial Intelligence-Based Tools for Addressing Clinical Queries: ChatGPT Versus Google Gemini.

Artificial intelligence (AI)-based text generators, such as ChatGPT (OpenAI) and Google Bard (now Google Gemini), have demonstrated proficiency in predicting words and providing responses to various questions. However, their performance in answering clinical queries has not been well assessed. This comparative analysis aimed to assess the capabilities of ChatGPT and Google Gemini in addressing clinical questions. Separate interactions with ChatGPT and Google Gemini were conducted to obtain responses to the clinical question, posed in a PICOT (patient, intervention, comparison, outcome, time) format. To ascertain the accuracy of the information provided by the AI chat bots, a thorough examination of full-text articles was conducted. Although ChatGPT exhibited relative accuracy in generating bibliographic information, it displayed some inconsistencies in clinical content. Conversely, Google Gemini generated citations and summaries that were entirely fabricated. Despite generating responses that may appear credible, both AI-based tools exhibited factual inaccuracies, raising substantial concerns about their reliability as potential sources of clinical information. [J Nurs Educ. 2024;63(8):556-559.].

Email subjects generation with large language models: GPT-3.5, PaLM 2, and BERT

In order to enhance marketing efforts and improve the performance of marketing campaigns, the effectiveness of language generation models needs to be evaluated. This study examines the performance of large language models (LLMs), namely GPT-3.5, PaLM 2, and bidirectional encoder representations from transformers (BERT), in generating email subjects for advertising campaigns. By comparing their results, the authors evaluate the efficacy of these models in enhancing marketing efforts. The objective is to explore how LLMs contribute to creating compelling email subject lines and improving opening rates and campaign performance, which gives us an insight into the impact of these models in digital marketing. In this paper, the authors first go over the different types of language models and the differences between them, before giving an overview of the most popular ones that will be used in the study, such as GPT-3.5, PaLM 2, and BERT. This study assesses the relevance, engagement, and uniqueness of GPT-3.5, PaLM 2, and BERT by training and fine-tuning them on marketing texts. The findings provide insights into the major positive impact of artificial intelligence (AI) on digital marketing, enabling informed decision-making for AI-driven email marketing strategies.

Customer service chatbot enhancement with attention-based transfer learning

Customer service is an important and expensive aspect of business, often being the largest department in most companies. With growing societal acceptance and increasing cost efficiency due to mass production, service robots are beginning to cross from the industrial domain to the social domain. Currently, customer service robots tend to be digital and emulate social interactions through on-screen text, but state-of-the-art research points towards physical robots soon providing customer service in person. This article explores the feasibility of Transfer Learning different customer service domains to improve chatbot models. In our proposed approach, transfer learning-based chatbot models are initially assigned to learn one domain from an initial random weight distribution. Each model is then tasked with learning another domain by transferring knowledge from the previous domains. To evaluate our approach, a range of 19 companies from domains such as e-Commerce, telecommunications, and technology are selected through social interaction with X (formerly Twitter) customer support accounts. The results show that the majority of models are improved when transferring knowledge from at least one other domain, particularly those more data-scarce than others. General language transfer learning is observed, as well as higher-level transfer of similar domain knowledge. For each of the 19 domains, the Wilcoxon signed-rank test suggests that 16 have statistically significant distributions between transfer and non-transfer learning. Finally, feasibility is explored for the deployment of chatbot models to physical robot platforms including “Pepper”, a semi-humanoid robot manufactured by SoftBank Robotics, and “Temi”, a personal assistant robot.

Top-ranked U.S. and U.K.’s universities’ first responses to GenAI: key themes, emotions, and pedagogical implications for teaching and learning

The emergence of ChatGPT, a Generative AI program, has sparked discussions about its teaching and learning value, and concerns about academic integrity in higher education (HE). An extant review of the literature indicates that a scarcity of research exists on GenAI, specifically a synthesis of the official views, guidelines and articles of top-ranked universities on the use, limitations, challenges, and opportunities brought by ChatGPT in the early phase when ChatGPT was released in 2022 until early May 2023, which can offer insights into the concerns and recommendations for educators. Using the corpus assisted discourse analysis approach, this study identified the key themes and emotions elicited by evaluations of the ChatGPT situation from a self-built corpus containing 151 articles from 47 top-ranked U.S. universities and 34 U.K. ones. Our findings indicated three prominent themes discussed on official websites, including ChatGPT as a text and content generator, use of ChatGPT in teaching and learning, and potential implications and opportunities of using ChatGPT in HE. Further examination revealed that bias, concern, worry, threat, fear, and trust were the prevailing emotions relating to ChatGPT. Illustrated with examples collected from our corpus, this paper offers an in-depth discussion of universities’ first responses to the use of ChatGPT. The insights gained have some pedagogical implications for academics, researchers and educators and may inform policy and practice in HE on the use of GenAI.

Cross-modal knowledge distillation for continuous sign language recognition

Continuous Sign Language Recognition (CSLR) is a task which converts a sign language video into a gloss sequence. The existing deep learning based sign language recognition methods usually rely on large-scale training data and rich supervised information. However, current sign language datasets are limited, and they are only annotated at sentence-level rather than frame-level. Inadequate supervision of sign language data poses a serious challenge for sign language recognition, which may result in insufficient training of sign language recognition models. To address above problems, we propose a cross-modal knowledge distillation method for continuous sign language recognition, which contains two teacher models and one student model. One of the teacher models is the Sign2Text dialogue teacher model, which takes a sign language video and a dialogue sentence as input and outputs the sign language recognition result. The other teacher model is the Text2Gloss translation teacher model, which targets to translate a text sentence into a gloss sequence. Both teacher models can provide information-rich soft labels to assist the training of the student model, which is a general sign language recognition model. We conduct extensive experiments on multiple commonly used sign language datasets, i.e., PHOENIX 2014T, CSL-Daily and QSL, the results show that the proposed cross-modal knowledge distillation method can effectively improve the sign language recognition accuracy by transferring multi-modal information from teacher models to the student model. Code is available at https://github.com/glq-1992/cross-modal-knowledge-distillation_new.

Automatic Generation Algorithm of Movie and TV Scripts Based on ChatGPT

A movie or TV script is a meticulously crafted document that serves as the blueprint for a visual storytelling project. It outlines the dialogue, actions, and scene descriptions that will unfold on screen, guiding actors, directors, and crew members in bringing the story to life. Each script is divided into scenes and acts, with clear instructions on character entrances and exits, camera angles, and pacing. This paper introduces a novel framework for enhancing the quality of movie and TV scripts through the integration of the Combinational Multi-Stage Genetic Optimization (CMSGO) model with ChatGPT, a state-of-the-art language generation model. The CMSGO model utilizes iterative optimization techniques to systematically refine and enhance script elements such as coherence, dialogue flow, character development, and overall narrative structure. The proposed CMSGO model comprises the Combinational model with the genetic optimization function. The function CMSGO model examines the fitness function with the Multi-Stage Optimization process. The proposed CMSGO model uses the estimation of features in the Multi-stage optimization model with the computation of features related to the scripts. Through 20 generations of optimization, the CMSGO model demonstrates its effectiveness in improving script quality, as evidenced by a steady increase in average script quality scores. Additionally, the multi-stage optimization approach targets specific aspects of script quality, allowing for targeted adjustments to parameters related to character motivations, plot coherence, and tone. Viewer opinions further validate the efficacy of the generated scripts, with positive evaluations across various aspects such as audience engagement, coherence, emotional impact, and originality. The proposed framework offers a robust and data-driven approach to scriptwriting, enabling the creation of high-quality movie and TV scripts that captivate and resonate with audiences, thus enriching the overall viewing experience.

Generality and Exception in Islamic Legal Theory: Intent, Language, and the Jurist’s Role

Abstract Whenever discussions arise about the generality or specificity of law and legal provisions, they occur on at least two different levels. On the external level, there is an expectation that law should be general insofar as similar cases and factual situations are concerned, and specific with regards to different and exceptional cases. This is often invoked as a requirement for what constitutes good law or proper governance. On the internal level, generality refers to the expectation that law must offer coherent rules, principles, or standards of action in some form. General and specific language (al-ʿām wal-khāṣṣ) in Islamic legal theory (uṣūl al-fiqh) enjoyed a noteworthy degree of attention in modern scholarship. This attention largely centers on those concepts’ external potential for driving reform in specific areas of law. This Article, by contrast, focuses on the generality–specificity binary as a tool for ensuring the law’s internal cohesiveness, including its claims to legitimacy and its enforceability. To this end, the Article offers an overview of the examination of general and specific expressions in the thought of two influential twelfth-century C.E. (sixth-centuryA.H.) jurist-theologians of different juristic and theological affiliation: the Shāfiʿī-Ashʿarī Abū Ḥāmid al-Ghazālī (d. 1111 C.E./505 A.H.) and the Central Asian Ḥanafī-Māturīdī ʿAlāʾ al-Dīn al-Samarqandī (d. 1145 C.E./539 A.H.). The Article then goes on to argue that ensuring a coherent relation between intent, language, and norms was a central guiding purpose of these studies. We can see this with particular clarity in the fact that, despite reaching comparable practical positions concerning how norms ought to follow from general language, each jurist attained his conclusion through a different path, reflecting different understandings of the relation of norms to divine intent.

Applying the Markov chain in natural language processing and three-pool model

This paper delves into the application of Markov chains in Natural Language Processing (NLP), and the Markov Chain Monte Carlo (MCMC) methodology relevant to the three-pool model. The former outlines the basic principles of Markov chains, highlighting their utility in predicting word sequences in language modelling and text generation, despite certain limitations. Also, the former describes mathematical frameworks like n-gram models that enhance prediction accuracy by considering multiple preceding words. It acknowledges challenges in NLP such as oversimplification and emotional depth, as well as computational issues in higher-order models. It concludes by discussing the integration of Markov chains with other models to mitigate these limitations, and their enduring relevance in computational linguistics. The later investigates the MCMC methodology, a seminal development in the field of statistical inference, which is especially useful when analysing complicated systems when traditional statistical techniques are inadequate. Moreover, this later explores the fundamental concepts of MCMC, clarifies how it is inherently related to Markov chains, presents the three-pool model that is commonly applied to models of physical, chemical, or ecological systems, and discusses how MCMC can be used to analyse these models.

PIM GPT a hybrid process in memory accelerator for autoregressive transformers

Decoder-only Transformer models such as Generative Pre-trained Transformers (GPT) have demonstrated exceptional performance in text generation by autoregressively predicting the next token. However, the efficiency of running GPT on current hardware systems is bounded by low compute-to-memory-ratio and high memory access. In this work, we propose a Process-in-memory (PIM) GPT accelerator, PIM-GPT, which achieves end-to-end acceleration of GPT inference with high performance and high energy efficiency. PIM-GPT leverages DRAM-based PIM designs for executing multiply-accumulate (MAC) operations directly in the DRAM chips, eliminating the need to move matrix data off-chip. Non-linear functions and data communication are supported by an application specific integrated chip (ASIC). At the software level, mapping schemes are designed to maximize data locality and computation parallelism. Overall, PIM-GPT achieves 41 − 137 × , 631 − 1074 × speedup and 123 − 383 × , 320 − 602 × energy efficiency over GPU and CPU baseline on 8 GPT models with up to 1.4 billion parameters.

Sign language video to text conversion via optimised LSTM with improved motion estimation

ABSTRACT A method of communication for the deaf and the dumb is sign language (SL). Knowing SL allows speakers and listeners to converse effectively with deaf and dumb people. Yet, while they can acquire sign language (SL) to communicate with the dumb and the deaf, untrained persons are unable to speak with them. For such individuals, an SL to text system is more beneficial in enabling them to communicate more effectively with everyday people. To interact with the deaf and dumb, SL is a tactile gesture that employs the hands and eyes. This proposes a new model for translating SL video to text conversion or generation. Frame Conversion is the initial step, where the given input video is converted into frames. Subsequently, features considered are hierarchy of skeleton and Improved Motion estimation (ME). For generating the text, this work uses an optimised LSTM model, where training of weights is done by hybrid Combined Pelican and BES algorithm (CP-BES). Moreover, the mean value for dataset ISL-CSLTR has achieved the proposed method is 27.3%, 31.6%, 28.7%, 33.4%, 29.8%, 24.5%, 36.2%, and 34.8% greater than the conventional schemes like CNN, FSU-CNN, ASL – CNN, RNN, LSTM +POA, LSTM +ARCHOA, LSTM +HGS, LSTM +PRO and LSTM +BES.

Future-proofing geotechnics workflows: accelerating problem-solving with large language models

ABSTRACT The integration of Large Language Models (LLMs), such as ChatGPT, into the workflows of geotechnical engineering has a high potential to transform how the discipline approaches problem-solving and decision-making. This paper investigates the practical uses of LLMs in addressing geotechnical challenges based on opinions from a diverse group, including students, researchers, and professionals from academia, industry, and government sectors gathered from a workshop dedicated to this study. After introducing the key concepts of LLMs, we present preliminary LLM solutions for four distinct practical geotechnical problems as illustrative examples. In addition to the basic text generation ability, each problem is designed to cover different extended functionalities of LLMs that cannot be achieved by conventional machine learning tools, including multimodal modelling under a unified framework, programming ability, knowledge extraction, and text embedding. We also address the potentials and challenges in implementing LLMs, particularly in achieving high precision and accuracy in specialised tasks, and underscore the need for expert oversight. The findings demonstrate the effectiveness of LLMs in enhancing efficiency, data processing, and decision-making in geotechnical engineering, suggesting a paradigm shift towards more integrated, data-driven approaches in this field.

Communication Protocols and QECC From the Perspective of TQFT, Part II: QECCs as Spacetimes

AbstractTopological quantum field theories (TQFTs) provide a general, minimal‐assumption language for describing quantum‐state preparation and measurement. They therefore provide a general language in which to express multi‐agent communication protocols, e.g., local operations, classical communication (LOCC) protocols. In the accompanying Part I, we construct LOCC protocols using TQFT, and show that LOCC protocols induce quantum error‐correcting codes (QECCs) on the agent‐environment boundary. Such QECCs can be regarded as implementing or inducing the emergence of spacetimes on such boundaries. Here connection between inter‐agent communication and spacetime is investigated, by exploiting different realizations of TQFT. The authors delved into TQFTs that support on their boundaries spin‐networks as computational systems: these are known as topological quantum neural networks (TQNNs). TQNNs, which have a natural representation as tensor networks, implement QECC. The HaPPY code is recognized to be a paradigmatic example. How generic QECCs, as bulk‐boundary codes, induce effective spacetimes is then shown. The effective spatial and temporal separations that take place in QECC enables LOCC protocols between spatially separated observers. The implementation of QECCs in BF and Chern‐Simons theories are then considered, and QECC‐induced spacetimes are shown to provide the classical redundancy required for LOCC. Finally, the topological M‐theory is considered as an implementation of QECC in higher spacetime dimensions.

Communication Protocols and QECCs from the Perspective of TQFT, Part I: Constructing LOCC Protocols and QECCs from TQFTs

AbstractTopological quantum field theories (TQFTs) provide a general, minimal‐assumption language for describing quantum‐state preparation and measurement. They therefore provide a general language in which to express multi‐agent communication protocols, e.g., local operations, classical communication (LOCC) protocols. Here, LOCC protocols are constructed using TQFT and it is shown that LOCC protocols generically induce quantum error‐correcting codes (QECCs). Using multi‐observer scenarios described by quantum Darwinism and Bell/EPR experiments as examples, it is shown how these LOCC‐induced QECCs effectively convert entanglement into classical redundancy. In the accompanying Part II, it is shown that such QECCs can be regarded as implementing, or inducing the emergence of, spacetimes on the boundaries between interacting systems. The connection between inter‐agent communication and spacetime using BF and Chern‐Simons theories, and then using topological M‐theory is investigated.

Unlocking authentic judicial reasoning: A Template-Based Legal Information Generation framework for judicial views

In the field of legal AI, the research on the task of generating judicial views is relatively weak. Judicial views typically include judgment information on the merits such as charges, law articles, and sentencing circumstances. There are several issues with using universal algorithms to generate text, such as a lack of professional expression required for legal documents, and missing or incorrect judgment information on the merits. Judicial views often involve judgment information on the merits, rather than summarizing and excerpting known information. This could be why the application of conventional text generation algorithms may not produce satisfactory results. In actual judicial scenarios, judges generate their views step by step. Firstly, identify multiple judgment information on the merits in the judicial views, and then connect them with legal templates to achieve the generation of judicial views. This article proposes a Template Based Legal Information Generation (TBLIG) framework that generates higher quality judicial views based on the authentic judicial reasoning methods used by judges. The method consists of two steps: The first step involves determining the judgment information on the merits in the judicial view, while the second step involves synchronously generating a legal template based on the case. We have carried out a large number of experiments on the dataset of indictments and non-indictments. Through comparison with other models and ablation experiments, we have proved the effectiveness of our proposed method.

Survey for Detecting AI-generated Content

In large language models (LLMs) field, the rapid advancements have significantly improved text generation, which has blured the distinction between AI-generated and human-written texts. These developments have sparked concerns about potential risks, such as disseminating fake information or engaging in academic cheating. As the responsible use of LLMs becomes imperative, the detection of AI-generated content has become a crucial task. Most existing surveys on AI-generated text (AIGT) Detection have analysed the detection approaches from a computational perspective, with less attention to linguistic aspects. This survey seeks to provide a fresh perspective to drive progress in the area of LLM-generated text detection. Futhermore, in order to make the assessment more explainable, we emphasize the great importence of leveraging specific parameters or metrics to linguistically evaluate the candidate text.

Acceptance of artificial intelligence in education: opportunities, concerns and need for action

The spread of AI text generators such as ChatGPT in education has reached an enormous reach in a short period, which has led to various questions regarding the acceptance of artificial intelligence among teachers and student teachers. This study examines the acceptance of AI among teachers and student teachers. In particular, it considers crucial aspects for planning teaching and teacher training. The results show that despite fundamentally positive attitudes towards AI applications, there are concerns regarding data ethics and legal standards. The correlation between the intention to use AI and trust in AI is significant. The findings should help gain a more comprehensive understanding of the acceptance of AI in the education sector and help teachers plan training and further education accordingly.

AI-generated text in otolaryngology publications: a comparative analysis before and after the release of ChatGPT.

This study delves into the broader implications of artificial intelligence (AI) text generation technologies, including large language models (LLMs) and chatbots, on the scientific literature of otolaryngology. By observing trends in AI-generated text within published otolaryngology studies, this investigation aims to contextualize the impact of AI-driven tools that are reshaping scientific writing and communication. Text from 143 original articles published in JAMA Otolaryngology - Head and Neck Surgery was collected, representing periods before and after ChatGPT's release in November 2022. The text from each article's abstract, introduction, methods, results, and discussion were entered into ZeroGPT.com to estimate the percentage of AI-generated content. Statistical analyses, including T-Tests and Fligner-Killeen's tests, were conducted using R. A significant increase was observed in the mean percentage of AI-generated text post-ChatGPT release, especially in the abstract (from 34.36 to 46.53%, p = 0.004), introduction (from 32.43 to 45.08%, p = 0.010), and discussion sections (from 15.73 to 25.03%, p = 0.015). Publications of authors from non-English speaking countries demonstrated a higher percentage of AI-generated text. This study found that the advent of ChatGPT has significantly impacted writing practices among researchers publishing in JAMA Otolaryngology - Head and Neck Surgery, raising concerns over the accuracy of AI-created content and potential misinformation risks. This manuscript highlights the evolving dynamics between AI technologies, scientific communication, and publication integrity, emphasizing the urgent need for continued research in this dynamic field. The findings also suggest an increasing reliance on AI tools like ChatGPT, raising questions about their broader implications for scientific publishing.