Multi-Model Dialectical Evaluation of LLM Reasoning Chains: A Structured Framework with Dual Scoring Agents

How Can <i>IJDS</i> Authors, Reviewers, and Editors Use (and Misuse) Generative AI?

Large language models (LLMs) have demonstrated remarkable capabilities in text generation, which also raise numerous concerns about their potential misuse, especially in educational exercises and academic writing. Accurately identifying and tracing the origins of LLM-generated content is crucial for accountability and transparency, ensuring the responsible use of LLMs in educational and academic environments. Previous methods utilize binary classifiers to discriminate whether a piece of text was written by a human or generated by a specific LLM or employ multi-class classifiers to trace the source LLM from a fixed set. These methods, however, are restricted to one or several pre-specified LLMs and cannot generalize to new LLMs, which are continually emerging. This study formulates source LLM tracing in a class-incremental learning (CIL) fashion, where new LLMs continually emerge, and a model incrementally learns to identify new LLMs without forgetting old ones. A training-free continual learning method is further devised for the task, the idea of which is to continually extract prototypes for emerging LLMs, using a frozen encoder, and then to perform origin tracing via prototype matching after a delicate decorrelation process. For evaluation, two datasets are constructed, one in English and one in Chinese. These datasets simulate a scenario where six LLMs emerge over time and are used to generate student essays, and an LLM detector has to incrementally expand its recognition scope as new LLMs appear. Experimental results show that the proposed method achieves an average accuracy of 97.04% on the English dataset and 91.23% on the Chinese dataset. These results validate the feasibility of continual origin tracing of LLM-generated text and verify its effectiveness in detecting cheating in student coursework.

Incorporating explicit personas into dialogue models is critical for generating responses that fulfill specific user needs and preferences, creating a more personalized and engaging interaction. Early works on persona-based dialogue generation directly concatenate the persona descriptions and dialogue history into relatively small pre-trained language models (PLMs) for response generation, which leads to uninformative and inferior results due to the sparse persona information and the limited model generation capabilities. Recently, large language models (LLMs) have shown their surprising capabilities in language generation. Prompting the LLMs with the persona descriptions for role-playing dialogue generation has also achieved promising results. However, deploying LLMs is challenging for practical applications due to their large scale, spurring efforts to distill the generation capabilities into more concise and compact models through teacher-student learning. In this article, we propose an efficient compact K nowledge-grounded P ersona-based D ialogue model enhanced by LLM D istillation (KPDD). Specifically, first, we propose to enrich the annotated persona descriptions by integrating external knowledge graphs (KGs) with a mixed encoding network, coupled with a mixture of experts (MoE) module for both informative and diverse response generation. The mixed encoding network contains multiple layers of modality interaction operations, enabling information from both modalities propagates to the other. Second, to fully exploit the generation capabilities of LLMs, we turn to the distillation technique to improve the generation capabilities of our model, facilitated by a natural language inference (NLI)-based filtering mechanism to extract high-quality information from LLMs. In addition, we employ a curriculum learning strategy to train our model on the high-quality filtered distilled data and progressively on the relatively noisy original data, enhancing its adaptability and performance. Extensive experiments show that KPDD outperforms state-of-the-art baselines in terms of both automatic and human evaluation.

Are large language models qualified reviewers in originality evaluation?

A survey on large language model (LLM) security and privacy: The Good, The Bad, and The Ugly

Understanding and interpreting code is a crucial task in intelligent software engineering, aiding developers and users in adjusting code for correctness and robustness. The emergence of large language models (LLMs) provides new perspectives for code interpretation tasks. However, current LLM-based code interpretation remains restricted to limited dimensions, lacks a unified evaluation standard, and is missing a comprehensive and systematic assessment methodology. To address this issue, this paper proposes an LLM code understanding evaluation method based on a multi-granularity voting mechanism, aiming to systematically investigate and analyze LLMs' performance in code interpretation tasks. First, we carefully select code snippets from open-source GitHub projects and preprocess them for LLM analysis. Second, we use identical prompts and inputs to test three popular LLMs, recording their output. During this process, we apply prompt engineering techniques to specific target code snippets and conduct repeated experiments to explore the impact of prompt engineering on LLM-generated code explanations. Next, we design evaluation metrics to quantify the LLM outputs and assess their effectiveness based on the obtained scores. Experimental results demonstrate significant differences in code analysis and generation capabilities among the evaluated general-purpose LLMs from different vendors when given identical prompts and inputs. When multiple dimensions are considered in evaluating the generated content, different LLMs exhibit varying strengths in different aspects. Additionally, applying specific prompt engineering techniques can moderate the discrepancies in code analysis and generation capabilities among different LLMs.

A Conversational Information Retrieval (CIR) system can be defined as an information retrieval (IR) system characterized by a conversational interface that facilitates user interaction with the system to obtain information through multi-turn dialogues in natural language, whether in spoken or written modalities. Information Retrieval (IR) has undergone considerable transformation, transcending conventional search methodologies to address a wide array of user information requirements. IR models, LLMs, and human users establishes a novel technical paradigm that is significantly more effective for information seeking. IR models deliver timely and pertinent information, LLMs supply intrinsic knowledge, and humans assume a pivotal role as both demanders and assessors of the reliability of information services. Large Language Models (LLMs) have exhibited remarkable proficiency in text comprehension, generation, and knowledge inference, thereby unveiling promising prospects for research within the eld of IR. LLMs not only enhance the process of generative retrieval but also provide superior frameworks for user comprehension, model assessment, and user-system engagement. However, substantial challenges persist, encompassing computational expenses, issues of credibility, limitations specific to certain domains, and ethical implications. The brisk and extraordinary progress made in the eld of Large Language Models (LLMs) has dramatically rede ned the framework of natural language processing, paving the way for the advent of ever more elaborate and sophisticated conversational search functionalities that had previously seemed impossible. Large Language Models (LLMs) have shown an exceptionally proficiency in various essential areas such as text comprehension, text generation, and the inference of knowledge, which consequently opens up a plethora of promising opportunities for further research and exploration within the discipline of Information Retrieval (IR). Moreover, LLMs not only significantly enhance the mechanisms involved in generative retrieval processes but also offer advanced and superior frameworks that contribute to improved user comprehension, comprehensive model assessment, and enriched engagement between users and systems. The effective capture and interpretation of user intent in complex contextual search scenarios continues to represent a substantial and critical challenge that must be addressed in order to optimize these interactions. Through extensive experimentation and evaluation, we demonstrate the effectiveness of the proposed framework in improving search relevance, user satisfaction, and interaction efficiency.

Machine translation, which uses computers to translate one language into another, is one of the most challenging tasks in artificial intelligence. During the last decade, neural machine translation (NMT), which builds translation models based on deep neural networks, has achieved significant improvement. However, NMT still faces several challenges. For example, the translation quality of an NMT system greatly depends on the amount of bilingual training data, which is expensive to acquire. Furthermore, it is difficult to incorporate external knowledge into an NMT system to obtain further improvement for a specific domain. Recently, large language models (LLMs) have demonstrated remarkable capabilities in language understanding and generation. This raises interesting questions about whether LLMs can be good translators and whether it is easy to adapt LLMs to new domains or to meet specific requirements. In this study, we hypothesized that LLMs can be adapted to perform translation by using prompts or fine-tuning and these adapted LLMs would outperform the conventional NMT model in four aspects: translation quality, interactive ability, knowledge incorporation ability, and domain adaptation. We compared GPT-4 and Google Translate, the representative LLM and NMT models, respectively, on the WMT 2019 (Fourth conference on machine translation) dataset. Experimental results showed that GPT-4 outperformed Google Translate in the above four aspects by exploiting appropriate prompts. Further experiments on Llama, an open-source LLM developed by Meta, showed that the translation quality of LLMs can be further improved by fine-tuning on limited language-related bilingual corpus, demonstrating strong adaptation abilities of LLMs.

Recent advancements in Artificial Intelligence (AI), particularly in the advanced machine learning for the Natural Language Processing (NLP) paradigm, have led to the development of powerful Large Language Models (LLMs) capable of impressive feats in tasks like translation, text summarisation, text generation and code generation. However, a critical challenge hindering their real-world deployment is their susceptibility to hallucinations, where they generate plausible looking but factually incorrect outputs. These limitations come with adverse effects, such as the propagation of misinformation and reducing user trustworthiness in the related technologies, even when they possess transformative potential in various sectors. This study aims to enhance the performance of LLMs by presenting a new strategy that combines grammar-aware prompt engineering (GAPE) and formal methods (FMs) to leverage their synergy in the LLM process logic. We argue that by combining linguistic principles using GAPE and constructing a basis of formal structures using FMs, we could improve the LLM's ability to analyse language, decrease ambiguity in prompts, improve consistency in output, and eventually, greatly diminish LLM hallucinations. To do this, we propose a collaboration between linguists and AI experts while also providing specialised training for LLMs that emphasises linguistic precision. Additionally, we suggest implementing iterative design and development procedures for LLMs that use GAPE and FM principles to continuously enhance the performance of LLMs. By following these techniques, we may create a future in which LLMs are more trustworthy for a wide range of users and use cases with reliable LLM technologies and efficient advancements in practical situations

The growing interest in advanced large language models (LLMs) like ChatGPT has sparked debate about how best to use them in various human activities. However, a neglected issue in the debate concerning the applications of LLMs is whether they can reason logically and follow rules in novel contexts, which are critical for our understanding and applications of LLMs. To address this knowledge gap, this study investigates five LLMs (ChatGPT-4o, Claude, Gemini, Meta AI, and Mistral) using word ladder puzzles to assess their logical reasoning and rule-adherence capabilities. Our two-phase methodology involves (1) explicit instructions about word ladder puzzles and rules regarding how to solve the puzzles and then evaluate rule understanding, followed by (2) assessing LLMs’ ability to create and solve word ladder puzzles while adhering to rules. Additionally, we test their ability to implicitly recognize and avoid HIPAA privacy rule violations as an example of a real-world scenario. Our findings reveal that LLMs show a persistent lack of logical reasoning and systematically fail to follow puzzle rules. Furthermore, all LLMs except Claude prioritized task completion (text writing) over ethical considerations in the HIPAA test. Our findings expose critical flaws in LLMs’ reasoning and rule-following capabilities, raising concerns about their reliability in critical tasks requiring strict rule-following and logical reasoning. Therefore, we urge caution when integrating LLMs into critical fields and highlight the need for further research into their capabilities and limitations to ensure responsible AI development.

Recently, Large Language Models (LLMs) have transformed information retrieval, becoming widely adopted across various domains due to their ability to process extensive textual data and generate diverse insights. Biodiversity literature, with its broad range of topics, is no exception to this trend (Boyko et al. 2023, Castro et al. 2024). LLMs can help in information extraction and synthesis, text annotation and classification, and many other natural language processing tasks. We leverage LLMs to automate the information retrieval task from biodiversity publications, building upon data sourced from our previous work (Ahmed et al. 2024). In our previous work (Ahmed et al. 2023, Ahmed et al. 2024), we assessed the reproducibility of deep learning (DL) methods used in biodiversity research. We developed a manual pipeline to extract key information on DL pipelines—dataset, source code, open-source frameworks, model architecture, hyperparameters, software and hardware specs, randomness, averaging result and evaluation metrics from 61 publications (Ahmed et al. 2024). While this allowed analysis, it required extensive manual effort by domain experts, limiting scalability. To address this, we propose an automatic information extraction pipeline using LLMs with the Retrieval Augmented Generation (RAG) technique. RAG combines the retrieval of relevant documents with the generative capabilities of LLMs to enhance the quality and relevance of the extracted information. We employed an open-source LLM, Hugging Face implementation of Mixtral 8x7B (Jiang et al. 2024), a mixture of expert models in our pipeline (Fig. 1) and adapted the RAG pipeline from earlier work (Kommineni et al. 2024). The pipeline was run on a single NVIDIA A100 40GB graphics processing unit with 4-bit quantization. To evaluate our pipeline, we compared the expert-assisted manual approach with the LLM-assisted automatic approach. We measured their consistency using the inter-annotator agreement (IAA) and quantified it with the Cohen Kappa score (Pedregosa et al. 2011), where a higher score indicates more reliable and aligned outputs (1: maximum agreement, -1: no agreement). The Kappa score among human experts (annotators 1 and 2) was 0.54 (moderate agreement), while the scores comparing human experts with the LLM were 0.16 and 0.12 (slight agreement). The difference is partly due to human annotators having access to more information (including code, dataset, figures, tables and supplementary materials) than the LLM, which was restricted to the text itself. Given these restrictions, the results are promising but also show the potential to improve them by adding further modalities to the LLM inputs. Future work will involve several key improvements to our LLM-assisted information retrieval pipeline: Incorporating multimodal data (e.g., figures, tables, code, etc.) as input to the LLM, alongside text, to enhance the accuracy and comprehensiveness of the information retrieved from publications. Optimizing the retrieval component of the RAG framework with advanced techniques like semantic search, hybrid search or relevance feedback can improve the quality of outputs. Expanding the evaluation to a larger corpus of biodiversity literature could provide a more comprehensive understanding of pipeline capabilities, and this paves the way for pipeline optimization. A human-in-the-loop approach for evaluating the LLM-generated outputs by matching the ground truth values from the respective publications, will increase the quality of the overall pipeline. Employing more metrics for the evaluation beyond the Cohen Kappa score to better understand the LLM-assisted outputs. Incorporating multimodal data (e.g., figures, tables, code, etc.) as input to the LLM, alongside text, to enhance the accuracy and comprehensiveness of the information retrieved from publications. Optimizing the retrieval component of the RAG framework with advanced techniques like semantic search, hybrid search or relevance feedback can improve the quality of outputs. Expanding the evaluation to a larger corpus of biodiversity literature could provide a more comprehensive understanding of pipeline capabilities, and this paves the way for pipeline optimization. A human-in-the-loop approach for evaluating the LLM-generated outputs by matching the ground truth values from the respective publications, will increase the quality of the overall pipeline. Employing more metrics for the evaluation beyond the Cohen Kappa score to better understand the LLM-assisted outputs. Leveraging LLMs to automate information retrieval from biodiversity publications signifies a notable advancement in the scalable and efficient analysis of biodiversity literature. Initial results show promise, yet there is substantial potential for enhancement through the integration of multimodal data, optimized retrieval mechanisms, and comprehensive evaluation. By addressing these areas, we aim to improve the accuracy and utility of our pipeline, ultimately enabling broader and more in-depth analysis of biodiversity literature.

The integration of large language models (LLMs) into clinical nutrition marks a transformative advancement, offering promising solutions for enhancing patient care, personalizing dietary recommendations, and supporting evidence-based clinical decision-making. Trained on extensive text corpora and powered by transformer-based architectures, LLMs demonstrate remarkable capabilities in natural language understanding and generation. This review provides an overview of their current and potential applications in clinical nutrition, focusing on key technologies including prompt engineering, fine-tuning, retrieval-augmented generation, and multimodal integration. These enhancements increase domain relevance, factual accuracy, and contextual responsiveness, enabling LLMs to deliver more reliable outputs in nutrition-related tasks. Recent studies have shown LLMs’ utility in dietary planning, nutritional education, obesity management, and malnutrition risk assessment. Despite these advances, challenges remain. Limitations in reasoning, factual accuracy, and domain specificity, along with risks of bias and hallucination, underscore the need for rigorous validation and human oversight. Furthermore, ethical considerations, environmental costs, and infrastructural integration must be addressed before widespread adoption. Future directions include combining LLMs with predictive analytics, integrating them with electronic health records and wearables, and adapting them for multilingual, culturally sensitive dietary guidance. LLMs also hold potential as research and educational tools, assisting in literature synthesis and patient engagement. Their transformative promise depends on cross-disciplinary collaboration, responsible deployment, and clinician training. Ultimately, while LLMs are not a replacement for healthcare professionals, they offer powerful augmentation tools for delivering scalable, personalized, and data-driven nutritional care in an increasingly complex healthcare environment.

A lot of hype has accompanied the increasing number of generative artificial intelligence-powered large language models (LLMs). Similarly, much has been written about what currently available LLMs can and cannot do, including their benefits and risks, especially in higher education. However, few use cases have investigated the performance and generative capabilities of LLMs in low-resource languages. With this in mind, one of the purposes of the current study was to explore the extent to which seven, currently available, free-to-use versions of LLMs (ChatGPT, Claude, Copilot, Gemini, GroqChat, Perplexity, and YouChat) perform in five low-resource languages (isiZulu, Sesotho, Yoruba, M?ori, and Mi’kmaq) in their generative multilingual capabilities. Employing a common input prompt, in which the only change was to insert the name of a given low-resource language and English in each case, this study collected its datasets by inputting this common prompt into the seven LLMs. Three of the findings of this study are noteworthy. First, the seven LLMs displayed a significant lack of generative multilingual capabilities in the five low-resource languages. Second, they hallucinated and produced nonsensical, meaningless, and irrelevant responses in their low-resource language outputs. Third, their English responses were far better in quality, relevance, depth, detail, and nuance than their low-resource language only and English responses for the five low-resource languages. The paper ends by offering the implications and making the conclusions of the study in terms of LLMs’ generative capabilities in low-resource languages.

Large Language Models (LLMs) have garnered increasing attention in neurosurgery and possess significant potential to improve the field. However, the breadth and performance of LLMs across diverse neurosurgical tasks have not been systematically examined, and LLMs come with their own challenges and unique terminology. We seek to identify key models, establish reporting guidelines for replicability, and highlight progress in key application areas of LLM use in the neurosurgical literature. We searched PubMed and Google Scholar using terms related to LLMs and neurosurgery ("large language model" OR "LLM" OR "ChatGPT" OR "GPT-3" OR "GPT3" OR "GPT-3.5" OR "GPT3.5" OR "GPT-4" OR "GPT4" OR "LLAMA" OR "MISTRAL" OR "BARD") AND "neurosurgery". The final set of articles was reviewed for publication year, application area, specific LLM(s) used, control/comparison groups used to evaluate LLM performance, whether the article reported specific LLM prompts, prompting strategy types used, whether the LLM query could be reproduced in its entirety (including both the prompt used and any adjoining data), measures of hallucination, and reported performance measures. Fifty-one articles met inclusion criteria, and were categorized into six application areas, with the most common being Generation of Text for Direct Clinical Use (n = 14, 27.5%), Answering Standardized Exam Questions (n = 12, 23.5%), and Clinical Judgement and Decision-Making Support (n = 11, 21.6%). The most frequently used LLMs were GPT-3.5 (n = 30, 58.8%), GPT-4 (n = 20, 39.2%), Bard (n = 9, 17.6%), and Bing (n = 6, 11.8%). Most studies (n = 43, 84.3%) used LLMs directly out-of-the-box, while 8 studies (15.7%) conducted advanced pre-training or fine-tuning. Large language models show advanced capabilities in complex tasks and hold potential to transform neurosurgery. However, research typically addresses basic applications and overlooks enhancing LLM performance, facing reproducibility issues. Standardizing detailed reporting, considering LLM stochasticity, and using advanced methods beyond basic validation are essential for progress.

Large Language Models (LLMs) excel in natural language-relevant tasks like text generation and question answering Q&A. To further expand their application, efforts focus on enabling LLMs to utilize real-world tools. However, their tool-use ability in professional GIS remains under explored due to two main challenges. Firstly, LLMs are usually trained on general-domain corpora, lacking sufficient and comprehensive GIS-specific data to align with professional knowledge, including understanding the functions of GIS tools. Secondly, researchers often need to combine multiple GIS tools to solve geospatial tasks. To address these challenges, we propose a trainable method to enable LLMs to master GIS tools. We curated a comprehensive set of resources: instruction-response data (GeoTool, 1950 instructions) to enhance the understanding of LLMs for GIS tools, instruction-solution data (GeoSolution, 3645 instructions) to improve their ability to generate tool-use solutions for geospatial tasks, and annotated instruction-solution evaluation data (GeoTask, 300 instructions) for evaluating LLMs’ GIS tool-use proficiency. Using the collected training data (GeoTool and GeoSolution), we fine-tuned a professional-domain LLM called GeoTool-GPT based on an open-source general-domain LLM, the LLaMA-2-7b model. The experiment based on evaluation data validates our method’s effectiveness in enhancing the tool-use ability of general-domain LLMs in the professional GIS domain, with the performance of our model closely approaching that of GPT-4.