This communication presents preliminary findings from comparing two recent chatbots, OpenAI's ChatGPT and Google's Bard, in the context of fire engineering by evaluating their responses in handling fire safety related queries. A diverse range of fire engineering questions and scenarios were created and examined, including structural fire design, fire prevention strategies, evacuation, building code compliance, and fire suppression systems (some of which resemble those commonly present in the Fire Protection exam (FPE)). The results reveal some key differences in the performance of the chatbots, with ChatGPT demonstrating a relatively superior performance. Then, this communication highlights the potential for chatbot technology to revolutionize fire engineering practices by providing instant access to critical information while outlining areas for further improvement and research. Evidently, and when it matures, this technology will likely be elemental to our engineers' practice and education.
Recent studies of the applications of conversational AI tools, such as chatbots powered by large language models, to complex real-world knowledge work have shown limitations related to reasoning and multi-step problem solving. Specifically, while existing chatbots simulate shallow reasoning and understanding they are prone to errors as problem complexity increases. The failure of these systems to address complex knowledge work is due to the fact that they do not perform any actual cognition. In this position paper, we present Cognitive AI, a higher-level framework for implementing programmatically defined neuro-symbolic cognition above and outside of large language models. Specifically, we propose a dual-layer functional architecture for Cognitive AI that serves as a roadmap for AI systems that can perform complex multi-step knowledge work. We propose that Cognitive AI is a necessary precursor for the evolution of higher forms of AI, such as AGI, and specifically claim that AGI cannot be achieved by probabilistic approaches on their own. We conclude with a discussion of the implications for large language models, adoption cycles in AI, and commercial Cognitive AI development.
Cognitive assistants (CA) are chatbots that provide context-aware support to human workers in knowledge-intensive tasks. Traditionally, cognitive assistants respond in specific ways to predefined user intents and conversation patterns. However, this rigidness does not handle the diversity of natural language well. Recent advances in natural language processing (NLP), powering large language models (LLM) such as GPT-4, Llama2, and Gemini, could enable CAs to converse in a more flexible, human-like manner. However, the additional degrees of freedom may have unforeseen consequences, especially in knowledge-intensive contexts where accuracy is crucial. As a preliminary step to assessing the potential of using LLMs in these contexts, we conducted a user study comparing an LLM-based CA to an intent-based system regarding interaction efficiency, user experience, workload, and usability. This revealed that LLM-based CAs exhibited better user experience, task completion rate, usability, and perceived performance than intent-based systems, suggesting that switching NLP techniques should be investigated further.
We explore a strategy to handle controversial topics in LLM-based chatbots based on Wikipedia's Neutral Point of View (NPOV) principle: acknowledge the absence of a single true answer and surface multiple perspectives. We frame this as retrieval augmented generation, where perspectives are retrieved from a knowledge base and the LLM is tasked with generating a fluent and faithful response from the given perspectives. As a starting point, we use a deterministic retrieval system and then focus on common LLM failure modes that arise during this approach to text generation, namely hallucination and coverage errors. We propose and evaluate three methods to detect such errors based on (1) word-overlap, (2) salience, and (3) LLM-based classifiers. Our results demonstrate that LLM-based classifiers, even when trained only on synthetic errors, achieve high error detection performance, with ROC AUC scores of 95.3% for hallucination and 90.5% for coverage error detection on unambiguous error cases. We show that when no training data is available, our other methods still yield good results on hallucination (84.0%) and coverage error (85.2%) detection.
The advent of ChatGPT has sparked over a year of regulatory frenzy. However, few existing studies have rigorously questioned the assumption that, if left unregulated, AI chatbot's output would inflict tangible, severe real harm on human affairs. Most researchers have overlooked the critical possibility that the information market itself can effectively mitigate these risks and, as a result, they tend to use regulatory tools to address the issue directly. This Article develops a yardstick for reevaluating both AI-related content risks and corresponding regulatory proposals by focusing on inter-informational competition among various outlets. The decades-long history of regulating information and communications technologies indicates that regulators tend to err too much on the side of caution and to put forward excessive regulatory measures when encountering the uncertainties brought about by new technologies. In fact, a trove of empirical evidence has demonstrated that market competition among information outlets can effectively mitigate most risks and that overreliance on regulation is not only unnecessary but detrimental, as well. This Article argues that sufficient competition among chatbots and other information outlets in the information marketplace can sufficiently mitigate and even resolve most content risks posed by generative AI technologies. This renders certain loudly advocated regulatory strategies, like mandatory prohibitions, licensure, curation of datasets, and notice-and-response regimes, truly unnecessary and even toxic to desirable competition and innovation throughout the AI industry. Ultimately, the ideas that I advance in this Article should pour some much-needed cold water on the regulatory frenzy over generative AI and steer the issue back to a rational track.
Identifying scientific publications that are within a dynamic field of research often requires costly annotation by subject-matter experts. Resources like widely-accepted classification criteria or field taxonomies are unavailable for a domain like artificial intelligence (AI), which spans emerging topics and technologies. We address these challenges by inferring a functional definition of AI research from existing expert labels, and then evaluating state-of-the-art chatbot models on the task of expert data annotation. Using the arXiv publication database as ground-truth, we experiment with prompt engineering for GPT chatbot models to identify an alternative, automated expert annotation pipeline that assigns AI labels with 94% accuracy. For comparison, we fine-tune SPECTER, a transformer language model pre-trained on scientific publications, that achieves 96% accuracy (only 2% higher than GPT) on classifying AI publications. Our results indicate that with effective prompt engineering, chatbots can be used as reliable data annotators even where subject-area expertise is required. To evaluate the utility of chatbot-annotated datasets on downstream classification tasks, we train a new classifier on GPT-labeled data and compare its performance to the arXiv-trained model. The classifier trained on GPT-labeled data outperforms the arXiv-trained model by nine percentage points, achieving 82% accuracy.
Large language models (LLMs) have profoundly transformed natural language applications, with a growing reliance on instruction-based definitions for designing chatbots. However, post-deployment the chatbot definitions are fixed and are vulnerable to attacks by malicious users, emphasizing the need to prevent unethical applications and financial losses. Existing studies explore user prompts' impact on LLM-based chatbots, yet practical methods to contain attacks on application-specific chatbots remain unexplored. This paper presents System Prompt Meta Language (SPML), a domain-specific language for refining prompts and monitoring the inputs to the LLM-based chatbots. SPML actively checks attack prompts, ensuring user inputs align with chatbot definitions to prevent malicious execution on the LLM backbone, optimizing costs. It also streamlines chatbot definition crafting with programming language capabilities, overcoming natural language design challenges. Additionally, we introduce a groundbreaking benchmark with 1.8k system prompts and 20k user inputs, offering the inaugural language and benchmark for chatbot definition evaluation. Experiments across datasets demonstrate SPML's proficiency in understanding attacker prompts, surpassing models like GPT-4, GPT-3.5, and LLAMA. Our data and codes are publicly available at: https://prompt-compiler.github.io/SPML/.
Large language models (LLMs) like ChatGPT demonstrate the remarkable progress of artificial intelligence. However, their tendency to hallucinate -- generate plausible but false information -- poses a significant challenge. This issue is critical, as seen in recent court cases where ChatGPT's use led to citations of non-existent legal rulings. This paper explores how Retrieval-Augmented Generation (RAG) can counter hallucinations by integrating external knowledge with prompts. We empirically evaluate RAG against standard LLMs using prompts designed to induce hallucinations. Our results show that RAG increases accuracy in some cases, but can still be misled when prompts directly contradict the model's pre-trained understanding. These findings highlight the complex nature of hallucinations and the need for more robust solutions to ensure LLM reliability in real-world applications. We offer practical recommendations for RAG deployment and discuss implications for the development of more trustworthy LLMs.
Large language models (LLMs) exhibit powerful general intelligence across diverse scenarios, including their integration into chatbots. However, a vital challenge of LLM-based chatbots is that they may produce hallucinated content in responses, which significantly limits their applicability. Various efforts have been made to alleviate hallucination, such as retrieval augmented generation and reinforcement learning with human feedback, but most of them require additional training and data annotation. In this paper, we propose a novel post-hoc Citation-Enhanced Generation (CEG) approach combined with retrieval argumentation. Unlike previous studies that focus on preventing hallucinations during generation, our method addresses this issue in a post-hoc way. It incorporates a retrieval module to search for supporting documents relevant to the generated content, and employs a natural language inference-based citation generation module. Once the statements in the generated content lack of reference, our model can regenerate responses until all statements are supported by citations. Note that our method is a training-free plug-and-play plugin that is capable of various LLMs. Experiments on various hallucination-related datasets show our framework outperforms state-of-the-art methods in both hallucination detection and response regeneration on three benchmarks. Our codes and dataset will be publicly available.
Purpose: To assess the alignment of GPT-4-based evaluation to human clinician experts, for the evaluation of responses to ophthalmology-related patient queries generated by fine-tuned LLM chatbots. Methods: 400 ophthalmology questions and paired answers were created by ophthalmologists to represent commonly asked patient questions, divided into fine-tuning (368; 92%), and testing (40; 8%). We find-tuned 5 different LLMs, including LLAMA2-7b, LLAMA2-7b-Chat, LLAMA2-13b, and LLAMA2-13b-Chat. For the testing dataset, additional 8 glaucoma QnA pairs were included. 200 responses to the testing dataset were generated by 5 fine-tuned LLMs for evaluation. A customized clinical evaluation rubric was used to guide GPT-4 evaluation, grounded on clinical accuracy, relevance, patient safety, and ease of understanding. GPT-4 evaluation was then compared against ranking by 5 clinicians for clinical alignment. Results: Among all fine-tuned LLMs, GPT-3.5 scored the highest (87.1%), followed by LLAMA2-13b (80.9%), LLAMA2-13b-chat (75.5%), LLAMA2-7b-Chat (70%) and LLAMA2-7b (68.8%) based on the GPT-4 evaluation. GPT-4 evaluation demonstrated significant agreement with human clinician rankings, with Spearman and Kendall Tau correlation coefficients of 0.90 and 0.80 respectively; while correlation based on Cohen Kappa was more modest at 0.50. Notably, qualitative analysis and the glaucoma sub-analysis revealed clinical inaccuracies in the LLM-generated responses, which were appropriately identified by the GPT-4 evaluation. Conclusion: The notable clinical alignment of GPT-4 evaluation highlighted its potential to streamline the clinical evaluation of LLM chatbot responses to healthcare-related queries. By complementing the existing clinician-dependent manual grading, this efficient and automated evaluation could assist the validation of future developments in LLM applications for healthcare.