The emergence of large language models (LLMs) further improves the capabilities of open-domain dialogue systems and can generate fluent, coherent, and diverse responses. However, LLMs still lack a crucial ability: communication skills. This limitation renders them more like information seeking tools rather than anthropomorphic chatbots. Communication skills, such as topic transition, proactively asking questions, concept guidance, empathy, and summarising often should be taken into consideration, to make LLMs more anthropomorphic and proactive during the conversation, thereby increasing the interest of users and attracting them to chat for longer. However, enabling these communication skills in black-box LLMs remains a key challenge because they do not have the same utterance formation mode as real people: think before speaking. Inspired by linguistics and cognitive science, we empower LLMs with communication skills through inner monologues. To evaluate various communication skills, we construct a benchmark named Cskills, which can also more comprehensively evaluate the dialogue generation ability of the model. Experimental results show that the proposed CSIM strategy improves the backbone models and outperforms the baselines.
Two ways has been discussed to unlock the reasoning capability of a large language model. The first one is prompt engineering and the second one is to combine the multiple inferences of large language models, or the multi-agent discussion. Theoretically, this paper justifies the multi-agent discussion mechanisms from the symmetry of agents. Empirically, this paper reports the empirical results of the interplay of prompts and discussion mechanisms, revealing the empirical state-of-the-art performance of complex multi-agent mechanisms can be approached by carefully developed prompt engineering. This paper also proposes a scalable discussion mechanism based on conquer and merge, providing a simple multi-agent discussion solution with simple prompts but state-of-the-art performance.
Prompt engineering is a challenging yet crucial task for optimizing the performance of large language models on customized tasks. It requires complex reasoning to examine the model's errors, hypothesize what is missing or misleading in the current prompt, and communicate the task with clarity. While recent works indicate that large language models can be meta-prompted to perform automatic prompt engineering, we argue that their potential is limited due to insufficient guidance for complex reasoning in the meta-prompt. We fill this gap by infusing into the meta-prompt three key components: detailed descriptions, context specification, and a step-by-step reasoning template. The resulting method, named PE2, exhibits remarkable versatility across diverse language tasks. It finds prompts that outperform "let's think step by step" by 6.3% on MultiArith and 3.1% on GSM8K, and outperforms competitive baselines on counterfactual tasks by 6.9%. Further, we show that PE2 can make targeted and highly specific prompt edits, rectify erroneous prompts, and induce multi-step plans for complex tasks.
This paper reports on the use of prompt engineering and GPT-3.5 for biomedical query-focused multi-document summarisation. Using GPT-3.5 and appropriate prompts, our system achieves top ROUGE-F1 results in the task of obtaining short-paragraph-sized answers to biomedical questions in the 2023 BioASQ Challenge (BioASQ 11b). This paper confirms what has been observed in other domains: 1) Prompts that incorporated few-shot samples generally improved on their counterpart zero-shot variants; 2) The largest improvement was achieved by retrieval augmented generation. The fact that these prompts allow our top runs to rank within the top two runs of BioASQ 11b demonstrate the power of using adequate prompts for Large Language Models in general, and GPT-3.5 in particular, for query-focused summarisation.
Identifying contextual integrity (CI) and governing knowledge commons (GKC) parameters in privacy policy texts can facilitate normative privacy analysis. However, GKC-CI annotation has heretofore required manual or crowdsourced effort. This paper demonstrates that high-accuracy GKC-CI parameter annotation of privacy policies can be performed automatically using large language models. We fine-tune 50 open-source and proprietary models on 21,588 ground truth GKC-CI annotations from 16 privacy policies. Our best performing model has an accuracy of 90.65%, which is comparable to the accuracy of experts on the same task. We apply our best performing model to 456 privacy policies from a variety of online services, demonstrating the effectiveness of scaling GKC-CI annotation for privacy policy exploration and analysis. We publicly release our model training code, training and testing data, an annotation visualizer, and all annotated policies for future GKC-CI research.
Online forums encourage the exchange and discussion of different stances on many topics. Not only do they provide an opportunity to present one's own arguments, but may also gather a broad cross-section of others' arguments. However, the resulting long discussions are difficult to overview. This paper presents a novel unsupervised approach using large language models (LLMs) to generating indicative summaries for long discussions that basically serve as tables of contents. Our approach first clusters argument sentences, generates cluster labels as abstractive summaries, and classifies the generated cluster labels into argumentation frames resulting in a two-level summary. Based on an extensively optimized prompt engineering approach, we evaluate 19~LLMs for generative cluster labeling and frame classification. To evaluate the usefulness of our indicative summaries, we conduct a purpose-driven user study via a new visual interface called Discussion Explorer: It shows that our proposed indicative summaries serve as a convenient navigation tool to explore long discussions.
Neural text detectors are models trained to detect whether a given text was generated by a language model or written by a human. In this paper, we investigate three simple and resource-efficient strategies (parameter tweaking, prompt engineering, and character-level mutations) to alter texts generated by GPT-3.5 that are unsuspicious or unnoticeable for humans but cause misclassification by neural text detectors. The results show that especially parameter tweaking and character-level mutations are effective strategies.
Recent studies have demonstrated the great potential of Large Language Models (LLMs) serving as zero-shot relevance rankers. The typical approach involves making comparisons between pairs or lists of documents. Although effective, these listwise and pairwise methods are not efficient and also heavily rely on intricate prompt engineering. To tackle this problem, we introduce a novel instruction distillation method. The key idea is to distill the pairwise ranking ability of open-sourced LLMs to a simpler but more efficient pointwise ranking. Specifically, given the same LLM, we first rank documents using the effective pairwise approach with complex instructions, and then distill the teacher predictions to the pointwise approach with simpler instructions. Evaluation results on the BEIR, TREC, and ReDial datasets demonstrate that instruction distillation can improve efficiency by 10 to 100x and also enhance the ranking performance of LLMs. Furthermore, our approach surpasses the performance of existing supervised methods like monoT5 and is on par with the state-of-the-art zero-shot methods. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.
Recent advances in prompt engineering enable large language models (LLMs) to solve multi-hop logical reasoning problems with impressive accuracy. However, there is little existing work investigating the robustness of LLMs with few-shot prompting techniques. Therefore, we introduce a systematic approach to test the robustness of LLMs in multi-hop reasoning tasks via domain-agnostic perturbations. We include perturbations at multiple levels of abstractions (e.g. lexical perturbations such as typos, and semantic perturbations such as the inclusion of intermediate reasoning steps in the questions) to conduct behavioral analysis on the LLMs. Throughout our experiments, we find that models are more sensitive to certain perturbations such as replacing words with their synonyms. We also demonstrate that increasing the proportion of perturbed exemplars in the prompts improves the robustness of few-shot prompting methods.
Large language models (LLMs) have demonstrated their potential in social science research by emulating human perceptions and behaviors, a concept referred to as algorithmic fidelity. This study assesses the algorithmic fidelity and bias of LLMs by utilizing two nationally representative climate change surveys. The LLMs were conditioned on demographics and/or psychological covariates to simulate survey responses. The findings indicate that LLMs can effectively capture presidential voting behaviors but encounter challenges in accurately representing global warming perspectives when relevant covariates are not included. GPT-4 exhibits improved performance when conditioned on both demographics and covariates. However, disparities emerge in LLM estimations of the views of certain groups, with LLMs tending to underestimate worry about global warming among Black Americans. While highlighting the potential of LLMs to aid social science research, these results underscore the importance of meticulous conditioning, model selection, survey question format, and bias assessment when employing LLMs for survey simulation. Further investigation into prompt engineering and algorithm auditing is essential to harness the power of LLMs while addressing their inherent limitations.
Although large language models (LLMs) have advanced the state-of-the-art in NLP significantly, deploying them for downstream applications is still challenging due to cost, responsiveness, control, or concerns around privacy and security. As such, trainable models are still the preferred option in some cases. However, these models still require human-labeled data for optimal performance, which is expensive and time-consuming to obtain. In order to address this issue, several techniques to reduce human effort involve labeling or generating data using LLMs. Although these methods are effective for certain applications, in practice they encounter difficulties in real-world scenarios. Labeling data requires careful data selection, while generating data necessitates task-specific prompt engineering. In this paper, we propose a unified data creation pipeline that requires only a single formatting example, and which is applicable to a broad range of tasks, including traditionally problematic ones with semantically devoid label spaces. In our experiments we demonstrate that instruction-following LLMs are highly cost-effective data creators, and that models trained with these data exhibit performance better than those trained with human-labeled data (by up to 17.5%) on out-of-distribution evaluation, while maintaining comparable performance on in-distribution tasks. These results have important implications for the robustness of NLP systems deployed in the real-world.
People with blindness and low vision (pBLV) encounter substantial challenges when it comes to comprehensive scene recognition and precise object identification in unfamiliar environments. Additionally, due to the vision loss, pBLV have difficulty in accessing and identifying potential tripping hazards on their own. In this paper, we present a pioneering approach that leverages a large vision-language model to enhance visual perception for pBLV, offering detailed and comprehensive descriptions of the surrounding environments and providing warnings about the potential risks. Our method begins by leveraging a large image tagging model (i.e., Recognize Anything (RAM)) to identify all common objects present in the captured images. The recognition results and user query are then integrated into a prompt, tailored specifically for pBLV using prompt engineering. By combining the prompt and input image, a large vision-language model (i.e., InstructBLIP) generates detailed and comprehensive descriptions of the environment and identifies potential risks in the environment by analyzing the environmental objects and scenes, relevant to the prompt. We evaluate our approach through experiments conducted on both indoor and outdoor datasets. Our results demonstrate that our method is able to recognize objects accurately and provide insightful descriptions and analysis of the environment for pBLV.
The springing up of Large Language Models (LLMs) has shifted the community from single-task-orientated natural language processing (NLP) research to a holistic end-to-end multi-task learning paradigm. Along this line of research endeavors in the area, LLM-based prompting methods have attracted much attention, partially due to the technological advantages brought by prompt engineering (PE) as well as the underlying NLP principles disclosed by various prompting methods. Traditional supervised learning usually requires training a model based on labeled data and then making predictions. In contrast, PE methods directly use the powerful capabilities of existing LLMs (i.e., GPT-3 and GPT-4) via composing appropriate prompts, especially under few-shot or zero-shot scenarios. Facing the abundance of studies related to the prompting and the ever-evolving nature of this field, this article aims to (i) illustrate a novel perspective to review existing PE methods, within the well-established communication theory framework; (ii) facilitate a better/deeper understanding of developing trends of existing PE methods used in four typical tasks; (iii) shed light on promising research directions for future PE methods.
To address prevalent issues in medical imaging, such as data acquisition challenges and label availability, transfer learning from natural to medical image domains serves as a viable strategy to produce reliable segmentation results. However, several existing barriers between domains need to be broken down, including addressing contrast discrepancies, managing anatomical variability, and adapting 2D pretrained models for 3D segmentation tasks. In this paper, we propose ProMISe,a prompt-driven 3D medical image segmentation model using only a single point prompt to leverage knowledge from a pretrained 2D image foundation model. In particular, we use the pretrained vision transformer from the Segment Anything Model (SAM) and integrate lightweight adapters to extract depth-related (3D) spatial context without updating the pretrained weights. For robust results, a hybrid network with complementary encoders is designed, and a boundary-aware loss is proposed to achieve precise boundaries. We evaluate our model on two public datasets for colon and pancreas tumor segmentations, respectively. Compared to the state-of-the-art segmentation methods with and without prompt engineering, our proposed method achieves superior performance. The code is publicly available at https://github.com/MedICL-VU/ProMISe.
The advanced capabilities of Large Language Models (LLMs) have made them invaluable across various applications, from conversational agents and content creation to data analysis, research, and innovation. However, their effectiveness and accessibility also render them susceptible to abuse for generating malicious content, including phishing attacks. This study explores the potential of using four popular commercially available LLMs, i.e., ChatGPT (GPT 3.5 Turbo), GPT 4, Claude, and Bard, to generate functional phishing attacks using a series of malicious prompts. We discover that these LLMs can generate both phishing websites and emails that can convincingly imitate well-known brands and also deploy a range of evasive tactics that are used to elude detection mechanisms employed by anti-phishing systems. These attacks can be generated using unmodified or "vanilla" versions of these LLMs without requiring any prior adversarial exploits such as jailbreaking. We evaluate the performance of the LLMs towards generating these attacks and find that they can also be utilized to create malicious prompts that, in turn, can be fed back to the model to generate phishing scams - thus massively reducing the prompt-engineering effort required by attackers to scale these threats. As a countermeasure, we build a BERT-based automated detection tool that can be used for the early detection of malicious prompts to prevent LLMs from generating phishing content. Our model is transferable across all four commercial LLMs, attaining an average accuracy of 96% for phishing website prompts and 94% for phishing email prompts. We also disclose the vulnerabilities to the concerned LLMs, with Google acknowledging it as a severe issue. Our detection model is available for use at Hugging Face, as well as a ChatGPT Actions plugin.
Question generation has numerous applications in the educational context. Question generation can prove helpful for students when reviewing content and testing themselves. Furthermore, a question generation model can aid teachers by lessening the burden of creating assessments and other practice material. This paper aims to find the best method to generate questions from textual data through a transformer model and prompt engineering. In this research, we finetuned a pretrained distilBERT model on the SQuAD question answering dataset to generate questions. In addition to training a transformer model, prompt engineering was applied to generate questions effectively using the LLaMA model. The generated questions were compared against the baseline questions in the SQuAD dataset to evaluate the effectiveness of four different prompts. All four prompts demonstrated over 60% similarity on average. Of the prompt-generated questions, 30% achieved a high similarity score greater than 70%.
Open-ended questions, which require students to produce multi-word, nontrivial responses, are a popular tool for formative assessment as they provide more specific insights into what students do and don't know. However, grading open-ended questions can be time-consuming leading teachers to resort to simpler question formats or conduct fewer formative assessments. While there has been a longstanding interest in automating of short-answer grading (ASAG), but previous approaches have been technically complex, limiting their use in formative assessment contexts. The newest generation of Large Language Models (LLMs) potentially makes grading short answer questions more feasible. This paper investigates the potential for the newest version of LLMs to be used in ASAG, specifically in the grading of short answer questions for formative assessments, in two ways. First, it introduces a novel dataset of short answer reading comprehension questions, drawn from a set of reading assessments conducted with over 150 students in Ghana. This dataset allows for the evaluation of LLMs in a new context, as they are predominantly designed and trained on data from high-income North American countries. Second, the paper empirically evaluates how well various configurations of generative LLMs grade student short answer responses compared to expert human raters. The findings show that GPT-4, with minimal prompt engineering, performed extremely well on grading the novel dataset (QWK 0.92, F1 0.89), reaching near parity with expert human raters. To our knowledge this work is the first to empirically evaluate the performance of generative LLMs on short answer reading comprehension questions using real student data, with low technical hurdles to attaining this performance. These findings suggest that generative LLMs could be used to grade formative literacy assessment tasks.
We propose a modular framework that leverages the expertise of different foundation models over different modalities and domains in order to perform a single, complex, multi-modal task, without relying on prompt engineering or otherwise tailor-made multi-modal training. Our approach enables decentralized command execution and allows each model to both contribute and benefit from the expertise of the other models. Our method can be extended to a variety of foundation models (including audio and vision), above and beyond only language models, as it does not depend on prompts. We demonstrate our approach on two tasks. On the well-known task of stylized image captioning, our experiments show that our approach outperforms semi-supervised state-of-the-art models, while being zero-shot and avoiding costly training, data collection, and prompt engineering. We further demonstrate this method on a novel task, audio-aware image captioning, in which an image and audio are given and the task is to generate text that describes the image within the context of the provided audio. Our code is available on GitHub.
Purpose: This study aimed to develop an open-source multimodal large language model (CXR-LLAVA) for interpreting chest X-ray images (CXRs), leveraging recent advances in large language models (LLMs) to potentially replicate the image interpretation skills of human radiologists Materials and Methods: For training, we collected 592,580 publicly available CXRs, of which 374,881 had labels for certain radiographic abnormalities (Dataset 1) and 217,699 provided free-text radiology reports (Dataset 2). After pre-training a vision transformer with Dataset 1, we integrated it with an LLM influenced by the LLAVA network. Then, the model was fine-tuned, primarily using Dataset 2. The model's diagnostic performance for major pathological findings was evaluated, along with the acceptability of radiologic reports by human radiologists, to gauge its potential for autonomous reporting. Results: The model demonstrated impressive performance in test sets, achieving an average F1 score of 0.81 for six major pathological findings in the MIMIC internal test set and 0.62 for seven major pathological findings in the external test set. The model's F1 scores surpassed those of GPT-4-vision and Gemini-Pro-Vision in both test sets. In human radiologist evaluations of the external test set, the model achieved a 72.7% success rate in autonomous reporting, slightly below the 84.0% rate of ground truth reports. Conclusion: This study highlights the significant potential of multimodal LLMs for CXR interpretation, while also acknowledging the performance limitations. Despite these challenges, we believe that making our model open-source will catalyze further research, expanding its effectiveness and applicability in various clinical contexts. CXR-LLAVA is available at https://github.com/ECOFRI/CXR_LLAVA.
Large language models (LLMs) offer unprecedented text completion capabilities. As general models, they can fulfill a wide range of roles, including those of more specialized models. We assess the performance of GPT-4 and GPT-3.5 in zero shot, few shot and fine-tuned settings on the aspect-based sentiment analysis (ABSA) task. Fine-tuned GPT-3.5 achieves a state-of-the-art F1 score of 83.8 on the joint aspect term extraction and polarity classification task of the SemEval-2014 Task 4, improving upon InstructABSA [@scaria_instructabsa_2023] by 5.7%. However, this comes at the price of 1000 times more model parameters and thus increased inference cost. We discuss the the cost-performance trade-offs of different models, and analyze the typical errors that they make. Our results also indicate that detailed prompts improve performance in zero-shot and few-shot settings but are not necessary for fine-tuned models. This evidence is relevant for practioners that are faced with the choice of prompt engineering versus fine-tuning when using LLMs for ABSA.