The rapid advancement of large language models has revolutionized various applications but also raised crucial concerns about their potential to perpetuate biases and unfairness when deployed in social media contexts. Evaluating LLMs' potential biases and fairness has become crucial, as existing methods rely on limited prompts focusing on just a few groups, lacking a comprehensive categorical perspective. In this paper, we propose evaluating LLM biases from a group fairness lens using a novel hierarchical schema characterizing diverse social groups. Specifically, we construct a dataset, GFair, encapsulating target-attribute combinations across multiple dimensions. In addition, we introduce statement organization, a new open-ended text generation task, to uncover complex biases in LLMs. Extensive evaluations of popular LLMs reveal inherent safety concerns. To mitigate the biases of LLM from a group fairness perspective, we pioneer a novel chain-of-thought method GF-Think to mitigate biases of LLMs from a group fairness perspective. Experimental results demonstrate its efficacy in mitigating bias in LLMs to achieve fairness.

Large language models have shown unprecedented abilities in generating linguistically coherent and syntactically correct natural language output. However, they often return incorrect and inconsistent answers to input questions. Due to the complexity and uninterpretability of the internally learned representations, it is challenging to modify language models such that they provide correct and consistent results. The data management community has developed various methods and tools for providing consistent answers over inconsistent datasets. In these methods, users specify the desired properties of data in a domain in the form of high-level declarative constraints. This approach has provided usable and scalable methods to delivering consistent information from inconsistent datasets. We aim to build upon this success and leverage these methods to modify language models such that they deliver consistent and accurate results. We investigate the challenges of using these ideas to obtain consistent and relevant answers from language models and report some preliminary empirical studies.

The recently released Google Gemini class of models are the first to comprehensively report results that rival the OpenAI GPT series across a wide variety of tasks. In this paper, we do an in-depth exploration of Gemini's language abilities, making two contributions. First, we provide a third-party, objective comparison of the abilities of the OpenAI GPT and Google Gemini models with reproducible code and fully transparent results. Second, we take a closer look at the results, identifying areas where one of the two model classes excels. We perform this analysis over 10 datasets testing a variety of language abilities, including reasoning, answering knowledge-based questions, solving math problems, translating between languages, generating code, and acting as instruction-following agents. From this analysis, we find that Gemini Pro achieves accuracy that is close but slightly inferior to the corresponding GPT 3.5 Turbo on all English-language tasks that we benchmarked, but find that Gemini Pro excels in translation into other languages for the languages that it supports. We further provide explanations for some of the under-performing tasks, including failures in mathematical reasoning with many digits, sensitivity to multiple-choice answer ordering, and others. We also identify areas where Gemini Pro demonstrates comparably high performance, such as handling longer and more complex reasoning chains.

This paper explores the image synthesis capabilities of GPT-4, a leading multi-modal large language model. We establish a benchmark for evaluating the fidelity of texture features in images generated by GPT-4, comprising manually painted pictures and their AI-generated counterparts. The contributions of this study are threefold: First, we provide an in-depth analysis of the fidelity of image synthesis features based on GPT-4, marking the first such study on this state-of-the-art model. Second, the quantitative and qualitative experiments fully reveals the limitations of the GPT-4 model in image synthesis. Third, we have compiled a unique benchmark of manual drawings and corresponding GPT-4-generated images, introducing a new task to advance fidelity research in AI-generated content (AIGC). The dataset is available at: \url{https://github.com/rickwang28574/DeepArt}.

Few-shot prompting elicits the remarkable abilities of large language models by equipping them with a few demonstration examples in the input. However, the traditional method of providing large language models with all demonstration input-output pairs at once may not effectively guide large language models to learn the specific input-output mapping relationship. In this paper, inspired by the regulatory and supportive role of metacognition in students' learning, we propose a novel metacognition-enhanced few-shot prompting, which guides large language models to reflect on their thought processes to comprehensively learn the given demonstration examples. Furthermore, considering that positive reinforcement can improve students' learning motivation, we introduce positive reinforcement into our metacognition-enhanced few-shot prompting to promote the few-shot learning of large language models by providing response-based positive feedback. The experimental results on two real-world datasets show that our metacognition-enhanced few-shot prompting with positive reinforcement surpasses traditional few-shot prompting in classification accuracy and macro F1.

Language models have been increasingly popular for automatic creativity assessment, generating semantic distances to objectively measure the quality of creative ideas. However, there is currently a lack of an automatic assessment system for evaluating creative ideas in the Chinese language. To address this gap, we developed TransDis, a scoring system using transformer-based language models, capable of providing valid originality (quality) and flexibility (variety) scores for Alternative Uses Task (AUT) responses in Chinese. Study 1 demonstrated that the latent model-rated originality factor, comprised of three transformer-based models, strongly predicted human originality ratings, and the model-rated flexibility strongly correlated with human flexibility ratings as well. Criterion validity analyses indicated that model-rated originality and flexibility positively correlated to other creativity measures, demonstrating similar validity to human ratings. Study 2 & 3 showed that TransDis effectively distinguished participants instructed to provide creative vs. common uses (Study 2) and participants instructed to generate ideas in a flexible vs. persistent way (Study 3). Our findings suggest that TransDis can be a reliable and low-cost tool for measuring idea originality and flexibility in Chinese language, potentially paving the way for automatic creativity assessment in other languages. We offer an open platform to compute originality and flexibility for AUT responses in Chinese and over 50 other languages (https://osf.io/59jv2/).

In this paper, we present our finding that prepending a Task-Agnostic Prefix Prompt (TAPP) to the input improves the instruction-following ability of various Large Language Models (LLMs) during inference. TAPP is different from canonical prompts for LLMs in that it is a fixed prompt prepended to the beginning of every input regardless of the target task for zero-shot generalization. We observe that both base LLMs (i.e. not fine-tuned to follow instructions) and instruction-tuned models benefit from TAPP, resulting in 34.58% and 12.26% improvement on average, respectively. This implies that the instruction-following ability of LLMs can be improved during inference time with a fixed prompt constructed with simple heuristics. We hypothesize that TAPP assists language models to better estimate the output distribution by focusing more on the instruction of the target task during inference. In other words, such ability does not seem to be sufficiently activated in not only base LLMs but also many instruction-fine-tuned LLMs. All experiments are reproducible from https://github.com/seonghyeonye/TAPP.

Understanding Transformer-based models has attracted significant attention, as they lie at the heart of recent technological advances across machine learning. While most interpretability methods rely on running models over inputs, recent work has shown that a zero-pass approach, where parameters are interpreted directly without a forward/backward pass is feasible for some Transformer parameters, and for two-layer attention networks. In this work, we present a theoretical analysis where all parameters of a trained Transformer are interpreted by projecting them into the embedding space, that is, the space of vocabulary items they operate on. We derive a simple theoretical framework to support our arguments and provide ample evidence for its validity. First, an empirical analysis showing that parameters of both pretrained and fine-tuned models can be interpreted in embedding space. Second, we present two applications of our framework: (a) aligning the parameters of different models that share a vocabulary, and (b) constructing a classifier without training by ``translating'' the parameters of a fine-tuned classifier to parameters of a different model that was only pretrained. Overall, our findings open the door to interpretation methods that, at least in part, abstract away from model specifics and operate in the embedding space only.

Although large language models (LLMs) often produce impressive outputs, it remains unclear how they perform in real-world scenarios requiring strong reasoning skills and expert domain knowledge. We set out to investigate whether close- and open-source models (GPT-3.5, LLama-2, etc.) can be applied to answer and reason about difficult real-world-based questions. We focus on three popular medical benchmarks (MedQA-USMLE, MedMCQA, and PubMedQA) and multiple prompting scenarios: Chain-of-Thought (CoT, think step-by-step), few-shot and retrieval augmentation. Based on an expert annotation of the generated CoTs, we found that InstructGPT can often read, reason and recall expert knowledge. Last, by leveraging advances in prompt engineering (few-shot and ensemble methods), we demonstrated that GPT-3.5 not only yields calibrated predictive distributions, but also reaches the passing score on three datasets: MedQA-USMLE 60.2%, MedMCQA 62.7% and PubMedQA 78.2%. Open-source models are closing the gap: Llama-2 70B also passed the MedQA-USMLE with 62.5% accuracy.

The ability of large language models (LLMs) to follow natural language instructions with human-level fluency suggests many opportunities in healthcare to reduce administrative burden and improve quality of care. However, evaluating LLMs on realistic text generation tasks for healthcare remains challenging. Existing question answering datasets for electronic health record (EHR) data fail to capture the complexity of information needs and documentation burdens experienced by clinicians. To address these challenges, we introduce MedAlign, a benchmark dataset of 983 natural language instructions for EHR data. MedAlign is curated by 15 clinicians (7 specialities), includes clinician-written reference responses for 303 instructions, and provides 276 longitudinal EHRs for grounding instruction-response pairs. We used MedAlign to evaluate 6 general domain LLMs, having clinicians rank the accuracy and quality of each LLM response. We found high error rates, ranging from 35% (GPT-4) to 68% (MPT-7B-Instruct), and an 8.3% drop in accuracy moving from 32k to 2k context lengths for GPT-4. Finally, we report correlations between clinician rankings and automated natural language generation metrics as a way to rank LLMs without human review. We make MedAlign available under a research data use agreement to enable LLM evaluations on tasks aligned with clinician needs and preferences.