Performance Evaluation of Large Language Models in Statistical Programming

The programming capabilities of large language models (LLMs) have revolutionized automatic code generation and opened new avenues for automatic statistical analysis. However, the validity and quality of these generated codes need to be systematically evaluated before they can be widely adopted. Despite their growing prominence, a comprehensive evaluation of statistical code generated by LLMs remains scarce in the literature. In this paper, we assess the performance of LLMs, including two versions of ChatGPT and one version of Llama, in the domain of SAS programming for statistical analysis. Our study utilizes a set of statistical analysis tasks encompassing diverse statistical topics and datasets. Each task includes a problem description, dataset information, and human-verified SAS code. We conduct a comprehensive assessment of the quality of SAS code generated by LLMs through human expert evaluation based on correctness, effectiveness, readability, executability, and the accuracy of output results. The analysis of rating scores reveals that while LLMs demonstrate usefulness in generating syntactically correct code, they struggle with tasks requiring deep domain understanding and may produce redundant or incorrect results. This study offers valuable insights into the capabilities and limitations of LLMs in statistical programming, providing guidance for future advancements in AI-assisted coding systems for statistical analysis.

Decoupling Dark Knowledge via Block-wise Logit Distillation for Feature-level Alignment

Knowledge Distillation (KD), a learning manner with a larger teacher network guiding a smaller student network, transfers dark knowledge from the teacher to the student via logits or intermediate features, with the aim of producing a well-performed lightweight model. Notably, many subsequent feature-based KD methods outperformed the earliest logit-based KD method and iteratively generated numerous state-of-the-art distillation methods. Nevertheless, recent work has uncovered the potential of the logit-based method, bringing the simple KD form based on logits back into the limelight. Features or logits? They partially implement the KD with entirely distinct perspectives; therefore, choosing between logits and features is not straightforward. This paper provides a unified perspective of feature alignment in order to obtain a better comprehension of their fundamental distinction. Inheriting the design philosophy and insights of feature-based and logit-based methods, we introduce a block-wise logit distillation framework to apply implicit logit-based feature alignment by gradually replacing teacher's blocks as intermediate stepping-stone models to bridge the gap between the student and the teacher. Our method obtains comparable or superior results to state-of-the-art distillation methods. This paper demonstrates the great potential of combining logit and features, and we hope it will inspire future research to revisit KD from a higher vantage point.

FairMT-Bench: Benchmarking Fairness for Multi-turn Dialogue in Conversational LLMs

The growing use of large language model (LLM)-based chatbots has raised concerns about fairness. Fairness issues in LLMs can lead to severe consequences, such as bias amplification, discrimination, and harm to marginalized communities. While existing fairness benchmarks mainly focus on single-turn dialogues, multi-turn scenarios, which in fact better reflect real-world conversations, present greater challenges due to conversational complexity and potential bias accumulation. In this paper, we propose a comprehensive fairness benchmark for LLMs in multi-turn dialogue scenarios, \textbf{FairMT-Bench}. Specifically, we formulate a task taxonomy targeting LLM fairness capabilities across three stages: context understanding, user interaction, and instruction trade-offs, with each stage comprising two tasks. To ensure coverage of diverse bias types and attributes, we draw from existing fairness datasets and employ our template to construct a multi-turn dialogue dataset, \texttt{FairMT-10K}. For evaluation, GPT-4 is applied, alongside bias classifiers including Llama-Guard-3 and human validation to ensure robustness. Experiments and analyses on \texttt{FairMT-10K} reveal that in multi-turn dialogue scenarios, current LLMs are more likely to generate biased responses, and there is significant variation in performance across different tasks and models. Based on this, we curate a challenging dataset, \texttt{FairMT-1K}, and test 15 current state-of-the-art (SOTA) LLMs on this dataset. The results show the current state of fairness in LLMs and showcase the utility of this novel approach for assessing fairness in more realistic multi-turn dialogue contexts, calling for future work to focus on LLM fairness improvement and the adoption of \texttt{FairMT-1K} in such efforts.

Modality-Fair Preference Optimization for Trustworthy MLLM Alignment

Direct Preference Optimization (DPO) is effective for aligning large language models (LLMs), but when applied to multimodal models (MLLMs), it often favors text over image information, leading to unreliable outputs and visual hallucinations. To address this, we propose Modality-Fair Preference Optimization (MFPO) to balance text and image preferences. First, we found that the lack of image-related rewards in preference data biases optimization toward text, so we created automated, fine-grained image preference data to correct this. Then, we designed a learning objective to ensure the model captures both text and image preferences while maintaining high-quality outputs. Finally, we use a multi-stage alignment approach to stabilize training and improve learning across both modalities. Extensive experiments demonstrate that MFPO significantly enhances MLLM trustworthiness. On models like LLaVA-v1.5 (7B, 13B), our approach reduces hallucinations substantially. On the 7B model, MFPO outperforms GPT-4V and achieves a nearly 40\% improvement over previous methods on Object HalBench, as well as achieving state-of-the-art performance on both Object HalBench and AMBER when combined with the latest LLaVA-v1.6. Code will be released.

PAD: Personalized Alignment at Decoding-Time

Aligning with personalized preferences, which vary significantly across cultural, educational, and political differences, poses a significant challenge due to the computational costs and data demands of traditional alignment methods. In response, this paper presents Personalized Alignment at Decoding-time (PAD), a novel framework designed to align LLM outputs with diverse personalized preferences during the inference phase, eliminating the need for additional training. By introducing a unique personalized reward modeling strategy, this framework decouples the text generation process from personalized preferences, facilitating the generation of generalizable token-level personalized rewards. The PAD algorithm leverages these rewards to guide the decoding process, dynamically tailoring the base model's predictions to personalized preferences. Extensive experimental results demonstrate that PAD not only outperforms existing training-based alignment methods in terms of aligning with diverse preferences but also shows significant generalizability to preferences unseen during training and scalability across different base models. This work advances the capability of LLMs to meet user needs in real-time applications, presenting a substantial step forward in personalized LLM alignment.

GEIC: Universal and Multilingual Named Entity Recognition with Large Language Models

Large Language Models (LLMs) have supplanted traditional methods in numerous natural language processing tasks. Nonetheless, in Named Entity Recognition (NER), existing LLM-based methods underperform compared to baselines and require significantly more computational resources, limiting their application. In this paper, we introduce the task of generation-based extraction and in-context classification (GEIC), designed to leverage LLMs' prior knowledge and self-attention mechanisms for NER tasks. We then propose CascadeNER, a universal and multilingual GEIC framework for few-shot and zero-shot NER. CascadeNER employs model cascading to utilize two small-parameter LLMs to extract and classify independently, reducing resource consumption while enhancing accuracy. We also introduce AnythingNER, the first NER dataset specifically designed for LLMs, including 8 languages, 155 entity types and a novel dynamic categorization system. Experiments show that CascadeNER achieves state-of-the-art performance on low-resource and fine-grained scenarios, including CrossNER and FewNERD. Our work is openly accessible.

Editable Fairness: Fine-Grained Bias Mitigation in Language Models

Generating fair and accurate predictions plays a pivotal role in deploying large language models (LLMs) in the real world. However, existing debiasing methods inevitably generate unfair or incorrect predictions as they are designed and evaluated to achieve parity across different social groups but leave aside individual commonsense facts, resulting in modified knowledge that elicits unreasonable or undesired predictions. In this paper, we first establish a new bias mitigation benchmark, BiaScope, which systematically assesses performance by leveraging newly constructed datasets and metrics on knowledge retention and generalization. Then, we propose a novel debiasing approach, Fairness Stamp (FAST), which enables fine-grained calibration of individual social biases. FAST identifies the decisive layer responsible for storing social biases and then calibrates its outputs by integrating a small modular network, considering both bias mitigation and knowledge-preserving demands. Comprehensive experiments demonstrate that FAST surpasses state-of-the-art baselines with superior debiasing performance while not compromising the overall model capability for knowledge retention and downstream predictions. This highlights the potential of fine-grained debiasing strategies to achieve fairness in LLMs. Code will be publicly available.

VersusDebias: Universal Zero-Shot Debiasing for Text-to-Image Models via SLM-Based Prompt Engineering and Generative Adversary

With the rapid development of Text-to-Image models, biases in human image generation against demographic groups social attract more and more concerns. Existing methods are designed based on certain models with fixed prompts, unable to accommodate the trend of high-speed updating of Text-to-Image (T2I) models and variable prompts in practical scenes. Additionally, they fail to consider the possibility of hallucinations, leading to deviations between expected and actual results. To address this issue, we introduce VersusDebias, a novel and universal debiasing framework for biases in T2I models, consisting of one generative adversarial mechanism (GAM) and one debiasing generation mechanism using a small language model (SLM). The self-adaptive GAM generates specialized attribute arrays for each prompts for diminishing the influence of hallucinations from T2I models. The SLM uses prompt engineering to generate debiased prompts for the T2I model, providing zero-shot debiasing ability and custom optimization for different models. Extensive experiments demonstrate VersusDebias's capability to rectify biases on arbitrary models across multiple protected attributes simultaneously, including gender, race, and age. Furthermore, VersusDebias outperforms existing methods in both zero-shot and few-shot situations, illustrating its extraordinary utility. Our work is openly accessible to the research community to ensure the reproducibility.

BIGbench: A Unified Benchmark for Social Bias in Text-to-Image Generative Models Based on Multi-modal LLM

Text-to-Image (T2I) generative models are becoming more crucial in terms of their ability to generate complex and high-quality images, which also raises concerns about the social biases in their outputs, especially in human generation. Sociological research has established systematic classifications of bias; however, existing research of T2I models often conflates different types of bias, hindering the progress of these methods. In this paper, we introduce BIGbench, a unified benchmark for Biases of Image Generation with a well-designed dataset. In contrast to existing benchmarks, BIGbench classifies and evaluates complex biases into four dimensions: manifestation of bias, visibility of bias, acquired attributes, and protected attributes. Additionally, BIGbench applies advanced multi-modal large language models (MLLM), achieving fully automated evaluation while maintaining high accuracy. We apply BIGbench to evaluate eight recent general T2I models and three debiased methods. We also conduct human evaluation, whose results demonstrated the effectiveness of BIGbench in aligning images and identifying various biases. Besides, our study also revealed new research directions about biases, including the side-effect of irrelevant protected attributes and distillation. Our dataset and benchmark is openly accessible to the research community to ensure the reproducibility.

BiasAlert: A Plug-and-play Tool for Social Bias Detection in LLMs

Evaluating the bias in Large Language Models (LLMs) becomes increasingly crucial with their rapid development. However, existing evaluation methods rely on fixed-form outputs and cannot adapt to the flexible open-text generation scenarios of LLMs (e.g., sentence completion and question answering). To address this, we introduce BiasAlert, a plug-and-play tool designed to detect social bias in open-text generations of LLMs. BiasAlert integrates external human knowledge with inherent reasoning capabilities to detect bias reliably. Extensive experiments demonstrate that BiasAlert significantly outperforms existing state-of-the-art methods like GPT4-as-A-Judge in detecting bias. Furthermore, through application studies, we demonstrate the utility of BiasAlert in reliable LLM bias evaluation and bias mitigation across various scenarios. Model and code will be publicly released.