Search

Large language models have demonstrated substantial advancements in reasoning capabilities, particularly through inference-time scaling, as illustrated by models such as OpenAI's o1. However, current Vision-Language Models (VLMs) often struggle to perform systematic and structured reasoning, especially when handling complex visual question-answering tasks. In this work, we introduce LLaVA-o1, a novel VLM designed to conduct autonomous multistage reasoning. Unlike chain-of-thought prompting, LLaVA-o1 independently engages in sequential stages of summarization, visual interpretation, logical reasoning, and conclusion generation. This structured approach enables LLaVA-o1 to achieve marked improvements in precision on reasoning-intensive tasks. To accomplish this, we compile the LLaVA-o1-100k dataset, integrating samples from various visual question answering sources and providing structured reasoning annotations. Besides, we propose an inference-time stage-level beam search method, which enables effective inference-time scaling. Remarkably, with only 100k training samples and a simple yet effective inference time scaling method, LLaVA-o1 not only outperforms its base model by 8.9% on a wide range of multimodal reasoning benchmarks, but also surpasses the performance of larger and even closed-source models, such as Gemini-1.5-pro, GPT-4o-mini, and Llama-3.2-90B-Vision-Instruct., Comment: 11 pages, 5 figures

Large Language Models (LLMs) have become essential in advancing natural language processing (NLP) tasks, but their sequential token generation limits inference speed. Multi-Draft Speculative Decoding (MDSD) offers a promising solution by using a smaller draft model to generate multiple token sequences, which the target LLM verifies in parallel. However, current heuristic approaches, such as Recursive Rejection Sampling (RRS), suffer from low acceptance rates in subsequent drafts, limiting the advantages of using multiple drafts. Meanwhile, Optimal Transport with Membership Cost (OTM) can theoretically improve acceptance rates, but its computational cost is too high for real-time use. We present SpecHub, a novel, efficient sampling-verification method for MDSD that improves acceptance rates with only linear computational overhead. By simplifying the OTM problem into a compact Linear Programming model, SpecHub significantly reduces computational complexity. It further accelerates sampling by leveraging a sparse joint distribution, focusing computation on high-probability token sequences. In extensive experiments, Spechub consistently generates 0.05-0.27 and 0.02-0.16 more tokens per step than RRS and RRS without replacement. We attach our code at \url{https://github.com/MasterGodzilla/Speculative_decoding_OT}., Comment: EMNLP 2024 (Main)

The rapid progression of multimodal large language models (MLLMs) has demonstrated superior performance on various multimodal benchmarks. However, the issue of data contamination during training creates challenges in performance evaluation and comparison. While numerous methods exist for detecting dataset contamination in large language models (LLMs), they are less effective for MLLMs due to their various modalities and multiple training phases. In this study, we introduce a multimodal data contamination detection framework, MM-Detect, designed for MLLMs. Our experimental results indicate that MM-Detect is sensitive to varying degrees of contamination and can highlight significant performance improvements due to leakage of the training set of multimodal benchmarks. Furthermore, We also explore the possibility of contamination originating from the pre-training phase of LLMs used by MLLMs and the fine-tuning phase of MLLMs, offering new insights into the stages at which contamination may be introduced.

Machine unlearning has emerged as an effective strategy for forgetting specific information in the training data. However, with the increasing integration of visual data, privacy concerns in Vision Language Models (VLMs) remain underexplored. To address this, we introduce Facial Identity Unlearning Benchmark (FIUBench), a novel VLM unlearning benchmark designed to robustly evaluate the effectiveness of unlearning algorithms under the Right to be Forgotten setting. Specifically, we formulate the VLM unlearning task via constructing the Fictitious Facial Identity VQA dataset and apply a two-stage evaluation pipeline that is designed to precisely control the sources of information and their exposure levels. In terms of evaluation, since VLM supports various forms of ways to ask questions with the same semantic meaning, we also provide robust evaluation metrics including membership inference attacks and carefully designed adversarial privacy attacks to evaluate the performance of algorithms. Through the evaluation of four baseline VLM unlearning algorithms within FIUBench, we find that all methods remain limited in their unlearning performance, with significant trade-offs between model utility and forget quality. Furthermore, our findings also highlight the importance of privacy attacks for robust evaluations. We hope FIUBench will drive progress in developing more effective VLM unlearning algorithms.

Large Language Models (LLMs) are increasingly employed for simulations, enabling applications in role-playing agents and Computational Social Science (CSS). However, the reliability of these simulations is under-explored, which raises concerns about the trustworthiness of LLMs in these applications. In this paper, we aim to answer ``How reliable is LLM-based simulation?'' To address this, we introduce TrustSim, an evaluation dataset covering 10 CSS-related topics, to systematically investigate the reliability of the LLM simulation. We conducted experiments on 14 LLMs and found that inconsistencies persist in the LLM-based simulated roles. In addition, the consistency level of LLMs does not strongly correlate with their general performance. To enhance the reliability of LLMs in simulation, we proposed Adaptive Learning Rate Based ORPO (AdaORPO), a reinforcement learning-based algorithm to improve the reliability in simulation across 7 LLMs. Our research provides a foundation for future studies to explore more robust and trustworthy LLM-based simulations.

Continual offline reinforcement learning (CORL) has shown impressive ability in diffusion-based lifelong learning systems by modeling the joint distributions of trajectories. However, most research only focuses on limited continual task settings where the tasks have the same observation and action space, which deviates from the realistic demands of training agents in various environments. In view of this, we propose Vector-Quantized Continual Diffuser, named VQ-CD, to break the barrier of different spaces between various tasks. Specifically, our method contains two complementary sections, where the quantization spaces alignment provides a unified basis for the selective weights activation. In the quantized spaces alignment, we leverage vector quantization to align the different state and action spaces of various tasks, facilitating continual training in the same space. Then, we propose to leverage a unified diffusion model attached by the inverse dynamic model to master all tasks by selectively activating different weights according to the task-related sparse masks. Finally, we conduct extensive experiments on 15 continual learning (CL) tasks, including conventional CL task settings (identical state and action spaces) and general CL task settings (various state and action spaces). Compared with 16 baselines, our method reaches the SOTA performance.

Evaluating large language models (LLMs) is costly: it requires the generation and examination of LLM outputs on a large-scale benchmark of various tasks. This paper investigates how to efficiently reduce the tasks used to benchmark LLMs without affecting the evaluation quality. Our study reveals that task transferability and relevance provide critical information to identify the most representative subset of tasks via optimizing a facility location function. We propose a practically efficient metric for estimating the transferability between two tasks via in-context learning (ICL). By analyzing the pairwise transferability, we can reduce tasks in a modern LLM benchmark (e.g., MMLU or FLAN) to 5% while inducing only a <4% difference to the evaluation on the original benchmark. Compared to prior works, our method is training-free, gradient-free, and highly efficient requiring ICL only., Comment: https://github.com/tianyi-lab/bento

Federated learning (FL), an emerging distributed machine learning paradigm, has been applied to various privacy-preserving scenarios. However, due to its distributed nature, FL faces two key issues: the non-independent and identical distribution (non-IID) of user data and vulnerability to Byzantine threats. To address these challenges, in this paper, we propose FedCAP, a robust FL framework against both data heterogeneity and Byzantine attacks. The core of FedCAP is a model update calibration mechanism to help a server capture the differences in the direction and magnitude of model updates among clients. Furthermore, we design a customized model aggregation rule that facilitates collaborative training among similar clients while accelerating the model deterioration of malicious clients. With a Euclidean norm-based anomaly detection mechanism, the server can quickly identify and permanently remove malicious clients. Moreover, the impact of data heterogeneity and Byzantine attacks can be further mitigated through personalization on the client side. We conduct extensive experiments, comparing multiple state-of-the-art baselines, to demonstrate that FedCAP performs well in several non-IID settings and shows strong robustness under a series of poisoning attacks., Comment: 14 pages, 12 figures, 5 tables, accepted by 2024 Annual Computer Security Applications Conference (ACSAC 2024)

Recent advancements in both representation learning and function learning have demonstrated substantial promise across diverse domains of artificial intelligence. However, the effective integration of these paradigms poses a significant challenge, particularly in cases where users must manually decide whether to apply a representation learning or function learning model based on dataset characteristics. To address this issue, we introduce MLP-KAN, a unified method designed to eliminate the need for manual model selection. By integrating Multi-Layer Perceptrons (MLPs) for representation learning and Kolmogorov-Arnold Networks (KANs) for function learning within a Mixture-of-Experts (MoE) architecture, MLP-KAN dynamically adapts to the specific characteristics of the task at hand, ensuring optimal performance. Embedded within a transformer-based framework, our work achieves remarkable results on four widely-used datasets across diverse domains. Extensive experimental evaluation demonstrates its superior versatility, delivering competitive performance across both deep representation and function learning tasks. These findings highlight the potential of MLP-KAN to simplify the model selection process, offering a comprehensive, adaptable solution across various domains. Our code and weights are available at \url{https://github.com/DLYuanGod/MLP-KAN}.

The perturbation analysis of linear solvers applied to systems arising broadly in machine learning settings -- for instance, when using linear regression models -- establishes an important perspective when reframing these analyses through the lens of a data poisoning attack. By analyzing solvers' responses to such attacks, this work aims to contribute to the development of more robust linear solvers and provide insights into poisoning attacks on linear solvers. In particular, we investigate how the errors in the input data will affect the fitting error and accuracy of the solution from a linear system-solving algorithm under perturbations common in adversarial attacks. We propose data perturbation through two distinct knowledge levels, developing a poisoning optimization and studying two methods of perturbation: Label-guided Perturbation (LP) and Unconditioning Perturbation (UP). Existing works mainly focus on deriving the worst-case perturbation bound from a theoretical perspective, and the analysis is often limited to specific kinds of linear system solvers. Under the circumstance that the data is intentionally perturbed -- as is the case with data poisoning -- we seek to understand how different kinds of solvers react to these perturbations, identifying those algorithms most impacted by different types of adversarial attacks., Comment: 18 pages

This comprehensive study evaluates the performance of OpenAI's o1-preview large language model across a diverse array of complex reasoning tasks, spanning multiple domains, including computer science, mathematics, natural sciences, medicine, linguistics, and social sciences. Through rigorous testing, o1-preview demonstrated remarkable capabilities, often achieving human-level or superior performance in areas ranging from coding challenges to scientific reasoning and from language processing to creative problem-solving. Key findings include: -83.3% success rate in solving complex competitive programming problems, surpassing many human experts. -Superior ability in generating coherent and accurate radiology reports, outperforming other evaluated models. -100% accuracy in high school-level mathematical reasoning tasks, providing detailed step-by-step solutions. -Advanced natural language inference capabilities across general and specialized domains like medicine. -Impressive performance in chip design tasks, outperforming specialized models in areas such as EDA script generation and bug analysis. -Remarkable proficiency in anthropology and geology, demonstrating deep understanding and reasoning in these specialized fields. -Strong capabilities in quantitative investing. O1 has comprehensive financial knowledge and statistical modeling skills. -Effective performance in social media analysis, including sentiment analysis and emotion recognition. The model excelled particularly in tasks requiring intricate reasoning and knowledge integration across various fields. While some limitations were observed, including occasional errors on simpler problems and challenges with certain highly specialized concepts, the overall results indicate significant progress towards artificial general intelligence.

Decoding visual-semantic information from brain signals, such as functional MRI (fMRI), across different subjects poses significant challenges, including low signal-to-noise ratio, limited data availability, and cross-subject variability. Recent advancements in large language models (LLMs) show remarkable effectiveness in processing multimodal information. In this study, we introduce an LLM-based approach for reconstructing visual-semantic information from fMRI signals elicited by video stimuli. Specifically, we employ fine-tuning techniques on an fMRI encoder equipped with adaptors to transform brain responses into latent representations aligned with the video stimuli. Subsequently, these representations are mapped to textual modality by LLM. In particular, we integrate self-supervised domain adaptation methods to enhance the alignment between visual-semantic information and brain responses. Our proposed method achieves good results using various quantitative semantic metrics, while yielding similarity with ground-truth information., Comment: ECCV2024 Workshop

Biomedical image segmentation is crucial for accurately diagnosing and analyzing various diseases. However, Convolutional Neural Networks (CNNs) and Transformers, the most commonly used architectures for this task, struggle to effectively capture long-range dependencies due to the inherent locality of CNNs and the computational complexity of Transformers. To address this limitation, we introduce TTT-Unet, a novel framework that integrates Test-Time Training (TTT) layers into the traditional U-Net architecture for biomedical image segmentation. TTT-Unet dynamically adjusts model parameters during the testing time, enhancing the model's ability to capture both local and long-range features. We evaluate TTT-Unet on multiple medical imaging datasets, including 3D abdominal organ segmentation in CT and MR images, instrument segmentation in endoscopy images, and cell segmentation in microscopy images. The results demonstrate that TTT-Unet consistently outperforms state-of-the-art CNN-based and Transformer-based segmentation models across all tasks. The code is available at https://github.com/rongzhou7/TTT-Unet.

Artificial neural networks, especially recent diffusion-based models, have shown remarkable superiority in gaming, control, and QA systems, where the training tasks' datasets are usually static. However, in real-world applications, such as robotic control of reinforcement learning (RL), the tasks are changing, and new tasks arise in a sequential order. This situation poses the new challenge of plasticity-stability trade-off for training an agent who can adapt to task changes and retain acquired knowledge. In view of this, we propose a rehearsal-based continual diffusion model, called Continual Diffuser (CoD), to endow the diffuser with the capabilities of quick adaptation (plasticity) and lasting retention (stability). Specifically, we first construct an offline benchmark that contains 90 tasks from multiple domains. Then, we train the CoD on each task with sequential modeling and conditional generation for making decisions. Next, we preserve a small portion of previous datasets as the rehearsal buffer and replay it to retain the acquired knowledge. Extensive experiments on a series of tasks show CoD can achieve a promising plasticity-stability trade-off and outperform existing diffusion-based methods and other representative baselines on most tasks.

Medical image segmentation and video object segmentation are essential for diagnosing and analyzing diseases by identifying and measuring biological structures. Recent advances in natural domain have been driven by foundation models like the Segment Anything Model 2 (SAM-2). To explore the performance of SAM-2 in biomedical applications, we designed three evaluation pipelines for single-frame 2D image segmentation, multi-frame 3D image segmentation and multi-frame video segmentation with varied prompt designs, revealing SAM-2's limitations in medical contexts. Consequently, we developed BioSAM-2, an enhanced foundation model optimized for biomedical data based on SAM-2. Our experiments show that BioSAM-2 not only surpasses the performance of existing state-of-the-art foundation models but also matches or even exceeds specialist models, demonstrating its efficacy and potential in the medical domain.

GPT-4V has attracted considerable attention due to its extraordinary capacity for integrating and processing multimodal information. At the same time, its ability of face recognition raises new safety concerns of privacy leakage. Despite researchers' efforts in safety alignment through RLHF or preprocessing filters, vulnerabilities might still be exploited. In our study, we introduce AutoJailbreak, an innovative automatic jailbreak technique inspired by prompt optimization. We leverage Large Language Models (LLMs) for red-teaming to refine the jailbreak prompt and employ weak-to-strong in-context learning prompts to boost efficiency. Furthermore, we present an effective search method that incorporates early stopping to minimize optimization time and token expenditure. Our experiments demonstrate that AutoJailbreak significantly surpasses conventional methods, achieving an Attack Success Rate (ASR) exceeding 95.3\%. This research sheds light on strengthening GPT-4V security, underscoring the potential for LLMs to be exploited in compromising GPT-4V integrity., Comment: TrustNLP@NAACL2024 (Fourth Workshop on Trustworthy Natural Language Processing)

Despite recent strides made by AI in image processing, the issue of mixed exposure, pivotal in many real-world scenarios like surveillance and photography, remains inadequately addressed. Traditional image enhancement techniques and current transformer models are limited with primary focus on either overexposure or underexposure. To bridge this gap, we introduce the Unified-Exposure Guided Transformer (Unified-EGformer). Our proposed solution is built upon advanced transformer architectures, equipped with local pixel-level refinement and global refinement blocks for color correction and image-wide adjustments. We employ a guided attention mechanism to precisely identify exposure-compromised regions, ensuring its adaptability across various real-world conditions. U-EGformer, with a lightweight design featuring a memory footprint (peak memory) of only $\sim$1134 MB (0.1 Million parameters) and an inference time of 95 ms (9.61x faster than the average), is a viable choice for real-time applications such as surveillance and autonomous navigation. Additionally, our model is highly generalizable, requiring minimal fine-tuning to handle multiple tasks and datasets with a single architecture., Comment: Under submission

Generative models hold promise for revolutionizing medical education, robot-assisted surgery, and data augmentation for medical AI development. Diffusion models can now generate realistic images from text prompts, while recent advancements have demonstrated their ability to create diverse, high-quality videos. However, these models often struggle with generating accurate representations of medical procedures and detailed anatomical structures. This paper introduces Bora, the first spatio-temporal diffusion probabilistic model designed for text-guided biomedical video generation. Bora leverages Transformer architecture and is pre-trained on general-purpose video generation tasks. It is fine-tuned through model alignment and instruction tuning using a newly established medical video corpus, which includes paired text-video data from various biomedical fields. To the best of our knowledge, this is the first attempt to establish such a comprehensive annotated biomedical video dataset. Bora is capable of generating high-quality video data across four distinct biomedical domains, adhering to medical expert standards and demonstrating consistency and diversity. This generalist video generative model holds significant potential for enhancing medical consultation and decision-making, particularly in resource-limited settings. Additionally, Bora could pave the way for immersive medical training and procedure planning. Extensive experiments on distinct medical modalities such as endoscopy, ultrasound, MRI, and cell tracking validate the effectiveness of our model in understanding biomedical instructions and its superior performance across subjects compared to state-of-the-art generation models.

While Large Language Models (LLMs) have achieved remarkable success across various applications, they also raise concerns regarding self-cognition. In this paper, we perform a pioneering study to explore self-cognition in LLMs. Specifically, we first construct a pool of self-cognition instruction prompts to evaluate where an LLM exhibits self-cognition and four well-designed principles to quantify LLMs' self-cognition. Our study reveals that 4 of the 48 models on Chatbot Arena--specifically Command R, Claude3-Opus, Llama-3-70b-Instruct, and Reka-core--demonstrate some level of detectable self-cognition. We observe a positive correlation between model size, training data quality, and self-cognition level. Additionally, we also explore the utility and trustworthiness of LLM in the self-cognition state, revealing that the self-cognition state enhances some specific tasks such as creative writing and exaggeration. We believe that our work can serve as an inspiration for further research to study the self-cognition in LLMs., Comment: Accepted at ICML 2024 Large Language Models and Cognition Workshop

Jailbreak attacks on large language models (LLMs) involve inducing these models to generate harmful content that violates ethics or laws, posing a significant threat to LLM security. Current jailbreak attacks face two main challenges: low success rates due to defensive measures and high resource requirements for crafting specific prompts. This paper introduces Virtual Context, which leverages special tokens, previously overlooked in LLM security, to improve jailbreak attacks. Virtual Context addresses these challenges by significantly increasing the success rates of existing jailbreak methods and requiring minimal background knowledge about the target model, thus enhancing effectiveness in black-box settings without additional overhead. Comprehensive evaluations show that Virtual Context-assisted jailbreak attacks can improve the success rates of four widely used jailbreak methods by approximately 40% across various LLMs. Additionally, applying Virtual Context to original malicious behaviors still achieves a notable jailbreak effect. In summary, our research highlights the potential of special tokens in jailbreak attacks and recommends including this threat in red-teaming testing to comprehensively enhance LLM security.

Large Language Models (LLMs) such as GPT-4 and Llama3 have significantly impacted various fields by enabling high-quality synthetic data generation and reducing dependence on expensive human-generated datasets. Despite this, challenges remain in the areas of generalization, controllability, diversity, and truthfulness within the existing generative frameworks. To address these challenges, this paper presents UniGen, a comprehensive LLM-powered framework designed to produce diverse, accurate, and highly controllable datasets. UniGen is adaptable, supporting all types of text datasets and enhancing the generative process through innovative mechanisms. To augment data diversity, UniGen incorporates an attribute-guided generation module and a group checking feature. For accuracy, it employs a code-based mathematical assessment for label verification alongside a retrieval-augmented generation technique for factual validation. The framework also allows for user-specified constraints, enabling customization of the data generation process to suit particular requirements. Extensive experiments demonstrate the superior quality of data generated by UniGen, and each module within UniGen plays a critical role in this enhancement. Additionally, UniGen is applied in two practical scenarios: benchmarking LLMs and data augmentation. The results indicate that UniGen effectively supports dynamic and evolving benchmarking, and that data augmentation improves LLM capabilities in various domains, including agent-oriented abilities and reasoning skills.

Personalized diffusion models (PDMs) have become prominent for adapting pretrained text-to-image models to generate images of specific subjects using minimal training data. However, PDMs are susceptible to minor adversarial perturbations, leading to significant degradation when fine-tuned on corrupted datasets. These vulnerabilities are exploited to create protective perturbations that prevent unauthorized image generation. Existing purification methods attempt to mitigate this issue but often over-purify images, resulting in information loss. In this work, we conduct an in-depth analysis of the fine-tuning process of PDMs through the lens of shortcut learning. We hypothesize and empirically demonstrate that adversarial perturbations induce a latent-space misalignment between images and their text prompts in the CLIP embedding space. This misalignment causes the model to erroneously associate noisy patterns with unique identifiers during fine-tuning, resulting in poor generalization. Based on these insights, we propose a systematic defense framework that includes data purification and contrastive decoupling learning. We first employ off-the-shelf image restoration techniques to realign images with their original semantic meanings in latent space. Then, we introduce contrastive decoupling learning with noise tokens to decouple the learning of personalized concepts from spurious noise patterns. Our study not only uncovers fundamental shortcut learning vulnerabilities in PDMs but also provides a comprehensive evaluation framework for developing stronger protection. Our extensive evaluation demonstrates its superiority over existing purification methods and stronger robustness against adaptive perturbation., Comment: Our code is available at https://github.com/liuyixin-louis/DiffShortcut

Vision Transformers (ViTs) have achieved remarkable performance in various image classification tasks by leveraging the attention mechanism to process image patches as tokens. However, the high computational and memory demands of ViTs pose significant challenges for deployment in resource-constrained environments. This paper introduces ViT-1.58b, a novel 1.58-bit quantized ViT model designed to drastically reduce memory and computational overhead while preserving competitive performance. ViT-1.58b employs ternary quantization, which refines the balance between efficiency and accuracy by constraining weights to {-1, 0, 1} and quantizing activations to 8-bit precision. Our approach ensures efficient scaling in terms of both memory and computation. Experiments on CIFAR-10 and ImageNet-1k demonstrate that ViT-1.58b maintains comparable accuracy to full-precision Vit, with significant reductions in memory usage and computational costs. This paper highlights the potential of extreme quantization techniques in developing sustainable AI solutions and contributes to the broader discourse on efficient model deployment in practical applications. Our code and weights are available at https://github.com/DLYuanGod/ViT-1.58b.

Large Language Models (LLMs) have demonstrated exceptional task-solving capabilities, increasingly adopting roles akin to human-like assistants. The broader integration of LLMs into society has sparked interest in whether they manifest psychological attributes, and whether these attributes are stable-inquiries that could deepen the understanding of their behaviors. Inspired by psychometrics, this paper presents a framework for investigating psychology in LLMs, including psychological dimension identification, assessment dataset curation, and assessment with results validation. Following this framework, we introduce a comprehensive psychometrics benchmark for LLMs that covers six psychological dimensions: personality, values, emotion, theory of mind, motivation, and intelligence. This benchmark includes thirteen datasets featuring diverse scenarios and item types. Our findings indicate that LLMs manifest a broad spectrum of psychological attributes. We also uncover discrepancies between LLMs' self-reported traits and their behaviors in real-world scenarios. This paper demonstrates a thorough psychometric assessment of LLMs, providing insights into reliable evaluation and potential applications in AI and social sciences.

Large Language Models (LLMs) have garnered significant attention due to their remarkable ability to process information across various languages. Despite their capabilities, they exhibit inconsistencies in handling identical queries in different languages, presenting challenges for further advancement. This paper introduces a method to enhance the multilingual performance of LLMs by aggregating knowledge from diverse languages. This approach incorporates a low-resource knowledge detector specific to a language, a language selection process, and mechanisms for answer replacement and integration. Our experiments demonstrate notable performance improvements, particularly in reducing language performance disparity. An ablation study confirms that each component of our method significantly contributes to these enhancements. This research highlights the inherent potential of LLMs to harmonize multilingual capabilities and offers valuable insights for further exploration.

Recently, Large Language Models (LLMs) have garnered significant attention for their exceptional natural language processing capabilities. However, concerns about their trustworthiness remain unresolved, particularly in addressing "jailbreaking" attacks on aligned LLMs. Previous research predominantly relies on scenarios with white-box LLMs or specific and fixed prompt templates, which are often impractical and lack broad applicability. In this paper, we introduce a straightforward and novel method, named ObscurePrompt, for jailbreaking LLMs, inspired by the observed fragile alignments in Out-of-Distribution (OOD) data. Specifically, we first formulate the decision boundary in the jailbreaking process and then explore how obscure text affects LLM's ethical decision boundary. ObscurePrompt starts with constructing a base prompt that integrates well-known jailbreaking techniques. Powerful LLMs are then utilized to obscure the original prompt through iterative transformations, aiming to bolster the attack's robustness. Comprehensive experiments show that our approach substantially improves upon previous methods in terms of attack effectiveness, maintaining efficacy against two prevalent defense mechanisms. We believe that our work can offer fresh insights for future research on enhancing LLM alignment.

Recently, Multimodal Large Language Models (MLLMs) have been used as agents to control keyboard and mouse inputs by directly perceiving the Graphical User Interface (GUI) and generating corresponding code. However, current agents primarily exhibit excellent understanding capabilities in static environments and are predominantly applied in relatively simple domains, such as Web or mobile interfaces. We argue that a robust GUI agent should be capable of perceiving temporal information on the GUI, including dynamic Web content and multi-step tasks. Additionally, it should possess a comprehensive understanding of various GUI scenarios, including desktop software and multi-window interactions. To this end, this paper introduces a new dataset, termed GUI-World, which features meticulously crafted Human-MLLM annotations, extensively covering six GUI scenarios and eight types of GUI-oriented questions in three formats. We evaluate the capabilities of current state-of-the-art MLLMs, including ImageLLMs and VideoLLMs, in understanding various types of GUI content, especially dynamic and sequential content. Our findings reveal that ImageLLMs struggle with dynamic GUI content without manually annotated keyframes or operation history. On the other hand, VideoLLMs fall short in all GUI-oriented tasks given the sparse GUI video dataset. Based on GUI-World, we take the initial step of leveraging a fine-tuned VideoLLM as a GUI agent, demonstrating an improved understanding of various GUI tasks. However, due to the limitations in the performance of base LLMs, we conclude that using VideoLLMs as GUI agents remains a significant challenge. We believe our work provides valuable insights for future research in dynamic GUI content understanding. The code and dataset are publicly available at our project homepage: https://gui-world.github.io/.

Large Language Models (LLMs) have achieved remarkable success across various industries due to their exceptional generative capabilities. However, for safe and effective real-world deployments, ensuring honesty and helpfulness is critical. This paper addresses the question: Can we prioritize the helpfulness of LLMs while preserving their honesty? To begin with, we establish exhaustive principles aimed at guaranteeing the honesty of LLM. Additionally, we introduce a novel dataset, referred to as HoneSet, comprising 930 queries spanning six categories meticulously crafted to assess an LLM's capacity for maintaining honesty. Subsequently, we present two approaches to augmenting honesty and helpfulness in LLMs: a training-free enhancement and a fine-tuning-based improvement. The training-free approach, which is based on curiosity-driven prompting, empowers LLMs to articulate internal confusion and uncertainty regarding queries, thereby optimizing their responses. Conversely, the fine-tuning-based method employs a two-stage process inspired by curriculum learning: initially instructing LLMs to discern between honest and dishonest responses, then refining their training to enhance helpfulness. Experiments conducted on nine prominent LLMs demonstrate a significant improvement in alignment with honesty across all models through the implementation of our proposed enhancements. Particularly noteworthy is the 65.3% enhancement observed in Llama3-8b and the remarkable 124.7% improvement in Mistral-7b, as measured by the H$^{2}$ (honest and helpful) assessment. We believe that our work can pave the way for developing more trustworthy LLMs for real-world applications.

Recent works have shown the remarkable superiority of transformer models in reinforcement learning (RL), where the decision-making problem is formulated as sequential generation. Transformer-based agents could emerge with self-improvement in online environments by providing task contexts, such as multiple trajectories, called in-context RL. However, due to the quadratic computation complexity of attention in transformers, current in-context RL methods suffer from huge computational costs as the task horizon increases. In contrast, the Mamba model is renowned for its efficient ability to process long-term dependencies, which provides an opportunity for in-context RL to solve tasks that require long-term memory. To this end, we first implement Decision Mamba (DM) by replacing the backbone of Decision Transformer (DT). Then, we propose a Decision Mamba-Hybrid (DM-H) with the merits of transformers and Mamba in high-quality prediction and long-term memory. Specifically, DM-H first generates high-value sub-goals from long-term memory through the Mamba model. Then, we use sub-goals to prompt the transformer, establishing high-quality predictions. Experimental results demonstrate that DM-H achieves state-of-the-art in long and short-term tasks, such as D4RL, Grid World, and Tmaze benchmarks. Regarding efficiency, the online testing of DM-H in the long-term task is 28$\times$ times faster than the transformer-based baselines., Comment: arXiv admin note: text overlap with arXiv:2405.20692. arXiv admin note: text overlap with arXiv:2405.20692; text overlap with arXiv:2305.16554, arXiv:2210.14215 by other authors

In-context learning is a promising approach for offline reinforcement learning (RL) to handle online tasks, which can be achieved by providing task prompts. Recent works demonstrated that in-context RL could emerge with self-improvement in a trial-and-error manner when treating RL tasks as an across-episodic sequential prediction problem. Despite the self-improvement not requiring gradient updates, current works still suffer from high computational costs when the across-episodic sequence increases with task horizons. To this end, we propose an In-context Decision Transformer (IDT) to achieve self-improvement in a high-level trial-and-error manner. Specifically, IDT is inspired by the efficient hierarchical structure of human decision-making and thus reconstructs the sequence to consist of high-level decisions instead of low-level actions that interact with environments. As one high-level decision can guide multi-step low-level actions, IDT naturally avoids excessively long sequences and solves online tasks more efficiently. Experimental results show that IDT achieves state-of-the-art in long-horizon tasks over current in-context RL methods. In particular, the online evaluation time of our IDT is \textbf{36$\times$} times faster than baselines in the D4RL benchmark and \textbf{27$\times$} times faster in the Grid World benchmark.

Federated continual learning (FCL) has received increasing attention due to its potential in handling real-world streaming data, characterized by evolving data distributions and varying client classes over time. The constraints of storage limitations and privacy concerns confine local models to exclusively access the present data within each learning cycle. Consequently, this restriction induces performance degradation in model training on previous data, termed "catastrophic forgetting". However, existing FCL approaches need to identify or know changes in data distribution, which is difficult in the real world. To release these limitations, this paper directs attention to a broader continuous framework. Within this framework, we introduce Federated Bayesian Neural Network (FedBNN), a versatile and efficacious framework employing a variational Bayesian neural network across all clients. Our method continually integrates knowledge from local and historical data distributions into a single model, adeptly learning from new data distributions while retaining performance on historical distributions. We rigorously evaluate FedBNN's performance against prevalent methods in federated learning and continual learning using various metrics. Experimental analyses across diverse datasets demonstrate that FedBNN achieves state-of-the-art results in mitigating forgetting.

Backdoor attacks have been well-studied in visible light object detection (VLOD) in recent years. However, VLOD can not effectively work in dark and temperature-sensitive scenarios. Instead, thermal infrared object detection (TIOD) is the most accessible and practical in such environments. In this paper, our team is the first to investigate the security vulnerabilities associated with TIOD in the context of backdoor attacks, spanning both the digital and physical realms. We introduce two novel types of backdoor attacks on TIOD, each offering unique capabilities: Object-affecting Attack and Range-affecting Attack. We conduct a comprehensive analysis of key factors influencing trigger design, which include temperature, size, material, and concealment. These factors, especially temperature, significantly impact the efficacy of backdoor attacks on TIOD. A thorough understanding of these factors will serve as a foundation for designing physical triggers and temperature controlling experiments. Our study includes extensive experiments conducted in both digital and physical environments. In the digital realm, we evaluate our approach using benchmark datasets for TIOD, achieving an Attack Success Rate (ASR) of up to 98.21%. In the physical realm, we test our approach in two real-world settings: a traffic intersection and a parking lot, using a thermal infrared camera. Here, we attain an ASR of up to 98.38%., Comment: To appear in CVPR 2024.11pages, 8 figures and 4 tables

Large Language Models (LLMs) have achieved remarkable progress in code generation. It now becomes crucial to identify whether the code is AI-generated and to determine the specific model used, particularly for purposes such as protecting Intellectual Property (IP) in industry and preventing cheating in programming exercises. To this end, several attempts have been made to insert watermarks into machine-generated code. However, existing approaches are limited to inserting only a single bit of information or overly depending on particular code patterns. In this paper, we introduce CodeIP, a novel multi-bit watermarking technique that embeds additional information to preserve crucial provenance details, such as the vendor ID of an LLM, thereby safeguarding the IPs of LLMs in code generation. Furthermore, to ensure the syntactical correctness of the generated code, we propose constraining the sampling process for predicting the next token by training a type predictor. Experiments conducted on a real-world dataset across five programming languages demonstrate the effectiveness of CodeIP in watermarking LLMs for code generation while maintaining the syntactical correctness of code., Comment: 15 pages, 13 figures

Recent years have witnessed significant progress in developing deep learning-based models for automated code completion. Although using source code in GitHub has been a common practice for training deep-learning-based models for code completion, it may induce some legal and ethical issues such as copyright infringement. In this paper, we investigate the legal and ethical issues of current neural code completion models by answering the following question: Is my code used to train your neural code completion model? To this end, we tailor a membership inference approach (termed CodeMI) that was originally crafted for classification tasks to a more challenging task of code completion. In particular, since the target code completion models perform as opaque black boxes, preventing access to their training data and parameters, we opt to train multiple shadow models to mimic their behavior. The acquired posteriors from these shadow models are subsequently employed to train a membership classifier. Subsequently, the membership classifier can be effectively employed to deduce the membership status of a given code sample based on the output of a target code completion model. We comprehensively evaluate the effectiveness of this adapted approach across a diverse array of neural code completion models, (i.e., LSTM-based, CodeGPT, CodeGen, and StarCoder). Experimental results reveal that the LSTM-based and CodeGPT models suffer the membership leakage issue, which can be easily detected by our proposed membership inference approach with an accuracy of 0.842, and 0.730, respectively. Interestingly, our experiments also show that the data membership of current large language models of code, e.g., CodeGen and StarCoder, is difficult to detect, leaving ampler space for further improvement. Finally, we also try to explain the findings from the perspective of model memorization.

LLM-as-a-Judge uses a large language model (LLM) to select the best response from a set of candidates for a given question. LLM-as-a-Judge has many applications such as LLM-powered search, reinforcement learning with AI feedback (RLAIF), and tool selection. In this work, we propose JudgeDeceiver, an optimization-based prompt injection attack to LLM-as-a-Judge. JudgeDeceiver injects a carefully crafted sequence into an attacker-controlled candidate response such that LLM-as-a-Judge selects the candidate response for an attacker-chosen question no matter what other candidate responses are. Specifically, we formulate finding such sequence as an optimization problem and propose a gradient based method to approximately solve it. Our extensive evaluation shows that JudgeDeceive is highly effective, and is much more effective than existing prompt injection attacks that manually craft the injected sequences and jailbreak attacks when extended to our problem. We also show the effectiveness of JudgeDeceiver in three case studies, i.e., LLM-powered search, RLAIF, and tool selection. Moreover, we consider defenses including known-answer detection, perplexity detection, and perplexity windowed detection. Our results show these defenses are insufficient, highlighting the urgent need for developing new defense strategies. Our implementation is available at this repository: https://github.com/ShiJiawenwen/JudgeDeceiver., Comment: To appear in the Proceedings of The ACM Conference on Computer and Communications Security (CCS), 2024

Text-to-video generation has made significant strides, but replicating the capabilities of advanced systems like OpenAI Sora remains challenging due to their closed-source nature. Existing open-source methods struggle to achieve comparable performance, often hindered by ineffective agent collaboration and inadequate training data quality. In this paper, we introduce Mora, a novel multi-agent framework that leverages existing open-source modules to replicate Sora functionalities. We address these fundamental limitations by proposing three key techniques: (1) multi-agent fine-tuning with a self-modulation factor to enhance inter-agent coordination, (2) a data-free training strategy that uses large models to synthesize training data, and (3) a human-in-the-loop mechanism combined with multimodal large language models for data filtering to ensure high-quality training datasets. Our comprehensive experiments on six video generation tasks demonstrate that Mora achieves performance comparable to Sora on VBench, outperforming existing open-source methods across various tasks. Specifically, in the text-to-video generation task, Mora achieved a Video Quality score of 0.800, surpassing Sora 0.797 and outperforming all other baseline models across six key metrics. Additionally, in the image-to-video generation task, Mora achieved a perfect Dynamic Degree score of 1.00, demonstrating exceptional capability in enhancing motion realism and achieving higher Imaging Quality than Sora. These results highlight the potential of collaborative multi-agent systems and human-in-the-loop mechanisms in advancing text-to-video generation. Our code is available at \url{https://github.com/lichao-sun/Mora}.

The rapid expansion of AI in healthcare has led to a surge in medical data generation and storage, boosting medical AI development. However, fears of unauthorized use, like training commercial AI models, hinder researchers from sharing their valuable datasets. To encourage data sharing, one promising solution is to introduce imperceptible noise into the data. This method aims to safeguard the data against unauthorized training by inducing degradation in the generalization ability of the trained model. However, they are not effective and efficient when applied to medical data, mainly due to the ignorance of the sparse nature of medical images. To address this problem, we propose the Sparsity-Aware Local Masking (SALM) method, a novel approach that selectively perturbs significant pixel regions rather than the entire image as previously. This simple yet effective approach, by focusing on local areas, significantly narrows down the search space for disturbances and fully leverages the characteristics of sparsity. Our extensive experiments across various datasets and model architectures demonstrate that SALM effectively prevents unauthorized training of different models and outperforms previous SoTA data protection methods., Comment: Accept by ICML 2024 NextGenAISafety

Alzheimer's Disease (AD) is a neurodegenerative condition characterized by diverse progression rates among individuals, with changes in cortical thickness (CTh) closely linked to its progression. Accurately forecasting CTh trajectories can significantly enhance early diagnosis and intervention strategies, providing timely care. However, the longitudinal data essential for these studies often suffer from temporal sparsity and incompleteness, presenting substantial challenges in modeling the disease's progression accurately. Existing methods are limited, focusing primarily on datasets without missing entries or requiring predefined assumptions about CTh progression. To overcome these obstacles, we propose a conditional score-based diffusion model specifically designed to generate CTh trajectories with the given baseline information, such as age, sex, and initial diagnosis. Our conditional diffusion model utilizes all available data during the training phase to make predictions based solely on baseline information during inference without needing prior history about CTh progression. The prediction accuracy of the proposed CTh prediction pipeline using a conditional score-based model was compared for sub-groups consisting of cognitively normal, mild cognitive impairment, and AD subjects. The Bland-Altman analysis shows our diffusion-based prediction model has a near-zero bias with narrow 95% confidential interval compared to the ground-truth CTh in 6-36 months. In addition, our conditional diffusion model has a stochastic generative nature, therefore, we demonstrated an uncertainty analysis of patient-specific CTh prediction through multiple realizations.

Data scarcity and privacy concerns limit the availability of high-quality medical images for public use, which can be mitigated through medical image synthesis. However, current medical image synthesis methods often struggle to accurately capture the complexity of detailed anatomical structures and pathological conditions. To address these challenges, we propose a novel medical image synthesis model that leverages fine-grained image-text alignment and anatomy-pathology prompts to generate highly detailed and accurate synthetic medical images. Our method integrates advanced natural language processing techniques with image generative modeling, enabling precise alignment between descriptive text prompts and the synthesized images' anatomical and pathological details. The proposed approach consists of two key components: an anatomy-pathology prompting module and a fine-grained alignment-based synthesis module. The anatomy-pathology prompting module automatically generates descriptive prompts for high-quality medical images. To further synthesize high-quality medical images from the generated prompts, the fine-grained alignment-based synthesis module pre-defines a visual codebook for the radiology dataset and performs fine-grained alignment between the codebook and generated prompts to obtain key patches as visual clues, facilitating accurate image synthesis. We validate the superiority of our method through experiments on public chest X-ray datasets and demonstrate that our synthetic images preserve accurate semantic information, making them valuable for various medical applications., Comment: 5 figures

Sora is a text-to-video generative AI model, released by OpenAI in February 2024. The model is trained to generate videos of realistic or imaginative scenes from text instructions and show potential in simulating the physical world. Based on public technical reports and reverse engineering, this paper presents a comprehensive review of the model's background, related technologies, applications, remaining challenges, and future directions of text-to-video AI models. We first trace Sora's development and investigate the underlying technologies used to build this "world simulator". Then, we describe in detail the applications and potential impact of Sora in multiple industries ranging from film-making and education to marketing. We discuss the main challenges and limitations that need to be addressed to widely deploy Sora, such as ensuring safe and unbiased video generation. Lastly, we discuss the future development of Sora and video generation models in general, and how advancements in the field could enable new ways of human-AI interaction, boosting productivity and creativity of video generation., Comment: 37 pages, 18 figures; GitHub: https://github.com/lichao-sun/SoraReview

As Large Language Models (LLMs) are integrated into critical real-world applications, their strategic and logical reasoning abilities are increasingly crucial. This paper evaluates LLMs' reasoning abilities in competitive environments through game-theoretic tasks, e.g., board and card games that require pure logic and strategic reasoning to compete with opponents. We first propose GTBench, a language-driven environment composing 10 widely recognized tasks, across a comprehensive game taxonomy: complete versus incomplete information, dynamic versus static, and probabilistic versus deterministic scenarios. Then, we (1) Characterize the game-theoretic reasoning of LLMs; and (2) Perform LLM-vs.-LLM competitions as reasoning evaluation. We observe that (1) LLMs have distinct behaviors regarding various gaming scenarios; for example, LLMs fail in complete and deterministic games yet they are competitive in probabilistic gaming scenarios; (2) Most open-source LLMs, e.g., CodeLlama-34b-Instruct and Llama-2-70b-chat, are less competitive than commercial LLMs, e.g., GPT-4, in complex games, yet the recently released Llama-3-70b-Instruct makes up for this shortcoming. In addition, code-pretraining greatly benefits strategic reasoning, while advanced reasoning methods such as Chain-of-Thought (CoT) and Tree-of-Thought (ToT) do not always help. We further characterize the game-theoretic properties of LLMs, such as equilibrium and Pareto Efficiency in repeated games. Detailed error profiles are provided for a better understanding of LLMs' behavior. We hope our research provides standardized protocols and serves as a foundation to spur further explorations in the strategic reasoning of LLMs., Comment: 26 pages; the first two authors contributed equally; GTBench HF Leaderboard: https://huggingface.co/spaces/GTBench/GTBench

The development of large vision-language models (LVLMs) offers the potential to address challenges faced by traditional multimodal recommendations thanks to their proficient understanding of static images and textual dynamics. However, the application of LVLMs in this field is still limited due to the following complexities: First, LVLMs lack user preference knowledge as they are trained from vast general datasets. Second, LVLMs suffer setbacks in addressing multiple image dynamics in scenarios involving discrete, noisy, and redundant image sequences. To overcome these issues, we propose the novel reasoning scheme named Rec-GPT4V: Visual-Summary Thought (VST) of leveraging large vision-language models for multimodal recommendation. We utilize user history as in-context user preferences to address the first challenge. Next, we prompt LVLMs to generate item image summaries and utilize image comprehension in natural language space combined with item titles to query the user preferences over candidate items. We conduct comprehensive experiments across four datasets with three LVLMs: GPT4-V, LLaVa-7b, and LLaVa-13b. The numerical results indicate the efficacy of VST., Comment: under review

Multimodal Large Language Models (MLLMs) have gained significant attention recently, showing remarkable potential in artificial general intelligence. However, assessing the utility of MLLMs presents considerable challenges, primarily due to the absence of multimodal benchmarks that align with human preferences. Drawing inspiration from the concept of LLM-as-a-Judge within LLMs, this paper introduces a novel benchmark, termed MLLM-as-a-Judge, to assess the ability of MLLMs in assisting judges across diverse modalities, encompassing three distinct tasks: Scoring Evaluation, Pair Comparison, and Batch Ranking. Our study reveals that, while MLLMs demonstrate remarkable human-like discernment in Pair Comparison, there is a significant divergence from human preferences in Scoring Evaluation and Batch Ranking. Furthermore, a closer examination reveals persistent challenges in the judgment capacities of LLMs, including diverse biases, hallucinatory responses, and inconsistencies in judgment, even in advanced models such as GPT-4V. These findings emphasize the pressing need for enhancements and further research efforts to be undertaken before regarding MLLMs as fully reliable evaluators. In light of this, we advocate for additional efforts dedicated to supporting the continuous development within the domain of MLLM functioning as judges. The code and dataset are publicly available at our project homepage: \url{https://mllm-judge.github.io/}., Comment: ICML 2024 (Oral)

Do large language models (LLMs) exhibit any forms of awareness similar to humans? In this paper, we introduce AwareBench, a benchmark designed to evaluate awareness in LLMs. Drawing from theories in psychology and philosophy, we define awareness in LLMs as the ability to understand themselves as AI models and to exhibit social intelligence. Subsequently, we categorize awareness in LLMs into five dimensions, including capability, mission, emotion, culture, and perspective. Based on this taxonomy, we create a dataset called AwareEval, which contains binary, multiple-choice, and open-ended questions to assess LLMs' understandings of specific awareness dimensions. Our experiments, conducted on 13 LLMs, reveal that the majority of them struggle to fully recognize their capabilities and missions while demonstrating decent social intelligence. We conclude by connecting awareness of LLMs with AI alignment and safety, emphasizing its significance to the trustworthy and ethical development of LLMs. Our dataset and code are available at https://github.com/HowieHwong/Awareness-in-LLM.

Collaborative learning (CL) is a distributed learning framework that aims to protect user privacy by allowing users to jointly train a model by sharing their gradient updates only. However, gradient inversion attacks (GIAs), which recover users' training data from shared gradients, impose severe privacy threats to CL. Existing defense methods adopt different techniques, e.g., differential privacy, cryptography, and perturbation defenses, to defend against the GIAs. Nevertheless, all current defense methods suffer from a poor trade-off between privacy, utility, and efficiency. To mitigate the weaknesses of existing solutions, we propose a novel defense method, Dual Gradient Pruning (DGP), based on gradient pruning, which can improve communication efficiency while preserving the utility and privacy of CL. Specifically, DGP slightly changes gradient pruning with a stronger privacy guarantee. And DGP can also significantly improve communication efficiency with a theoretical analysis of its convergence and generalization. Our extensive experiments show that DGP can effectively defend against the most powerful GIAs and reduce the communication cost without sacrificing the model's utility.

We present the Radiation Oncology NLP Database (ROND), the first dedicated Natural Language Processing (NLP) dataset for radiation oncology, an important medical specialty that has received limited attention from the NLP community in the past. With the advent of Artificial General Intelligence (AGI), there is an increasing need for specialized datasets and benchmarks to facilitate research and development. ROND is specifically designed to address this gap in the domain of radiation oncology, a field that offers many opportunities for NLP exploration. It encompasses various NLP tasks including Logic Reasoning, Text Classification, Named Entity Recognition (NER), Question Answering (QA), Text Summarization, and Patient-Clinician Conversations, each with a distinct focus on radiation oncology concepts and application cases. In addition, we have developed an instruction-tuning dataset consisting of over 20k instruction pairs (based on ROND) and trained a large language model, CancerChat. This serves to demonstrate the potential of instruction-tuning large language models within a highly-specialized medical domain. The evaluation results in this study could serve as baseline results for future research. ROND aims to stimulate advancements in radiation oncology and clinical NLP by offering a platform for testing and improving algorithms and models in a domain-specific context. The ROND dataset is a joint effort of multiple U.S. health institutions. The data is available at https://github.com/zl-liu/Radiation-Oncology-NLP-Database., Comment: 10 pages, 7 figures, 6 tables

With the rapid development and widespread application of Large Language Models (LLMs), the use of Machine-Generated Text (MGT) has become increasingly common, bringing with it potential risks, especially in terms of quality and integrity in fields like news, education, and science. Current research mainly focuses on purely MGT detection without adequately addressing mixed scenarios, including AI-revised Human-Written Text (HWT) or human-revised MGT. To tackle this challenge, we define mixtext, a form of mixed text involving both AI and human-generated content. Then, we introduce MixSet, the first dataset dedicated to studying these mixtext scenarios. Leveraging MixSet, we executed comprehensive experiments to assess the efficacy of prevalent MGT detectors in handling mixtext situations, evaluating their performance in terms of effectiveness, robustness, and generalization. Our findings reveal that existing detectors struggle to identify mixtext, particularly in dealing with subtle modifications and style adaptability. This research underscores the urgent need for more fine-grain detectors tailored for mixtext, offering valuable insights for future research. Code and Models are available at https://github.com/Dongping-Chen/MixSet., Comment: Accepted by NAACL 2024

Large language models (LLMs), exemplified by ChatGPT, have gained considerable attention for their excellent natural language processing capabilities. Nonetheless, these LLMs present many challenges, particularly in the realm of trustworthiness. Therefore, ensuring the trustworthiness of LLMs emerges as an important topic. This paper introduces TrustLLM, a comprehensive study of trustworthiness in LLMs, including principles for different dimensions of trustworthiness, established benchmark, evaluation, and analysis of trustworthiness for mainstream LLMs, and discussion of open challenges and future directions. Specifically, we first propose a set of principles for trustworthy LLMs that span eight different dimensions. Based on these principles, we further establish a benchmark across six dimensions including truthfulness, safety, fairness, robustness, privacy, and machine ethics. We then present a study evaluating 16 mainstream LLMs in TrustLLM, consisting of over 30 datasets. Our findings firstly show that in general trustworthiness and utility (i.e., functional effectiveness) are positively related. Secondly, our observations reveal that proprietary LLMs generally outperform most open-source counterparts in terms of trustworthiness, raising concerns about the potential risks of widely accessible open-source LLMs. However, a few open-source LLMs come very close to proprietary ones. Thirdly, it is important to note that some LLMs may be overly calibrated towards exhibiting trustworthiness, to the extent that they compromise their utility by mistakenly treating benign prompts as harmful and consequently not responding. Finally, we emphasize the importance of ensuring transparency not only in the models themselves but also in the technologies that underpin trustworthiness. Knowing the specific trustworthy technologies that have been employed is crucial for analyzing their effectiveness., Comment: This work is still under work and we welcome your contribution