The widespread adoption of large language models (LLMs) underscores the urgent need to ensure their fairness. However, LLMs frequently present dominant viewpoints while ignoring alternative perspectives from minority parties, resulting in potential biases. We hypothesize that these fairness-violating behaviors occur because LLMs express their viewpoints using a human personality that represents the majority of training data. In response to this, we validate that prompting LLMs with specific roles can allow LLMs to express diverse viewpoints. Building on this insight and observation, we develop FairThinking, a pipeline designed to automatically generate roles that enable LLMs to articulate diverse perspectives for fair expressions. To evaluate FairThinking, we create a dataset with a thousand items covering three fairness-related topics and conduct experiments on GPT-3.5, GPT-4, Llama2, and Mistral to demonstrate its superior performance.
The emerging success of large language models (LLMs) heavily relies on collecting abundant training data from external (untrusted) sources. Despite substantial efforts devoted to data cleaning and curation, well-constructed LLMs have been reported to suffer from copyright infringement, data poisoning, and/or privacy violations, which would impede practical deployment of LLMs. In this study, we propose a simple and easily implementable method for purifying LLMs from the negative effects caused by uncurated data, namely, through ensembling LLMs with benign and small language models (SLMs). Aside from theoretical guarantees, we perform comprehensive experiments to empirically confirm the efficacy of ensembling LLMs with SLMs, which can effectively preserve the performance of LLMs while mitigating issues such as copyright infringement, data poisoning, and privacy violations.
Although Large Language Models (LLMs) have demonstrated strong performance on a wide range of tasks, they still face reliability challenges such as hallucination. Previous studies reveal that highly capable LLMs like GPT-4 are effective in judging the reliability of individual responses, while less capable ones are often tuned to evaluate the relative reliability of responses to the same query. To enable less capable LLMs to effectively judge the reliability of individual responses, we propose a novel method named $\textit{Meta}$ $\textit{Ranking}$ (MR). Unlike previous methods, which assess the response directly, we achieve the judgement by comparing the target query-response pair with reference query-response pairs. We found its remarkable effectiveness in error detection for LLM responses on reasoning tasks, where less capable LLMs could outperform strong baselines, even without fine-tuning. We further demonstrate that MR can be used to enhance the performance of LLMs in two practical applications: query routing and iterative training data filtering. The former achieves GPT-4-turbo comparable performance with less than half the token consumption, while the latter makes the instruction-tuned LLaMA-7B and Phi-2, a 2.7B model, significantly surpass Alpaca-13B over fewer training samples, underscoring the high potential of our proposed method.
With the prevalence of text-to-image generative models, their safety becomes a critical concern. adversarial testing techniques have been developed to probe whether such models can be prompted to produce Not-Safe-For-Work (NSFW) content. However, existing solutions face several challenges, including low success rate and inefficiency. We introduce Groot, the first automated framework leveraging tree-based semantic transformation for adversarial testing of text-to-image models. Groot employs semantic decomposition and sensitive element drowning strategies in conjunction with LLMs to systematically refine adversarial prompts. Our comprehensive evaluation confirms the efficacy of Groot, which not only exceeds the performance of current state-of-the-art approaches but also achieves a remarkable success rate (93.66%) on leading text-to-image models such as DALL-E 3 and Midjourney.
State-of-the-art Large Multi-Modal Models (LMMs) have demonstrated exceptional capabilities in vision-language tasks. Despite their advanced functionalities, the performances of LMMs are still limited in challenging scenarios that require complex reasoning with multiple levels of visual information. Existing prompting techniques for LMMs focus on either improving textual reasoning or leveraging tools for image preprocessing, lacking a simple and general visual prompting scheme to promote vision-language coordination in LMMs. In this work, we propose Scaffold prompting that scaffolds coordinates to promote vision-language coordination. Specifically, Scaffold overlays a dot matrix within the image as visual information anchors and leverages multi-dimensional coordinates as textual positional references. Extensive experiments on a wide range of challenging vision-language tasks demonstrate the superiority of Scaffold over GPT-4V with the textual CoT prompting. Our code is released in https://github.com/leixy20/Scaffold.
With the development of LLMs, the security threats of LLMs are getting more and more attention. Numerous jailbreak attacks have been proposed to assess the security defense of LLMs. Current jailbreak attacks primarily utilize scenario camouflage techniques. However their explicitly mention of malicious intent will be easily recognized and defended by LLMs. In this paper, we propose an indirect jailbreak attack approach, Puzzler, which can bypass the LLM's defense strategy and obtain malicious response by implicitly providing LLMs with some clues about the original malicious query. In addition, inspired by the wisdom of "When unable to attack, defend" from Sun Tzu's Art of War, we adopt a defensive stance to gather clues about the original malicious query through LLMs. Extensive experimental results show that Puzzler achieves a query success rate of 96.6% on closed-source LLMs, which is 57.9%-82.7% higher than baselines. Furthermore, when tested against the state-of-the-art jailbreak detection approaches, Puzzler proves to be more effective at evading detection compared to baselines.
Graph contrastive learning (GCL) has emerged as a pivotal technique in the domain of graph representation learning. A crucial aspect of effective GCL is the caliber of generated positive and negative samples, which is intrinsically dictated by their resemblance to the original data. Nevertheless, precise control over similarity during sample generation presents a formidable challenge, often impeding the effective discovery of representative graph patterns. To address this challenge, we propose an innovative framework: Adversarial Curriculum Graph Contrastive Learning (ACGCL), which capitalizes on the merits of pair-wise augmentation to engender graph-level positive and negative samples with controllable similarity, alongside subgraph contrastive learning to discern effective graph patterns therein. Within the ACGCL framework, we have devised a novel adversarial curriculum training methodology that facilitates progressive learning by sequentially increasing the difficulty of distinguishing the generated samples. Notably, this approach transcends the prevalent sparsity issue inherent in conventional curriculum learning strategies by adaptively concentrating on more challenging training data. Finally, a comprehensive assessment of ACGCL is conducted through extensive experiments on six well-known benchmark datasets, wherein ACGCL conspicuously surpasses a set of state-of-the-art baselines.
LLM hallucination, i.e. generating factually incorrect yet seemingly convincing answers, is currently a major threat to the trustworthiness and reliability of LLMs. The first step towards solving this complicated problem is to measure it. However, existing hallucination metrics require to have a benchmark dataset with gold-standard answers, i.e. "best" or "correct" answers written by humans. Such requirement makes hallucination measurement costly and prone to human errors. In this work, we propose Factualness Evaluations via Weighting LLMs (FEWL), the first hallucination metric that is specifically designed for the scenario when gold-standard answers are absent. FEWL leverages the answers from off-the-shelf LLMs that serve as a proxy of gold-standard answers. The key challenge is how to quantify the expertise of reference LLMs resourcefully. We show FEWL has certain theoretical guarantees and demonstrate empirically it gives more accurate hallucination measures than naively using reference LLMs. We also show how to leverage FEWL to reduce hallucination through both in-context learning and supervised finetuning. Last, we build a large-scale benchmark dataset to facilitate LLM hallucination research.
We explore machine unlearning (MU) in the domain of large language models (LLMs), referred to as LLM unlearning. This initiative aims to eliminate undesirable data influence (e.g., sensitive or illegal information) and the associated model capabilities, while maintaining the integrity of essential knowledge generation and not affecting causally unrelated information. We envision LLM unlearning becoming a pivotal element in the life-cycle management of LLMs, potentially standing as an essential foundation for developing generative AI that is not only safe, secure, and trustworthy, but also resource-efficient without the need of full retraining. We navigate the unlearning landscape in LLMs from conceptual formulation, methodologies, metrics, and applications. In particular, we highlight the often-overlooked aspects of existing LLM unlearning research, e.g., unlearning scope, data-model interaction, and multifaceted efficacy assessment. We also draw connections between LLM unlearning and related areas such as model editing, influence functions, model explanation, adversarial training, and reinforcement learning. Furthermore, we outline an effective assessment framework for LLM unlearning and explore its applications in copyright and privacy safeguards and sociotechnical harm reduction.
The rapid progress of foundation models has led to the prosperity of autonomous agents, which leverage the universal capabilities of foundation models to conduct reasoning, decision-making, and environmental interaction. However, the efficacy of agents remains limited when operating in intricate, realistic environments. In this work, we introduce the principles of $\mathbf{U}$nified $\mathbf{A}$lignment for $\mathbf{A}$gents ($\mathbf{UA}^2$), which advocate for the simultaneous alignment of agents with human intentions, environmental dynamics, and self-constraints such as the limitation of monetary budgets. From the perspective of $\mathbf{UA}^2$, we review the current agent research and highlight the neglected factors in existing agent benchmarks and method candidates. We also conduct proof-of-concept studies by introducing realistic features to WebShop, including user profiles to demonstrate intentions, personalized reranking for complex environmental dynamics, and runtime cost statistics to reflect self-constraints. We then follow the principles of $\mathbf{UA}^2$ to propose an initial design of our agent, and benchmark its performance with several candidate baselines in the retrofitted WebShop. The extensive experimental results further prove the importance of the principles of $\mathbf{UA}^2$. Our research sheds light on the next steps of autonomous agent research with improved general problem-solving abilities.