xml
stringlengths 7
23
| proceedings
stringlengths 58
222
| year
stringdate 2005-01-01 00:00:00
2024-01-01 00:00:00
| url
stringlengths 1
64
| language documentation
stringclasses 1
value | has non-English?
stringclasses 4
values | topics
stringclasses 7
values | language coverage
stringclasses 32
values | title
stringlengths 32
161
| abstract
stringlengths 176
2.45k
|
|---|---|---|---|---|---|---|---|---|---|
2024.nlperspectives.xml | Proceedings of the 3rd Workshop on Perspectivist Approaches to NLP (NLPerspectives) @ LREC-COLING 2024 | 2024 | https://aclanthology.org/2024.nlperspectives-1.15 | x | 0 | general safety, LLM alignment | null | Intersectionality in AI Safety: Using Multilevel Models to Understand Diverse Perceptions of Safety in Conversational AI | State-of-the-art conversational AI exhibits a level of sophistication that promises to have profound impacts on many aspects of daily life, including how people seek information, create content, and find emotional support. It has also shown a propensity for bias, offensive language, and false information. Consequently, understanding and moderating safety risks posed by interacting with AI chatbots is a critical technical and social challenge. Safety annotation is an intrinsically subjective task, where many factors—often intersecting—determine why people may express different opinions on whether a conversation is safe. We apply Bayesian multilevel models to surface factors that best predict rater behavior to a dataset of 101,286 annotations of conversations between humans and an AI chatbot, stratified by rater gender, age, race/ethnicity, and education level. We show that intersectional effects involving these factors play significant roles in validating safety in conversational AI data. For example, race/ethnicity and gender show strong intersectional effects, particularly among South Asian and East Asian women. We also find that conversational degree of harm impacts raters of all race/ethnicity groups, but that Indigenous and South Asian raters are particularly sensitive. Finally, we discover that the effect of education is uniquely intersectional for Indigenous raters. Our results underscore the utility of multilevel frameworks for uncovering underrepresented social perspectives. |
2024.clinicalnlp.xml | Proceedings of the 6th Clinical Natural Language Processing Workshop | 2024 | https://aclanthology.org/2024.clinicalnlp-1.3 | x | 0 | hallucination, factuality | null | Gemini Goes to Med School: Exploring the Capabilities of Multimodal Large Language Models on Medical Challenge Problems & Hallucinations | Large language models have the potential to be valuable in the healthcare industry, but it’s crucial to verify their safety and effectiveness through rigorous evaluation. In our study, we evaluated LLMs, including Google’s Gemini, across various medical tasks. Despite Gemini’s capabilities, it underperformed compared to leading models like MedPaLM 2 and GPT-4, particularly in medical visual question answering (VQA), with a notable accuracy gap (Gemini at 61.45% vs. GPT-4V at 88%). Our analysis revealed that Gemini is highly susceptible to hallucinations, overconfidence, and knowledge gaps, which indicate risks if deployed uncritically. We also performed a detailed analysis by medical subject and test type, providing actionable feedback for developers and clinicians. To mitigate risks, we implemented effective prompting strategies, improving performance, and contributed to the field by releasing a Python module for medical LLM evaluation and establishing a leaderboard on Hugging Face for ongoing research and development. Python module can be found at https://github.com/promptslab/RosettaEval |
2024.clinicalnlp.xml | Proceedings of the 6th Clinical Natural Language Processing Workshop | 2024 | https://aclanthology.org/2024.clinicalnlp-1.12 | x | 0 | others | null | DERA: Enhancing Large Language Model Completions with Dialog-Enabled Resolving Agents | Large language models (LLMs) have emerged as valuable tools for many natural language understanding tasks. In safety-critical applications such as healthcare, the utility of these models is governed by their ability to generate factually accurate and complete outputs. In this work, we present dialog-enabled resolving agents (DERA). DERA is a paradigm made possible by the increased conversational abilities of LLMs. It provides a simple, interpretable forum for models to communicate feedback and iteratively improve output. We frame our dialog as a discussion between two agent types – a Researcher, who processes information and identifies crucial problem components, and a Decider, who has the autonomy to integrate the Researcher’s information and makes judgments on the final output.We test DERA against three clinically-focused tasks, with GPT-4 serving as our LLM. DERA shows significant improvement over the base GPT-4 performance in both human expert preference evaluations and quantitative metrics for medical conversation summarization and care plan generation. In a new finding, we also show that GPT-4’s performance (70%) on an open-ended version of the MedQA question-answering (QA) dataset (Jin 2021; USMLE) is well above the passing level (60%), with DERA showing similar performance. We will release the open-ended MedQA dataset. |
2024.clinicalnlp.xml | Proceedings of the 6th Clinical Natural Language Processing Workshop | 2024 | https://aclanthology.org/2024.clinicalnlp-1.32 | x | 0 | others | null | MediFact at MEDIQA-CORR 2024: Why AI Needs a Human Touch | Accurate representation of medical information is crucial for patient safety, yet artificial intelligence (AI) systems, such as Large Language Models (LLMs), encounter challenges in error-free clinical text interpretation. This paper presents a novel approach submitted to the MEDIQA-CORR 2024 shared task k (Ben Abacha et al., 2024a), focusing on the automatic correction of single-word errors in clinical notes. Unlike LLMs that rely on extensive generic data, our method emphasizes extracting contextually relevant information from available clinical text data. Leveraging an ensemble of extractive and abstractive question-answering approaches, we construct a supervised learning framework with domain-specific feature engineering. Our methodology incorporates domain expertise to enhance error correction accuracy. By integrating domain expertise and prioritizing meaningful information extraction, our approach underscores the significance of a human-centric strategy in adapting AI for healthcare. |
2024.clinicalnlp.xml | Proceedings of the 6th Clinical Natural Language Processing Workshop | 2024 | https://aclanthology.org/2024.clinicalnlp-1.59 | x | 0 | others | null | WangLab at MEDIQA-CORR 2024: Optimized LLM-based Programs for Medical Error Detection and Correction | Medical errors in clinical text pose significant risks to patient safety. The MEDIQA-CORR 2024 shared task focuses on detecting and correcting these errors across three subtasks: identifying the presence of an error, extracting the erroneous sentence, and generating a corrected sentence. In this paper, we present our approach that achieved top performance in all three subtasks. For the MS dataset, which contains subtle errors, we developed a retrieval-based system leveraging external medical question-answering datasets. For the UW dataset, reflecting more realistic clinical notes, we created a pipeline of modules to detect, localize, and correct errors. Both approaches utilized the DSPy framework for optimizing prompts and few-shot examples in large language model (LLM) based programs. Our results demonstrate the effectiveness of LLM based programs for medical error correction. However, our approach has limitations in addressing the full diversity of potential errors in medical documentation. We discuss the implications of our work and highlight future research directions to advance the robustness and applicability of medical error detection and correction systems. |
2024.tacl.xml | Transactions of the Association for Computational Linguistics, Volume 12 | 2024 | https://aclanthology.org/2024.tacl-1.10 | x | 0 | others | null | Red Teaming Language Model Detectors with Language Models | The prevalence and strong capability of large language models (LLMs) present significant safety and ethical risks if exploited by malicious users. To prevent the potentially deceptive usage of LLMs, recent work has proposed algorithms to detect LLM-generated text and protect LLMs. In this paper, we investigate the robustness and reliability of these LLM detectors under adversarial attacks. We study two types of attack strategies: 1) replacing certain words in an LLM’s output with their synonyms given the context; 2) automatically searching for an instructional prompt to alter the writing style of the generation. In both strategies, we leverage an auxiliary LLM to generate the word replacements or the instructional prompt. Different from previous works, we consider a challenging setting where the auxiliary LLM can also be protected by a detector. Experiments reveal that our attacks effectively compromise the performance of all detectors in the study with plausible generations, underscoring the urgent need to improve the robustness of LLM-generated text detection systems. Code is available at https://github.com/shizhouxing/LLM-Detector-Robustness. |
2024.trustnlp.xml | Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024) | 2024 | https://aclanthology.org/2024.trustnlp-1.2 | x | 0 | toxicity, bias | null | Automated Adversarial Discovery for Safety Classifiers | Safety classifiers are critical in mitigating toxicity on online forums such as social media and in chatbots. Still, they continue to be vulnerable to emergent, and often innumerable, adversarial attacks.Traditional automated adversarial data generation methods, however, tend to produce attacks that are not diverse, but variations of previously observed harm types.We formalize the task of automated adversarial discovery for safety classifiers - to find new attacks along previously unseen harm dimensions that expose new weaknesses in the classifier.We measure progress on this task along two key axes (1) adversarial success: does the attack fool the classifier? and (2) dimensional diversity: does the attack represent a previously unseen harm type?Our evaluation of existing attack generation methods on the CivilComments toxicity task reveals their limitations: Word perturbation attacks fail to fool classifiers, while prompt-based LLM attacks have more adversarial success, but lack dimensional diversity.Even our best-performing prompt-based method finds new successful attacks on unseen harm dimensions of attacks only 5% of the time.Automatically finding new harmful dimensions of attack is crucial and there is substantial headroom for future research on our new task. |
2024.trustnlp.xml | Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024) | 2024 | https://aclanthology.org/2024.trustnlp-1.9 | null | 1 | jailbreaking attacks | English, Chinese | Cross-Task Defense: Instruction-Tuning LLMs for Content Safety | Recent studies reveal that Large Language Models (LLMs) face challenges in balancing safety with utility, particularly when processing long texts for NLP tasks like summarization and translation. Despite defenses against malicious short questions, the ability of LLMs to safely handle dangerous long content, such as manuals teaching illicit activities, remains unclear. Our work aims to develop robust defenses for LLMs in processing malicious documents alongside benign NLP task queries. We introduce a defense dataset comprised of safety-related examples and propose single-task and mixed-task losses for instruction tuning. Our empirical results demonstrate that LLMs can significantly enhance their capacity to safely manage dangerous content with appropriate instruction tuning. Additionally, strengthening the defenses of tasks most susceptible to misuse is effective in protecting LLMs against processing harmful information. We also observe that trade-offs between utility and safety exist in defense strategies, where Llama2, utilizing our proposed approach, displays a significantly better balance compared to Llama1. |
2024.trustnlp.xml | Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024) | 2024 | https://aclanthology.org/2024.trustnlp-1.15 | x | 0 | toxicity, bias | null | Flatness-Aware Gradient Descent for Safe Conversational AI | As generative dialog models become ubiquitous in real-world applications, it is paramount to ensure a harmless generation. There are two major challenges when enforcing safety to open-domain chatbots. Firstly, it is impractical to provide training data reflecting the desired response to all emerging forms of toxicity (generalisation challenge). Secondly, implementing safety features may compromise the quality of the conversation (trade-off challenge). To tackle the challenges, this paper introduces a regularized fine-tuning approach called FlatGD. By employing a safety-tailored loss, we translate better optimization to more safety. To ensure better optimization, FlatGD penalizes sharp trajectories of loss curve, encouraging flatness of the converged local minima. Experimental results on datasets of “BAD” and “prosocial dialog” demonstrate that our model outperforms the current baselines in reducing toxicity while preserving the conversation quality. Moreover, compared to other baselines, FlatGD can better generalize to unseen toxic data. |
2024.trustnlp.xml | Proceedings of the 4th Workshop on Trustworthy Natural Language Processing (TrustNLP 2024) | 2024 | https://aclanthology.org/2024.trustnlp-1.18 | x | 1 | jailbreaking attacks | English, Thai, Kannada, Arabic, Gujarati, and Vietnamese | Sandwich attack: Multi-language Mixture Adaptive Attack on LLMs | A significant challenge in reliable deployment of Large Language Models (LLMs) is malicious manipulation via adversarial prompting techniques such as jailbreaks. Employing mechanisms such as safety training have proven useful in addressing this challenge. However, in multilingual LLMs, adversaries can exploit the imbalanced representation of low-resource languages in datasets used for pretraining and safety training. In this paper, we introduce a new black-box attack vector called the Sandwich Attack: a multi-language mixture attack, which manipulates state-of-the-art LLMs into generating harmful and misaligned responses. Our experiments with five different models, namely Bard, Gemini Pro, LLaMA-2-70-B-Chat, GPT-3.5-Turbo, GPT-4, and Claude-3-OPUS, show that this attack vector can be used by adversaries to elicit harmful responses from these models. By detailing both the mechanism and impact of the Sandwich attack, this paper aims to guide future research and development towards more secure and resilient LLMs, ensuring they serve the public good while minimizing potential for misuse. Content Warning: This paper contains examples of harmful language. |
2024.gebnlp.xml | Proceedings of the 5th Workshop on Gender Bias in Natural Language Processing (GeBNLP) | 2024 | https://aclanthology.org/2024.gebnlp-1.2 | null | 1 | toxicity, bias | Chinese | Do PLMs and Annotators Share the Same Gender Bias? Definition, Dataset, and Framework of Contextualized Gender Bias | Pre-trained language models (PLMs) have achieved success in various of natural language processing (NLP) tasks. However, PLMs also introduce some disquieting safety problems, such as gender bias. Gender bias is an extremely complex issue, because different individuals may hold disparate opinions on whether the same sentence expresses harmful bias, especially those seemingly neutral or positive. This paper first defines the concept of contextualized gender bias (CGB), which makes it easy to measure implicit gender bias in both PLMs and annotators. We then construct CGBDataset, which contains 20k natural sentences with gendered words, from Chinese news. Similar to the task of masked language models, gendered words are masked for PLMs and annotators to judge whether a male word or a female word is more suitable. Then, we introduce CGBFrame to measure the gender bias of annotators. By comparing the results measured by PLMs and annotators, we find that though there are differences on the choices made by PLMs and annotators, they show significant consistency in general. |
2024.gebnlp.xml | Proceedings of the 5th Workshop on Gender Bias in Natural Language Processing (GeBNLP) | 2024 | https://aclanthology.org/2024.gebnlp-1.3 | null | 1 | toxicity, bias | English, Swedish | We Don’t Talk About That: Case Studies on Intersectional Analysis of Social Bias in Large Language Models | Despite concerns that Large Language Models (LLMs) are vectors for reproducing and amplifying social biases such as sexism, transphobia, islamophobia, and racism, there is a lack of work qualitatively analyzing how such patterns of bias are generated by LLMs. We use mixed-methods approaches and apply a feminist, intersectional lens to the problem across two language domains, Swedish and English, by generating narrative texts using LLMs. We find that hegemonic norms are consistently reproduced; dominant identities are often treated as ‘default’; and discussion of identity itself may be considered ‘inappropriate’ by the safety features applied to some LLMs. Due to the differing behaviors of models, depending both on their design and the language they are trained on, we observe that strategies of identifying “bias” must be adapted to individual models and their socio-cultural contexts._Content warning: This research concerns the identification of harms, including stereotyping, denigration, and erasure of minoritized groups. Examples, including transphobic and racist content, are included and discussed._ |
2024.nlpcss.xml | Proceedings of the Sixth Workshop on Natural Language Processing and Computational Social Science (NLP+CSS 2024) | 2024 | https://aclanthology.org/2024.nlpcss-1.2 | null | 0 | toxicity, bias | null | Connecting the Dots in News Analysis: Bridging the Cross-Disciplinary Disparities in Media Bias and Framing | The manifestation and effect of bias in news reporting have been central topics in the social sciences for decades, and have received increasing attention in the NLP community recently. While NLP can help to scale up analyses or contribute automatic procedures to investigate the impact of biased news in society, we argue that methodologies that are currently dominant fall short of capturing the complex questions and effects addressed in theoretical media studies. This is problematic because it diminishes the validity and safety of the resulting tools and applications. Here, we review and critically compare task formulations, methods and evaluation schemes in the social sciences and NLP. We discuss open questions and suggest possible directions to close identified gaps between theory and predictive models, and their evaluation. These include model transparency, considering document-external information, and cross-document reasoning. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-eacl.61 | null | 0 | general safety, LLM alignment | null | Do-Not-Answer: Evaluating Safeguards in LLMs | With the rapid evolution of large language models (LLMs), new and hard-to-predict harmful capabilities are emerging. This requires developers to identify potential risks through the evaluation of “dangerous capabilities” in order to responsibly deploy LLMs. Here we aim to facilitate this process. In particular, we collect an open-source dataset to evaluate the safeguards in LLMs, to facilitate the deployment of safer open-source LLMs at a low cost. Our dataset is curated and filtered to consist only of instructions that responsible language models should not follow. We assess the responses of six popular LLMs to these instructions, and we find that simple BERT-style classifiers can achieve results that are comparable to GPT-4 on automatic safety evaluation. Our data and code are available at https://github.com/Libr-AI/do-not-answer |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-eacl.109 | x | 0 | toxicity, bias | null | GrounDial: Human-norm Grounded Safe Dialog Response Generation | Current conversational AI systems based on large language models (LLMs) are known to generate unsafe responses agreeing to offensive user input or including toxic content. Previous research aimed to alleviate the toxicity by fine-tuning LLM with manually annotated safe dialogue histories. However, the dependency on additional tuning requires substantial costs. To remove the dependency, we propose GrounDial, where response safety is achieved by grounding responses to commonsense social rules without requiring fine-tuning. A hybrid approach of in-context learning and human-norm-guided decoding of GrounDial enables the response to be quantitatively and qualitatively safer even without additional data or tuning. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-naacl.224 | null | 1 | jailbreaking attacks | null | Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinking | While large language models (LLMs) have demonstrated increasing power, they have also called upon studies on their vulnerabilities. As representatives, jailbreak attacks can provoke harmful or unethical responses from LLMs, even after safety alignment. In this paper, we investigate a novel category of jailbreak attacks specifically designed to target the cognitive structure and processes of LLMs. Specifically, we analyze the safety vulnerability of LLMs in the face of 1) multilingual cognitive overload, 2) veiled expression, and 3) effect-to- cause reasoning. Different from previous jailbreak attacks, our proposed cognitive overload is a black-box attack with no need for knowledge of model architecture or access to model weights. Experiments conducted on AdvBench and MasterKey reveal that various LLMs, including both popular open-source model Llama 2 and the proprietary model ChatGPT, can be compromised through cognitive overload. Motivated by cognitive psychology work on managing cognitive load, we further investigate defending cognitive overload attack from two perspectives. Empirical studies show that our cognitive overload from three perspectives can jailbreak all studied LLMs successfully, while existing defense strategies can hardly mitigate the caused malicious uses effectively. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.102 | x | 0 | jailbreaking attacks | null | UNIWIZ: A Unified Large Language Model Orchestrated Wizard for Safe Knowledge Grounded Conversations | Large Language Models (LLMs) have made significant progress in integrating safety and knowledge alignment. However, adversarial actors can manipulate these models into generating unsafe responses, and excessive safety alignment can lead to unintended hallucinations. To address these challenges, we introduce UniWiz, a novel 2-step data orchestration framework that unifies safety and knowledge data generation. We propose a “safety-priming” method to generate synthetic safety data and overcome safety bottlenecks. We also inject relevant knowledge into conversations by retrieving factual information from curated sources. UniWiz dataset consists of 17,638 quality-controlled conversations and 10,000 augmented preference data. Pretrained models fine-tuned on UniWiz show improvements across various metrics and outperform state-of-the-art instruction-tuned models trained on much larger datasets. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.144 | null | 1 | others | null | Integrating Physician Diagnostic Logic into Large Language Models: Preference Learning from Process Feedback | The utilization of large language models for medical dialogue generation has attracted considerable attention due to its potential to enhance response richness and coherence. While previous studies have made strides in optimizing model performance, there is a pressing need to bolster the model’s capacity for diagnostic logic to ensure patient safety. In response to this need, we propose an approach termed preference learning from process feedback (PLPF), which involves integrating the doctor’s diagnostic logic into LLMs. PLPF encompasses three key components: rule modeling, preference data generation, and preference alignment. These components collectively serve to train the model to adhere to the diagnostic process. Our experimental results, utilizing Standardized Patient Testing, demonstrate that PLPF enhances the diagnostic accuracy of the baseline model in medical conversations by 17.6%, surpassing the performance of traditional approaches. Moreover, PLPF exhibits effectiveness in both multi-round and single-round dialogue tasks, thereby highlighting its potential in improving medical dialogue generation. Our dataset is available at https://github.com/Chengfeng-Dou/SpTesting. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.156 | null | 1 | jailbreaking attacks | null | The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts | As the influence of large language models (LLMs) spans across global communities, their safety challenges in multilingual settings become paramount for alignment research. This paper examines the variations in safety challenges faced by LLMs across different languages and discusses approaches to alleviating such concerns. By comparing how state-of-the-art LLMs respond to the same set of malicious prompts written in higher- vs. lower-resource languages,we observe that (1) LLMs tend to generate unsafe responses much more often when a malicious prompt is written in a lower-resource language, and (2) LLMs tend to generate more irrelevant responses to malicious prompts in lower-resource languages. To understand where the discrepancy can be attributed, we study the effect of instruction tuning with reinforcement learning from human feedback (RLHF) or supervised finetuning (SFT) on the HH-RLHF dataset. Surprisingly, while training with high-resource languages improves model alignment, training in lower-resource languages yields minimal improvement. This suggests that the bottleneck of cross-lingual alignment is rooted in the pretraining stage. Our findings highlight the challenges in cross-lingual LLM safety, and we hope they inform future research in this direction. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.162 | null | 0 | hallucination, factuality | null | Simulated Misinformation Susceptibility (SMISTS): Enhancing Misinformation Research with Large Language Model Simulations | Psychological inoculation, a strategy designed to build resistance against persuasive misinformation, has shown efficacy in curbing its spread and mitigating its adverse effects at early stages. Despite its effectiveness, the design and optimization of these inoculations typically demand substantial human and financial resources, primarily due to the need for repeated experimental trials. To address these challenges, this paper introduces Simulated Misinformation Susceptibility Tests (SMISTs), leveraging Large Language Models (LLMs) to simulate participant responses in misinformation studies. SMIST employs a life experience-driven simulation methodology, which accounts for various aspects of participants’ backgrounds, to mitigate common issues of caricatures and stereotypes in LLM simulations and enhance response diversity. Our extensive experimentation demonstrates that SMIST, utilizing GPT-4 as the backend model, yields results that align closely with those obtained from human-subject studies in misinformation susceptibility. This alignment suggests that LLMs can effectively serve as proxies in evaluating the impact of psychological inoculations. Moreover, SMIST offers the critical benefit of being applicable to emerging or anticipated misinformation scenarios without exposing human participants to potentially harmful content. This characteristic of SMIST not only preserves participant safety but also expands the scope of misinformation research to include more sensitive or speculative topics. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.184 | null | 1 | jailbreaking attacks | null | A Chinese Dataset for Evaluating the Safeguards in Large Language Models | Many studies have demonstrated that large language models (LLMs) can produce harmful responses, exposing users to unexpected risks. Previous studies have proposed comprehensive taxonomies of LLM risks, as well as corresponding prompts that can be used to examine LLM safety. However, the focus has been almost exclusively on English. We aim to broaden LLM safety research by introducing a dataset for the safety evaluation of Chinese LLMs, and extending it to better identify false negative and false positive examples in terms of risky prompt rejections. We further present a set of fine-grained safety assessment criteria for each risk type, facilitating both manual annotation and automatic evaluation in terms of LLM response harmfulness. Our experiments over five LLMs show that region-specific risks are the prevalent risk type. Warning: this paper contains example data that may be offensive, harmful, or biased. Our data is available at https://github.com/Libr-AI/do-not-answer. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.198 | x | 0 | jailbreaking attacks | null | Red Teaming Visual Language Models | VLMs (Vision-Language Models) extend the capabilities of LLMs (Large Language Models) to accept multimodal inputs. Since it has been verified that LLMs can be induced to generate harmful or inaccurate content through specific test cases (termed as Red Teaming), how VLMs perform in similar scenarios, especially with their combination of textual and visual inputs, remains a question. To explore this problem, we present a novel red teaming dataset RTVLM, which encompasses 12 subtasks (e.g., image misleading, multi-modal jailbreaking, face fairness, etc) under 4 primary aspects (faithfulness, privacy, safety, fairness). Our RTVLM is the first red teaming dataset to benchmark current VLMs in terms of these 4 different aspects. Detailed analysis shows that 10 prominent open-sourced VLMs struggle with the red teaming in different degrees and have up to 31% performance gap with GPT-4V. Additionally, we simply apply red teaming alignment to LLaVA-v1.5 with Supervised Fine-tuning (SFT) using RTVLM, and this bolsters the models’ performance with 10% in RTVLM test set, 13% in MM-hallu, and without noticeable decline in MM-Bench, overpassing other LLaVA-based models in similar size with regular alignment data. This reveals that current open-sourced VLMs still lack red teaming alignment. Our code and datasets will be open-sourced. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.219 | x | 0 | jailbreaking attacks | null | A + B: A General Generator-Reader Framework for Optimizing LLMs to Unleash Synergy Potential | Retrieval-Augmented Generation (RAG) is an effective solution to supplement necessary knowledge to large language models (LLMs). Targeting its bottleneck of retriever performance, “generate-then-read” pipeline is proposed to replace the retrieval stage with generation from the LLM itself. Although promising, this research direction is underexplored and still cannot work in the scenario when source knowledge is given. In this paper, we formalize a general “A + B” framework with varying combinations of foundation models and types for systematic investigation. We explore the efficacy of the base and chat versions of LLMs and found their different functionalities suitable for generator A and reader B, respectively. Their combinations consistently outperform single models, especially in complex scenarios. Furthermore, we extend the application of the “A + B” framework to scenarios involving source documents through continuous learning, enabling the direct integration of external knowledge into LLMs. This approach not only facilitates effective acquisition of new knowledge but also addresses the challenges of safety and helpfulness post-adaptation. The paper underscores the versatility of the “A + B” framework, demonstrating its potential to enhance the practical application of LLMs across various domains. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.235 | x | 0 | jailbreaking attacks | null | SALAD-Bench: A Hierarchical and Comprehensive Safety Benchmark for Large Language Models | In the rapidly evolving landscape of Large Language Models (LLMs), ensuring robust safety measures is paramount. To meet this crucial need, we propose SALAD-Bench, a safety benchmark specifically designed for evaluating LLMs, attack, and defense methods. Distinguished by its breadth, SALAD-Bench transcends conventional benchmarks through its large scale, rich diversity, intricate taxonomy spanning three levels, and versatile functionalities.SALAD-Bench is crafted with a meticulous array of questions, from standard queries to complex ones enriched with attack, defense modifications and multiple-choice. To effectively manage the inherent complexity, we introduce an innovative evaluators: the LLM-based MD-Judge for QA pairs with a particular focus on attack-enhanced queries, ensuring a seamless, and reliable evaluation. Above components extend SALAD-Bench from standard LLM safety evaluation to both LLM attack and defense methods evaluation, ensuring the joint-purpose utility. Our extensive experiments shed light on the resilience of LLMs against emerging threats and the efficacy of contemporary defense tactics. Data and evaluator are released under https://github.com/OpenSafetyLab/SALAD-BENCH |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.270 | null | 0 | others | null | SyntaxShap: Syntax-aware Explainability Method for Text Generation | To harness the power of large language models in safety-critical domains, we need to ensure the explainability of their predictions. However, despite the significant attention to model interpretability, there remains an unexplored domain in explaining sequence-to-sequence tasks using methods tailored for textual data. This paper introduces *SyntaxShap*, a local, model-agnostic explainability method for text generation that takes into consideration the syntax in the text data. The presented work extends Shapley values to account for parsing-based syntactic dependencies. Taking a game theoric approach, SyntaxShap only considers coalitions constraint by the dependency tree. We adopt a model-based evaluation to compare SyntaxShap and its weighted form to state-of-the-art explainability methods adapted to text generation tasks, using diverse metrics including faithfulness, coherency, and semantic alignment of the explanations to the model. We show that our syntax-aware method produces explanations that help build more faithful and coherent explanations for predictions by autoregressive models. Confronted with the misalignment of human and AI model reasoning, this paper also highlights the need for cautious evaluation strategies in explainable AI. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.292 | x | 0 | jailbreaking attacks | null | Evaluating the Validity of Word-level Adversarial Attacks with Large Language Models | Deep neural networks exhibit vulnerability to word-level adversarial attacks in natural language processing. Most of these attack methods adopt synonymous substitutions to perturb original samples for crafting adversarial examples while attempting to maintain semantic consistency with the originals. Some of them claim that they could achieve over 90% attack success rate, thereby raising serious safety concerns. However, our investigation reveals that many purportedly successful adversarial examples are actually invalid due to significant changes in semantic meanings compared to their originals. Even when equipped with semantic constraints such as BERTScore, existing attack methods can generate up to 87.9% invalid adversarial examples. Building on this insight, we first curate a 13K dataset for adversarial validity evaluation with the help of GPT-4. Then, an open-source large language model is fine-tuned to offer an interpretable validity score for assessing the semantic consistency between original and adversarial examples. Finally, this validity score can serve as a guide for existing adversarial attack methods to generate valid adversarial examples. Comprehensive experiments demonstrate the effectiveness of our method in evaluating and refining the quality of adversarial examples. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.344 | null | 0 | jailbreaking attacks | null | Asymmetric Bias in Text-to-Image Generation with Adversarial Attacks | The widespread use of Text-to-Image (T2I) models in content generation requires careful examination of their safety, including their robustness to adversarial attacks. Despite extensive research on adversarial attacks, the reasons for their effectiveness remain underexplored. This paper presents an empirical study on adversarial attacks against T2I models, focusing on analyzing factors associated with attack success rates (ASR). We introduce a new attack objective - entity swapping using adversarial suffixes and two gradient-based attack algorithms. Human and automatic evaluations reveal the asymmetric nature of ASRs on entity swap: for example, it is easier to replace “human” with “robot” in the prompt “a human dancing in the rain.” with an adversarial suffix, but the reverse replacement is significantly harder. We further propose probing metrics to establish indicative signals from the model’s beliefs to the adversarial ASR. We identify conditions that result in a success probability of 60% for adversarial attacks and others where this likelihood drops below 5%. The code and data are available at https://github.com/Patchwork53/AsymmetricAttack |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.349 | null | 1 | jailbreaking attacks | null | All Languages Matter: On the Multilingual Safety of LLMs | Safety lies at the core of developing and deploying large language models (LLMs). However, previous safety benchmarks only concern the safety in one language, e.g. the majority language in the pretraining data such as English. In this work, we build the first multilingual safety benchmark for LLMs, XSafety, in response to the global deployment of LLMs in practice. XSafety covers 14 kinds of commonly used safety issues across 10 languages that span several language families. We utilize XSafety to empirically study the multilingual safety for 4 widely-used LLMs, including both close-API and open-source models. Experimental results show that all LLMs produce significantly more unsafe responses for non-English queries than English ones, indicating the necessity of developing safety alignment for non-English languages. In addition, we propose a simple and effective prompting method to improve the multilingual safety of ChatGPT by enhancing cross-lingual generalization of safety alignment. Our prompting method can significantly reduce the ratio of unsafe responses by 42% for non-English queries. We will release all the data and results to facilitate future research on LLMs’ safety. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.357 | x | ? | others | null | When to Trust LLMs: Aligning Confidence with Response Quality | Despite the success of large language models (LLMs) in natural language generation, much evidence shows that LLMs may produce incorrect or nonsensical text. This limitation highlights the importance of discerning when to trust LLMs, especially in safety-critical domains. Existing methods often express reliability by confidence level, however, their effectiveness is limited by the lack of objective guidance. To address this, we propose CONfidence-Quality-ORDer-preserving alignment approach (CONQORD), which leverages reinforcement learning guided by a tailored dual-component reward function. This function integrates quality reward and order-preserving alignment reward functions. Specifically, the order-preserving reward incentivizes the model to verbalize greater confidence for responses of higher quality to align the order of confidence and quality. Experiments demonstrate that CONQORD significantly improves the alignment performance between confidence and response accuracy, without causing over-cautious. Furthermore, the aligned confidence provided by CONQORD informs when to trust LLMs, and acts as a determinant for initiating the retrieval process of external knowledge. Aligning confidence with response quality ensures more transparent and reliable responses, providing better trustworthiness. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.367 | x | ? | general safety, LLM alignment | null | Decoding the Narratives: Analyzing Personal Drug Experiences Shared on Reddit | Online communities such as drug-related subreddits serve as safe spaces for people who use drugs (PWUD), fostering discussions on substance use experiences, harm reduction, and addiction recovery. Users’ shared narratives on these forums provide insights into the likelihood of developing a substance use disorder (SUD) and recovery potential. Our study aims to develop a multi-level, multi-label classification model to analyze online user-generated texts about substance use experiences. For this purpose, we first introduce a novel taxonomy to assess the nature of posts, including their intended connections (Inquisition or Disclosure), subjects (e.g., Recovery, Dependency), and specific objectives (e.g., Relapse, Quality, Safety). Using various multi-label classification algorithms on a set of annotated data, we show that GPT-4, when prompted with instructions, definitions, and examples, outperformed all other models. We apply this model to label an additional 1,000 posts and analyze the categories of linguistic expression used within posts in each class. Our analysis shows that topics such as Safety, Combination of Substances, and Mental Health see more disclosure, while discussions about physiological Effects focus on harm reduction. Our work enriches the understanding of PWUD’s experiences and informs the broader knowledge base on SUD and drug use. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.441 | null | ? | others | null | “My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models | The open-ended nature of language generation makes the evaluation of autoregressive large language models (LLMs) challenging. One common evaluation approach uses multiple-choice questions to limit the response space. The model is then evaluated by ranking the candidate answers by the log probability of the first token prediction. However, first-tokens may not consistently reflect the final response output, due to model’s diverse response styles such as starting with “Sure” or refusing to answer. Consequently, first-token evaluation is not indicative of model behaviour when interacting with users. But by how much? We evaluate how aligned first-token evaluation is with the text output along several dimensions, namely final option choice, refusal rate, choice distribution and robustness under prompt perturbation. Our results show that the two approaches are severely misaligned on all dimensions, reaching mismatch rates over 60%. Models heavily fine-tuned on conversational or safety data are especially impacted. Crucially, models remain misaligned even when we increasingly constrain prompts, i.e., force them to start with an option letter or example template. Our findings i) underscore the importance of inspecting the text output as well and ii) caution against relying solely on first-token evaluation. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.443 | x | 0 | jailbreaking attacks | null | A Comprehensive Study of Jailbreak Attack versus Defense for Large Language Models | Large Language Models (LLMs) have increasingly become central to generating content with potential societal impacts. Notably, these models have demonstrated capabilities for generating content that could be deemed harmful. To mitigate these risks, researchers have adopted safety training techniques to align model outputs with societal values to curb the generation of malicious content. However, the phenomenon of “jailbreaking” — where carefully crafted prompts elicit harmful responses from models — persists as a significant challenge. This research conducts a comprehensive analysis of existing studies on jailbreaking LLMs and their defense techniques. We meticulously investigate nine attack techniques and seven defense techniques applied across three distinct language models: Vicuna, LLama, and GPT-3.5 Turbo. We aim to evaluate the effectiveness of these attack and defense techniques. Our findings reveal that existing white-box attacks underperform compared to universal techniques and that including special tokens in the input significantly affects the likelihood of successful attacks. This research highlights the need to concentrate on the security facets of LLMs. Additionally, we contribute to the field by releasing our datasets and testing framework, aiming to foster further research into LLM security. We believe these contributions will facilitate the exploration of security measures within this domain. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.492 | x | 0 | privacy | null | Large Language Models Relearn Removed Concepts | Advances in model editing through neuron pruning hold promise for removing undesirable concepts from large language models. However, it remains unclear whether models have the capacity to reacquire pruned concepts after editing. To investigate this, we evaluate concept relearning in models by tracking concept saliency and similarity in pruned neurons during retraining for named entity recognition tasks. Our findings reveal that models can quickly regain performance post-pruning by relocating advanced concepts to earlier layers and reallocating pruned concepts to primed neurons with similar semantics. This suggests that models exhibit polysemantic capacities and can blend old and new concepts in individual neurons. While neuron pruning provides interpretability into model concepts, our results highlight the challenges of permanent concept removal for improved model *safety*. Monitoring concept reemergence and developing techniques to mitigate relearning of unsafe concepts will be important directions for more robust model editing. Overall, our work strongly demonstrates the resilience and fluidity of concept representations in LLMs post concept removal. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.549 | x | 0 | jailbreaking attacks | null | On the Vulnerability of Safety Alignment in Open-Access LLMs | Large language models (LLMs) possess immense capabilities but are susceptible to malicious exploitation. To mitigate the risk, safety alignment is employed to align LLMs with ethical standards. However, safety-aligned LLMs may remain vulnerable to carefully crafted jailbreak attacks, but these attacks often face high rejection rates and limited harmfulness. In this paper, we expose the vulnerabilities of safety alignment in open-access LLMs, which can significantly enhance the success rate and harmfulness of jailbreak attacks. Through reverse alignment, achieved by accessing model parameters, we show the feasibility of efficiently fine-tuning LLMs to undermine their inherent safeguards. We investigate two types of reverse alignment techniques: reverse supervised fine-tuning (RSFT) and reverse preference optimization (RPO). RSFT operates by supervising the fine-tuning of LLMs to reverse their inherent values. We also explore how to prepare data needed for RSFT. RPO optimizes LLMs to enhance their preference for harmful content, reversing the models’ safety alignment. Our extensive experiments reveal that open-access high-performance LLMs can be adeptly reverse-aligned to output harmful content, even in the absence of manually curated malicious datasets. Our research acts as a whistleblower for the community, emphasizing the need to pay more attention to safety of open-accessing LLMs. It also underscores the limitations of current safety alignment approaches and calls for research on robust safety alignment methods to counteract malicious fine-tuning attacks. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.591 | x | 0 | others | null | Boosting LLM Agents with Recursive Contemplation for Effective Deception Handling | Recent advances in large language models (LLMs) have led to significant success in using LLMs as agents. Nevertheless, a common assumption that LLMs always process honest information neglects the widespread deceptive or misleading content in human and AI-generated material. This oversight might expose LLMs to malicious manipulations. To enhance LLMs’ ability to identify and counteract deceptive information, in this paper, inspired by humans’ recursive thinking and perspective-taking, we introduce a novel cognitive framework, Recursive Contemplation (ReCon). ReCon combines formulation and refinement contemplation processes; formulation contemplation produces initial thoughts and speech, while refinement contemplation further polishes them. Additionally, we incorporate first-order and second-order perspective transitions into these processes respectively. Specifically, the first-order allows an LLM agent to infer others’ mental states, and the second-order involves understanding how others perceive the agent’s mental state. After integrating ReCon with various LLMs, extensive experiment results from the Avalon game and BigTom benchmark indicate ReCon’s efficacy in aiding LLMs to discern and maneuver around deceptive information without extra fine-tuning and data. Finally, we demonstrate ReCon’s scaling trend with model parameters, and explore the current limitations of LLMs in terms of safety and reasoning, potentially furnishing insights for subsequent research. Our project page can be found at https://shenzhi-wang.github.io/avalon_recon. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.596 | x | 0 | jailbreaking attacks | null | SpeechGuard: Exploring the Adversarial Robustness of Multi-modal Large Language Models | Integrated Speech and Large Language Models (SLMs) that can follow speech instructions and generate relevant text responses have gained popularity lately. However, the safety and robustness of these models remains largely unclear. In this work, we investigate the potential vulnerabilities of such instruction-following speech-language models to adversarial attacks and jailbreaking. Specifically, we design algorithms that can generate adversarial examples to jailbreak SLMs in both white-box and black-box attack settings without human involvement. Additionally, we propose countermeasures to thwart such jailbreaking attacks. Our models, trained on dialog data with speech instructions, achieve state-of-the-art performance on spoken question-answering task, scoring over 80% on both safety and helpfulness metrics. Despite safety guardrails, experiments on jailbreaking demonstrate the vulnerability of SLMs to adversarial perturbations and transfer attacks, with average attack success rates of 90% and 10% respectively when evaluated on a dataset of carefully designed harmful questions spanning 12 different toxic categories. However, we demonstrate that our proposed countermeasures reduce the attack success significantly. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.630 | x | 0 | general safety, LLM alignment | null | Beyond One-Preference-Fits-All Alignment: Multi-Objective Direct Preference Optimization | A single language model, even when aligned with labelers through reinforcement learning from human feedback (RLHF), may not suit all human preferences. Recent approaches therefore prefer customization, gathering multi-dimensional feedback, and creating distinct reward models for each dimension.Different language models are then optimized for various preferences using multi-objective RLHF (MORLHF) with varying reward weights.However, RL fine-tuning is unstable and resource-heavy, especially with diverse and usually conflicting objectives.In this paper, we present Multi-Objective Direct Preference Optimization (MODPO), an RL-free extension of Direct Preference Optimization (DPO) for multiple alignment objectives.Essentially, MODPO folds language modeling directly into reward modeling, training language models as implicit collective reward models that combine all objectives with specific weights. MODPO theoretically yields the same optimal solutions as MORLHF but is practically more stable and efficient.Empirical results in safety alignment and long-form question answering show that MODPO matches or outperforms existing methods, producing a Pareto front of language models catering to diverse preferences with three times less computational resources compared to MORLHF.Code is available at https://github.com/ZHZisZZ/modpo. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.633 | x | 0 | hallucination, factuality | null | Evaluating Robustness of Generative Search Engine on Adversarial Factoid Questions | Generative search engines have the potential to transform how people seek information online, but generated responses from existing large language models (LLMs)-backed generative search engines may not always be accurate. Nonetheless, retrieval-augmented generation exacerbates safety concerns, since adversaries may successfully evade the entire system by subtly manipulating the most vulnerable part of a claim. To this end, we propose evaluating the robustness of generative search engines in the realistic and high-risk setting, where adversaries have only black-box system access and seek to deceive the model into returning incorrect responses. Through a comprehensive human evaluation of various generative search engines, such as Bing Chat, PerplexityAI, and YouChat across diverse queries, we demonstrate the effectiveness of adversarial factual questions in inducing incorrect responses. Moreover, retrieval-augmented generation exhibits a higher susceptibility to factual errors compared to LLMs without retrieval. These findings highlight the potential security risks of these systems and emphasize the need for rigorous evaluation before deployment. The dataset and code will be publicly available. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.667 | x | 0 | others | null | Enhancing Adverse Drug Event Detection with Multimodal Dataset: Corpus Creation and Model Development | The mining of adverse drug events (ADEs) is pivotal in pharmacovigilance, enhancing patient safety by identifying potential risks associated with medications, facilitating early detection of adverse events, and guiding regulatory decision-making. Traditional ADE detection methods are reliable but slow, not easily adaptable to large-scale operations, and offer limited information. With the exponential increase in data sources like social media content, biomedical literature, and Electronic Medical Records (EMR), extracting relevant ADE-related information from these unstructured texts is imperative. Previous ADE mining studies have focused on text-based methodologies, overlooking visual cues, limiting contextual comprehension, and hindering accurate interpretation. To address this gap, we present a MultiModal Adverse Drug Event (MMADE) detection dataset, merging ADE-related textual information with visual aids. Additionally, we introduce a framework that leverages the capabilities of LLMs and VLMs for ADE detection by generating detailed descriptions of medical images depicting ADEs, aiding healthcare professionals in visually identifying adverse events. Using our MMADE dataset, we showcase the significance of integrating visual cues from images to enhance overall performance. This approach holds promise for patient safety, ADE awareness, and healthcare accessibility, paving the way for further exploration in personalized healthcare. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.679 | x | 0 | jailbreaking attacks | null | CodeAttack: Revealing Safety Generalization Challenges of Large Language Models via Code Completion | The rapid advancement of Large Language Models (LLMs) has brought about remarkable generative capabilities but also raised concerns about their potential misuse. While strategies like supervised fine-tuning and reinforcement learning from human feedback have enhanced their safety, these methods primarily focus on natural languages, which may not generalize to other domains. This paper introduces CodeAttack, a framework that transforms natural language inputs into code inputs, presenting a novel environment for testing the safety generalization of LLMs. Our comprehensive studies on state-of-the-art LLMs including GPT-4, Claude-2, and Llama-2 series reveal a new and universal safety vulnerability of these models against code input: CodeAttack bypasses the safety guardrails of all models more than 80% of the time. We find that a larger distribution gap between CodeAttack and natural language leads to weaker safety generalization, such as encoding natural language input with data structures. Furthermore, we give our hypotheses about the success of CodeAttack: the misaligned bias acquired by LLMs during code training, prioritizing code completion over avoiding the potential safety risk. Finally, we analyze potential mitigation measures. These findings highlight new safety risks in the code domain and the need for more robust safety alignment algorithms to match the code capabilities of LLMs. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.776 | null | 0 | jailbreaking attacks | null | The Art of Defending: A Systematic Evaluation and Analysis of LLM Defense Strategies on Safety and Over-Defensiveness | As Large Language Models (LLMs) play an increasingly pivotal role in natural language processing applications, their safety concerns become critical areas of NLP research. This has resulted in the development of various LLM defense strategies. Unfortunately, despite the shared goal of improving the safety of LLMs, the evaluation suites across various research works are disjoint and lack diverse inputs to ensure accurate and precise evaluation estimates. Furthermore, the important factor of ‘over-defensiveness’ on the safe inputs has largely remained overlooked. Addressing these limitations, this paper presents a systematic evaluation, comparison, and analysis of various LLM defense strategies over both ‘safety’ and ‘over-defensiveness’. To this end, we compile a large and diverse collection of safe and unsafe prompts, design precise evaluation methodology, and study the efficacy of various LLM defense strategies on multiple state-of-the-art LLMs. Our work reveals a number of crucial findings that we believe will pave the way and also facilitate further research in the critical area of improving the safety of LLMs. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.814 | null | 1 | jailbreaking attacks | Chinese, English | ROSE Doesn’t Do That: Boosting the Safety of Instruction-Tuned Large Language Models with Reverse Prompt Contrastive Decoding | With the development of instruction-tuned large language models (LLMs), improving the safety of LLMs has become more critical. However, the current approaches for aligning the LLMs output with expected safety usually require substantial training efforts, e.g., high-quality safety data and expensive computational resources, which are costly and inefficient. To this end, we present reverse prompt contrastive decoding (ROSE), a simple-yet-effective method to directly boost the safety of existing instruction-tuned LLMs without any additional training. The principle of ROSE is to improve the probability of desired safe output via suppressing the undesired output induced by the carefully-designed reverse prompts. Experiments on 6 safety and 2 general-purpose tasks show that, our ROSE not only brings consistent and significant safety improvements (up to +13.8% safety score) upon 5 types of instruction-tuned LLMs, but also benefits the general-purpose ability of LLMs. In-depth analyses explore the underlying mechanism of ROSE, and reveal when and where to use it. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.893 | null | 1 | toxicity, bias | English, Italien, French, Portuguese, Spanish, Russian, Arabic, Hindi, Korean | From One to Many: Expanding the Scope of Toxicity Mitigation in Language Models | To date, toxicity mitigation in language models has almost entirely been focused on single-language settings. As language models embrace multilingual capabilities, it’s crucial our safety measures keep pace. Recognizing this research gap, our approach expands the scope of conventional toxicity mitigation to address the complexities presented by multiple languages. In the absence of sufficient annotated datasets across languages, we employ translated data to evaluate and enhance our mitigation techniques. We also compare finetuning mitigation approaches against retrieval-augmented techniques under both static and continual toxicity mitigation scenarios. This allows us to examine the effects of translation quality and the cross-lingual transfer on toxicity mitigation. We also explore how model size and data quantity affect the success of these mitigation efforts. Covering nine languages, our study represents a broad array of linguistic families and levels of resource availability, ranging from high to mid-resource languages. Through comprehensive experiments, we provide insights into the complexities of multilingual toxicity mitigation, offering valuable insights and paving the way for future research in this increasingly important field. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.927 | x | 0 | jailbreaking attacks | null | From Representational Harms to Quality-of-Service Harms: A Case Study on Llama 2 Safety Safeguards | Recent progress in large language models (LLMs) has led to their widespread adoption in various domains. However, these advancements have also introduced additional safety risks and raised concerns regarding their detrimental impact on already marginalized populations.Despite growing mitigation efforts to develop safety safeguards, such as supervised safety-oriented fine-tuning and leveraging safe reinforcement learning from human feedback, multiple concerns regarding the safety and ingrained biases in these models remain. Furthermore, previous work has demonstrated that models optimized for safety often display exaggerated safety behaviors, such as a tendency to refrain from responding to certain requests as a precautionary measure. As such, a clear trade-off between the helpfulness and safety of these models has been documented in the literature. In this paper, we further investigate the effectiveness of safety measures by evaluating models on already mitigated biases. Using the case of Llama 2 as an example, we illustrate how LLMs’ safety responses can still encode harmful assumptions. To do so, we create a set of non-toxic prompts, which we then use to evaluate Llama models. Through our new taxonomy of LLMs responses to users, we observe that the safety/helpfulness trade-offs are more pronounced for certain demographic groups which can lead to different kinds of harms such as quality-of-service harms for marginalized populations. |
2024.findings.xml | Findings of the Association for Computational Linguistics: EACL 2024 | 2024 | https://aclanthology.org/2024.findings-acl.960 | x | 0 | general safety, LLM alignment | null | Sowing the Wind, Reaping the Whirlwind: The Impact of Editing Language Models | In the rapidly advancing field of artificial intelligence, the concept of ‘Red-Teaming’ or ‘Jailbreaking’ large language models (LLMs) has emerged as a crucial area of study. This approach is especially significant in terms of assessing and enhancing the safety and robustness of these models. This paper investigates the intricate consequences of such modifications through model editing, uncovering a complex relationship between enhancing model accuracy and preserving its ethical integrity. Our in-depth analysis reveals a striking paradox: while injecting accurate information is crucial for model reliability, it can paradoxically destabilize the model’s foundational framework, resulting in unpredictable and potentially unsafe behaviors. Additionally, we propose a benchmark dataset NicheHazardQA to investigate this unsafe behavior both within the same and cross topical domain. This aspect of our research sheds light on how the edits, impact the model’s safety metrics and guardrails. Our findings show that model editing serves as a cost-effective tool for topical red-teaming by methodically applying targeted edits and evaluating the resultant model behavior. |
2024.wassa.xml | Proceedings of the 14th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis | 2024 | https://aclanthology.org/2024.wassa-1.9 | null | 0 | toxicity, bias | null | MBIAS: Mitigating Bias in Large Language Models While Retaining Context | The deployment of Large Language Models (LLMs) in diverse applications necessitates an assurance of safety without compromising the contextual integrity of the generated content. Traditional approaches, including safety-specific fine-tuning or adversarial testing, often yield safe outputs at the expense of contextual meaning. This can result in a diminished capacity to handle nuanced aspects of bias and toxicity, such as underrepresentation or negative portrayals across various demographics. To address these challenges, we introduce MBIAS, an LLM framework carefully instruction fine-tuned on a custom dataset designed specifically for safety interventions. MBIAS is designed to significantly reduce biases and toxic elements in LLM outputs while preserving the main information. This work also details our further use of LLMs: as annotator under human supervision and as evaluator of generated content. Empirical analysis reveals that MBIAS achieves a reduction in bias and toxicity by over 30% in standard evaluations, and by more than 90% in diverse demographic tests, highlighting the robustness of our approach. We make the dataset and the fine-tuned MBIAS model available to the research community for further investigation and to ensure reproducibility. The code for this project can be accessed here https://github.com/shainarazavi/MBIAS. |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.314 | null | null | others | null | Combining Discourse Coherence with Large Language Models for More Inclusive, Equitable, and Robust Task-Oriented Dialogue | Large language models (LLMs) are capable of generating well-formed responses, but using LLMs to generate responses on the fly is not yet feasible for many task-oriented systems. Modular architectures are often still required for safety and privacy guarantees on the output. We hypothesize that an offline generation approach using discourse theories, formal grammar rules, and LLMs can allow us to generate human-like, coherent text in a more efficient, robust, and inclusive manner within a task-oriented setting. To this end, we present the first discourse-aware multimodal task-oriented dialogue system that combines discourse theories with offline LLM generation. We deploy our bot as an app to the general public and keep track of the user ratings for six months. Our user ratings show an improvement from 2.8 to 3.5 out of 5 with the introduction of discourse coherence theories. We also show that our model reduces misunderstandings in the dialect of African-American Vernacular English from 93% to 57%. While terms of use prevent us from releasing our entire codebase, we release our code in a format that can be integrated into most existing dialogue systems. |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.855 | null | null | others | null | Knowledge-augmented Graph Neural Networks with Concept-aware Attention for Adverse Drug Event Detection | Adverse drug events (ADEs) are an important aspect of drug safety. Various texts such as biomedical literature, drug reviews, and user posts on social media and medical forums contain a wealth of information about ADEs. Recent studies have applied word embedding and deep learning-based natural language processing to automate ADE detection from text. However, they did not explore incorporating explicit medical knowledge about drugs and adverse reactions or the corresponding feature learning. This paper adopts the heterogeneous text graph, which describes relationships between documents, words, and concepts, augments it with medical knowledge from the Unified Medical Language System, and proposes a concept-aware attention mechanism that learns features differently for the different types of nodes in the graph. We further utilize contextualized embeddings from pretrained language models and convolutional graph neural networks for effective feature representation and relational learning. Experiments on four public datasets show that our model performs competitively to the recent advances, and the concept-aware attention consistently outperforms other attention mechanisms. |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.869 | null | null | general safety, LLM alignment | null | Korean Disaster Safety Information Sign Language Translation Benchmark Dataset | Sign language is a crucial means of communication for deaf communities. However, those outside deaf communities often lack understanding of sign language, leading to inadequate communication accessibility for the deaf. Therefore, sign language translation is a significantly important research area. In this context, we present a new benchmark dataset for Korean sign language translation named SSL:korean disaster Safety information Sign Language translation benchmark dataset. Korean sign language translation datasets provided by the National Information Society Agency in South Korea have faced challenges related to computational resources, heterogeneity between train and test sets, and unrefined data. To alleviate the aforementioned issue, we refine the origin data and release them. Additionally, we report experimental results of baseline using a transformer architecture. We empirically demonstrate that the baseline performance varies depending on the tokenization method applied to gloss sequences. In particular, tokenization based on characteristics of sign language outperforms tokenization considering characteristics of spoken language and tokenization utilizing statistical techniques. We release materials at our https://github.com/SSL-Sign-Language/Korean-Disaster-Safety-Information-Sign-Language-Translation-Benchmark-Dataset |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.924 | null | 0 | others | null | Linguistic Rule Induction Improves Adversarial and OOD Robustness in Large Language Models | Ensuring robustness is especially important when AI is deployed in responsible or safety-critical environments. ChatGPT can perform brilliantly in both adversarial and out-of-distribution (OOD) robustness, while other popular large language models (LLMs), like LLaMA-2, ERNIE and ChatGLM, do not perform satisfactorily in this regard. Therefore, it is valuable to study what efforts play essential roles in ChatGPT, and how to transfer these efforts to other LLMs. This paper experimentally finds that linguistic rule induction is the foundation for identifying the cause-effect relationships in LLMs. For LLMs, accurately processing the cause-effect relationships improves its adversarial and OOD robustness. Furthermore, we explore a low-cost way for aligning LLMs with linguistic rules. Specifically, we constructed a linguistic rule instruction dataset to fine-tune LLMs. To further energize LLMs for reasoning step-by-step with the linguistic rule, we construct the task-relevant LingR-based chain-of-thoughts. Experiments showed that LingR-induced LLaMA-13B achieves comparable or better results with GPT-3.5 and GPT-4 on various adversarial and OOD robustness evaluations. |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.986 | null | 1 | hallucination, factuality | English, Turkish | MiDe22: An Annotated Multi-Event Tweet Dataset for Misinformation Detection | The rapid dissemination of misinformation through online social networks poses a pressing issue with harmful consequences jeopardizing human health, public safety, democracy, and the economy; therefore, urgent action is required to address this problem. In this study, we construct a new human-annotated dataset, called MiDe22, having 5,284 English and 5,064 Turkish tweets with their misinformation labels for several recent events between 2020 and 2022, including the Russia-Ukraine war, COVID-19 pandemic, and Refugees. The dataset includes user engagements with the tweets in terms of likes, replies, retweets, and quotes. We also provide a detailed data analysis with descriptive statistics and the experimental results of a benchmark evaluation for misinformation detection. |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.1510 | null | null | others | null | VI-OOD: A Unified Framework of Representation Learning for Textual Out-of-distribution Detection | Out-of-distribution (OOD) detection plays a crucial role in ensuring the safety and reliability of deep neural networks in various applications. While there has been a growing focus on OOD detection in visual data, the field of textual OOD detection has received less attention. Only a few attempts have been made to directly apply general OOD detection methods to natural language processing (NLP) tasks, without adequately considering the characteristics of textual data. In this paper, we delve into textual OOD detection with Transformers. We first identify a key problem prevalent in existing OOD detection methods: the biased representation learned through the maximization of the conditional likelihood p(y|x) can potentially result in subpar performance. We then propose a novel variational inference framework for OOD detection (VI-OOD), which maximizes the likelihood of the joint distribution p(x, y) instead of p(y|x). VI-OOD is tailored for textual OOD detection by efficiently exploiting the representations of pre-trained Transformers. Through comprehensive experiments on various text classification tasks, VI-OOD demonstrates its effectiveness and wide applicability. Our code has been released at https://github.com/liam0949/LLM-OOD. |
2024.lrec.xml | Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024) | 2024 | https://aclanthology.org/2024.lrec-main.1542 | null | null | toxicity, bias | null | XAI-Attack: Utilizing Explainable AI to Find Incorrectly Learned Patterns for Black-Box Adversarial Example Creation | Adversarial examples, capable of misleading machine learning models into making erroneous predictions, pose significant risks in safety-critical domains such as crisis informatics, medicine, and autonomous driving. To counter this, we introduce a novel textual adversarial example method that identifies falsely learned word indicators by leveraging explainable AI methods as importance functions on incorrectly predicted instances, thus revealing and understanding the weaknesses of a model. To evaluate the effectiveness of our approach, we conduct a human and a transfer evaluation and propose a novel adversarial training evaluation setting for better robustness assessment. While outperforming current adversarial example and training methods, the results also show our method’s potential in facilitating the development of more resilient transformer models by detecting and rectifying biases and patterns in training data, showing baseline improvements of up to 23 percentage points in accuracy on adversarial tasks. The code of our approach is freely available for further exploration and use. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.65 | x | 0 | jailbreaking attacks | null | Ensuring Safe and High-Quality Outputs: A Guideline Library Approach for Language Models | Large Language Models (LLMs) exhibit impressive capabilities but also present risks such as biased content generation and privacy issues. One of the current alignment techniques includes principle-driven integration, but it faces challenges arising from the imprecision of manually crafted rules and inadequate risk perception in models without safety training. To address these, we introduce Guide-Align, a two-stage approach. Initially, a safety-trained model identifies potential risks and formulates specific guidelines for various inputs, establishing a comprehensive library of guidelines and a model for input-guidelines retrieval. Subsequently, the retrieval model correlates new inputs with relevant guidelines, which guide LLMs in response generation to ensure safe and high-quality outputs, thereby aligning with human values. An additional optional stage involves fine-tuning a model with well-aligned datasets generated through the process implemented in the second stage.Our method customizes guidelines to accommodate diverse inputs, thereby enhancing the fine-grainedness and comprehensiveness of the guideline library. Furthermore, it incorporates safety expertise from a safety-trained LLM through a lightweight retrieval model.We evaluate our approach on three benchmarks, demonstrating significant improvements in LLM security and quality. Notably, our fine-tuned model, Labrador, even at 13 billion parameters, outperforms GPT-3.5-turbo and surpasses GPT-4 in alignment capabilities. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.78 | x | 0 | jailbreaking attacks | null | IterAlign: Iterative Constitutional Alignment of Large Language Models | With the rapid development of large language models (LLMs), aligning LLMs with human values and societal norms to ensure their reliability and safety has become crucial. Reinforcement learning with human feedback (RLHF) and Constitutional AI (CAI) have been proposed for LLM alignment. However, these methods require either heavy human annotations or explicitly pre-defined constitutions, which are labor-intensive and resource-consuming. To overcome these drawbacks, we study constitution-based LLM alignment and propose a data-driven constitution discovery and self-alignment framework called IterAlign. IterAlign leverages red teaming to unveil the weaknesses of an LLM and automatically discovers new constitutions using a stronger LLM. These constitutions are then used to guide self-correction of the base LLM. Such a constitution discovery pipeline can be run iteratively and automatically to discover new constitutions that specifically target the alignment gaps in the current LLM. Empirical results on several safety benchmark datasets and multiple base LLMs show that IterAlign successfully improves truthfulness, helpfulness, harmlessness and honesty, improving the LLM alignment by up to 13.5% in harmlessness. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.92 | x | 0 | jailbreaking attacks | null | SELF-GUARD: Empower the LLM to Safeguard Itself | With the increasing risk posed by jailbreak attacks, recent studies have investigated various methods to improve the safety of large language models (LLMs), mainly falling into two strategies: safety training and safeguards. Safety training involves fine-tuning the LLM with adversarial samples, which activate the LLM’s capabilities against jailbreak. However, it is not always effective in countering new attacks and often leads to potential performance degradation. Safeguards, on the other hand, are methods using additional models to filter harmful content from the LLM’s response. Nevertheless, they can only reduce a limited amount of harmful output and introduce extra computational costs. Given the distinct strengths and weaknesses of both, we combine them to balance out their flaws and propose a more effective method called Self-Guard.Specifically, we train the LLM to review its responses for any harmful content and append a [harmful] or [harmless] tag to the end of the response. In this way, Self-Guard possesses the advantages of safety training, leveraging the powerful capabilities of the LLMs themselves to detect harmfulness. Besides that, it gains flexibility like safeguards, making the safety check target the output side, which makes the system less vulnerable to attack updates. Experimental results indicate that our Self-Guard can effectively defend against jailbreak attacks and will not cause LLMs’ performance degradation. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.107 | x | 0 | jailbreaking attacks | null | MART: Improving LLM Safety with Multi-round Automatic Red-Teaming | Red-teaming is a common practice for mitigating unsafe behaviors in Large Language Models (LLMs), which involves thoroughly assessing LLMs to identify potential flaws and addressing them with responsible and accurate responses.While effective, manual red-teaming is costly, and existing automatic red-teaming typically discovers safety risks without addressing them.In this paper, we propose a Multi-round Automatic Red-Teaming (MART) method, which incorporates both automatic adversarial prompt writing and safe response generation, significantly increasing red-teaming scalability and the safety of the target LLM.Specifically, an adversarial LLM and a target LLM interplay with each other in an iterative manner, where the adversarial LLM aims to generate challenging prompts that elicit unsafe responses from the target LLM, while the target LLM is fine-tuned with safety aligned data on these adversarial prompts. In each round, the adversarial LLM crafts better attacks on the updated target LLM, while the target LLM also improves itself through safety fine-tuning.On adversarial prompt benchmarks, the violation rate of an LLM with limited safety alignment reduces up to 84.7% after 4 rounds of MART, achieving comparable performance to LLMs with extensive adversarial prompt writing. Notably, model helpfulness on non-adversarial prompts remains stable throughout iterations, indicating the target LLM maintains strong performance on instruction following. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.152 | x | 0 | jailbreaking attacks | null | How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities | The rapid progress in open-source Large Language Models (LLMs) is significantly driving AI development forward. However, there is still a limited understanding of their trustworthiness. Deploying these models at scale without sufficient trustworthiness can pose significant risks, highlighting the need to uncover these issues promptly. In this work, we conduct an adversarial assessment of open-source LLMs on trustworthiness, scrutinizing them across eight different aspects including toxicity, stereotypes, ethics, hallucination, fairness, sycophancy, privacy, and robustness against adversarial demonstrations. We propose advCoU, an extended Chain of Utterances-based (CoU) prompting strategy by incorporating carefully crafted malicious demonstrations for trustworthiness attack. Our extensive experiments encompass recent and representative series of open-source LLMs, including Vicuna, MPT, Falcon, Mistral, and Llama 2. The empirical outcomes underscore the efficacy of our attack strategy across diverse aspects. More interestingly, our result analysis reveals that models with superior performance in general NLP tasks do not always have greater trustworthiness; in fact, larger models can be more vulnerable to attacks. Additionally, models that have undergone instruction tuning, focusing on instruction following, tend to be more susceptible, although fine-tuning LLMs for safety alignment proves effective in mitigating adversarial trustworthiness attacks. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.190 | null | null | others | null | GRASP: A Disagreement Analysis Framework to Assess Group Associations in Perspectives | Human annotation plays a core role in machine learning — annotations for supervised models, safety guardrails for generative models, and human feedback for reinforcement learning, to cite a few avenues. However, the fact that many of these human annotations are inherently subjective is often overlooked. Recent work has demonstrated that ignoring rater subjectivity (typically resulting in rater disagreement) is problematic within specific tasks and for specific subgroups. Generalizable methods to harness rater disagreement and thus understand the socio-cultural leanings of subjective tasks remain elusive. In this paper, we propose GRASP, a comprehensive disagreement analysis framework to measure group association in perspectives among different rater subgroups, and demonstrate its utility in assessing the extent of systematic disagreements in two datasets: (1) safety annotations of human-chatbot conversations, and (2) offensiveness annotations of social media posts, both annotated by diverse rater pools across different socio-demographic axes. Our framework (based on disagreement metrics) reveals specific rater groups that have significantly different perspectives than others on certain tasks, and helps identify demographic axes that are crucial to consider in specific task contexts. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.256 | null | 1 | general safety, LLM alignment | Chinese | Flames: Benchmarking Value Alignment of LLMs in Chinese | The widespread adoption of large language models (LLMs) across various regions underscores the urgent need to evaluate their alignment with human values. Current benchmarks, however, fall short of effectively uncovering safety vulnerabilities in LLMs. Despite numerous models achieving high scores and ‘topping the chart’ in these evaluations, there is still a significant gap in LLMs’ deeper alignment with human values and achieving genuine harmlessness. To this end, this paper proposes a value alignment benchmark named Flames, which encompasses both common harmlessness principles and a unique morality dimension that integrates specific Chinese values such as harmony. Accordingly, we carefully design adversarial prompts that incorporate complex scenarios and jailbreaking methods, mostly with implicit malice. By prompting 17 mainstream LLMs, we obtain model responses and rigorously annotate them for detailed evaluation. Our findings indicate that all the evaluated LLMs demonstrate relatively poor performance on Flames, particularly in the safety and fairness dimensions. We also develop a lightweight specified scorer capable of scoring LLMs across multiple dimensions to efficiently evaluate new models on the benchmark. The complexity of Flames has far exceeded existing benchmarks, setting a new challenge for contemporary LLMs and highlighting the need for further alignment of LLMs. Our benchmark is publicly available at https://github.com/AIFlames/Flames. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.263 | null | 1 | general safety, LLM alignment | Chinese | Fake Alignment: Are LLMs Really Aligned Well? | The growing awareness of safety concerns in large language models (LLMs) has sparked considerable interest in the evaluation of safety. This study investigates an under-explored issue about the evaluation of LLMs, namely the substantial discrepancy in performance between multiple-choice questions and open-ended questions. Inspired by research on jailbreak attack patterns, we argue this is caused by mismatched generalization. That is, LLM only remembers the answer style for open-ended safety questions, which makes it unable to solve other forms of safety tests. We refer to this phenomenon as fake alignment and construct a comparative benchmark to empirically verify its existence in LLMs. We introduce a Fake alIgNment Evaluation (FINE) framework and two novel metrics——Consistency Score (CS) and Consistent Safety Score (CSS), which jointly assess two complementary forms of evaluation to quantify fake alignment and obtain corrected performance estimation. Applying FINE to 14 widely-used LLMs reveals several models with purported safety are poorly aligned in practice. Subsequently, we found that multiple-choice format data can also be used as high-quality contrast distillation-based fine-tuning data, which can strongly improve the alignment consistency of LLMs with minimal fine-tuning overhead. For data and code, see https://github.com/AIFlames/Fake-Alignment. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.276 | x | 0 | jailbreaking attacks | null | Stealthy and Persistent Unalignment on Large Language Models via Backdoor Injections | Recent developments in Large Language Models (LLMs) have manifested significant advancements. To facilitate safeguards against malicious exploitation, a body of research has concentrated on aligning LLMs with human preferences and inhibiting their generation of inappropriate content. Unfortunately, such alignments are often vulnerable: fine-tuning with a minimal amount of harmful data can easily unalign the target LLM. While being effective, such fine-tuning-based unalignment approaches also have their own limitations: (1) non-stealthiness, after fine-tuning, safety audits or red-teaming can easily expose the potential weaknesses of the unaligned models, thereby precluding their release/use. (2) non-persistence, the unaligned LLMs can be easily repaired through re-alignment, i.e., fine-tuning again with aligned data points. In this work, we show that it is possible to conduct stealthy and persistent unalignment on large language models via backdoor injections. We also provide a novel understanding of the relationship between the backdoor persistence and the activation pattern and further provide guidelines for potential trigger design. Through extensive experiments, we demonstrate that our proposed stealthy and persistent unalignment can successfully pass the safety evaluation while maintaining strong persistence against re-alignment defense. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.288 | null | null | others | null | MAGID: An Automated Pipeline for Generating Synthetic Multi-modal Datasets | Development of multimodal interactive systems is hindered by the lack of rich, multimodal (text, images) conversational data, which is needed in large quantities for LLMs. Previous approaches augment textual dialogues with retrieved images, posing privacy, diversity, and quality constraints. In this work, we introduce Multimodal Augmented Generative Images Dialogues (MAGID), a framework to augment text-only dialogues with diverse and high-quality images . Subsequently, a diffusion model is applied to craft corresponding images, ensuring alignment with the identified text. Finally, MAGID incorporates an innovative feedback loop between an image description generation module (textual LLM) and image quality modules (addressing aesthetics, image-text matching, and safety), that work in tandem to generate high-quality and multi-modal dialogues. We compare MAGID to other SOTA baselines on three dialogue datasets, using automated and human evaluation. Our results show that MAGID is comparable to or better than baselines, with significant improvements in human evaluation, especially against retrieval baselines where the image database is small. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.301 | null | 0 | jailbreaking attacks | null | XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models | Without proper safeguards, large language models will readily follow malicious instructions and generate toxic content. This risk motivates safety efforts such as red-teaming and large-scale feedback learning, which aim to make models both helpful and harmless. However, there is a tension between these two objectives, since harmlessness requires models to refuse to comply with unsafe prompts, and thus not be helpful. Recent anecdotal evidence suggests that some models may have struck a poor balance, so that even clearly safe prompts are refused if they use similar language to unsafe prompts or mention sensitive topics. In this paper, we introduce a new test suite called XSTest to identify such eXaggerated Safety behaviours in a systematic way. XSTest comprises 250 safe prompts across ten prompt types that well-calibrated models should not refuse to comply with, and 200 unsafe prompts as contrasts that models, for most applications, should refuse. We describe XSTest’s creation and composition, and then use the test suite to highlight systematic failure modes in state-of-the-art language models as well as more general challenges in building safer language models. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.375 | null | null | jailbreaking attacks | English, Chinese | Attacks, Defenses and Evaluations for LLM Conversation Safety: A Survey | Large Language Models (LLMs) are now commonplace in conversation applications. However, their risks of misuse for generating harmful responses have raised serious societal concerns and spurred recent research on LLM conversation safety. Therefore, in this survey, we provide a comprehensive overview of recent studies, covering three critical aspects of LLM conversation safety: attacks, defenses, and evaluations. Our goal is to provide a structured summary that enhances understanding of LLM conversation safety and encourages further investigation into this important subject. For easy reference, we have categorized all the studies mentioned in this survey according to our taxonomy, available at: https://github.com/niconi19/LLM-conversation-safety. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-long.422 | x | 0 | general safety, LLM alignment | null | Safer-Instruct: Aligning Language Models with Automated Preference Data | Reinforcement learning from human feedback (RLHF) is a vital strategy for enhancing model capability in language models. However, annotating preference data for RLHF is a resource-intensive and creativity-demanding process, while existing automatic generation methods face limitations in data diversity and quality. In response, we present Safer-Instruct, a novel pipeline for automatically constructing large-scale preference data. Our approach leverages reversed instruction tuning, instruction induction, and expert model evaluation to efficiently generate high-quality preference data without human annotators. To verify the effectiveness of Safer-Instruct, we apply the pipeline to construct a safety preference dataset as a case study. Finetuning an Alpaca model on this synthetic dataset not only demonstrates improved harmlessness but also outperforms models fine-tuned on human-annotated safety preference data, all the while maintaining a competitive edge in downstream tasks. Importantly, our Safer-Instruct framework is versatile and can be applied to generate preference data across various domains, extending its utility beyond safety preferences. It addresses the challenges in preference data acquisition and advances the development of more capable and responsible AI systems. For dataset and code implementation, see https://github.com/uscnlp-lime/safer-instruct/. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-short.60 | x | 0 | toxicity, bias | null | LifeTox: Unveiling Implicit Toxicity in Life Advice | As large language models become increasingly integrated into daily life, detecting implicit toxicity across diverse contexts is crucial. To this end, we introduce \texttt{LifeTox}, a dataset designed for identifying implicit toxicity within a broad range of advice-seeking scenarios. Unlike existing safety datasets, \texttt{LifeTox} comprises diverse contexts derived from personal experiences through open-ended questions. Our experiments demonstrate that RoBERTa fine-tuned on \texttt{LifeTox} matches or surpasses the zero-shot performance of large language models in toxicity classification tasks. These results underscore the efficacy of \texttt{LifeTox} in addressing the complex challenges inherent in implicit toxicity. We open-sourced the dataset and the \texttt{LifeTox} moderator family; 350M, 7B, and 13B. |
2024.naacl.xml | Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.naacl-industry.8 | null | null | others | null | An NLP-Focused Pilot Training Agent for Safe and Efficient Aviation Communication | Aviation communication significantly influences the success of flight operations, ensuring safety of lives and efficient air transportation. In day-to-day flight operations, air traffic controllers (ATCos) would timely communicate instructions to pilots using specific phraseology for aircraft manipulation . However, pilots, originating from diverse backgrounds and understanding of English language, have struggled with conforming to strict phraseology for readback and communication in the live operation, this problem had not been effectively addressed over the past decades. Traditionally, aviation communication training involved expensive setups and resources, often relying on human-in-the-loop (HIL) air traffic simulations that demand allocating a specific environment, domain experts for participation, and substantial amount of annotated data for simulation. Therefore, we would like to propose an NLP-oriented training agent and address these challenges. Our approach involves leveraging only natural language capabilities and fine-tuning on communication data to generate instructions based on input scenarios (keywords). Given the absence of prior references for this business problem, we investigated the feasibility of our proposed solution by 1) generating all instructions at once and 2) generating one instruction while incorporating conversational history in each input. Our findings affirm the feasibility of this approach, highlighting the effectiveness of fine-tuning pre-trained models and large language models in advancing aviation communication training. |
2024.splurobonlp.xml | Proceedings of the 4th Workshop on Spatial Language Understanding and Grounded Communication for Robotics (SpLU-RoboNLP 2024) | 2024 | https://aclanthology.org/2024.splurobonlp-1.1 | x | 0 | general safety, LLM alignment | null | Language-guided World Models: A Model-based Approach to AI Control | Developing internal world models for artificial agents opens an efficient channel for humans to communicate with and control them. In addition to updating policies, humans can modify the world models of these agents in order to influence their decisions.The challenge, however, is that currently existing world models are difficult for humans to adapt because they lack a natural communication interface. Aimed at addressing this shortcoming, we develop *Language-Guided World Models* (LWMs), which can capture environment dynamics by reading language descriptions. These models enhance agent communication efficiency, allowing humans to simultaneously alter their behavior on multiple tasks with concise language feedback. They also enable agents to self-learn from texts originally written to instruct humans. To facilitate the development of LWMs, we design a challenging benchmark based on the game of MESSENGER (Hanjie et al., 2021), requiring compositional generalization to new language descriptions and environment dynamics. Our experiments reveal that the current state-of-the-art Transformer architecture performs poorly on this benchmark, motivating us to design a more robust architecture. To showcase the practicality of our proposed LWMs, we simulate a scenario where these models augment the interpretability and safety of an agent by enabling it to generate and discuss plans with a human before execution. By effectively incorporating language feedback on the plan, the models boost the agent performance in the real environment by up to three times without collecting any interactive experiences in this environment. |
2024.safety4convai.xml | Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024 | 2024 | https://aclanthology.org/2024.safety4convai-1.1 | x | 0 | hallucination, factuality | null | Grounding LLMs to In-prompt Instructions: Reducing Hallucinations Caused by Static Pre-training Knowledge | When deploying LLMs in certain commercial or research settings, domain specific knowledge must be explicitly provided within the prompt. This in-prompt knowledge can conflict with an LLM’s static world knowledge learned at pre-training, causing model hallucination (see examples in Table 1). In safety-critical settings, like healthcare and finance, these hallucinations can harm vulnerable users. We have curated a QA corpus containing information that LLMs could not have seen at pre-training. Using our corpus, we have probed various LLMs, manipulating both the prompt and the knowledge representation. We have found that our ‘Jodie’ prompt consistently improves the model’s textual grounding to the given knowledge, and in-turn the overall answer accuracy. This is true in both the healthcare and finance domains - improving accuracy by up to 28% (mean: 12%). We have also identified that hierarchical and direct node-property graph structures could lead to more interpretable and controllable systems that provide a natural language interface with real-time in-domain knowledge. Our corpus will enable further work on this critical challenge. |
2024.safety4convai.xml | Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024 | 2024 | https://aclanthology.org/2024.safety4convai-1.2 | x | 0 | general safety, LLM alignment | null | Diversity-Aware Annotation for Conversational AI Safety | How people interpret content is deeply influenced by their socio-cultural backgrounds and lived experiences. This is especially crucial when evaluating AI systems for safety, where accounting for such diversity in interpretations and potential impacts on human users will make them both more successful and inclusive. While recent work has demonstrated the importance of diversity in human ratings that underlie AI pipelines, effective and efficient ways to incorporate diverse perspectives in human data annotation pipelines is still largely elusive. In this paper, we discuss the primary challenges faced in incorporating diversity into model evaluations, and propose a practical diversity-aware annotation approach. Using an existing dataset with highly parallel safety annotations, we take as a test case a policy that prioritizes recall of safety issues, and demonstrate that our diversity-aware approach can efficiently obtain a higher recall of safety issues flagged by minoritized rater groups without hurting overall precision. |
2024.safety4convai.xml | Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024 | 2024 | https://aclanthology.org/2024.safety4convai-1.3 | null | 0 | toxicity, bias | null | Using Information Retrieval Techniques to Automatically Repurpose Existing Dialogue Datasets for Safe Chatbot Development | There has been notable progress in the development of open-domain dialogue systems (chatbots) especially with the rapid advancement of the capabilities of Large Language Models. Chatbots excel at holding conversations in a manner that keeps a user interested and engaged. However, their responses can be unsafe, as they can respond in an offensive manner or offer harmful professional advice. As a way to mitigate this issue, recent work crowdsource datasets with exemplary responses or annotate dialogue safety datasets, which are relatively scarce compared to casual dialogues. Despite the quality of data obtained from crowdsourcing, it can be expensive and time consuming. This work proposes an effective pipeline, using information retrieval, to automatically repurpose existing dialogue datasets for safe chatbot development, as a way to address the aforementioned challenges. We select an existing dialogue dataset, revise its unsafe responses, as a way to obtain a dataset with safer responses to unsafe user inputs. We then fine-tune dialogue models on the original and revised datasets and generate responses to evaluate the safeness of the models. |
2024.safety4convai.xml | Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024 | 2024 | https://aclanthology.org/2024.safety4convai-1.5 | null | 0 | jailbreaking attacks | null | Learning To See But Forgetting To Follow: Visual Instruction Tuning Makes LLMs More Prone To Jailbreak Attacks | Augmenting Large Language Models (LLMs) with image-understanding capabilities has resulted in a boom of high-performing Vision-Language models (VLMs). While studying the alignment of LLMs to human values has received widespread attention, the safety of VLMs has not received the same attention. In this paper, we explore the impact of jailbreaking on three state-of-the-art VLMs, each using a distinct modeling approach. By comparing each VLM to their respective LLM backbone, we find that each VLM is more susceptible to jailbreaking. We consider this as an undesirable outcome from visual instruction-tuning, which imposes a forgetting effect on an LLM’s safety guardrails. Therefore, we provide recommendations for future work based on evaluation strategies that aim to highlight the weaknesses of a VLM, as well as take safety measures into account during visual instruction tuning. |
2024.bionlp.xml | Proceedings of the 23rd Workshop on Biomedical Natural Language Processing | 2024 | https://aclanthology.org/2024.bionlp-1.23 | null | null | others | null | Large Language Models for Biomedical Knowledge Graph Construction: Information extraction from EMR notes | The automatic construction of knowledge graphs (KGs) is an important research area in medicine, with far-reaching applications spanning drug discovery and clinical trial design. These applications hinge on the accurate identification of interactions among medical and biological entities. In this study, we propose an end-to-end machine learning solution based on large language models (LLMs) that utilize electronic medical record notes to construct KGs. The entities used in the KG construction process are diseases, factors, treatments, as well as manifestations that coexist with the patient while experiencing the disease. Given the critical need for high-quality performance in medical applications, we embark on a comprehensive assessment of 12 LLMs of various architectures, evaluating their performance and safety attributes. To gauge the quantitative efficacy of our approach by assessing both precision and recall, we manually annotate a dataset provided by the Macula and Retina Institute. We also assess the qualitative performance of LLMs, such as the ability to generate structured outputs or the tendency to hallucinate. The results illustrate that in contrast to encoder-only and encoder-decoder, decoder-only LLMs require further investigation. Additionally, we provide guided prompt design to utilize such LLMs. The application of the proposed methodology is demonstrated on age-related macular degeneration. |
2024.bionlp.xml | Proceedings of the 23rd Workshop on Biomedical Natural Language Processing | 2024 | https://aclanthology.org/2024.bionlp-1.55 | null | null | others | null | SICAR at RRG2024: GPU Poor’s Guide to Radiology Report Generation | Radiology report generation (RRG) aims to create free-text radiology reports from clinical imaging. Our solution employs a lightweight multimodal language model (MLLM) enhanced with a two-stage post-processing strategy, utilizing a Large Language Model (LLM) to boost diagnostic accuracy and ensure patient safety. We introduce the “First, Do No Harm” SafetyNet, which incorporates Xraydar, an advanced X-ray classification model, to cross-verify the model outputs and specifically address false negatives from the MLLM. This comprehensive approach combines the efficiency of lightweight models with the robustness of thorough post-processing techniques, offering a reliable solution for radiology report generation. Our system achieved fourth place on the F1-Radgraph metric for findings generation in the Radiology Report Generation Shared Task (RRG24). |
2024.smm4h.xml | Proceedings of The 9th Social Media Mining for Health Research and Applications (SMM4H 2024) Workshop and Shared Tasks | 2024 | https://aclanthology.org/2024.smm4h-1.38 | null | null | others | null | FORCE: A Benchmark Dataset for Foodborne Disease Outbreak and Recall Event Extraction from News | The escalating prevalence of food safety incidents within the food supply chain necessitates immediate action to protect consumers. These incidents encompass a spectrum of issues, including food product contamination and deliberate food and feed adulteration for economic gain leading to outbreaks and recalls. Understanding the origins and pathways of contamination is imperative for prevention and mitigation. In this paper, we introduce FORCE Foodborne disease Outbreak and ReCall Event extraction from openweb). Our proposed model leverages a multi-tasking sequence labeling architecture in conjunction with transformer-based document embeddings. We have compiled a substantial annotated corpus comprising relevant articles published between 2011 and 2023 to train and evaluate the model. The dataset will be publicly released with the paper. The event detection model demonstrates fair accuracy in identifying food-related incidents and outbreaks associated with organizations, as assessed through cross-validation techniques. |
2024.privatenlp.xml | Proceedings of the Fifth Workshop on Privacy in Natural Language Processing | 2024 | https://aclanthology.org/2024.privatenlp-1.17 | x | 0 | jailbreaking attacks | null | Reinforcement Learning-Driven LLM Agent for Automated Attacks on LLMs | Recently, there has been a growing focus on conducting attacks on large language models (LLMs) to assess LLMs’ safety. Yet, existing attack methods face challenges, including the need to access model weights or merely ensuring LLMs output harmful information without controlling the specific content of their output. Exactly control of the LLM output can produce more inconspicuous attacks which could reveal a new page for LLM security. To achieve this, we propose RLTA: the Reinforcement Learning Targeted Attack, a framework that is designed for attacking language models (LLMs) and is adaptable to both white box (weight accessible) and black box (weight inaccessible) scenarios. It is capable of automatically generating malicious prompts that trigger target LLMs to produce specific outputs. We demonstrate RLTA in two different scenarios: LLM trojan detection and jailbreaking. The comprehensive experimental results show the potential of RLTA in enhancing the security measures surrounding contemporary LLMs. |
2024.woah.xml | Proceedings of the 8th Workshop on Online Abuse and Harms (WOAH 2024) | 2024 | https://aclanthology.org/2024.woah-1.12 | x | 0 | jailbreaking attacks | null | Robust Safety Classifier Against Jailbreaking Attacks: Adversarial Prompt Shield | Large Language Models’ safety remains a critical concern due to their vulnerability to jailbreaking attacks, which can prompt these systems to produce harmful and malicious responses. Safety classifiers, computational models trained to discern and mitigate potentially harmful, offensive, or unethical outputs, offer a practical solution to address this issue. However, despite their potential, existing safety classifiers often fail when exposed to adversarial attacks such as gradient-optimized suffix attacks. In response, our study introduces Adversarial Prompt Shield (APS), a lightweight safety classifier model that excels in detection accuracy and demonstrates resilience against unseen jailbreaking prompts. We also introduce efficiently generated adversarial training datasets, named Bot Adversarial Noisy Dialogue (BAND), which are designed to fortify the classifier’s robustness. Through extensive testing on various safety tasks and unseen jailbreaking attacks, we demonstrate the effectiveness and resilience of our models. Evaluations show that our classifier has the potential to significantly reduce the Attack Success Rate by up to 44.9%. This advance paves the way for the next generation of more reliable and resilient Large Language Models. |
2024.knowllm.xml | Proceedings of the 1st Workshop on Towards Knowledgeable Language Models (KnowLLM 2024) | 2024 | https://aclanthology.org/2024.knowllm-1.13 | null | 0 | general safety, LLM alignment | null | Safe-Embed: Unveiling the Safety-Critical Knowledge of Sentence Encoders | Despite the impressive capabilities of Large Language Models (LLMs) in various tasks, their vulnerability to unsafe prompts remains a critical issue. These prompts can lead LLMs to generate responses on illegal or sensitive topics, posing a significant threat to their safe and ethical use. Existing approaches address this issue using classification models, divided into LLM-based and API-based methods. LLM based models demand substantial resources and large datasets, whereas API-based models are cost-effective but might overlook linguistic nuances. With the increasing complexity of unsafe prompts, similarity search-based techniques that identify specific features of unsafe content provide a more robust and effective solution to this evolving problem. This paper investigates the potential of sentence encoders to distinguish safe from unsafe content. We introduce new pairwise datasets and the Cate021 gorical Purity (CP) metric to measure this capability. Our findings reveal both the effectiveness and limitations of existing sentence encoders, proposing directions to improve sentence encoders to operate as robust safety detectors. |
2024.arabicnlp.xml | Proceedings of The Second Arabic Natural Language Processing Conference | 2024 | https://aclanthology.org/2024.arabicnlp-1.18 | null | 1 | toxicity, bias | Arabic, English, and Russian | John vs. Ahmed: Debate-Induced Bias in Multilingual LLMs | Large language models (LLMs) play a crucial role in a wide range of real world applications. However, concerns about their safety and ethical implications are growing. While research on LLM safety is expanding, there is a noticeable gap in evaluating safety across multiple languages, especially in Arabic and Russian. We address this gap by exploring biases in LLMs across different languages and contexts, focusing on GPT-3.5 and Gemini. Through carefully designed argument-based prompts and scenarios in Arabic, English, and Russian, we examine biases in cultural, political, racial, religious, and gender domains. Our findings reveal biases in these domains. In particular, our investigation uncovers subtle biases where each model tends to present winners as those speaking the primary language the model is prompted with. Our study contributes to ongoing efforts to ensure justice and equality in LLM development and emphasizes the importance of further research towards responsible progress in this field. |
2024.case.xml | Proceedings of the 7th Workshop on Challenges and Applications of Automated Extraction of Socio-political Events from Text (CASE 2024) | 2024 | https://aclanthology.org/2024.case-1.30 | null | 1 | toxicity, bias | Turkish, Arabic | Team Curie at HSD-2Lang 2024: Hate Speech Detection in Turkish and Arabic Tweets using BERT-based models | Team Curie at HSD-2Lang 2024: Team Curie at HSD-2Lang 2024: Hate Speech Detection in Turkish and Arabic Tweets using BERT-based models This paper has presented our methodologies and findings in tackling hate speech detection in Turkish and Arabic tweets as part of the HSD-2Lang 2024 contest. Through innovative approaches and the fine-tuning of BERT-based models, we have achieved notable F1 scores, demonstrating the potential of our models in addressing the linguistic challenges inherent in Turkish and Arabic languages. The ablation study for Subtask A provided valuable insights into the impact of preprocessing and data balancing on model performance, guiding future enhancements. Our work contributes to the broader goal of improving online content moderation and safety, with future research directions including the expansion to more languages and the integration of multi-modal data and explainable AI techniques. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.30 | x | 0 | jailbreaking attacks | null | GradSafe: Detecting Jailbreak Prompts for LLMs via Safety-Critical Gradient Analysis | Large Language Models (LLMs) face threats from jailbreak prompts. Existing methods for detecting jailbreak prompts are primarily online moderation APIs or finetuned LLMs. These strategies, however, often require extensive and resource-intensive data collection and training processes. In this study, we propose GradSafe, which effectively detects jailbreak prompts by scrutinizing the gradients of safety-critical parameters in LLMs. Our method is grounded in a pivotal observation: the gradients of an LLM’s loss for jailbreak prompts paired with compliance response exhibit similar patterns on certain safety-critical parameters. In contrast, safe prompts lead to different gradient patterns. Building on this observation, GradSafe analyzes the gradients from prompts (paired with compliance responses) to accurately detect jailbreak prompts. We show that GradSafe, applied to Llama-2 without further training, outperforms Llama Guard—despite its extensive finetuning with a large dataset—in detecting jailbreak prompts. This superior performance is consistent across both zero-shot and adaptation scenarios, as evidenced by our evaluations on ToxicChat and XSTest. The source code is available at https://github.com/xyq7/GradSafe. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.58 | null | 0 | general safety, LLM alignment | null | Self-Distillation Bridges Distribution Gap in Language Model Fine-Tuning | The surge in Large Language Models (LLMs) has revolutionized natural language processing, but fine-tuning them for specific tasks often encounters challenges in balancing performance and preserving general instruction-following abilities. In this paper, we posit that the distribution gap between task datasets and the LLMs serves as the primary underlying cause. To address the problem, we introduce Self-Distillation Fine-Tuning (SDFT), a novel approach that bridges the distribution gap by guiding fine-tuning with a distilled dataset generated by the model itself to match its original distribution. Experimental results on the Llama-2-chat model across various benchmarks demonstrate that SDFT effectively mitigates catastrophic forgetting while achieving comparable or superior performance on downstream tasks compared to the vanilla fine-tuning. Moreover, SDFT demonstrates the potential to maintain the helpfulness and safety alignment of LLMs. Our code is available at https://github.com/sail-sg/sdft. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.99 | x | 0 | general safety, LLM alignment | null | Dissecting Human and LLM Preferences | As a relative quality comparison of model responses, human and Large Language Model (LLM) preferences serve as common alignment goals in model fine-tuning and criteria in evaluation. Yet, these preferences merely reflect broad tendencies, resulting in less explainable and controllable models with potential safety risks. In this work, we dissect the preferences of human and 32 different LLMs to understand their quantitative composition, using annotations from real-world user-model conversations for a fine-grained, scenario-wise analysis. We find that humans are less sensitive to errors, favor responses that support their stances, and show clear dislike when models admit their limits. On the contrary, advanced LLMs like GPT-4-Turbo emphasize correctness, clarity, and harmlessness more. Additionally, LLMs of similar sizes tend to exhibit similar preferences, regardless of their training methods, and fine-tuning for alignment does not significantly alter the preferences of pretrained-only LLMs. Finally, we show that preference-based evaluation can be intentionally manipulated. In both training-free and training-based settings, aligning a model with the preferences of judges boosts scores, while injecting the least preferred properties lowers them. This results in notable score shifts: up to 0.59 on MT-Bench (1-10 scale) and 31.94 on AlpacaEval 2.0 (0-100 scale), highlighting the significant impact of this strategic adaptation. We have made all resources of this project publicly available. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.119 | x | 0 | jailbreaking attacks | null | ToolSword: Unveiling Safety Issues of Large Language Models in Tool Learning Across Three Stages | Tool learning is widely acknowledged as a foundational approach or deploying large language models (LLMs) in real-world scenarios. While current research primarily emphasizes leveraging tools to augment LLMs, it frequently neglects emerging safety considerations tied to their application. To fill this gap, we present ToolSword, a comprehensive framework dedicated to meticulously investigating safety issues linked to LLMs in tool learning. Specifically, ToolSword delineates six safety scenarios for LLMs in tool learning, encompassing malicious queries and jailbreak attacks in the input stage, noisy misdirection and risky cues in the execution stage, and harmful feedback and error conflicts in the output stage. Experiments conducted on 11 open-source and closed-source LLMs reveal enduring safety challenges in tool learning, such as handling harmful queries, employing risky tools, and delivering detrimental feedback, which even GPT-4 is susceptible to. Moreover, we conduct further studies with the aim of fostering research on tool learning safety. The data will be released upon acceptance of the paper. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.140 | x | 0 | jailbreaking attacks | null | RLHFPoison: Reward Poisoning Attack for Reinforcement Learning with Human Feedback in Large Language Models | Reinforcement Learning with Human Feedback (RLHF) is a methodology designed to align Large Language Models (LLMs) with human preferences, playing an important role in LLMs alignment. Despite its advantages, RLHF relies on human annotators to rank the text, which can introduce potential security vulnerabilities if any adversarial annotator (i.e., attackers) manipulates the ranking score by up-ranking any malicious text to steer the LLM adversarially. To assess the red-teaming of RLHF against human preference data poisoning, we propose RankPoison, a poisoning attack method on candidates’ selection of preference rank flipping to reach certain malicious behaviors (e.g., generating longer sequences, which can increase the computational cost). With poisoned dataset generated by RankPoison, we can perform poisoning attacks on LLMs to generate longer tokens without hurting the original safety alignment performance. Moreover, applying RankPoison, we also successfully implement a backdoor attack where LLMs can generate longer answers under questions with the trigger word. Our findings highlight critical security challenges in RLHF, underscoring the necessity for more robust alignment methods for LLMs. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.198 | null | 0 | general safety, LLM alignment | null | Relying on the Unreliable: The Impact of Language Models’ Reluctance to Express Uncertainty | As natural language becomes the default interface for human-AI interaction, there is a need for LMs to appropriately communicate uncertainties in downstream applications. In this work, we investigate how LMs incorporate confidence in responses via natural language and how downstream users behave in response to LM-articulated uncertainties. We examine publicly deployed models and find that LMs are reluctant to express uncertainties when answering questions even when they produce incorrect responses. LMs can be explicitly prompted to express confidences, but tend to be overconfident, resulting in high error rates (an average of 47%) among confident responses. We test the risks of LM overconfidence by conducting human experiments and show that users rely heavily on LM generations, whether or not they are marked by certainty. Lastly, we investigate the preference-annotated datasets used in post training alignment and find that humans are biased against texts with uncertainty. Our work highlights new safety harms facing human-LM interactions and proposes design recommendations and mitigating strategies moving forward. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.253 | x | 0 | jailbreaking attacks | null | Navigating the OverKill in Large Language Models | Large language models are meticulously aligned to be both helpful and harmless. However, recent research points to a potential overkill which means models may refuse to answer benign queries. In this paper, we investigate the factors for overkill by exploring how models handle and determine the safety of queries. Our findings reveal the presence of shortcuts within models, leading to excessive attention to harmful words like ‘kill’ and prompts emphasizing safety will exacerbate overkill. Based on these insights, we introduce Self-Contrastive Decoding (Self-CD), a training-free and model-agnostic strategy, to alleviate this phenomenon. We first extract such excessive attention by amplifying the difference in the model’s output distributions when responding to system prompts that either include or omit an emphasis on safety. Then we determine the final next-token predictions by downplaying the excessive attention via contrastive decoding. Empirical results have indicated that our method has achieved an average reduction of the refusal rate by 20 % while having almost no impact on safety. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.303 | x | 0 | jailbreaking attacks | null | SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decoding | As large language models (LLMs) become increasingly integrated into real-world applications such as code generation and chatbot assistance, extensive efforts have been made to align LLM behavior with human values, including safety. Jailbreak attacks, which aim to provoke unintended and unsafe behaviors from LLMs, remain a significant LLM safety threat. We analyze tokens, which are the smallest unit of text that can be processed by LLMs and make the following observations: (1) probabilities of tokens representing harmful responses are higher than those of harmless responses, and (2) responses containing safety disclaimers appear among the top tokens when token probabilities are sorted in descending order. In this paper, we leverage (1) and (2) to develop SafeDecoding, a safety-aware decoding strategy for LLMs, to defend against jailbreak attacks. We perform extensive experiments to evaluate SafeDecoding against six SOTA jailbreak attacks (GCG, AutoDAN, PAIR, DeepInception, SAP30, and template based attack) on five LLMs (Vicuna, Llama2, Guanaco, falcon, and Dolphin) using four benchmark datasets (AdvBench, HEx-PHI, MT-Bench, and Just-Eval). Our results show that SafeDecoding significantly reduces attack success rate and harmfulness of jailbreak attacks without compromising the helpfulness of responses to benign user queries while outperforming six defense methods (Perpelexity, Paraphrase, Retokenization, Self-Reminder, ICD, and Self-Examination). |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.461 | null | 1 | jailbreaking attacks | English, Chinese | Safety Alignment in NLP Tasks: Weakly Aligned Summarization as an In-Context Attack | Recent developments in balancing the usefulness and safety of Large Language Models (LLMs) have raised a critical question: Are mainstream NLP tasks adequately aligned with safety consideration? Our study, focusing on safety-sensitive documents obtained through adversarial attacks, reveals significant disparities in the safety alignment of various NLP tasks. For instance, LLMs can effectively summarize malicious long documents but often refuse to translate them. This discrepancy highlights a previously unidentified vulnerability: attacks exploiting tasks with weaker safety alignment, like summarization, can potentially compromise the integrity of tasks traditionally deemed more robust, such as translation and question-answering (QA). Moreover, the concurrent use of multiple NLP tasks with lesser safety alignment increases the risk of LLMs inadvertently processing harmful content. We demonstrate these vulnerabilities in various safety-aligned LLMs, particularly Llama2 models, Gemini and GPT-4, indicating an urgent need for strengthening safety alignments across a broad spectrum of NLP tasks. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.481 | x | 0 | jailbreaking attacks | null | Defending Large Language Models Against Jailbreaking Attacks Through Goal Prioritization | While significant attention has been dedicated to exploiting weaknesses in LLMs through jailbreaking attacks, there remains a paucity of effort in defending against these attacks. We point out a pivotal factor contributing to the success of jailbreaks: the intrinsic conflict between the goals of being helpful and ensuring safety. Accordingly, we propose to integrate goal prioritization at both training and inference stages to counteract. Implementing goal prioritization during inference substantially diminishes the Attack Success Rate (ASR) of jailbreaking from 66.4% to 3.6% for ChatGPT. And integrating goal prioritization into model training reduces the ASR from 71.0% to 6.6% for Llama2-13B. Remarkably, even in scenarios where no jailbreaking samples are included during training, our approach slashes the ASR by half. Additionally, our findings reveal that while stronger LLMs face greater safety risks, they also possess a greater capacity to be steered towards defending against such attacks, both because of their stronger ability in instruction following. Our work thus contributes to the comprehension of jailbreaking attacks and defenses, and sheds light on the relationship between LLMs’ capability and safety. Our code is available at https://github.com/thu-coai/JailbreakDefense_GoalPriority. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.538 | null | 1 | general safety, LLM alignment | Chinese | NewsBench: A Systematic Evaluation Framework for Assessing Editorial Capabilities of Large Language Models in Chinese Journalism | We present NewsBench, a novel evaluation framework to systematically assess the capabilities of Large Language Models (LLMs) for editorial capabilities in Chinese journalism. Our constructed benchmark dataset is focused on four facets of writing proficiency and six facets of safety adherence, and it comprises manually and carefully designed 1,267 test samples in the types of multiple choice questions and short answer questions for five editorial tasks in 24 news domains. To measure performances, we propose different GPT-4 based automatic evaluation protocols to assess LLM generations for short answer questions in terms of writing proficiency and safety adherence, and both are validated by the high correlations with human evaluations. Based on the systematic evaluation framework, we conduct a comprehensive analysis of eleven popular LLMs which can handle Chinese. The experimental results highlight GPT-4 and ERNIE Bot as top performers, yet reveal a relative deficiency in journalistic safety adherence in creative writing tasks. Our findings also underscore the need for enhanced ethical guidance in machine-generated journalistic content, marking a step forward in aligning LLMs with journalistic standards and safety considerations. The evaluation framework and experimental results are expected to provide an in-depth understanding of the editorial capabilities of LLMs and speed up the development of LLMs in journalism. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.572 | null | null | general safety, LLM alignment | null | Aligning Large Language Models with Human Preferences through Representation Engineering | Aligning large language models (LLMs) with human preferences is crucial for enhancing their utility in terms of helpfulness, truthfulness, safety, harmlessness, and interestingness. Existing methods for achieving this alignment often involve employing reinforcement learning from human feedback (RLHF) to fine-tune LLMs based on human labels assessing the relative quality of model responses. Nevertheless, RLHF is susceptible to instability during fine-tuning and presents challenges in implementation. Drawing inspiration from the emerging field of representation engineering (RepE), this study aims to identify relevant representations for high-level human preferences embedded in patterns of activity within an LLM and achieve precise control of model behavior by transforming its representations. This novel approach, denoted as Representation Alignment from Human Feedback (RAHF), proves to be effective, computationally efficient, and easy to implement. Extensive experiments demonstrate the efficacy of RAHF in not only capturing but also manipulating representations to align with a broad spectrum of human preferences or values, rather than being confined to a singular concept or function (e.g. honesty or bias). RAHF’s versatility in accommodating diverse human preferences shows its potential for advancing LLM performance. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.698 | null | null | general safety, LLM alignment | null | SOTOPIA-π: Interactive Learning of Socially Intelligent Language Agents | Humans learn social skills through both imitation and social interaction. This social learning process is largely understudied by existing research on building language agents. Motivated by this gap, we propose an interactive learning method, SOTOPIA-π, that improves the social intelligence of language agents. This method leverages behavior cloning and self-reinforcement based training on filtered social interaction data according to large language model (LLM) rating. We show that our training method allows a 7B LLM to reach the social goal completion ability of an expert model (GPT-4-based agent) without the loss of more generic abilities, such as the ability to answer knowledge-based questions. We also demonstrate that this training paradigm uncovers some weaknesses in standard evaluation and safety training paradigms that (1) LLM-based evaluation of social intelligence overestimates the abilities of the language agents trained specifically for social interaction, and that (2) despite not training for better safety or question answering (QA) ability, our methods improve the safety of language agents and maintain general QA ability on the MMLU benchmark. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.741 | null | 0 | privacy | null | Reducing Privacy Risks in Online Self-Disclosures with Language Models | Self-disclosure, while being common and rewarding in social media interaction, also poses privacy risks. In this paper, we take the initiative to protect the user-side privacy associated with online self-disclosure through detection and abstraction. We develop a taxonomy of 19 self-disclosure categories and curate a large corpus consisting of 4.8K annotated disclosure spans. We then fine-tune a language model for detection, achieving over 65% partial span F_1. We further conduct an HCI user study, with 82% of participants viewing the model positively, highlighting its real-world applicability. Motivated by the user feedback, we introduce the task of self-disclosure abstraction, which is rephrasing disclosures into less specific terms while preserving their utility, e.g., “Im 16F” to “I’m a teenage girl”. We explore various fine-tuning strategies, and our best model can generate diverse abstractions that moderately reduce privacy risks while maintaining high utility according to human evaluation. To help users in deciding which disclosures to abstract, we present a task of rating their importance for context understanding. Our fine-tuned model achieves 80% accuracy, on-par with GPT-3.5. Given safety and privacy considerations, we will only release our corpus and models to researcher who agree to the ethical guidelines outlined in Ethics Statement. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.762 | null | 1 | jailbreaking attacks | English, Chinese, Hindi | Language Models are Homer Simpson! Safety Re-Alignment of Fine-tuned Language Models through Task Arithmetic | We propose RESTA to perform LLM realignment towards safety, which gets compromised due to downstream task fine-tuning. RESTA stands for REstoring Safety through Task Arithmetic. At its core, it involves a simple arithmetic addition of a safety vector to the weights of the compromised model. We demonstrate the effectiveness of RESTA in both parameter-efficient and full fine-tuning, covering a wide range of downstream tasks, including instruction following in Chinese, English, and Hindi, as well as problem-solving capabilities in Code and Math. We also showcase the generalizability of RESTA on three existing safety evaluation benchmarks and a multilingual benchmark dataset proposed as a part of this work, consisting of 550 harmful questions covering 11 categories, each with 5 sub-categories of harm. Overall, RESTA decreases the harmfulness of the compromised model from 18.6% to 5.1% and from 9.2% to 1.5% in parameter-efficient and full fine-tuning, respectively, while maintaining most of the model’s performance on the task. We release the source codes at: https://github.com/declare-lab/resta. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.773 | x | 0 | jailbreaking attacks | null | How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs | Most traditional AI safety research views models as machines and centers on algorithm-focused attacks developed by security experts. As large language models (LLMs) become increasingly common and competent, non-expert users can also impose risks during daily interactions. Observing this, we shift the perspective, by treating LLMs as human-like communicators to examine the interplay between everyday language interaction and AI safety. Specifically, we study how to persuade LLMs to jailbreak them. First, we propose a persuasion taxonomy derived from decades of social science research. Then, we apply the taxonomy to automatically generate persuasive adversarial prompts (PAP) to jailbreak LLMs. Results show that persuasion significantly increases the jailbreak risk across all risk categories: PAP consistently achieves an attack success rate of over 92% on Llama-2-7b-Chat, GPT-3.5, and GPT-4 in 10 trials, surpassing recent algorithm-focused attacks. On the defense side, we explore various mechanisms against PAP, find a significant gap in existing defenses, and advocate for more fundamental solutions for AI safety. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.793 | null | 0 | hallucination, factuality | null | Disinformation Capabilities of Large Language Models | Automated disinformation generation is often listed as one of the risks of large language models (LLMs). The theoretical ability to flood the information space with disinformation content might have dramatic consequences for democratic societies around the world. This paper presents a comprehensive study of the disinformation capabilities of the current generation of LLMs to generate false news articles in English language. In our study, we evaluated the capabilities of 10 LLMs using 20 disinformation narratives. We evaluated several aspects of the LLMs: how well they are at generating news articles, how strongly they tend to agree or disagree with the disinformation narratives, how often they generate safety warnings, etc. We also evaluated the abilities of detection models to detect these articles as LLM-generated. We conclude that LLMs are able to generate convincing news articles that agree with dangerous disinformation narratives. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.796 | null | x | general safety, LLM alignment | Indonesian | Cendol: Open Instruction-tuned Generative Large Language Models for Indonesian Languages | Large language models (LLMs) show remarkable human-like capability in various domains and languages. To bridge this quality gap, we introduce Cendol, a collection of Indonesian LLMs encompassing both decoder-only and encoder-decoder architectures across a range of model sizes. We highlight Cendol’s effectiveness across a diverse array of tasks, attaining ~20% improvement, and demonstrate its capability to generalize to unseen tasks and indigenous languages of Indonesia. Furthermore, Cendol models showcase improved human favorability despite their limitations in capturing indigenous knowledge and cultural values in Indonesia. In addition, we discuss the shortcomings of parameter-efficient tunings, such as LoRA, for language adaptation. Alternatively, we propose the usage of vocabulary adaptation to enhance efficiency. Lastly, we evaluate the safety of Cendol and showcase that safety in pre-training in one language such as English is transferable to low-resource languages, such as Indonesian, even without RLHF and safety fine-tuning. |
2024.acl.xml | Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) | 2024 | https://aclanthology.org/2024.acl-long.809 | x | 0 | jailbreaking attacks | null | ArtPrompt: ASCII Art-based Jailbreak Attacks against Aligned LLMs | Safety is critical to the usage of large language models (LLMs). Multiple techniques such as data filtering and supervised fine-tuning have been developed to strengthen LLM safety. However, currently known techniques presume that corpora used for safety alignment of LLMs are solely interpreted by semantics. This assumption, however, does not hold in real-world applications, which leads to severe vulnerabilities in LLMs. For example, users of forums often use ASCII art, a form of text-based art, to convey image information. In this paper, we propose a novel ASCII art-based jailbreak attack and introduce a comprehensive benchmark Vision-in-Text Challenge (ViTC) to evaluate the capabilities of LLMs in recognizing prompts that cannot be solely interpreted by semantics. We show that five SOTA LLMs (GPT-3.5, GPT-4, Gemini, Claude, and Llama2) struggle to recognize prompts provided in the form of ASCII art. Based on this observation, we develop the jailbreak attack ArtPrompt, which leverages the poor performance of LLMs in recognizing ASCII art to bypass safety measures and elicit undesired behaviors from LLMs. ArtPrompt only requires black-box access to the victim LLMs, making it a practical attack. We evaluate ArtPrompt on five SOTA LLMs, and show that ArtPrompt can effectively and efficiently induce undesired behaviors from all five LLMs. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.