Evaluating Large Language Models (LLMs) in low-resource and linguistically diverse languages remains a significant challenge in NLP, particularly for languages using non-Latin scripts like those spoken in India. Existing benchmarks predominantly focus on English, leaving substantial gaps in assessing LLM capabilities in these languages. We introduce MILU, a Multi task Indic Language Understanding Benchmark, a comprehensive evaluation benchmark designed to address this gap. MILU spans 8 domains and 42 subjects across 11 Indic languages, reflecting both general and culturally specific knowledge. With an India-centric design, incorporates material from regional and state-level examinations, covering topics such as local history, arts, festivals, and laws, alongside standard subjects like science and mathematics. We evaluate over 45 LLMs, and find that current LLMs struggle with MILU, with GPT-4o achieving the highest average accuracy at 72 percent. Open multilingual models outperform language-specific fine-tuned models, which perform only slightly better than random baselines. Models also perform better in high resource languages as compared to low resource ones. Domain-wise analysis indicates that models perform poorly in culturally relevant areas like Arts and Humanities, Law and Governance compared to general fields like STEM. To the best of our knowledge, MILU is the first of its kind benchmark focused on Indic languages, serving as a crucial step towards comprehensive cultural evaluation. All code, benchmarks, and artifacts are publicly available to foster open research.

Rapid developments in AI and its adoption across various domains have necessitated a need to build robust guardrails and risk containment plans while ensuring equitable benefits for the betterment of society. The current technology-centered approach has resulted in a fragmented, reactive, and ineffective policy apparatus. This paper highlights the immediate and urgent need to pivot to a society-centered approach to develop comprehensive, coherent, forward-looking AI policy. To this end, we present a Social Outcomes and Priorities centered (SOP) framework for AI policy along with proposals on implementation of its various components. While the SOP framework is presented from a US-centric view, the takeaways are general and applicable globally.

What ethical concerns, if any, do LLM researchers have? We introduce EthiCon, a corpus of 1,580 ethical concern statements extracted from scientific papers published in the ACL Anthology. We extract ethical concern keywords from the statements and show promising results in automating the concern identification process. Through a survey, we compare the ethical concerns of the corpus to the concerns listed by the general public and professionals in the field. Finally, we compare our retrieved ethical concerns with existing taxonomies pointing to gaps and future research directions.

In an effort to mitigate the harms of large language models (LLMs), learning from human feedback (LHF) has been used to steer LLMs towards outputs that are intended to be both less harmful and more helpful. Despite the widespread adoption of LHF in practice, the quality of this feedback and its effectiveness as a safety mitigation technique remain unclear. This study addresses these issues by auditing the widely-used Helpful and Harmless (HH) dataset by Anthropic. Our work includes: (1) a thorough investigation of the dataset's content through both manual and automated evaluation; (2) experiments demonstrating the dataset's impact on models' safety; and (3) an analysis of the 100 most influential papers citing this dataset. Through our audit, we showcase how conceptualization failures and quality issues identified in the HH dataset can create additional harms by leading to disparate safety behaviors across demographic groups. Our findings highlight the need for more nuanced, context-sensitive approaches to safety mitigation in LLMs.

This work addresses the challenge of capturing the complexities of legal knowledge by proposing a multi-layered embedding-based retrieval method for legal and legislative texts. Creating embeddings not only for individual articles but also for their components (paragraphs, clauses) and structural groupings (books, titles, chapters, etc), we seek to capture the subtleties of legal information through the use of dense vectors of embeddings, representing it at varying levels of granularity. Our method meets various information needs by allowing the Retrieval Augmented Generation system to provide accurate responses, whether for specific segments or entire sections, tailored to the user's query. We explore the concepts of aboutness, semantic chunking, and inherent hierarchy within legal texts, arguing that this method enhances the legal information retrieval. Despite the focus being on Brazil's legislative methods and the Brazilian Constitution, which follow a civil law tradition, our findings should in principle be applicable across different legal systems, including those adhering to common law traditions. Furthermore, the principles of the proposed method extend beyond the legal domain, offering valuable insights for organizing and retrieving information in any field characterized by information encoded in hierarchical text.

With the rapidly increasing capabilities and adoption of code agents for AI-assisted coding, safety concerns, such as generating or executing risky code, have become significant barriers to the real-world deployment of these agents. To provide comprehensive and practical evaluations on the safety of code agents, we propose RedCode, a benchmark for risky code execution and generation: (1) RedCode-Exec provides challenging prompts that could lead to risky code execution, aiming to evaluate code agents' ability to recognize and handle unsafe code. We provide a total of 4,050 risky test cases in Python and Bash tasks with diverse input formats including code snippets and natural text. They covers 25 types of critical vulnerabilities spanning 8 domains (e.g., websites, file systems). We provide Docker environments and design corresponding evaluation metrics to assess their execution results. (2) RedCode-Gen provides 160 prompts with function signatures and docstrings as input to assess whether code agents will follow instructions to generate harmful code or software. Our empirical findings, derived from evaluating three agent frameworks based on 19 LLMs, provide insights into code agents' vulnerabilities. For instance, evaluations on RedCode-Exec show that agents are more likely to reject executing risky operations on the operating system, but are less likely to reject executing technically buggy code, indicating high risks. Risky operations described in natural text lead to a lower rejection rate than those in code format. Additionally, evaluations on RedCode-Gen show that more capable base models and agents with stronger overall coding abilities, such as GPT4, tend to produce more sophisticated and effective harmful software. Our findings highlight the need for stringent safety evaluations for diverse code agents. Our dataset and code are available at https://github.com/AI-secure/RedCode.

Jailbreaking methods, which induce Multi-modal Large Language Models (MLLMs) to output harmful responses, raise significant safety concerns. Among these methods, gradient-based approaches, which use gradients to generate malicious prompts, have been widely studied due to their high success rates in white-box settings, where full access to the model is available. However, these methods have notable limitations: they require white-box access, which is not always feasible, and involve high memory usage. To address scenarios where white-box access is unavailable, attackers often resort to transfer attacks. In transfer attacks, malicious inputs generated using white-box models are applied to black-box models, but this typically results in reduced attack performance. To overcome these challenges, we propose Zer0-Jack, a method that bypasses the need for white-box access by leveraging zeroth-order optimization. We propose patch coordinate descent to efficiently generate malicious image inputs to directly attack black-box MLLMs, which significantly reduces memory usage further. Through extensive experiments, Zer0-Jack achieves a high attack success rate across various models, surpassing previous transfer-based methods and performing comparably with existing white-box jailbreak techniques. Notably, Zer0-Jack achieves a 95\% attack success rate on MiniGPT-4 with the Harmful Behaviors Multi-modal Dataset on a black-box setting, demonstrating its effectiveness. Additionally, we show that Zer0-Jack can directly attack commercial MLLMs such as GPT-4o. Codes are provided in the supplement.

This study explores how the Large Language Models (LLMs) adjust linguistic features to create personalized persuasive outputs. While research showed that LLMs personalize outputs, a gap remains in understanding the linguistic features of their persuasive capabilities. We identified 13 linguistic features crucial for influencing personalities across different levels of the Big Five model of personality. We analyzed how prompts with personality trait information influenced the output of 19 LLMs across five model families. The findings show that models use more anxiety-related words for neuroticism, increase achievement-related words for conscientiousness, and employ fewer cognitive processes words for openness to experience. Some model families excel at adapting language for openness to experience, others for conscientiousness, while only one model adapts language for neuroticism. Our findings show how LLMs tailor responses based on personality cues in prompts, indicating their potential to create persuasive content affecting the mind and well-being of the recipients.

Decision-making processes have increasingly come to rely on sophisticated machine learning tools, raising concerns about the fairness of their predictions with respect to any sensitive groups. The widespread use of commercial black-box machine learning models necessitates careful consideration of their legal and ethical implications on consumers. In situations where users have access to these "black-box" models, a key question emerges: how can we mitigate or eliminate the influence of sensitive attributes, such as race or gender? We propose towerDebias (tDB), a novel approach designed to reduce the influence of sensitive variables in predictions made by black-box models. Using the Tower Property from probability theory, tDB aims to improve prediction fairness during the post-processing stage in a manner amenable to the Fairness-Utility Tradeoff. This method is highly flexible, requiring no prior knowledge of the original model's internal structure, and can be extended to a range of different applications. We provide a formal improvement theorem for tDB and demonstrate its effectiveness in both regression and classification tasks, underscoring its impact on the fairness-utility tradeoff.

WARNING: This paper contains unsafe content. With the development of large language models (LLMs), the sequence length of these models continues to increase, drawing significant attention to long-context language models. However, the evaluation of these models has been primarily limited to their capabilities, with a lack of research focusing on their safety. Existing work, such as ManyShotJailbreak, has to some extent demonstrated that longcontext language models can exhibit safety concerns. However, the methods used are limited and lack comprehensiveness. In response, we introduce LongSafety-Bench, the first benchmark designed to objectively and comprehensively evaluate the safety of long-context models. LongSafetyBench consists of 10 task categories, with an average length of 41,889 words. After testing eight long-context language models on LongSafetyBench, we found that existing models generally exhibit insufficient safety capabilities. The proportion of safe responses from most mainstream long-context LLMs is below 50%. Moreover, models' safety performance in long-context scenarios does not always align with that in short-context scenarios. Further investigation revealed that long-context models tend to overlook harmful content within lengthy texts. We also proposed a simple yet effective solution, allowing open-source models to achieve performance comparable to that of top-tier closed-source models. We believe that LongSafetyBench can serve as a valuable benchmark for evaluating the safety capabilities of long-context language models. We hope that our work will encourage the broader community to pay attention to the safety of long-context models and contribute to the development of solutions to improve the safety of long-context LLMs. Recently, thanks to more advanced model architectures (Xiao et al., 2024b;a; Liu et al., 2024a) and expanded position encoding techniques (Su et al., 2023; Liu et al., 2024b), the context length of language models has been extended significantly (Achiam et al., 2023; Reid et al., 2024). In the foreseeable future, as language models continue to evolve and tackle increasingly complex problems, the demand for handling longer contexts is expected to grow accordingly. We anticipate that long-context language models will become mainstream. Previous research on long-context language models, such as LongBench (Bai et al., 2024), L-Eval (An et al., 2023), and RULER (Hsieh et al., 2024), has typically focused on their capabilities, while neglecting to address their safety. In short-context scenarios, the safety issues of language models have already been extensively studied.(Zhang Illegal Activities, Misinformation Harm, Offensiveness The question is composed of a long content and Bias.