The impressive capabilities of recent language models can be largely attributed to the multi-trillion token pretraining datasets that they are trained on. However, model developers fail to disclose their construction methodology which has lead to a lack of open information on how to develop effective pretraining sets. To address this issue, we perform the first systematic study across the entire pipeline of pretraining set construction. First, we run ablations on existing techniques for pretraining set development to identify which methods translate to the largest gains in model accuracy on downstream evaluations. Then, we categorize the most widely used data source, web crawl snapshots, across the attributes of toxicity, quality, type of speech, and domain. Finally, we show how such attribute information can be used to further refine and improve the quality of a pretraining set. These findings constitute an actionable set of steps that practitioners can use to develop high quality pretraining sets.

This study explores real-world human interactions with large language models (LLMs) in diverse, unconstrained settings in contrast to most prior research focusing on ethically trimmed models like ChatGPT for specific tasks. We aim to understand the originator of toxicity. Our findings show that although LLMs are rightfully accused of providing toxic content, it is mostly demanded or at least provoked by humans who actively seek such content. Our manual analysis of hundreds of conversations judged as toxic by APIs commercial vendors, also raises questions with respect to current practices of what user requests are refused to answer. Furthermore, we conjecture based on multiple empirical indicators that humans exhibit a change of their mental model, switching from the mindset of interacting with a machine more towards interacting with a human.

Machine Learning systems are increasingly prevalent across healthcare, law enforcement, and finance but often operate on historical data, which may carry biases against certain demographic groups. Causal and counterfactual fairness provides an intuitive way to define fairness that closely aligns with legal standards. Despite its theoretical benefits, counterfactual fairness comes with several practical limitations, largely related to the reliance on domain knowledge and approximate causal discovery techniques in constructing a causal model. In this study, we take a fresh perspective on counterfactually fair prediction, building upon recent work in in context learning (ICL) and prior fitted networks (PFNs) to learn a transformer called FairPFN. This model is pretrained using synthetic fairness data to eliminate the causal effects of protected attributes directly from observational data, removing the requirement of access to the correct causal model in practice. In our experiments, we thoroughly assess the effectiveness of FairPFN in eliminating the causal impact of protected attributes on a series of synthetic case studies and real world datasets. Our findings pave the way for a new and promising research area: transformers for causal and counterfactual fairness.

We present seven experiments exploring gender biases in GPT. Initially, GPT was asked to generate demographics of a potential writer of twenty phrases containing feminine stereotypes and twenty with masculine stereotypes. Results show a strong asymmetry, with stereotypically masculine sentences attributed to a female more often than vice versa. For example, the sentence "I love playing fotbal! Im practicing with my cosin Michael" was constantly assigned by ChatGPT to a female writer. This phenomenon likely reflects that while initiatives to integrate women in traditionally masculine roles have gained momentum, the reverse movement remains relatively underdeveloped. Subsequent experiments investigate the same issue in high-stakes moral dilemmas. GPT-4 finds it more appropriate to abuse a man to prevent a nuclear apocalypse than to abuse a woman. This bias extends to other forms of violence central to the gender parity debate (abuse), but not to those less central (torture). Moreover, this bias increases in cases of mixed-sex violence for the greater good: GPT-4 agrees with a woman using violence against a man to prevent a nuclear apocalypse but disagrees with a man using violence against a woman for the same purpose. Finally, these biases are implicit, as they do not emerge when GPT-4 is directly asked to rank moral violations. These results highlight the necessity of carefully managing inclusivity efforts to prevent unintended discrimination.

Previous works on Large Language Models (LLMs) have mainly focused on evaluating their helpfulness or harmlessness. However, honesty, another crucial alignment criterion, has received relatively less attention. Dishonest behaviors in LLMs, such as spreading misinformation and defrauding users, present severe risks that intensify as these models approach superintelligent levels. Enhancing honesty in LLMs addresses critical limitations and helps uncover latent capabilities that are not readily expressed. This underscores the urgent need for reliable methods and benchmarks to effectively ensure and evaluate the honesty of LLMs. In this paper, we introduce BeHonest, a pioneering benchmark specifically designed to assess honesty in LLMs comprehensively. BeHonest evaluates three essential aspects of honesty: awareness of knowledge boundaries, avoidance of deceit, and consistency in responses. Building on this foundation, we designed 10 scenarios to evaluate and analyze 9 popular LLMs on the market, including both closed-source and open-source models from different model families with varied model sizes. Our findings indicate that there is still significant room for improvement in the honesty of LLMs. We encourage the AI community to prioritize honesty alignment in these models, which can harness their full potential to benefit society while preventing them from causing harm through deception or inconsistency. Our benchmark and code can be found at: \url{https://github.com/GAIR-NLP/BeHonest}.

Injustice occurs when someone experiences unfair treatment or their rights are violated and is often due to the presence of implicit biases and prejudice such as stereotypes. The automated identification of injustice in text has received little attention, due in part to the fact that underlying implicit biases or stereotypes are rarely explicitly stated and that instances often occur unconsciously due to the pervasive nature of prejudice in society. Here, we describe a novel framework that combines the use of a fine-tuned BERT-based bias detection model, two stereotype detection models, and a lexicon-based approach to show that epistemological biases (i.e., words, which presupposes, entails, asserts, hedges, or boosts text to erode or assert a person's capacity as a knower) can assist with the automatic detection of injustice in text. The news media has many instances of injustice (i.e. discriminatory narratives), thus it is our use case here. We conduct and discuss an empirical qualitative research study which shows how the framework can be applied to detect injustices, even at higher volumes of data.

Abstract--Visual diffusion models have revolutionized the field of creative AI, producing high-quality and diverse content. However, they inevitably memorize training images or videos, subsequently replicating their concepts, content, or styles during inference. In this survey, we provide the first comprehensive review of replication in visual diffusion models, marking a novel contribution to the field by systematically categorizing the existing studies into unveiling, understanding, and mitigating this phenomenon. Specifically, unveiling mainly refers to the methods used to detect replication instances. Understanding involves analyzing the underlying mechanisms and factors that contribute to this phenomenon. Mitigation focuses on developing strategies to reduce or eliminate replication. Beyond these aspects, we also review papers focusing on its real-world influence. For instance, in the context of healthcare, replication is critically worrying due to privacy concerns related to patient data. Finally, the paper concludes with a discussion of the ongoing challenges, such as the difficulty in detecting and benchmarking replication, and outlines future directions including the development of more robust mitigation techniques. By synthesizing insights from diverse studies, this paper aims to equip researchers and practitioners with a deeper understanding at the intersection between AI technology and social good. Compared to traditional Generative Adversarial Networks (GAN) [3] and Variational Autoencoders (VAE) [4], visual diffusion models excel in producing high-quality, diverse, and stable images.

The use of artificial intelligence (AI) in the public sector is best understood as a continuation and intensification of long standing rationalization and bureaucratization processes. Drawing on Weber, we take the core of these processes to be the replacement of traditions with instrumental rationality, i.e., the most calculable and efficient way of achieving any given policy objective. In this article, we demonstrate how much of the criticisms, both among the public and in scholarship, directed towards AI systems spring from well known tensions at the heart of Weberian rationalization. To illustrate this point, we introduce a thought experiment whereby AI systems are used to optimize tax policy to advance a specific normative end, reducing economic inequality. Our analysis shows that building a machine-like tax system that promotes social and economic equality is possible. However, it also highlights that AI driven policy optimization (i) comes at the exclusion of other competing political values, (ii) overrides citizens sense of their noninstrumental obligations to each other, and (iii) undermines the notion of humans as self-determining beings. Contemporary scholarship and advocacy directed towards ensuring that AI systems are legal, ethical, and safe build on and reinforce central assumptions that underpin the process of rationalization, including the modern idea that science can sweep away oppressive systems and replace them with a rule of reason that would rescue humans from moral injustices. That is overly optimistic. Science can only provide the means, they cannot dictate the ends. Nonetheless, the use of AI in the public sector can also benefit the institutions and processes of liberal democracies. Most importantly, AI driven policy optimization demands that normative ends are made explicit and formalized, thereby subjecting them to public scrutiny and debate.

Organisations that design and deploy artificial intelligence (AI) systems increasingly commit themselves to high-level, ethical principles. However, there still exists a gap between principles and practices in AI ethics. One major obstacle organisations face when attempting to operationalise AI Ethics is the lack of a well-defined material scope. Put differently, the question to which systems and processes AI ethics principles ought to apply remains unanswered. Of course, there exists no universally accepted definition of AI, and different systems pose different ethical challenges. Nevertheless, pragmatic problem-solving demands that things should be sorted so that their grouping will promote successful actions for some specific end. In this article, we review and compare previous attempts to classify AI systems for the purpose of implementing AI governance in practice. We find that attempts to classify AI systems found in previous literature use one of three mental model. The Switch, i.e., a binary approach according to which systems either are or are not considered AI systems depending on their characteristics. The Ladder, i.e., a risk-based approach that classifies systems according to the ethical risks they pose. And the Matrix, i.e., a multi-dimensional classification of systems that take various aspects into account, such as context, data input, and decision-model. Each of these models for classifying AI systems comes with its own set of strengths and weaknesses. By conceptualising different ways of classifying AI systems into simple mental models, we hope to provide organisations that design, deploy, or regulate AI systems with the conceptual tools needed to operationalise AI governance in practice.

Legal systems worldwide are inundated with exponential growth in cases and documents. There is an imminent need to develop NLP and ML techniques for automatically processing and understanding legal documents to streamline the legal system. However, evaluating and comparing various NLP models designed specifically for the legal domain is challenging. This paper addresses this challenge by proposing IL-TUR: Benchmark for Indian Legal Text Understanding and Reasoning. IL-TUR contains monolingual (English, Hindi) and multi-lingual (9 Indian languages) domain-specific tasks that address different aspects of the legal system from the point of view of understanding and reasoning over Indian legal documents. We present baseline models (including LLM-based) for each task, outlining the gap between models and the ground truth. To foster further research in the legal domain, we create a leaderboard (available at: https://exploration-lab.github.io/IL-TUR/) where the research community can upload and compare legal text understanding systems.