The last two years have seen a rapid growth in concerns around the safety of large language models (LLMs). Researchers and practitioners have met these concerns by introducing an abundance of new datasets for evaluating and improving LLM safety. However, much of this work has happened in parallel, and with very different goals in mind, ranging from the mitigation of near-term risks around bias and toxic content generation to the assessment of longer-term catastrophic risk potential. This makes it difficult for researchers and practitioners to find the most relevant datasets for a given use case, and to identify gaps in dataset coverage that future work may fill. To remedy these issues, we conduct a first systematic review of open datasets for evaluating and improving LLM safety. We review 102 datasets, which we identified through an iterative and community-driven process over the course of several months. We highlight patterns and trends, such as a a trend towards fully synthetic datasets, as well as gaps in dataset coverage, such as a clear lack of non-English datasets. We also examine how LLM safety datasets are used in practice -- in LLM release publications and popular LLM benchmarks -- finding that current evaluation practices are highly idiosyncratic and make use of only a small fraction of available datasets. Our contributions are based on SafetyPrompts.com, a living catalogue of open datasets for LLM safety, which we commit to updating continuously as the field of LLM safety develops.

This paper investigates the influence of anthropomorphized descriptions of so-called "AI" (artificial intelligence) systems on people's self-assessment of trust in the system. Building on prior work, we define four categories of anthropomorphization (1. Properties of a cognizer, 2. Agency, 3. Biological metaphors, and 4. Properties of a communicator). We use a survey-based approach (n=954) to investigate whether participants are likely to trust one of two (fictitious) "AI" systems by randomly assigning people to see either an anthropomorphized or a de-anthropomorphized description of the systems. We find that participants are no more likely to trust anthropomorphized over de-anthropmorphized product descriptions overall. The type of product or system in combination with different anthropomorphic categories appears to exert greater influence on trust than anthropomorphizing language alone, and age is the only demographic factor that significantly correlates with people's preference for anthropomorphized or de-anthropomorphized descriptions. When elaborating on their choices, participants highlight factors such as lesser of two evils, lower or higher stakes contexts, and human favoritism as driving motivations when choosing between product A and B, irrespective of whether they saw an anthropomorphized or a de-anthropomorphized description of the product. Our results suggest that "anthropomorphism" in "AI" descriptions is an aggregate concept that may influence different groups differently, and provide nuance to the discussion of whether anthropomorphization leads to higher trust and over-reliance by the general public in systems sold as "AI".

Background: Clinical diagnosis is typically reached by following a series of steps recommended by guidelines authored by colleges of experts. Accordingly, guidelines play a crucial role in rationalizing clinical decisions but suffer from limitations as they are built to cover the majority of the population and fail at covering patients with uncommon conditions. Moreover, their updates are long and expensive, making them unsuitable for emerging diseases and practices. Methods: Inspired by guidelines, we formulate the task of diagnosis as a sequential decision-making problem and study the use of Deep Reinforcement Learning (DRL) algorithms to learn the optimal sequence of actions to perform in order to obtain a correct diagnosis from Electronic Health Records (EHRs). We apply DRL on synthetic, but realistic EHRs and develop two clinical use cases: Anemia diagnosis, where the decision pathways follow the schema of a decision tree; and Systemic Lupus Erythematosus (SLE) diagnosis, which follows a weighted criteria score. We particularly evaluate the robustness of our approaches to noisy and missing data since these frequently occur in EHRs. Results: In both use cases, and in the presence of imperfect data, our best DRL algorithms exhibit competitive performance when compared to the traditional classifiers, with the added advantage that they enable the progressive generation of a pathway to the suggested diagnosis which can both guide and explain the decision-making process. Conclusion: DRL offers the opportunity to learn personalized decision pathways to diagnosis. We illustrate with our two use cases their advantages: they generate step-by-step pathways that are self-explanatory; and their correctness is competitive when compared to state-of-the-art approaches.

In some situations, the interpretability of the machine learning models plays a role as important as the model accuracy. Interpretability comes from the need to trust the prediction model, verify some of its properties, or even enforce them to improve fairness. Many model-agnostic explanatory methods exists to provide explanations for black-box models. In the regression task, the practitioner can use white-boxes or gray-boxes models to achieve more interpretable results, which is the case of symbolic regression. When using an explanatory method, and since interpretability lacks a rigorous definition, there is a need to evaluate and compare the quality and different explainers. This paper proposes a benchmark scheme to evaluate explanatory methods to explain regression models, mainly symbolic regression models. Experiments were performed using 100 physics equations with different interpretable and non-interpretable regression methods and popular explanation methods, evaluating the performance of the explainers performance with several explanation measures. In addition, we further analyzed four benchmarks from the GP community. The results have shown that Symbolic Regression models can be an interesting alternative to white-box and black-box models that is capable of returning accurate models with appropriate explanations. Regarding the explainers, we observed that Partial Effects and SHAP were the most robust explanation models, with Integrated Gradients being unstable only with tree-based models. This benchmark is publicly available for further experiments.

LLMs are increasingly transforming our digital ecosystem, but they often inherit societal biases learned from their training data, for instance stereotypes associating certain attributes with specific identity groups. While whether and how these biases are mitigated may depend on the specific use cases, being able to effectively detect instances of stereotype perpetuation is a crucial first step. Current methods to assess presence of stereotypes in generated language rely on simple template or co-occurrence based measures, without accounting for the variety of sentential contexts they manifest in. We argue that understanding the sentential context is crucial for detecting instances of generalization. We distinguish two types of generalizations: (1) language that merely mentions the presence of a generalization ("people think the French are very rude"), and (2) language that reinforces such a generalization ("as French they must be rude"), from non-generalizing context ("My French friends think I am rude"). For meaningful stereotype evaluations, we need to reliably distinguish such instances of generalizations. We introduce the new task of detecting generalization in language, and build GeniL, a multilingual dataset of over 50K sentences from 9 languages (English, Arabic, Bengali, Spanish, French, Hindi, Indonesian, Malay, and Portuguese) annotated for instances of generalizations. We demonstrate that the likelihood of a co-occurrence being an instance of generalization is usually low, and varies across different languages, identity groups, and attributes. We build classifiers to detect generalization in language with an overall PR-AUC of 58.7, with varying degrees of performance across languages. Our research provides data and tools to enable a nuanced understanding of stereotype perpetuation, a crucial step towards more inclusive and responsible language technologies.

Symbolic regression (SR) searches for parametric models that accurately fit a dataset, prioritizing simplicity and interpretability. Despite this secondary objective, studies point out that the models are often overly complex due to redundant operations, introns, and bloat that arise during the iterative process, and can hinder the search with repeated exploration of bloated segments. Applying a fast heuristic algebraic simplification may not fully simplify the expression and exact methods can be infeasible depending on size or complexity of the expressions. We propose a novel agnostic simplification and bloat control for SR employing an efficient memoization with locality-sensitive hashing (LHS). The idea is that expressions and their sub-expressions traversed during the iterative simplification process are stored in a dictionary using LHS, enabling efficient retrieval of similar structures. We iterate through the expression, replacing subtrees with others of same hash if they result in a smaller expression. Empirical results shows that applying this simplification during evolution performs equal or better than without simplification in minimization of error, significantly reducing the number of nonlinear functions. This technique can learn simplification rules that work in general or for a specific problem, and improves convergence while reducing model complexity.

The transformer is a powerful data modelling framework responsible for remarkable performance on a wide range of tasks. However, they are limited in terms of scalability as it is suboptimal and inefficient to process long-sequence data. To this purpose we introduce BLRP (Bidirectional Long-Range Parser), a novel and versatile attention mechanism designed to increase performance and efficiency on long-sequence tasks. It leverages short and long range heuristics in the form of a local sliding window approach combined with a global bidirectional latent space synthesis technique. We show the benefits and versatility of our approach on vision and language domains by demonstrating competitive results against state-of-the-art methods on the Long-Range-Arena and CIFAR benchmarks together with ablations demonstrating the computational efficiency.

Multiple benchmarks have been developed to assess the alignment between deep neural networks (DNNs) and human vision. In almost all cases these benchmarks are observational in the sense they are composed of behavioural and brain responses to naturalistic images that have not been manipulated to test hypotheses regarding how DNNs or humans perceive and identify objects. Here we introduce the toolbox MindSet: Vision, consisting of a collection of image datasets and related scripts designed to test DNNs on 30 psychological findings. In all experimental conditions, the stimuli are systematically manipulated to test specific hypotheses regarding human visual perception and object recognition. In addition to providing pre-generated datasets of images, we provide code to regenerate these datasets, offering many configurable parameters which greatly extend the dataset versatility for different research contexts, and code to facilitate the testing of DNNs on these image datasets using three different methods (similarity judgments, out-ofdistribution classification, and decoder method), accessible at https://github.

In this paper, we present a method for enhancing the accuracy of scene text recognition tasks by judging whether the image and text match each other. While previous studies focused on generating the recognition results from input images, our approach also considers the model's misrecognition results to understand its error tendencies, thus improving the text recognition pipeline. This method boosts text recognition accuracy by providing explicit feedback on the data that the model is likely to misrecognize by predicting correct or incorrect between the image and text. The experimental results on publicly available datasets demonstrate that our proposed method outperforms the baseline and state-of-the-art methods in scene text recognition.

In recent years, generative AI (GenAI), like large language models and text-to-image models, has received significant attention across various domains. However, ensuring the responsible generation of content by these models is crucial for their real-world applicability. This raises an interesting question: \textit{What should responsible GenAI generate, and what should it not?} To answer the question, this paper investigates the practical responsible requirements of both textual and visual generative models, outlining five key considerations: generating truthful content, avoiding toxic content, refusing harmful instruction, leaking no training data-related content, and ensuring generated content identifiable. Specifically, we review recent advancements and challenges in addressing these requirements. Besides, we discuss and emphasize the importance of responsible GenAI across healthcare, education, finance, and artificial general intelligence domains. Through a unified perspective on both textual and visual generative models, this paper aims to provide insights into practical safety-related issues and further benefit the community in building responsible GenAI.