Extant work shows that generative AI models such as GPT-3.5 and 4 perpetuate social stereotypes and biases. One concerning but less explored source of bias is ideology. Do GPT models take ideological stances on politically sensitive topics? In this article, we provide an original approach to identifying ideological bias in generative models, showing that bias can stem from both the training data and the filtering algorithm. We leverage linguistic variation in countries with contrasting political attitudes to evaluate bias in average GPT responses to sensitive political topics in those languages. First, we find that GPT output is more conservative in languages that map well onto conservative societies (i.e., Polish), and more liberal in languages used uniquely in liberal societies (i.e., Swedish). This result provides strong evidence of training data bias in GPT models. Second, differences across languages observed in GPT-3.5 persist in GPT-4, even though GPT-4 is significantly more liberal due to OpenAI's filtering policy. Our main takeaway is that generative model training must focus on high-quality, curated datasets to reduce bias, even if it entails a compromise in training data size. Filtering responses after training only introduces new biases and does not remove the underlying training biases.

Recent advances in language modelling has significantly decreased the need of labelled data in text classification tasks. Transformer-based models, pre-trained on unlabeled data, can outmatch the performance of models trained from scratch for each task. However, the amount of labelled data need to fine-tune such type of model is still considerably high for domains requiring expert-level annotators, like the legal domain. This paper investigates the best strategies for optimizing the use of a small labeled dataset and large amounts of unlabeled data and perform a classification task in the legal area with 50 predefined topics. More specifically, we use the records of demands to a Brazilian Public Prosecutor's Office aiming to assign the descriptions in one of the subjects, which currently demands deep legal knowledge for manual filling. The task of optimizing the performance of classifiers in this scenario is especially challenging, given the low amount of resources available regarding the Portuguese language, especially in the legal domain. Our results demonstrate that classic supervised models such as logistic regression and SVM and the ensembles random forest and gradient boosting achieve better performance along with embeddings extracted with word2vec when compared to BERT language model. The latter demonstrates superior performance in association with the architecture of the model itself as a classifier, having surpassed all previous models in that regard. The best result was obtained with Unsupervised Data Augmentation (UDA), which jointly uses BERT, data augmentation, and strategies of semi-supervised learning, with an accuracy of 80.7% in the aforementioned task.

The increasing availability of real-world conversation data offers exciting opportunities for researchers to study user-chatbot interactions. However, the sheer volume of this data makes manually examining individual conversations impractical. To overcome this challenge, we introduce WildVis, an interactive tool that enables fast, versatile, and large-scale conversation analysis. WildVis provides search and visualization capabilities in the text and embedding spaces based on a list of criteria. To manage million-scale datasets, we implemented optimizations including search index construction, embedding precomputation and compression, and caching to ensure responsive user interactions within seconds. We demonstrate WildVis' utility through three case studies: facilitating chatbot misuse research, visualizing and comparing topic distributions across datasets, and characterizing user-specific conversation patterns. WildVis is open-source and designed to be extendable, supporting additional datasets and customized search and visualization functionalities.

Information retrieval (IR) is the task of finding relevant documents in response to a user query. Although Spanish is the second most spoken native language, current IR benchmarks lack Spanish data, hindering the development of information access tools for Spanish speakers. We introduce MessIRve, a large-scale Spanish IR dataset with around 730 thousand queries from Google's autocomplete API and relevant documents sourced from Wikipedia. MessIRve's queries reflect diverse Spanish-speaking regions, unlike other datasets that are translated from English or do not consider dialectal variations. The large size of the dataset allows it to cover a wide variety of topics, unlike smaller datasets. We provide a comprehensive description of the dataset, comparisons with existing datasets, and baseline evaluations of prominent IR models. Our contributions aim to advance Spanish IR research and improve information access for Spanish speakers.

In this paper, we introduce PhishLang, an open-source, lightweight language model specifically designed for phishing website detection through contextual analysis of the website. Unlike traditional heuristic or machine learning models that rely on static features and struggle to adapt to new threats, and deep learning models that are computationally intensive, our model leverages MobileBERT, a fast and memory-efficient variant of the BERT architecture, to learn granular features characteristic of phishing attacks. PhishLang operates with minimal data preprocessing and offers performance comparable to leading deep learning anti-phishing tools, while being significantly faster and less resource-intensive. Over a 3.5-month testing period, PhishLang successfully identified 25,796 phishing URLs, many of which were undetected by popular antiphishing blocklists, thus demonstrating its potential to enhance current detection measures. Capitalizing on PhishLang's resource efficiency, we release the first open-source fully client-side Chromium browser extension that provides inference locally without requiring to consult an online blocklist and can be run on low-end systems with no impact on inference times. Our implementation not only outperforms prevalent (server-side) phishing tools, but is significantly more effective than the limited commercial client-side measures available. Furthermore, we study how PhishLang can be integrated with GPT-3.5 Turbo to create explainable blocklisting -- which, upon detection of a website, provides users with detailed contextual information about the features that led to a website being marked as phishing.

Automated content moderation systems have long been used to help reduce the occurrence of violent, hateful, sexual, or otherwise undesired user-generated content online, including in online comment sections and by social media platforms [7, 19, 24]. As content is generated by AI systems, automated content moderation techniques are being applied to the text generated by these systems to filter unwanted content before it is shown to users [21, 22]. However, content moderation is known to suffer from identity-related biases, such that speech by or about marginalized identities is more likely to be incorrectly flagged as inappropriate content [5, 10, 27]. In this paper, we conduct an audit of five content moderation systems to measure identity-related speech suppression, introducing benchmark datasets and definitions to quantify these biases in the context of generative AI systems. Previous assessments of content moderation systems have used benchmark datasets to measure effectiveness and bias. These include datasets composed of user-generated content, such as tweets or internet comments, that have been hand-labeled according to a content moderation rubric [2, 8]. However, most of these datasets are composed of short-form content and do not include the types of text involved in generative AI systems, be they user-generated prompts or system-provided responses. Automated content moderation systems applied in generative AI settings may have unexpected or undesired results, for example flagging PG-rated movie scripts as inappropriate content [21]. As generative AI is increasingly used for creative and expressive text generation from schools to Hollywood, this paper is motivated by this question: whose stories won't be told?

The $k$-nearest neighbor ($k$-NN) algorithm is one of the most popular methods for nonparametric classification. However, a relevant limitation concerns the definition of the number of neighbors $k$. This parameter exerts a direct impact on several properties of the classifier, such as the bias-variance tradeoff, smoothness of decision boundaries, robustness to noise, and class imbalance handling. In the present paper, we introduce a new adaptive $k$-nearest neighbours ($kK$-NN) algorithm that explores the local curvature at a sample to adaptively defining the neighborhood size. The rationale is that points with low curvature could have larger neighborhoods (locally, the tangent space approximates well the underlying data shape), whereas points with high curvature could have smaller neighborhoods (locally, the tangent space is a loose approximation). We estimate the local Gaussian curvature by computing an approximation to the local shape operator in terms of the local covariance matrix as well as the local Hessian matrix. Results on many real-world datasets indicate that the new $kK$-NN algorithm yields superior balanced accuracy compared to the established $k$-NN method and also another adaptive $k$-NN algorithm. This is particularly evident when the number of samples in the training data is limited, suggesting that the $kK$-NN is capable of learning more discriminant functions with less data considering many relevant cases.

Attention mechanisms and Transformer architectures have revolutionized Natural Language Processing (NLP) by enabling exceptional modeling of long-range dependencies and capturing intricate linguistic patterns. However, their inherent reliance on linear operations in the form of matrix multiplications limits their ability to fully capture inter-token relationships on their own. We propose MöbiusAttention, a novel approach that integrates Möbius transformations within the attention mechanism of Transformer-based models. Möbius transformations are non-linear operations in spaces over complex numbers with the ability to map between various geometries. By incorporating these properties, MöbiusAttention empowers models to learn more intricate geometric relationships between tokens and capture a wider range of information through complex-valued weight vectors. We build and pre-train a BERT and a RoFormer version enhanced with MöbiusAttention, which we then finetune on the GLUE benchmark. We evaluate empirically our approach against the baseline BERT and RoFormer models on a range of downstream tasks. Our approach compares favorably against the baseline models, even with smaller number of parameters suggesting the enhanced expressivity of MöbiusAttention. This research paves the way for exploring the potential of Möbius transformations in the complex projective space to enhance the expressivity and performance of foundation models. At the heart of their success lies the attention mechanism (Vaswani et al., 2017), a powerful tool that enables them to identify relationships between different parts of the data, be it words in a sentence or image patches in a scene. Despite their remarkable impact, current transformers face limitations. A key constraint is the inherent linearity of the attention mechanism, which primarily relies on weights learned through linear transformations, matrix multiplications, and the softmax function. While softmax is a non-linear operation, it is only used to produce a probability distribution over the elements signaling their relative importance in comparison to the others, and not to introduce non-linear interdependencies. Predominantly linear operations restrict the ability of models to capture complex linguistic dependencies, leading to potential information loss within each attention layer as shown by recent research (Zhang, 2023). Figure 1: Various Möbius transformations: Each sub-figure shows flows from a single point after successive transformations. Elliptic Möbius has two fixed points at the centers of two circular flows.

Table-based Question Answering (TQA) involves answering questions based on tabular data. The complexity of table structures and question logic makes this task difficult even for Large Language Models (LLMs). This paper improves TQA performance by leveraging LLMs' reasoning capabilities. Inspired by how humans solve TQA tasks, we propose a Seek-and-Solve pipeline that instructs the LLM to first seek relevant information and then answer questions. The two stages are integrated at the reasoning level, and their Chain of Thought (CoT) paths are integrated into a coherent Seek-and-Solve CoT (SS-CoT). Furthermore, we present a compact single-stage TQA-solving prompt distilled from the pipeline. Experiments demonstrate that under In-Context Learning settings, using samples with SS-CoT paths as demonstrations, the TQA-solving prompt can effectively guide the LLM to solve complex TQA tasks, resulting in improved performance and reliability. Our results highlight the importance of properly eliciting LLMs' reasoning capabilities in solving complex TQA tasks.

The influence of individual differences on the perception and evaluation of interactions with robots has been researched for decades. Some human demographic characteristics have been shown to affect how individuals perceive interactions with robots. Still, it is to-date not clear whether, which and to what extent individual differences influence how we perceive robots, and even less is known about human factors and their effect on the perception of robot movements. In addition, most results on the relevance of individual differences investigate human-robot interactions with humanoid or social robots whereas interactions with industrial robots are underrepresented. We present a literature review on the relationship of robot movements and the influence of demographic variation. Our review reveals a limited comparability of existing findings due to a lack of standardized robot manipulations, various dependent variables used and differing experimental setups including different robot types. In addition, most studies have insufficient sample sizes to derive generalizable results. To overcome these shortcomings, we report the results from a Web-based experiment with 930 participants that studies the effect of demographic characteristics on the evaluation of movement behaviors of an articulated robot arm. Our findings demonstrate that most participants prefer an approach from the side, a large movement range, conventional numbers of rotations, smooth movements and neither fast nor slow movement speeds. Regarding individual differences, most of these preferences are robust to demographic variation, and only gender and age was found to cause slight preference differences between slow and fast movements.