Sexism in online media comments is a pervasive challenge that often manifests subtly, complicating moderation efforts as interpretations of what constitutes sexism can vary among individuals. We study monolingual and multilingual open-source text embeddings to reliably detect sexism and misogyny in German-language online comments from an Austrian newspaper. We observed classifiers trained on text embeddings to mimic closely the individual judgements of human annotators. Our method showed robust performance in the GermEval 2024 GerMS-Detect Subtask 1 challenge, achieving an average macro F1 score of 0.597 (4th place, as reported on Codabench). It also accurately predicted the distribution of human annotations in GerMS-Detect Subtask 2, with an average Jensen-Shannon distance of 0.301 (2nd place). The computational efficiency of our approach suggests potential for scalable applications across various languages and linguistic contexts.

We introduce Cognitive Kernel, an open-source agent system towards the goal of generalist autopilots. Unlike copilot systems, which primarily rely on users to provide essential state information (e.g., task descriptions) and assist users by answering questions or auto-completing contents, autopilot systems must complete tasks from start to finish independently, which requires the system to acquire the state information from the environments actively. To achieve this, an autopilot system should be capable of understanding user intents, actively gathering necessary information from various real-world sources, and making wise decisions. Cognitive Kernel adopts a model-centric design. In our implementation, the central policy model (a fine-tuned LLM) initiates interactions with the environment using a combination of atomic actions, such as opening files, clicking buttons, saving intermediate results to memory, or calling the LLM itself. This differs from the widely used environment-centric design, where a task-specific environment with predefined actions is fixed, and the policy model is limited to selecting the correct action from a given set of options. Our design facilitates seamless information flow across various sources and provides greater flexibility. We evaluate our system in three use cases: real-time information management, private information management, and long-term memory management. The results demonstrate that Cognitive Kernel achieves better or comparable performance to other closed-source systems in these scenarios. Cognitive Kernel is fully dockerized, ensuring everyone can deploy it privately and securely. We open-source the system and the backbone model to encourage further research on LLM-driven autopilot systems.

Despite being an integral tool for finding health-related information online, YouTube has faced criticism for disseminating COVID-19 misinformation globally to its users. Yet, prior audit studies have predominantly investigated YouTube within the Global North contexts, often overlooking the Global South. To address this gap, we conducted a comprehensive 10-day geolocation-based audit on YouTube to compare the prevalence of COVID-19 misinformation in search results between the United States (US) and South Africa (SA), the countries heavily affected by the pandemic in the Global North and the Global South, respectively. For each country, we selected 3 geolocations and placed sock-puppets, or bots emulating "real" users, that collected search results for 48 search queries sorted by 4 search filters for 10 days, yielding a dataset of 915K results. We found that 31.55% of the top-10 search results contained COVID-19 misinformation. Among the top-10 search results, bots in SA faced significantly more misinformative search results than their US counterparts. Overall, our study highlights the contrasting algorithmic behaviors of YouTube search between two countries, underscoring the need for the platform to regulate algorithmic behavior consistently across different regions of the Globe.

With over 781 billion website visits globally each month [51], their popularity highlights the growing need for developers to maintain high standards of quality and functionality. Traditional manual web testing approaches, however, can be time-consuming and challenging [8], leading to the increased adoption of automated testing methodologies to streamline the quality assurance process [5, 12, 13, 19, 24, 27, 30, 44, 48, 53, 56, 64]. Despite these advances, conventional testing tools may exhibit challenges and shortcomings regarding testing coverage, potentially overlooking critical bugs and usability issues [18, 19]. The discrepancy between tests generated by conventional methods and real user interactions further compounds these challenges [63], resulting in suboptimal testing outcomes. Web applications typically encompass a spectrum of actions, including form submissions, button clicks, and navigation through various pages. Automated testing tools for web applications encounter challenges stemming from the intricate and dynamic nature of modern web interfaces, which can feature diverse layouts, interactions, and non-deterministic states [3]. To address these challenges and mitigate the limitations of the traditional test generation methods, there has been a growing interest in leveraging deep learning (DL) [12, 13] and reinforcement learning (RL) [22, 23, 26, 27, 30, 31, 48, 64] techniques for automated testing in web applications. By assimilating insights from human testers' behavior, such automated testing approaches aim to emulate human-like interactions with web interfaces, thereby improving the comprehensiveness and effectiveness of testing. However, these DL and RL-based methodologies are not without their constraints.

As the demand for human-like interactions with LLMs continues to grow, so does the interest in manipulating their personality traits, which has emerged as a key area of research. Methods like prompt-based In-Context Knowledge Editing (IKE) and gradient-based Model Editor Networks (MEND) have been explored but show irregularity and variability. IKE depends on the prompt, leading to variability and sensitivity, while MEND yields inconsistent and gibberish outputs. To address this, we employed Opinion QA Based Parameter-Efficient Fine-Tuning (PEFT), specifically Quantized Low-Rank Adaptation (QLORA), to manipulate the Big Five personality traits: Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism. After PEFT, models such as Mistral-7B-Instruct and Llama-2-7B-chat began generating emojis, despite their absence in the PEFT data. For instance, Llama-2-7B-chat generated emojis in 99.5% of extraversion-related test instances, while Mistral-8B-Instruct did so in 92.5% of openness-related test instances. Explainability analysis indicated that the LLMs used emojis intentionally to express these traits. This paper provides a number of novel contributions.

Large Language Model (LLM)-based recommendation systems provide more comprehensive recommendations than traditional systems by deeply analyzing content and user behavior. However, these systems often exhibit biases, favoring mainstream content while marginalizing non-traditional options due to skewed training data. This study investigates the intricate relationship between bias and LLM-based recommendation systems, with a focus on music, song, and book recommendations across diverse demographic and cultural groups. Through a comprehensive analysis conducted over different LLM-models, this paper evaluates the impact of bias on recommendation outcomes. Our findings reveal that bias is so deeply ingrained within these systems that even a simpler intervention like prompt engineering can significantly reduce bias, underscoring the pervasive nature of the issue. Moreover, factors like intersecting identities and contextual information, such as socioeconomic status, further amplify these biases, demonstrating the complexity and depth of the challenges faced in creating fair recommendations across different groups.

This Transformer architectures are the backbone of the modern AI would allow more straightforward tokens present in the input, revolution. However, they are based on simply stacking the which are easier to predict, to directly learn features in shallow same blocks in dozens of layers and processing information layers to predict the outout. It can thus better utilize and sequentially from one block to another. In this paper, we propose reserve deeper complex self-attention blocks for tougher token to challenge this and introduce adaptive computations predictions. Mainly two ideas inspired this paper - First, for LLM-like setups, which allow the final layer to attend to several papers utilize intermediate layers that capture information all of the intermediate layers as it deems fit through the attention at various scales [10, 11, 12, 13], for solving downstream mechanism, thereby introducing computational attention tasks in natural language such as sentiment analysis, shortcuts. These shortcuts can thus make the architecture word representations, etc.

Debiasing is often challenging due to computational costs, data constraints, and potential degradation of multi-task language capabilities. This work introduces a novel approach utilizing ChatGPT to generate synthetic training data, aiming to enhance the debiasing of LLMs. We propose two strategies: Targeted Prompting, which provides effective debiasing for known biases but necessitates prior specification of bias in question; and General Prompting, which, while slightly less effective, offers debiasing across various categories. We leverage resource-efficient LLM debiasing using adapter tuning and compare the effectiveness of our synthetic data to existing debiasing datasets. Our results reveal that: (1) ChatGPT can efficiently produce high-quality training data for debiasing other LLMs; (2) data produced via our approach surpasses existing datasets in debiasing performance while also preserving internal knowledge of a pre-trained LLM; and (3) synthetic data exhibits generalizability across categories, effectively mitigating various biases, including intersectional ones. These findings underscore the potential of synthetic data in advancing the fairness of LLMs with minimal retraining cost.

We introduce Playground v3 (PGv3), our latest text-to-image model that achieves state-of-the-art (SoTA) performance across multiple testing benchmarks, excels in graphic design abilities and introduces new capabilities. Unlike traditional text-to-image generative models that rely on pre-trained language models like T5 or CLIP text encoders, our approach fully integrates Large Language Models (LLMs) with a novel structure that leverages text conditions exclusively from a decoder-only LLM. Additionally, to enhance image captioning quality-we developed an in-house captioner, capable of generating captions with varying levels of detail, enriching the diversity of text structures. We also introduce a new benchmark CapsBench to evaluate detailed image captioning performance. Experimental results demonstrate that PGv3 excels in text prompt adherence, complex reasoning, and accurate text rendering. User preference studies indicate the super-human graphic design ability of our model for common design applications, such as stickers, posters, and logo designs. Furthermore, PGv3 introduces new capabilities, including precise RGB color control and robust multilingual understanding.

Structural network embedding is a crucial step in enabling effective downstream tasks for complex systems that aims to project a network into a lower-dimensional space while preserving similarities among nodes. We introduce a simple and efficient embedding technique based on approximate variants of equitable partitions. The approximation consists in introducing a user-tunable tolerance parameter relaxing the otherwise strict condition for exact equitable partitions that can be hardly found in real-world networks. We exploit a relationship between equitable partitions and equivalence relations for Markov chains and ordinary differential equations to develop a partition refinement algorithm for computing an approximate equitable partition in polynomial time. We compare our method against state-of-the-art embedding techniques on benchmark networks. We report comparable -- when not superior -- performance for visualization, classification, and regression tasks at a cost between one and three orders of magnitude smaller using a prototype implementation, enabling the embedding of large-scale networks which could not be efficiently handled by most of the competing techniques.