Benchmarking Adversarial Robustness to Bias Elicitation in Large Language Models: Scalable Automated Assessment with LLM-as-a-Judge

Large Language Models (LLMs) have revolutionized artificial intelligence, driving advancements in machine translation, summarization, and conversational agents. However, their increasing integration into critical societal domains has raised concerns about embedded biases, which can perpetuate stereotypes and compromise fairness. These biases stem from various sources, including historical inequalities in training data, linguistic imbalances, and adversarial manipulation. Despite mitigation efforts, recent studies indicate that LLMs remain vulnerable to adversarial attacks designed to elicit biased responses. This work proposes a scalable benchmarking framework to evaluate LLM robustness against adversarial bias elicitation. Our methodology involves (i) systematically probing models with a multi-task approach targeting biases across various sociocultural dimensions, (ii) quantifying robustness through safety scores using an LLM-as-a-Judge approach for automated assessment of model responses, and (iii) employing jailbreak techniques to investigate vulnerabilities in safety mechanisms. Our analysis examines prevalent biases in both small and large state-of-the-art models and their impact on model safety. Additionally, we assess the safety of domain-specific models fine-tuned for critical fields, such as medicine. Finally, we release a curated dataset of bias-related prompts, CLEAR-Bias, to facilitate systematic vulnerability benchmarking. Our findings reveal critical trade-offs between model size and safety, aiding the development of fairer and more robust future language models.

Dynamic hashtag recommendation in social media with trend shift detection and adaptation

The widespread use of social media platforms results in the generation of vast amounts of user-generated content, which requires efficient methods for categorization and search. Hashtag recommendation systems have emerged as a crucial tool for automatically suggesting relevant hashtags and improving content discoverability. However, existing static models struggle to adapt to the highly dynamic and real-time nature of social media conversations, where new hashtags emerge and existing ones undergo semantic shifts. To address these challenges, this paper presents H-ADAPTS (Hashtag recommendAtion by Detecting and adAPting to Trend Shifts), a BERT-based hashtag recommendation methodology that can detect and adapt to shifts in the main trends and topics underlying social media conversation. Our approach introduces a trend-aware detection mechanism to identify changes in hashtag usage, triggering efficient model adaptation on a (small) set of recent posts. The framework leverages Apache Storm for real-time stream processing, enabling scalable and fault-tolerant analysis of high-velocity social data. Experimental results on two real-world case studies, including the COVID-19 pandemic and the 2020 US presidential election, demonstrate the ability to maintain high recommendation accuracy by adapting to emerging trends. Our methodology significantly outperforms existing solutions, ensuring timely and relevant hashtag recommendations in dynamic environments.

Are Large Language Models Really Bias-Free? Jailbreak Prompts for Assessing Adversarial Robustness to Bias Elicitation

Large Language Models (LLMs) have revolutionized artificial intelligence, demonstrating remarkable computational power and linguistic capabilities. However, these models are inherently prone to various biases stemming from their training data. These include selection, linguistic, and confirmation biases, along with common stereotypes related to gender, ethnicity, sexual orientation, religion, socioeconomic status, disability, and age. This study explores the presence of these biases within the responses given by the most recent LLMs, analyzing the impact on their fairness and reliability. We also investigate how known prompt engineering techniques can be exploited to effectively reveal hidden biases of LLMs, testing their adversarial robustness against jailbreak prompts specially crafted for bias elicitation. Extensive experiments are conducted using the most widespread LLMs at different scales, confirming that LLMs can still be manipulated to produce biased or inappropriate responses, despite their advanced capabilities and sophisticated alignment processes. Our findings underscore the importance of enhancing mitigation techniques to address these safety issues, toward a more sustainable and inclusive artificial intelligence.

Block size estimation for data partitioning in HPC applications using machine learning techniques

The extensive use of HPC infrastructures and frameworks for running data-intensive applications has led to a growing interest in data partitioning techniques and strategies. In fact, finding an effective partitioning, i.e. a suitable size for data blocks, is a key strategy to speed-up parallel data-intensive applications and increase scalability. This paper describes a methodology for data block size estimation in HPC applications, which relies on supervised machine learning techniques. The implementation of the proposed methodology was evaluated using as a testbed dislib, a distributed computing library highly focused on machine learning algorithms built on top of the PyCOMPSs framework. We assessed the effectiveness of our solution through an extensive experimental evaluation considering different algorithms, datasets, and infrastructures, including the MareNostrum 4 supercomputer. The results we obtained show that the methodology is able to efficiently determine a suitable way to split a given dataset, thus enabling the efficient execution of data-parallel applications in high performance environments.