Activation steering has emerged as a powerful method for guiding the behavior of generative models towards desired outcomes such as toxicity mitigation. However, most existing methods apply interventions uniformly across all inputs, degrading model performance when steering is unnecessary. We introduce Dynamically Scaled Activation Steering (DSAS), a method-agnostic steering framework that decouples when to steer from how to steer. DSAS adaptively modulates the strength of existing steering transformations across layers and inputs, intervening strongly only when undesired behavior is detected. At generation time, DSAS computes context-dependent scaling factors that selectively adjust the strength of any steering method. We also show how DSAS can be jointly optimized end-to-end together with the steering function. When combined with existing steering methods, DSAS consistently improves the Pareto front with respect to steering alone, achieving a better trade-off between toxicity mitigation and utility preservation. We further demonstrate DSAS's generality by applying it to a text-to-image diffusion model, showing how adaptive steering allows the modulation of specific concepts. Finally, DSAS introduces minimal computational overhead while improving interpretability, pinpointing which tokens require steering and by how much.

Organization's communication is essential for public trust, but the rise of generative AI models has introduced significant challenges by generating persuasive content that can form competing narratives with official messages from government and commercial organizations at speed and scale. This has left agencies in a reactive position, often unaware of how these models construct their persuasive strategies, making it more difficult to sustain communication effectiveness. In this paper, we introduce a large LLM-generated persuasion attack dataset, which includes 134,136 attacks generated by GPT-4, Gemma 2, and Llama 3.1 on agency news. These attacks span 23 persuasive techniques from SemEval 2023 Task 3, directed toward 972 press releases from ten agencies. The generated attacks come in two mediums, press release statements and social media posts, covering both long-form and short-form communication strategies. We analyzed the moral resonance of these persuasion attacks to understand their attack vectors. GPT-4's attacks mainly focus on Care, with Authority and Loyalty also playing a role. Gemma 2 emphasizes Care and Authority, while Llama 3.1 centers on Loyalty and Care. Analyzing LLM-generated persuasive attacks across models will enable proactive defense, allow to create the reputation armor for organizations, and propel the development of both effective and resilient communications in the information ecosystem.

Cloud-based large language models (LLMs) and their variants have significantly influenced real-world applications. Deploying smaller models (i.e., small language models (SLMs)) on edge devices offers additional advantages, such as reduced latency and independence from network connectivity. However, edge devices' limited computing resources and constrained energy budgets challenge efficient deployment. This study evaluates the power efficiency of five representative SLMs - Llama 3.2, Phi-3 Mini, TinyLlama, and Gemma 2 on Raspberry Pi 5, Jetson Nano, and Jetson Orin Nano (CPU and GPU configurations). Results show that Jetson Orin Nano with GPU acceleration achieves the highest energy-to-performance ratio, significantly outperforming CPU-based setups. Llama 3.2 provides the best balance of accuracy and power efficiency, while TinyLlama is well-suited for low-power environments at the cost of reduced accuracy. In contrast, Phi-3 Mini consumes the most energy despite its high accuracy. In addition, GPU acceleration, memory bandwidth, and model architecture are key in optimizing inference energy efficiency. Our empirical analysis offers practical insights for AI, smart systems, and mobile ad-hoc platforms to leverage tradeoffs from accuracy, inference latency, and power efficiency in energy-constrained environments.

As multilingual Large Language Models (LLMs) gain traction across South Asia, their alignment with local ethical norms, particularly for Bengali, which is spoken by over 285 million people and ranked 6th globally, remains underexplored. Existing ethics benchmarks are largely English-centric and shaped by Western frameworks, overlooking cultural nuances critical for real-world deployment. To address this, we introduce BengaliMoralBench, the first large-scale ethics benchmark for the Bengali language and socio-cultural contexts. It covers five moral domains, Daily Activities, Habits, Parenting, Family Relationships, and Religious Activities, subdivided into 50 culturally relevant subtopics. Each scenario is annotated via native-speaker consensus using three ethical lenses: Virtue, Commonsense, and Justice ethics. We conduct systematic zero-shot evaluation of prominent multilingual LLMs, including Llama, Gemma, Qwen, and DeepSeek, using a unified prompting protocol and standard metrics. Performance varies widely (50-91% accuracy), with qualitative analysis revealing consistent weaknesses in cultural grounding, commonsense reasoning, and moral fairness. BengaliMoralBench provides a foundation for responsible localization, enabling culturally aligned evaluation and supporting the deployment of ethically robust AI in diverse, low-resource multilingual settings such as Bangladesh.

The growing spread of online misinformation has created an urgent need for scalable, reliable fact-checking solutions. Crowdsourced fact-checking - where non-experts evaluate claim veracity - offers a cost-effective alternative to expert verification, despite concerns about variability in quality and bias. Encouraged by promising results in certain contexts, major platforms such as X (formerly Twitter), Facebook, and Instagram have begun shifting from centralized moderation to decentralized, crowd-based approaches. In parallel, advances in Large Language Models (LLMs) have shown strong performance across core fact-checking tasks, including claim detection and evidence evaluation. However, their potential role in crowdsourced workflows remains unexplored. This paper investigates whether LLM-powered generative agents - autonomous entities that emulate human behavior and decision-making - can meaningfully contribute to fact-checking tasks traditionally reserved for human crowds. Using the protocol of La Barbera et al. (2024), we simulate crowds of generative agents with diverse demographic and ideological profiles. Agents retrieve evidence, assess claims along multiple quality dimensions, and issue final veracity judgments. Our results show that agent crowds outperform human crowds in truthfulness classification, exhibit higher internal consistency, and show reduced susceptibility to social and cognitive biases. Compared to humans, agents rely more systematically on informative criteria such as Accuracy, Precision, and Informativeness, suggesting a more structured decision-making process. Overall, our findings highlight the potential of generative agents as scalable, consistent, and less biased contributors to crowd-based fact-checking systems.

Large Language Models (LLMs) demonstrate human-level or even superior language abilities, effectively modeling syntactic structures, yet the specific computational modules responsible remain unclear. A key question is whether LLM behavioral capabilities stem from mechanisms akin to those in the human brain. To address these questions, we introduce the Hierarchical Frequency Tagging Probe (HFTP), a tool that utilizes frequency-domain analysis to identify neuron-wise components of LLMs (e.g., individual Multilayer Perceptron (MLP) neurons) and cortical regions (via intracranial recordings) encoding syntactic structures. Our results show that models such as GPT-2, Gemma, Gemma 2, Llama 2, Llama 3.1, and GLM-4 process syntax in analogous layers, while the human brain relies on distinct cortical regions for different syntactic levels. Representational similarity analysis reveals a stronger alignment between LLM representations and the left hemisphere of the brain (dominant in language processing). Notably, upgraded models exhibit divergent trends: Gemma 2 shows greater brain similarity than Gemma, while Llama 3.1 shows less alignment with the brain compared to Llama 2. These findings offer new insights into the interpretability of LLM behavioral improvements, raising questions about whether these advancements are driven by human-like or non-human-like mechanisms, and establish HFTP as a valuable tool bridging computational linguistics and cognitive neuroscience. This project is available at https://github.com/LilTiger/HFTP.

The rise of Large Language Models has not been inclusive of all cultures. The models are mostly trained on English texts and culture which makes them underperform in other languages and cultural contexts. By developing a generalizable method for preparing culturally relevant datasets and post-training the Gemma 2 model, this project aimed to increase the performance of Gemma 2 for an underrepresented language and showcase how others can do the same to unlock the power of Generative AI in their country and preserve their cultural heritage.

The legal field already uses various large language models (LLMs) in actual applications, but their quantitative performance and reasons for it are underexplored. We evaluated several open-source and proprietary LLMs -- including GPT-series, Anthropic, Deepseek and Llama-3, variants -- on parts of the European Qualifying Examination (EQE) for future European Patent Attorneys. OpenAI o1 led with 0.82 accuracy and 0.81 F1 score, whereas (Amazon Web Services) AWS Llama 3.1 8B lagged at 0.50 accuracy, and a Python-deployed Llama 3.1 8B scored 0.55. The latter two are within the range of mere guessing for the two-answer forced-choice design. None of the evaluated models could have passed the examination fully, as accuracy never exceeded the average threshold of 0.90 required for professional-level standards -- also not models that are regularly promoted for their assumed beyond-PhD- and bar-admitted-lawyer-level performance. GPT-4o excelled at integrating text and graphics, while Claude 3 Opus often lost formatting coherence. Human patent experts evaluated the textual justifications and uncovered various critical shortcomings of each model. They valued clarity and legal rationale over the raw correctness of the answers, which revealed misalignment between automatic metrics and expert judgment. Model outputs were sensitive to modest temperature changes and prompt wording, which underscores the remaining necessity of expert oversight. Future work should target logical consistency, robust multimodality, and adaptive prompting to approach human-level patent proficiency. In summary, despite the outstanding performance of recent large models, the general public might overestimate their performance. The field has a long way to go to develop a virtual patent attorney. This paper wants to point out several specific limitations that need solutions.

Using Large Language Models (LLMs) in real-world applications presents significant challenges, particularly in balancing computational efficiency with model performance. Optimizing acceleration after fine-tuning and during inference is critical for building efficient architectures. This paper introduces Zero-Shot Adjustable Acceleration, a novel training and inference method that dynamically adjusts hardware utilization during inference without requiring additional fine-tuning. The proposed approach is applied to recent LLMs and evaluated across multiple classification and text generation tasks. Experimental results demonstrate that the method supports a wide range of zero-shot acceleration and achieves up to 11x speedup compared to the baseline.

Large Language Models (LLMs) have demonstrated remarkable performance across various Natural Language Processing (NLP) tasks, largely due to their generalisability and ability to perform tasks without additional training. However, their effectiveness for low-resource languages remains limited. In this study, we evaluate the performance of 55 publicly available LLMs on Maltese, a low-resource language, using a newly introduced benchmark covering 11 discriminative and generative tasks. Our experiments highlight that many models perform poorly, particularly on generative tasks, and that smaller fine-tuned models often perform better across all tasks. From our multidimensional analysis, we investigate various factors impacting performance. We conclude that prior exposure to Maltese during pre-training and instruction-tuning emerges as the most important factor. We also examine the trade-offs between fine-tuning and prompting, highlighting that while fine-tuning requires a higher initial cost, it yields better performance and lower inference costs. Through this work, we aim to highlight the need for more inclusive language technologies and recommend that researchers working with low-resource languages consider more "traditional" language modelling approaches.