Large Language Models (LLMs) have made significant strides in natural language processing, and a precise understanding of the internal mechanisms driving their success is essential. We regard LLMs as transforming embeddings via a discrete, coupled, nonlinear, dynamical system in high dimensions. This perspective motivates tracing the trajectories of individual tokens as they pass through transformer blocks, and linearizing the system along these trajectories through their Jacobian matrices. In our analysis of 38 openly available LLMs, we uncover the alignment of top left and right singular vectors of Residual Jacobians, as well as the emergence of linearity and layer-wise exponential growth. Notably, we discover that increased alignment $\textit{positively correlates}$ with model performance. Metrics evaluated post-training show significant improvement in comparison to measurements made with randomly initialized weights, highlighting the significant effects of training in transformers. These findings reveal a remarkable level of regularity that has previously been overlooked, reinforcing the dynamical interpretation and paving the way for deeper understanding and optimization of LLM architectures.

The availability of highly convincing audio deepfake generators highlights the need for designing robust audio deepfake detectors. Existing works often rely solely on real and fake data available in the training set, which may lead to overfitting, thereby reducing the robustness to unseen manipulations. To enhance the generalization capabilities of audio deepfake detectors, we propose a novel augmentation method for generating audio pseudo-fakes targeting the decision boundary of the model. Inspired by adversarial attacks, we perturb original real data to synthesize pseudo-fakes with ambiguous prediction probabilities. Comprehensive experiments on two well-known architectures demonstrate that the proposed augmentation contributes to improving the generalization capabilities of these architectures.

In this study, we conduct a comprehensive review of smart grid security, exploring system architectures, attack methodologies, defense strategies, and future research opportunities. We provide an in-depth analysis of various attack vectors, focusing on new attack surfaces introduced by advanced components in smart grids. The review particularly includes an extensive analysis of coordinated attacks that incorporate multiple attack strategies and exploit vulnerabilities across various smart grid components to increase their adverse impact, demonstrating the complexity and potential severity of these threats. Following this, we examine innovative detection and mitigation strategies, including game theory, graph theory, blockchain, and machine learning, discussing their advancements in counteracting evolving threats and associated research challenges. In particular, our review covers a thorough examination of widely used machine learning-based mitigation strategies, analyzing their applications and research challenges spanning across supervised, unsupervised, semi-supervised, ensemble, and reinforcement learning. Further, we outline future research directions and explore new techniques and concerns. We first discuss the research opportunities for existing and emerging strategies, and then explore the potential role of new techniques, such as large language models (LLMs), and the emerging threat of adversarial machine learning in the future of smart grid security.

PaliGemma is an open Vision-Language Model (VLM) that is based on the SigLIP-So400m vision encoder and the Gemma-2B language model. It is trained to be a versatile and broadly knowledgeable base model that is effective to transfer. It achieves strong performance on a wide variety of open-world tasks. We evaluate PaliGemma on almost 40 diverse tasks including standard VLM benchmarks, but also more specialized tasks such as remote-sensing and segmentation.

Large language models (LLMs) are increasingly deployed in real-world scenarios with the help of recent model compression techniques. Such momentum towards local deployment means the use of compressed LLMs will widely impact a large population. However, prior analysis works often prioritize on preserving perplexity which is a direct analogy to training loss. The impact of compression method on other critical aspects of model behavior, particularly safety, still calls for a systematic assessment. To this end, we investigate the impact of model compression on four dimensions: (1) degeneration harm, i.e., bias and toxicity in generation; (2) representational harm, i.e., biases in discriminative tasks; (3) dialect bias; (4) language modeling and downstream task performance. We cover a wide spectrum of LLM compression techniques, including unstructured pruning, semi-structured pruning and quantization. Our analysis reveals that compression can lead to unexpected consequences. Although compression may unintentionally remedy LLMs' degeneration harm, it can still exacerbate on the representational harm axis. Although compression may unintentionally remedy LLMs' degeneration harm, it can still exacerbate on the representational harm axis. Moreover, there is a divergent impact on different protected groups as the compression rate grows. Finally, different compression methods have drastically different safety impacts, e.g., quantization mostly preserves bias while pruning degrades quickly. Our findings underscore the importance of integrating safety assessments into the development of compressed LLMs to ensure their reliability across real-world applications. Our full results are available here: \url{https://github.com/zhichaoxu-shufe/Beyond-Perplexity-Compression-Safety-Eval}

The reconfiguration of human-LM interactions from simple sentence completions to complex, multi-domain, humanlike engagements necessitates new methodologies to understand how humans choose to rely on LMs. In our work, we contend that reliance is influenced by numerous factors within the interactional context of a generation, a departure from prior work that used verbalized confidence (e.g., "I'm certain the answer is...") as the key determinant of reliance. Here, we introduce Rel-A.I., an in situ, system-level evaluation approach to measure human reliance on LM-generated epistemic markers (e.g., "I think it's..", "Undoubtedly it's..."). Using this methodology, we measure reliance rates in three emergent human-LM interaction settings: long-term interactions, anthropomorphic generations, and variable subject matter. Our findings reveal that reliance is not solely based on verbalized confidence but is significantly affected by other features of the interaction context. Prior interactions, anthropomorphic cues, and subject domain all contribute to reliance variability. An expression such as, "I'm pretty sure it's...", can vary up to 20% in reliance frequency depending on its interactional context. Our work underscores the importance of context in understanding human reliance and offers future designers and researchers with a methodology to conduct such measurements.

The advent of large language models (LLMs) has heralded a new era in natural language processing (NLP), offering capabilities that range from sophisticated text generation to nuanced language understanding. These advancements have been propelled by significant improvements in model architectures, algorithms, and, crucially, the availability of extensive datasets for training. Given the data-intensive nature of these models, the quest for high-quality, diverse, and substantial datasets has become paramount. In this context, massive web-mined corpora have emerged as a vital resource, offering an abundance of textual data that mirrors the vastness and variety of human language and interaction [22, 35, 37, 42]. The internet, with its exponential growth and dynamic content, presents a near-infinite source of text data, spanning every conceivable topic, language, and style. This richness makes web-mined data an attractive foundation for training LLMs, aiming to equip them with a broad understanding of language and its applications. However, the use of such data is not without its challenges. The process of web mining--extracting data from websites--entails navigating a complex landscape of technical, legal, ethical, and quality-related issues [12, 13, 15, 43, 46]. By critically examining the use of web-mined corpora in the pre-training of LLMs, this article contributes to a nuanced understanding of the current landscape and future directions in large-scale language model development.

Hallucination is often regarded as a major impediment for using large language models (LLMs), especially for knowledge-intensive tasks. Even when the training corpus consists solely of true statements, language models still generate hallucinations in the form of amalgamations of multiple facts. We coin this phenomenon as ``knowledge overshadowing'': when we query knowledge from a language model with multiple conditions, some conditions overshadow others, leading to hallucinated outputs. This phenomenon partially stems from training data imbalance, which we verify on both pretrained models and fine-tuned models, over a wide range of LM model families and sizes.From a theoretical point of view, knowledge overshadowing can be interpreted as over-generalization of the dominant conditions (patterns). We show that the hallucination rate grows with both the imbalance ratio (between the popular and unpopular condition) and the length of dominant condition description, consistent with our derived generalization bound. Finally, we propose to utilize overshadowing conditions as a signal to catch hallucination before it is produced, along with a training-free self-contrastive decoding method to alleviate hallucination during inference. Our proposed approach showcases up to 82% F1 for hallucination anticipation and 11.2% to 39.4% hallucination control, with different models and datasets.

In light of prominent discourse around the negative implications of generative AI, an emerging area of research is investigating the current and estimated impacts of AI-generated propaganda on African citizens participating in elections. Throughout Africa, there have already been suspected cases of AI-generated propaganda influencing electoral outcomes or precipitating coups in countries like Nigeria, Burkina Faso, and Gabon, underscoring the need for comprehensive research in this domain. This paper aims to highlight the risks associated with the spread of generative AI-driven disinformation within Africa while concurrently examining the roles of government, civil society, academia, and the general public in the responsible development, practical use, and robust governance of AI. To understand how African governments might effectively counteract the impact of AI-generated propaganda, this paper presents case studies illustrating the current usage of generative AI for election-related propaganda in Africa. Subsequently, this paper discusses efforts by fact-checking organisations to mitigate the negative impacts of disinformation, explores the potential for new initiatives to actively engage citizens in literacy efforts to combat disinformation spread, and advocates for increased governmental regulatory measures. Overall, this research seeks to increase comprehension of the potential ramifications of AI-generated propaganda on democratic processes within Africa and propose actionable strategies for stakeholders to address these multifaceted challenges.

This review introduces the transformative potential of generative Artificial Intelligence (AI) and foundation models, including large language models (LLMs), for health technology assessment (HTA). We explore their applications in four critical areas, evidence synthesis, evidence generation, clinical trials and economic modeling: (1) Evidence synthesis: Generative AI has the potential to assist in automating literature reviews and meta-analyses by proposing search terms, screening abstracts, and extracting data with notable accuracy; (2) Evidence generation: These models can potentially facilitate automating the process and analyze the increasingly available large collections of real-world data (RWD), including unstructured clinical notes and imaging, enhancing the speed and quality of real-world evidence (RWE) generation; (3) Clinical trials: Generative AI can be used to optimize trial design, improve patient matching, and manage trial data more efficiently; and (4) Economic modeling: Generative AI can also aid in the development of health economic models, from conceptualization to validation, thus streamlining the overall HTA process. Despite their promise, these technologies, while rapidly improving, are still nascent and continued careful evaluation in their applications to HTA is required. To ensure their responsible use and implementation, both developers and users of research incorporating these tools, should familiarize themselves with their current limitations, including the issues related to scientific validity, risk of bias, and consider equity and ethical implications. We also surveyed the current policy landscape and provide suggestions for HTA agencies on responsibly integrating generative AI into their workflows, emphasizing the importance of human oversight and the fast-evolving nature of these tools.