32584 research outputs found
Sort by
Exploring the Use and Misuse of Large Language Models
No full textLanguage modeling has evolved from simple rule-based systems into complex assistants capable of tackling a multitude of tasks. State-of-the-art large language models (LLMs) are capable of scoring highly on proficiency benchmarks, and as a result have been deployed across industries to increase productivity and convenience. However, the prolific nature of such tools has provided threat actors with the ability to leverage them for attack development. Our paper describes the current state of LLMs, their availability, and their role in benevolent and malicious applications. In addition, we propose how an LLM can be combined with text-to-speech (TTS) voice cloning to create a framework capable of carrying out social engineering attacks. Our case study analyzes the realism of two different open-source TTS models, Tortoise TTS and Coqui XTTS-v2, by calculating similarity scores between generated and real audio samples from four participants. Our results demonstrate that Tortoise is able to generate realistic voice clone audios for native English speaking males, which indicates that easily accessible resources can be leveraged to create deceptive social engineering attacks. As such tools become more advanced, defenses such as awareness, detection, and red teaming may not be able to keep up with dangerously equipped adversaries
High Abundances of Crustose Coralline Algae Inside Cryptic Coral Habitats Linked to Coral Reef Functioning
No full textCryptic habitats on coral reefs consist of crevices, tunnels, and holes that are estimated to comprise 30%–75% of the total reef surface area. These large habitats are vastly understudied because they are often difficult to access. Crustose coralline algae (CCA) are thought to perform important ecological functions, including calcium accretion, both inside cryptic habitats and on the exposed (top) reef. Using GoPro cameras, we surveyed 250 coral crevices from 13 reefs on West Maui, Hawai‘i. We compared relative CCA cover between crevice microhabitats and the well-studied top-reef and identified abiotic and biotic factors that correlated with CCA abundance. We found that crevices had approximately 3.1 times more CCA cover than the top-reef and that CCA cover was highest on crevice ceilings and entrances, which had low sediment and macroalgal cover, compared to the back walls of crevices, which had higher sediment abundance. These results suggest that crevice openings and ceilings are key microhabitats for CCA and therefore may also be important for reef building and stabilization. Thus, these findings help establish functional links between coral reef structural complexity, cryptic habitats, and reef resilience, and highlight the importance of studying cryptic habitats to inform approaches to coral reef conservation
Cosmopolitan Spirits: Islam and the Experimental Cosmologies of Egypt\u27s Spiritualist Movement, 1947–1960
No full textThis article considers the history of Egypt’s mid-twentieth-century Spiritualist movement through an examination of its periodical, a monthly Arabic magazine called Alam al-Ruh (The World of the Spirit) (1947–1960). As it shows, Egyptian Spiritualists defended and promoted their project by crafting an experimental cosmology, one that blended claims about empirical verification with elements of Islamic and Spiritualist cosmologies. It further shows how this combination of scientism and cosmology reflects a core dynamic within the history of scientific exploration in the Islamic world. Like spiritual seekers and occult practitioners in Muslim societies elsewhere, Egyptian Spiritualists positioned their project as one of eradicating superstition from religion, modernizing the nation, and advancing science. By attending to the Egyptian Spiritualist effort to scientize religion and spiritualize science, this article foregrounds the Islamic tradition’s entanglements with scientific discourses and navigates beyond claims about epistemological rupture that often characterize the study of Islam’s relationship to modern science
Bridging SQL Mastery and Career Confidence for Undergraduate Students Through Simulated Job Interviews
No full textEmployers increasingly prioritize candidates who can solve real-world Structured Query Language (SQL) problems, particularly during technical interviews. However, many undergraduate students feel underprepared for these interviews because they have not engaged in the deep learning needed to apply SQL concepts confidently. Additionally, students often fail to recognize the career relevance of SQL skills. This Teaching Tip introduces an immersive SQL lesson designed to bridge the gap between conceptual learning and practical application. The lesson includes a mock SQL technical interview, where students apply their knowledge to solve real-world business problems, class discussions on SQL-related careers, and a post-interview debrief to foster reflection and feedback. Results from pre- and post-lesson surveys indicate significant benefits, including enhanced student confidence in their SQL knowledge, student intention to continue learning and using SQL in the future, and student confidence in their ability to perform well in real SQL interviews. Open-ended survey responses support these findings and further reveal that the SQL lesson positively impacts students by clarifying concepts, reinforcing learned skills, and demonstrating the applicability of SQL in real-world scenarios. This approach demonstrates a practical and scalable framework for integrating immersive professional experiences into technical coursework that may be adapted to different class types (e.g., adopting an abridged version) and different courses (e.g., data analysis)
Rondo: A Minimal Single Page Application for Digital Exhibits
No full textMinimal computing is a promising conceptual framework for digital humanities infrastructure, but the static site architecture most commonly associated with minimal computing can present a steep learning curve, particularly in a workshop or classroom context. This article introduces Rondo, a new minimal framework for digital exhibits which requires no software installation or command line interaction. Rondo differentiates itself from static site tools by adopting a single page application architecture with data stored in a Google Sheet, but it includes tools to disconnect from Google Sheets and create a static site with no external dependencies. Rondo’s reliance on Google Sheets reflects an approach to computing which draws inspiration from Agnès Varda’s film The Gleaners and I. Gleaning, the practice of collecting food or other resources left behind by commercial enterprises like farming, is proposed as a productive framework for exploring the relationship between digital humanities and the technology industry. Rondo’s integration with the Digital Public Library of America presents another way to explore the possibilities of small acts of curation, criticism, and juxtaposition using resources gleaned from larger institutions and corporations
Deep Learning-Based Drowsiness Detection System for Driver’s Safety
No full textDriver drowsiness is a leading contributor to road accidents, accounting for over 100,000 crashes and approximately ~1000 fatalities each year in the United States alone, as per National Safety Council (NSC). To mitigate this urgent public-safety risk, we propose a real-time Driver Drowsiness Detection System that achieves both high accuracy and fast inference on standard hardware. Our key idea is to combine two complementary deep-learning strategies: 1) a custom Convolutional Neural Network (CNN) paired with a Support Vector Machine (SVM) classifier, and 2) a lightweight transfer-learning model built upon a pre-trained convolutional backbone. We evaluate these approaches on two datasets: a four-class Kaggle collection of open/closed eyes and yawn/no-yawn images, and the 37-subject MRL eye dataset. For the custom CNN+SVM pipeline, we optimized the split ratios, dropout rates, and L2 regularization to achieve 100% training accuracy and 99.7% validation accuracy. For the transfer-learning model, we leveraged an existing network to accelerate training, achieving 99.4% training accuracy and 99.1% validation accuracy. Finally, we compare both models using metrics including loss curves, confusion matrices, precision, recall, and F1-score. Our system demonstrates that real-time, highly accurate drowsiness detection is achievable without specialized hardware, paving the way for broader deployment in road-safety applications
Performance and Interpretability Analysis of Code Generation Large Language Models
No full textAs Large Language Models (LLMs) are increasingly getting integrated into software development workflows, understanding their reliability, error patterns and interpretability in real-world development scenarios is crucial for establishing their practical utility. This study evaluates and interprets the performance of 15 open-source LLM models, including Code LLaMa, Granite Code, DeepSeek-Coder-V2, and Yi-Coder on code translation and generation from requirements using the Rosetta Code dataset across diverse programming languages and tasks. Syntactic correctness and code quality are quantified using metrics such as CodeBLEU, chrF, and METEOR. Interpretability is explored through Feature Ablation and Shapley Value Sampling to elucidate prompt processing mechanisms. Results indicate high syntactic correctness and quality scores for models such as DeepSeek-Coder-V2 and Yi-Coder, alongside observed sensitivities to specific prompt components. This research provides quantitative and qualitative insights into the capabilities and limitations of open-source code-generating LLMs, informing model selection and the understanding of LLM-generated code
Temperature and Pressure Shaped the Evolution of Antifreeze Proteins in Polar and Deep Sea Zoarcoid Fishes
No full textAntifreeze proteins (AFPs) have enabled teleost fishes to repeatedly colonize polar seas. Four AFP types have convergently evolved in several fish lineages. AFPs inhibit ice crystal growth and lower tissue freezing point. In lineages with AFPs, species inhabiting colder environments may possess more AFP copies. Elucidating how differences in AFP copy number evolve is challenging due to the genes\u27 tandem array structure and consequently poor resolution of these repetitive regions. Here, we explore the evolution of type III AFPs (AFP III) in the globally distributed suborder Zoarcoidei, leveraging six new long-read genome assemblies. Zoarcoidei has fewer genomic resources relative to other polar fish clades while it is one of the few groups of fishes adapted to both the Arctic and Southern Oceans. Combining these new assemblies with additional long-read genomes available for Zoarcoidei, we conducted a comprehensive phylogenetic test of AFP III evolution and modeled the effects of thermal habitat and depth on AFP III gene family evolution. We confirm a single origin of AFP III via neofunctionalization of the enzyme sialic acid synthase B. We also show that AFP copy number increased under low temperature but decreased with depth, potentially because pressure lowers freezing point. Associations between the environment and AFP III copy number were driven by duplications of paralogs that were translocated out of the ancestral locus at which AFP III arose. Our results reveal novel environmental effects on AFP evolution and demonstrate the value of high-quality genomic resources for studying how structural genomic variation shapes convergent adaptation
The impact of Artificial Intelligence on Ecojustice and Ethics
No full textArtificial Intelligence (AI) has attained human-level performance in tasks like text summarization, machine translation, and code generation while handling multimodal data. Despite these advances, AI poses challenges such as generating false or biased outputs, causing discrimination, and exhibiting opaque decision-making processes that hinder bias mitigation. Additionally, training large language models (LLMs) for AI consumes substantial energy and can enable autonomous warfare, posing a threat to eco-justice, and raising sustainability concerns. Their ability to create realistic content poses the risk of misuse and perpetuating misinformation. This article addresses these ethical issues and presents solutions through regulation and technology, with examples of global legislation tackling algorithmic discrimination and consumer rights. It emphasizes the need for a workforce skilled in navigating an AI-driven world. Future research directions touch upon mechanistic interpretability to understand AI outputs, new approaches to natural language processing and multimodal data understanding, and developing more interpretable neural network architectures with fewer parameters. This article summarizes the tutorial on the topic that includes some of the author’s work in related areas
Synthetic Social Engineering Scenario Generation Using LLMs for Awareness-Based Attack Resilience
No full textSocial engineering is found in a strong majority of cyberattacks today, as it is a powerful manipulation tactic that does not require the technical skills of hacking. Calculated social engineers utilize simple communication to deceive and exploit their victims, all by capitalizing on the vulnerabilities of human nature: trust and fear. When successful, this inconspicuous technique can lead to millions of dollars in losses. Social engineering is not a one-dimensional technique; criminals often leverage a combination of strategies to craft a robust yet subtle attack. In addition, offenders are continually evolving their methods in efforts to surpass preventive measures. A common utility to defend against social engineering attacks is detection-based software. Security awareness, however, is a valuable approach that is often eclipsed by automated tech solutions. Awareness establishes a strong first line of defense against these ever-changing attacks. This study utilizes four data-supplemented large language models to generate custom social engineering scenarios with the goal of supporting strong example-driven security awareness programs. The performances of BERT, T5, GPT-3.5, and Llama 3.1 are comparatively analyzed, with Llama 3.1 producing the highest quality scenarios based on multiple metrics, including LLM-as-a-judge. Through chain-of-thought prompting, the Llama 3.1: Interactive version is capable of generating superiorly realistic, detailed, and specific attack scenarios, which are invaluable to improving hands-on curriculums and security exercises