The diffusion of generative AI copilots such as Microsoft Copilot and OpenAI ChatGPT has produced a new class of unsanctioned employee tool usage—commonly labeled shadow AI. This article provides a systematic quantification of shadow AI adoption across Fortune 500 enterprises and evaluates its net effect on the organizational capability gap. Employing a mixed‑methods design that integrates a la...
The rapid proliferation of AI‑generated text in scholarly and professional contexts demands robust, multidimensional evaluation tools that go beyond traditional readability formulas such as Flesch‑Kincaid. This article introduces a composite quality metric that integrates four independent dimensions: (1) surface‑level readability, (2) conceptual density, (3) argumentative coherence, and (4) exp...
Explainability in AI-driven predictive maintenance remains a critical but under‑quantified factor in industrial deployments. This article investigates how the reliability and accuracy of AI-generated explanations affect maintenance decision outcomes in large‑scale manufacturing environments. We define explanation quality along three dimensions—clarity, fidelity, and actionable insight—and const...
Supply chain attacks targeting machine l[REDACTED]g (ML) pipelines have emerged as a critical threat vector, compromising model integrity through poisoning, backdoor insertion, and Trojan triggers embedded in ostensibly benign training data. This article systematically reviews the state of the art in model poisoning attack vectors within open-source ML ecosystems, focusing on Hugging Face Hub a...
Unreported income among small and medium enterprises (SMEs) remains a critical challenge for tax authorities globally, with recent studies indicating that undeclared income accounts for 15-25% of the tax gap in European OECD nations. This article addresses the gap in current tax compliance tools by developing and evaluating machine l[REDACTED]g models specifically designed to detect anomalous c...
The rapid evolution of large language models (LLMs) has e[REDACTED]sed scalability bottlenecks inherent in the Transformer architecture, particularly its quadratic complexity in attention computation. Recent advances propose alternative paradigms—state‑space models (SSMs) such as Mamba and RWKV, as well as hybrid architectures that blend linear attention with selective state propagation—as viab...
Property-based testing (PBT) has emerged as a systematic method for uncovering edge-case failures in complex software systems [1]. Recent extensions to nondeterministic domains, particularly large language models (LLMs), enable the definition of invariants that must hold across varying model outputs [2]. This article introduces a framework for applying PBT to LLM-powered systems, focusing on hy...
Speculative decoding is an inference acceleration technique that leverages a lightweight draft model to propose tokens which are subsequently verified by a target model. This abstract outlines a production-focused benchmark of three speculative decoding implementations — Medusa, Eagle, and SpecTr — evaluated across a diverse set of real-world workloads. We quantify throughput improvements, late...
Enterprise adoption of large language models (LLMs) has progressed from experimental pilots to core production workloads, yet most organizations continue to compute return on investment (ROI) using GPU‐hour pricing as the sole cost driver. This narrow view systematically underestimates the true economic burden of LLMs, omitting fine‑tuning expenses, retrieval‑augmented generation (RAG) infrastr...
The proliferation of AI-driven geopolitical risk forecasting has transformed conflict prediction methodologies, yet systematic validation against real-world outcomes remains incomplete. This study conducts a retrospective evaluation of five major forecasting platforms—including PredictIt, Metaculus, and three commercial vendors—against 47 documented conflict escalations between January 2024 and...