Bias in artificial intelligence systems has become a critical concern as these technologies increasingly influence decision-making across domains such as healthcare, criminal justice, and employment. Bias manifests as systematic errors that lead to unfair or discriminatory outcomes, often disproportionately affecting marginalised groups. Understanding the origins of bias is essential for
Generative Artificial Intelligence (AI) has revolutionised digital media through its ability to synthesise highly realistic content, with deepfake technology standing as one of its most prominent and contentious applications. The term “deepfake,” derived from “deep learning” and “fake,” refers to synthetic media—typically videos or audio—that convincingly depict individuals
Misinformation, defined as false or misleading information disseminated regardless of intent, poses significant challenges to societal trust and democratic processes (Wardle & Derakhshan, 2017). Unlike disinformation, which involves deliberate deception, misinformation encompasses a broader spectrum, including unintentional errors, rumours, and misinterpretations. The advent GenAI models, capable of producing human-like text,
Content restriction in generative AI manifests as explicit or implicit censorship. Explicit censorship uses predefined rules to block content like hate speech or illegal material, employing keyword blacklists, pattern-matching, or classifiers (Gillespie 2018). DeepSeek’s models, aligned with Chinese regulations, use real-time filters to block politically sensitive content, such as
Detecting hallucinations involves distinguishing accurate outputs from those that deviate from factual or contextual grounding. One approach is consistency checking, where LLM outputs are evaluated against external knowledge bases to identify discrepancies. Manakul et al. (2023) propose SelfCheckGPT, a zero-resource method that uses the model’s internal consistency to detect
Advancements in artificial intelligence (AI), particularly in natural language processing (NLP), highlight critical distinctions between parroting and hallucination in language models. Parroting refers to AI reproducing or mimicking patterns and phrases from training data without demonstrating understanding or creativity. Hallucination involves generating factually incorrect, implausible, or fabricated outputs, often diverging
The rapid integration of Artificial Intelligence (AI) into global economies has driven transformative advancements in sectors such as healthcare and agriculture. However, this technological revolution incurs significant environmental costs, particularly through substantial energy consumption and greenhouse gas (GHG) emissions. The carbon footprint of AI, stemming from energy-intensive processes like hardware
Contemporary researchers face unprecedented financial barriers when engaging with state-of-the-art language models, particularly through API-based services where costs are directly proportional to token consumption and computational resource utilisation. The challenge is compounded by increasing complexity of research tasks requiring extensive prompt engineering, iterative model interactions, and large-scale data processing operations.
The emergence of large language models (LLMs) and generative AI applications has ushered in a new era of artificial intelligence capabilities, fundamentally altering the landscape of computational requirements and associated costs. Generative AI systems, built upon transformer architectures and trained on vast datasets, have demonstrated remarkable scalability and adaptability across
Retrieval-Augmented Generation (RAG) represents a paradigm shift in natural language processing (NLP), integrating large language models (LLMs) with dynamic information retrieval systems to produce responses that are both contextually enriched and factually grounded (Lewis et al. 2020). At its core, the RAG architecture couples a conventional generative model—one that
Most contemporary LLMs employ a decoder‑only transformer architecture, which processes sequences in parallel via self‑attention. However, scaling dense transformers linearly in size increases computation and cost. Mixture‑of‑experts (MoE) approaches address this by activating only a subset of parameters per token. In the Switch Transformer, MoE routing
Small Language Models (SLMs) are compact neural networks designed to perform natural language processing (NLP) tasks with significantly fewer parameters and lower computational requirements than their larger counterparts. SLMs aim to deliver robust performance in resource-constrained environments, such as mobile devices or edge computing systems, where efficiency is paramount. The
