|
Tulsi Ram Sharma
Chetan Swami
Keywords:
Generative AI; foundation models; large language models; transformers; diffusion models; multimodal AI; RAG; agentic AI; responsible AI; AI safety.
Abstract:
Generative Artificial Intelligence (GenAI) has evolved from specialized generative models into a broad computational paradigm capable of producing text, images, audio, video, software, scientific hypotheses, and multimodal content. The field is increasingly organized around foundation models, transformer-based language models, diffusion and flow-based generators, multimodal architectures, mixture-of-experts (MoE) systems, retrieval-augmented generation (RAG), and emerging agentic architectures. This review synthesizes the architectural foundations of modern GenAI, examines training and adaptation strategies, surveys major application domains, and identifies unresolved technical, societal, and governance challenges. Particular attention is given to the transition from isolated content generation toward systems that combine generation with retrieval, reasoning, planning, tool use, memory, verification, and human feedback. The review further discusses efficiency, hallucination, robustness, privacy, security, provenance, intellectual property, evaluation, and responsible deployment. Future research is expected to emphasize efficient multimodal reasoning, trustworthy agentic systems, personalized and private generation, controllable scientific and industrial models, stronger evaluation frameworks, and hardware–software co-design. The overall trajectory suggests that GenAI will increasingly function as a general-purpose interface between humans, knowledge, software tools, and physical or digital environments.
|
|
International Journal of Recent Research and Review
ISSN: 2277-8322
Vol. XIX, Issue 3
September 2026
|
PDF View
PUBLISHED
September 2026
ISSUE
Vol. XIX, Issue 3
SECTION
Articles
|
