Prompt engineering and prompt-tuning: Foundations, advancements and research direction

The rapid evolution of large language models (LLMs) has shifted adaptation strategies away from full model fine-tuning and toward prompt-driven control. Prompt engineering enables LLMs to perform new tasks through carefully structured natural-language instructions, while prompt-tuning and related continuous prompting techniques introduce efficient mechanisms for task customization without modifying underlying model parameters. This paper presents an integrated examination of prompt-based methodologies, outlining the foundational developments that established prompting as a central paradigm in modern AI systems. It further analyzes key distinctions between discrete, continuous, and dynamic prompting approaches, highlighting their conceptual connections and performance characteristics. Through a detailed and structured review of influential literature, the article synthesizes how prompting methods have advanced cross-domain adaptation, semantic controllability, code generation, security analysis, multimodal retrieval, and other application areas. The paper concludes by identifying research opportunities related to interpretability, automatically generated prompts, multimodal extensions, robustness under adversarial or variable inputs, and the role of prompting in autonomous and human-centered AI systems.