AI Glossary: Essential Terminology You Need to Know Now

With the proliferation of AI, specialized terminology and acronyms are entering everyday conversation and business settings at an unprecedented pace. Although most of these terms originated in technical circles, they now frequently appear in news and product descriptions, making it increasingly important for those unfamiliar with AI to grasp their basic meanings. This article organizes and explains commonly used important AI-related terminology.

First, "Generative AI" is a collective term for AI capable of creating new content such as text, images, audio, and code. ChatGPT and Midjourney are prime examples, and they differ significantly from traditional classification and search-type AI in that they generate output from scratch in response to given instructions (prompts). As this technology became widely adopted, the various concepts introduced subsequently also became widely used among the general public.

"LLM (Large Language Model)" refers to a large-scale AI model trained on vast amounts of text data. By learning patterns from human-written text, it can naturally perform tasks such as answering questions, summarizing texts, and translating. Models like GPT-4, Claude, and Gemini exemplify this, and they form the core of many generative AI services. "Parameters" refer to the number of internal variables that the model adjusts during the learning process, and it is said that larger numbers tend to enable handling of more complex expressions.

"Prompt" refers to an instruction given to AI, and the quality of output changes significantly depending on how it is written. The technique of refining this writing style is called "prompt engineering" and has gained attention as a practical skill in AI utilization. Meanwhile, the phenomenon where AI plausibly generates information that contradicts facts is called "hallucination," and special care is needed in fields such as healthcare, law, and journalism where accuracy is essential.

"RAG (Retrieval-Augmented Generation)" is a mechanism that allows AI to search and reference external databases or documents when generating answers. While the knowledge from LLM's training data has temporal and scope limitations, combining RAG can supplement the latest information and specialized knowledge in specific domains. It is widely used when companies build AI assistants based on internal documents. "Fine-tuning" is the opposite approach, referring to the technique of adding specific data to an existing model for additional training to specialize it for particular purposes.

"Agent (AI Agent)" refers to AI that can act autonomously not just by answering questions, but by using tools and executing tasks sequentially. It is characterized by the ability to achieve complex goals by combining web search, code execution, file manipulation, and other capabilities, and it is a concept whose importance has rapidly increased since 2024. "Multimodal" is a term referring to the capability of AI to handle multiple formats of data simultaneously—not just text, but also images, audio, and video—and most latest-generation models possess this capability.

Understanding these terms provides a foundation for accurately interpreting AI-related news and product information. As technology evolves daily, new concepts and acronyms are expected to continue increasing. By taking time once to organize the meanings of these terms, you gain the ability to evaluate the features and limitations of each service on your own, taking a first step toward more proactive use of AI.