AI glossary

Common AI words explained in plain language: LLM, token, RAG, fine-tuning, hallucination and more.

LLM (large language model): An AI model trained on huge amounts of text to predict the next word. That simple skill lets it answer questions, write, summarize, translate and write code. ChatGPT, Claude and Gemini are built on LLMs.
Token: The small piece of text an AI model reads and writes — often part of a word. Prices and limits are counted in tokens. In English 1 token is about 4 characters; Korean, Japanese and Chinese usually use more tokens per character.
Context window: How much text a model can look at in one go — your messages, attached files and its own answers together. A bigger window lets it work with long documents, but costs more per request.
Prompt: The instruction or question you give an AI. Clear goals, examples and the format you want usually get much better answers than a one-line request.
Fine-tuning: Training an existing model a little more on your own examples so it follows a specific style or task. Cheaper than training from scratch, but needs good data.
RAG (retrieval-augmented generation): Letting a model look things up in your documents before answering. It searches, puts the relevant parts into the prompt, then answers — so answers stay up to date without retraining.
Embedding: Turning text or images into a list of numbers so that similar meanings end up close together. Used for search, recommendations and RAG.
Parameters (7B, 70B…): The adjustable numbers inside a model. “7B” means 7 billion. More parameters often means a more capable model, but also more memory and slower answers.
Quantization: Storing a model’s numbers with fewer bits (for example 4-bit instead of 16-bit) so it fits on a normal PC or phone. It gets much smaller and faster, with a small loss in quality.
Open weights: A model whose trained numbers can be downloaded and run on your own hardware. Check each model’s license: some allow commercial use, some do not.
Hallucination: When an AI states something false with confidence — a fake quote, a wrong number, a source that does not exist. Always check important facts against the original source.
Multimodal: A model that understands more than text — images, audio or video too. You can show it a photo or chart and ask questions about it.
Diffusion model: The kind of model behind most AI image and video generators. It starts from random noise and removes it step by step until a picture that matches your prompt appears.
AI agent: An AI that does multi-step work on its own — searching, using tools, writing files, checking results — instead of only answering one message.
Inference: Running a trained model to get an answer. Training happens once and is expensive; inference happens every time someone uses the model, so its speed and cost matter most for services.
Benchmark: A standard test used to compare models, such as math problems or coding tasks. Useful as a rough guide, but a high score does not guarantee the model is best for your own work.