LLM, token, RAG: AI terms explained for beginners
The AI words in every job post and tutorial, explained one at a time with everyday comparisons. No maths, no prior AI knowledge needed.
On this page
Job posts, tutorials and interviews are full of AI words: LLM, token, prompt, embedding, RAG. If you are a student or a fresher, it is easy to nod along without being sure what they mean. This post explains the AI terms beginners hear most, one at a time, with an everyday comparison for each.
Model and LLM
A model is a program that has learned patterns from a lot of examples, instead of following rules a person wrote by hand.
An LLM (large language model) is a model trained on a huge amount of text. Its core skill is predicting which words are likely to come next. That single skill, at large scale, lets it answer questions, summarise, translate and write code.
Everyday comparison: the suggestions your phone keyboard shows while you type, but far more capable.
Prompt
A prompt is the text you give a model: your question plus any instructions or examples. Clear prompts get better answers, the same way a clear request to a colleague gets better work.
A weak prompt: "Explain loops." A better prompt: "Explain Python for-loops to a beginner in five sentences, with one example that adds up the numbers in a list."
Token
Models do not read words; they read tokens, which are small chunks of text. A token might be a whole short word, part of a longer word, or a punctuation mark.
Tokens matter for two practical reasons:
- Cost: paid AI services usually charge per token, for both your input and the model's output.
- Limits: each model can only handle a certain number of tokens at once. This limit is called the context window.
Everyday comparison: the context window is like the model's desk. Only what fits on the desk can be looked at right now.
Training and inference
Training is when a model learns from data. It takes a lot of computing power and is usually done by large companies or research labs.
Inference is when you use a trained model: you send a prompt and get an answer. Almost all developer jobs work with inference, not training.
Hallucination
A hallucination is when a model gives an answer that sounds confident but is wrong or made up, such as a function that does not exist or a fake reference.
This is why the skill of checking AI output matters so much. Our post on learning to code with AI covers habits that help.
Embedding
An embedding turns a piece of text into a list of numbers that represents its meaning. Texts with similar meaning get similar numbers, even if they use different words.
Everyday comparison: placing every sentence as a dot on a map, where "how do I reset my password" and "I forgot my login" land close together.
Embeddings power search that understands meaning instead of only matching exact keywords.
Vector database
A vector database stores embeddings and can quickly find the ones closest to a given embedding. Postgres with the pgvector extension is one popular choice.
RAG
RAG (retrieval-augmented generation) combines the ideas above:
- Turn your documents into embeddings and store them.
- When a question comes in, find the most relevant pieces of your documents.
- Give those pieces to the LLM together with the question, so it answers from your information.
Everyday comparison: an open-book exam. The model does not need to have memorised your company's policies; it reads the right page before answering.
If you want to build one yourself, our tutorial on building a RAG chatbot with Ollama and pgvector walks through it on a normal laptop.
API key
An API key is a secret string that identifies you when your code calls a paid AI service. Treat it like a password: never put it in code you push to GitHub. If APIs are new to you, read that guide first.
Local model
A local model runs on your own computer instead of a company's servers. It costs nothing per request and keeps your data private, which makes it great for learning. Tools like Ollama make this easy; see our comparison of local model runners.
You do not need to memorise this list. Bookmark it, and come back when a term in a job post or tutorial confuses you.
Quick reference
| Term | One-line meaning |
|---|---|
| LLM | A model trained on text that predicts what comes next |
| Prompt | What you send to the model |
| Token | A small chunk of text; the unit of cost and limits |
| Context window | How many tokens the model can handle at once |
| Inference | Using a trained model to get answers |
| Hallucination | A confident but wrong answer |
| Embedding | Text turned into numbers that capture meaning |
| Vector database | Stores embeddings and finds similar ones |
| RAG | Look up relevant documents, then answer |