A program that has been trained to spot patterns and make predictions based on examples.
Example
Think of it like a student who has read millions of books. They have not memorised every word, but they have learned how language, ideas, and answers tend to fit together. An AI model does the same thing, with data instead of books.
Why it matters
When people say "the model got it wrong" they mean the trained program made a mistake, not that someone is lying to you.
Attention
Architecture
Definition
The mechanism inside a Transformer that decides which parts of the input text are most relevant to each other.
Example
In the sentence "The trophy did not fit in the bag because it was too big," attention helps the model figure out that "it" refers to the trophy, not the bag. It does this by looking at the relationships between all the words at once.
Why it matters
Attention is the core idea behind why Transformers are so powerful. It lets the model understand long-range connections in text that older architectures missed.
A chatbot made by OpenAI that lets you have a text conversation with a large language model.
Example
You type a question or request, and ChatGPT generates a response. It is not searching the web or looking things up in a database. It is producing text based on patterns it learned during training.
Why it matters
ChatGPT was the product that made AI feel accessible to everyday people. It is one popular chatbot, not the whole category. Claude, Gemini, and other tools are similar products from other companies.
The maximum amount of text an AI can see at one time, like its working memory for a conversation.
Example
Imagine the AI has a desk. The context window is the size of the desk. Once it is full, older parts of the conversation fall off the edge. The AI can only work with what is currently on the desk.
Why it matters
Long conversations or big documents can overflow the context window, causing the AI to forget earlier messages.
Short for Generative Pre-trained Transformer. The family of language models built by OpenAI that powers ChatGPT.
Example
GPT stands for three things: Generative (it creates new text), Pre-trained (it learned from a huge amount of text before you used it), and Transformer (the architecture that makes it work). GPT-3.5, GPT-4, and GPT-4o are all versions in this family.
Why it matters
Knowing what GPT stands for helps you decode a lot of AI product names and news headlines. It also points to three real ideas worth understanding separately.
When an AI confidently states something that is factually wrong or completely made up.
Example
Ask an AI for a book recommendation and it might invent a title that does not exist, full author name, ISBN, and all, without any sign that it is guessing. That is a hallucination.
Why it matters
AI models predict what sounds right, not what is true. Always double-check facts that matter, especially names, dates, and citations.
A type of AI model trained on enormous amounts of text so it can generate, summarise, translate, and reason about language.
Example
ChatGPT is OpenAI's chatbot, Claude is Anthropic's chatbot, and Gemini is Google's chatbot. They are all built on large language models. The LLM is the engine under the hood. The chatbot is the interface you talk to.
Why it matters
LLM is the most important term in modern AI. When people debate AI safety, AI costs, or AI capabilities, they are usually talking about LLMs.
"Summarise this email in three bullet points" is a prompt. So is "write me a poem about Mondays." Whatever you type in the chat box, that is your prompt.
Why it matters
The quality of what you get back depends heavily on how clearly you ask. Better prompts produce better answers.
A technique where the AI looks things up in a document or database before answering, so it can give more accurate, up-to-date answers.
Example
Imagine the AI is taking an open-book exam. Instead of relying purely on memory (training), it can flip open the textbook (retrieval) and then write its answer (generation). That is RAG.
Why it matters
RAG reduces hallucinations and lets AI answer questions about specific documents or recent events it was not trained on.
A small chunk of text, roughly a word or part of a word, that the AI processes as one unit.
Example
The sentence "I love pizza" might become three tokens: "I", " love", " pizza". Longer or rarer words often split into more pieces. The AI never reads full words the way you do. It works with these puzzle pieces instead.
Why it matters
Tokens affect cost and limits. Most AI tools cap how many tokens you can send or receive in one go. That is the context limit.
The process of teaching a model by showing it huge amounts of examples so it learns patterns.
Example
It is like practising for a sport. The model runs through billions of examples, adjusting its internal settings each time it gets something wrong. After enough practice, it gets pretty good at the task.
Why it matters
Training happens before you ever use the model. When you chat with an AI, the training is already done. You are talking to the finished product.
Transformer
Architecture
Definition
The neural network architecture that powers almost every modern language model, including GPT, Claude, and Gemini.
Example
Before Transformers, AI read text word by word and often forgot the beginning of long sentences. A Transformer looks at the whole input at once and figures out which parts relate to each other. That is why it is so much better at understanding context.
Why it matters
Every time you hear about a "language model," there is almost certainly a Transformer underneath. It is the breakthrough that made modern AI possible.