AI Vocabulary: The Words Behind Chatbots and Models

AI has brought a small set of new English words into ordinary use, and a larger set of old words given new work. You do not need all of them. The core below is what you need to talk about the tools, read about them, and ask for what you want.

The nouns

• model — the system itself; the thing that was trained. ChatGPT, Claude, Gemini are each a model. This model is good at code.
• chatbot — the conversation you type into, built on top of a model. The chatbot answered instantly.
• prompt — your message to it. My prompt was too short.
• input — what goes in (your prompt, a file, a pasted text). The input was a long article.
• output — what comes out. The output was three paragraphs.
• response — the model's answer. The response ignored my question.
• context — the surrounding information the model is using. I gave it the context of the whole email thread.

Two pairs worth separating, because they are easily confused:

• input / output describe a flow, like water in and water out.
• prompt / response describe a conversation, your message and its answer.

The verbs

• generate — the model produces text. It generated a summary.
• respond / answer — the neutral verbs for what it does back.
• prompt — you give it an instruction. I prompted it again with more detail.
• train — build a model by showing it examples. Not something you do — something done to the model.
• fine-tune — adjust a model for one specific job. The company fine-tuned it for legal documents.
• run — operate a model. They run the model on their own servers.

Note the direction. You prompt a model; it generates a response. The model prompted a response is wrong — prompting is what the human does.

What the model does not do

English learners overuse two verbs around AI, both worth correcting:

• understand — a model does not understand in the human sense. It predicts. Saying the model doesn't understand my question is fine in everyday speech, but technically it predicts the most likely next words. The model predicts what comes next.
• think — same caution. The model "thinks" step by step is a metaphor, not a fact. In careful English, the model reasons or processes, in quotation marks if you want to be strict.

The safe everyday register. In normal conversation, understand, think and know about a model are fine — everyone uses them. The distinction matters only when you want to be precise, for example in a job interview or when writing about AI.

The words that changed meaning

Several ordinary words now carry a second, AI-specific sense:

Word	Old sense	AI sense
hallucinate	see things that are not there	invent a confident-sounding but false answer
temperature	how hot something is	how creative or random the output is
token	a coin or voucher	a chunk of text the model reads at once
weight	how heavy something is	a number the model learned during training

You meet these fully in how models understand you. For now, notice the pattern: AI borrowed concrete words and gave them abstract work.

Using the words in a sentence

• The model ignored part of my prompt.
• I gave it the context and asked for a short output.
• It generated a response that was too generic.

And the sentence that ties the whole set together: I prompted the chatbot with the input, and its response missed my context.

Test yourself

The quiz below checks the nouns, the verbs, and the pairs people mix up.