What an AI assistant is, in plain terms

Step 1 Warm-up recall

Write down one way you have used, or seen someone use, an AI assistant at work. Note what you thought of the result.

Your answer

Private: nothing you type here is sent or stored.

Step 2 Learning goal

By the end of this lesson you can:

• describe what a large language model does in outline (Foundation);
• explain why an answer is a likely continuation rather than a lookup (Core);
• say why you remain responsible for what you use (Core).

Step 3 A familiar example

Your phone suggests the next word as you type. An AI assistant works on a vastly larger scale, but the core idea is similar: given the text so far, produce what is likely to come next.

Step 4 Clear explanation

A large language model is a system trained on very large amounts of text so that it can produce text that fits a given start. An AI assistant is a product built on such a model and further trained to follow instructions in a helpful way. When you send a request, the model builds a reply piece by piece, each piece chosen because it is a likely continuation of what came before. See the encyclopedia entries on the large language model and natural language processing.

Three consequences matter at work. First, the reply is generated, not looked up: it can sound authoritative even where it contains an error. Second, it depends on your wording: a different request can give a different answer. Third, it has no accountability. You, not the tool, answer to your manager, client or regulator for what you send.

This is a simplified account. Real systems add many components, and some assistants can search the web or use tools, which changes what they can do. The principle still holds: treat the output as a draft from a fast, well-read assistant who sometimes bluffs. For wider context, see What Is Artificial Intelligence, Really?

Step 5 Worked example (4 marks)

A colleague says: “The AI said it, so it must be in the regulation.” Respond.

1. The reply is a likely continuation of text, not a lookup. (1 mark)
2. Likely wording can still be wrong or invented. (1 mark)
3. Ask for the source, then read the regulation itself. (1 mark)
4. You remain accountable for what you rely on. (1 mark)

Step 6 Learner practice (instant-marked)

Step 7 Recap

• A language model produces likely continuations of text.
• An assistant is a product built on it and trained to follow instructions.
• You, not the tool, are accountable for what you use.

Step 8 Exit ticket (ungraded)

Write two sentences explaining to a colleague why an AI answer needs checking.

Your answer

Private: nothing you type here is sent or stored.

Step 9 Go further (Extension)

Read the encyclopedia entry on the large language model and note one thing the simplified account here leaves out.