What AI Is (and Isn’t)

Foundations

What AI Is (and Isn’t)

A plain-English look at what large language models actually do — and where a teacher’s judgment has to stay in charge.

This video was created by AI.

Learning Outcomes

By the end of this lesson, you will be able to:

Define AI and large language models in plain, everyday language.
Explain, at a high level, how an LLM is trained — from raw text to useful responses.
Recognise the key limitations of LLMs — including hallucination and lack of real-world understanding — and explain why those limits matter in a classroom context.
Apply the “humans-in-the-loop” principle to your own practice, keeping professional judgment at the centre of every AI-assisted decision.

What Is a Large Language Model?

A large language model (LLM) is the technology behind tools like Gemini and ChatGPT. Here is what that actually means in practice:

It is a program trained on billions of words to predict the next word in a sequence.
It learns patterns from huge amounts of text — books, articles, and the open web.
It responds in fluent, human-like language, which can make its output feel authoritative.
It is predicting, not understanding — there is no comprehension or awareness behind the words it produces.

How an LLM Is Trained

Stage One: Pre-training

The model is fed enormous amounts of text — books, articles, the open web — so it learns how language fits together.

Stage Two: Human Feedback

The model’s internal settings — its weights — are tuned using human feedback so that its answers become more useful and accurate.

Stage Three: Fine-tuning

The model is further trained for specific jobs — such as drafting a lesson plan — so it performs well in targeted contexts.

What It Learns

Through training, the model learns statistical patterns in language — which words and ideas tend to follow others — not facts or meaning in any human sense.

What “Weights” Are

Weights are the billions of numerical settings inside the model that are adjusted during training — they encode everything the model has learned about language.

Predicting, Not Understanding

Even after all three stages, the model is still predicting the next word — it is not reasoning, understanding, or making judgments the way a teacher does.

The Real Limits of LLMs

LLMs have significant limitations every teacher should know. They can state false or invented information with total confidence — a phenomenon called hallucination. They have no real-world understanding or awareness; they are pattern-matching, not thinking. And they become increasingly unreliable on anything far outside their training data.

The single most important principle to carry into your AI practice is humans-in-the-loop: the teacher remains the decision-maker at every step. AI can draft a worksheet, brainstorm discussion questions, or suggest differentiation strategies — but it cannot weigh up a student’s needs, read a classroom’s mood, or take professional responsibility for what gets taught. That part never gets delegated.

Check Your Understanding

A quick reflection to lock in the foundations before you move on.

Capstone Activity

Apply What You Learned

Name one task you’d trust an LLM to draft this week — and one you wouldn’t hand it. What makes the difference? Write down the line you draw, and describe the check you’ll run before using anything it produces.

Your Response:

[Capstone Box]

Curriki Launches Curriki AI and Grants Initiative to Support Revolutionary Learning Tools.

Curriki Launches Curriki AI and Grants Initiative to Support Revolutionary Learning Tools. Learn More