What Is a Large Language Model? A Parent's Guide

Quick answer

A large language model, or LLM, is the technology behind tools like ChatGPT. It is a computer program trained on enormous amounts of text that learns one core skill: predicting the most likely next word in a sentence, over and over, at remarkable speed. That single trick, scaled up, produces the fluent essays, explanations and answers your teenager sees on screen. What an LLM does not do is understand, check or believe anything it writes. It has no idea whether a sentence is true, only whether it is likely. That is why it can explain photosynthesis beautifully and also invent a textbook that does not exist, in the same confident tone. Knowing this one fact changes how a family supervises homework use.

Where the name comes from

Each part of the term is doing honest work. "Language model" means a statistical model of language: a system that has absorbed patterns in how words follow other words. "Large" refers to scale, both in the training text (a meaningful slice of the internet, plus books and articles) and in the model itself, which has billions of internal settings, called parameters, that get tuned during training.

The training process is oddly simple to describe. The model is shown a passage with the next word hidden, makes a guess, gets corrected, and adjusts those internal dials a fraction. Repeat that billions of times and you get a system with an uncanny feel for what word comes next in almost any context.

The most useful mental image for a parent: the world's most well-read autocomplete. That is not an insult. Predicting text extremely well turns out to require absorbing an enormous amount of how humans explain, argue and describe, which is why the results feel so much like understanding.

Prediction, not understanding

Here is the gentle but important part. When your teenager asks an LLM to explain the causes of Federation, the model is not consulting a checked database of Australian history. It is generating the kind of text that typically follows a question like that, one word at a time.

Often the answer is accurate, because the training text contains a great deal of truth and the patterns encode it. But the model has no concept of true or false, only likely and unlikely. It cannot notice that it is wrong. It has no memory of a specific source it can point to, and no ability to feel uncertain the way a person does.

A comparison your teenager will recognise: an LLM is like a student who has read everything and verified nothing. Brilliant company, fluent in every subject, never once checked a footnote.

Why it sometimes makes things up

Once you see the prediction mechanism, the strangest behaviour of these tools stops being strange. Ask an LLM for a quote from a novel and it may produce a sentence the author never wrote, because it is generating quote-shaped text, not retrieving a quote. Ask for sources and it can produce citation-shaped references to papers that do not exist.

These failures are called hallucinations, and they follow directly from how the technology works rather than being an occasional glitch. The model is doing exactly what it was built to do: continue the text plausibly. Plausible and true usually overlap. Where they do not, the model cannot tell the difference, and neither can a Year 9 student in a hurry. We unpack the failure mode, and the habits that answer it, in what AI hallucinations are and why they matter for students.

What this means for supervising homework

You do not need to police every chat. Elevate Education's survey work suggests roughly three-quarters of Australian high-schoolers already use AI at least a few times a week, so the realistic parental job is shaping how, not whether. The broader picture sits in our pillar guide to AI education for teenagers in Australia.

Three habits follow directly from the prediction insight:

1. Treat fluency as decoration, not evidence. A confident tone tells you nothing about accuracy. Teach your teenager to ask "how would I check this?" as a reflex.
2. Never accept a citation, quote or number unverified. These are precisely where prediction fails most often, and where a teacher's red pen lands hardest.
3. Use it for explanation, not for facts. "Explain this a different way" plays to the model's strength. "Give me the statistic on X" plays to its weakness.

Common misunderstandings worth clearing up

A few reassurances, since the term invites science-fiction imagery. An LLM is not conscious and has no goals; between prompts it is a text machine at rest. It is not looking things up on the internet unless it has explicitly been given a search tool. And it does not learn from your teenager's chats in real time - the model was trained beforehand, though anything typed in may be stored by the company, which is a privacy conversation worth having separately.

If you want the fuller journey from your teenager's question to the answer on screen, including fine-tuning and guardrails, see how AI chatbots actually work. None of it requires maths to supervise. The single sentence that carries a family a long way: it predicts, it does not know.

The recommendation: teach your teenager the one-line definition - an LLM predicts the next word, it does not understand - and let every household rule follow from it. Encourage its use as an explainer, ban unverified citations and statistics, and make "how did you check that?" as normal a question as "have you done your homework?". A teenager who genuinely gets what the tool is will use it better than one who has only been told when they may touch it.