What is an LLM for kids?

A large language model is a next-word guesser trained on enormous piles of text. It does not remember your family, it does not look things up unless a tool is attached, and it does not 'know' the way a person knows. When it sounds sure, that is fluency — a pattern that often looks like an answer.

Imagine a very fast autocomplete that has read millions of stories, recipes, and homework forums. If you start a sentence, it continues in a style that statistically fits. That is powerful for drafts, jokes, and first explanations. It is a poor oracle for 'what did I score on last Tuesday's quiz?' or 'is this mushroom safe?'

Ask the same model two questions: 'What is a black hole?' and 'What did my teacher say yesterday?' Compare the tone. The first may be useful if you verify it. The second should fail honestly. In WonderMind, Sprout learners meet this as a guessing robot; Pioneer learners name tokens, context windows, and calibration.

Use an LLM to generate options, not to be the adult in the room. A calculator is honest about arithmetic. A search engine can show sources. A language model can invent a source that looks real. That difference is the first AI-literacy skill.