Your child types a question into ChatGPT. Two seconds later, a full, fluent answer appears. It's natural to wonder: does it actually understand what I asked?
The one-sentence version
A large language model is a system that has read an enormous amount of text and learned, statistically, which word is likely to come next after any given sequence of words, and it uses that skill, one word at a time, to write its answer to you.
Why it feels like it "understands"
Language turns out to encode an enormous amount of the structure of the world. To reliably predict the next word in context, a system has to implicitly capture something like cause-and-effect and everyday knowledge, even though it was never explicitly taught any of that as a fact.
It isn't retrieving knowledge. It's reconstructing the shape of an answer from the shape of everything it has read before, which is why it can sound completely confident while being completely wrong.
What this means for your family
- Treat it like a very fast first draft, not a final answer.
- Ask "how do you know?" out loud.
- Use it together, at least at first.