Lesson 2 of 9
Does it actually understand?
5 min read
It finishes your sentences, explains jokes, writes code. Surely it understands? The mechanism is humbler than it looks.
It predicts the next word
At its core, a language model does one thing: given the text so far, guess the next word — then the next, then the next. Everything it 'says' is that loop running fast. It's ranking likely continuations from patterns in its training, not consulting a mind that grasps meaning.
Fluency is next-word prediction at scale, not comprehension.
Meaning we read in
So does it understand? It has no beliefs, senses, or intentions — but its predictions are so good that useful behaviour falls out. The honest answer: it models language brilliantly and understanding poorly, and it's easy to mistake the first for the second.
It's a master of the form of language, not the experience of meaning.
This is why it can be confidently wrong: a fluent, likely-sounding sentence and a true one aren't the same thing.
What to remember
- —It predicts the next word, over and over.
- —Fluency isn't the same as understanding.
- —Plausible-sounding ≠ correct — check important claims.
How does a language model produce its answers?
Continue in the app
Take the whole AI, Explained course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.