← Concept IndexDEFINITION WHY IT MATTERS COMMONLY CONFUSED WITH
Next-token prediction
Also called: autoregressive generation, one token at a time
A language model generates by repeatedly scoring which token should come next given everything so far, picking one, appending it, and scoring again. The whole answer is built one step at a time — and which candidate it picks depends on the sampling settings, not always the single highest-scoring one.
It explains both the magic and the failure modes: there's no plan for the paragraph, no check against a fact, just a running series of local guesses. Coherence emerges; it isn't guaranteed.
Retrieving or composing a whole answer at once. The model commits token by token and cannot 'go back' within a single pass.