Each Token Narrows What Comes Next | Visual AI Learning Guide

The loop just appends one token at a time, so where does the coherence come from? Start with the first token of a sentence, and almost anything could follow it.

0:00 / 0:00

Auto-play

Every token appended to the sequence becomes part of the context for the next prediction. Early in a sentence, many continuations are plausible; as tokens accumulate, most of those branches stop being sensible and the remaining candidates cluster around fewer options. Coherence is therefore not planned in advance — it emerges because each new token constrains the next distribution, and the constraint compounds over the whole answer.