How It Works
Autoregressive Models
An AI model that generates output one piece at a time, where each new piece is predicted based on everything generated so far. This is how most text-generating AI actually writes: predicting the next word based on all previous words, one word at a time.
Origin
“Autoregression” is a statistics term predating AI by nearly a century, introduced by statistician Udny Yule in a 1927 paper modeling a value as a function of its own past values. The term was carried into machine learning as sequence models were described as generating output step-by-step from prior output.