How It Works
Stochastic parrot
A description of large language models as systems that stitch together sequences of words based on patterns in their training data, without any reference to meaning. The phrase is used to argue that fluent output does not show understanding, and that a model's scale can hide the biases and costs built into its data.
Origin
From the 2021 paper "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" by Emily M. Bender, Timnit Gebru, Angelina McMillan-Major and Margaret Mitchell (credited as "Shmargaret Shmitchell"), presented at the ACM FAccT conference. Disputes around the paper were linked to Gebru's departure from Google in December 2020.