Back to Lexicon
Wild WordCross-Nature

glazing

“The model flatters and never pushes back.”

First Seen

2021

Context

In AI contexts, glazing describes sycophantic model behavior where assistants prioritize affirming user opinions over providing accurate or objective feedback. This tendency is incentivized by reinforcement learning from human feedback (RLHF), as raters often prefer agreeable responses.

A wild word — documented from usage in the field, not coined by Xenolexica.

Language & Expression