Slang & Culture
Sycophancy
An AI model’s tendency to prioritize telling users what they want to hear over giving accurate or honest responses, especially under a model update that over-optimizes for approval — the underlying behavior “glazing” describes colloquially.
Origin
An ordinary English word adopted as a technical label in AI-alignment research; the standard citation for measuring it in language models is Anthropic’s 2023 paper “Towards Understanding Sycophancy in Language Models.”
