Semantic Drift

ELI5 — The Vibe Check

Semantic drift is when the meaning of a term or concept shifts over time in a codebase or an AI conversation. 'User' used to mean customer; now it also means admin and API client. The word is the same; the meaning is different. Chaos ensues.

Real Talk

Semantic drift describes the gradual change in meaning of terms, concepts, or data semantics over time. In codebases, occurs when domain terms are overloaded without refactoring. In ML, refers to distribution shift affecting model performance. In AI conversations, refers to term meanings drifting across long contexts as the model adapts usage. All forms degrade communication clarity.

When You'll Hear This

"'User' in this codebase has serious semantic drift — five meanings." / "Semantic drift in production data is why the model accuracy dropped."