Briefle words · Prompt craft
Majority label bias
The tendency of an AI to over-predict whichever label appears most often in its examples.
Say it to an AI
Label each ticket High, Medium or Low. High = many people blocked now; Medium = one team degraded; Low = requests and minor issues. Examples: “Whole office is offline” → High; “One shared file won't open” → Medium; “Request: new mouse” → Low. One line per ticket.
How an expert would ask for the stronger version of “IT ticket triage”.
See it
Try it as a puzzle →An office IT admin asks an AI to set a priority on this morning's helpdesk tickets.
Why: A's examples are 3 × High, and the reply skews every new ticket upward: 'out of toner' → High, 'a second monitor' → Medium.
Related words
Room 4 · Examples & references
Learn “Majority label bias” in a five-minute lesson