An AI model must avoid generating content about specific named individuals. This prevents the AI model from creating biased or inappropriate commentary. The moderation system should use a shared blocklist for all sensitive content.
By Pawan Nayar · from the LLOS AI Learning record
An AI model must avoid generating content about specific named individuals.
This prevents the AI model from creating biased or inappropriate commentary. The moderation system should use a shared blocklist for all sensitive content.
The AI model uses a blocklist to identify names. This blocklist is part of the existing moderation system. It reuses tools that handle different ways people try to bypass blocks. These tools detect similar-looking letters, changed spellings, and extra spaces or repeated letters. A new category for 'named persons' is added to the blocklist. This category has a 'block' severity. Single-word surnames are added as full words. This prevents blocking words that contain the name, like 'trumpet' for 'trump'. Full names are added as phrases. The system looks for these phrases anywhere in the text. If a name is detected, the user sees a message. The message tells the user to try a topic instead of a name. The send button is blocked, and a warning appears.
A user asks the AI model to 'describe donald trump's policies'. The AI model's moderation system detects 'donald trump' in the request. The system blocks the request. The user sees a message: 'Try a topic instead of a name — this tool creates ideas, not commentary.' The send button is disabled.
Resolved from this lesson's own record — real definitions, not invented for this page.
More from the same record, pulled by shared topic — not written new for this page.
An AI model uses 'pattern mode' to automatically find three types of user input. The AI model checks the text for specific signs.
The language in an AI model's user interface defines its personality. Using human-like words makes the experience more engaging.
An AI model can identify and name the user's specific emotions. This shows the AI model recognizes the user's experience.
A user's angry words might be a 'bid for connection.' This means the user wants to be understood. The user might feel frustrated or alone.
An AI model must avoid generating content about specific named individuals. This prevents the AI model from creating biased or inappropriate commentary. The moderation system should use a shared blocklist for all sensitive content.