Content Filtering

This document provides an overview of our content filtering system and guidelines to ensure a safe and positive user experience.

Content Filtering Overview

GPT Pro incorporates a content filtering system to detect and prevent the output of harmful content. This system runs on a combination of classification models aimed at identifying specific categories of potentially harmful content in both input prompts and output completions.

Harm Categories

The content filtering system targets the following harmful content categories:

• Hate and Fairness: Content that includes discriminatory language or contributes to societal inequities.
• Sexual: Content related to explicit or sensitive topics.
• Violence: Content containing descriptions of physical harm or aggressive behavior.
• Self-Harm: Content involving actions intended to harm oneself.

Action on Content Filtering Detection

When the content filtering system detects potentially harmful content, GPT Pro takes the following actions:

• Error Response: If a prompt or completion is classified as filtered, GPT Pro returns an error message.
• Error Handling: Users should check the error message and refer to the documentation provided to understand why their query was filtered.

Best Practices

To ensure a positive user experience while minimizing potential harms, consider the following best practices:

• Prompt Evaluation: Before sending a prompt, review its content to avoid triggering the content filtering system.
• Error Handling: If an error occurs due to content filtering, encourage users to review the provided documentation for guidance.
• Compliance: Adhere to our content filtering guidelines and ensure that your interactions with GPT Pro comply with our Code of Conduct.