Module 10.3 Safety and Integrity

While OpenAI has implemented safety features to enhance ChatGPT’s reliability, the system isn’t flawless. These safety measures include:

• Detecting and filtering harmful or inappropriate language.
• Moderating prompts that are dangerous or sensitive.
• Redirecting or disengaging from potentially harmful conversations.

However, mistakes can still occur. To mitigate risks, be aware of the following potential challenges:

1. AI Bias

1. AI bias occurs when the system generates responses that are harmful, inaccurate, or discriminatory. Examples include:
   • Stereotypes or negative portrayals of certain identities or cultural practices.
   • Limited or unequal responses based on users’ traits, such as language ability or literacy level.
2. How to address it:
   • Critically evaluate outputs for fairness and inclusivity.
   • Use reporting features to flag inappropriate content.
   • Craft thoughtful prompts to minimize potential bias.
3. Inaccuracies

2. ChatGPT is powered by large language models (LLMs), which predict likely responses based on input. While these models are advanced, they may still produce incorrect information.
3. How to address it:
   • Verify all outputs before sharing them with colleagues, parents, or students.
   • Double-check facts, references, and citations for accuracy.
4. Inappropriate Use

3. Generative AI can inadvertently be used for unintended purposes, such as generating inappropriate content or relying on it too heavily for sensitive communication.
4. Always exercise judgment and ensure outputs align with ethical and professional standards.