Using AI models

Using AI models is extremely beneficial, but it's essential to understand how to feed AI models with data and do it correctly.

Given the numerous questions from friends and colleagues about using globally available AI models, I decided to write a brief blog post about it. Here I will emphasize the most important rules and tips related to interacting with AI models such as Google Bard and OpenAI ChatGPT, in a secure way.

This post is not dedicated to tech-savvy people. Its aim is to explain the concept and rules for working with AI models in a secure way to a broad audience.

For those who want to delve deeper into the topic, I recommend
exploring the "Prompt Engineering" terms, which we plan to discuss in future blog posts.

Inserting sensitive data into AI prompts (Bard, ChatGPT...) is a serious mistake that could lead to potentially significant consequences.

By sensitive data, we consider personal data such as name, address, number of personal documents, bank cards, etc.

Best practices for working with AI models in a secure way

Here are some guidelines and best practices for working with AI models in a secure way:

• Use in isolated Environments: If possible, use versions of AI models that have an isolated environment. In this way, you will run the model within your control, without the risk of data leakage. It's not new that you have to pay for it and the free version doesn't support that feature generally.

  • Anonymization: When entering data into AI, consider anonymization or pseudonymization. Thus, even with unintentional exposure, the data will be unidentifiable.
  • Data Masking: Use masking techniques to replace sensitive information, maintaining data structure but without real information.
  • Regular Checks: Review interactions with AI and establish a monitoring mechanism, but without storing sensitive information in the records.
  • Role-Based Access Control(RBAC): Limit access to the AI system based on roles within your organization.
  • Training and Awareness: Ensure users and developers are aware of the rules. They shouldn't input sensitive data unless absolutely necessary.
  • Model Customization: Consider fine-tuning the model with specific, non-sensitive data.
  • Data Retention Policy: Determine how long data will be stored and when it will be deleted or anonymized.
  • Data Encryption: Ensure data encryption at all stages - both at rest and in transit. (You can do that on the Cloud!)
  • Incident Response Plan: Have a plan for security incidents, including steps for resolution, investigation, and notification. (Prepare the scenario, contacts...)

By applying these practices, you can safely use tools like Google Bart and ChatGPT.

10 prompts for safe usage

Now, let's move on to something specific, hands-on examples and cases that you can use right now:

1. Instead of: "How can I pay my apartment bill at [Address]?"
   Use: "How can I pay an electricity bill online?"

2. Instead of: "What are diabetes symptoms? My father, [Father's Name], believes he might have it."
   Use: "What are the symptoms of diabetes?"

3. Instead of: "I have a meeting at [Company Name] on [Exact Address]. How should I get there?"
   Use: "How do I commute from the suburbs to the city center?"

4. Instead of: "My son, [Son's Name], attends [School Name]. What are his academic responsibilities?"
   Use: "What are typical responsibilities for elementary school students?"

5. Instead of: "I lost my credit card with the number [Card Number]. What's my next step?"
   Use: "What should I do if my credit card is lost?"

6. Instead of: "My birthday is on [Exact Date]. Were there any significant events on this day?"
   Use: "What major events occurred on [Exact Date]?"

7. Instead of: "My child, [Child's Name], struggles with math and has attention issues. Can you assist?"
   Use: "I have challenges with math and maintaining attention. Can you help?"

8. Instead of: "My phone number is [Phone Number]. How do I block unwarranted calls?"
   Use: "How can I block unwanted calls on my phone?"

9. Instead of: "I reside at [Exact Address]. Where's the closest pharmacy?"
   Use: "Where can I locate a downtown pharmacy?"

10. Instead of: "I am [Full Name]. I'm seeking career advice in the IT sector."
    Use: "I'm looking for IT sector career advice."

It's a straightforward pattern: be careful, use it, and harness the immense power of AI judiciously!

In this AI age, data privacy and security can't be overstated.

I believe this article equips you with a better understanding and the means to adopt the best practices. Remember, your safety is YOUR priority!

For further inquiries or insights about AI, Data, and Cloud technologies, consider joining our workshops at the Google Developers Group in Sarajevo. For more information, visit www.GDGSarajevo.org.

Best regards,
Kemal C.