AI Snippet 15: Iterative improvement

In last week's post, we showed how use cases are already familiar to some LLMs and can be used as templates. However, AI is not immune to error. If you're unsatisfied with the results or need to adjust the input, you can easily instruct the AI (like an LLM) to make improvements through an interactive dialogue. As an example, we’ll continue the use case description from our last post and ask ChatGPT-4 to correct the actor’s name.??

Prompt: Can you please adjust Use Case 3: Update Delivery Status. The actor shall be the printer instead of the delivery system.?

ChatGPT: Certainly! I'll adjust the Use Case 3: Update Delivery Status, changing the actor from the delivery system to the printer, which likely refers to the entity responsible for managing the initial stages of order fulfillment, such as printing or packaging the Use Case 3: Update Delivery Status?

Use Case Name: Update Delivery Status?

Actor: Printer?

Preconditions: An order has been placed and is ready for the printing or packaging phase.?

Postconditions: The delivery status of the order is updated in the system.?

Main Success Scenario (Basic Flow):?

1. Printer processes the order for printing or packaging.?

2. Once processing is complete, Printer updates the order status in the system as "Ready for Shipping."?

3. System reflects the updated status, which is accessible to the customer and the delivery team.?

Extensions (Alternative Flows):?

1a. Printing or packaging issues:?

1. Printer identifies an issue with printing or packaging.?

2. Printer updates the order status as "Printing/Packaging Delayed."?

3. System notifies the customer support team to inform the customer and manage expectations.?

Special Requirements: System must support real-time updates and notifications to ensure timely communication between the printer, delivery team, and customer.?

Trigger: Completion of printing or packaging of an order.?

Frequency of Use: Moderate to high, depending on order volume.?

Not much was expected to change here; only the actor should have been updated to "printer." However, ChatGPT made some assumptions before providing the updated use case, resulting in a completely different outcome. This error highlights the importance of carefully reviewing every output from ChatGPT, even when only a minor change is requested.?

?

By Andrea Herrmann and the IREB AI interest group – and ChatGPT 4?