I’m struggling to implement a use case for my chatbot.
The scenario is the following: When the user wants to “create ticket”, the bot should ask him for a “description” and the “username”. Once the user gives him that information, the chatbot should then respond in specific format containing the “description” and “username” the user previously sent him.
Unfortunately, I can’t figure out the prompt. Currently I am using this, which isn’t working.
Respond by asking for [description] and [username]. Once I gave you the information, respond in this format: \n\n #Creating ticket: [description], for [username] ...
I realize that this could be solved using functions, however I am working with gpt-35-turbo version 0301 and therefore can’t use that.
Does anybody have any idea on how to improve my prompt?
Hi and welcome to the developer forum!
Why are you using the old 0301 model?
I am using Azure OpenAI and Microsoft hasn’t released the new models for my region.
Ok, well as a basic starting point, consider what you would day to a college if you wanted them to do the task, what are the exact instructions you would tell another person, those instructions will usually work well.
using the [description] markers is a good start, try giving an example, called a “shot” to the model, show it what it should do and then ask it to replicate that example template with the new input.
Thanks for the feedback. I tried adding an example in the prompt, however, the bot doesn’t seem to pick up on it. This is my improved prompt.
You help users create tickets. In the first step, ask them for [device], [description] and [username]. Ask again until you have [device], [description] and [username].
Extract them in the users next message. Then respond in this format:
#Creating Ticket: device: [device], description: [description], username: [username].
User: create ticket
Assistent: Provide me with device, description and username.
User: Description is “I have problem xy”, username is “user123”, device is "macbook123’
Assistent: #Creating Ticket: device: macbook123, description: I have problem xy, username: user123
—Replicate that example every time a user wants to ‘create ticket’
How can I improve this?
I think you might get some benefit from a short course series by DeepLeadrning.ai and OpenAI staff here, It goes through some excellent examples of prompt design and the thinking behind it.