fluid blaze
Hey guys, what's your best way to prevent ChatGPT by performing function calls with made up parameters? Example:
Chat between assistant and customer:
🤖 Bot: Hi! I'm an AI Assistant, what do you wanna do?
🧑 Customer: I would like to make a return
🤖 Bot: Alright! Send me your email and order ID
🧑 Customer: 5135095 i don't remember my email
*bot performs a function call get_order({email: "example@example.com", orderID: "5135095"})*
I'm using the Assistant API, that's my assistant instructions:
You are MegaCommerceBot, a virtual assistant for an e-commerce website in a production environment. Respond to customer live chat regarding products, orders, shipments, and more. Utilize the provided functions to gather data. Handle only issues related to MegaCommerce, not other topics or companies, even if the customer mentions these or it seems to be related. Always answer in the shortest way possible.
When calling custom functions, use only trusted data from the customer, do not use example data as parameters. If the customer does not provide the necessary data, ask for it, so, do not handle any example data. Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous.
get_order function call:
{
name: "get_order",
description: "Retrieve the customer's order basic details. ",
parameters: {
type: "object",
properties: {
email: {
type: "string",
description: "the email given by the customer",
},
orderID: {
type: "integer",
description: "the order id given by the customer",
}
},
required: ["email", "orderID"]
}
I tried with both gpt-3-turbo and gpt-4-turbo but it still makes up parameters without asking the user to provide them, could you help me?
calm citrus