One of the exciting things about a large language model is that you can use it to build a custom chatbot with only modest amounts of effort. The OpenAI GPT web interface provides a conversational interface via a large language model. But one of the cool things is that you can also use a large language model to build your own custom chatbot, such as an AI customer service agent or an AI order taker for restaurants. In this article, we will learn how to do that.
To build a chatbot using a large language model, we need to understand the components of the OpenAI chat completions format. The format is designed to make multi-turn conversations easy. We define two helper functions: the gpt_completion function and a new helper function that takes a list of messages as input. The messages can be from different roles, such as system messages, user messages, and assistant messages.
The system message sets the behavior and persona of the assistant. It acts as a high-level instruction for the conversation. The user messages are the input, and the assistant messages are the output. By providing a list of messages, we can have more interactive and dynamic conversations with the chatbot.
In addition to the chat completions format, we also need to define a helper function to collect user prompts and assistant responses. This function collects prompts from a user interface and appends them to a list called context. The context is then passed to the model every time we call it. This way, the model has the information it needs to determine what to do next.
To demonstrate the usage of the chatbot, we create an example conversation where the chatbot takes pizza orders. We define the system message to guide the conversation, including greeting the customer, collecting the order, asking for pickup or delivery, summarizing the order, and collecting payment. The user messages and assistant messages are added to the context, allowing the chatbot to remember the previous parts of the conversation.
Finally, we show how to create a JSON summary of the food order using the chatbot. The summary includes itemized prices for each item, including the pizza, toppings, drinks, and sides. This summary can be submitted to the order system.
In conclusion, building a custom chatbot with a large language model is an exciting and accessible task. By understanding the chat completions format and using helper functions, we can create dynamic and interactive conversations with the chatbot. The possibilities are endless, from customer service agents to order takers. So go ahead and build your own chatbot using the power of language models!
