Are you interested in learning how to use large language models like ChatGPT to develop applications and products? There is an exciting video course available that offers best practices for using models like ChatGPT, and many common use cases and a section on custom chatbots.
ChatGPT Prompt Engineering for Developers is available for a limited time at no cost. This course will provide best practices for product development and applications for ecommerce, as the course explores the many ways to develop applications using an LLM and the iterative prompting process using the OpenAI API. Students will learn to think about generating good prompts and will learn the iterative approach for developing good prompts. Students will also learn to develop a custom chatbot. The course is appropriate for new learners, and is still practical for an experienced machine learning engineer.
The course about ChatGPT prompt engineering for developers is presented by Isa Fulford, who is employed by OpenAI. The course presents best practice for using and developing applications using LLMs. The course presents materials discussing appropriate use for software development, typical use cases for developing products and applications and how to create a chatbot using an LLM.
There are two types of LLMs; the base LLM and instruction-tuned LLM. The base LLM predicts words based on a training test dataset and instruction tuned LLMs predict words based on giving instructions. Because instruction-tuned LLM’s are the “easiest use” of the LLM class, and the “safer” LLM’s to use, all proposed real-life use cases are asking to use instruction-set LLM’s as the primary option.
When you are using an instruction-set LLM, you want to provide clear, direct and specific instructions so that it is easier for the LLM to produce the outputs or actions that you want for your business case. Determining when to provide instructions to an instruction-set LLM is not dissimilar to determining when to provide instructions to someone else. In the next video, we will provide examples for providing clear, direct, and specific instruction, together with an additional principle of prompting: giving the LLM time to think.
Key Principles to Using Language Model APIs
Across this course, you will learn two principles for working with language models (like ChatGPT): write clear, and specific instructions and give the model time to think. We will again stress that these two themes are fundamental to achieving the best results in your applications, and we will investigate these themes at the use case level through the Jupyter Notebook material.
Installing and Configuring the OpenAI package
The course material describes how to install the openai library via pip, and how to configure your API key and input to the service. You will want to consider strategies for giving fair, clear, and specific instructions to the LLM that should include some consideration of delimiters, output requests with or without structure or condition checking, as well as the ideas related to “few-shot prompting”.
The potential of large language models
In this course, you will explore well-known and potentially unknown potential of large language models such as summarising text, inferring from text, transforming text and extending short text. You will learn to create a departmental summary, extract sentiment and emotion from reviews, translate text into other languages, and convert formats – from markdown to HTML and JSON.
Building a custom chatbot
One of the most satisfying things you will create in this course is a custom chatbot that uses the OpenAI ChatCompletions format, and a large language model, such as ChatGPT. The chatbot in the example you will build is an “orderbot” for a pizza restaurant. You will learn about the components of the ChatCompletions and how to guide the conversation with system, user and wizard messages.
The course also includes utilities to help manage conversations over multiple shifts, and demonstrations/examples of how the chatbot actually works. Context also matters. The model needs the relevant messages in the conversation as input to even understand the conversation and if the right response would be generated.
Getting started
This video course provides you with a very solid starting point to begin building applications that use large language models. Throughout the course, you will take the knowledge and practical experience acquired to build things that a very small number of people are able to do today.