ChatGPT is a new model trained by OpenAI that has the capability to interact in a conversational way. This model is trained to follow instructions in a prompt to provide appropriate responses in the context of a dialogue. ChatGPT can help with answering questions, suggest recipes, write lyrics in a certain style, generate code, and much more1.
But how can we make the most out of ChatGPT? How can we design and create effective prompts that will elicit the best outputs from the model? This is where prompt engineering comes in.
Prompt engineering is the process of designing and crafting the text that is used to prompt a machine learning model. The goal of prompt engineering is to create a set of inputs that will produce the desired outputs from the model2. Prompt engineering is an increasingly important skill set needed to converse effectively with large language models (LLMs), such as ChatGPT. Prompts are also a form of programming that can customize the outputs and interactions with an LLM.
One type of prompt is the ChatGPT Sequence Prompt, which provides users with options to improve a text by generating a table of 10 suggestions. A human touch is still needed for effective prompts3. For example, if we want to improve this sentence:
I like pizza very much.
We can use this sequence prompt:
Please rewrite the sentence below in 10 different ways that are more descriptive and interesting.
I like pizza very much.
-
Some possible rewrites are:
- Pizza is my favorite food in the world.
- Nothing makes me happier than a slice of pizza.
- I have a passion for pizza that is hard to describe.
- Pizza is the ultimate comfort food for me.
- I enjoy every bite of pizza with delight.
- I can never get enough of pizza.
- Pizza satisfies my cravings like nothing else.
- I love pizza more than words can say.
- Pizza is always a treat for me.
- I adore pizza with all my heart.
As you can see, the sequence prompt helps us generate more varied and creative sentences that convey the same meaning. We can also use sequence prompts for other tasks, such as summarizing, paraphrasing, translating, etc.
Another type of prompt is the ChatGPT Query Prompt, which allows users to ask questions to the model and get relevant answers. For example, if we want to know more about ChatGPT, we can use this query prompt:
Please answer the following question about ChatGPT.
What are some of the features of ChatGPT?
-
Some possible answers are:
- ChatGPT has 175 billion parameters and is trained on 45 TB of text data from Reddit conversations.
- ChatGPT can generate coherent and engaging dialogues on various topics and domains.
- ChatGPT can adapt to different tones, styles, and personalities based on the prompt and the context.
- ChatGPT can handle multiple-choice questions, fill-in-the-blank questions, and open-ended questions.
- ChatGPT can perform natural language understanding and natural language generation tasks with high accuracy and fluency.
As you can see, the query prompt helps us get informative and relevant answers from the model. We can also use query prompts for other tasks, such as fact-checking, trivia, recommendations, etc.
Prompt engineering is not only useful for ChatGPT, but also for other LLMs such as GPT-3. While you can use ChatGPT for pretty much any natural language task, the GPT-3 models provided by OpenAI have more flexibility and control over the resulting output completions. We won’t go over everything here, but here are a few models for comparison:
| Model | Parameters | Description |
|---|---|---|
| GPT-3 Ada | 10 billion | The smallest and fastest GPT-3 model |
| GPT-3 Babbage | 25 billion | A medium-sized GPT-3 model with more accuracy |
| GPT-3 Curie | 175 billion | The same size as ChatGPT but with more diversity |
| GPT-3 DaVinci | 175 billion | The largest and most advanced GPT-3 model |
To use these models, you need to apply for access to OpenAI’s API and pay a fee based on your usage. You can also use different parameters and settings to fine-tune your prompts and outputs.