This is a long post about how to craft good prompts for ChatGPT. Good prompts create good outputs. The post provides a lot of information about how to craft these prompts, and it is relevant even for people who are just looking for ChatGPT to write short bits of text like bulleted lists or marketing copy. The intended audience is non-technical, though the principles explored in this post can also be used by more technical people who are looking for ChatGPT to write a Python script or similar task.
If you’ve spent any time using ChatGPT, you will have figured out that crafting a good prompt is critically important. This is frequently called “prompt engineering,” but there need not be anything engineering-related, or even quantitative, about the prompt.
Below is a prompt that I created, and below the prompt is the output that ChatGPT generated. I will separate ChatGPT’s output with section breaks, so that it is clear to the reader which parts of this post were generated by AI and which were not. After the ChatGPT output, I’ll spend a bit of time analyzing what’s good about it and what’s not so great. Finally, I’ll give some ideas about how I would iterate on my initial prompt, in order to get an even better result.
The goal here is to make concrete the notion of crafting a good prompt. It’s important to note that while this prompt explicitly calls for an essay-length response, not all prompts require such lengthy output. You could just as easily write a prompt which requests a product description, a few paragraphs for a blog post, a bulleted list of divergent ideas, etc.
Here’s my prompt:
Write a 1,000 word essay about how to design good prompts for a tool like ChatGPT. The intended audience is non-technical. The use cases that these readers will be interested in could be anything from creating sales scripts, to drafting emails, to writing marketing and SEO copy, to brainstorming about business negotiations, to using divergent thinking to come up with growth ideas for their company. The essay should provide an overview of how generative AI tools like ChatGPT work from a non-technical perspective, so that readers have some context when they learn how to create good prompts.
[AI begins]
Generative AI tools like ChatGPT have revolutionized the way we communicate and create content. With the ability to generate natural language responses and suggestions, these tools can help individuals and businesses alike with tasks like drafting emails, writing marketing copy, and brainstorming growth ideas. However, the quality of the output largely depends on the quality of the prompts given to the tool. In this essay, we will discuss how to design good prompts for a tool like ChatGPT and explore some best practices for generating high-quality responses.
Before we dive into the specifics of prompt design, it’s important to understand how generative AI tools like ChatGPT work. At a high level, these tools use deep learning models to predict the most likely next word or sequence of words given the input text. This is accomplished by training the model on large amounts of data, which allows it to identify patterns and relationships between words and phrases. Once the model has been trained, it can generate new text by selecting the most probable sequence of words based on the input prompt.
Now that we have a basic understanding of how generative AI tools work, let’s explore how to design good prompts for these tools. The first step is to identify the purpose of the prompt. What task are you hoping to accomplish with the tool? Do you need help generating sales scripts, drafting emails, or coming up with growth ideas for your business? Once you have a clear idea of the purpose, you can begin to craft a prompt that will elicit the desired response.
When designing a prompt, it’s important to keep in mind that the tool will only generate responses based on the input it receives. Therefore, the prompt should be as clear and specific as possible. For example, if you want help drafting an email, your prompt should include key details like the recipient, the subject, and the main message you want to convey. The more specific your prompt, the more likely it is that the tool will generate a relevant and useful response.
Another important consideration when designing a prompt is the tone and style of the response you want to receive. If you’re drafting a marketing copy, for example, you may want to convey a certain tone or voice. To achieve this, you can craft a prompts that includes specific language or phrases that reflect the desired tone. Similarly, if you’re brainstorming business ideas, you can design a prompt that encourages divergent thinking by using open-ended questions or prompts that encourage creativity.
One of the key challenges when designing prompts for generative AI tools is balancing specificity with flexibility. On the one hand, you want to provide enough detail to elicit a relevant and useful response. On the other hand, you don’t want to constrain the tool too much by providing a prompt that is too rigid or specific. To achieve the right balance, it can be helpful to experiment with different prompts and see what type of responses you receive. Over time, you can refine your prompts to strike the right balance between specificity and flexibility.
Another important consideration when designing prompts is the context in which they will be used. For example, if you’re using a generative AI tool to draft sales scripts, you may want to take into account the industry or market you’re targeting. Similarly, if you’re drafting marketing copy, you may want to consider the brand identity and messaging you want to convey. By taking into account these contextual factors, you can design prompts that are more likely to generate relevant and useful responses.
Finally, it’s worth noting that designing good prompts is not a one-time process. As you use a generative tool like ChatGPT, you’ll likely encounter new challenges and opportunities where you can leverage the tool to achieve your goals. To make the most of these opportunities, it’s important to continually refine and improve your prompt design skills. This can involve experimenting with different types of prompts, analyzing the responses you receive, and iteratively refining your approach over time.
In summary, designing good prompts for a tool like ChatGPT requires a combination of specificity, flexibility, and contextual awareness. By identifying the purpose of the prompt, crafting clear and specific language, balancing rigidity and flexibility, and considering the context in which the tool will be used, you can generate high-quality responses that are tailored to your specific needs. With practice and refinement, you can unlock the full potential of generative AI tools to help you achieve your business and personal goals.
Overall, generative AI tools like ChatGPT are powerful resources that can help individuals and businesses alike with a variety of tasks, from drafting emails to brainstorming growth ideas. However, the quality of the output depends largely on the quality of the prompts given to the tool. By following the best practices outlined in this essay, you can design good prompts that generate relevant and useful responses, unlocking the full potential of these powerful tools.
[AI ends]
The first thing to note is that ChatGPT only generated 835 words, when I asked for 1,000. This has to do with some technical limitations that OpenAI places on ChatGPT’s output. If you want a technical explanation of this limit, this may help.
Aside from that minor quibble, I think that this is a pretty good response to the prompt given. You can see that the prompt is very specific: it includes certain salient key words (“non-technical”, “divergent thinking”, “growth ideas”, etc.) which tell ChatGPT what you are looking for. The prompt contains a sufficient level of specifcity such that ChatGPT’s output sounds natural, it addresses the user’s needs, and it provides a coherent and verbose answer to the question being sought.
Another thing that you should experiment with is how the output changes as the prompt changes. For example, I asked ChatGPT to rewrite this for a more technical audience, and it came up with the following:
Generative AI tools like ChatGPT rely on deep learning models to predict the most probable next word or sequence of words given an input text prompt. These models are trained on large amounts of data to identify patterns and relationships between words and phrases. Once trained, the models can generate new text by selecting the most probable sequence of words based on the input prompt.
Designing good prompts for generative AI tools is critical to achieving high-quality output. The first step is to identify the purpose of the prompt, which may include generating sales scripts, drafting emails, or brainstorming business ideas. The prompt should be clear, specific, and balanced to elicit relevant and useful responses.
To achieve the appropriate level of specificity and flexibility, experiment with different types of prompts and analyze the responses received. The context in which the tool will be used, such as the industry or market targeted, should be taken into account. Continual refinement of prompt design skills is necessary to take full advantage of generative AI tools.
Overall, generative AI tools like ChatGPT have the potential to help individuals and businesses with a variety of tasks, but high-quality output is contingent upon good prompt design. Understanding the workings of these deep learning models and their training process is essential in designing effective prompts.
I’m not as thrilled with this output as I am with the first one, possibly because the first output contains more concrete examples of what you should do to generate a good output from a prompt. But perhaps the assumption being made is that a “more technical audience” is one which has sufficient knowledge about large learning models, and technology in general, to intuit what a good prompt looks like.
In any event, the more you experiment with prompts, the better you will become at writing them. Generative AI tools like ChatGPT promise to upend much white collar knowledge work, and getting ahead of the curve by learning how to generate good prompts will pay off.