GIGO. It dates back to 1957, when William Mellin of the U.S. Army explained to a newspaper reporter that computers were limited by their input. It means "Garbage In, Garbage Out" and is just as applicable to today's remarkable generative AI tools as it was to the clunky tube-based computing devices of 1957.

In the case of an artificial intelligence such as ChatGPT, there are numerous opportunities for garbage in and garbage out. Remember that we have no notion what it has been fed during its training. We are aware that its perception of facts is so fluid that it could almost serve as an American politician. John Schulman, co-founder of OpenAI (the creators of ChatGPT), states, "Our greatest concern was with regards to veracity, as the model likes to invent things."

Regardless of how effective your prompts are, there is always the possibility that the AI will simply invent information. However, there are many things you can do to ensure the best possible outcome when crafting prompts. This will be the focus of this tutorial.

ALSO READ:ChatGPT enables me to fix code quicker, but at what cost?

How to construct efficient ChatGPT prompts

To avoid the GIGO effect, it is best to create questions that encourage ChatGPT's LLM (large language model) to provide the best potential answers.

In fact, writing effective prompts has evolved into its own lucrative field, "prompt engineering." Sabrina Ortiz of ZDNET authored an article titled "Do you enjoy posing ChatGPT questions? This article explored the novel employment category "You Could Get Paid (A Lot) for It"

Consequently, focus on the remainder of this article. Who can say? It could help you develop the skills necessary to become one of those highly compensated, quick engineers. These jobs apparently pay between $175,000 and $335,000 per year. Hmm... See ya!

1. Communicate with the AI as you would with a person.

One of the most intriguing aspects of working with ChatGPT is that you do not program it; instead, you converse with it. As a programmer with formal training, I've had to abandon numerous behaviors when interacting with AI. Conversation with it (and with it) requires a mental adjustment.

When I say to address it as a person, I mean to address it as you would a coworker or teammate. If that is difficult, give it a name. If Alexa is already taken, consider "Bob." This is helpful because when you speak with Bob, you can include conversational details, anecdotes that add texture to your narrative, a variety of possibilities, and more.

When communicating with a person, it is natural to anticipate that they will initially misunderstand your point and request clarification. It would be natural for the person you are speaking with to stray from the topic at hand and require reorientation. It would be natural to fill in the backstory and ask complex questions, and it would also be natural to have to probe deeper and rephrase some of those questions based on the responses you received.

This is how you should communicate with ChatGPT.

2. Provide a setting and context

Writing a ChatGPT query requires more than a simple one-sentence inquiry. It frequently entails providing pertinent context information to establish the context of the query.

Suppose you wish to prepare for a marathon. You may inquire of ChatGPT:

How should I train for a marathon?

However, you will receive a significantly more nuanced inquiry if you state that you are training for your first marathon. The responses you receive will be more tailored to your requirements, as in:

I am a novice runner who has never completed a marathon, but I intend to do so within six months. How should I train for a marathon?

Do you see how you're providing the AI with significantly more information so that it can provide you with a more focused and useful response? Specifically, I do not run, dance, or leap. This is essentially an illustration. No way am I going to run a marathon (unless there's a V-Twin engine under my saddle). Here are two additional examples of context-establishing questions:

In a few months, I intend to travel to Spain and would like to learn some basic Spanish in order to communicate with locals. I'm searching for online resources that are appropriate for beginners and offer a structured and all-encompassing approach to learning the language. Can you recommend some online resources for a novice learning Spanish?

In this instance, rather than simply requesting learning resources, the context helps the AI learn how to communicate with locals on the ground. Here's another illustration:

I am a business owner interested in exploring how blockchain technology can be used to enhance the efficiency and transparency of supply chains. I am seeking a clear and concise explanation of the technology as well as examples of its application within the context of supply chain management. Can you describe the concept of blockchain technology and its prospective supply chain management applications?

In this example, rather than simply requesting information on blockchain and how it operates, the focus is on blockchain for supply chain efficacy and how it could be applied in a real-world setting. Please input that prompt into ChatGPT. Its response is extremely intriguing.

I'll give you one more example that is considerably more detailed. It demonstrates how to compose a comprehensive prompt. One note: I limit the answer to 500 words because ChatGPT seems to break between 500 and 700 words, leaving the stories in the middle of a sentence and failing to continue when requested to do so. I'm expecting that future iterations will provide more in-depth answers, as premise like the one I'm about to describe tend to generate really entertaining story beginnings.

Write a 500-word or less short story for me.

The plot is set in Boston in 2339. The entire tale takes place inside a bookstore in the Victorian style that would be at home in Diagon Alley. There are the following human beings inside the store:

Give this character a name and at least one ability or characteristic that influences their backstory and, potentially, the entire short story.

ALSO READ:How to compose code using ChatGPT

The helper: This is a clerk in the store, the assistant. He is called Todd.

The client and his companion: Jackson and Ophelia, two consumers, entered the store together. Jackson is dressed as if he is attending a Steampunk convention, whereas Ophelia is clearly returning home from a day of office work.

Evangeline, in her mid-40s, is another regular client of the store. Archibald, another patron, is a man who could be anywhere between 40 and 70 years old. He exudes an aura of mystery and comes across as both grandiose and clandestine. Something about Archibald makes the other individuals uneasy.

A common belief in retail sales is that there is always additional stock "in the back," in a storage area for items that may not be displayed on the shelves where customers peruse. This story's premise is that there is something very unusual "in the back" of this store.

Put it all together and convey an engaging and entertaining story.

You can see how more information gives the AI more to work with. First, enter "Write me a story about a bookstore" into ChatGPT and observe the output. Then enter the prompt above, and you will observe the difference.

3. Instruct the AI to assume an identity or career

One of the coolest features of ChatGPT is that it can write from the perspective of a particular individual or profession. A few months ago, I demonstrated how to make ChatGPT write as a pirate or Shakespeare, but you can also make it write as a teacher, marketing executive, journalist, or any other perspective you desire.

Here is an illustration. I will request that ChatGPT describe the Amazon Echo smart home device from the perspectives of a product manager, a caregiver, and a journalist. These are the three cues:

Describe the Amazon Echo Alexa device from the point of view of its product manager.

Describe the Amazon Echo Alexa device from the perspective of an adult offspring caring for a senior parent.

From a journalist's perspective, characterize the Amazon Echo Alexa device.

These three prompts can be entered into ChatGPT to view its full response. However, a few sentences will demonstrate how ChatGPT assumes distinct roles to provide various responses.

I can confidently state this is one of the most innovative and revolutionary products in the smart home industry, according to the response provided by the AI assuming the identity of the product manager.

The AI's response using the caregiver identity included the following: The device's ability to set reminders and alarms can be especially useful for elderly individuals who may have difficulty remembering to take their medication or attend appointments.

And the AI's response using the journalist identity included this sentence: From a journalistic standpoint, the Echo has made headlines due to privacy concerns regarding the collection and storage of user data.

You can see how various identities enable the AI to respond with a variety of perspectives. I'll elaborate on this notion by demonstrating how to program an AI to conduct a thought experiment. Let's examine some of the considerations that went into the development of Alexa:

It is the year 2012. Siri has been available on the iPhone for approximately a year, but no Alexa-like smart home device has been released. At an Amazon board meeting, the Echo intelligent assistant based on Alexa has just been proposed.

Provide the pro and con arguments that board members at that meeting would have likely discussed in determining whether or not to approve spending to invest in developing the device.

Include engineering design specialists and product champions if that provides a more comprehensive perspective.

Here is a useful advice. Minor modifications to your prompts can drastically alter ChatGPT's response. For instance, when I changed "Provide the pro and con arguments that..." to "Provide the pro and con arguments as dialog that...", ChatGPT rewrote its response from a list of pros and cons to a dialogue between participants.

Additional writing techniques

1. Please feel free to pose the question again. ChatGPT frequently modifies its response for each query.

2. Make minor adjustments to your prompts in order to elicit an improved response.

3. ChatGPT will remember previous conversations for as long as the current page remains active. If you abandon the page, its awareness will be lost. To clarify, ChatGPT will also occasionally lose the conversation thread "just because," so be aware that you may need to restart the conversation on occasion.

4. Similarly, turning the page will initiate new responses to the discussion.

5. Occasionally, answers longer than 500 words break down. Be careful to specify the desired response length.

6. You can modify and elucidate prompts based on how the AI has previously responded. If it misunderstands you, you may be able to simply explain what it missed and proceed.

7. Rephrase queries if it refuses to answer them as asked. Use personas to elicit responses that the target might not otherwise provide.

8. If you require citations, inform it to support or justify its answers with sources.

9. Continue to experiment.

10. Occasionally, ChatGPT simply fails. Continue to attempt, but also be willing to abandon and move on to alternative tools. It is not yet flawless.

What kinds of prompts are most effective with ChatGPT?

You can query virtually anything of ChatGPT, which is part of its allure. However, bear in mind that it is intended for written responses. If you want a list of websites, Google is your best option. Wolfram Alpha is the place to go if you require computation. Give ChatGPT open-ended prompts, encourage creativity (yeah, and we assumed only humans could be creative), don't be afraid to share personal experiences or emotions, and remember that the AI's knowledge will expire in 2021.

ALSO READ:How to use ChatGPT to build an application

What should I do if ChatGPT refuses to answer or if I disagree with its response?

ChatGPT incorporates a number of safeguards. For example, it tends to shut down when asked political queries. This is a feature of the system. Although you might be able to glean a response, it is unlikely to be of much use. Nonetheless, feel free to continue posing queries using alternative wording or perspectives.

You may ask as many inquiries as you like, although I was once told I had asked too many and advised to return later. Yes, it essentially stated, "Go away, kid, you're bothering me." After hours of interrogation, I probably was. This can be as or even more addictive than viewing YouTube videos of puppies.