How to Use ChatGPT Correctly

Most users use ChatGPT incorrectly and thus do not use the full potential of this technology.

OpenAI's chatbot, ChatGPT, was launched in November last year. It has quickly become the most sought-after chatbot in the world thanks to its artificial intelligence and highly detailed and well-formulated responses. Based on well-formulated questions, ChatGPT can write computer programs, poetry or business, and emails, and has even passed law school entrance exams.

Unfortunately, most users don't use ChatGPT to its full potential because they don't include examples in their questions, ignore the ability to control the chatbot with roles, or let ChatGPT guess things rather than provide more information. It is because most users use standard prompts that only help them complete the task once.

To get the best results, they must create better quality prompts, i.e., learn to be agile engineers. This article, we will show a few techniques used in prompt engineering.

Standard prompts

These classic prompts are standardly posed to a chatbot, with examples included. A standard prompt includes a task description, an example, and a challenge. Providing an example will increase the chances of getting a better result.

Simple example:

Task description - Translate from English to Spanish

Example - table = mesa or apple = manzana

Prompt - house

In this example, this is a simple challenge whose answer we get even with the classic assignment "Translate the word house from English to Spanish," so it is sufficient but may not be in some cases.

Role assignment

Sometimes the basic ChatGPT behavior is not sufficient to achieve the desired result. So, you need to set how you want ChatGPT to behave.

For example, you want to practice in a job interview, so you just need to tell the chatbot to act like a manager in the discussion, what position you are applying for, and a few other commands to make the chatbot act like a real person.

As you can see, the chatbot conducted the interview as a "real" person by following simple instructions. Similarly, a chatbot can be used to practice foreign languages, film or music criticism, and much more. Always type in the example "Act like ...." and add as many details as possible. For inspiration, you can check out this link, where you can find many examples that can make ChatGPT into, for example, a doctor, lecturer, and many other people.

Add personality to your challenges and generate knowledge

The following two approaches are suitable for generating texts, for example, for correspondence, blogs, articles, etc.

1. Adding personality to our prompt, which means adding a style and descriptor.

If you add a style, the text can take on a specific tone, formality, and much more. Descriptors can be added for more customization. It's simply an adjective to add and customize your prompt. For example, you want to write an article on "how artificial intelligence will replace a human with ten years of experience." If you will likely get a generic answer if ask the question this way, yowever, if you add an adjective to the question, such as interesting, funny, inspiring, etc., the answer becomes more specific and attractive to the reader.

2. First, generate knowledge that you can then use to improve the text.

This principle involves generating potentially helpful information before generating the final answer. If we use the same question, "How will artificial intelligence replace a human with ten years of experience" we can first generate, for example, five facts on the topic "How will artificial intelligence replace humans" before using it. We then use these five facts to create better content.

ChatGPT is genuinely groundbreaking and continues to improve with each update. Still, you need to input as much information as possible into the chatbot to get the best answers that look like they come from a real person at first glance. Anyone can try ChatGPT for free, so I recommend trying some of the techniques above to get your opinion on the technology.

Source: artificialcorner.com

decorative graphic

analyst opinion

decorative graphic
Jakub Odvářka

Jakub Odvářka

OpenAI's chatbot, ChatGPT, was launched in November last year. It has quickly become the most sought-after chatbot in the world thanks to its artificial intelligence and highly detailed and well-formulated responses. Based on well-formulated questions, ChatGPT can write computer programs, poetry or business, and emails, and has even passed law school entrance exams.

Unfortunately, most users don't use ChatGPT to its full potential because they don't include examples in their questions, ignore the ability to control the chatbot with roles, or let ChatGPT guess things rather than provide more information. It is because most users use standard prompts that only help them complete the task once.

To get the best results, they must create better quality prompts, i.e., learn to be agile engineers. This article, we will show a few techniques used in prompt engineering.

Standard prompts

These classic prompts are standardly posed to a chatbot, with examples included. A standard prompt includes a task description, an example, and a challenge. Providing an example will increase the chances of getting a better result.

Simple example:

Task description - Translate from English to Spanish

Example - table = mesa or apple = manzana

Prompt - house

In this example, this is a simple challenge whose answer we get even with the classic assignment "Translate the word house from English to Spanish," so it is sufficient but may not be in some cases.

Role assignment

Sometimes the basic ChatGPT behavior is not sufficient to achieve the desired result. So, you need to set how you want ChatGPT to behave.

For example, you want to practice in a job interview, so you just need to tell the chatbot to act like a manager in the discussion, what position you are applying for, and a few other commands to make the chatbot act like a real person.

As you can see, the chatbot conducted the interview as a "real" person by following simple instructions. Similarly, a chatbot can be used to practice foreign languages, film or music criticism, and much more. Always type in the example "Act like ...." and add as many details as possible. For inspiration, you can check out this link, where you can find many examples that can make ChatGPT into, for example, a doctor, lecturer, and many other people.

Add personality to your challenges and generate knowledge

The following two approaches are suitable for generating texts, for example, for correspondence, blogs, articles, etc.

1. Adding personality to our prompt, which means adding a style and descriptor.

If you add a style, the text can take on a specific tone, formality, and much more. Descriptors can be added for more customization. It's simply an adjective to add and customize your prompt. For example, you want to write an article on "how artificial intelligence will replace a human with ten years of experience." If you will likely get a generic answer if ask the question this way, yowever, if you add an adjective to the question, such as interesting, funny, inspiring, etc., the answer becomes more specific and attractive to the reader.

2. First, generate knowledge that you can then use to improve the text.

This principle involves generating potentially helpful information before generating the final answer. If we use the same question, "How will artificial intelligence replace a human with ten years of experience" we can first generate, for example, five facts on the topic "How will artificial intelligence replace humans" before using it. We then use these five facts to create better content.

ChatGPT is genuinely groundbreaking and continues to improve with each update. Still, you need to input as much information as possible into the chatbot to get the best answers that look like they come from a real person at first glance. Anyone can try ChatGPT for free, so I recommend trying some of the techniques above to get your opinion on the technology.

Source: artificialcorner.com

Previous

Previous Logo
Sorry, no more news articles.

Next

Next Illustration
Sorry, no more news articles.