Are you using Generative AI such as ChatGPT and DALL-E in your work? In order to get the desired result from the Generative AI properly, the user's question, that is, the prompt sentence, must be elegant and clear. This is called "Prompt Engineering." Prompt engineering refers to the technique of designing the appropriate prompt to obtain the desired result when interacting with an AI model. How to get the best results with the same content! I will tell you the know-how to use Generative AI smarter.

7 easy-to-know, prompt engineering methods

Prompt engineering is useful for both Large Language Model (LLM) and Specialized Large Language Model (sLLM), but it is especially important for LLM. This is because LLM such as ChatGPT is designed to answer a wide range of topics, so more sophisticated prompts need to be entered for the exact results that users want. Lee Ho-soo, a first-generation AI expert from IBM Watson Lab, introduced prompt engineering methods that are good to apply to the field in his book 'Trend of Utilization by Generative AI Industry'. We have summarized seven good ways to apply them.

1) Give the role of speaker and listener!
You can get a better result if you assume that AI is the troubleshooter of an issue or problem and identify yourself or the person you are using. For example, rather than asking "the cause and prevention of norovirus infection", you can get a better answer when asking, "You are a family medicine and pediatrics specialist. Explain easily the causes and prevention of norovirus infection to parents with young children in their daily lives."

2) Be more specific with the situation and context!
I need to provide sufficient background information so that AI can understand my requests properly. It is about setting targets, difficulties, conditions, etc., such as 'easy enough for middle school students to understand' or 'Please analyze around three key keywords'. You can also consider the language or communication level appropriate for you by specifying your age group. This will help you come up with a much more accurate response.

3) Break up your complicated requests step by step!
If it's too complicated, AI can also split it into phases to ask questions and then derive the final result, so that it can answer without confusion. Even if you don't ask questions one by one, ask "Analyze the cause of this problem, check the results, and explain the solution," and AI will show you the right answer for each step.

4) Give us some advance information!
In order to derive in-depth insights into a specific field, it is also advisable to present specific information to AI first. It is a method of entering prior information that cannot be learned or searched elsewhere and then asking related questions. You can get more in-depth answers based on prior information.

5) Give me an example!
If the format of the desired result is clear, it is also good to provide a sample. You can pre-specify the desired output style, such as "Please answer in a pattern similar to the example below." If you clarify the format of the answer, the quality of the result can be further improved.

6) Collect new angles!
Even for questions with the same content, the answer can vary depending on who and from what perspective. You can collect opinions from a completely different perspective by changing the role you initially assigned to AI. For example, if you ask each of the perspectives of companies related to global warming and the perspective of the government, you will get different insights.

7) Experiment and repeat!
Try various prompts for the same question and test which method works best. It is a method of requesting in various expressions to compare the results, or requesting further improvement based on the initial answer. This way, AI can provide more specific answers by providing a more accurate understanding of the user's intentions. For example, if you enter "Explain it in detail" and think it is insufficient, you can persistently ask "Explain it in a more specific case" to achieve a high-quality result.

Text vs images, tips for better results

We've gathered a few more helpful tips depending on the format of the results you'll get from the Generative AI.

[텍스트 도출 시]

* USP : Unique Selling Propositoin 

[이미지 도출 시]

Subscribe to our newsletter