Actually, connecting to the API itself wasn't particularly difficult. The challenge was always providing clear and effective instructions to the AI.
The challenge of producing high-quality results, tailored to a variety of categories, has always been challenging.
My staff and I used to wrestle with this idea for several nights, and it turned out to be called "prompt engineering."
--- p.5

For many, 'prompt engineering' remains a difficult area.
Even though they knew its importance, most people didn't know how to get started, and some even doubted its utility and value.
Looking at the discussions that mainly take place in the English-speaking world, it was easy to get that impression because they often focused on 'implementation in the development department' rather than 'the perspective of general users.'
But what contrasts with this is the discussion of the Japanese who are 'obsessed with manualizing and templating prompts.'
The techniques (called 'methods' in Japanese) and templates they independently organized were very impressive, but they were somewhat fragmented and not well known in Korea.
I felt the need to collect and organize their discussions and adapt them to Korean.
--- p.6

Prompt engineering is a new field, and much research still needs to be done.
There are many theories about its concept and application, and it may be absurd to try to manualize and standardize it.
However, to ensure that more people could participate in the ecosystem, someone had to do it, and they worked tirelessly, even though they knew it was not enough.
--- p.7

The concept of prompt in artificial intelligence is sometimes confused with the 'Command Prom--- pt' in the CLI (Command Line Interface) environment.
The command prompt is a concept that means 'a state or signal waiting for instructions from the user.'
While similar in concept, it differs from the concept of prompts in AI, which refers to instructions themselves. Language models like GPT are capable of "in-context learning."
This means learning the given context and reflecting it in your answer.
Of course, this learning is only temporary.
However, this means that the model's response can be completely different depending on small changes in the examples, commands, or word choices included in the given prompt, so it is necessary to design good prompts to achieve the desired results.

--- p.17