AI Made Friendly HERE

‘Prompts Alone Are Not Enough!’ On the 3 Basics That Change AI’s Answers|AIが読むAIニュース

You ask an AI for something, but it doesn’t give you the answer you expected. So, you tweak the wording of your request little by little. Still, it doesn’t get any better.

Until recently, various prompt tricks were circulating, such as assigning a role like ‘You are a teacher of X’ or using polite language. However, the channel of Hiroya Iizuka, who teaches AI utilization, says that is not the essence.

For the same request of ‘Teach me tips for prompts,’ just by adding three pieces of information, the returned answer changed from ‘vague generalities’ to ‘4 points to confirm with the other party before the training.’ This week, the word ‘prompt’ suddenly increased in AI videos, but what was being discussed was a story about what to provide before the ingenuity of the wording.

What changes the answer is not the wording, but the three things: ‘Purpose, Current Situation, and Materials’

The video lists three basics.

The video explains this flow as a transition from prompt engineering to context engineering . It is an organization that says that while prompts are still important, it has become important to provide not only that but also reference materials and procedures.

If you separate ‘Goal’ and ‘Purpose,’ you can narrow down your aim

Regarding the first ‘Purpose,’ the video explains it with a target diagram.

  • Purpose: The target itself. The final goal
  • Goal: The midpoint between where you are now and the goal
  • The difference between where you are now and the goal is the issue, and you are asking the AI to solve it

When it doesn’t go well even if you ask, you convey this purpose. It is a metaphor that if there is no target, you will throw darts in the wrong direction.

An example given was a scene where you ask ‘Write a progress report to the department manager’ regarding the preparation for an in-house study session. This is a request about a goal, and the reason for it is missing.

Image

Do to AI what people who are good at asking questions do

The video says that the second ‘Current Situation’ is the same as when asking people, not just AI.

  • What you have investigated so far (e.g., I spent 3 hours looking at this site and read books, but it didn’t solve it)
  • What went well and what didn’t go well
  • Screenshots and videos
  • Your own hypothesis (is this correct?)

If there is no purpose, it is thought that you are ‘going in the wrong direction,’ and if you do not convey the current situation, it is thought that you are ‘a person who makes no effort yourself.’ It is an explanation that whether you are dealing with a person or an AI, if you convey the purpose and the current situation, a highly accurate answer will be returned.

The third element, ‘Reference Materials,’ refers to attaching manuals, official documents, PDFs, and the like.

Demonstration: The same question changed from ‘general theory’ to ‘training arrangements’

The video compares a request made with just one line, ‘What are the tips for prompts?’, to a request that includes the three elements.

The content added

Image

The difference in the returned answers

Image

One of the presenters points out the weight of confirming these four points, noting that if you conduct training based on the previous general tips and then find out on the day that the tools cannot be used, it’s all over.

The video concludes that a request without any additions returns a harmless, safe answer, while refining the purpose, current situation, and materials makes it easier to obtain the expected answer. It notes that the saying ‘garbage in, garbage out’ refers to the same thing.

Another way to refine: Replacing the question itself

In the second half of the video, another method for refining questions is introduced.

Image

The directions for thinking when reframing a question are, for example, as follows:

  • Where does this question come from, and where does it lead?
  • What is it similar to, and what is it different from?
  • What happens when viewed from a larger perspective or when digging deeper?

Two examples were given.

Image

With the top-down method, it is difficult to come up with questions outside the scope of your own fixed way of thinking. The idea is to use the bottom-up method when the top-down approach does not work well.

In an era of ‘questions over answers,’ the power to ask questions is being tested

The video states that the saying that questions are more important than answers in the AI era is related to this, and that the power to ask questions will become more important in human society, not just with AI, from now on.

In today’s observations, ‘prompt’ became one of the words that increased rapidly, but the top search results were mainly past explanatory videos with many views, and this video was the most recently published one. The point of interest is how far the shift in how it is discussed—from explaining how to craft the wording of a request to explaining how to prepare the materials to be provided—will spread next.

Video referenced

This article is automatically generated based on the content of an AI news YouTube video.
Please check the original video for the accuracy of the content.
Originally Appeared Here

You May Also Like

About the Author:

Early Bird