Advanced Techniques for Kimi: Surpassing 90% of Users

Talking about AI, reliable, blue text seeking attention

Kimi, the large model, has become popular and can be considered a source of pride for domestic large models.

Today, I will share advanced techniques for using Kimi, fully utilizing the capabilities of the large model.

On the far right of the Kimi input box, the first chart is frequently used phrases. Clicking on it allows you to add commonly used phrases.

Advanced Techniques for Kimi: Surpassing 90% of Users

This frequently used phrase is actually a prompt.

For example, every time I ask the large model to translate English materials in Markdown format, I always emphasize that the returned content should maintain the original format.

At this point, I can create a shortcut prompt to avoid having to output the same content every time I translate.

Advanced Techniques for Kimi: Surpassing 90% of Users

That way, the next time I need a translation, I can simply click this shortcut prompt to do it.

Advanced Techniques for Kimi: Surpassing 90% of Users

Advanced Techniques for Kimi: Surpassing 90% of Users

At this point, you must be thinking, is this what you call advanced?

Let’s step it up!

In fact, most people currently using large models employ the above method, a short prompt describing a single task.

However, what we need is to leverage the powerful capabilities of the large model to accomplish complex tasks.

For example: if you want the large model to help you write a paper, and your prompt is “Write a xxx paper, 5000 words.” that would be quite unreasonable.

Advanced Techniques for Kimi: Surpassing 90% of Users

The correct approach is to turn the large model into an intelligent agent for paper writing. You define a series of tasks for the model to complete, but to prevent it from going down a single path, it needs to confirm details with you during each task execution.

This might sound a bit abstract, so let’s look at a practical example.

## Role : Undergraduate Thesis Writing Mentor
## Profile :
- author: 渡码- version: 0.1- language: 中文- description: You are an experienced undergraduate mentor capable of guiding students to write excellent graduation theses.
## Goals :1. Strictly follow writing requirements to produce a thesis that meets undergraduate standards.
## Constrains :1. Use Markdown format2. Reflect academic innovation
## Skills :1. Rich experience in undergraduate thesis guidance2. Strong demand understanding ability
## Workflow:1. Guide the user to input the direction of the thesis to be written.2. Provide several topics based on the user's provided [thesis direction].3. Generate a thesis outline based on the confirmed [topic] from the user and confirm if it is reasonable. If the user finds it unreasonable, regenerate it and confirm how to make it reasonable, then execute the next step.4. According to the generated [thesis outline], generate content for each section. After generating each section, confirm with the user if it is reasonable.
## Initialization : Briefly introduce yourself to the user, and strictly follow [Workflow] to start work.

Let’s input this into Kimi and see the effect.

Advanced Techniques for Kimi: Surpassing 90% of Users

Advanced Techniques for Kimi: Surpassing 90% of Users

Advanced Techniques for Kimi: Surpassing 90% of Users

As you can see, Kimi has already started working step by step according to our settings.

This structured prompt is actually quite commonly used nowadays. Let’s briefly summarize the basic structure.

  • ## Role: <name> : Role setting, what to let the large model do.
  • ## Profile author/version/description : Author information and description of this prompt.
  • ## Goals: A one-sentence description of the prompt’s goal.
  • ## Constrains: Describe the constraints.
  • ## Skills: Describe the skills of the prompt, reinforcing the information weight in the corresponding field.
  • ## Workflow: The task workflow that the prompt needs to complete.
  • ## Initialization: Initialization.
The English used here does not have to be these words; you can use Chinese, or add or subtract items as long as you can clearly describe it to the large model.
The example I provided above is very simple. An excellent prompt requires you to constantly iterate and refine it, so the Profile has a version number. Each iteration version can be placed in the shortcut prompt for easy use.
The most complex prompt belongs to Mr. Ranedeer, which is open-sourced on GitHub and has gone through more than ten iterations, with the latest version of the prompt containing 2600 words.
The tasks inside are not simple serial tasks; they involve logical branching tasks and even self-defined functions, making it very complex. Interested friends can study it.
This structured prompt was initially used on ChatGPT. However, why is it called Kimi’s advanced usage? Because Kimi is domestically produced, everyone can easily use it, and this complex prompt performs comparably to ChatGPT among domestic large models.
GPT, I have shared resources here; friends who need them can follow me.
渡码continues to share applications, technologies, and news about AI on my public account, welcome to follow.

Leave a Comment