This article is the 10th in a series by Teacher Wang Jue introducing AIGC. For previous articles, please refer to:
-
#ai Teacher Wang Jue’s AIGC Educational Application Collection
When controlling large models to complete general tasks,
you only need to give instructions in natural language.
For example, in Kimi, we can directly type to request:
-
Help me generate a summary of a certain article/link, etc.
However, for some more complex, specific, and restrictive tasks,
we must use prompts.
Academically speaking, Prompt is a method for “fine-tuning large models,
guiding the model to output according to specific requirements
by inputting specific prompts.
For example, before asking a question, we first tell the large model:
Who you are, What you should do, How to do it,
and sometimes specify what you cannot do!
In this way:
the all-powerful general AI
is transformed into a specialized AI for a specific scenario!
This is the simplest method for adapting AI to specific scenarios.
For example, everyone can try this “Celebrity Database” AI Bot (just enter a celebrity’s name, such as “Yang Zhenning”)
We can see that the AI’s output is no longer in an “article” style,
but in a structured presentation:
The same AI
can complete different tasks using different prompts!
Let’s take a look at what the above prompt looks like:
(The following is the prompt written in coze)
Of course, coze is different from general large models,
it can enhance the capabilities of large models beyond the scope of “general language”
through “plugins,” “knowledge bases,” etc.,
providing richer and more professional services!
—— This is why Teacher Wang Jue recommends everyone use coze to develop AI functions
(If you don’t know about coze, you can refer to:《coze: Creating the World’s Strongest “AI Agent”》)
In the prompt,
not only can we specify the output content and format,
but we can also specify the “user interaction method”
What is “user interaction method”?
Everyone can first scan the code to experience the “Idiom Solitaire Game” made with coze
The above example shows
how the large model plays a role,
interacting with the user!
Moreover, it can even specify
the dependency relationship between the introduction and conclusion
The large model will check this relationship according to the prompt’s requirements,
which is truly amazing!
This function seems to be much more complicated than looking up information,
but for large models,
it can still be controlled using prompts!
Below is a screenshot of the prompt in coze:
Some may wonder, Teacher Wang Jue is really amazing
to write such good and comprehensive prompts!
Actually, it’s not me who is amazing, but coze is amazing!
We only need to write down the requirements we think of in the “character setup and response logic” (which is actually the prompt)
and then click the upper right “Optimize” button
coze will automatically help you optimize the prompt.
For example, the lengthy prompt for “Idiom Solitaire”
I actually just input four characters “Idiom Solitaire” 
(Of course, after coze optimized it, I added a little new requirement, which can be flexibly adjusted and tried, somewhat like natural language programming, which is very interesting!)
————————————————
Additionally, we used coze as an example
to introduce the role of prompts and how to write them.
In fact, most prompts
are universal for various large models,
so if needed,
we can copy the above prompts
and paste them directly into Kimi’s text box.
Kimi will tell you:
It has already become an “Idiom Solitaire Dialogue Machine”:
Also, in Kimi,
if you want to save this prompt,
for direct use next time,
that requires clicking the cube button
and then clicking “Add Common Phrase” (which is actually the prompt)
Paste the prompt into the text box:
Don’t forget to give it a name
so that in the future, you can call this prompt by entering “@name”
as shown below:
————————————————
Moreover, coze is a development tool for creating your own “AI agent” that is specialized,
while we can directly use general large models like Kimi.
We have seen how coze optimizes prompts,
saving us a lot of trouble!
In fact, Kimi also provides prompt optimization functions
with a powerful “Prompt Expert” feature in Kimi+:
(To learn more about Kimi and Kimi+, please refer to:《Kimi+ is here, no need to write prompts anymore!》)
Just input the prompts you want to write into the text box
(write as many points as you think of)
Kimi’s “Prompt Expert” will automatically help us write standardized prompts according to writing norms:
Finally, “Breaking News”:
The more than ten typical “scenarios” provided in Kimi+
including “Prompt Expert,” “PPT Assistant,”
are essentially built-in prompts,
controlling the large model to do “specialized tasks”!
If everyone has used Kimi+’s functions
or experienced the two mini-games with coze today,
you will definitely realize how important prompts are
for making large models into specialized tools, and
how powerful the function of large models as “knowledge tools” is!
To facilitate communication about the application of AI technology in education and teaching, Teacher Wang Jue has created a WeChat group, welcome to join:
This article is part of Teacher Wang Jue’s training course “Essential Skills for Teachers in the AI Era”. To introduce the above training course to your institution, please enter “AI training” in this public account to learn about contact information.
-
wk: Complete Guide to Micro-Courses, covering design, production, and application -
ppt: Collection of PPT Techniques -
xxkx: Articles on Learning Science Research
